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 #include <xen/interface/io/protocols.h>
  51
  52 #include <asm/xen/hypervisor.h>
  53
  54 enum blkif_state {
  55         BLKIF_STATE_DISCONNECTED,
  56         BLKIF_STATE_CONNECTED,
  57         BLKIF_STATE_SUSPENDED,
  58 };
  59
  60 struct blk_shadow {
  61         struct blkif_request req;
  62         unsigned long request;
  63         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  64 };
  65
  66 static struct block_device_operations xlvbd_block_fops;
  67
  68 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  69
  70 /*
  71  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  72  * hang in private_data off the gendisk structure. We may end up
  73  * putting all kinds of interesting stuff here :-)
  74  */
  75 struct blkfront_info
  76 {
  77         struct xenbus_device *xbdev;
  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         int is_ready;
  92
  93         /**
  94          * The number of people holding this device open.  We won't allow a
  95          * hot-unplug unless this is 0.
  96          */
  97         int users;
  98 };
  99
 100 static DEFINE_SPINLOCK(blkif_io_lock);
 101
 102 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 103         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 104 #define GRANT_INVALID_REF       0
 105
 106 #define PARTS_PER_DISK          16
 107
 108 #define BLKIF_MAJOR(dev) ((dev)>>8)
 109 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 110
 111 #define DEV_NAME        "xvd"   /* name in /dev */
 112
 113 /* Information about our VBDs. */
 114 #define MAX_VBDS 64
 115 static LIST_HEAD(vbds_list);
 116
 117 static int get_id_from_freelist(struct blkfront_info *info)
 118 {
 119         unsigned long free = info->shadow_free;
 120         BUG_ON(free > BLK_RING_SIZE);
 121         info->shadow_free = info->shadow[free].req.id;
 122         info->shadow[free].req.id = 0x0fffffee; /* debug */
 123         return free;
 124 }
 125
 126 static void add_id_to_freelist(struct blkfront_info *info,
 127                                unsigned long id)
 128 {
 129         info->shadow[id].req.id  = info->shadow_free;
 130         info->shadow[id].request = 0;
 131         info->shadow_free = id;
 132 }
 133
 134 static void blkif_restart_queue_callback(void *arg)
 135 {
 136         struct blkfront_info *info = (struct blkfront_info *)arg;
 137         schedule_work(&info->work);
 138 }
 139
 140 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 141 {
 142         /* We don't have real geometry info, but let's at least return
 143            values consistent with the size of the device */
 144         sector_t nsect = get_capacity(bd->bd_disk);
 145         sector_t cylinders = nsect;
 146
 147         hg->heads = 0xff;
 148         hg->sectors = 0x3f;
 149         sector_div(cylinders, hg->heads * hg->sectors);
 150         hg->cylinders = cylinders;
 151         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 152                 hg->cylinders = 0xffff;
 153         return 0;
 154 }
 155
 156 /*
 157  * blkif_queue_request
 158  *
 159  * request block io
 160  *
 161  * id: for guest use only.
 162  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 163  * buffer: buffer to read/write into. this should be a
 164  *   virtual address in the guest os.
 165  */
 166 static int blkif_queue_request(struct request *req)
 167 {
 168         struct blkfront_info *info = req->rq_disk->private_data;
 169         unsigned long buffer_mfn;
 170         struct blkif_request *ring_req;
 171         struct req_iterator iter;
 172         struct bio_vec *bvec;
 173         unsigned long id;
 174         unsigned int fsect, lsect;
 175         int ref;
 176         grant_ref_t gref_head;
 177
 178         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 179                 return 1;
 180
 181         if (gnttab_alloc_grant_references(
 182                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 183                 gnttab_request_free_callback(
 184                         &info->callback,
 185                         blkif_restart_queue_callback,
 186                         info,
 187                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 188                 return 1;
 189         }
 190
 191         /* Fill out a communications ring structure. */
 192         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 193         id = get_id_from_freelist(info);
 194         info->shadow[id].request = (unsigned long)req;
 195
 196         ring_req->id = id;
 197         ring_req->sector_number = (blkif_sector_t)req->sector;
 198         ring_req->handle = info->handle;
 199
 200         ring_req->operation = rq_data_dir(req) ?
 201                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 202         if (blk_barrier_rq(req))
 203                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 204
 205         ring_req->nr_segments = 0;
 206         rq_for_each_segment(bvec, req, iter) {
 207                 BUG_ON(ring_req->nr_segments == BLKIF_MAX_SEGMENTS_PER_REQUEST);
 208                 buffer_mfn = pfn_to_mfn(page_to_pfn(bvec->bv_page));
 209                 fsect = bvec->bv_offset >> 9;
 210                 lsect = fsect + (bvec->bv_len >> 9) - 1;
 211                 /* install a grant reference. */
 212                 ref = gnttab_claim_grant_reference(&gref_head);
 213                 BUG_ON(ref == -ENOSPC);
 214
 215                 gnttab_grant_foreign_access_ref(
 216                                 ref,
 217                                 info->xbdev->otherend_id,
 218                                 buffer_mfn,
 219                                 rq_data_dir(req) );
 220
 221                 info->shadow[id].frame[ring_req->nr_segments] =
 222                                 mfn_to_pfn(buffer_mfn);
 223
 224                 ring_req->seg[ring_req->nr_segments] =
 225                                 (struct blkif_request_segment) {
 226                                         .gref       = ref,
 227                                         .first_sect = fsect,
 228                                         .last_sect  = lsect };
 229
 230                 ring_req->nr_segments++;
 231         }
 232
 233         info->ring.req_prod_pvt++;
 234
 235         /* Keep a private copy so we can reissue requests when recovering. */
 236         info->shadow[id].req = *ring_req;
 237
 238         gnttab_free_grant_references(gref_head);
 239
 240         return 0;
 241 }
 242
 243
 244 static inline void flush_requests(struct blkfront_info *info)
 245 {
 246         int notify;
 247
 248         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 249
 250         if (notify)
 251                 notify_remote_via_irq(info->irq);
 252 }
 253
 254 /*
 255  * do_blkif_request
 256  *  read a block; request is in a request queue
 257  */
 258 static void do_blkif_request(struct request_queue *rq)
 259 {
 260         struct blkfront_info *info = NULL;
 261         struct request *req;
 262         int queued;
 263
 264         pr_debug("Entered do_blkif_request\n");
 265
 266         queued = 0;
 267
 268         while ((req = elv_next_request(rq)) != NULL) {
 269                 info = req->rq_disk->private_data;
 270                 if (!blk_fs_request(req)) {
 271                         end_request(req, 0);
 272                         continue;
 273                 }
 274
 275                 if (RING_FULL(&info->ring))
 276                         goto wait;
 277
 278                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 279                          "(%u/%li) buffer:%p [%s]\n",
 280                          req, req->cmd, (unsigned long)req->sector,
 281                          req->current_nr_sectors,
 282                          req->nr_sectors, req->buffer,
 283                          rq_data_dir(req) ? "write" : "read");
 284
 285
 286                 blkdev_dequeue_request(req);
 287                 if (blkif_queue_request(req)) {
 288                         blk_requeue_request(rq, req);
 289 wait:
 290                         /* Avoid pointless unplugs. */
 291                         blk_stop_queue(rq);
 292                         break;
 293                 }
 294
 295                 queued++;
 296         }
 297
 298         if (queued != 0)
 299                 flush_requests(info);
 300 }
 301
 302 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 303 {
 304         struct request_queue *rq;
 305
 306         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 307         if (rq == NULL)
 308                 return -1;
 309
 310         elevator_init(rq, "noop");
 311
 312         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 313         blk_queue_hardsect_size(rq, sector_size);
 314         blk_queue_max_sectors(rq, 512);
 315
 316         /* Each segment in a request is up to an aligned page in size. */
 317         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 318         blk_queue_max_segment_size(rq, PAGE_SIZE);
 319
 320         /* Ensure a merged request will fit in a single I/O ring slot. */
 321         blk_queue_max_phys_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 322         blk_queue_max_hw_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 323
 324         /* Make sure buffer addresses are sector-aligned. */
 325         blk_queue_dma_alignment(rq, 511);
 326
 327         gd->queue = rq;
 328
 329         return 0;
 330 }
 331
 332
 333 static int xlvbd_barrier(struct blkfront_info *info)
 334 {
 335         int err;
 336
 337         err = blk_queue_ordered(info->rq,
 338                                 info->feature_barrier ? QUEUE_ORDERED_DRAIN : QUEUE_ORDERED_NONE,
 339                                 NULL);
 340
 341         if (err)
 342                 return err;
 343
 344         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 345                info->gd->disk_name,
 346                info->feature_barrier ? "enabled" : "disabled");
 347         return 0;
 348 }
 349
 350
 351 static int xlvbd_alloc_gendisk(int minor, blkif_sector_t capacity,
 352                                int vdevice, u16 vdisk_info, u16 sector_size,
 353                                struct blkfront_info *info)
 354 {
 355         struct gendisk *gd;
 356         int nr_minors = 1;
 357         int err = -ENODEV;
 358
 359         BUG_ON(info->gd != NULL);
 360         BUG_ON(info->rq != NULL);
 361
 362         if ((minor % PARTS_PER_DISK) == 0)
 363                 nr_minors = PARTS_PER_DISK;
 364
 365         gd = alloc_disk(nr_minors);
 366         if (gd == NULL)
 367                 goto out;
 368
 369         if (nr_minors > 1)
 370                 sprintf(gd->disk_name, "%s%c", DEV_NAME,
 371                         'a' + minor / PARTS_PER_DISK);
 372         else
 373                 sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 374                         'a' + minor / PARTS_PER_DISK,
 375                         minor % PARTS_PER_DISK);
 376
 377         gd->major = XENVBD_MAJOR;
 378         gd->first_minor = minor;
 379         gd->fops = &xlvbd_block_fops;
 380         gd->private_data = info;
 381         gd->driverfs_dev = &(info->xbdev->dev);
 382         set_capacity(gd, capacity);
 383
 384         if (xlvbd_init_blk_queue(gd, sector_size)) {
 385                 del_gendisk(gd);
 386                 goto out;
 387         }
 388
 389         info->rq = gd->queue;
 390         info->gd = gd;
 391
 392         if (info->feature_barrier)
 393                 xlvbd_barrier(info);
 394
 395         if (vdisk_info & VDISK_READONLY)
 396                 set_disk_ro(gd, 1);
 397
 398         if (vdisk_info & VDISK_REMOVABLE)
 399                 gd->flags |= GENHD_FL_REMOVABLE;
 400
 401         if (vdisk_info & VDISK_CDROM)
 402                 gd->flags |= GENHD_FL_CD;
 403
 404         return 0;
 405
 406  out:
 407         return err;
 408 }
 409
 410 static void kick_pending_request_queues(struct blkfront_info *info)
 411 {
 412         if (!RING_FULL(&info->ring)) {
 413                 /* Re-enable calldowns. */
 414                 blk_start_queue(info->rq);
 415                 /* Kick things off immediately. */
 416                 do_blkif_request(info->rq);
 417         }
 418 }
 419
 420 static void blkif_restart_queue(struct work_struct *work)
 421 {
 422         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 423
 424         spin_lock_irq(&blkif_io_lock);
 425         if (info->connected == BLKIF_STATE_CONNECTED)
 426                 kick_pending_request_queues(info);
 427         spin_unlock_irq(&blkif_io_lock);
 428 }
 429
 430 static void blkif_free(struct blkfront_info *info, int suspend)
 431 {
 432         /* Prevent new requests being issued until we fix things up. */
 433         spin_lock_irq(&blkif_io_lock);
 434         info->connected = suspend ?
 435                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 436         /* No more blkif_request(). */
 437         if (info->rq)
 438                 blk_stop_queue(info->rq);
 439         /* No more gnttab callback work. */
 440         gnttab_cancel_free_callback(&info->callback);
 441         spin_unlock_irq(&blkif_io_lock);
 442
 443         /* Flush gnttab callback work. Must be done with no locks held. */
 444         flush_scheduled_work();
 445
 446         /* Free resources associated with old device channel. */
 447         if (info->ring_ref != GRANT_INVALID_REF) {
 448                 gnttab_end_foreign_access(info->ring_ref, 0,
 449                                           (unsigned long)info->ring.sring);
 450                 info->ring_ref = GRANT_INVALID_REF;
 451                 info->ring.sring = NULL;
 452         }
 453         if (info->irq)
 454                 unbind_from_irqhandler(info->irq, info);
 455         info->evtchn = info->irq = 0;
 456
 457 }
 458
 459 static void blkif_completion(struct blk_shadow *s)
 460 {
 461         int i;
 462         for (i = 0; i < s->req.nr_segments; i++)
 463                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 464 }
 465
 466 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 467 {
 468         struct request *req;
 469         struct blkif_response *bret;
 470         RING_IDX i, rp;
 471         unsigned long flags;
 472         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 473         int error;
 474
 475         spin_lock_irqsave(&blkif_io_lock, flags);
 476
 477         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 478                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 479                 return IRQ_HANDLED;
 480         }
 481
 482  again:
 483         rp = info->ring.sring->rsp_prod;
 484         rmb(); /* Ensure we see queued responses up to 'rp'. */
 485
 486         for (i = info->ring.rsp_cons; i != rp; i++) {
 487                 unsigned long id;
 488                 int ret;
 489
 490                 bret = RING_GET_RESPONSE(&info->ring, i);
 491                 id   = bret->id;
 492                 req  = (struct request *)info->shadow[id].request;
 493
 494                 blkif_completion(&info->shadow[id]);
 495
 496                 add_id_to_freelist(info, id);
 497
 498                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 499                 switch (bret->operation) {
 500                 case BLKIF_OP_WRITE_BARRIER:
 501                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 502                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 503                                        info->gd->disk_name);
 504                                 error = -EOPNOTSUPP;
 505                                 info->feature_barrier = 0;
 506                                 xlvbd_barrier(info);
 507                         }
 508                         /* fall through */
 509                 case BLKIF_OP_READ:
 510                 case BLKIF_OP_WRITE:
 511                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 512                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 513                                         "request: %x\n", bret->status);
 514
 515                         ret = __blk_end_request(req, error, blk_rq_bytes(req));
 516                         BUG_ON(ret);
 517                         break;
 518                 default:
 519                         BUG();
 520                 }
 521         }
 522
 523         info->ring.rsp_cons = i;
 524
 525         if (i != info->ring.req_prod_pvt) {
 526                 int more_to_do;
 527                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 528                 if (more_to_do)
 529                         goto again;
 530         } else
 531                 info->ring.sring->rsp_event = i + 1;
 532
 533         kick_pending_request_queues(info);
 534
 535         spin_unlock_irqrestore(&blkif_io_lock, flags);
 536
 537         return IRQ_HANDLED;
 538 }
 539
 540
 541 static int setup_blkring(struct xenbus_device *dev,
 542                          struct blkfront_info *info)
 543 {
 544         struct blkif_sring *sring;
 545         int err;
 546
 547         info->ring_ref = GRANT_INVALID_REF;
 548
 549         sring = (struct blkif_sring *)__get_free_page(GFP_KERNEL);
 550         if (!sring) {
 551                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 552                 return -ENOMEM;
 553         }
 554         SHARED_RING_INIT(sring);
 555         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 556
 557         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 558         if (err < 0) {
 559                 free_page((unsigned long)sring);
 560                 info->ring.sring = NULL;
 561                 goto fail;
 562         }
 563         info->ring_ref = err;
 564
 565         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 566         if (err)
 567                 goto fail;
 568
 569         err = bind_evtchn_to_irqhandler(info->evtchn,
 570                                         blkif_interrupt,
 571                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 572         if (err <= 0) {
 573                 xenbus_dev_fatal(dev, err,
 574                                  "bind_evtchn_to_irqhandler failed");
 575                 goto fail;
 576         }
 577         info->irq = err;
 578
 579         return 0;
 580 fail:
 581         blkif_free(info, 0);
 582         return err;
 583 }
 584
 585
 586 /* Common code used when first setting up, and when resuming. */
 587 static int talk_to_backend(struct xenbus_device *dev,
 588                            struct blkfront_info *info)
 589 {
 590         const char *message = NULL;
 591         struct xenbus_transaction xbt;
 592         int err;
 593
 594         /* Create shared ring, alloc event channel. */
 595         err = setup_blkring(dev, info);
 596         if (err)
 597                 goto out;
 598
 599 again:
 600         err = xenbus_transaction_start(&xbt);
 601         if (err) {
 602                 xenbus_dev_fatal(dev, err, "starting transaction");
 603                 goto destroy_blkring;
 604         }
 605
 606         err = xenbus_printf(xbt, dev->nodename,
 607                             "ring-ref", "%u", info->ring_ref);
 608         if (err) {
 609                 message = "writing ring-ref";
 610                 goto abort_transaction;
 611         }
 612         err = xenbus_printf(xbt, dev->nodename,
 613                             "event-channel", "%u", info->evtchn);
 614         if (err) {
 615                 message = "writing event-channel";
 616                 goto abort_transaction;
 617         }
 618         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 619                             XEN_IO_PROTO_ABI_NATIVE);
 620         if (err) {
 621                 message = "writing protocol";
 622                 goto abort_transaction;
 623         }
 624
 625         err = xenbus_transaction_end(xbt, 0);
 626         if (err) {
 627                 if (err == -EAGAIN)
 628                         goto again;
 629                 xenbus_dev_fatal(dev, err, "completing transaction");
 630                 goto destroy_blkring;
 631         }
 632
 633         xenbus_switch_state(dev, XenbusStateInitialised);
 634
 635         return 0;
 636
 637  abort_transaction:
 638         xenbus_transaction_end(xbt, 1);
 639         if (message)
 640                 xenbus_dev_fatal(dev, err, "%s", message);
 641  destroy_blkring:
 642         blkif_free(info, 0);
 643  out:
 644         return err;
 645 }
 646
 647
 648 /**
 649  * Entry point to this code when a new device is created.  Allocate the basic
 650  * structures and the ring buffer for communication with the backend, and
 651  * inform the backend of the appropriate details for those.  Switch to
 652  * Initialised state.
 653  */
 654 static int blkfront_probe(struct xenbus_device *dev,
 655                           const struct xenbus_device_id *id)
 656 {
 657         int err, vdevice, i;
 658         struct blkfront_info *info;
 659
 660         /* FIXME: Use dynamic device id if this is not set. */
 661         err = xenbus_scanf(XBT_NIL, dev->nodename,
 662                            "virtual-device", "%i", &vdevice);
 663         if (err != 1) {
 664                 xenbus_dev_fatal(dev, err, "reading virtual-device");
 665                 return err;
 666         }
 667
 668         info = kzalloc(sizeof(*info), GFP_KERNEL);
 669         if (!info) {
 670                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 671                 return -ENOMEM;
 672         }
 673
 674         info->xbdev = dev;
 675         info->vdevice = vdevice;
 676         info->connected = BLKIF_STATE_DISCONNECTED;
 677         INIT_WORK(&info->work, blkif_restart_queue);
 678
 679         for (i = 0; i < BLK_RING_SIZE; i++)
 680                 info->shadow[i].req.id = i+1;
 681         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 682
 683         /* Front end dir is a number, which is used as the id. */
 684         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 685         dev->dev.driver_data = info;
 686
 687         err = talk_to_backend(dev, info);
 688         if (err) {
 689                 kfree(info);
 690                 dev->dev.driver_data = NULL;
 691                 return err;
 692         }
 693
 694         return 0;
 695 }
 696
 697
 698 static int blkif_recover(struct blkfront_info *info)
 699 {
 700         int i;
 701         struct blkif_request *req;
 702         struct blk_shadow *copy;
 703         int j;
 704
 705         /* Stage 1: Make a safe copy of the shadow state. */
 706         copy = kmalloc(sizeof(info->shadow), GFP_KERNEL);
 707         if (!copy)
 708                 return -ENOMEM;
 709         memcpy(copy, info->shadow, sizeof(info->shadow));
 710
 711         /* Stage 2: Set up free list. */
 712         memset(&info->shadow, 0, sizeof(info->shadow));
 713         for (i = 0; i < BLK_RING_SIZE; i++)
 714                 info->shadow[i].req.id = i+1;
 715         info->shadow_free = info->ring.req_prod_pvt;
 716         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 717
 718         /* Stage 3: Find pending requests and requeue them. */
 719         for (i = 0; i < BLK_RING_SIZE; i++) {
 720                 /* Not in use? */
 721                 if (copy[i].request == 0)
 722                         continue;
 723
 724                 /* Grab a request slot and copy shadow state into it. */
 725                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 726                 *req = copy[i].req;
 727
 728                 /* We get a new request id, and must reset the shadow state. */
 729                 req->id = get_id_from_freelist(info);
 730                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 731
 732                 /* Rewrite any grant references invalidated by susp/resume. */
 733                 for (j = 0; j < req->nr_segments; j++)
 734                         gnttab_grant_foreign_access_ref(
 735                                 req->seg[j].gref,
 736                                 info->xbdev->otherend_id,
 737                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 738                                 rq_data_dir(
 739                                         (struct request *)
 740                                         info->shadow[req->id].request));
 741                 info->shadow[req->id].req = *req;
 742
 743                 info->ring.req_prod_pvt++;
 744         }
 745
 746         kfree(copy);
 747
 748         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 749
 750         spin_lock_irq(&blkif_io_lock);
 751
 752         /* Now safe for us to use the shared ring */
 753         info->connected = BLKIF_STATE_CONNECTED;
 754
 755         /* Send off requeued requests */
 756         flush_requests(info);
 757
 758         /* Kick any other new requests queued since we resumed */
 759         kick_pending_request_queues(info);
 760
 761         spin_unlock_irq(&blkif_io_lock);
 762
 763         return 0;
 764 }
 765
 766 /**
 767  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 768  * driver restart.  We tear down our blkif structure and recreate it, but
 769  * leave the device-layer structures intact so that this is transparent to the
 770  * rest of the kernel.
 771  */
 772 static int blkfront_resume(struct xenbus_device *dev)
 773 {
 774         struct blkfront_info *info = dev->dev.driver_data;
 775         int err;
 776
 777         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 778
 779         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 780
 781         err = talk_to_backend(dev, info);
 782         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 783                 err = blkif_recover(info);
 784
 785         return err;
 786 }
 787
 788
 789 /*
 790  * Invoked when the backend is finally 'ready' (and has told produced
 791  * the details about the physical device - #sectors, size, etc).
 792  */
 793 static void blkfront_connect(struct blkfront_info *info)
 794 {
 795         unsigned long long sectors;
 796         unsigned long sector_size;
 797         unsigned int binfo;
 798         int err;
 799
 800         if ((info->connected == BLKIF_STATE_CONNECTED) ||
 801             (info->connected == BLKIF_STATE_SUSPENDED) )
 802                 return;
 803
 804         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
 805                 __func__, info->xbdev->otherend);
 806
 807         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 808                             "sectors", "%llu", &sectors,
 809                             "info", "%u", &binfo,
 810                             "sector-size", "%lu", &sector_size,
 811                             NULL);
 812         if (err) {
 813                 xenbus_dev_fatal(info->xbdev, err,
 814                                  "reading backend fields at %s",
 815                                  info->xbdev->otherend);
 816                 return;
 817         }
 818
 819         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
 820                             "feature-barrier", "%lu", &info->feature_barrier,
 821                             NULL);
 822         if (err)
 823                 info->feature_barrier = 0;
 824
 825         err = xlvbd_alloc_gendisk(BLKIF_MINOR(info->vdevice),
 826                                   sectors, info->vdevice,
 827                                   binfo, sector_size, info);
 828         if (err) {
 829                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
 830                                  info->xbdev->otherend);
 831                 return;
 832         }
 833
 834         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 835
 836         /* Kick pending requests. */
 837         spin_lock_irq(&blkif_io_lock);
 838         info->connected = BLKIF_STATE_CONNECTED;
 839         kick_pending_request_queues(info);
 840         spin_unlock_irq(&blkif_io_lock);
 841
 842         add_disk(info->gd);
 843
 844         info->is_ready = 1;
 845 }
 846
 847 /**
 848  * Handle the change of state of the backend to Closing.  We must delete our
 849  * device-layer structures now, to ensure that writes are flushed through to
 850  * the backend.  Once is this done, we can switch to Closed in
 851  * acknowledgement.
 852  */
 853 static void blkfront_closing(struct xenbus_device *dev)
 854 {
 855         struct blkfront_info *info = dev->dev.driver_data;
 856         unsigned long flags;
 857
 858         dev_dbg(&dev->dev, "blkfront_closing: %s removed\n", dev->nodename);
 859
 860         if (info->rq == NULL)
 861                 goto out;
 862
 863         spin_lock_irqsave(&blkif_io_lock, flags);
 864
 865         del_gendisk(info->gd);
 866
 867         /* No more blkif_request(). */
 868         blk_stop_queue(info->rq);
 869
 870         /* No more gnttab callback work. */
 871         gnttab_cancel_free_callback(&info->callback);
 872         spin_unlock_irqrestore(&blkif_io_lock, flags);
 873
 874         /* Flush gnttab callback work. Must be done with no locks held. */
 875         flush_scheduled_work();
 876
 877         blk_cleanup_queue(info->rq);
 878         info->rq = NULL;
 879
 880  out:
 881         xenbus_frontend_closed(dev);
 882 }
 883
 884 /**
 885  * Callback received when the backend's state changes.
 886  */
 887 static void backend_changed(struct xenbus_device *dev,
 888                             enum xenbus_state backend_state)
 889 {
 890         struct blkfront_info *info = dev->dev.driver_data;
 891         struct block_device *bd;
 892
 893         dev_dbg(&dev->dev, "blkfront:backend_changed.\n");
 894
 895         switch (backend_state) {
 896         case XenbusStateInitialising:
 897         case XenbusStateInitWait:
 898         case XenbusStateInitialised:
 899         case XenbusStateUnknown:
 900         case XenbusStateClosed:
 901                 break;
 902
 903         case XenbusStateConnected:
 904                 blkfront_connect(info);
 905                 break;
 906
 907         case XenbusStateClosing:
 908                 bd = bdget_disk(info->gd, 0);
 909                 if (bd == NULL)
 910                         xenbus_dev_fatal(dev, -ENODEV, "bdget failed");
 911
 912                 mutex_lock(&bd->bd_mutex);
 913                 if (info->users > 0)
 914                         xenbus_dev_error(dev, -EBUSY,
 915                                          "Device in use; refusing to close");
 916                 else
 917                         blkfront_closing(dev);
 918                 mutex_unlock(&bd->bd_mutex);
 919                 bdput(bd);
 920                 break;
 921         }
 922 }
 923
 924 static int blkfront_remove(struct xenbus_device *dev)
 925 {
 926         struct blkfront_info *info = dev->dev.driver_data;
 927
 928         dev_dbg(&dev->dev, "blkfront_remove: %s removed\n", dev->nodename);
 929
 930         blkif_free(info, 0);
 931
 932         kfree(info);
 933
 934         return 0;
 935 }
 936
 937 static int blkfront_is_ready(struct xenbus_device *dev)
 938 {
 939         struct blkfront_info *info = dev->dev.driver_data;
 940
 941         return info->is_ready;
 942 }
 943
 944 static int blkif_open(struct inode *inode, struct file *filep)
 945 {
 946         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 947         info->users++;
 948         return 0;
 949 }
 950
 951 static int blkif_release(struct inode *inode, struct file *filep)
 952 {
 953         struct blkfront_info *info = inode->i_bdev->bd_disk->private_data;
 954         info->users--;
 955         if (info->users == 0) {
 956                 /* Check whether we have been instructed to close.  We will
 957                    have ignored this request initially, as the device was
 958                    still mounted. */
 959                 struct xenbus_device *dev = info->xbdev;
 960                 enum xenbus_state state = xenbus_read_driver_state(dev->otherend);
 961
 962                 if (state == XenbusStateClosing)
 963                         blkfront_closing(dev);
 964         }
 965         return 0;
 966 }
 967
 968 static struct block_device_operations xlvbd_block_fops =
 969 {
 970         .owner = THIS_MODULE,
 971         .open = blkif_open,
 972         .release = blkif_release,
 973         .getgeo = blkif_getgeo,
 974 };
 975
 976
 977 static struct xenbus_device_id blkfront_ids[] = {
 978         { "vbd" },
 979         { "" }
 980 };
 981
 982 static struct xenbus_driver blkfront = {
 983         .name = "vbd",
 984         .owner = THIS_MODULE,
 985         .ids = blkfront_ids,
 986         .probe = blkfront_probe,
 987         .remove = blkfront_remove,
 988         .resume = blkfront_resume,
 989         .otherend_changed = backend_changed,
 990         .is_ready = blkfront_is_ready,
 991 };
 992
 993 static int __init xlblk_init(void)
 994 {
 995         if (!is_running_on_xen())
 996                 return -ENODEV;
 997
 998         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
 999                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1000                        XENVBD_MAJOR, DEV_NAME);
1001                 return -ENODEV;
1002         }
1003
1004         return xenbus_register_frontend(&blkfront);
1005 }
1006 module_init(xlblk_init);
1007
1008
1009 static void xlblk_exit(void)
1010 {
1011         return xenbus_unregister_driver(&blkfront);
1012 }
1013 module_exit(xlblk_exit);
1014
1015 MODULE_DESCRIPTION("Xen virtual block device frontend");
1016 MODULE_LICENSE("GPL");
1017 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1018 MODULE_ALIAS("xen:vbd");
1019 MODULE_ALIAS("xenblk");