drivers/block/xen-blkfront.c

   1 /*
   2  * blkfront.c
   3  *
   4  * XenLinux virtual block device driver.
   5  *
   6  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
   7  * Modifications by Mark A. Williamson are (c) Intel Research Cambridge
   8  * Copyright (c) 2004, Christian Limpach
   9  * Copyright (c) 2004, Andrew Warfield
  10  * Copyright (c) 2005, Christopher Clark
  11  * Copyright (c) 2005, XenSource Ltd
  12  *
  13  * This program is free software; you can redistribute it and/or
  14  * modify it under the terms of the GNU General Public License version 2
  15  * as published by the Free Software Foundation; or, when distributed
  16  * separately from the Linux kernel or incorporated into other
  17  * software packages, subject to the following license:
  18  *
  19  * Permission is hereby granted, free of charge, to any person obtaining a copy
  20  * of this source file (the "Software"), to deal in the Software without
  21  * restriction, including without limitation the rights to use, copy, modify,
  22  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  23  * and to permit persons to whom the Software is furnished to do so, subject to
  24  * the following conditions:
  25  *
  26  * The above copyright notice and this permission notice shall be included in
  27  * all copies or substantial portions of the Software.
  28  *
  29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  30  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  31  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  32  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  33  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  34  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  35  * IN THE SOFTWARE.
  36  */
  37
  38 #include <linux/interrupt.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/hdreg.h>
  41 #include <linux/cdrom.h>
  42 #include <linux/module.h>
  43 #include <linux/slab.h>
  44 #include <linux/smp_lock.h>
  45 #include <linux/scatterlist.h>
  46
  47 #include <xen/xen.h>
  48 #include <xen/xenbus.h>
  49 #include <xen/grant_table.h>
  50 #include <xen/events.h>
  51 #include <xen/page.h>
  52 #include <xen/platform_pci.h>
  53
  54 #include <xen/interface/grant_table.h>
  55 #include <xen/interface/io/blkif.h>
  56 #include <xen/interface/io/protocols.h>
  57
  58 #include <asm/xen/hypervisor.h>
  59
  60 enum blkif_state {
  61         BLKIF_STATE_DISCONNECTED,
  62         BLKIF_STATE_CONNECTED,
  63         BLKIF_STATE_SUSPENDED,
  64 };
  65
  66 struct blk_shadow {
  67         struct blkif_request req;
  68         unsigned long request;
  69         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  70 };
  71
  72 static const struct block_device_operations xlvbd_block_fops;
  73
  74 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  75
  76 /*
  77  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  78  * hang in private_data off the gendisk structure. We may end up
  79  * putting all kinds of interesting stuff here :-)
  80  */
  81 struct blkfront_info
  82 {
  83         struct mutex mutex;
  84         struct xenbus_device *xbdev;
  85         struct gendisk *gd;
  86         int vdevice;
  87         blkif_vdev_t handle;
  88         enum blkif_state connected;
  89         int ring_ref;
  90         struct blkif_front_ring ring;
  91         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  92         unsigned int evtchn, irq;
  93         struct request_queue *rq;
  94         struct work_struct work;
  95         struct gnttab_free_callback callback;
  96         struct blk_shadow shadow[BLK_RING_SIZE];
  97         unsigned long shadow_free;
  98         int feature_barrier;
  99         int is_ready;
 100 };
 101
 102 static DEFINE_SPINLOCK(blkif_io_lock);
 103
 104 static unsigned int nr_minors;
 105 static unsigned long *minors;
 106 static DEFINE_SPINLOCK(minor_lock);
 107
 108 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 109         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 110 #define GRANT_INVALID_REF       0
 111
 112 #define PARTS_PER_DISK          16
 113 #define PARTS_PER_EXT_DISK      256
 114
 115 #define BLKIF_MAJOR(dev) ((dev)>>8)
 116 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 117
 118 #define EXT_SHIFT 28
 119 #define EXTENDED (1<<EXT_SHIFT)
 120 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 121 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 122
 123 #define DEV_NAME        "xvd"   /* name in /dev */
 124
 125 static int get_id_from_freelist(struct blkfront_info *info)
 126 {
 127         unsigned long free = info->shadow_free;
 128         BUG_ON(free >= BLK_RING_SIZE);
 129         info->shadow_free = info->shadow[free].req.id;
 130         info->shadow[free].req.id = 0x0fffffee; /* debug */
 131         return free;
 132 }
 133
 134 static void add_id_to_freelist(struct blkfront_info *info,
 135                                unsigned long id)
 136 {
 137         info->shadow[id].req.id  = info->shadow_free;
 138         info->shadow[id].request = 0;
 139         info->shadow_free = id;
 140 }
 141
 142 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 143 {
 144         unsigned int end = minor + nr;
 145         int rc;
 146
 147         if (end > nr_minors) {
 148                 unsigned long *bitmap, *old;
 149
 150                 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
 151                                  GFP_KERNEL);
 152                 if (bitmap == NULL)
 153                         return -ENOMEM;
 154
 155                 spin_lock(&minor_lock);
 156                 if (end > nr_minors) {
 157                         old = minors;
 158                         memcpy(bitmap, minors,
 159                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 160                         minors = bitmap;
 161                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 162                 } else
 163                         old = bitmap;
 164                 spin_unlock(&minor_lock);
 165                 kfree(old);
 166         }
 167
 168         spin_lock(&minor_lock);
 169         if (find_next_bit(minors, end, minor) >= end) {
 170                 for (; minor < end; ++minor)
 171                         __set_bit(minor, minors);
 172                 rc = 0;
 173         } else
 174                 rc = -EBUSY;
 175         spin_unlock(&minor_lock);
 176
 177         return rc;
 178 }
 179
 180 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 181 {
 182         unsigned int end = minor + nr;
 183
 184         BUG_ON(end > nr_minors);
 185         spin_lock(&minor_lock);
 186         for (; minor < end; ++minor)
 187                 __clear_bit(minor, minors);
 188         spin_unlock(&minor_lock);
 189 }
 190
 191 static void blkif_restart_queue_callback(void *arg)
 192 {
 193         struct blkfront_info *info = (struct blkfront_info *)arg;
 194         schedule_work(&info->work);
 195 }
 196
 197 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 198 {
 199         /* We don't have real geometry info, but let's at least return
 200            values consistent with the size of the device */
 201         sector_t nsect = get_capacity(bd->bd_disk);
 202         sector_t cylinders = nsect;
 203
 204         hg->heads = 0xff;
 205         hg->sectors = 0x3f;
 206         sector_div(cylinders, hg->heads * hg->sectors);
 207         hg->cylinders = cylinders;
 208         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 209                 hg->cylinders = 0xffff;
 210         return 0;
 211 }
 212
 213 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 214                        unsigned command, unsigned long argument)
 215 {
 216         struct blkfront_info *info = bdev->bd_disk->private_data;
 217         int i;
 218
 219         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 220                 command, (long)argument);
 221
 222         switch (command) {
 223         case CDROMMULTISESSION:
 224                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 225                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 226                         if (put_user(0, (char __user *)(argument + i)))
 227                                 return -EFAULT;
 228                 return 0;
 229
 230         case CDROM_GET_CAPABILITY: {
 231                 struct gendisk *gd = info->gd;
 232                 if (gd->flags & GENHD_FL_CD)
 233                         return 0;
 234                 return -EINVAL;
 235         }
 236
 237         default:
 238                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 239                   command);*/
 240                 return -EINVAL; /* same return as native Linux */
 241         }
 242
 243         return 0;
 244 }
 245
 246 /*
 247  * blkif_queue_request
 248  *
 249  * request block io
 250  *
 251  * id: for guest use only.
 252  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 253  * buffer: buffer to read/write into. this should be a
 254  *   virtual address in the guest os.
 255  */
 256 static int blkif_queue_request(struct request *req)
 257 {
 258         struct blkfront_info *info = req->rq_disk->private_data;
 259         unsigned long buffer_mfn;
 260         struct blkif_request *ring_req;
 261         unsigned long id;
 262         unsigned int fsect, lsect;
 263         int i, ref;
 264         grant_ref_t gref_head;
 265         struct scatterlist *sg;
 266
 267         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 268                 return 1;
 269
 270         if (gnttab_alloc_grant_references(
 271                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 272                 gnttab_request_free_callback(
 273                         &info->callback,
 274                         blkif_restart_queue_callback,
 275                         info,
 276                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 277                 return 1;
 278         }
 279
 280         /* Fill out a communications ring structure. */
 281         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 282         id = get_id_from_freelist(info);
 283         info->shadow[id].request = (unsigned long)req;
 284
 285         ring_req->id = id;
 286         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
 287         ring_req->handle = info->handle;
 288
 289         ring_req->operation = rq_data_dir(req) ?
 290                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 291         if (req->cmd_flags & REQ_HARDBARRIER)
 292                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 293
 294         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
 295         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 296
 297         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
 298                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 299                 fsect = sg->offset >> 9;
 300                 lsect = fsect + (sg->length >> 9) - 1;
 301                 /* install a grant reference. */
 302                 ref = gnttab_claim_grant_reference(&gref_head);
 303                 BUG_ON(ref == -ENOSPC);
 304
 305                 gnttab_grant_foreign_access_ref(
 306                                 ref,
 307                                 info->xbdev->otherend_id,
 308                                 buffer_mfn,
 309                                 rq_data_dir(req) );
 310
 311                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 312                 ring_req->seg[i] =
 313                                 (struct blkif_request_segment) {
 314                                         .gref       = ref,
 315                                         .first_sect = fsect,
 316                                         .last_sect  = lsect };
 317         }
 318
 319         info->ring.req_prod_pvt++;
 320
 321         /* Keep a private copy so we can reissue requests when recovering. */
 322         info->shadow[id].req = *ring_req;
 323
 324         gnttab_free_grant_references(gref_head);
 325
 326         return 0;
 327 }
 328
 329
 330 static inline void flush_requests(struct blkfront_info *info)
 331 {
 332         int notify;
 333
 334         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 335
 336         if (notify)
 337                 notify_remote_via_irq(info->irq);
 338 }
 339
 340 /*
 341  * do_blkif_request
 342  *  read a block; request is in a request queue
 343  */
 344 static void do_blkif_request(struct request_queue *rq)
 345 {
 346         struct blkfront_info *info = NULL;
 347         struct request *req;
 348         int queued;
 349
 350         pr_debug("Entered do_blkif_request\n");
 351
 352         queued = 0;
 353
 354         while ((req = blk_peek_request(rq)) != NULL) {
 355                 info = req->rq_disk->private_data;
 356
 357                 if (RING_FULL(&info->ring))
 358                         goto wait;
 359
 360                 blk_start_request(req);
 361
 362                 if (req->cmd_type != REQ_TYPE_FS) {
 363                         __blk_end_request_all(req, -EIO);
 364                         continue;
 365                 }
 366
 367                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 368                          "(%u/%u) buffer:%p [%s]\n",
 369                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 370                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 371                          req->buffer, rq_data_dir(req) ? "write" : "read");
 372
 373                 if (blkif_queue_request(req)) {
 374                         blk_requeue_request(rq, req);
 375 wait:
 376                         /* Avoid pointless unplugs. */
 377                         blk_stop_queue(rq);
 378                         break;
 379                 }
 380
 381                 queued++;
 382         }
 383
 384         if (queued != 0)
 385                 flush_requests(info);
 386 }
 387
 388 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 389 {
 390         struct request_queue *rq;
 391
 392         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 393         if (rq == NULL)
 394                 return -1;
 395
 396         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 397
 398         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 399         blk_queue_logical_block_size(rq, sector_size);
 400         blk_queue_max_hw_sectors(rq, 512);
 401
 402         /* Each segment in a request is up to an aligned page in size. */
 403         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 404         blk_queue_max_segment_size(rq, PAGE_SIZE);
 405
 406         /* Ensure a merged request will fit in a single I/O ring slot. */
 407         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 408
 409         /* Make sure buffer addresses are sector-aligned. */
 410         blk_queue_dma_alignment(rq, 511);
 411
 412         /* Make sure we don't use bounce buffers. */
 413         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 414
 415         gd->queue = rq;
 416
 417         return 0;
 418 }
 419
 420
 421 static int xlvbd_barrier(struct blkfront_info *info)
 422 {
 423         int err;
 424         const char *barrier;
 425
 426         switch (info->feature_barrier) {
 427         case QUEUE_ORDERED_DRAIN:       barrier = "enabled (drain)"; break;
 428         case QUEUE_ORDERED_TAG:         barrier = "enabled (tag)"; break;
 429         case QUEUE_ORDERED_NONE:        barrier = "disabled"; break;
 430         default:                        return -EINVAL;
 431         }
 432
 433         err = blk_queue_ordered(info->rq, info->feature_barrier);
 434
 435         if (err)
 436                 return err;
 437
 438         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 439                info->gd->disk_name, barrier);
 440         return 0;
 441 }
 442
 443
 444 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 445                                struct blkfront_info *info,
 446                                u16 vdisk_info, u16 sector_size)
 447 {
 448         struct gendisk *gd;
 449         int nr_minors = 1;
 450         int err = -ENODEV;
 451         unsigned int offset;
 452         int minor;
 453         int nr_parts;
 454
 455         BUG_ON(info->gd != NULL);
 456         BUG_ON(info->rq != NULL);
 457
 458         if ((info->vdevice>>EXT_SHIFT) > 1) {
 459                 /* this is above the extended range; something is wrong */
 460                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 461                 return -ENODEV;
 462         }
 463
 464         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 465                 minor = BLKIF_MINOR(info->vdevice);
 466                 nr_parts = PARTS_PER_DISK;
 467         } else {
 468                 minor = BLKIF_MINOR_EXT(info->vdevice);
 469                 nr_parts = PARTS_PER_EXT_DISK;
 470         }
 471
 472         if ((minor % nr_parts) == 0)
 473                 nr_minors = nr_parts;
 474
 475         err = xlbd_reserve_minors(minor, nr_minors);
 476         if (err)
 477                 goto out;
 478         err = -ENODEV;
 479
 480         gd = alloc_disk(nr_minors);
 481         if (gd == NULL)
 482                 goto release;
 483
 484         offset = minor / nr_parts;
 485
 486         if (nr_minors > 1) {
 487                 if (offset < 26)
 488                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 489                 else
 490                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 491                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 492         } else {
 493                 if (offset < 26)
 494                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 495                                 'a' + offset,
 496                                 minor & (nr_parts - 1));
 497                 else
 498                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 499                                 'a' + ((offset / 26) - 1),
 500                                 'a' + (offset % 26),
 501                                 minor & (nr_parts - 1));
 502         }
 503
 504         gd->major = XENVBD_MAJOR;
 505         gd->first_minor = minor;
 506         gd->fops = &xlvbd_block_fops;
 507         gd->private_data = info;
 508         gd->driverfs_dev = &(info->xbdev->dev);
 509         set_capacity(gd, capacity);
 510
 511         if (xlvbd_init_blk_queue(gd, sector_size)) {
 512                 del_gendisk(gd);
 513                 goto release;
 514         }
 515
 516         info->rq = gd->queue;
 517         info->gd = gd;
 518
 519         xlvbd_barrier(info);
 520
 521         if (vdisk_info & VDISK_READONLY)
 522                 set_disk_ro(gd, 1);
 523
 524         if (vdisk_info & VDISK_REMOVABLE)
 525                 gd->flags |= GENHD_FL_REMOVABLE;
 526
 527         if (vdisk_info & VDISK_CDROM)
 528                 gd->flags |= GENHD_FL_CD;
 529
 530         return 0;
 531
 532  release:
 533         xlbd_release_minors(minor, nr_minors);
 534  out:
 535         return err;
 536 }
 537
 538 static void xlvbd_release_gendisk(struct blkfront_info *info)
 539 {
 540         unsigned int minor, nr_minors;
 541         unsigned long flags;
 542
 543         if (info->rq == NULL)
 544                 return;
 545
 546         spin_lock_irqsave(&blkif_io_lock, flags);
 547
 548         /* No more blkif_request(). */
 549         blk_stop_queue(info->rq);
 550
 551         /* No more gnttab callback work. */
 552         gnttab_cancel_free_callback(&info->callback);
 553         spin_unlock_irqrestore(&blkif_io_lock, flags);
 554
 555         /* Flush gnttab callback work. Must be done with no locks held. */
 556         flush_scheduled_work();
 557
 558         del_gendisk(info->gd);
 559
 560         minor = info->gd->first_minor;
 561         nr_minors = info->gd->minors;
 562         xlbd_release_minors(minor, nr_minors);
 563
 564         blk_cleanup_queue(info->rq);
 565         info->rq = NULL;
 566
 567         put_disk(info->gd);
 568         info->gd = NULL;
 569 }
 570
 571 static void kick_pending_request_queues(struct blkfront_info *info)
 572 {
 573         if (!RING_FULL(&info->ring)) {
 574                 /* Re-enable calldowns. */
 575                 blk_start_queue(info->rq);
 576                 /* Kick things off immediately. */
 577                 do_blkif_request(info->rq);
 578         }
 579 }
 580
 581 static void blkif_restart_queue(struct work_struct *work)
 582 {
 583         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 584
 585         spin_lock_irq(&blkif_io_lock);
 586         if (info->connected == BLKIF_STATE_CONNECTED)
 587                 kick_pending_request_queues(info);
 588         spin_unlock_irq(&blkif_io_lock);
 589 }
 590
 591 static void blkif_free(struct blkfront_info *info, int suspend)
 592 {
 593         /* Prevent new requests being issued until we fix things up. */
 594         spin_lock_irq(&blkif_io_lock);
 595         info->connected = suspend ?
 596                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 597         /* No more blkif_request(). */
 598         if (info->rq)
 599                 blk_stop_queue(info->rq);
 600         /* No more gnttab callback work. */
 601         gnttab_cancel_free_callback(&info->callback);
 602         spin_unlock_irq(&blkif_io_lock);
 603
 604         /* Flush gnttab callback work. Must be done with no locks held. */
 605         flush_scheduled_work();
 606
 607         /* Free resources associated with old device channel. */
 608         if (info->ring_ref != GRANT_INVALID_REF) {
 609                 gnttab_end_foreign_access(info->ring_ref, 0,
 610                                           (unsigned long)info->ring.sring);
 611                 info->ring_ref = GRANT_INVALID_REF;
 612                 info->ring.sring = NULL;
 613         }
 614         if (info->irq)
 615                 unbind_from_irqhandler(info->irq, info);
 616         info->evtchn = info->irq = 0;
 617
 618 }
 619
 620 static void blkif_completion(struct blk_shadow *s)
 621 {
 622         int i;
 623         for (i = 0; i < s->req.nr_segments; i++)
 624                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 625 }
 626
 627 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 628 {
 629         struct request *req;
 630         struct blkif_response *bret;
 631         RING_IDX i, rp;
 632         unsigned long flags;
 633         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 634         int error;
 635
 636         spin_lock_irqsave(&blkif_io_lock, flags);
 637
 638         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 639                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 640                 return IRQ_HANDLED;
 641         }
 642
 643  again:
 644         rp = info->ring.sring->rsp_prod;
 645         rmb(); /* Ensure we see queued responses up to 'rp'. */
 646
 647         for (i = info->ring.rsp_cons; i != rp; i++) {
 648                 unsigned long id;
 649
 650                 bret = RING_GET_RESPONSE(&info->ring, i);
 651                 id   = bret->id;
 652                 req  = (struct request *)info->shadow[id].request;
 653
 654                 blkif_completion(&info->shadow[id]);
 655
 656                 add_id_to_freelist(info, id);
 657
 658                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 659                 switch (bret->operation) {
 660                 case BLKIF_OP_WRITE_BARRIER:
 661                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 662                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 663                                        info->gd->disk_name);
 664                                 error = -EOPNOTSUPP;
 665                                 info->feature_barrier = QUEUE_ORDERED_NONE;
 666                                 xlvbd_barrier(info);
 667                         }
 668                         /* fall through */
 669                 case BLKIF_OP_READ:
 670                 case BLKIF_OP_WRITE:
 671                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 672                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 673                                         "request: %x\n", bret->status);
 674
 675                         __blk_end_request_all(req, error);
 676                         break;
 677                 default:
 678                         BUG();
 679                 }
 680         }
 681
 682         info->ring.rsp_cons = i;
 683
 684         if (i != info->ring.req_prod_pvt) {
 685                 int more_to_do;
 686                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 687                 if (more_to_do)
 688                         goto again;
 689         } else
 690                 info->ring.sring->rsp_event = i + 1;
 691
 692         kick_pending_request_queues(info);
 693
 694         spin_unlock_irqrestore(&blkif_io_lock, flags);
 695
 696         return IRQ_HANDLED;
 697 }
 698
 699
 700 static int setup_blkring(struct xenbus_device *dev,
 701                          struct blkfront_info *info)
 702 {
 703         struct blkif_sring *sring;
 704         int err;
 705
 706         info->ring_ref = GRANT_INVALID_REF;
 707
 708         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 709         if (!sring) {
 710                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 711                 return -ENOMEM;
 712         }
 713         SHARED_RING_INIT(sring);
 714         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 715
 716         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 717
 718         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 719         if (err < 0) {
 720                 free_page((unsigned long)sring);
 721                 info->ring.sring = NULL;
 722                 goto fail;
 723         }
 724         info->ring_ref = err;
 725
 726         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 727         if (err)
 728                 goto fail;
 729
 730         err = bind_evtchn_to_irqhandler(info->evtchn,
 731                                         blkif_interrupt,
 732                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 733         if (err <= 0) {
 734                 xenbus_dev_fatal(dev, err,
 735                                  "bind_evtchn_to_irqhandler failed");
 736                 goto fail;
 737         }
 738         info->irq = err;
 739
 740         return 0;
 741 fail:
 742         blkif_free(info, 0);
 743         return err;
 744 }
 745
 746
 747 /* Common code used when first setting up, and when resuming. */
 748 static int talk_to_blkback(struct xenbus_device *dev,
 749                            struct blkfront_info *info)
 750 {
 751         const char *message = NULL;
 752         struct xenbus_transaction xbt;
 753         int err;
 754
 755         /* Create shared ring, alloc event channel. */
 756         err = setup_blkring(dev, info);
 757         if (err)
 758                 goto out;
 759
 760 again:
 761         err = xenbus_transaction_start(&xbt);
 762         if (err) {
 763                 xenbus_dev_fatal(dev, err, "starting transaction");
 764                 goto destroy_blkring;
 765         }
 766
 767         err = xenbus_printf(xbt, dev->nodename,
 768                             "ring-ref", "%u", info->ring_ref);
 769         if (err) {
 770                 message = "writing ring-ref";
 771                 goto abort_transaction;
 772         }
 773         err = xenbus_printf(xbt, dev->nodename,
 774                             "event-channel", "%u", info->evtchn);
 775         if (err) {
 776                 message = "writing event-channel";
 777                 goto abort_transaction;
 778         }
 779         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 780                             XEN_IO_PROTO_ABI_NATIVE);
 781         if (err) {
 782                 message = "writing protocol";
 783                 goto abort_transaction;
 784         }
 785
 786         err = xenbus_transaction_end(xbt, 0);
 787         if (err) {
 788                 if (err == -EAGAIN)
 789                         goto again;
 790                 xenbus_dev_fatal(dev, err, "completing transaction");
 791                 goto destroy_blkring;
 792         }
 793
 794         xenbus_switch_state(dev, XenbusStateInitialised);
 795
 796         return 0;
 797
 798  abort_transaction:
 799         xenbus_transaction_end(xbt, 1);
 800         if (message)
 801                 xenbus_dev_fatal(dev, err, "%s", message);
 802  destroy_blkring:
 803         blkif_free(info, 0);
 804  out:
 805         return err;
 806 }
 807
 808 /**
 809  * Entry point to this code when a new device is created.  Allocate the basic
 810  * structures and the ring buffer for communication with the backend, and
 811  * inform the backend of the appropriate details for those.  Switch to
 812  * Initialised state.
 813  */
 814 static int blkfront_probe(struct xenbus_device *dev,
 815                           const struct xenbus_device_id *id)
 816 {
 817         int err, vdevice, i;
 818         struct blkfront_info *info;
 819
 820         /* FIXME: Use dynamic device id if this is not set. */
 821         err = xenbus_scanf(XBT_NIL, dev->nodename,
 822                            "virtual-device", "%i", &vdevice);
 823         if (err != 1) {
 824                 /* go looking in the extended area instead */
 825                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 826                                    "%i", &vdevice);
 827                 if (err != 1) {
 828                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 829                         return err;
 830                 }
 831         }
 832
 833         if (xen_hvm_domain()) {
 834                 char *type;
 835                 int len;
 836                 /* no unplug has been done: do not hook devices != xen vbds */
 837                 if (xen_platform_pci_unplug & XEN_UNPLUG_IGNORE) {
 838                         int major;
 839
 840                         if (!VDEV_IS_EXTENDED(vdevice))
 841                                 major = BLKIF_MAJOR(vdevice);
 842                         else
 843                                 major = XENVBD_MAJOR;
 844
 845                         if (major != XENVBD_MAJOR) {
 846                                 printk(KERN_INFO
 847                                                 "%s: HVM does not support vbd %d as xen block device\n",
 848                                                 __FUNCTION__, vdevice);
 849                                 return -ENODEV;
 850                         }
 851                 }
 852                 /* do not create a PV cdrom device if we are an HVM guest */
 853                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 854                 if (IS_ERR(type))
 855                         return -ENODEV;
 856                 if (strncmp(type, "cdrom", 5) == 0) {
 857                         kfree(type);
 858                         return -ENODEV;
 859                 }
 860                 kfree(type);
 861         }
 862         info = kzalloc(sizeof(*info), GFP_KERNEL);
 863         if (!info) {
 864                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 865                 return -ENOMEM;
 866         }
 867
 868         mutex_init(&info->mutex);
 869         info->xbdev = dev;
 870         info->vdevice = vdevice;
 871         info->connected = BLKIF_STATE_DISCONNECTED;
 872         INIT_WORK(&info->work, blkif_restart_queue);
 873
 874         for (i = 0; i < BLK_RING_SIZE; i++)
 875                 info->shadow[i].req.id = i+1;
 876         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 877
 878         /* Front end dir is a number, which is used as the id. */
 879         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 880         dev_set_drvdata(&dev->dev, info);
 881
 882         err = talk_to_blkback(dev, info);
 883         if (err) {
 884                 kfree(info);
 885                 dev_set_drvdata(&dev->dev, NULL);
 886                 return err;
 887         }
 888
 889         return 0;
 890 }
 891
 892
 893 static int blkif_recover(struct blkfront_info *info)
 894 {
 895         int i;
 896         struct blkif_request *req;
 897         struct blk_shadow *copy;
 898         int j;
 899
 900         /* Stage 1: Make a safe copy of the shadow state. */
 901         copy = kmalloc(sizeof(info->shadow),
 902                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 903         if (!copy)
 904                 return -ENOMEM;
 905         memcpy(copy, info->shadow, sizeof(info->shadow));
 906
 907         /* Stage 2: Set up free list. */
 908         memset(&info->shadow, 0, sizeof(info->shadow));
 909         for (i = 0; i < BLK_RING_SIZE; i++)
 910                 info->shadow[i].req.id = i+1;
 911         info->shadow_free = info->ring.req_prod_pvt;
 912         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 913
 914         /* Stage 3: Find pending requests and requeue them. */
 915         for (i = 0; i < BLK_RING_SIZE; i++) {
 916                 /* Not in use? */
 917                 if (copy[i].request == 0)
 918                         continue;
 919
 920                 /* Grab a request slot and copy shadow state into it. */
 921                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 922                 *req = copy[i].req;
 923
 924                 /* We get a new request id, and must reset the shadow state. */
 925                 req->id = get_id_from_freelist(info);
 926                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 927
 928                 /* Rewrite any grant references invalidated by susp/resume. */
 929                 for (j = 0; j < req->nr_segments; j++)
 930                         gnttab_grant_foreign_access_ref(
 931                                 req->seg[j].gref,
 932                                 info->xbdev->otherend_id,
 933                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 934                                 rq_data_dir(
 935                                         (struct request *)
 936                                         info->shadow[req->id].request));
 937                 info->shadow[req->id].req = *req;
 938
 939                 info->ring.req_prod_pvt++;
 940         }
 941
 942         kfree(copy);
 943
 944         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 945
 946         spin_lock_irq(&blkif_io_lock);
 947
 948         /* Now safe for us to use the shared ring */
 949         info->connected = BLKIF_STATE_CONNECTED;
 950
 951         /* Send off requeued requests */
 952         flush_requests(info);
 953
 954         /* Kick any other new requests queued since we resumed */
 955         kick_pending_request_queues(info);
 956
 957         spin_unlock_irq(&blkif_io_lock);
 958
 959         return 0;
 960 }
 961
 962 /**
 963  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 964  * driver restart.  We tear down our blkif structure and recreate it, but
 965  * leave the device-layer structures intact so that this is transparent to the
 966  * rest of the kernel.
 967  */
 968 static int blkfront_resume(struct xenbus_device *dev)
 969 {
 970         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 971         int err;
 972
 973         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 974
 975         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 976
 977         err = talk_to_blkback(dev, info);
 978         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 979                 err = blkif_recover(info);
 980
 981         return err;
 982 }
 983
 984 static void
 985 blkfront_closing(struct blkfront_info *info)
 986 {
 987         struct xenbus_device *xbdev = info->xbdev;
 988         struct block_device *bdev = NULL;
 989
 990         mutex_lock(&info->mutex);
 991
 992         if (xbdev->state == XenbusStateClosing) {
 993                 mutex_unlock(&info->mutex);
 994                 return;
 995         }
 996
 997         if (info->gd)
 998                 bdev = bdget_disk(info->gd, 0);
 999
1000         mutex_unlock(&info->mutex);
1001
1002         if (!bdev) {
1003                 xenbus_frontend_closed(xbdev);
1004                 return;
1005         }
1006
1007         mutex_lock(&bdev->bd_mutex);
1008
1009         if (bdev->bd_openers) {
1010                 xenbus_dev_error(xbdev, -EBUSY,
1011                                  "Device in use; refusing to close");
1012                 xenbus_switch_state(xbdev, XenbusStateClosing);
1013         } else {
1014                 xlvbd_release_gendisk(info);
1015                 xenbus_frontend_closed(xbdev);
1016         }
1017
1018         mutex_unlock(&bdev->bd_mutex);
1019         bdput(bdev);
1020 }
1021
1022 /*
1023  * Invoked when the backend is finally 'ready' (and has told produced
1024  * the details about the physical device - #sectors, size, etc).
1025  */
1026 static void blkfront_connect(struct blkfront_info *info)
1027 {
1028         unsigned long long sectors;
1029         unsigned long sector_size;
1030         unsigned int binfo;
1031         int err;
1032         int barrier;
1033
1034         switch (info->connected) {
1035         case BLKIF_STATE_CONNECTED:
1036                 /*
1037                  * Potentially, the back-end may be signalling
1038                  * a capacity change; update the capacity.
1039                  */
1040                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1041                                    "sectors", "%Lu", &sectors);
1042                 if (XENBUS_EXIST_ERR(err))
1043                         return;
1044                 printk(KERN_INFO "Setting capacity to %Lu\n",
1045                        sectors);
1046                 set_capacity(info->gd, sectors);
1047                 revalidate_disk(info->gd);
1048
1049                 /* fall through */
1050         case BLKIF_STATE_SUSPENDED:
1051                 return;
1052
1053         default:
1054                 break;
1055         }
1056
1057         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1058                 __func__, info->xbdev->otherend);
1059
1060         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1061                             "sectors", "%llu", &sectors,
1062                             "info", "%u", &binfo,
1063                             "sector-size", "%lu", &sector_size,
1064                             NULL);
1065         if (err) {
1066                 xenbus_dev_fatal(info->xbdev, err,
1067                                  "reading backend fields at %s",
1068                                  info->xbdev->otherend);
1069                 return;
1070         }
1071
1072         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1073                             "feature-barrier", "%lu", &barrier,
1074                             NULL);
1075
1076         /*
1077          * If there's no "feature-barrier" defined, then it means
1078          * we're dealing with a very old backend which writes
1079          * synchronously; draining will do what needs to get done.
1080          *
1081          * If there are barriers, then we can do full queued writes
1082          * with tagged barriers.
1083          *
1084          * If barriers are not supported, then there's no much we can
1085          * do, so just set ordering to NONE.
1086          */
1087         if (err)
1088                 info->feature_barrier = QUEUE_ORDERED_DRAIN;
1089         else if (barrier)
1090                 info->feature_barrier = QUEUE_ORDERED_TAG;
1091         else
1092                 info->feature_barrier = QUEUE_ORDERED_NONE;
1093
1094         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1095         if (err) {
1096                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1097                                  info->xbdev->otherend);
1098                 return;
1099         }
1100
1101         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1102
1103         /* Kick pending requests. */
1104         spin_lock_irq(&blkif_io_lock);
1105         info->connected = BLKIF_STATE_CONNECTED;
1106         kick_pending_request_queues(info);
1107         spin_unlock_irq(&blkif_io_lock);
1108
1109         add_disk(info->gd);
1110
1111         info->is_ready = 1;
1112 }
1113
1114 /**
1115  * Callback received when the backend's state changes.
1116  */
1117 static void blkback_changed(struct xenbus_device *dev,
1118                             enum xenbus_state backend_state)
1119 {
1120         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1121
1122         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1123
1124         switch (backend_state) {
1125         case XenbusStateInitialising:
1126         case XenbusStateInitWait:
1127         case XenbusStateInitialised:
1128         case XenbusStateUnknown:
1129         case XenbusStateClosed:
1130                 break;
1131
1132         case XenbusStateConnected:
1133                 blkfront_connect(info);
1134                 break;
1135
1136         case XenbusStateClosing:
1137                 blkfront_closing(info);
1138                 break;
1139         }
1140 }
1141
1142 static int blkfront_remove(struct xenbus_device *xbdev)
1143 {
1144         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1145         struct block_device *bdev = NULL;
1146         struct gendisk *disk;
1147
1148         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1149
1150         blkif_free(info, 0);
1151
1152         mutex_lock(&info->mutex);
1153
1154         disk = info->gd;
1155         if (disk)
1156                 bdev = bdget_disk(disk, 0);
1157
1158         info->xbdev = NULL;
1159         mutex_unlock(&info->mutex);
1160
1161         if (!bdev) {
1162                 kfree(info);
1163                 return 0;
1164         }
1165
1166         /*
1167          * The xbdev was removed before we reached the Closed
1168          * state. See if it's safe to remove the disk. If the bdev
1169          * isn't closed yet, we let release take care of it.
1170          */
1171
1172         mutex_lock(&bdev->bd_mutex);
1173         info = disk->private_data;
1174
1175         dev_warn(disk_to_dev(disk),
1176                  "%s was hot-unplugged, %d stale handles\n",
1177                  xbdev->nodename, bdev->bd_openers);
1178
1179         if (info && !bdev->bd_openers) {
1180                 xlvbd_release_gendisk(info);
1181                 disk->private_data = NULL;
1182                 kfree(info);
1183         }
1184
1185         mutex_unlock(&bdev->bd_mutex);
1186         bdput(bdev);
1187
1188         return 0;
1189 }
1190
1191 static int blkfront_is_ready(struct xenbus_device *dev)
1192 {
1193         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1194
1195         return info->is_ready && info->xbdev;
1196 }
1197
1198 static int blkif_open(struct block_device *bdev, fmode_t mode)
1199 {
1200         struct gendisk *disk = bdev->bd_disk;
1201         struct blkfront_info *info;
1202         int err = 0;
1203
1204         lock_kernel();
1205
1206         info = disk->private_data;
1207         if (!info) {
1208                 /* xbdev gone */
1209                 err = -ERESTARTSYS;
1210                 goto out;
1211         }
1212
1213         mutex_lock(&info->mutex);
1214
1215         if (!info->gd)
1216                 /* xbdev is closed */
1217                 err = -ERESTARTSYS;
1218
1219         mutex_unlock(&info->mutex);
1220
1221 out:
1222         unlock_kernel();
1223         return err;
1224 }
1225
1226 static int blkif_release(struct gendisk *disk, fmode_t mode)
1227 {
1228         struct blkfront_info *info = disk->private_data;
1229         struct block_device *bdev;
1230         struct xenbus_device *xbdev;
1231
1232         lock_kernel();
1233
1234         bdev = bdget_disk(disk, 0);
1235         bdput(bdev);
1236
1237         if (bdev->bd_openers)
1238                 goto out;
1239
1240         /*
1241          * Check if we have been instructed to close. We will have
1242          * deferred this request, because the bdev was still open.
1243          */
1244
1245         mutex_lock(&info->mutex);
1246         xbdev = info->xbdev;
1247
1248         if (xbdev && xbdev->state == XenbusStateClosing) {
1249                 /* pending switch to state closed */
1250                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1251                 xlvbd_release_gendisk(info);
1252                 xenbus_frontend_closed(info->xbdev);
1253         }
1254
1255         mutex_unlock(&info->mutex);
1256
1257         if (!xbdev) {
1258                 /* sudden device removal */
1259                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1260                 xlvbd_release_gendisk(info);
1261                 disk->private_data = NULL;
1262                 kfree(info);
1263         }
1264
1265 out:
1266         unlock_kernel();
1267         return 0;
1268 }
1269
1270 static const struct block_device_operations xlvbd_block_fops =
1271 {
1272         .owner = THIS_MODULE,
1273         .open = blkif_open,
1274         .release = blkif_release,
1275         .getgeo = blkif_getgeo,
1276         .ioctl = blkif_ioctl,
1277 };
1278
1279
1280 static const struct xenbus_device_id blkfront_ids[] = {
1281         { "vbd" },
1282         { "" }
1283 };
1284
1285 static struct xenbus_driver blkfront = {
1286         .name = "vbd",
1287         .owner = THIS_MODULE,
1288         .ids = blkfront_ids,
1289         .probe = blkfront_probe,
1290         .remove = blkfront_remove,
1291         .resume = blkfront_resume,
1292         .otherend_changed = blkback_changed,
1293         .is_ready = blkfront_is_ready,
1294 };
1295
1296 static int __init xlblk_init(void)
1297 {
1298         if (!xen_domain())
1299                 return -ENODEV;
1300
1301         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1302                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1303                        XENVBD_MAJOR, DEV_NAME);
1304                 return -ENODEV;
1305         }
1306
1307         return xenbus_register_frontend(&blkfront);
1308 }
1309 module_init(xlblk_init);
1310
1311
1312 static void __exit xlblk_exit(void)
1313 {
1314         return xenbus_unregister_driver(&blkfront);
1315 }
1316 module_exit(xlblk_exit);
1317
1318 MODULE_DESCRIPTION("Xen virtual block device frontend");
1319 MODULE_LICENSE("GPL");
1320 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1321 MODULE_ALIAS("xen:vbd");
1322 MODULE_ALIAS("xenblk");