Pileus Git - ~andy/linux/blob - drivers/block/xen-blkfront.c

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