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/mutex.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 DEFINE_MUTEX(blkfront_mutex);
  73 static const struct block_device_operations xlvbd_block_fops;
  74
  75 #define BLK_RING_SIZE __RING_SIZE((struct blkif_sring *)0, PAGE_SIZE)
  76
  77 /*
  78  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  79  * hang in private_data off the gendisk structure. We may end up
  80  * putting all kinds of interesting stuff here :-)
  81  */
  82 struct blkfront_info
  83 {
  84         struct mutex mutex;
  85         struct xenbus_device *xbdev;
  86         struct gendisk *gd;
  87         int vdevice;
  88         blkif_vdev_t handle;
  89         enum blkif_state connected;
  90         int ring_ref;
  91         struct blkif_front_ring ring;
  92         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  93         unsigned int evtchn, irq;
  94         struct request_queue *rq;
  95         struct work_struct work;
  96         struct gnttab_free_callback callback;
  97         struct blk_shadow shadow[BLK_RING_SIZE];
  98         unsigned long shadow_free;
  99         unsigned int feature_flush;
 100         int is_ready;
 101 };
 102
 103 static DEFINE_SPINLOCK(blkif_io_lock);
 104
 105 static unsigned int nr_minors;
 106 static unsigned long *minors;
 107 static DEFINE_SPINLOCK(minor_lock);
 108
 109 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 110         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 111 #define GRANT_INVALID_REF       0
 112
 113 #define PARTS_PER_DISK          16
 114 #define PARTS_PER_EXT_DISK      256
 115
 116 #define BLKIF_MAJOR(dev) ((dev)>>8)
 117 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 118
 119 #define EXT_SHIFT 28
 120 #define EXTENDED (1<<EXT_SHIFT)
 121 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 122 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 123
 124 #define DEV_NAME        "xvd"   /* name in /dev */
 125
 126 static int get_id_from_freelist(struct blkfront_info *info)
 127 {
 128         unsigned long free = info->shadow_free;
 129         BUG_ON(free >= BLK_RING_SIZE);
 130         info->shadow_free = info->shadow[free].req.id;
 131         info->shadow[free].req.id = 0x0fffffee; /* debug */
 132         return free;
 133 }
 134
 135 static void add_id_to_freelist(struct blkfront_info *info,
 136                                unsigned long id)
 137 {
 138         info->shadow[id].req.id  = info->shadow_free;
 139         info->shadow[id].request = 0;
 140         info->shadow_free = id;
 141 }
 142
 143 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 144 {
 145         unsigned int end = minor + nr;
 146         int rc;
 147
 148         if (end > nr_minors) {
 149                 unsigned long *bitmap, *old;
 150
 151                 bitmap = kzalloc(BITS_TO_LONGS(end) * sizeof(*bitmap),
 152                                  GFP_KERNEL);
 153                 if (bitmap == NULL)
 154                         return -ENOMEM;
 155
 156                 spin_lock(&minor_lock);
 157                 if (end > nr_minors) {
 158                         old = minors;
 159                         memcpy(bitmap, minors,
 160                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 161                         minors = bitmap;
 162                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 163                 } else
 164                         old = bitmap;
 165                 spin_unlock(&minor_lock);
 166                 kfree(old);
 167         }
 168
 169         spin_lock(&minor_lock);
 170         if (find_next_bit(minors, end, minor) >= end) {
 171                 for (; minor < end; ++minor)
 172                         __set_bit(minor, minors);
 173                 rc = 0;
 174         } else
 175                 rc = -EBUSY;
 176         spin_unlock(&minor_lock);
 177
 178         return rc;
 179 }
 180
 181 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 182 {
 183         unsigned int end = minor + nr;
 184
 185         BUG_ON(end > nr_minors);
 186         spin_lock(&minor_lock);
 187         for (; minor < end; ++minor)
 188                 __clear_bit(minor, minors);
 189         spin_unlock(&minor_lock);
 190 }
 191
 192 static void blkif_restart_queue_callback(void *arg)
 193 {
 194         struct blkfront_info *info = (struct blkfront_info *)arg;
 195         schedule_work(&info->work);
 196 }
 197
 198 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 199 {
 200         /* We don't have real geometry info, but let's at least return
 201            values consistent with the size of the device */
 202         sector_t nsect = get_capacity(bd->bd_disk);
 203         sector_t cylinders = nsect;
 204
 205         hg->heads = 0xff;
 206         hg->sectors = 0x3f;
 207         sector_div(cylinders, hg->heads * hg->sectors);
 208         hg->cylinders = cylinders;
 209         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 210                 hg->cylinders = 0xffff;
 211         return 0;
 212 }
 213
 214 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 215                        unsigned command, unsigned long argument)
 216 {
 217         struct blkfront_info *info = bdev->bd_disk->private_data;
 218         int i;
 219
 220         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 221                 command, (long)argument);
 222
 223         switch (command) {
 224         case CDROMMULTISESSION:
 225                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 226                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 227                         if (put_user(0, (char __user *)(argument + i)))
 228                                 return -EFAULT;
 229                 return 0;
 230
 231         case CDROM_GET_CAPABILITY: {
 232                 struct gendisk *gd = info->gd;
 233                 if (gd->flags & GENHD_FL_CD)
 234                         return 0;
 235                 return -EINVAL;
 236         }
 237
 238         default:
 239                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 240                   command);*/
 241                 return -EINVAL; /* same return as native Linux */
 242         }
 243
 244         return 0;
 245 }
 246
 247 /*
 248  * blkif_queue_request
 249  *
 250  * request block io
 251  *
 252  * id: for guest use only.
 253  * operation: BLKIF_OP_{READ,WRITE,PROBE}
 254  * buffer: buffer to read/write into. this should be a
 255  *   virtual address in the guest os.
 256  */
 257 static int blkif_queue_request(struct request *req)
 258 {
 259         struct blkfront_info *info = req->rq_disk->private_data;
 260         unsigned long buffer_mfn;
 261         struct blkif_request *ring_req;
 262         unsigned long id;
 263         unsigned int fsect, lsect;
 264         int i, ref;
 265         grant_ref_t gref_head;
 266         struct scatterlist *sg;
 267
 268         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 269                 return 1;
 270
 271         if (gnttab_alloc_grant_references(
 272                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 273                 gnttab_request_free_callback(
 274                         &info->callback,
 275                         blkif_restart_queue_callback,
 276                         info,
 277                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 278                 return 1;
 279         }
 280
 281         /* Fill out a communications ring structure. */
 282         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 283         id = get_id_from_freelist(info);
 284         info->shadow[id].request = (unsigned long)req;
 285
 286         ring_req->id = id;
 287         ring_req->sector_number = (blkif_sector_t)blk_rq_pos(req);
 288         ring_req->handle = info->handle;
 289
 290         ring_req->operation = rq_data_dir(req) ?
 291                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 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 void xlvbd_flush(struct blkfront_info *info)
 421 {
 422         blk_queue_flush(info->rq, info->feature_flush);
 423         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 424                info->gd->disk_name,
 425                info->feature_flush ? "enabled" : "disabled");
 426 }
 427
 428
 429 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 430                                struct blkfront_info *info,
 431                                u16 vdisk_info, u16 sector_size)
 432 {
 433         struct gendisk *gd;
 434         int nr_minors = 1;
 435         int err = -ENODEV;
 436         unsigned int offset;
 437         int minor;
 438         int nr_parts;
 439
 440         BUG_ON(info->gd != NULL);
 441         BUG_ON(info->rq != NULL);
 442
 443         if ((info->vdevice>>EXT_SHIFT) > 1) {
 444                 /* this is above the extended range; something is wrong */
 445                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 446                 return -ENODEV;
 447         }
 448
 449         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 450                 minor = BLKIF_MINOR(info->vdevice);
 451                 nr_parts = PARTS_PER_DISK;
 452         } else {
 453                 minor = BLKIF_MINOR_EXT(info->vdevice);
 454                 nr_parts = PARTS_PER_EXT_DISK;
 455         }
 456
 457         if ((minor % nr_parts) == 0)
 458                 nr_minors = nr_parts;
 459
 460         err = xlbd_reserve_minors(minor, nr_minors);
 461         if (err)
 462                 goto out;
 463         err = -ENODEV;
 464
 465         gd = alloc_disk(nr_minors);
 466         if (gd == NULL)
 467                 goto release;
 468
 469         offset = minor / nr_parts;
 470
 471         if (nr_minors > 1) {
 472                 if (offset < 26)
 473                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 474                 else
 475                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 476                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 477         } else {
 478                 if (offset < 26)
 479                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 480                                 'a' + offset,
 481                                 minor & (nr_parts - 1));
 482                 else
 483                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 484                                 'a' + ((offset / 26) - 1),
 485                                 'a' + (offset % 26),
 486                                 minor & (nr_parts - 1));
 487         }
 488
 489         gd->major = XENVBD_MAJOR;
 490         gd->first_minor = minor;
 491         gd->fops = &xlvbd_block_fops;
 492         gd->private_data = info;
 493         gd->driverfs_dev = &(info->xbdev->dev);
 494         set_capacity(gd, capacity);
 495
 496         if (xlvbd_init_blk_queue(gd, sector_size)) {
 497                 del_gendisk(gd);
 498                 goto release;
 499         }
 500
 501         info->rq = gd->queue;
 502         info->gd = gd;
 503
 504         xlvbd_flush(info);
 505
 506         if (vdisk_info & VDISK_READONLY)
 507                 set_disk_ro(gd, 1);
 508
 509         if (vdisk_info & VDISK_REMOVABLE)
 510                 gd->flags |= GENHD_FL_REMOVABLE;
 511
 512         if (vdisk_info & VDISK_CDROM)
 513                 gd->flags |= GENHD_FL_CD;
 514
 515         return 0;
 516
 517  release:
 518         xlbd_release_minors(minor, nr_minors);
 519  out:
 520         return err;
 521 }
 522
 523 static void xlvbd_release_gendisk(struct blkfront_info *info)
 524 {
 525         unsigned int minor, nr_minors;
 526         unsigned long flags;
 527
 528         if (info->rq == NULL)
 529                 return;
 530
 531         spin_lock_irqsave(&blkif_io_lock, flags);
 532
 533         /* No more blkif_request(). */
 534         blk_stop_queue(info->rq);
 535
 536         /* No more gnttab callback work. */
 537         gnttab_cancel_free_callback(&info->callback);
 538         spin_unlock_irqrestore(&blkif_io_lock, flags);
 539
 540         /* Flush gnttab callback work. Must be done with no locks held. */
 541         flush_scheduled_work();
 542
 543         del_gendisk(info->gd);
 544
 545         minor = info->gd->first_minor;
 546         nr_minors = info->gd->minors;
 547         xlbd_release_minors(minor, nr_minors);
 548
 549         blk_cleanup_queue(info->rq);
 550         info->rq = NULL;
 551
 552         put_disk(info->gd);
 553         info->gd = NULL;
 554 }
 555
 556 static void kick_pending_request_queues(struct blkfront_info *info)
 557 {
 558         if (!RING_FULL(&info->ring)) {
 559                 /* Re-enable calldowns. */
 560                 blk_start_queue(info->rq);
 561                 /* Kick things off immediately. */
 562                 do_blkif_request(info->rq);
 563         }
 564 }
 565
 566 static void blkif_restart_queue(struct work_struct *work)
 567 {
 568         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 569
 570         spin_lock_irq(&blkif_io_lock);
 571         if (info->connected == BLKIF_STATE_CONNECTED)
 572                 kick_pending_request_queues(info);
 573         spin_unlock_irq(&blkif_io_lock);
 574 }
 575
 576 static void blkif_free(struct blkfront_info *info, int suspend)
 577 {
 578         /* Prevent new requests being issued until we fix things up. */
 579         spin_lock_irq(&blkif_io_lock);
 580         info->connected = suspend ?
 581                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 582         /* No more blkif_request(). */
 583         if (info->rq)
 584                 blk_stop_queue(info->rq);
 585         /* No more gnttab callback work. */
 586         gnttab_cancel_free_callback(&info->callback);
 587         spin_unlock_irq(&blkif_io_lock);
 588
 589         /* Flush gnttab callback work. Must be done with no locks held. */
 590         flush_scheduled_work();
 591
 592         /* Free resources associated with old device channel. */
 593         if (info->ring_ref != GRANT_INVALID_REF) {
 594                 gnttab_end_foreign_access(info->ring_ref, 0,
 595                                           (unsigned long)info->ring.sring);
 596                 info->ring_ref = GRANT_INVALID_REF;
 597                 info->ring.sring = NULL;
 598         }
 599         if (info->irq)
 600                 unbind_from_irqhandler(info->irq, info);
 601         info->evtchn = info->irq = 0;
 602
 603 }
 604
 605 static void blkif_completion(struct blk_shadow *s)
 606 {
 607         int i;
 608         for (i = 0; i < s->req.nr_segments; i++)
 609                 gnttab_end_foreign_access(s->req.seg[i].gref, 0, 0UL);
 610 }
 611
 612 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 613 {
 614         struct request *req;
 615         struct blkif_response *bret;
 616         RING_IDX i, rp;
 617         unsigned long flags;
 618         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 619         int error;
 620
 621         spin_lock_irqsave(&blkif_io_lock, flags);
 622
 623         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 624                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 625                 return IRQ_HANDLED;
 626         }
 627
 628  again:
 629         rp = info->ring.sring->rsp_prod;
 630         rmb(); /* Ensure we see queued responses up to 'rp'. */
 631
 632         for (i = info->ring.rsp_cons; i != rp; i++) {
 633                 unsigned long id;
 634
 635                 bret = RING_GET_RESPONSE(&info->ring, i);
 636                 id   = bret->id;
 637                 req  = (struct request *)info->shadow[id].request;
 638
 639                 blkif_completion(&info->shadow[id]);
 640
 641                 add_id_to_freelist(info, id);
 642
 643                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 644                 switch (bret->operation) {
 645                 case BLKIF_OP_WRITE_BARRIER:
 646                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 647                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 648                                        info->gd->disk_name);
 649                                 error = -EOPNOTSUPP;
 650                                 info->feature_flush = 0;
 651                                 xlvbd_flush(info);
 652                         }
 653                         /* fall through */
 654                 case BLKIF_OP_READ:
 655                 case BLKIF_OP_WRITE:
 656                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 657                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 658                                         "request: %x\n", bret->status);
 659
 660                         __blk_end_request_all(req, error);
 661                         break;
 662                 default:
 663                         BUG();
 664                 }
 665         }
 666
 667         info->ring.rsp_cons = i;
 668
 669         if (i != info->ring.req_prod_pvt) {
 670                 int more_to_do;
 671                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 672                 if (more_to_do)
 673                         goto again;
 674         } else
 675                 info->ring.sring->rsp_event = i + 1;
 676
 677         kick_pending_request_queues(info);
 678
 679         spin_unlock_irqrestore(&blkif_io_lock, flags);
 680
 681         return IRQ_HANDLED;
 682 }
 683
 684
 685 static int setup_blkring(struct xenbus_device *dev,
 686                          struct blkfront_info *info)
 687 {
 688         struct blkif_sring *sring;
 689         int err;
 690
 691         info->ring_ref = GRANT_INVALID_REF;
 692
 693         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 694         if (!sring) {
 695                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 696                 return -ENOMEM;
 697         }
 698         SHARED_RING_INIT(sring);
 699         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 700
 701         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 702
 703         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 704         if (err < 0) {
 705                 free_page((unsigned long)sring);
 706                 info->ring.sring = NULL;
 707                 goto fail;
 708         }
 709         info->ring_ref = err;
 710
 711         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 712         if (err)
 713                 goto fail;
 714
 715         err = bind_evtchn_to_irqhandler(info->evtchn,
 716                                         blkif_interrupt,
 717                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 718         if (err <= 0) {
 719                 xenbus_dev_fatal(dev, err,
 720                                  "bind_evtchn_to_irqhandler failed");
 721                 goto fail;
 722         }
 723         info->irq = err;
 724
 725         return 0;
 726 fail:
 727         blkif_free(info, 0);
 728         return err;
 729 }
 730
 731
 732 /* Common code used when first setting up, and when resuming. */
 733 static int talk_to_blkback(struct xenbus_device *dev,
 734                            struct blkfront_info *info)
 735 {
 736         const char *message = NULL;
 737         struct xenbus_transaction xbt;
 738         int err;
 739
 740         /* Create shared ring, alloc event channel. */
 741         err = setup_blkring(dev, info);
 742         if (err)
 743                 goto out;
 744
 745 again:
 746         err = xenbus_transaction_start(&xbt);
 747         if (err) {
 748                 xenbus_dev_fatal(dev, err, "starting transaction");
 749                 goto destroy_blkring;
 750         }
 751
 752         err = xenbus_printf(xbt, dev->nodename,
 753                             "ring-ref", "%u", info->ring_ref);
 754         if (err) {
 755                 message = "writing ring-ref";
 756                 goto abort_transaction;
 757         }
 758         err = xenbus_printf(xbt, dev->nodename,
 759                             "event-channel", "%u", info->evtchn);
 760         if (err) {
 761                 message = "writing event-channel";
 762                 goto abort_transaction;
 763         }
 764         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 765                             XEN_IO_PROTO_ABI_NATIVE);
 766         if (err) {
 767                 message = "writing protocol";
 768                 goto abort_transaction;
 769         }
 770
 771         err = xenbus_transaction_end(xbt, 0);
 772         if (err) {
 773                 if (err == -EAGAIN)
 774                         goto again;
 775                 xenbus_dev_fatal(dev, err, "completing transaction");
 776                 goto destroy_blkring;
 777         }
 778
 779         xenbus_switch_state(dev, XenbusStateInitialised);
 780
 781         return 0;
 782
 783  abort_transaction:
 784         xenbus_transaction_end(xbt, 1);
 785         if (message)
 786                 xenbus_dev_fatal(dev, err, "%s", message);
 787  destroy_blkring:
 788         blkif_free(info, 0);
 789  out:
 790         return err;
 791 }
 792
 793 /**
 794  * Entry point to this code when a new device is created.  Allocate the basic
 795  * structures and the ring buffer for communication with the backend, and
 796  * inform the backend of the appropriate details for those.  Switch to
 797  * Initialised state.
 798  */
 799 static int blkfront_probe(struct xenbus_device *dev,
 800                           const struct xenbus_device_id *id)
 801 {
 802         int err, vdevice, i;
 803         struct blkfront_info *info;
 804
 805         /* FIXME: Use dynamic device id if this is not set. */
 806         err = xenbus_scanf(XBT_NIL, dev->nodename,
 807                            "virtual-device", "%i", &vdevice);
 808         if (err != 1) {
 809                 /* go looking in the extended area instead */
 810                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 811                                    "%i", &vdevice);
 812                 if (err != 1) {
 813                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 814                         return err;
 815                 }
 816         }
 817
 818         if (xen_hvm_domain()) {
 819                 char *type;
 820                 int len;
 821                 /* no unplug has been done: do not hook devices != xen vbds */
 822                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
 823                         int major;
 824
 825                         if (!VDEV_IS_EXTENDED(vdevice))
 826                                 major = BLKIF_MAJOR(vdevice);
 827                         else
 828                                 major = XENVBD_MAJOR;
 829
 830                         if (major != XENVBD_MAJOR) {
 831                                 printk(KERN_INFO
 832                                                 "%s: HVM does not support vbd %d as xen block device\n",
 833                                                 __FUNCTION__, vdevice);
 834                                 return -ENODEV;
 835                         }
 836                 }
 837                 /* do not create a PV cdrom device if we are an HVM guest */
 838                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 839                 if (IS_ERR(type))
 840                         return -ENODEV;
 841                 if (strncmp(type, "cdrom", 5) == 0) {
 842                         kfree(type);
 843                         return -ENODEV;
 844                 }
 845                 kfree(type);
 846         }
 847         info = kzalloc(sizeof(*info), GFP_KERNEL);
 848         if (!info) {
 849                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 850                 return -ENOMEM;
 851         }
 852
 853         mutex_init(&info->mutex);
 854         info->xbdev = dev;
 855         info->vdevice = vdevice;
 856         info->connected = BLKIF_STATE_DISCONNECTED;
 857         INIT_WORK(&info->work, blkif_restart_queue);
 858
 859         for (i = 0; i < BLK_RING_SIZE; i++)
 860                 info->shadow[i].req.id = i+1;
 861         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 862
 863         /* Front end dir is a number, which is used as the id. */
 864         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 865         dev_set_drvdata(&dev->dev, info);
 866
 867         err = talk_to_blkback(dev, info);
 868         if (err) {
 869                 kfree(info);
 870                 dev_set_drvdata(&dev->dev, NULL);
 871                 return err;
 872         }
 873
 874         return 0;
 875 }
 876
 877
 878 static int blkif_recover(struct blkfront_info *info)
 879 {
 880         int i;
 881         struct blkif_request *req;
 882         struct blk_shadow *copy;
 883         int j;
 884
 885         /* Stage 1: Make a safe copy of the shadow state. */
 886         copy = kmalloc(sizeof(info->shadow),
 887                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 888         if (!copy)
 889                 return -ENOMEM;
 890         memcpy(copy, info->shadow, sizeof(info->shadow));
 891
 892         /* Stage 2: Set up free list. */
 893         memset(&info->shadow, 0, sizeof(info->shadow));
 894         for (i = 0; i < BLK_RING_SIZE; i++)
 895                 info->shadow[i].req.id = i+1;
 896         info->shadow_free = info->ring.req_prod_pvt;
 897         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 898
 899         /* Stage 3: Find pending requests and requeue them. */
 900         for (i = 0; i < BLK_RING_SIZE; i++) {
 901                 /* Not in use? */
 902                 if (copy[i].request == 0)
 903                         continue;
 904
 905                 /* Grab a request slot and copy shadow state into it. */
 906                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 907                 *req = copy[i].req;
 908
 909                 /* We get a new request id, and must reset the shadow state. */
 910                 req->id = get_id_from_freelist(info);
 911                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 912
 913                 /* Rewrite any grant references invalidated by susp/resume. */
 914                 for (j = 0; j < req->nr_segments; j++)
 915                         gnttab_grant_foreign_access_ref(
 916                                 req->seg[j].gref,
 917                                 info->xbdev->otherend_id,
 918                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 919                                 rq_data_dir(
 920                                         (struct request *)
 921                                         info->shadow[req->id].request));
 922                 info->shadow[req->id].req = *req;
 923
 924                 info->ring.req_prod_pvt++;
 925         }
 926
 927         kfree(copy);
 928
 929         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 930
 931         spin_lock_irq(&blkif_io_lock);
 932
 933         /* Now safe for us to use the shared ring */
 934         info->connected = BLKIF_STATE_CONNECTED;
 935
 936         /* Send off requeued requests */
 937         flush_requests(info);
 938
 939         /* Kick any other new requests queued since we resumed */
 940         kick_pending_request_queues(info);
 941
 942         spin_unlock_irq(&blkif_io_lock);
 943
 944         return 0;
 945 }
 946
 947 /**
 948  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 949  * driver restart.  We tear down our blkif structure and recreate it, but
 950  * leave the device-layer structures intact so that this is transparent to the
 951  * rest of the kernel.
 952  */
 953 static int blkfront_resume(struct xenbus_device *dev)
 954 {
 955         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 956         int err;
 957
 958         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 959
 960         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 961
 962         err = talk_to_blkback(dev, info);
 963         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 964                 err = blkif_recover(info);
 965
 966         return err;
 967 }
 968
 969 static void
 970 blkfront_closing(struct blkfront_info *info)
 971 {
 972         struct xenbus_device *xbdev = info->xbdev;
 973         struct block_device *bdev = NULL;
 974
 975         mutex_lock(&info->mutex);
 976
 977         if (xbdev->state == XenbusStateClosing) {
 978                 mutex_unlock(&info->mutex);
 979                 return;
 980         }
 981
 982         if (info->gd)
 983                 bdev = bdget_disk(info->gd, 0);
 984
 985         mutex_unlock(&info->mutex);
 986
 987         if (!bdev) {
 988                 xenbus_frontend_closed(xbdev);
 989                 return;
 990         }
 991
 992         mutex_lock(&bdev->bd_mutex);
 993
 994         if (bdev->bd_openers) {
 995                 xenbus_dev_error(xbdev, -EBUSY,
 996                                  "Device in use; refusing to close");
 997                 xenbus_switch_state(xbdev, XenbusStateClosing);
 998         } else {
 999                 xlvbd_release_gendisk(info);
1000                 xenbus_frontend_closed(xbdev);
1001         }
1002
1003         mutex_unlock(&bdev->bd_mutex);
1004         bdput(bdev);
1005 }
1006
1007 /*
1008  * Invoked when the backend is finally 'ready' (and has told produced
1009  * the details about the physical device - #sectors, size, etc).
1010  */
1011 static void blkfront_connect(struct blkfront_info *info)
1012 {
1013         unsigned long long sectors;
1014         unsigned long sector_size;
1015         unsigned int binfo;
1016         int err;
1017         int barrier;
1018
1019         switch (info->connected) {
1020         case BLKIF_STATE_CONNECTED:
1021                 /*
1022                  * Potentially, the back-end may be signalling
1023                  * a capacity change; update the capacity.
1024                  */
1025                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1026                                    "sectors", "%Lu", &sectors);
1027                 if (XENBUS_EXIST_ERR(err))
1028                         return;
1029                 printk(KERN_INFO "Setting capacity to %Lu\n",
1030                        sectors);
1031                 set_capacity(info->gd, sectors);
1032                 revalidate_disk(info->gd);
1033
1034                 /* fall through */
1035         case BLKIF_STATE_SUSPENDED:
1036                 return;
1037
1038         default:
1039                 break;
1040         }
1041
1042         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1043                 __func__, info->xbdev->otherend);
1044
1045         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1046                             "sectors", "%llu", &sectors,
1047                             "info", "%u", &binfo,
1048                             "sector-size", "%lu", &sector_size,
1049                             NULL);
1050         if (err) {
1051                 xenbus_dev_fatal(info->xbdev, err,
1052                                  "reading backend fields at %s",
1053                                  info->xbdev->otherend);
1054                 return;
1055         }
1056
1057         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1058                             "feature-barrier", "%lu", &barrier,
1059                             NULL);
1060
1061         /*
1062          * If there's no "feature-barrier" defined, then it means
1063          * we're dealing with a very old backend which writes
1064          * synchronously; nothing to do.
1065          *
1066          * If there are barriers, then we use flush.
1067          */
1068         info->feature_flush = 0;
1069
1070         /*
1071          * The driver doesn't properly handled empty flushes, so
1072          * lets disable barrier support for now.
1073          */
1074 #if 0
1075         if (!err && barrier)
1076                 info->feature_flush = REQ_FLUSH;
1077 #endif
1078
1079         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1080         if (err) {
1081                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1082                                  info->xbdev->otherend);
1083                 return;
1084         }
1085
1086         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1087
1088         /* Kick pending requests. */
1089         spin_lock_irq(&blkif_io_lock);
1090         info->connected = BLKIF_STATE_CONNECTED;
1091         kick_pending_request_queues(info);
1092         spin_unlock_irq(&blkif_io_lock);
1093
1094         add_disk(info->gd);
1095
1096         info->is_ready = 1;
1097 }
1098
1099 /**
1100  * Callback received when the backend's state changes.
1101  */
1102 static void blkback_changed(struct xenbus_device *dev,
1103                             enum xenbus_state backend_state)
1104 {
1105         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1106
1107         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1108
1109         switch (backend_state) {
1110         case XenbusStateInitialising:
1111         case XenbusStateInitWait:
1112         case XenbusStateInitialised:
1113         case XenbusStateReconfiguring:
1114         case XenbusStateReconfigured:
1115         case XenbusStateUnknown:
1116         case XenbusStateClosed:
1117                 break;
1118
1119         case XenbusStateConnected:
1120                 blkfront_connect(info);
1121                 break;
1122
1123         case XenbusStateClosing:
1124                 blkfront_closing(info);
1125                 break;
1126         }
1127 }
1128
1129 static int blkfront_remove(struct xenbus_device *xbdev)
1130 {
1131         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1132         struct block_device *bdev = NULL;
1133         struct gendisk *disk;
1134
1135         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1136
1137         blkif_free(info, 0);
1138
1139         mutex_lock(&info->mutex);
1140
1141         disk = info->gd;
1142         if (disk)
1143                 bdev = bdget_disk(disk, 0);
1144
1145         info->xbdev = NULL;
1146         mutex_unlock(&info->mutex);
1147
1148         if (!bdev) {
1149                 kfree(info);
1150                 return 0;
1151         }
1152
1153         /*
1154          * The xbdev was removed before we reached the Closed
1155          * state. See if it's safe to remove the disk. If the bdev
1156          * isn't closed yet, we let release take care of it.
1157          */
1158
1159         mutex_lock(&bdev->bd_mutex);
1160         info = disk->private_data;
1161
1162         dev_warn(disk_to_dev(disk),
1163                  "%s was hot-unplugged, %d stale handles\n",
1164                  xbdev->nodename, bdev->bd_openers);
1165
1166         if (info && !bdev->bd_openers) {
1167                 xlvbd_release_gendisk(info);
1168                 disk->private_data = NULL;
1169                 kfree(info);
1170         }
1171
1172         mutex_unlock(&bdev->bd_mutex);
1173         bdput(bdev);
1174
1175         return 0;
1176 }
1177
1178 static int blkfront_is_ready(struct xenbus_device *dev)
1179 {
1180         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1181
1182         return info->is_ready && info->xbdev;
1183 }
1184
1185 static int blkif_open(struct block_device *bdev, fmode_t mode)
1186 {
1187         struct gendisk *disk = bdev->bd_disk;
1188         struct blkfront_info *info;
1189         int err = 0;
1190
1191         mutex_lock(&blkfront_mutex);
1192
1193         info = disk->private_data;
1194         if (!info) {
1195                 /* xbdev gone */
1196                 err = -ERESTARTSYS;
1197                 goto out;
1198         }
1199
1200         mutex_lock(&info->mutex);
1201
1202         if (!info->gd)
1203                 /* xbdev is closed */
1204                 err = -ERESTARTSYS;
1205
1206         mutex_unlock(&info->mutex);
1207
1208 out:
1209         mutex_unlock(&blkfront_mutex);
1210         return err;
1211 }
1212
1213 static int blkif_release(struct gendisk *disk, fmode_t mode)
1214 {
1215         struct blkfront_info *info = disk->private_data;
1216         struct block_device *bdev;
1217         struct xenbus_device *xbdev;
1218
1219         mutex_lock(&blkfront_mutex);
1220
1221         bdev = bdget_disk(disk, 0);
1222         bdput(bdev);
1223
1224         if (bdev->bd_openers)
1225                 goto out;
1226
1227         /*
1228          * Check if we have been instructed to close. We will have
1229          * deferred this request, because the bdev was still open.
1230          */
1231
1232         mutex_lock(&info->mutex);
1233         xbdev = info->xbdev;
1234
1235         if (xbdev && xbdev->state == XenbusStateClosing) {
1236                 /* pending switch to state closed */
1237                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1238                 xlvbd_release_gendisk(info);
1239                 xenbus_frontend_closed(info->xbdev);
1240         }
1241
1242         mutex_unlock(&info->mutex);
1243
1244         if (!xbdev) {
1245                 /* sudden device removal */
1246                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1247                 xlvbd_release_gendisk(info);
1248                 disk->private_data = NULL;
1249                 kfree(info);
1250         }
1251
1252 out:
1253         mutex_unlock(&blkfront_mutex);
1254         return 0;
1255 }
1256
1257 static const struct block_device_operations xlvbd_block_fops =
1258 {
1259         .owner = THIS_MODULE,
1260         .open = blkif_open,
1261         .release = blkif_release,
1262         .getgeo = blkif_getgeo,
1263         .ioctl = blkif_ioctl,
1264 };
1265
1266
1267 static const struct xenbus_device_id blkfront_ids[] = {
1268         { "vbd" },
1269         { "" }
1270 };
1271
1272 static struct xenbus_driver blkfront = {
1273         .name = "vbd",
1274         .owner = THIS_MODULE,
1275         .ids = blkfront_ids,
1276         .probe = blkfront_probe,
1277         .remove = blkfront_remove,
1278         .resume = blkfront_resume,
1279         .otherend_changed = blkback_changed,
1280         .is_ready = blkfront_is_ready,
1281 };
1282
1283 static int __init xlblk_init(void)
1284 {
1285         if (!xen_domain())
1286                 return -ENODEV;
1287
1288         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1289                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1290                        XENVBD_MAJOR, DEV_NAME);
1291                 return -ENODEV;
1292         }
1293
1294         return xenbus_register_frontend(&blkfront);
1295 }
1296 module_init(xlblk_init);
1297
1298
1299 static void __exit xlblk_exit(void)
1300 {
1301         return xenbus_unregister_driver(&blkfront);
1302 }
1303 module_exit(xlblk_exit);
1304
1305 MODULE_DESCRIPTION("Xen virtual block device frontend");
1306 MODULE_LICENSE("GPL");
1307 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1308 MODULE_ALIAS("xen:vbd");
1309 MODULE_ALIAS("xenblk");