drivers/block/xen-blkfront.c

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