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         struct request *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 __CONST_RING_SIZE(blkif, 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 = NULL;
 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  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 249  * and writes are handled as expected.  Since we lack a loose flush
 250  * request, we map flushes into a full ordered barrier.
 251  *
 252  * @req: a request struct
 253  */
 254 static int blkif_queue_request(struct request *req)
 255 {
 256         struct blkfront_info *info = req->rq_disk->private_data;
 257         unsigned long buffer_mfn;
 258         struct blkif_request *ring_req;
 259         unsigned long id;
 260         unsigned int fsect, lsect;
 261         int i, ref;
 262         grant_ref_t gref_head;
 263         struct scatterlist *sg;
 264
 265         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 266                 return 1;
 267
 268         if (gnttab_alloc_grant_references(
 269                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 270                 gnttab_request_free_callback(
 271                         &info->callback,
 272                         blkif_restart_queue_callback,
 273                         info,
 274                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 275                 return 1;
 276         }
 277
 278         /* Fill out a communications ring structure. */
 279         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 280         id = get_id_from_freelist(info);
 281         info->shadow[id].request = req;
 282
 283         ring_req->id = id;
 284         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 285         ring_req->handle = info->handle;
 286
 287         ring_req->operation = rq_data_dir(req) ?
 288                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 289
 290         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 291                 /*
 292                  * Ideally we could just do an unordered
 293                  * flush-to-disk, but all we have is a full write
 294                  * barrier at the moment.  However, a barrier write is
 295                  * a superset of FUA, so we can implement it the same
 296                  * way.  (It's also a FLUSH+FUA, since it is
 297                  * guaranteed ordered WRT previous writes.)
 298                  */
 299                 ring_req->operation = BLKIF_OP_WRITE_BARRIER;
 300         }
 301
 302         ring_req->nr_segments = blk_rq_map_sg(req->q, req, info->sg);
 303         BUG_ON(ring_req->nr_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 304
 305         for_each_sg(info->sg, sg, ring_req->nr_segments, i) {
 306                 buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 307                 fsect = sg->offset >> 9;
 308                 lsect = fsect + (sg->length >> 9) - 1;
 309                 /* install a grant reference. */
 310                 ref = gnttab_claim_grant_reference(&gref_head);
 311                 BUG_ON(ref == -ENOSPC);
 312
 313                 gnttab_grant_foreign_access_ref(
 314                                 ref,
 315                                 info->xbdev->otherend_id,
 316                                 buffer_mfn,
 317                                 rq_data_dir(req) );
 318
 319                 info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 320                 ring_req->u.rw.seg[i] =
 321                                 (struct blkif_request_segment) {
 322                                         .gref       = ref,
 323                                         .first_sect = fsect,
 324                                         .last_sect  = lsect };
 325         }
 326
 327         info->ring.req_prod_pvt++;
 328
 329         /* Keep a private copy so we can reissue requests when recovering. */
 330         info->shadow[id].req = *ring_req;
 331
 332         gnttab_free_grant_references(gref_head);
 333
 334         return 0;
 335 }
 336
 337
 338 static inline void flush_requests(struct blkfront_info *info)
 339 {
 340         int notify;
 341
 342         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 343
 344         if (notify)
 345                 notify_remote_via_irq(info->irq);
 346 }
 347
 348 /*
 349  * do_blkif_request
 350  *  read a block; request is in a request queue
 351  */
 352 static void do_blkif_request(struct request_queue *rq)
 353 {
 354         struct blkfront_info *info = NULL;
 355         struct request *req;
 356         int queued;
 357
 358         pr_debug("Entered do_blkif_request\n");
 359
 360         queued = 0;
 361
 362         while ((req = blk_peek_request(rq)) != NULL) {
 363                 info = req->rq_disk->private_data;
 364
 365                 if (RING_FULL(&info->ring))
 366                         goto wait;
 367
 368                 blk_start_request(req);
 369
 370                 if (req->cmd_type != REQ_TYPE_FS) {
 371                         __blk_end_request_all(req, -EIO);
 372                         continue;
 373                 }
 374
 375                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 376                          "(%u/%u) buffer:%p [%s]\n",
 377                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 378                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 379                          req->buffer, rq_data_dir(req) ? "write" : "read");
 380
 381                 if (blkif_queue_request(req)) {
 382                         blk_requeue_request(rq, req);
 383 wait:
 384                         /* Avoid pointless unplugs. */
 385                         blk_stop_queue(rq);
 386                         break;
 387                 }
 388
 389                 queued++;
 390         }
 391
 392         if (queued != 0)
 393                 flush_requests(info);
 394 }
 395
 396 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 397 {
 398         struct request_queue *rq;
 399
 400         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 401         if (rq == NULL)
 402                 return -1;
 403
 404         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 405
 406         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 407         blk_queue_logical_block_size(rq, sector_size);
 408         blk_queue_max_hw_sectors(rq, 512);
 409
 410         /* Each segment in a request is up to an aligned page in size. */
 411         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 412         blk_queue_max_segment_size(rq, PAGE_SIZE);
 413
 414         /* Ensure a merged request will fit in a single I/O ring slot. */
 415         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 416
 417         /* Make sure buffer addresses are sector-aligned. */
 418         blk_queue_dma_alignment(rq, 511);
 419
 420         /* Make sure we don't use bounce buffers. */
 421         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 422
 423         gd->queue = rq;
 424
 425         return 0;
 426 }
 427
 428
 429 static void xlvbd_flush(struct blkfront_info *info)
 430 {
 431         blk_queue_flush(info->rq, info->feature_flush);
 432         printk(KERN_INFO "blkfront: %s: barriers %s\n",
 433                info->gd->disk_name,
 434                info->feature_flush ? "enabled" : "disabled");
 435 }
 436
 437
 438 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 439                                struct blkfront_info *info,
 440                                u16 vdisk_info, u16 sector_size)
 441 {
 442         struct gendisk *gd;
 443         int nr_minors = 1;
 444         int err = -ENODEV;
 445         unsigned int offset;
 446         int minor;
 447         int nr_parts;
 448
 449         BUG_ON(info->gd != NULL);
 450         BUG_ON(info->rq != NULL);
 451
 452         if ((info->vdevice>>EXT_SHIFT) > 1) {
 453                 /* this is above the extended range; something is wrong */
 454                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 455                 return -ENODEV;
 456         }
 457
 458         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 459                 minor = BLKIF_MINOR(info->vdevice);
 460                 nr_parts = PARTS_PER_DISK;
 461         } else {
 462                 minor = BLKIF_MINOR_EXT(info->vdevice);
 463                 nr_parts = PARTS_PER_EXT_DISK;
 464         }
 465
 466         if ((minor % nr_parts) == 0)
 467                 nr_minors = nr_parts;
 468
 469         err = xlbd_reserve_minors(minor, nr_minors);
 470         if (err)
 471                 goto out;
 472         err = -ENODEV;
 473
 474         gd = alloc_disk(nr_minors);
 475         if (gd == NULL)
 476                 goto release;
 477
 478         offset = minor / nr_parts;
 479
 480         if (nr_minors > 1) {
 481                 if (offset < 26)
 482                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 483                 else
 484                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 485                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 486         } else {
 487                 if (offset < 26)
 488                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 489                                 'a' + offset,
 490                                 minor & (nr_parts - 1));
 491                 else
 492                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 493                                 'a' + ((offset / 26) - 1),
 494                                 'a' + (offset % 26),
 495                                 minor & (nr_parts - 1));
 496         }
 497
 498         gd->major = XENVBD_MAJOR;
 499         gd->first_minor = minor;
 500         gd->fops = &xlvbd_block_fops;
 501         gd->private_data = info;
 502         gd->driverfs_dev = &(info->xbdev->dev);
 503         set_capacity(gd, capacity);
 504
 505         if (xlvbd_init_blk_queue(gd, sector_size)) {
 506                 del_gendisk(gd);
 507                 goto release;
 508         }
 509
 510         info->rq = gd->queue;
 511         info->gd = gd;
 512
 513         xlvbd_flush(info);
 514
 515         if (vdisk_info & VDISK_READONLY)
 516                 set_disk_ro(gd, 1);
 517
 518         if (vdisk_info & VDISK_REMOVABLE)
 519                 gd->flags |= GENHD_FL_REMOVABLE;
 520
 521         if (vdisk_info & VDISK_CDROM)
 522                 gd->flags |= GENHD_FL_CD;
 523
 524         return 0;
 525
 526  release:
 527         xlbd_release_minors(minor, nr_minors);
 528  out:
 529         return err;
 530 }
 531
 532 static void xlvbd_release_gendisk(struct blkfront_info *info)
 533 {
 534         unsigned int minor, nr_minors;
 535         unsigned long flags;
 536
 537         if (info->rq == NULL)
 538                 return;
 539
 540         spin_lock_irqsave(&blkif_io_lock, flags);
 541
 542         /* No more blkif_request(). */
 543         blk_stop_queue(info->rq);
 544
 545         /* No more gnttab callback work. */
 546         gnttab_cancel_free_callback(&info->callback);
 547         spin_unlock_irqrestore(&blkif_io_lock, flags);
 548
 549         /* Flush gnttab callback work. Must be done with no locks held. */
 550         flush_work_sync(&info->work);
 551
 552         del_gendisk(info->gd);
 553
 554         minor = info->gd->first_minor;
 555         nr_minors = info->gd->minors;
 556         xlbd_release_minors(minor, nr_minors);
 557
 558         blk_cleanup_queue(info->rq);
 559         info->rq = NULL;
 560
 561         put_disk(info->gd);
 562         info->gd = NULL;
 563 }
 564
 565 static void kick_pending_request_queues(struct blkfront_info *info)
 566 {
 567         if (!RING_FULL(&info->ring)) {
 568                 /* Re-enable calldowns. */
 569                 blk_start_queue(info->rq);
 570                 /* Kick things off immediately. */
 571                 do_blkif_request(info->rq);
 572         }
 573 }
 574
 575 static void blkif_restart_queue(struct work_struct *work)
 576 {
 577         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 578
 579         spin_lock_irq(&blkif_io_lock);
 580         if (info->connected == BLKIF_STATE_CONNECTED)
 581                 kick_pending_request_queues(info);
 582         spin_unlock_irq(&blkif_io_lock);
 583 }
 584
 585 static void blkif_free(struct blkfront_info *info, int suspend)
 586 {
 587         /* Prevent new requests being issued until we fix things up. */
 588         spin_lock_irq(&blkif_io_lock);
 589         info->connected = suspend ?
 590                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 591         /* No more blkif_request(). */
 592         if (info->rq)
 593                 blk_stop_queue(info->rq);
 594         /* No more gnttab callback work. */
 595         gnttab_cancel_free_callback(&info->callback);
 596         spin_unlock_irq(&blkif_io_lock);
 597
 598         /* Flush gnttab callback work. Must be done with no locks held. */
 599         flush_work_sync(&info->work);
 600
 601         /* Free resources associated with old device channel. */
 602         if (info->ring_ref != GRANT_INVALID_REF) {
 603                 gnttab_end_foreign_access(info->ring_ref, 0,
 604                                           (unsigned long)info->ring.sring);
 605                 info->ring_ref = GRANT_INVALID_REF;
 606                 info->ring.sring = NULL;
 607         }
 608         if (info->irq)
 609                 unbind_from_irqhandler(info->irq, info);
 610         info->evtchn = info->irq = 0;
 611
 612 }
 613
 614 static void blkif_completion(struct blk_shadow *s)
 615 {
 616         int i;
 617         for (i = 0; i < s->req.nr_segments; i++)
 618                 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
 619 }
 620
 621 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 622 {
 623         struct request *req;
 624         struct blkif_response *bret;
 625         RING_IDX i, rp;
 626         unsigned long flags;
 627         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 628         int error;
 629
 630         spin_lock_irqsave(&blkif_io_lock, flags);
 631
 632         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 633                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 634                 return IRQ_HANDLED;
 635         }
 636
 637  again:
 638         rp = info->ring.sring->rsp_prod;
 639         rmb(); /* Ensure we see queued responses up to 'rp'. */
 640
 641         for (i = info->ring.rsp_cons; i != rp; i++) {
 642                 unsigned long id;
 643
 644                 bret = RING_GET_RESPONSE(&info->ring, i);
 645                 id   = bret->id;
 646                 req  = info->shadow[id].request;
 647
 648                 blkif_completion(&info->shadow[id]);
 649
 650                 add_id_to_freelist(info, id);
 651
 652                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 653                 switch (bret->operation) {
 654                 case BLKIF_OP_WRITE_BARRIER:
 655                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 656                                 printk(KERN_WARNING "blkfront: %s: write barrier op failed\n",
 657                                        info->gd->disk_name);
 658                                 error = -EOPNOTSUPP;
 659                         }
 660                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 661                                      info->shadow[id].req.nr_segments == 0)) {
 662                                 printk(KERN_WARNING "blkfront: %s: empty write barrier op failed\n",
 663                                        info->gd->disk_name);
 664                                 error = -EOPNOTSUPP;
 665                         }
 666                         if (unlikely(error)) {
 667                                 if (error == -EOPNOTSUPP)
 668                                         error = 0;
 669                                 info->feature_flush = 0;
 670                                 xlvbd_flush(info);
 671                         }
 672                         /* fall through */
 673                 case BLKIF_OP_READ:
 674                 case BLKIF_OP_WRITE:
 675                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 676                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 677                                         "request: %x\n", bret->status);
 678
 679                         __blk_end_request_all(req, error);
 680                         break;
 681                 default:
 682                         BUG();
 683                 }
 684         }
 685
 686         info->ring.rsp_cons = i;
 687
 688         if (i != info->ring.req_prod_pvt) {
 689                 int more_to_do;
 690                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 691                 if (more_to_do)
 692                         goto again;
 693         } else
 694                 info->ring.sring->rsp_event = i + 1;
 695
 696         kick_pending_request_queues(info);
 697
 698         spin_unlock_irqrestore(&blkif_io_lock, flags);
 699
 700         return IRQ_HANDLED;
 701 }
 702
 703
 704 static int setup_blkring(struct xenbus_device *dev,
 705                          struct blkfront_info *info)
 706 {
 707         struct blkif_sring *sring;
 708         int err;
 709
 710         info->ring_ref = GRANT_INVALID_REF;
 711
 712         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 713         if (!sring) {
 714                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 715                 return -ENOMEM;
 716         }
 717         SHARED_RING_INIT(sring);
 718         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 719
 720         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 721
 722         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 723         if (err < 0) {
 724                 free_page((unsigned long)sring);
 725                 info->ring.sring = NULL;
 726                 goto fail;
 727         }
 728         info->ring_ref = err;
 729
 730         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 731         if (err)
 732                 goto fail;
 733
 734         err = bind_evtchn_to_irqhandler(info->evtchn,
 735                                         blkif_interrupt,
 736                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 737         if (err <= 0) {
 738                 xenbus_dev_fatal(dev, err,
 739                                  "bind_evtchn_to_irqhandler failed");
 740                 goto fail;
 741         }
 742         info->irq = err;
 743
 744         return 0;
 745 fail:
 746         blkif_free(info, 0);
 747         return err;
 748 }
 749
 750
 751 /* Common code used when first setting up, and when resuming. */
 752 static int talk_to_blkback(struct xenbus_device *dev,
 753                            struct blkfront_info *info)
 754 {
 755         const char *message = NULL;
 756         struct xenbus_transaction xbt;
 757         int err;
 758
 759         /* Create shared ring, alloc event channel. */
 760         err = setup_blkring(dev, info);
 761         if (err)
 762                 goto out;
 763
 764 again:
 765         err = xenbus_transaction_start(&xbt);
 766         if (err) {
 767                 xenbus_dev_fatal(dev, err, "starting transaction");
 768                 goto destroy_blkring;
 769         }
 770
 771         err = xenbus_printf(xbt, dev->nodename,
 772                             "ring-ref", "%u", info->ring_ref);
 773         if (err) {
 774                 message = "writing ring-ref";
 775                 goto abort_transaction;
 776         }
 777         err = xenbus_printf(xbt, dev->nodename,
 778                             "event-channel", "%u", info->evtchn);
 779         if (err) {
 780                 message = "writing event-channel";
 781                 goto abort_transaction;
 782         }
 783         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 784                             XEN_IO_PROTO_ABI_NATIVE);
 785         if (err) {
 786                 message = "writing protocol";
 787                 goto abort_transaction;
 788         }
 789
 790         err = xenbus_transaction_end(xbt, 0);
 791         if (err) {
 792                 if (err == -EAGAIN)
 793                         goto again;
 794                 xenbus_dev_fatal(dev, err, "completing transaction");
 795                 goto destroy_blkring;
 796         }
 797
 798         xenbus_switch_state(dev, XenbusStateInitialised);
 799
 800         return 0;
 801
 802  abort_transaction:
 803         xenbus_transaction_end(xbt, 1);
 804         if (message)
 805                 xenbus_dev_fatal(dev, err, "%s", message);
 806  destroy_blkring:
 807         blkif_free(info, 0);
 808  out:
 809         return err;
 810 }
 811
 812 /**
 813  * Entry point to this code when a new device is created.  Allocate the basic
 814  * structures and the ring buffer for communication with the backend, and
 815  * inform the backend of the appropriate details for those.  Switch to
 816  * Initialised state.
 817  */
 818 static int blkfront_probe(struct xenbus_device *dev,
 819                           const struct xenbus_device_id *id)
 820 {
 821         int err, vdevice, i;
 822         struct blkfront_info *info;
 823
 824         /* FIXME: Use dynamic device id if this is not set. */
 825         err = xenbus_scanf(XBT_NIL, dev->nodename,
 826                            "virtual-device", "%i", &vdevice);
 827         if (err != 1) {
 828                 /* go looking in the extended area instead */
 829                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 830                                    "%i", &vdevice);
 831                 if (err != 1) {
 832                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 833                         return err;
 834                 }
 835         }
 836
 837         if (xen_hvm_domain()) {
 838                 char *type;
 839                 int len;
 840                 /* no unplug has been done: do not hook devices != xen vbds */
 841                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
 842                         int major;
 843
 844                         if (!VDEV_IS_EXTENDED(vdevice))
 845                                 major = BLKIF_MAJOR(vdevice);
 846                         else
 847                                 major = XENVBD_MAJOR;
 848
 849                         if (major != XENVBD_MAJOR) {
 850                                 printk(KERN_INFO
 851                                                 "%s: HVM does not support vbd %d as xen block device\n",
 852                                                 __FUNCTION__, vdevice);
 853                                 return -ENODEV;
 854                         }
 855                 }
 856                 /* do not create a PV cdrom device if we are an HVM guest */
 857                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 858                 if (IS_ERR(type))
 859                         return -ENODEV;
 860                 if (strncmp(type, "cdrom", 5) == 0) {
 861                         kfree(type);
 862                         return -ENODEV;
 863                 }
 864                 kfree(type);
 865         }
 866         info = kzalloc(sizeof(*info), GFP_KERNEL);
 867         if (!info) {
 868                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 869                 return -ENOMEM;
 870         }
 871
 872         mutex_init(&info->mutex);
 873         info->xbdev = dev;
 874         info->vdevice = vdevice;
 875         info->connected = BLKIF_STATE_DISCONNECTED;
 876         INIT_WORK(&info->work, blkif_restart_queue);
 877
 878         for (i = 0; i < BLK_RING_SIZE; i++)
 879                 info->shadow[i].req.id = i+1;
 880         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 881
 882         /* Front end dir is a number, which is used as the id. */
 883         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
 884         dev_set_drvdata(&dev->dev, info);
 885
 886         err = talk_to_blkback(dev, info);
 887         if (err) {
 888                 kfree(info);
 889                 dev_set_drvdata(&dev->dev, NULL);
 890                 return err;
 891         }
 892
 893         return 0;
 894 }
 895
 896
 897 static int blkif_recover(struct blkfront_info *info)
 898 {
 899         int i;
 900         struct blkif_request *req;
 901         struct blk_shadow *copy;
 902         int j;
 903
 904         /* Stage 1: Make a safe copy of the shadow state. */
 905         copy = kmalloc(sizeof(info->shadow),
 906                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
 907         if (!copy)
 908                 return -ENOMEM;
 909         memcpy(copy, info->shadow, sizeof(info->shadow));
 910
 911         /* Stage 2: Set up free list. */
 912         memset(&info->shadow, 0, sizeof(info->shadow));
 913         for (i = 0; i < BLK_RING_SIZE; i++)
 914                 info->shadow[i].req.id = i+1;
 915         info->shadow_free = info->ring.req_prod_pvt;
 916         info->shadow[BLK_RING_SIZE-1].req.id = 0x0fffffff;
 917
 918         /* Stage 3: Find pending requests and requeue them. */
 919         for (i = 0; i < BLK_RING_SIZE; i++) {
 920                 /* Not in use? */
 921                 if (!copy[i].request)
 922                         continue;
 923
 924                 /* Grab a request slot and copy shadow state into it. */
 925                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 926                 *req = copy[i].req;
 927
 928                 /* We get a new request id, and must reset the shadow state. */
 929                 req->id = get_id_from_freelist(info);
 930                 memcpy(&info->shadow[req->id], &copy[i], sizeof(copy[i]));
 931
 932                 /* Rewrite any grant references invalidated by susp/resume. */
 933                 for (j = 0; j < req->nr_segments; j++)
 934                         gnttab_grant_foreign_access_ref(
 935                                 req->u.rw.seg[j].gref,
 936                                 info->xbdev->otherend_id,
 937                                 pfn_to_mfn(info->shadow[req->id].frame[j]),
 938                                 rq_data_dir(info->shadow[req->id].request));
 939                 info->shadow[req->id].req = *req;
 940
 941                 info->ring.req_prod_pvt++;
 942         }
 943
 944         kfree(copy);
 945
 946         xenbus_switch_state(info->xbdev, XenbusStateConnected);
 947
 948         spin_lock_irq(&blkif_io_lock);
 949
 950         /* Now safe for us to use the shared ring */
 951         info->connected = BLKIF_STATE_CONNECTED;
 952
 953         /* Send off requeued requests */
 954         flush_requests(info);
 955
 956         /* Kick any other new requests queued since we resumed */
 957         kick_pending_request_queues(info);
 958
 959         spin_unlock_irq(&blkif_io_lock);
 960
 961         return 0;
 962 }
 963
 964 /**
 965  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 966  * driver restart.  We tear down our blkif structure and recreate it, but
 967  * leave the device-layer structures intact so that this is transparent to the
 968  * rest of the kernel.
 969  */
 970 static int blkfront_resume(struct xenbus_device *dev)
 971 {
 972         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
 973         int err;
 974
 975         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
 976
 977         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
 978
 979         err = talk_to_blkback(dev, info);
 980         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
 981                 err = blkif_recover(info);
 982
 983         return err;
 984 }
 985
 986 static void
 987 blkfront_closing(struct blkfront_info *info)
 988 {
 989         struct xenbus_device *xbdev = info->xbdev;
 990         struct block_device *bdev = NULL;
 991
 992         mutex_lock(&info->mutex);
 993
 994         if (xbdev->state == XenbusStateClosing) {
 995                 mutex_unlock(&info->mutex);
 996                 return;
 997         }
 998
 999         if (info->gd)
1000                 bdev = bdget_disk(info->gd, 0);
1001
1002         mutex_unlock(&info->mutex);
1003
1004         if (!bdev) {
1005                 xenbus_frontend_closed(xbdev);
1006                 return;
1007         }
1008
1009         mutex_lock(&bdev->bd_mutex);
1010
1011         if (bdev->bd_openers) {
1012                 xenbus_dev_error(xbdev, -EBUSY,
1013                                  "Device in use; refusing to close");
1014                 xenbus_switch_state(xbdev, XenbusStateClosing);
1015         } else {
1016                 xlvbd_release_gendisk(info);
1017                 xenbus_frontend_closed(xbdev);
1018         }
1019
1020         mutex_unlock(&bdev->bd_mutex);
1021         bdput(bdev);
1022 }
1023
1024 /*
1025  * Invoked when the backend is finally 'ready' (and has told produced
1026  * the details about the physical device - #sectors, size, etc).
1027  */
1028 static void blkfront_connect(struct blkfront_info *info)
1029 {
1030         unsigned long long sectors;
1031         unsigned long sector_size;
1032         unsigned int binfo;
1033         int err;
1034         int barrier;
1035
1036         switch (info->connected) {
1037         case BLKIF_STATE_CONNECTED:
1038                 /*
1039                  * Potentially, the back-end may be signalling
1040                  * a capacity change; update the capacity.
1041                  */
1042                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1043                                    "sectors", "%Lu", &sectors);
1044                 if (XENBUS_EXIST_ERR(err))
1045                         return;
1046                 printk(KERN_INFO "Setting capacity to %Lu\n",
1047                        sectors);
1048                 set_capacity(info->gd, sectors);
1049                 revalidate_disk(info->gd);
1050
1051                 /* fall through */
1052         case BLKIF_STATE_SUSPENDED:
1053                 return;
1054
1055         default:
1056                 break;
1057         }
1058
1059         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1060                 __func__, info->xbdev->otherend);
1061
1062         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1063                             "sectors", "%llu", &sectors,
1064                             "info", "%u", &binfo,
1065                             "sector-size", "%lu", &sector_size,
1066                             NULL);
1067         if (err) {
1068                 xenbus_dev_fatal(info->xbdev, err,
1069                                  "reading backend fields at %s",
1070                                  info->xbdev->otherend);
1071                 return;
1072         }
1073
1074         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1075                             "feature-barrier", "%lu", &barrier,
1076                             NULL);
1077
1078         /*
1079          * If there's no "feature-barrier" defined, then it means
1080          * we're dealing with a very old backend which writes
1081          * synchronously; nothing to do.
1082          *
1083          * If there are barriers, then we use flush.
1084          */
1085         info->feature_flush = 0;
1086
1087         if (!err && barrier)
1088                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1089
1090         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1091         if (err) {
1092                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1093                                  info->xbdev->otherend);
1094                 return;
1095         }
1096
1097         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1098
1099         /* Kick pending requests. */
1100         spin_lock_irq(&blkif_io_lock);
1101         info->connected = BLKIF_STATE_CONNECTED;
1102         kick_pending_request_queues(info);
1103         spin_unlock_irq(&blkif_io_lock);
1104
1105         add_disk(info->gd);
1106
1107         info->is_ready = 1;
1108 }
1109
1110 /**
1111  * Callback received when the backend's state changes.
1112  */
1113 static void blkback_changed(struct xenbus_device *dev,
1114                             enum xenbus_state backend_state)
1115 {
1116         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1117
1118         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1119
1120         switch (backend_state) {
1121         case XenbusStateInitialising:
1122         case XenbusStateInitWait:
1123         case XenbusStateInitialised:
1124         case XenbusStateReconfiguring:
1125         case XenbusStateReconfigured:
1126         case XenbusStateUnknown:
1127         case XenbusStateClosed:
1128                 break;
1129
1130         case XenbusStateConnected:
1131                 blkfront_connect(info);
1132                 break;
1133
1134         case XenbusStateClosing:
1135                 blkfront_closing(info);
1136                 break;
1137         }
1138 }
1139
1140 static int blkfront_remove(struct xenbus_device *xbdev)
1141 {
1142         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1143         struct block_device *bdev = NULL;
1144         struct gendisk *disk;
1145
1146         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1147
1148         blkif_free(info, 0);
1149
1150         mutex_lock(&info->mutex);
1151
1152         disk = info->gd;
1153         if (disk)
1154                 bdev = bdget_disk(disk, 0);
1155
1156         info->xbdev = NULL;
1157         mutex_unlock(&info->mutex);
1158
1159         if (!bdev) {
1160                 kfree(info);
1161                 return 0;
1162         }
1163
1164         /*
1165          * The xbdev was removed before we reached the Closed
1166          * state. See if it's safe to remove the disk. If the bdev
1167          * isn't closed yet, we let release take care of it.
1168          */
1169
1170         mutex_lock(&bdev->bd_mutex);
1171         info = disk->private_data;
1172
1173         dev_warn(disk_to_dev(disk),
1174                  "%s was hot-unplugged, %d stale handles\n",
1175                  xbdev->nodename, bdev->bd_openers);
1176
1177         if (info && !bdev->bd_openers) {
1178                 xlvbd_release_gendisk(info);
1179                 disk->private_data = NULL;
1180                 kfree(info);
1181         }
1182
1183         mutex_unlock(&bdev->bd_mutex);
1184         bdput(bdev);
1185
1186         return 0;
1187 }
1188
1189 static int blkfront_is_ready(struct xenbus_device *dev)
1190 {
1191         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1192
1193         return info->is_ready && info->xbdev;
1194 }
1195
1196 static int blkif_open(struct block_device *bdev, fmode_t mode)
1197 {
1198         struct gendisk *disk = bdev->bd_disk;
1199         struct blkfront_info *info;
1200         int err = 0;
1201
1202         mutex_lock(&blkfront_mutex);
1203
1204         info = disk->private_data;
1205         if (!info) {
1206                 /* xbdev gone */
1207                 err = -ERESTARTSYS;
1208                 goto out;
1209         }
1210
1211         mutex_lock(&info->mutex);
1212
1213         if (!info->gd)
1214                 /* xbdev is closed */
1215                 err = -ERESTARTSYS;
1216
1217         mutex_unlock(&info->mutex);
1218
1219 out:
1220         mutex_unlock(&blkfront_mutex);
1221         return err;
1222 }
1223
1224 static int blkif_release(struct gendisk *disk, fmode_t mode)
1225 {
1226         struct blkfront_info *info = disk->private_data;
1227         struct block_device *bdev;
1228         struct xenbus_device *xbdev;
1229
1230         mutex_lock(&blkfront_mutex);
1231
1232         bdev = bdget_disk(disk, 0);
1233         bdput(bdev);
1234
1235         if (bdev->bd_openers)
1236                 goto out;
1237
1238         /*
1239          * Check if we have been instructed to close. We will have
1240          * deferred this request, because the bdev was still open.
1241          */
1242
1243         mutex_lock(&info->mutex);
1244         xbdev = info->xbdev;
1245
1246         if (xbdev && xbdev->state == XenbusStateClosing) {
1247                 /* pending switch to state closed */
1248                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1249                 xlvbd_release_gendisk(info);
1250                 xenbus_frontend_closed(info->xbdev);
1251         }
1252
1253         mutex_unlock(&info->mutex);
1254
1255         if (!xbdev) {
1256                 /* sudden device removal */
1257                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1258                 xlvbd_release_gendisk(info);
1259                 disk->private_data = NULL;
1260                 kfree(info);
1261         }
1262
1263 out:
1264         mutex_unlock(&blkfront_mutex);
1265         return 0;
1266 }
1267
1268 static const struct block_device_operations xlvbd_block_fops =
1269 {
1270         .owner = THIS_MODULE,
1271         .open = blkif_open,
1272         .release = blkif_release,
1273         .getgeo = blkif_getgeo,
1274         .ioctl = blkif_ioctl,
1275 };
1276
1277
1278 static const struct xenbus_device_id blkfront_ids[] = {
1279         { "vbd" },
1280         { "" }
1281 };
1282
1283 static struct xenbus_driver blkfront = {
1284         .name = "vbd",
1285         .owner = THIS_MODULE,
1286         .ids = blkfront_ids,
1287         .probe = blkfront_probe,
1288         .remove = blkfront_remove,
1289         .resume = blkfront_resume,
1290         .otherend_changed = blkback_changed,
1291         .is_ready = blkfront_is_ready,
1292 };
1293
1294 static int __init xlblk_init(void)
1295 {
1296         if (!xen_domain())
1297                 return -ENODEV;
1298
1299         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1300                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1301                        XENVBD_MAJOR, DEV_NAME);
1302                 return -ENODEV;
1303         }
1304
1305         return xenbus_register_frontend(&blkfront);
1306 }
1307 module_init(xlblk_init);
1308
1309
1310 static void __exit xlblk_exit(void)
1311 {
1312         return xenbus_unregister_driver(&blkfront);
1313 }
1314 module_exit(xlblk_exit);
1315
1316 MODULE_DESCRIPTION("Xen virtual block device frontend");
1317 MODULE_LICENSE("GPL");
1318 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1319 MODULE_ALIAS("xen:vbd");
1320 MODULE_ALIAS("xenblk");