drivers/block/xen-blkfront.c

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