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         unsigned int flush_op;
 101         unsigned int feature_discard:1;
 102         unsigned int feature_secdiscard:1;
 103         unsigned int discard_granularity;
 104         unsigned int discard_alignment;
 105         int is_ready;
 106 };
 107
 108 static DEFINE_SPINLOCK(blkif_io_lock);
 109
 110 static unsigned int nr_minors;
 111 static unsigned long *minors;
 112 static DEFINE_SPINLOCK(minor_lock);
 113
 114 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 115         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 116 #define GRANT_INVALID_REF       0
 117
 118 #define PARTS_PER_DISK          16
 119 #define PARTS_PER_EXT_DISK      256
 120
 121 #define BLKIF_MAJOR(dev) ((dev)>>8)
 122 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 123
 124 #define EXT_SHIFT 28
 125 #define EXTENDED (1<<EXT_SHIFT)
 126 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 127 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 128 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 129 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 130 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 131 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 132
 133 #define DEV_NAME        "xvd"   /* name in /dev */
 134
 135 static int get_id_from_freelist(struct blkfront_info *info)
 136 {
 137         unsigned long free = info->shadow_free;
 138         BUG_ON(free >= BLK_RING_SIZE);
 139         info->shadow_free = info->shadow[free].req.u.rw.id;
 140         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 141         return free;
 142 }
 143
 144 static void add_id_to_freelist(struct blkfront_info *info,
 145                                unsigned long id)
 146 {
 147         info->shadow[id].req.u.rw.id  = info->shadow_free;
 148         info->shadow[id].request = NULL;
 149         info->shadow_free = id;
 150 }
 151
 152 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 153 {
 154         unsigned int end = minor + nr;
 155         int rc;
 156
 157         if (end > nr_minors) {
 158                 unsigned long *bitmap, *old;
 159
 160                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 161                                  GFP_KERNEL);
 162                 if (bitmap == NULL)
 163                         return -ENOMEM;
 164
 165                 spin_lock(&minor_lock);
 166                 if (end > nr_minors) {
 167                         old = minors;
 168                         memcpy(bitmap, minors,
 169                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 170                         minors = bitmap;
 171                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 172                 } else
 173                         old = bitmap;
 174                 spin_unlock(&minor_lock);
 175                 kfree(old);
 176         }
 177
 178         spin_lock(&minor_lock);
 179         if (find_next_bit(minors, end, minor) >= end) {
 180                 for (; minor < end; ++minor)
 181                         __set_bit(minor, minors);
 182                 rc = 0;
 183         } else
 184                 rc = -EBUSY;
 185         spin_unlock(&minor_lock);
 186
 187         return rc;
 188 }
 189
 190 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 191 {
 192         unsigned int end = minor + nr;
 193
 194         BUG_ON(end > nr_minors);
 195         spin_lock(&minor_lock);
 196         for (; minor < end; ++minor)
 197                 __clear_bit(minor, minors);
 198         spin_unlock(&minor_lock);
 199 }
 200
 201 static void blkif_restart_queue_callback(void *arg)
 202 {
 203         struct blkfront_info *info = (struct blkfront_info *)arg;
 204         schedule_work(&info->work);
 205 }
 206
 207 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 208 {
 209         /* We don't have real geometry info, but let's at least return
 210            values consistent with the size of the device */
 211         sector_t nsect = get_capacity(bd->bd_disk);
 212         sector_t cylinders = nsect;
 213
 214         hg->heads = 0xff;
 215         hg->sectors = 0x3f;
 216         sector_div(cylinders, hg->heads * hg->sectors);
 217         hg->cylinders = cylinders;
 218         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 219                 hg->cylinders = 0xffff;
 220         return 0;
 221 }
 222
 223 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 224                        unsigned command, unsigned long argument)
 225 {
 226         struct blkfront_info *info = bdev->bd_disk->private_data;
 227         int i;
 228
 229         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 230                 command, (long)argument);
 231
 232         switch (command) {
 233         case CDROMMULTISESSION:
 234                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 235                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 236                         if (put_user(0, (char __user *)(argument + i)))
 237                                 return -EFAULT;
 238                 return 0;
 239
 240         case CDROM_GET_CAPABILITY: {
 241                 struct gendisk *gd = info->gd;
 242                 if (gd->flags & GENHD_FL_CD)
 243                         return 0;
 244                 return -EINVAL;
 245         }
 246
 247         default:
 248                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 249                   command);*/
 250                 return -EINVAL; /* same return as native Linux */
 251         }
 252
 253         return 0;
 254 }
 255
 256 /*
 257  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 258  * and writes are handled as expected.
 259  *
 260  * @req: a request struct
 261  */
 262 static int blkif_queue_request(struct request *req)
 263 {
 264         struct blkfront_info *info = req->rq_disk->private_data;
 265         unsigned long buffer_mfn;
 266         struct blkif_request *ring_req;
 267         unsigned long id;
 268         unsigned int fsect, lsect;
 269         int i, ref;
 270         grant_ref_t gref_head;
 271         struct scatterlist *sg;
 272
 273         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 274                 return 1;
 275
 276         if (gnttab_alloc_grant_references(
 277                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 278                 gnttab_request_free_callback(
 279                         &info->callback,
 280                         blkif_restart_queue_callback,
 281                         info,
 282                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 283                 return 1;
 284         }
 285
 286         /* Fill out a communications ring structure. */
 287         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 288         id = get_id_from_freelist(info);
 289         info->shadow[id].request = req;
 290
 291         ring_req->u.rw.id = id;
 292         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 293         ring_req->u.rw.handle = info->handle;
 294
 295         ring_req->operation = rq_data_dir(req) ?
 296                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 297
 298         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 299                 /*
 300                  * Ideally we can do an unordered flush-to-disk. In case the
 301                  * backend onlysupports barriers, use that. A barrier request
 302                  * a superset of FUA, so we can implement it the same
 303                  * way.  (It's also a FLUSH+FUA, since it is
 304                  * guaranteed ordered WRT previous writes.)
 305                  */
 306                 ring_req->operation = info->flush_op;
 307         }
 308
 309         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 310                 /* id, sector_number and handle are set above. */
 311                 ring_req->operation = BLKIF_OP_DISCARD;
 312                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 313                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 314                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 315                 else
 316                         ring_req->u.discard.flag = 0;
 317         } else {
 318                 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
 319                                                            info->sg);
 320                 BUG_ON(ring_req->u.rw.nr_segments >
 321                        BLKIF_MAX_SEGMENTS_PER_REQUEST);
 322
 323                 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
 324                         buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 325                         fsect = sg->offset >> 9;
 326                         lsect = fsect + (sg->length >> 9) - 1;
 327                         /* install a grant reference. */
 328                         ref = gnttab_claim_grant_reference(&gref_head);
 329                         BUG_ON(ref == -ENOSPC);
 330
 331                         gnttab_grant_foreign_access_ref(
 332                                         ref,
 333                                         info->xbdev->otherend_id,
 334                                         buffer_mfn,
 335                                         rq_data_dir(req));
 336
 337                         info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 338                         ring_req->u.rw.seg[i] =
 339                                         (struct blkif_request_segment) {
 340                                                 .gref       = ref,
 341                                                 .first_sect = fsect,
 342                                                 .last_sect  = lsect };
 343                 }
 344         }
 345
 346         info->ring.req_prod_pvt++;
 347
 348         /* Keep a private copy so we can reissue requests when recovering. */
 349         info->shadow[id].req = *ring_req;
 350
 351         gnttab_free_grant_references(gref_head);
 352
 353         return 0;
 354 }
 355
 356
 357 static inline void flush_requests(struct blkfront_info *info)
 358 {
 359         int notify;
 360
 361         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 362
 363         if (notify)
 364                 notify_remote_via_irq(info->irq);
 365 }
 366
 367 /*
 368  * do_blkif_request
 369  *  read a block; request is in a request queue
 370  */
 371 static void do_blkif_request(struct request_queue *rq)
 372 {
 373         struct blkfront_info *info = NULL;
 374         struct request *req;
 375         int queued;
 376
 377         pr_debug("Entered do_blkif_request\n");
 378
 379         queued = 0;
 380
 381         while ((req = blk_peek_request(rq)) != NULL) {
 382                 info = req->rq_disk->private_data;
 383
 384                 if (RING_FULL(&info->ring))
 385                         goto wait;
 386
 387                 blk_start_request(req);
 388
 389                 if ((req->cmd_type != REQ_TYPE_FS) ||
 390                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 391                     !info->flush_op)) {
 392                         __blk_end_request_all(req, -EIO);
 393                         continue;
 394                 }
 395
 396                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 397                          "(%u/%u) buffer:%p [%s]\n",
 398                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 399                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 400                          req->buffer, rq_data_dir(req) ? "write" : "read");
 401
 402                 if (blkif_queue_request(req)) {
 403                         blk_requeue_request(rq, req);
 404 wait:
 405                         /* Avoid pointless unplugs. */
 406                         blk_stop_queue(rq);
 407                         break;
 408                 }
 409
 410                 queued++;
 411         }
 412
 413         if (queued != 0)
 414                 flush_requests(info);
 415 }
 416
 417 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 418 {
 419         struct request_queue *rq;
 420         struct blkfront_info *info = gd->private_data;
 421
 422         rq = blk_init_queue(do_blkif_request, &blkif_io_lock);
 423         if (rq == NULL)
 424                 return -1;
 425
 426         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 427
 428         if (info->feature_discard) {
 429                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 430                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 431                 rq->limits.discard_granularity = info->discard_granularity;
 432                 rq->limits.discard_alignment = info->discard_alignment;
 433                 if (info->feature_secdiscard)
 434                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 435         }
 436
 437         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 438         blk_queue_logical_block_size(rq, sector_size);
 439         blk_queue_max_hw_sectors(rq, 512);
 440
 441         /* Each segment in a request is up to an aligned page in size. */
 442         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 443         blk_queue_max_segment_size(rq, PAGE_SIZE);
 444
 445         /* Ensure a merged request will fit in a single I/O ring slot. */
 446         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 447
 448         /* Make sure buffer addresses are sector-aligned. */
 449         blk_queue_dma_alignment(rq, 511);
 450
 451         /* Make sure we don't use bounce buffers. */
 452         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 453
 454         gd->queue = rq;
 455
 456         return 0;
 457 }
 458
 459
 460 static void xlvbd_flush(struct blkfront_info *info)
 461 {
 462         blk_queue_flush(info->rq, info->feature_flush);
 463         printk(KERN_INFO "blkfront: %s: %s: %s\n",
 464                info->gd->disk_name,
 465                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 466                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 467                 "flush diskcache" : "barrier or flush"),
 468                info->feature_flush ? "enabled" : "disabled");
 469 }
 470
 471 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 472 {
 473         int major;
 474         major = BLKIF_MAJOR(vdevice);
 475         *minor = BLKIF_MINOR(vdevice);
 476         switch (major) {
 477                 case XEN_IDE0_MAJOR:
 478                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 479                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 480                                 EMULATED_HD_DISK_MINOR_OFFSET;
 481                         break;
 482                 case XEN_IDE1_MAJOR:
 483                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 484                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 485                                 EMULATED_HD_DISK_MINOR_OFFSET;
 486                         break;
 487                 case XEN_SCSI_DISK0_MAJOR:
 488                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 489                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 490                         break;
 491                 case XEN_SCSI_DISK1_MAJOR:
 492                 case XEN_SCSI_DISK2_MAJOR:
 493                 case XEN_SCSI_DISK3_MAJOR:
 494                 case XEN_SCSI_DISK4_MAJOR:
 495                 case XEN_SCSI_DISK5_MAJOR:
 496                 case XEN_SCSI_DISK6_MAJOR:
 497                 case XEN_SCSI_DISK7_MAJOR:
 498                         *offset = (*minor / PARTS_PER_DISK) +
 499                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 500                                 EMULATED_SD_DISK_NAME_OFFSET;
 501                         *minor = *minor +
 502                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 503                                 EMULATED_SD_DISK_MINOR_OFFSET;
 504                         break;
 505                 case XEN_SCSI_DISK8_MAJOR:
 506                 case XEN_SCSI_DISK9_MAJOR:
 507                 case XEN_SCSI_DISK10_MAJOR:
 508                 case XEN_SCSI_DISK11_MAJOR:
 509                 case XEN_SCSI_DISK12_MAJOR:
 510                 case XEN_SCSI_DISK13_MAJOR:
 511                 case XEN_SCSI_DISK14_MAJOR:
 512                 case XEN_SCSI_DISK15_MAJOR:
 513                         *offset = (*minor / PARTS_PER_DISK) +
 514                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 515                                 EMULATED_SD_DISK_NAME_OFFSET;
 516                         *minor = *minor +
 517                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 518                                 EMULATED_SD_DISK_MINOR_OFFSET;
 519                         break;
 520                 case XENVBD_MAJOR:
 521                         *offset = *minor / PARTS_PER_DISK;
 522                         break;
 523                 default:
 524                         printk(KERN_WARNING "blkfront: your disk configuration is "
 525                                         "incorrect, please use an xvd device instead\n");
 526                         return -ENODEV;
 527         }
 528         return 0;
 529 }
 530
 531 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 532                                struct blkfront_info *info,
 533                                u16 vdisk_info, u16 sector_size)
 534 {
 535         struct gendisk *gd;
 536         int nr_minors = 1;
 537         int err;
 538         unsigned int offset;
 539         int minor;
 540         int nr_parts;
 541
 542         BUG_ON(info->gd != NULL);
 543         BUG_ON(info->rq != NULL);
 544
 545         if ((info->vdevice>>EXT_SHIFT) > 1) {
 546                 /* this is above the extended range; something is wrong */
 547                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 548                 return -ENODEV;
 549         }
 550
 551         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 552                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 553                 if (err)
 554                         return err;
 555                 nr_parts = PARTS_PER_DISK;
 556         } else {
 557                 minor = BLKIF_MINOR_EXT(info->vdevice);
 558                 nr_parts = PARTS_PER_EXT_DISK;
 559                 offset = minor / nr_parts;
 560                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 561                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 562                                         "emulated IDE disks,\n\t choose an xvd device name"
 563                                         "from xvde on\n", info->vdevice);
 564         }
 565         err = -ENODEV;
 566
 567         if ((minor % nr_parts) == 0)
 568                 nr_minors = nr_parts;
 569
 570         err = xlbd_reserve_minors(minor, nr_minors);
 571         if (err)
 572                 goto out;
 573         err = -ENODEV;
 574
 575         gd = alloc_disk(nr_minors);
 576         if (gd == NULL)
 577                 goto release;
 578
 579         if (nr_minors > 1) {
 580                 if (offset < 26)
 581                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 582                 else
 583                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 584                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 585         } else {
 586                 if (offset < 26)
 587                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 588                                 'a' + offset,
 589                                 minor & (nr_parts - 1));
 590                 else
 591                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 592                                 'a' + ((offset / 26) - 1),
 593                                 'a' + (offset % 26),
 594                                 minor & (nr_parts - 1));
 595         }
 596
 597         gd->major = XENVBD_MAJOR;
 598         gd->first_minor = minor;
 599         gd->fops = &xlvbd_block_fops;
 600         gd->private_data = info;
 601         gd->driverfs_dev = &(info->xbdev->dev);
 602         set_capacity(gd, capacity);
 603
 604         if (xlvbd_init_blk_queue(gd, sector_size)) {
 605                 del_gendisk(gd);
 606                 goto release;
 607         }
 608
 609         info->rq = gd->queue;
 610         info->gd = gd;
 611
 612         xlvbd_flush(info);
 613
 614         if (vdisk_info & VDISK_READONLY)
 615                 set_disk_ro(gd, 1);
 616
 617         if (vdisk_info & VDISK_REMOVABLE)
 618                 gd->flags |= GENHD_FL_REMOVABLE;
 619
 620         if (vdisk_info & VDISK_CDROM)
 621                 gd->flags |= GENHD_FL_CD;
 622
 623         return 0;
 624
 625  release:
 626         xlbd_release_minors(minor, nr_minors);
 627  out:
 628         return err;
 629 }
 630
 631 static void xlvbd_release_gendisk(struct blkfront_info *info)
 632 {
 633         unsigned int minor, nr_minors;
 634         unsigned long flags;
 635
 636         if (info->rq == NULL)
 637                 return;
 638
 639         spin_lock_irqsave(&blkif_io_lock, flags);
 640
 641         /* No more blkif_request(). */
 642         blk_stop_queue(info->rq);
 643
 644         /* No more gnttab callback work. */
 645         gnttab_cancel_free_callback(&info->callback);
 646         spin_unlock_irqrestore(&blkif_io_lock, flags);
 647
 648         /* Flush gnttab callback work. Must be done with no locks held. */
 649         flush_work_sync(&info->work);
 650
 651         del_gendisk(info->gd);
 652
 653         minor = info->gd->first_minor;
 654         nr_minors = info->gd->minors;
 655         xlbd_release_minors(minor, nr_minors);
 656
 657         blk_cleanup_queue(info->rq);
 658         info->rq = NULL;
 659
 660         put_disk(info->gd);
 661         info->gd = NULL;
 662 }
 663
 664 static void kick_pending_request_queues(struct blkfront_info *info)
 665 {
 666         if (!RING_FULL(&info->ring)) {
 667                 /* Re-enable calldowns. */
 668                 blk_start_queue(info->rq);
 669                 /* Kick things off immediately. */
 670                 do_blkif_request(info->rq);
 671         }
 672 }
 673
 674 static void blkif_restart_queue(struct work_struct *work)
 675 {
 676         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 677
 678         spin_lock_irq(&blkif_io_lock);
 679         if (info->connected == BLKIF_STATE_CONNECTED)
 680                 kick_pending_request_queues(info);
 681         spin_unlock_irq(&blkif_io_lock);
 682 }
 683
 684 static void blkif_free(struct blkfront_info *info, int suspend)
 685 {
 686         /* Prevent new requests being issued until we fix things up. */
 687         spin_lock_irq(&blkif_io_lock);
 688         info->connected = suspend ?
 689                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 690         /* No more blkif_request(). */
 691         if (info->rq)
 692                 blk_stop_queue(info->rq);
 693         /* No more gnttab callback work. */
 694         gnttab_cancel_free_callback(&info->callback);
 695         spin_unlock_irq(&blkif_io_lock);
 696
 697         /* Flush gnttab callback work. Must be done with no locks held. */
 698         flush_work_sync(&info->work);
 699
 700         /* Free resources associated with old device channel. */
 701         if (info->ring_ref != GRANT_INVALID_REF) {
 702                 gnttab_end_foreign_access(info->ring_ref, 0,
 703                                           (unsigned long)info->ring.sring);
 704                 info->ring_ref = GRANT_INVALID_REF;
 705                 info->ring.sring = NULL;
 706         }
 707         if (info->irq)
 708                 unbind_from_irqhandler(info->irq, info);
 709         info->evtchn = info->irq = 0;
 710
 711 }
 712
 713 static void blkif_completion(struct blk_shadow *s)
 714 {
 715         int i;
 716         /* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
 717          * flag. */
 718         for (i = 0; i < s->req.u.rw.nr_segments; i++)
 719                 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
 720 }
 721
 722 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 723 {
 724         struct request *req;
 725         struct blkif_response *bret;
 726         RING_IDX i, rp;
 727         unsigned long flags;
 728         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 729         int error;
 730
 731         spin_lock_irqsave(&blkif_io_lock, flags);
 732
 733         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 734                 spin_unlock_irqrestore(&blkif_io_lock, flags);
 735                 return IRQ_HANDLED;
 736         }
 737
 738  again:
 739         rp = info->ring.sring->rsp_prod;
 740         rmb(); /* Ensure we see queued responses up to 'rp'. */
 741
 742         for (i = info->ring.rsp_cons; i != rp; i++) {
 743                 unsigned long id;
 744
 745                 bret = RING_GET_RESPONSE(&info->ring, i);
 746                 id   = bret->id;
 747                 req  = info->shadow[id].request;
 748
 749                 if (bret->operation != BLKIF_OP_DISCARD)
 750                         blkif_completion(&info->shadow[id]);
 751
 752                 add_id_to_freelist(info, id);
 753
 754                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 755                 switch (bret->operation) {
 756                 case BLKIF_OP_DISCARD:
 757                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 758                                 struct request_queue *rq = info->rq;
 759                                 printk(KERN_WARNING "blkfront: %s: discard op failed\n",
 760                                            info->gd->disk_name);
 761                                 error = -EOPNOTSUPP;
 762                                 info->feature_discard = 0;
 763                                 info->feature_secdiscard = 0;
 764                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 765                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 766                         }
 767                         __blk_end_request_all(req, error);
 768                         break;
 769                 case BLKIF_OP_FLUSH_DISKCACHE:
 770                 case BLKIF_OP_WRITE_BARRIER:
 771                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 772                                 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
 773                                        info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 774                                        "barrier" :  "flush disk cache",
 775                                        info->gd->disk_name);
 776                                 error = -EOPNOTSUPP;
 777                         }
 778                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 779                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
 780                                 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
 781                                        info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 782                                        "barrier" :  "flush disk cache",
 783                                        info->gd->disk_name);
 784                                 error = -EOPNOTSUPP;
 785                         }
 786                         if (unlikely(error)) {
 787                                 if (error == -EOPNOTSUPP)
 788                                         error = 0;
 789                                 info->feature_flush = 0;
 790                                 info->flush_op = 0;
 791                                 xlvbd_flush(info);
 792                         }
 793                         /* fall through */
 794                 case BLKIF_OP_READ:
 795                 case BLKIF_OP_WRITE:
 796                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 797                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 798                                         "request: %x\n", bret->status);
 799
 800                         __blk_end_request_all(req, error);
 801                         break;
 802                 default:
 803                         BUG();
 804                 }
 805         }
 806
 807         info->ring.rsp_cons = i;
 808
 809         if (i != info->ring.req_prod_pvt) {
 810                 int more_to_do;
 811                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 812                 if (more_to_do)
 813                         goto again;
 814         } else
 815                 info->ring.sring->rsp_event = i + 1;
 816
 817         kick_pending_request_queues(info);
 818
 819         spin_unlock_irqrestore(&blkif_io_lock, flags);
 820
 821         return IRQ_HANDLED;
 822 }
 823
 824
 825 static int setup_blkring(struct xenbus_device *dev,
 826                          struct blkfront_info *info)
 827 {
 828         struct blkif_sring *sring;
 829         int err;
 830
 831         info->ring_ref = GRANT_INVALID_REF;
 832
 833         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 834         if (!sring) {
 835                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 836                 return -ENOMEM;
 837         }
 838         SHARED_RING_INIT(sring);
 839         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 840
 841         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 842
 843         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 844         if (err < 0) {
 845                 free_page((unsigned long)sring);
 846                 info->ring.sring = NULL;
 847                 goto fail;
 848         }
 849         info->ring_ref = err;
 850
 851         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 852         if (err)
 853                 goto fail;
 854
 855         err = bind_evtchn_to_irqhandler(info->evtchn,
 856                                         blkif_interrupt,
 857                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 858         if (err <= 0) {
 859                 xenbus_dev_fatal(dev, err,
 860                                  "bind_evtchn_to_irqhandler failed");
 861                 goto fail;
 862         }
 863         info->irq = err;
 864
 865         return 0;
 866 fail:
 867         blkif_free(info, 0);
 868         return err;
 869 }
 870
 871
 872 /* Common code used when first setting up, and when resuming. */
 873 static int talk_to_blkback(struct xenbus_device *dev,
 874                            struct blkfront_info *info)
 875 {
 876         const char *message = NULL;
 877         struct xenbus_transaction xbt;
 878         int err;
 879
 880         /* Create shared ring, alloc event channel. */
 881         err = setup_blkring(dev, info);
 882         if (err)
 883                 goto out;
 884
 885 again:
 886         err = xenbus_transaction_start(&xbt);
 887         if (err) {
 888                 xenbus_dev_fatal(dev, err, "starting transaction");
 889                 goto destroy_blkring;
 890         }
 891
 892         err = xenbus_printf(xbt, dev->nodename,
 893                             "ring-ref", "%u", info->ring_ref);
 894         if (err) {
 895                 message = "writing ring-ref";
 896                 goto abort_transaction;
 897         }
 898         err = xenbus_printf(xbt, dev->nodename,
 899                             "event-channel", "%u", info->evtchn);
 900         if (err) {
 901                 message = "writing event-channel";
 902                 goto abort_transaction;
 903         }
 904         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 905                             XEN_IO_PROTO_ABI_NATIVE);
 906         if (err) {
 907                 message = "writing protocol";
 908                 goto abort_transaction;
 909         }
 910
 911         err = xenbus_transaction_end(xbt, 0);
 912         if (err) {
 913                 if (err == -EAGAIN)
 914                         goto again;
 915                 xenbus_dev_fatal(dev, err, "completing transaction");
 916                 goto destroy_blkring;
 917         }
 918
 919         xenbus_switch_state(dev, XenbusStateInitialised);
 920
 921         return 0;
 922
 923  abort_transaction:
 924         xenbus_transaction_end(xbt, 1);
 925         if (message)
 926                 xenbus_dev_fatal(dev, err, "%s", message);
 927  destroy_blkring:
 928         blkif_free(info, 0);
 929  out:
 930         return err;
 931 }
 932
 933 /**
 934  * Entry point to this code when a new device is created.  Allocate the basic
 935  * structures and the ring buffer for communication with the backend, and
 936  * inform the backend of the appropriate details for those.  Switch to
 937  * Initialised state.
 938  */
 939 static int blkfront_probe(struct xenbus_device *dev,
 940                           const struct xenbus_device_id *id)
 941 {
 942         int err, vdevice, i;
 943         struct blkfront_info *info;
 944
 945         /* FIXME: Use dynamic device id if this is not set. */
 946         err = xenbus_scanf(XBT_NIL, dev->nodename,
 947                            "virtual-device", "%i", &vdevice);
 948         if (err != 1) {
 949                 /* go looking in the extended area instead */
 950                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 951                                    "%i", &vdevice);
 952                 if (err != 1) {
 953                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 954                         return err;
 955                 }
 956         }
 957
 958         if (xen_hvm_domain()) {
 959                 char *type;
 960                 int len;
 961                 /* no unplug has been done: do not hook devices != xen vbds */
 962                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
 963                         int major;
 964
 965                         if (!VDEV_IS_EXTENDED(vdevice))
 966                                 major = BLKIF_MAJOR(vdevice);
 967                         else
 968                                 major = XENVBD_MAJOR;
 969
 970                         if (major != XENVBD_MAJOR) {
 971                                 printk(KERN_INFO
 972                                                 "%s: HVM does not support vbd %d as xen block device\n",
 973                                                 __FUNCTION__, vdevice);
 974                                 return -ENODEV;
 975                         }
 976                 }
 977                 /* do not create a PV cdrom device if we are an HVM guest */
 978                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 979                 if (IS_ERR(type))
 980                         return -ENODEV;
 981                 if (strncmp(type, "cdrom", 5) == 0) {
 982                         kfree(type);
 983                         return -ENODEV;
 984                 }
 985                 kfree(type);
 986         }
 987         info = kzalloc(sizeof(*info), GFP_KERNEL);
 988         if (!info) {
 989                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 990                 return -ENOMEM;
 991         }
 992
 993         mutex_init(&info->mutex);
 994         info->xbdev = dev;
 995         info->vdevice = vdevice;
 996         info->connected = BLKIF_STATE_DISCONNECTED;
 997         INIT_WORK(&info->work, blkif_restart_queue);
 998
 999         for (i = 0; i < BLK_RING_SIZE; i++)
1000                 info->shadow[i].req.u.rw.id = i+1;
1001         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1002
1003         /* Front end dir is a number, which is used as the id. */
1004         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1005         dev_set_drvdata(&dev->dev, info);
1006
1007         err = talk_to_blkback(dev, info);
1008         if (err) {
1009                 kfree(info);
1010                 dev_set_drvdata(&dev->dev, NULL);
1011                 return err;
1012         }
1013
1014         return 0;
1015 }
1016
1017
1018 static int blkif_recover(struct blkfront_info *info)
1019 {
1020         int i;
1021         struct blkif_request *req;
1022         struct blk_shadow *copy;
1023         int j;
1024
1025         /* Stage 1: Make a safe copy of the shadow state. */
1026         copy = kmalloc(sizeof(info->shadow),
1027                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1028         if (!copy)
1029                 return -ENOMEM;
1030         memcpy(copy, info->shadow, sizeof(info->shadow));
1031
1032         /* Stage 2: Set up free list. */
1033         memset(&info->shadow, 0, sizeof(info->shadow));
1034         for (i = 0; i < BLK_RING_SIZE; i++)
1035                 info->shadow[i].req.u.rw.id = i+1;
1036         info->shadow_free = info->ring.req_prod_pvt;
1037         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1038
1039         /* Stage 3: Find pending requests and requeue them. */
1040         for (i = 0; i < BLK_RING_SIZE; i++) {
1041                 /* Not in use? */
1042                 if (!copy[i].request)
1043                         continue;
1044
1045                 /* Grab a request slot and copy shadow state into it. */
1046                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1047                 *req = copy[i].req;
1048
1049                 /* We get a new request id, and must reset the shadow state. */
1050                 req->u.rw.id = get_id_from_freelist(info);
1051                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1052
1053                 if (req->operation != BLKIF_OP_DISCARD) {
1054                 /* Rewrite any grant references invalidated by susp/resume. */
1055                         for (j = 0; j < req->u.rw.nr_segments; j++)
1056                                 gnttab_grant_foreign_access_ref(
1057                                         req->u.rw.seg[j].gref,
1058                                         info->xbdev->otherend_id,
1059                                         pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1060                                         rq_data_dir(info->shadow[req->u.rw.id].request));
1061                 }
1062                 info->shadow[req->u.rw.id].req = *req;
1063
1064                 info->ring.req_prod_pvt++;
1065         }
1066
1067         kfree(copy);
1068
1069         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1070
1071         spin_lock_irq(&blkif_io_lock);
1072
1073         /* Now safe for us to use the shared ring */
1074         info->connected = BLKIF_STATE_CONNECTED;
1075
1076         /* Send off requeued requests */
1077         flush_requests(info);
1078
1079         /* Kick any other new requests queued since we resumed */
1080         kick_pending_request_queues(info);
1081
1082         spin_unlock_irq(&blkif_io_lock);
1083
1084         return 0;
1085 }
1086
1087 /**
1088  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1089  * driver restart.  We tear down our blkif structure and recreate it, but
1090  * leave the device-layer structures intact so that this is transparent to the
1091  * rest of the kernel.
1092  */
1093 static int blkfront_resume(struct xenbus_device *dev)
1094 {
1095         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1096         int err;
1097
1098         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1099
1100         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1101
1102         err = talk_to_blkback(dev, info);
1103         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1104                 err = blkif_recover(info);
1105
1106         return err;
1107 }
1108
1109 static void
1110 blkfront_closing(struct blkfront_info *info)
1111 {
1112         struct xenbus_device *xbdev = info->xbdev;
1113         struct block_device *bdev = NULL;
1114
1115         mutex_lock(&info->mutex);
1116
1117         if (xbdev->state == XenbusStateClosing) {
1118                 mutex_unlock(&info->mutex);
1119                 return;
1120         }
1121
1122         if (info->gd)
1123                 bdev = bdget_disk(info->gd, 0);
1124
1125         mutex_unlock(&info->mutex);
1126
1127         if (!bdev) {
1128                 xenbus_frontend_closed(xbdev);
1129                 return;
1130         }
1131
1132         mutex_lock(&bdev->bd_mutex);
1133
1134         if (bdev->bd_openers) {
1135                 xenbus_dev_error(xbdev, -EBUSY,
1136                                  "Device in use; refusing to close");
1137                 xenbus_switch_state(xbdev, XenbusStateClosing);
1138         } else {
1139                 xlvbd_release_gendisk(info);
1140                 xenbus_frontend_closed(xbdev);
1141         }
1142
1143         mutex_unlock(&bdev->bd_mutex);
1144         bdput(bdev);
1145 }
1146
1147 static void blkfront_setup_discard(struct blkfront_info *info)
1148 {
1149         int err;
1150         char *type;
1151         unsigned int discard_granularity;
1152         unsigned int discard_alignment;
1153         unsigned int discard_secure;
1154
1155         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1156         if (IS_ERR(type))
1157                 return;
1158
1159         info->feature_secdiscard = 0;
1160         if (strncmp(type, "phy", 3) == 0) {
1161                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1162                         "discard-granularity", "%u", &discard_granularity,
1163                         "discard-alignment", "%u", &discard_alignment,
1164                         NULL);
1165                 if (!err) {
1166                         info->feature_discard = 1;
1167                         info->discard_granularity = discard_granularity;
1168                         info->discard_alignment = discard_alignment;
1169                 }
1170                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1171                             "discard-secure", "%d", &discard_secure,
1172                             NULL);
1173                 if (!err)
1174                         info->feature_secdiscard = discard_secure;
1175
1176         } else if (strncmp(type, "file", 4) == 0)
1177                 info->feature_discard = 1;
1178
1179         kfree(type);
1180 }
1181
1182 /*
1183  * Invoked when the backend is finally 'ready' (and has told produced
1184  * the details about the physical device - #sectors, size, etc).
1185  */
1186 static void blkfront_connect(struct blkfront_info *info)
1187 {
1188         unsigned long long sectors;
1189         unsigned long sector_size;
1190         unsigned int binfo;
1191         int err;
1192         int barrier, flush, discard;
1193
1194         switch (info->connected) {
1195         case BLKIF_STATE_CONNECTED:
1196                 /*
1197                  * Potentially, the back-end may be signalling
1198                  * a capacity change; update the capacity.
1199                  */
1200                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1201                                    "sectors", "%Lu", &sectors);
1202                 if (XENBUS_EXIST_ERR(err))
1203                         return;
1204                 printk(KERN_INFO "Setting capacity to %Lu\n",
1205                        sectors);
1206                 set_capacity(info->gd, sectors);
1207                 revalidate_disk(info->gd);
1208
1209                 /* fall through */
1210         case BLKIF_STATE_SUSPENDED:
1211                 return;
1212
1213         default:
1214                 break;
1215         }
1216
1217         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1218                 __func__, info->xbdev->otherend);
1219
1220         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1221                             "sectors", "%llu", &sectors,
1222                             "info", "%u", &binfo,
1223                             "sector-size", "%lu", &sector_size,
1224                             NULL);
1225         if (err) {
1226                 xenbus_dev_fatal(info->xbdev, err,
1227                                  "reading backend fields at %s",
1228                                  info->xbdev->otherend);
1229                 return;
1230         }
1231
1232         info->feature_flush = 0;
1233         info->flush_op = 0;
1234
1235         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1236                             "feature-barrier", "%d", &barrier,
1237                             NULL);
1238
1239         /*
1240          * If there's no "feature-barrier" defined, then it means
1241          * we're dealing with a very old backend which writes
1242          * synchronously; nothing to do.
1243          *
1244          * If there are barriers, then we use flush.
1245          */
1246         if (!err && barrier) {
1247                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1248                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1249         }
1250         /*
1251          * And if there is "feature-flush-cache" use that above
1252          * barriers.
1253          */
1254         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1255                             "feature-flush-cache", "%d", &flush,
1256                             NULL);
1257
1258         if (!err && flush) {
1259                 info->feature_flush = REQ_FLUSH;
1260                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1261         }
1262
1263         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1264                             "feature-discard", "%d", &discard,
1265                             NULL);
1266
1267         if (!err && discard)
1268                 blkfront_setup_discard(info);
1269
1270         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1271         if (err) {
1272                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1273                                  info->xbdev->otherend);
1274                 return;
1275         }
1276
1277         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1278
1279         /* Kick pending requests. */
1280         spin_lock_irq(&blkif_io_lock);
1281         info->connected = BLKIF_STATE_CONNECTED;
1282         kick_pending_request_queues(info);
1283         spin_unlock_irq(&blkif_io_lock);
1284
1285         add_disk(info->gd);
1286
1287         info->is_ready = 1;
1288 }
1289
1290 /**
1291  * Callback received when the backend's state changes.
1292  */
1293 static void blkback_changed(struct xenbus_device *dev,
1294                             enum xenbus_state backend_state)
1295 {
1296         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1297
1298         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1299
1300         switch (backend_state) {
1301         case XenbusStateInitialising:
1302         case XenbusStateInitWait:
1303         case XenbusStateInitialised:
1304         case XenbusStateReconfiguring:
1305         case XenbusStateReconfigured:
1306         case XenbusStateUnknown:
1307         case XenbusStateClosed:
1308                 break;
1309
1310         case XenbusStateConnected:
1311                 blkfront_connect(info);
1312                 break;
1313
1314         case XenbusStateClosing:
1315                 blkfront_closing(info);
1316                 break;
1317         }
1318 }
1319
1320 static int blkfront_remove(struct xenbus_device *xbdev)
1321 {
1322         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1323         struct block_device *bdev = NULL;
1324         struct gendisk *disk;
1325
1326         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1327
1328         blkif_free(info, 0);
1329
1330         mutex_lock(&info->mutex);
1331
1332         disk = info->gd;
1333         if (disk)
1334                 bdev = bdget_disk(disk, 0);
1335
1336         info->xbdev = NULL;
1337         mutex_unlock(&info->mutex);
1338
1339         if (!bdev) {
1340                 kfree(info);
1341                 return 0;
1342         }
1343
1344         /*
1345          * The xbdev was removed before we reached the Closed
1346          * state. See if it's safe to remove the disk. If the bdev
1347          * isn't closed yet, we let release take care of it.
1348          */
1349
1350         mutex_lock(&bdev->bd_mutex);
1351         info = disk->private_data;
1352
1353         dev_warn(disk_to_dev(disk),
1354                  "%s was hot-unplugged, %d stale handles\n",
1355                  xbdev->nodename, bdev->bd_openers);
1356
1357         if (info && !bdev->bd_openers) {
1358                 xlvbd_release_gendisk(info);
1359                 disk->private_data = NULL;
1360                 kfree(info);
1361         }
1362
1363         mutex_unlock(&bdev->bd_mutex);
1364         bdput(bdev);
1365
1366         return 0;
1367 }
1368
1369 static int blkfront_is_ready(struct xenbus_device *dev)
1370 {
1371         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1372
1373         return info->is_ready && info->xbdev;
1374 }
1375
1376 static int blkif_open(struct block_device *bdev, fmode_t mode)
1377 {
1378         struct gendisk *disk = bdev->bd_disk;
1379         struct blkfront_info *info;
1380         int err = 0;
1381
1382         mutex_lock(&blkfront_mutex);
1383
1384         info = disk->private_data;
1385         if (!info) {
1386                 /* xbdev gone */
1387                 err = -ERESTARTSYS;
1388                 goto out;
1389         }
1390
1391         mutex_lock(&info->mutex);
1392
1393         if (!info->gd)
1394                 /* xbdev is closed */
1395                 err = -ERESTARTSYS;
1396
1397         mutex_unlock(&info->mutex);
1398
1399 out:
1400         mutex_unlock(&blkfront_mutex);
1401         return err;
1402 }
1403
1404 static int blkif_release(struct gendisk *disk, fmode_t mode)
1405 {
1406         struct blkfront_info *info = disk->private_data;
1407         struct block_device *bdev;
1408         struct xenbus_device *xbdev;
1409
1410         mutex_lock(&blkfront_mutex);
1411
1412         bdev = bdget_disk(disk, 0);
1413         bdput(bdev);
1414
1415         if (bdev->bd_openers)
1416                 goto out;
1417
1418         /*
1419          * Check if we have been instructed to close. We will have
1420          * deferred this request, because the bdev was still open.
1421          */
1422
1423         mutex_lock(&info->mutex);
1424         xbdev = info->xbdev;
1425
1426         if (xbdev && xbdev->state == XenbusStateClosing) {
1427                 /* pending switch to state closed */
1428                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1429                 xlvbd_release_gendisk(info);
1430                 xenbus_frontend_closed(info->xbdev);
1431         }
1432
1433         mutex_unlock(&info->mutex);
1434
1435         if (!xbdev) {
1436                 /* sudden device removal */
1437                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1438                 xlvbd_release_gendisk(info);
1439                 disk->private_data = NULL;
1440                 kfree(info);
1441         }
1442
1443 out:
1444         mutex_unlock(&blkfront_mutex);
1445         return 0;
1446 }
1447
1448 static const struct block_device_operations xlvbd_block_fops =
1449 {
1450         .owner = THIS_MODULE,
1451         .open = blkif_open,
1452         .release = blkif_release,
1453         .getgeo = blkif_getgeo,
1454         .ioctl = blkif_ioctl,
1455 };
1456
1457
1458 static const struct xenbus_device_id blkfront_ids[] = {
1459         { "vbd" },
1460         { "" }
1461 };
1462
1463 static DEFINE_XENBUS_DRIVER(blkfront, ,
1464         .probe = blkfront_probe,
1465         .remove = blkfront_remove,
1466         .resume = blkfront_resume,
1467         .otherend_changed = blkback_changed,
1468         .is_ready = blkfront_is_ready,
1469 );
1470
1471 static int __init xlblk_init(void)
1472 {
1473         int ret;
1474
1475         if (!xen_domain())
1476                 return -ENODEV;
1477
1478         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1479                 return -ENODEV;
1480
1481         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1482                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1483                        XENVBD_MAJOR, DEV_NAME);
1484                 return -ENODEV;
1485         }
1486
1487         ret = xenbus_register_frontend(&blkfront_driver);
1488         if (ret) {
1489                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1490                 return ret;
1491         }
1492
1493         return 0;
1494 }
1495 module_init(xlblk_init);
1496
1497
1498 static void __exit xlblk_exit(void)
1499 {
1500         return xenbus_unregister_driver(&blkfront_driver);
1501 }
1502 module_exit(xlblk_exit);
1503
1504 MODULE_DESCRIPTION("Xen virtual block device frontend");
1505 MODULE_LICENSE("GPL");
1506 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1507 MODULE_ALIAS("xen:vbd");
1508 MODULE_ALIAS("xenblk");