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 #include <linux/bitmap.h>
  47
  48 #include <xen/xen.h>
  49 #include <xen/xenbus.h>
  50 #include <xen/grant_table.h>
  51 #include <xen/events.h>
  52 #include <xen/page.h>
  53 #include <xen/platform_pci.h>
  54
  55 #include <xen/interface/grant_table.h>
  56 #include <xen/interface/io/blkif.h>
  57 #include <xen/interface/io/protocols.h>
  58
  59 #include <asm/xen/hypervisor.h>
  60
  61 enum blkif_state {
  62         BLKIF_STATE_DISCONNECTED,
  63         BLKIF_STATE_CONNECTED,
  64         BLKIF_STATE_SUSPENDED,
  65 };
  66
  67 struct blk_shadow {
  68         struct blkif_request req;
  69         struct request *request;
  70         unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71 };
  72
  73 static DEFINE_MUTEX(blkfront_mutex);
  74 static const struct block_device_operations xlvbd_block_fops;
  75
  76 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
  77
  78 /*
  79  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  80  * hang in private_data off the gendisk structure. We may end up
  81  * putting all kinds of interesting stuff here :-)
  82  */
  83 struct blkfront_info
  84 {
  85         spinlock_t io_lock;
  86         struct mutex mutex;
  87         struct xenbus_device *xbdev;
  88         struct gendisk *gd;
  89         int vdevice;
  90         blkif_vdev_t handle;
  91         enum blkif_state connected;
  92         int ring_ref;
  93         struct blkif_front_ring ring;
  94         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  95         unsigned int evtchn, irq;
  96         struct request_queue *rq;
  97         struct work_struct work;
  98         struct gnttab_free_callback callback;
  99         struct blk_shadow shadow[BLK_RING_SIZE];
 100         unsigned long shadow_free;
 101         unsigned int feature_flush;
 102         unsigned int flush_op;
 103         unsigned int feature_discard:1;
 104         unsigned int feature_secdiscard:1;
 105         unsigned int discard_granularity;
 106         unsigned int discard_alignment;
 107         int is_ready;
 108 };
 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                 bitmap_set(minors, minor, nr);
 181                 rc = 0;
 182         } else
 183                 rc = -EBUSY;
 184         spin_unlock(&minor_lock);
 185
 186         return rc;
 187 }
 188
 189 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 190 {
 191         unsigned int end = minor + nr;
 192
 193         BUG_ON(end > nr_minors);
 194         spin_lock(&minor_lock);
 195         bitmap_clear(minors,  minor, nr);
 196         spin_unlock(&minor_lock);
 197 }
 198
 199 static void blkif_restart_queue_callback(void *arg)
 200 {
 201         struct blkfront_info *info = (struct blkfront_info *)arg;
 202         schedule_work(&info->work);
 203 }
 204
 205 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 206 {
 207         /* We don't have real geometry info, but let's at least return
 208            values consistent with the size of the device */
 209         sector_t nsect = get_capacity(bd->bd_disk);
 210         sector_t cylinders = nsect;
 211
 212         hg->heads = 0xff;
 213         hg->sectors = 0x3f;
 214         sector_div(cylinders, hg->heads * hg->sectors);
 215         hg->cylinders = cylinders;
 216         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 217                 hg->cylinders = 0xffff;
 218         return 0;
 219 }
 220
 221 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 222                        unsigned command, unsigned long argument)
 223 {
 224         struct blkfront_info *info = bdev->bd_disk->private_data;
 225         int i;
 226
 227         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 228                 command, (long)argument);
 229
 230         switch (command) {
 231         case CDROMMULTISESSION:
 232                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 233                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 234                         if (put_user(0, (char __user *)(argument + i)))
 235                                 return -EFAULT;
 236                 return 0;
 237
 238         case CDROM_GET_CAPABILITY: {
 239                 struct gendisk *gd = info->gd;
 240                 if (gd->flags & GENHD_FL_CD)
 241                         return 0;
 242                 return -EINVAL;
 243         }
 244
 245         default:
 246                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 247                   command);*/
 248                 return -EINVAL; /* same return as native Linux */
 249         }
 250
 251         return 0;
 252 }
 253
 254 /*
 255  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 256  * and writes are handled as expected.
 257  *
 258  * @req: a request struct
 259  */
 260 static int blkif_queue_request(struct request *req)
 261 {
 262         struct blkfront_info *info = req->rq_disk->private_data;
 263         unsigned long buffer_mfn;
 264         struct blkif_request *ring_req;
 265         unsigned long id;
 266         unsigned int fsect, lsect;
 267         int i, ref;
 268         grant_ref_t gref_head;
 269         struct scatterlist *sg;
 270
 271         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 272                 return 1;
 273
 274         if (gnttab_alloc_grant_references(
 275                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 276                 gnttab_request_free_callback(
 277                         &info->callback,
 278                         blkif_restart_queue_callback,
 279                         info,
 280                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 281                 return 1;
 282         }
 283
 284         /* Fill out a communications ring structure. */
 285         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 286         id = get_id_from_freelist(info);
 287         info->shadow[id].request = req;
 288
 289         ring_req->u.rw.id = id;
 290         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 291         ring_req->u.rw.handle = info->handle;
 292
 293         ring_req->operation = rq_data_dir(req) ?
 294                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 295
 296         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 297                 /*
 298                  * Ideally we can do an unordered flush-to-disk. In case the
 299                  * backend onlysupports barriers, use that. A barrier request
 300                  * a superset of FUA, so we can implement it the same
 301                  * way.  (It's also a FLUSH+FUA, since it is
 302                  * guaranteed ordered WRT previous writes.)
 303                  */
 304                 ring_req->operation = info->flush_op;
 305         }
 306
 307         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 308                 /* id, sector_number and handle are set above. */
 309                 ring_req->operation = BLKIF_OP_DISCARD;
 310                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 311                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 312                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 313                 else
 314                         ring_req->u.discard.flag = 0;
 315         } else {
 316                 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
 317                                                            info->sg);
 318                 BUG_ON(ring_req->u.rw.nr_segments >
 319                        BLKIF_MAX_SEGMENTS_PER_REQUEST);
 320
 321                 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
 322                         buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 323                         fsect = sg->offset >> 9;
 324                         lsect = fsect + (sg->length >> 9) - 1;
 325                         /* install a grant reference. */
 326                         ref = gnttab_claim_grant_reference(&gref_head);
 327                         BUG_ON(ref == -ENOSPC);
 328
 329                         gnttab_grant_foreign_access_ref(
 330                                         ref,
 331                                         info->xbdev->otherend_id,
 332                                         buffer_mfn,
 333                                         rq_data_dir(req));
 334
 335                         info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 336                         ring_req->u.rw.seg[i] =
 337                                         (struct blkif_request_segment) {
 338                                                 .gref       = ref,
 339                                                 .first_sect = fsect,
 340                                                 .last_sect  = lsect };
 341                 }
 342         }
 343
 344         info->ring.req_prod_pvt++;
 345
 346         /* Keep a private copy so we can reissue requests when recovering. */
 347         info->shadow[id].req = *ring_req;
 348
 349         gnttab_free_grant_references(gref_head);
 350
 351         return 0;
 352 }
 353
 354
 355 static inline void flush_requests(struct blkfront_info *info)
 356 {
 357         int notify;
 358
 359         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 360
 361         if (notify)
 362                 notify_remote_via_irq(info->irq);
 363 }
 364
 365 /*
 366  * do_blkif_request
 367  *  read a block; request is in a request queue
 368  */
 369 static void do_blkif_request(struct request_queue *rq)
 370 {
 371         struct blkfront_info *info = NULL;
 372         struct request *req;
 373         int queued;
 374
 375         pr_debug("Entered do_blkif_request\n");
 376
 377         queued = 0;
 378
 379         while ((req = blk_peek_request(rq)) != NULL) {
 380                 info = req->rq_disk->private_data;
 381
 382                 if (RING_FULL(&info->ring))
 383                         goto wait;
 384
 385                 blk_start_request(req);
 386
 387                 if ((req->cmd_type != REQ_TYPE_FS) ||
 388                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 389                     !info->flush_op)) {
 390                         __blk_end_request_all(req, -EIO);
 391                         continue;
 392                 }
 393
 394                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 395                          "(%u/%u) buffer:%p [%s]\n",
 396                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 397                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 398                          req->buffer, rq_data_dir(req) ? "write" : "read");
 399
 400                 if (blkif_queue_request(req)) {
 401                         blk_requeue_request(rq, req);
 402 wait:
 403                         /* Avoid pointless unplugs. */
 404                         blk_stop_queue(rq);
 405                         break;
 406                 }
 407
 408                 queued++;
 409         }
 410
 411         if (queued != 0)
 412                 flush_requests(info);
 413 }
 414
 415 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 416 {
 417         struct request_queue *rq;
 418         struct blkfront_info *info = gd->private_data;
 419
 420         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 421         if (rq == NULL)
 422                 return -1;
 423
 424         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 425
 426         if (info->feature_discard) {
 427                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 428                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 429                 rq->limits.discard_granularity = info->discard_granularity;
 430                 rq->limits.discard_alignment = info->discard_alignment;
 431                 if (info->feature_secdiscard)
 432                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 433         }
 434
 435         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 436         blk_queue_logical_block_size(rq, sector_size);
 437         blk_queue_max_hw_sectors(rq, 512);
 438
 439         /* Each segment in a request is up to an aligned page in size. */
 440         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 441         blk_queue_max_segment_size(rq, PAGE_SIZE);
 442
 443         /* Ensure a merged request will fit in a single I/O ring slot. */
 444         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 445
 446         /* Make sure buffer addresses are sector-aligned. */
 447         blk_queue_dma_alignment(rq, 511);
 448
 449         /* Make sure we don't use bounce buffers. */
 450         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 451
 452         gd->queue = rq;
 453
 454         return 0;
 455 }
 456
 457
 458 static void xlvbd_flush(struct blkfront_info *info)
 459 {
 460         blk_queue_flush(info->rq, info->feature_flush);
 461         printk(KERN_INFO "blkfront: %s: %s: %s\n",
 462                info->gd->disk_name,
 463                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 464                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 465                 "flush diskcache" : "barrier or flush"),
 466                info->feature_flush ? "enabled" : "disabled");
 467 }
 468
 469 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 470 {
 471         int major;
 472         major = BLKIF_MAJOR(vdevice);
 473         *minor = BLKIF_MINOR(vdevice);
 474         switch (major) {
 475                 case XEN_IDE0_MAJOR:
 476                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 477                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 478                                 EMULATED_HD_DISK_MINOR_OFFSET;
 479                         break;
 480                 case XEN_IDE1_MAJOR:
 481                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 482                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 483                                 EMULATED_HD_DISK_MINOR_OFFSET;
 484                         break;
 485                 case XEN_SCSI_DISK0_MAJOR:
 486                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 487                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 488                         break;
 489                 case XEN_SCSI_DISK1_MAJOR:
 490                 case XEN_SCSI_DISK2_MAJOR:
 491                 case XEN_SCSI_DISK3_MAJOR:
 492                 case XEN_SCSI_DISK4_MAJOR:
 493                 case XEN_SCSI_DISK5_MAJOR:
 494                 case XEN_SCSI_DISK6_MAJOR:
 495                 case XEN_SCSI_DISK7_MAJOR:
 496                         *offset = (*minor / PARTS_PER_DISK) +
 497                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 498                                 EMULATED_SD_DISK_NAME_OFFSET;
 499                         *minor = *minor +
 500                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 501                                 EMULATED_SD_DISK_MINOR_OFFSET;
 502                         break;
 503                 case XEN_SCSI_DISK8_MAJOR:
 504                 case XEN_SCSI_DISK9_MAJOR:
 505                 case XEN_SCSI_DISK10_MAJOR:
 506                 case XEN_SCSI_DISK11_MAJOR:
 507                 case XEN_SCSI_DISK12_MAJOR:
 508                 case XEN_SCSI_DISK13_MAJOR:
 509                 case XEN_SCSI_DISK14_MAJOR:
 510                 case XEN_SCSI_DISK15_MAJOR:
 511                         *offset = (*minor / PARTS_PER_DISK) +
 512                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 513                                 EMULATED_SD_DISK_NAME_OFFSET;
 514                         *minor = *minor +
 515                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 516                                 EMULATED_SD_DISK_MINOR_OFFSET;
 517                         break;
 518                 case XENVBD_MAJOR:
 519                         *offset = *minor / PARTS_PER_DISK;
 520                         break;
 521                 default:
 522                         printk(KERN_WARNING "blkfront: your disk configuration is "
 523                                         "incorrect, please use an xvd device instead\n");
 524                         return -ENODEV;
 525         }
 526         return 0;
 527 }
 528
 529 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 530                                struct blkfront_info *info,
 531                                u16 vdisk_info, u16 sector_size)
 532 {
 533         struct gendisk *gd;
 534         int nr_minors = 1;
 535         int err;
 536         unsigned int offset;
 537         int minor;
 538         int nr_parts;
 539
 540         BUG_ON(info->gd != NULL);
 541         BUG_ON(info->rq != NULL);
 542
 543         if ((info->vdevice>>EXT_SHIFT) > 1) {
 544                 /* this is above the extended range; something is wrong */
 545                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 546                 return -ENODEV;
 547         }
 548
 549         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 550                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 551                 if (err)
 552                         return err;
 553                 nr_parts = PARTS_PER_DISK;
 554         } else {
 555                 minor = BLKIF_MINOR_EXT(info->vdevice);
 556                 nr_parts = PARTS_PER_EXT_DISK;
 557                 offset = minor / nr_parts;
 558                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 559                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 560                                         "emulated IDE disks,\n\t choose an xvd device name"
 561                                         "from xvde on\n", info->vdevice);
 562         }
 563         err = -ENODEV;
 564
 565         if ((minor % nr_parts) == 0)
 566                 nr_minors = nr_parts;
 567
 568         err = xlbd_reserve_minors(minor, nr_minors);
 569         if (err)
 570                 goto out;
 571         err = -ENODEV;
 572
 573         gd = alloc_disk(nr_minors);
 574         if (gd == NULL)
 575                 goto release;
 576
 577         if (nr_minors > 1) {
 578                 if (offset < 26)
 579                         sprintf(gd->disk_name, "%s%c", DEV_NAME, 'a' + offset);
 580                 else
 581                         sprintf(gd->disk_name, "%s%c%c", DEV_NAME,
 582                                 'a' + ((offset / 26)-1), 'a' + (offset % 26));
 583         } else {
 584                 if (offset < 26)
 585                         sprintf(gd->disk_name, "%s%c%d", DEV_NAME,
 586                                 'a' + offset,
 587                                 minor & (nr_parts - 1));
 588                 else
 589                         sprintf(gd->disk_name, "%s%c%c%d", DEV_NAME,
 590                                 'a' + ((offset / 26) - 1),
 591                                 'a' + (offset % 26),
 592                                 minor & (nr_parts - 1));
 593         }
 594
 595         gd->major = XENVBD_MAJOR;
 596         gd->first_minor = minor;
 597         gd->fops = &xlvbd_block_fops;
 598         gd->private_data = info;
 599         gd->driverfs_dev = &(info->xbdev->dev);
 600         set_capacity(gd, capacity);
 601
 602         if (xlvbd_init_blk_queue(gd, sector_size)) {
 603                 del_gendisk(gd);
 604                 goto release;
 605         }
 606
 607         info->rq = gd->queue;
 608         info->gd = gd;
 609
 610         xlvbd_flush(info);
 611
 612         if (vdisk_info & VDISK_READONLY)
 613                 set_disk_ro(gd, 1);
 614
 615         if (vdisk_info & VDISK_REMOVABLE)
 616                 gd->flags |= GENHD_FL_REMOVABLE;
 617
 618         if (vdisk_info & VDISK_CDROM)
 619                 gd->flags |= GENHD_FL_CD;
 620
 621         return 0;
 622
 623  release:
 624         xlbd_release_minors(minor, nr_minors);
 625  out:
 626         return err;
 627 }
 628
 629 static void xlvbd_release_gendisk(struct blkfront_info *info)
 630 {
 631         unsigned int minor, nr_minors;
 632         unsigned long flags;
 633
 634         if (info->rq == NULL)
 635                 return;
 636
 637         spin_lock_irqsave(&info->io_lock, flags);
 638
 639         /* No more blkif_request(). */
 640         blk_stop_queue(info->rq);
 641
 642         /* No more gnttab callback work. */
 643         gnttab_cancel_free_callback(&info->callback);
 644         spin_unlock_irqrestore(&info->io_lock, flags);
 645
 646         /* Flush gnttab callback work. Must be done with no locks held. */
 647         flush_work_sync(&info->work);
 648
 649         del_gendisk(info->gd);
 650
 651         minor = info->gd->first_minor;
 652         nr_minors = info->gd->minors;
 653         xlbd_release_minors(minor, nr_minors);
 654
 655         blk_cleanup_queue(info->rq);
 656         info->rq = NULL;
 657
 658         put_disk(info->gd);
 659         info->gd = NULL;
 660 }
 661
 662 static void kick_pending_request_queues(struct blkfront_info *info)
 663 {
 664         if (!RING_FULL(&info->ring)) {
 665                 /* Re-enable calldowns. */
 666                 blk_start_queue(info->rq);
 667                 /* Kick things off immediately. */
 668                 do_blkif_request(info->rq);
 669         }
 670 }
 671
 672 static void blkif_restart_queue(struct work_struct *work)
 673 {
 674         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 675
 676         spin_lock_irq(&info->io_lock);
 677         if (info->connected == BLKIF_STATE_CONNECTED)
 678                 kick_pending_request_queues(info);
 679         spin_unlock_irq(&info->io_lock);
 680 }
 681
 682 static void blkif_free(struct blkfront_info *info, int suspend)
 683 {
 684         /* Prevent new requests being issued until we fix things up. */
 685         spin_lock_irq(&info->io_lock);
 686         info->connected = suspend ?
 687                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 688         /* No more blkif_request(). */
 689         if (info->rq)
 690                 blk_stop_queue(info->rq);
 691         /* No more gnttab callback work. */
 692         gnttab_cancel_free_callback(&info->callback);
 693         spin_unlock_irq(&info->io_lock);
 694
 695         /* Flush gnttab callback work. Must be done with no locks held. */
 696         flush_work_sync(&info->work);
 697
 698         /* Free resources associated with old device channel. */
 699         if (info->ring_ref != GRANT_INVALID_REF) {
 700                 gnttab_end_foreign_access(info->ring_ref, 0,
 701                                           (unsigned long)info->ring.sring);
 702                 info->ring_ref = GRANT_INVALID_REF;
 703                 info->ring.sring = NULL;
 704         }
 705         if (info->irq)
 706                 unbind_from_irqhandler(info->irq, info);
 707         info->evtchn = info->irq = 0;
 708
 709 }
 710
 711 static void blkif_completion(struct blk_shadow *s)
 712 {
 713         int i;
 714         /* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
 715          * flag. */
 716         for (i = 0; i < s->req.u.rw.nr_segments; i++)
 717                 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
 718 }
 719
 720 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 721 {
 722         struct request *req;
 723         struct blkif_response *bret;
 724         RING_IDX i, rp;
 725         unsigned long flags;
 726         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 727         int error;
 728
 729         spin_lock_irqsave(&info->io_lock, flags);
 730
 731         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 732                 spin_unlock_irqrestore(&info->io_lock, flags);
 733                 return IRQ_HANDLED;
 734         }
 735
 736  again:
 737         rp = info->ring.sring->rsp_prod;
 738         rmb(); /* Ensure we see queued responses up to 'rp'. */
 739
 740         for (i = info->ring.rsp_cons; i != rp; i++) {
 741                 unsigned long id;
 742
 743                 bret = RING_GET_RESPONSE(&info->ring, i);
 744                 id   = bret->id;
 745                 req  = info->shadow[id].request;
 746
 747                 if (bret->operation != BLKIF_OP_DISCARD)
 748                         blkif_completion(&info->shadow[id]);
 749
 750                 add_id_to_freelist(info, id);
 751
 752                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 753                 switch (bret->operation) {
 754                 case BLKIF_OP_DISCARD:
 755                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 756                                 struct request_queue *rq = info->rq;
 757                                 printk(KERN_WARNING "blkfront: %s: discard op failed\n",
 758                                            info->gd->disk_name);
 759                                 error = -EOPNOTSUPP;
 760                                 info->feature_discard = 0;
 761                                 info->feature_secdiscard = 0;
 762                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 763                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 764                         }
 765                         __blk_end_request_all(req, error);
 766                         break;
 767                 case BLKIF_OP_FLUSH_DISKCACHE:
 768                 case BLKIF_OP_WRITE_BARRIER:
 769                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 770                                 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
 771                                        info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 772                                        "barrier" :  "flush disk cache",
 773                                        info->gd->disk_name);
 774                                 error = -EOPNOTSUPP;
 775                         }
 776                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 777                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
 778                                 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
 779                                        info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 780                                        "barrier" :  "flush disk cache",
 781                                        info->gd->disk_name);
 782                                 error = -EOPNOTSUPP;
 783                         }
 784                         if (unlikely(error)) {
 785                                 if (error == -EOPNOTSUPP)
 786                                         error = 0;
 787                                 info->feature_flush = 0;
 788                                 info->flush_op = 0;
 789                                 xlvbd_flush(info);
 790                         }
 791                         /* fall through */
 792                 case BLKIF_OP_READ:
 793                 case BLKIF_OP_WRITE:
 794                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 795                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 796                                         "request: %x\n", bret->status);
 797
 798                         __blk_end_request_all(req, error);
 799                         break;
 800                 default:
 801                         BUG();
 802                 }
 803         }
 804
 805         info->ring.rsp_cons = i;
 806
 807         if (i != info->ring.req_prod_pvt) {
 808                 int more_to_do;
 809                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 810                 if (more_to_do)
 811                         goto again;
 812         } else
 813                 info->ring.sring->rsp_event = i + 1;
 814
 815         kick_pending_request_queues(info);
 816
 817         spin_unlock_irqrestore(&info->io_lock, flags);
 818
 819         return IRQ_HANDLED;
 820 }
 821
 822
 823 static int setup_blkring(struct xenbus_device *dev,
 824                          struct blkfront_info *info)
 825 {
 826         struct blkif_sring *sring;
 827         int err;
 828
 829         info->ring_ref = GRANT_INVALID_REF;
 830
 831         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 832         if (!sring) {
 833                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 834                 return -ENOMEM;
 835         }
 836         SHARED_RING_INIT(sring);
 837         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 838
 839         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 840
 841         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 842         if (err < 0) {
 843                 free_page((unsigned long)sring);
 844                 info->ring.sring = NULL;
 845                 goto fail;
 846         }
 847         info->ring_ref = err;
 848
 849         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 850         if (err)
 851                 goto fail;
 852
 853         err = bind_evtchn_to_irqhandler(info->evtchn,
 854                                         blkif_interrupt,
 855                                         IRQF_SAMPLE_RANDOM, "blkif", info);
 856         if (err <= 0) {
 857                 xenbus_dev_fatal(dev, err,
 858                                  "bind_evtchn_to_irqhandler failed");
 859                 goto fail;
 860         }
 861         info->irq = err;
 862
 863         return 0;
 864 fail:
 865         blkif_free(info, 0);
 866         return err;
 867 }
 868
 869
 870 /* Common code used when first setting up, and when resuming. */
 871 static int talk_to_blkback(struct xenbus_device *dev,
 872                            struct blkfront_info *info)
 873 {
 874         const char *message = NULL;
 875         struct xenbus_transaction xbt;
 876         int err;
 877
 878         /* Create shared ring, alloc event channel. */
 879         err = setup_blkring(dev, info);
 880         if (err)
 881                 goto out;
 882
 883 again:
 884         err = xenbus_transaction_start(&xbt);
 885         if (err) {
 886                 xenbus_dev_fatal(dev, err, "starting transaction");
 887                 goto destroy_blkring;
 888         }
 889
 890         err = xenbus_printf(xbt, dev->nodename,
 891                             "ring-ref", "%u", info->ring_ref);
 892         if (err) {
 893                 message = "writing ring-ref";
 894                 goto abort_transaction;
 895         }
 896         err = xenbus_printf(xbt, dev->nodename,
 897                             "event-channel", "%u", info->evtchn);
 898         if (err) {
 899                 message = "writing event-channel";
 900                 goto abort_transaction;
 901         }
 902         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 903                             XEN_IO_PROTO_ABI_NATIVE);
 904         if (err) {
 905                 message = "writing protocol";
 906                 goto abort_transaction;
 907         }
 908
 909         err = xenbus_transaction_end(xbt, 0);
 910         if (err) {
 911                 if (err == -EAGAIN)
 912                         goto again;
 913                 xenbus_dev_fatal(dev, err, "completing transaction");
 914                 goto destroy_blkring;
 915         }
 916
 917         xenbus_switch_state(dev, XenbusStateInitialised);
 918
 919         return 0;
 920
 921  abort_transaction:
 922         xenbus_transaction_end(xbt, 1);
 923         if (message)
 924                 xenbus_dev_fatal(dev, err, "%s", message);
 925  destroy_blkring:
 926         blkif_free(info, 0);
 927  out:
 928         return err;
 929 }
 930
 931 /**
 932  * Entry point to this code when a new device is created.  Allocate the basic
 933  * structures and the ring buffer for communication with the backend, and
 934  * inform the backend of the appropriate details for those.  Switch to
 935  * Initialised state.
 936  */
 937 static int blkfront_probe(struct xenbus_device *dev,
 938                           const struct xenbus_device_id *id)
 939 {
 940         int err, vdevice, i;
 941         struct blkfront_info *info;
 942
 943         /* FIXME: Use dynamic device id if this is not set. */
 944         err = xenbus_scanf(XBT_NIL, dev->nodename,
 945                            "virtual-device", "%i", &vdevice);
 946         if (err != 1) {
 947                 /* go looking in the extended area instead */
 948                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 949                                    "%i", &vdevice);
 950                 if (err != 1) {
 951                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 952                         return err;
 953                 }
 954         }
 955
 956         if (xen_hvm_domain()) {
 957                 char *type;
 958                 int len;
 959                 /* no unplug has been done: do not hook devices != xen vbds */
 960                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
 961                         int major;
 962
 963                         if (!VDEV_IS_EXTENDED(vdevice))
 964                                 major = BLKIF_MAJOR(vdevice);
 965                         else
 966                                 major = XENVBD_MAJOR;
 967
 968                         if (major != XENVBD_MAJOR) {
 969                                 printk(KERN_INFO
 970                                                 "%s: HVM does not support vbd %d as xen block device\n",
 971                                                 __FUNCTION__, vdevice);
 972                                 return -ENODEV;
 973                         }
 974                 }
 975                 /* do not create a PV cdrom device if we are an HVM guest */
 976                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 977                 if (IS_ERR(type))
 978                         return -ENODEV;
 979                 if (strncmp(type, "cdrom", 5) == 0) {
 980                         kfree(type);
 981                         return -ENODEV;
 982                 }
 983                 kfree(type);
 984         }
 985         info = kzalloc(sizeof(*info), GFP_KERNEL);
 986         if (!info) {
 987                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 988                 return -ENOMEM;
 989         }
 990
 991         mutex_init(&info->mutex);
 992         spin_lock_init(&info->io_lock);
 993         info->xbdev = dev;
 994         info->vdevice = vdevice;
 995         info->connected = BLKIF_STATE_DISCONNECTED;
 996         INIT_WORK(&info->work, blkif_restart_queue);
 997
 998         for (i = 0; i < BLK_RING_SIZE; i++)
 999                 info->shadow[i].req.u.rw.id = i+1;
1000         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1001
1002         /* Front end dir is a number, which is used as the id. */
1003         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1004         dev_set_drvdata(&dev->dev, info);
1005
1006         err = talk_to_blkback(dev, info);
1007         if (err) {
1008                 kfree(info);
1009                 dev_set_drvdata(&dev->dev, NULL);
1010                 return err;
1011         }
1012
1013         return 0;
1014 }
1015
1016
1017 static int blkif_recover(struct blkfront_info *info)
1018 {
1019         int i;
1020         struct blkif_request *req;
1021         struct blk_shadow *copy;
1022         int j;
1023
1024         /* Stage 1: Make a safe copy of the shadow state. */
1025         copy = kmalloc(sizeof(info->shadow),
1026                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1027         if (!copy)
1028                 return -ENOMEM;
1029         memcpy(copy, info->shadow, sizeof(info->shadow));
1030
1031         /* Stage 2: Set up free list. */
1032         memset(&info->shadow, 0, sizeof(info->shadow));
1033         for (i = 0; i < BLK_RING_SIZE; i++)
1034                 info->shadow[i].req.u.rw.id = i+1;
1035         info->shadow_free = info->ring.req_prod_pvt;
1036         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1037
1038         /* Stage 3: Find pending requests and requeue them. */
1039         for (i = 0; i < BLK_RING_SIZE; i++) {
1040                 /* Not in use? */
1041                 if (!copy[i].request)
1042                         continue;
1043
1044                 /* Grab a request slot and copy shadow state into it. */
1045                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1046                 *req = copy[i].req;
1047
1048                 /* We get a new request id, and must reset the shadow state. */
1049                 req->u.rw.id = get_id_from_freelist(info);
1050                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1051
1052                 if (req->operation != BLKIF_OP_DISCARD) {
1053                 /* Rewrite any grant references invalidated by susp/resume. */
1054                         for (j = 0; j < req->u.rw.nr_segments; j++)
1055                                 gnttab_grant_foreign_access_ref(
1056                                         req->u.rw.seg[j].gref,
1057                                         info->xbdev->otherend_id,
1058                                         pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1059                                         rq_data_dir(info->shadow[req->u.rw.id].request));
1060                 }
1061                 info->shadow[req->u.rw.id].req = *req;
1062
1063                 info->ring.req_prod_pvt++;
1064         }
1065
1066         kfree(copy);
1067
1068         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1069
1070         spin_lock_irq(&info->io_lock);
1071
1072         /* Now safe for us to use the shared ring */
1073         info->connected = BLKIF_STATE_CONNECTED;
1074
1075         /* Send off requeued requests */
1076         flush_requests(info);
1077
1078         /* Kick any other new requests queued since we resumed */
1079         kick_pending_request_queues(info);
1080
1081         spin_unlock_irq(&info->io_lock);
1082
1083         return 0;
1084 }
1085
1086 /**
1087  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1088  * driver restart.  We tear down our blkif structure and recreate it, but
1089  * leave the device-layer structures intact so that this is transparent to the
1090  * rest of the kernel.
1091  */
1092 static int blkfront_resume(struct xenbus_device *dev)
1093 {
1094         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1095         int err;
1096
1097         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1098
1099         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1100
1101         err = talk_to_blkback(dev, info);
1102         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1103                 err = blkif_recover(info);
1104
1105         return err;
1106 }
1107
1108 static void
1109 blkfront_closing(struct blkfront_info *info)
1110 {
1111         struct xenbus_device *xbdev = info->xbdev;
1112         struct block_device *bdev = NULL;
1113
1114         mutex_lock(&info->mutex);
1115
1116         if (xbdev->state == XenbusStateClosing) {
1117                 mutex_unlock(&info->mutex);
1118                 return;
1119         }
1120
1121         if (info->gd)
1122                 bdev = bdget_disk(info->gd, 0);
1123
1124         mutex_unlock(&info->mutex);
1125
1126         if (!bdev) {
1127                 xenbus_frontend_closed(xbdev);
1128                 return;
1129         }
1130
1131         mutex_lock(&bdev->bd_mutex);
1132
1133         if (bdev->bd_openers) {
1134                 xenbus_dev_error(xbdev, -EBUSY,
1135                                  "Device in use; refusing to close");
1136                 xenbus_switch_state(xbdev, XenbusStateClosing);
1137         } else {
1138                 xlvbd_release_gendisk(info);
1139                 xenbus_frontend_closed(xbdev);
1140         }
1141
1142         mutex_unlock(&bdev->bd_mutex);
1143         bdput(bdev);
1144 }
1145
1146 static void blkfront_setup_discard(struct blkfront_info *info)
1147 {
1148         int err;
1149         char *type;
1150         unsigned int discard_granularity;
1151         unsigned int discard_alignment;
1152         unsigned int discard_secure;
1153
1154         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1155         if (IS_ERR(type))
1156                 return;
1157
1158         info->feature_secdiscard = 0;
1159         if (strncmp(type, "phy", 3) == 0) {
1160                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1161                         "discard-granularity", "%u", &discard_granularity,
1162                         "discard-alignment", "%u", &discard_alignment,
1163                         NULL);
1164                 if (!err) {
1165                         info->feature_discard = 1;
1166                         info->discard_granularity = discard_granularity;
1167                         info->discard_alignment = discard_alignment;
1168                 }
1169                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1170                             "discard-secure", "%d", &discard_secure,
1171                             NULL);
1172                 if (!err)
1173                         info->feature_secdiscard = discard_secure;
1174
1175         } else if (strncmp(type, "file", 4) == 0)
1176                 info->feature_discard = 1;
1177
1178         kfree(type);
1179 }
1180
1181 /*
1182  * Invoked when the backend is finally 'ready' (and has told produced
1183  * the details about the physical device - #sectors, size, etc).
1184  */
1185 static void blkfront_connect(struct blkfront_info *info)
1186 {
1187         unsigned long long sectors;
1188         unsigned long sector_size;
1189         unsigned int binfo;
1190         int err;
1191         int barrier, flush, discard;
1192
1193         switch (info->connected) {
1194         case BLKIF_STATE_CONNECTED:
1195                 /*
1196                  * Potentially, the back-end may be signalling
1197                  * a capacity change; update the capacity.
1198                  */
1199                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1200                                    "sectors", "%Lu", &sectors);
1201                 if (XENBUS_EXIST_ERR(err))
1202                         return;
1203                 printk(KERN_INFO "Setting capacity to %Lu\n",
1204                        sectors);
1205                 set_capacity(info->gd, sectors);
1206                 revalidate_disk(info->gd);
1207
1208                 /* fall through */
1209         case BLKIF_STATE_SUSPENDED:
1210                 return;
1211
1212         default:
1213                 break;
1214         }
1215
1216         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1217                 __func__, info->xbdev->otherend);
1218
1219         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1220                             "sectors", "%llu", &sectors,
1221                             "info", "%u", &binfo,
1222                             "sector-size", "%lu", &sector_size,
1223                             NULL);
1224         if (err) {
1225                 xenbus_dev_fatal(info->xbdev, err,
1226                                  "reading backend fields at %s",
1227                                  info->xbdev->otherend);
1228                 return;
1229         }
1230
1231         info->feature_flush = 0;
1232         info->flush_op = 0;
1233
1234         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1235                             "feature-barrier", "%d", &barrier,
1236                             NULL);
1237
1238         /*
1239          * If there's no "feature-barrier" defined, then it means
1240          * we're dealing with a very old backend which writes
1241          * synchronously; nothing to do.
1242          *
1243          * If there are barriers, then we use flush.
1244          */
1245         if (!err && barrier) {
1246                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1247                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1248         }
1249         /*
1250          * And if there is "feature-flush-cache" use that above
1251          * barriers.
1252          */
1253         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1254                             "feature-flush-cache", "%d", &flush,
1255                             NULL);
1256
1257         if (!err && flush) {
1258                 info->feature_flush = REQ_FLUSH;
1259                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1260         }
1261
1262         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1263                             "feature-discard", "%d", &discard,
1264                             NULL);
1265
1266         if (!err && discard)
1267                 blkfront_setup_discard(info);
1268
1269         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1270         if (err) {
1271                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1272                                  info->xbdev->otherend);
1273                 return;
1274         }
1275
1276         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1277
1278         /* Kick pending requests. */
1279         spin_lock_irq(&info->io_lock);
1280         info->connected = BLKIF_STATE_CONNECTED;
1281         kick_pending_request_queues(info);
1282         spin_unlock_irq(&info->io_lock);
1283
1284         add_disk(info->gd);
1285
1286         info->is_ready = 1;
1287 }
1288
1289 /**
1290  * Callback received when the backend's state changes.
1291  */
1292 static void blkback_changed(struct xenbus_device *dev,
1293                             enum xenbus_state backend_state)
1294 {
1295         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1296
1297         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1298
1299         switch (backend_state) {
1300         case XenbusStateInitialising:
1301         case XenbusStateInitWait:
1302         case XenbusStateInitialised:
1303         case XenbusStateReconfiguring:
1304         case XenbusStateReconfigured:
1305         case XenbusStateUnknown:
1306         case XenbusStateClosed:
1307                 break;
1308
1309         case XenbusStateConnected:
1310                 blkfront_connect(info);
1311                 break;
1312
1313         case XenbusStateClosing:
1314                 blkfront_closing(info);
1315                 break;
1316         }
1317 }
1318
1319 static int blkfront_remove(struct xenbus_device *xbdev)
1320 {
1321         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1322         struct block_device *bdev = NULL;
1323         struct gendisk *disk;
1324
1325         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1326
1327         blkif_free(info, 0);
1328
1329         mutex_lock(&info->mutex);
1330
1331         disk = info->gd;
1332         if (disk)
1333                 bdev = bdget_disk(disk, 0);
1334
1335         info->xbdev = NULL;
1336         mutex_unlock(&info->mutex);
1337
1338         if (!bdev) {
1339                 kfree(info);
1340                 return 0;
1341         }
1342
1343         /*
1344          * The xbdev was removed before we reached the Closed
1345          * state. See if it's safe to remove the disk. If the bdev
1346          * isn't closed yet, we let release take care of it.
1347          */
1348
1349         mutex_lock(&bdev->bd_mutex);
1350         info = disk->private_data;
1351
1352         dev_warn(disk_to_dev(disk),
1353                  "%s was hot-unplugged, %d stale handles\n",
1354                  xbdev->nodename, bdev->bd_openers);
1355
1356         if (info && !bdev->bd_openers) {
1357                 xlvbd_release_gendisk(info);
1358                 disk->private_data = NULL;
1359                 kfree(info);
1360         }
1361
1362         mutex_unlock(&bdev->bd_mutex);
1363         bdput(bdev);
1364
1365         return 0;
1366 }
1367
1368 static int blkfront_is_ready(struct xenbus_device *dev)
1369 {
1370         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1371
1372         return info->is_ready && info->xbdev;
1373 }
1374
1375 static int blkif_open(struct block_device *bdev, fmode_t mode)
1376 {
1377         struct gendisk *disk = bdev->bd_disk;
1378         struct blkfront_info *info;
1379         int err = 0;
1380
1381         mutex_lock(&blkfront_mutex);
1382
1383         info = disk->private_data;
1384         if (!info) {
1385                 /* xbdev gone */
1386                 err = -ERESTARTSYS;
1387                 goto out;
1388         }
1389
1390         mutex_lock(&info->mutex);
1391
1392         if (!info->gd)
1393                 /* xbdev is closed */
1394                 err = -ERESTARTSYS;
1395
1396         mutex_unlock(&info->mutex);
1397
1398 out:
1399         mutex_unlock(&blkfront_mutex);
1400         return err;
1401 }
1402
1403 static int blkif_release(struct gendisk *disk, fmode_t mode)
1404 {
1405         struct blkfront_info *info = disk->private_data;
1406         struct block_device *bdev;
1407         struct xenbus_device *xbdev;
1408
1409         mutex_lock(&blkfront_mutex);
1410
1411         bdev = bdget_disk(disk, 0);
1412
1413         if (bdev->bd_openers)
1414                 goto out;
1415
1416         /*
1417          * Check if we have been instructed to close. We will have
1418          * deferred this request, because the bdev was still open.
1419          */
1420
1421         mutex_lock(&info->mutex);
1422         xbdev = info->xbdev;
1423
1424         if (xbdev && xbdev->state == XenbusStateClosing) {
1425                 /* pending switch to state closed */
1426                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1427                 xlvbd_release_gendisk(info);
1428                 xenbus_frontend_closed(info->xbdev);
1429         }
1430
1431         mutex_unlock(&info->mutex);
1432
1433         if (!xbdev) {
1434                 /* sudden device removal */
1435                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1436                 xlvbd_release_gendisk(info);
1437                 disk->private_data = NULL;
1438                 kfree(info);
1439         }
1440
1441 out:
1442         bdput(bdev);
1443         mutex_unlock(&blkfront_mutex);
1444         return 0;
1445 }
1446
1447 static const struct block_device_operations xlvbd_block_fops =
1448 {
1449         .owner = THIS_MODULE,
1450         .open = blkif_open,
1451         .release = blkif_release,
1452         .getgeo = blkif_getgeo,
1453         .ioctl = blkif_ioctl,
1454 };
1455
1456
1457 static const struct xenbus_device_id blkfront_ids[] = {
1458         { "vbd" },
1459         { "" }
1460 };
1461
1462 static DEFINE_XENBUS_DRIVER(blkfront, ,
1463         .probe = blkfront_probe,
1464         .remove = blkfront_remove,
1465         .resume = blkfront_resume,
1466         .otherend_changed = blkback_changed,
1467         .is_ready = blkfront_is_ready,
1468 );
1469
1470 static int __init xlblk_init(void)
1471 {
1472         int ret;
1473
1474         if (!xen_domain())
1475                 return -ENODEV;
1476
1477         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1478                 return -ENODEV;
1479
1480         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1481                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1482                        XENVBD_MAJOR, DEV_NAME);
1483                 return -ENODEV;
1484         }
1485
1486         ret = xenbus_register_frontend(&blkfront_driver);
1487         if (ret) {
1488                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1489                 return ret;
1490         }
1491
1492         return 0;
1493 }
1494 module_init(xlblk_init);
1495
1496
1497 static void __exit xlblk_exit(void)
1498 {
1499         return xenbus_unregister_driver(&blkfront_driver);
1500 }
1501 module_exit(xlblk_exit);
1502
1503 MODULE_DESCRIPTION("Xen virtual block device frontend");
1504 MODULE_LICENSE("GPL");
1505 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1506 MODULE_ALIAS("xen:vbd");
1507 MODULE_ALIAS("xenblk");