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 int add_id_to_freelist(struct blkfront_info *info,
 145                                unsigned long id)
 146 {
 147         if (info->shadow[id].req.u.rw.id != id)
 148                 return -EINVAL;
 149         if (info->shadow[id].request == NULL)
 150                 return -EINVAL;
 151         info->shadow[id].req.u.rw.id  = info->shadow_free;
 152         info->shadow[id].request = NULL;
 153         info->shadow_free = id;
 154         return 0;
 155 }
 156
 157 static const char *op_name(int op)
 158 {
 159         static const char *const names[] = {
 160                 [BLKIF_OP_READ] = "read",
 161                 [BLKIF_OP_WRITE] = "write",
 162                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 163                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 164                 [BLKIF_OP_DISCARD] = "discard" };
 165
 166         if (op < 0 || op >= ARRAY_SIZE(names))
 167                 return "unknown";
 168
 169         if (!names[op])
 170                 return "reserved";
 171
 172         return names[op];
 173 }
 174 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 175 {
 176         unsigned int end = minor + nr;
 177         int rc;
 178
 179         if (end > nr_minors) {
 180                 unsigned long *bitmap, *old;
 181
 182                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 183                                  GFP_KERNEL);
 184                 if (bitmap == NULL)
 185                         return -ENOMEM;
 186
 187                 spin_lock(&minor_lock);
 188                 if (end > nr_minors) {
 189                         old = minors;
 190                         memcpy(bitmap, minors,
 191                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 192                         minors = bitmap;
 193                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 194                 } else
 195                         old = bitmap;
 196                 spin_unlock(&minor_lock);
 197                 kfree(old);
 198         }
 199
 200         spin_lock(&minor_lock);
 201         if (find_next_bit(minors, end, minor) >= end) {
 202                 bitmap_set(minors, minor, nr);
 203                 rc = 0;
 204         } else
 205                 rc = -EBUSY;
 206         spin_unlock(&minor_lock);
 207
 208         return rc;
 209 }
 210
 211 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 212 {
 213         unsigned int end = minor + nr;
 214
 215         BUG_ON(end > nr_minors);
 216         spin_lock(&minor_lock);
 217         bitmap_clear(minors,  minor, nr);
 218         spin_unlock(&minor_lock);
 219 }
 220
 221 static void blkif_restart_queue_callback(void *arg)
 222 {
 223         struct blkfront_info *info = (struct blkfront_info *)arg;
 224         schedule_work(&info->work);
 225 }
 226
 227 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 228 {
 229         /* We don't have real geometry info, but let's at least return
 230            values consistent with the size of the device */
 231         sector_t nsect = get_capacity(bd->bd_disk);
 232         sector_t cylinders = nsect;
 233
 234         hg->heads = 0xff;
 235         hg->sectors = 0x3f;
 236         sector_div(cylinders, hg->heads * hg->sectors);
 237         hg->cylinders = cylinders;
 238         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 239                 hg->cylinders = 0xffff;
 240         return 0;
 241 }
 242
 243 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 244                        unsigned command, unsigned long argument)
 245 {
 246         struct blkfront_info *info = bdev->bd_disk->private_data;
 247         int i;
 248
 249         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 250                 command, (long)argument);
 251
 252         switch (command) {
 253         case CDROMMULTISESSION:
 254                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 255                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 256                         if (put_user(0, (char __user *)(argument + i)))
 257                                 return -EFAULT;
 258                 return 0;
 259
 260         case CDROM_GET_CAPABILITY: {
 261                 struct gendisk *gd = info->gd;
 262                 if (gd->flags & GENHD_FL_CD)
 263                         return 0;
 264                 return -EINVAL;
 265         }
 266
 267         default:
 268                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 269                   command);*/
 270                 return -EINVAL; /* same return as native Linux */
 271         }
 272
 273         return 0;
 274 }
 275
 276 /*
 277  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 278  * and writes are handled as expected.
 279  *
 280  * @req: a request struct
 281  */
 282 static int blkif_queue_request(struct request *req)
 283 {
 284         struct blkfront_info *info = req->rq_disk->private_data;
 285         unsigned long buffer_mfn;
 286         struct blkif_request *ring_req;
 287         unsigned long id;
 288         unsigned int fsect, lsect;
 289         int i, ref;
 290         grant_ref_t gref_head;
 291         struct scatterlist *sg;
 292
 293         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 294                 return 1;
 295
 296         if (gnttab_alloc_grant_references(
 297                 BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
 298                 gnttab_request_free_callback(
 299                         &info->callback,
 300                         blkif_restart_queue_callback,
 301                         info,
 302                         BLKIF_MAX_SEGMENTS_PER_REQUEST);
 303                 return 1;
 304         }
 305
 306         /* Fill out a communications ring structure. */
 307         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 308         id = get_id_from_freelist(info);
 309         info->shadow[id].request = req;
 310
 311         ring_req->u.rw.id = id;
 312         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 313         ring_req->u.rw.handle = info->handle;
 314
 315         ring_req->operation = rq_data_dir(req) ?
 316                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 317
 318         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 319                 /*
 320                  * Ideally we can do an unordered flush-to-disk. In case the
 321                  * backend onlysupports barriers, use that. A barrier request
 322                  * a superset of FUA, so we can implement it the same
 323                  * way.  (It's also a FLUSH+FUA, since it is
 324                  * guaranteed ordered WRT previous writes.)
 325                  */
 326                 ring_req->operation = info->flush_op;
 327         }
 328
 329         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 330                 /* id, sector_number and handle are set above. */
 331                 ring_req->operation = BLKIF_OP_DISCARD;
 332                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 333                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 334                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 335                 else
 336                         ring_req->u.discard.flag = 0;
 337         } else {
 338                 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
 339                                                            info->sg);
 340                 BUG_ON(ring_req->u.rw.nr_segments >
 341                        BLKIF_MAX_SEGMENTS_PER_REQUEST);
 342
 343                 for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
 344                         buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 345                         fsect = sg->offset >> 9;
 346                         lsect = fsect + (sg->length >> 9) - 1;
 347                         /* install a grant reference. */
 348                         ref = gnttab_claim_grant_reference(&gref_head);
 349                         BUG_ON(ref == -ENOSPC);
 350
 351                         gnttab_grant_foreign_access_ref(
 352                                         ref,
 353                                         info->xbdev->otherend_id,
 354                                         buffer_mfn,
 355                                         rq_data_dir(req));
 356
 357                         info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 358                         ring_req->u.rw.seg[i] =
 359                                         (struct blkif_request_segment) {
 360                                                 .gref       = ref,
 361                                                 .first_sect = fsect,
 362                                                 .last_sect  = lsect };
 363                 }
 364         }
 365
 366         info->ring.req_prod_pvt++;
 367
 368         /* Keep a private copy so we can reissue requests when recovering. */
 369         info->shadow[id].req = *ring_req;
 370
 371         gnttab_free_grant_references(gref_head);
 372
 373         return 0;
 374 }
 375
 376
 377 static inline void flush_requests(struct blkfront_info *info)
 378 {
 379         int notify;
 380
 381         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 382
 383         if (notify)
 384                 notify_remote_via_irq(info->irq);
 385 }
 386
 387 /*
 388  * do_blkif_request
 389  *  read a block; request is in a request queue
 390  */
 391 static void do_blkif_request(struct request_queue *rq)
 392 {
 393         struct blkfront_info *info = NULL;
 394         struct request *req;
 395         int queued;
 396
 397         pr_debug("Entered do_blkif_request\n");
 398
 399         queued = 0;
 400
 401         while ((req = blk_peek_request(rq)) != NULL) {
 402                 info = req->rq_disk->private_data;
 403
 404                 if (RING_FULL(&info->ring))
 405                         goto wait;
 406
 407                 blk_start_request(req);
 408
 409                 if ((req->cmd_type != REQ_TYPE_FS) ||
 410                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 411                     !info->flush_op)) {
 412                         __blk_end_request_all(req, -EIO);
 413                         continue;
 414                 }
 415
 416                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 417                          "(%u/%u) buffer:%p [%s]\n",
 418                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 419                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 420                          req->buffer, rq_data_dir(req) ? "write" : "read");
 421
 422                 if (blkif_queue_request(req)) {
 423                         blk_requeue_request(rq, req);
 424 wait:
 425                         /* Avoid pointless unplugs. */
 426                         blk_stop_queue(rq);
 427                         break;
 428                 }
 429
 430                 queued++;
 431         }
 432
 433         if (queued != 0)
 434                 flush_requests(info);
 435 }
 436
 437 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 438 {
 439         struct request_queue *rq;
 440         struct blkfront_info *info = gd->private_data;
 441
 442         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 443         if (rq == NULL)
 444                 return -1;
 445
 446         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 447
 448         if (info->feature_discard) {
 449                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 450                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 451                 rq->limits.discard_granularity = info->discard_granularity;
 452                 rq->limits.discard_alignment = info->discard_alignment;
 453                 if (info->feature_secdiscard)
 454                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 455         }
 456
 457         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 458         blk_queue_logical_block_size(rq, sector_size);
 459         blk_queue_max_hw_sectors(rq, 512);
 460
 461         /* Each segment in a request is up to an aligned page in size. */
 462         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 463         blk_queue_max_segment_size(rq, PAGE_SIZE);
 464
 465         /* Ensure a merged request will fit in a single I/O ring slot. */
 466         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 467
 468         /* Make sure buffer addresses are sector-aligned. */
 469         blk_queue_dma_alignment(rq, 511);
 470
 471         /* Make sure we don't use bounce buffers. */
 472         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 473
 474         gd->queue = rq;
 475
 476         return 0;
 477 }
 478
 479
 480 static void xlvbd_flush(struct blkfront_info *info)
 481 {
 482         blk_queue_flush(info->rq, info->feature_flush);
 483         printk(KERN_INFO "blkfront: %s: %s: %s\n",
 484                info->gd->disk_name,
 485                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 486                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 487                 "flush diskcache" : "barrier or flush"),
 488                info->feature_flush ? "enabled" : "disabled");
 489 }
 490
 491 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 492 {
 493         int major;
 494         major = BLKIF_MAJOR(vdevice);
 495         *minor = BLKIF_MINOR(vdevice);
 496         switch (major) {
 497                 case XEN_IDE0_MAJOR:
 498                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 499                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 500                                 EMULATED_HD_DISK_MINOR_OFFSET;
 501                         break;
 502                 case XEN_IDE1_MAJOR:
 503                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 504                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 505                                 EMULATED_HD_DISK_MINOR_OFFSET;
 506                         break;
 507                 case XEN_SCSI_DISK0_MAJOR:
 508                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 509                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 510                         break;
 511                 case XEN_SCSI_DISK1_MAJOR:
 512                 case XEN_SCSI_DISK2_MAJOR:
 513                 case XEN_SCSI_DISK3_MAJOR:
 514                 case XEN_SCSI_DISK4_MAJOR:
 515                 case XEN_SCSI_DISK5_MAJOR:
 516                 case XEN_SCSI_DISK6_MAJOR:
 517                 case XEN_SCSI_DISK7_MAJOR:
 518                         *offset = (*minor / PARTS_PER_DISK) +
 519                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 520                                 EMULATED_SD_DISK_NAME_OFFSET;
 521                         *minor = *minor +
 522                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 523                                 EMULATED_SD_DISK_MINOR_OFFSET;
 524                         break;
 525                 case XEN_SCSI_DISK8_MAJOR:
 526                 case XEN_SCSI_DISK9_MAJOR:
 527                 case XEN_SCSI_DISK10_MAJOR:
 528                 case XEN_SCSI_DISK11_MAJOR:
 529                 case XEN_SCSI_DISK12_MAJOR:
 530                 case XEN_SCSI_DISK13_MAJOR:
 531                 case XEN_SCSI_DISK14_MAJOR:
 532                 case XEN_SCSI_DISK15_MAJOR:
 533                         *offset = (*minor / PARTS_PER_DISK) +
 534                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 535                                 EMULATED_SD_DISK_NAME_OFFSET;
 536                         *minor = *minor +
 537                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 538                                 EMULATED_SD_DISK_MINOR_OFFSET;
 539                         break;
 540                 case XENVBD_MAJOR:
 541                         *offset = *minor / PARTS_PER_DISK;
 542                         break;
 543                 default:
 544                         printk(KERN_WARNING "blkfront: your disk configuration is "
 545                                         "incorrect, please use an xvd device instead\n");
 546                         return -ENODEV;
 547         }
 548         return 0;
 549 }
 550
 551 static char *encode_disk_name(char *ptr, unsigned int n)
 552 {
 553         if (n >= 26)
 554                 ptr = encode_disk_name(ptr, n / 26 - 1);
 555         *ptr = 'a' + n % 26;
 556         return ptr + 1;
 557 }
 558
 559 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 560                                struct blkfront_info *info,
 561                                u16 vdisk_info, u16 sector_size)
 562 {
 563         struct gendisk *gd;
 564         int nr_minors = 1;
 565         int err;
 566         unsigned int offset;
 567         int minor;
 568         int nr_parts;
 569         char *ptr;
 570
 571         BUG_ON(info->gd != NULL);
 572         BUG_ON(info->rq != NULL);
 573
 574         if ((info->vdevice>>EXT_SHIFT) > 1) {
 575                 /* this is above the extended range; something is wrong */
 576                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 577                 return -ENODEV;
 578         }
 579
 580         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 581                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 582                 if (err)
 583                         return err;
 584                 nr_parts = PARTS_PER_DISK;
 585         } else {
 586                 minor = BLKIF_MINOR_EXT(info->vdevice);
 587                 nr_parts = PARTS_PER_EXT_DISK;
 588                 offset = minor / nr_parts;
 589                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 590                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 591                                         "emulated IDE disks,\n\t choose an xvd device name"
 592                                         "from xvde on\n", info->vdevice);
 593         }
 594         if (minor >> MINORBITS) {
 595                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 596                         info->vdevice, minor);
 597                 return -ENODEV;
 598         }
 599
 600         if ((minor % nr_parts) == 0)
 601                 nr_minors = nr_parts;
 602
 603         err = xlbd_reserve_minors(minor, nr_minors);
 604         if (err)
 605                 goto out;
 606         err = -ENODEV;
 607
 608         gd = alloc_disk(nr_minors);
 609         if (gd == NULL)
 610                 goto release;
 611
 612         strcpy(gd->disk_name, DEV_NAME);
 613         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 614         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 615         if (nr_minors > 1)
 616                 *ptr = 0;
 617         else
 618                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 619                          "%d", minor & (nr_parts - 1));
 620
 621         gd->major = XENVBD_MAJOR;
 622         gd->first_minor = minor;
 623         gd->fops = &xlvbd_block_fops;
 624         gd->private_data = info;
 625         gd->driverfs_dev = &(info->xbdev->dev);
 626         set_capacity(gd, capacity);
 627
 628         if (xlvbd_init_blk_queue(gd, sector_size)) {
 629                 del_gendisk(gd);
 630                 goto release;
 631         }
 632
 633         info->rq = gd->queue;
 634         info->gd = gd;
 635
 636         xlvbd_flush(info);
 637
 638         if (vdisk_info & VDISK_READONLY)
 639                 set_disk_ro(gd, 1);
 640
 641         if (vdisk_info & VDISK_REMOVABLE)
 642                 gd->flags |= GENHD_FL_REMOVABLE;
 643
 644         if (vdisk_info & VDISK_CDROM)
 645                 gd->flags |= GENHD_FL_CD;
 646
 647         return 0;
 648
 649  release:
 650         xlbd_release_minors(minor, nr_minors);
 651  out:
 652         return err;
 653 }
 654
 655 static void xlvbd_release_gendisk(struct blkfront_info *info)
 656 {
 657         unsigned int minor, nr_minors;
 658         unsigned long flags;
 659
 660         if (info->rq == NULL)
 661                 return;
 662
 663         spin_lock_irqsave(&info->io_lock, flags);
 664
 665         /* No more blkif_request(). */
 666         blk_stop_queue(info->rq);
 667
 668         /* No more gnttab callback work. */
 669         gnttab_cancel_free_callback(&info->callback);
 670         spin_unlock_irqrestore(&info->io_lock, flags);
 671
 672         /* Flush gnttab callback work. Must be done with no locks held. */
 673         flush_work(&info->work);
 674
 675         del_gendisk(info->gd);
 676
 677         minor = info->gd->first_minor;
 678         nr_minors = info->gd->minors;
 679         xlbd_release_minors(minor, nr_minors);
 680
 681         blk_cleanup_queue(info->rq);
 682         info->rq = NULL;
 683
 684         put_disk(info->gd);
 685         info->gd = NULL;
 686 }
 687
 688 static void kick_pending_request_queues(struct blkfront_info *info)
 689 {
 690         if (!RING_FULL(&info->ring)) {
 691                 /* Re-enable calldowns. */
 692                 blk_start_queue(info->rq);
 693                 /* Kick things off immediately. */
 694                 do_blkif_request(info->rq);
 695         }
 696 }
 697
 698 static void blkif_restart_queue(struct work_struct *work)
 699 {
 700         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 701
 702         spin_lock_irq(&info->io_lock);
 703         if (info->connected == BLKIF_STATE_CONNECTED)
 704                 kick_pending_request_queues(info);
 705         spin_unlock_irq(&info->io_lock);
 706 }
 707
 708 static void blkif_free(struct blkfront_info *info, int suspend)
 709 {
 710         /* Prevent new requests being issued until we fix things up. */
 711         spin_lock_irq(&info->io_lock);
 712         info->connected = suspend ?
 713                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 714         /* No more blkif_request(). */
 715         if (info->rq)
 716                 blk_stop_queue(info->rq);
 717         /* No more gnttab callback work. */
 718         gnttab_cancel_free_callback(&info->callback);
 719         spin_unlock_irq(&info->io_lock);
 720
 721         /* Flush gnttab callback work. Must be done with no locks held. */
 722         flush_work(&info->work);
 723
 724         /* Free resources associated with old device channel. */
 725         if (info->ring_ref != GRANT_INVALID_REF) {
 726                 gnttab_end_foreign_access(info->ring_ref, 0,
 727                                           (unsigned long)info->ring.sring);
 728                 info->ring_ref = GRANT_INVALID_REF;
 729                 info->ring.sring = NULL;
 730         }
 731         if (info->irq)
 732                 unbind_from_irqhandler(info->irq, info);
 733         info->evtchn = info->irq = 0;
 734
 735 }
 736
 737 static void blkif_completion(struct blk_shadow *s)
 738 {
 739         int i;
 740         /* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
 741          * flag. */
 742         for (i = 0; i < s->req.u.rw.nr_segments; i++)
 743                 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
 744 }
 745
 746 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 747 {
 748         struct request *req;
 749         struct blkif_response *bret;
 750         RING_IDX i, rp;
 751         unsigned long flags;
 752         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 753         int error;
 754
 755         spin_lock_irqsave(&info->io_lock, flags);
 756
 757         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 758                 spin_unlock_irqrestore(&info->io_lock, flags);
 759                 return IRQ_HANDLED;
 760         }
 761
 762  again:
 763         rp = info->ring.sring->rsp_prod;
 764         rmb(); /* Ensure we see queued responses up to 'rp'. */
 765
 766         for (i = info->ring.rsp_cons; i != rp; i++) {
 767                 unsigned long id;
 768
 769                 bret = RING_GET_RESPONSE(&info->ring, i);
 770                 id   = bret->id;
 771                 /*
 772                  * The backend has messed up and given us an id that we would
 773                  * never have given to it (we stamp it up to BLK_RING_SIZE -
 774                  * look in get_id_from_freelist.
 775                  */
 776                 if (id >= BLK_RING_SIZE) {
 777                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
 778                              info->gd->disk_name, op_name(bret->operation), id);
 779                         /* We can't safely get the 'struct request' as
 780                          * the id is busted. */
 781                         continue;
 782                 }
 783                 req  = info->shadow[id].request;
 784
 785                 if (bret->operation != BLKIF_OP_DISCARD)
 786                         blkif_completion(&info->shadow[id]);
 787
 788                 if (add_id_to_freelist(info, id)) {
 789                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
 790                              info->gd->disk_name, op_name(bret->operation), id);
 791                         continue;
 792                 }
 793
 794                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 795                 switch (bret->operation) {
 796                 case BLKIF_OP_DISCARD:
 797                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 798                                 struct request_queue *rq = info->rq;
 799                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 800                                            info->gd->disk_name, op_name(bret->operation));
 801                                 error = -EOPNOTSUPP;
 802                                 info->feature_discard = 0;
 803                                 info->feature_secdiscard = 0;
 804                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 805                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 806                         }
 807                         __blk_end_request_all(req, error);
 808                         break;
 809                 case BLKIF_OP_FLUSH_DISKCACHE:
 810                 case BLKIF_OP_WRITE_BARRIER:
 811                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 812                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 813                                        info->gd->disk_name, op_name(bret->operation));
 814                                 error = -EOPNOTSUPP;
 815                         }
 816                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 817                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
 818                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
 819                                        info->gd->disk_name, op_name(bret->operation));
 820                                 error = -EOPNOTSUPP;
 821                         }
 822                         if (unlikely(error)) {
 823                                 if (error == -EOPNOTSUPP)
 824                                         error = 0;
 825                                 info->feature_flush = 0;
 826                                 info->flush_op = 0;
 827                                 xlvbd_flush(info);
 828                         }
 829                         /* fall through */
 830                 case BLKIF_OP_READ:
 831                 case BLKIF_OP_WRITE:
 832                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 833                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 834                                         "request: %x\n", bret->status);
 835
 836                         __blk_end_request_all(req, error);
 837                         break;
 838                 default:
 839                         BUG();
 840                 }
 841         }
 842
 843         info->ring.rsp_cons = i;
 844
 845         if (i != info->ring.req_prod_pvt) {
 846                 int more_to_do;
 847                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 848                 if (more_to_do)
 849                         goto again;
 850         } else
 851                 info->ring.sring->rsp_event = i + 1;
 852
 853         kick_pending_request_queues(info);
 854
 855         spin_unlock_irqrestore(&info->io_lock, flags);
 856
 857         return IRQ_HANDLED;
 858 }
 859
 860
 861 static int setup_blkring(struct xenbus_device *dev,
 862                          struct blkfront_info *info)
 863 {
 864         struct blkif_sring *sring;
 865         int err;
 866
 867         info->ring_ref = GRANT_INVALID_REF;
 868
 869         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 870         if (!sring) {
 871                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 872                 return -ENOMEM;
 873         }
 874         SHARED_RING_INIT(sring);
 875         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 876
 877         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 878
 879         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 880         if (err < 0) {
 881                 free_page((unsigned long)sring);
 882                 info->ring.sring = NULL;
 883                 goto fail;
 884         }
 885         info->ring_ref = err;
 886
 887         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 888         if (err)
 889                 goto fail;
 890
 891         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
 892                                         "blkif", info);
 893         if (err <= 0) {
 894                 xenbus_dev_fatal(dev, err,
 895                                  "bind_evtchn_to_irqhandler failed");
 896                 goto fail;
 897         }
 898         info->irq = err;
 899
 900         return 0;
 901 fail:
 902         blkif_free(info, 0);
 903         return err;
 904 }
 905
 906
 907 /* Common code used when first setting up, and when resuming. */
 908 static int talk_to_blkback(struct xenbus_device *dev,
 909                            struct blkfront_info *info)
 910 {
 911         const char *message = NULL;
 912         struct xenbus_transaction xbt;
 913         int err;
 914
 915         /* Create shared ring, alloc event channel. */
 916         err = setup_blkring(dev, info);
 917         if (err)
 918                 goto out;
 919
 920 again:
 921         err = xenbus_transaction_start(&xbt);
 922         if (err) {
 923                 xenbus_dev_fatal(dev, err, "starting transaction");
 924                 goto destroy_blkring;
 925         }
 926
 927         err = xenbus_printf(xbt, dev->nodename,
 928                             "ring-ref", "%u", info->ring_ref);
 929         if (err) {
 930                 message = "writing ring-ref";
 931                 goto abort_transaction;
 932         }
 933         err = xenbus_printf(xbt, dev->nodename,
 934                             "event-channel", "%u", info->evtchn);
 935         if (err) {
 936                 message = "writing event-channel";
 937                 goto abort_transaction;
 938         }
 939         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 940                             XEN_IO_PROTO_ABI_NATIVE);
 941         if (err) {
 942                 message = "writing protocol";
 943                 goto abort_transaction;
 944         }
 945
 946         err = xenbus_transaction_end(xbt, 0);
 947         if (err) {
 948                 if (err == -EAGAIN)
 949                         goto again;
 950                 xenbus_dev_fatal(dev, err, "completing transaction");
 951                 goto destroy_blkring;
 952         }
 953
 954         xenbus_switch_state(dev, XenbusStateInitialised);
 955
 956         return 0;
 957
 958  abort_transaction:
 959         xenbus_transaction_end(xbt, 1);
 960         if (message)
 961                 xenbus_dev_fatal(dev, err, "%s", message);
 962  destroy_blkring:
 963         blkif_free(info, 0);
 964  out:
 965         return err;
 966 }
 967
 968 /**
 969  * Entry point to this code when a new device is created.  Allocate the basic
 970  * structures and the ring buffer for communication with the backend, and
 971  * inform the backend of the appropriate details for those.  Switch to
 972  * Initialised state.
 973  */
 974 static int blkfront_probe(struct xenbus_device *dev,
 975                           const struct xenbus_device_id *id)
 976 {
 977         int err, vdevice, i;
 978         struct blkfront_info *info;
 979
 980         /* FIXME: Use dynamic device id if this is not set. */
 981         err = xenbus_scanf(XBT_NIL, dev->nodename,
 982                            "virtual-device", "%i", &vdevice);
 983         if (err != 1) {
 984                 /* go looking in the extended area instead */
 985                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 986                                    "%i", &vdevice);
 987                 if (err != 1) {
 988                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 989                         return err;
 990                 }
 991         }
 992
 993         if (xen_hvm_domain()) {
 994                 char *type;
 995                 int len;
 996                 /* no unplug has been done: do not hook devices != xen vbds */
 997                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
 998                         int major;
 999
1000                         if (!VDEV_IS_EXTENDED(vdevice))
1001                                 major = BLKIF_MAJOR(vdevice);
1002                         else
1003                                 major = XENVBD_MAJOR;
1004
1005                         if (major != XENVBD_MAJOR) {
1006                                 printk(KERN_INFO
1007                                                 "%s: HVM does not support vbd %d as xen block device\n",
1008                                                 __FUNCTION__, vdevice);
1009                                 return -ENODEV;
1010                         }
1011                 }
1012                 /* do not create a PV cdrom device if we are an HVM guest */
1013                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1014                 if (IS_ERR(type))
1015                         return -ENODEV;
1016                 if (strncmp(type, "cdrom", 5) == 0) {
1017                         kfree(type);
1018                         return -ENODEV;
1019                 }
1020                 kfree(type);
1021         }
1022         info = kzalloc(sizeof(*info), GFP_KERNEL);
1023         if (!info) {
1024                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1025                 return -ENOMEM;
1026         }
1027
1028         mutex_init(&info->mutex);
1029         spin_lock_init(&info->io_lock);
1030         info->xbdev = dev;
1031         info->vdevice = vdevice;
1032         info->connected = BLKIF_STATE_DISCONNECTED;
1033         INIT_WORK(&info->work, blkif_restart_queue);
1034
1035         for (i = 0; i < BLK_RING_SIZE; i++)
1036                 info->shadow[i].req.u.rw.id = i+1;
1037         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1038
1039         /* Front end dir is a number, which is used as the id. */
1040         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1041         dev_set_drvdata(&dev->dev, info);
1042
1043         err = talk_to_blkback(dev, info);
1044         if (err) {
1045                 kfree(info);
1046                 dev_set_drvdata(&dev->dev, NULL);
1047                 return err;
1048         }
1049
1050         return 0;
1051 }
1052
1053
1054 static int blkif_recover(struct blkfront_info *info)
1055 {
1056         int i;
1057         struct blkif_request *req;
1058         struct blk_shadow *copy;
1059         int j;
1060
1061         /* Stage 1: Make a safe copy of the shadow state. */
1062         copy = kmalloc(sizeof(info->shadow),
1063                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1064         if (!copy)
1065                 return -ENOMEM;
1066         memcpy(copy, info->shadow, sizeof(info->shadow));
1067
1068         /* Stage 2: Set up free list. */
1069         memset(&info->shadow, 0, sizeof(info->shadow));
1070         for (i = 0; i < BLK_RING_SIZE; i++)
1071                 info->shadow[i].req.u.rw.id = i+1;
1072         info->shadow_free = info->ring.req_prod_pvt;
1073         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1074
1075         /* Stage 3: Find pending requests and requeue them. */
1076         for (i = 0; i < BLK_RING_SIZE; i++) {
1077                 /* Not in use? */
1078                 if (!copy[i].request)
1079                         continue;
1080
1081                 /* Grab a request slot and copy shadow state into it. */
1082                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1083                 *req = copy[i].req;
1084
1085                 /* We get a new request id, and must reset the shadow state. */
1086                 req->u.rw.id = get_id_from_freelist(info);
1087                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1088
1089                 if (req->operation != BLKIF_OP_DISCARD) {
1090                 /* Rewrite any grant references invalidated by susp/resume. */
1091                         for (j = 0; j < req->u.rw.nr_segments; j++)
1092                                 gnttab_grant_foreign_access_ref(
1093                                         req->u.rw.seg[j].gref,
1094                                         info->xbdev->otherend_id,
1095                                         pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1096                                         rq_data_dir(info->shadow[req->u.rw.id].request));
1097                 }
1098                 info->shadow[req->u.rw.id].req = *req;
1099
1100                 info->ring.req_prod_pvt++;
1101         }
1102
1103         kfree(copy);
1104
1105         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1106
1107         spin_lock_irq(&info->io_lock);
1108
1109         /* Now safe for us to use the shared ring */
1110         info->connected = BLKIF_STATE_CONNECTED;
1111
1112         /* Send off requeued requests */
1113         flush_requests(info);
1114
1115         /* Kick any other new requests queued since we resumed */
1116         kick_pending_request_queues(info);
1117
1118         spin_unlock_irq(&info->io_lock);
1119
1120         return 0;
1121 }
1122
1123 /**
1124  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1125  * driver restart.  We tear down our blkif structure and recreate it, but
1126  * leave the device-layer structures intact so that this is transparent to the
1127  * rest of the kernel.
1128  */
1129 static int blkfront_resume(struct xenbus_device *dev)
1130 {
1131         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1132         int err;
1133
1134         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1135
1136         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1137
1138         err = talk_to_blkback(dev, info);
1139         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1140                 err = blkif_recover(info);
1141
1142         return err;
1143 }
1144
1145 static void
1146 blkfront_closing(struct blkfront_info *info)
1147 {
1148         struct xenbus_device *xbdev = info->xbdev;
1149         struct block_device *bdev = NULL;
1150
1151         mutex_lock(&info->mutex);
1152
1153         if (xbdev->state == XenbusStateClosing) {
1154                 mutex_unlock(&info->mutex);
1155                 return;
1156         }
1157
1158         if (info->gd)
1159                 bdev = bdget_disk(info->gd, 0);
1160
1161         mutex_unlock(&info->mutex);
1162
1163         if (!bdev) {
1164                 xenbus_frontend_closed(xbdev);
1165                 return;
1166         }
1167
1168         mutex_lock(&bdev->bd_mutex);
1169
1170         if (bdev->bd_openers) {
1171                 xenbus_dev_error(xbdev, -EBUSY,
1172                                  "Device in use; refusing to close");
1173                 xenbus_switch_state(xbdev, XenbusStateClosing);
1174         } else {
1175                 xlvbd_release_gendisk(info);
1176                 xenbus_frontend_closed(xbdev);
1177         }
1178
1179         mutex_unlock(&bdev->bd_mutex);
1180         bdput(bdev);
1181 }
1182
1183 static void blkfront_setup_discard(struct blkfront_info *info)
1184 {
1185         int err;
1186         char *type;
1187         unsigned int discard_granularity;
1188         unsigned int discard_alignment;
1189         unsigned int discard_secure;
1190
1191         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1192         if (IS_ERR(type))
1193                 return;
1194
1195         info->feature_secdiscard = 0;
1196         if (strncmp(type, "phy", 3) == 0) {
1197                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1198                         "discard-granularity", "%u", &discard_granularity,
1199                         "discard-alignment", "%u", &discard_alignment,
1200                         NULL);
1201                 if (!err) {
1202                         info->feature_discard = 1;
1203                         info->discard_granularity = discard_granularity;
1204                         info->discard_alignment = discard_alignment;
1205                 }
1206                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1207                             "discard-secure", "%d", &discard_secure,
1208                             NULL);
1209                 if (!err)
1210                         info->feature_secdiscard = discard_secure;
1211
1212         } else if (strncmp(type, "file", 4) == 0)
1213                 info->feature_discard = 1;
1214
1215         kfree(type);
1216 }
1217
1218 /*
1219  * Invoked when the backend is finally 'ready' (and has told produced
1220  * the details about the physical device - #sectors, size, etc).
1221  */
1222 static void blkfront_connect(struct blkfront_info *info)
1223 {
1224         unsigned long long sectors;
1225         unsigned long sector_size;
1226         unsigned int binfo;
1227         int err;
1228         int barrier, flush, discard;
1229
1230         switch (info->connected) {
1231         case BLKIF_STATE_CONNECTED:
1232                 /*
1233                  * Potentially, the back-end may be signalling
1234                  * a capacity change; update the capacity.
1235                  */
1236                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1237                                    "sectors", "%Lu", &sectors);
1238                 if (XENBUS_EXIST_ERR(err))
1239                         return;
1240                 printk(KERN_INFO "Setting capacity to %Lu\n",
1241                        sectors);
1242                 set_capacity(info->gd, sectors);
1243                 revalidate_disk(info->gd);
1244
1245                 /* fall through */
1246         case BLKIF_STATE_SUSPENDED:
1247                 return;
1248
1249         default:
1250                 break;
1251         }
1252
1253         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1254                 __func__, info->xbdev->otherend);
1255
1256         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1257                             "sectors", "%llu", &sectors,
1258                             "info", "%u", &binfo,
1259                             "sector-size", "%lu", &sector_size,
1260                             NULL);
1261         if (err) {
1262                 xenbus_dev_fatal(info->xbdev, err,
1263                                  "reading backend fields at %s",
1264                                  info->xbdev->otherend);
1265                 return;
1266         }
1267
1268         info->feature_flush = 0;
1269         info->flush_op = 0;
1270
1271         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1272                             "feature-barrier", "%d", &barrier,
1273                             NULL);
1274
1275         /*
1276          * If there's no "feature-barrier" defined, then it means
1277          * we're dealing with a very old backend which writes
1278          * synchronously; nothing to do.
1279          *
1280          * If there are barriers, then we use flush.
1281          */
1282         if (!err && barrier) {
1283                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1284                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1285         }
1286         /*
1287          * And if there is "feature-flush-cache" use that above
1288          * barriers.
1289          */
1290         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1291                             "feature-flush-cache", "%d", &flush,
1292                             NULL);
1293
1294         if (!err && flush) {
1295                 info->feature_flush = REQ_FLUSH;
1296                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1297         }
1298
1299         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1300                             "feature-discard", "%d", &discard,
1301                             NULL);
1302
1303         if (!err && discard)
1304                 blkfront_setup_discard(info);
1305
1306         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1307         if (err) {
1308                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1309                                  info->xbdev->otherend);
1310                 return;
1311         }
1312
1313         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1314
1315         /* Kick pending requests. */
1316         spin_lock_irq(&info->io_lock);
1317         info->connected = BLKIF_STATE_CONNECTED;
1318         kick_pending_request_queues(info);
1319         spin_unlock_irq(&info->io_lock);
1320
1321         add_disk(info->gd);
1322
1323         info->is_ready = 1;
1324 }
1325
1326 /**
1327  * Callback received when the backend's state changes.
1328  */
1329 static void blkback_changed(struct xenbus_device *dev,
1330                             enum xenbus_state backend_state)
1331 {
1332         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1333
1334         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1335
1336         switch (backend_state) {
1337         case XenbusStateInitialising:
1338         case XenbusStateInitWait:
1339         case XenbusStateInitialised:
1340         case XenbusStateReconfiguring:
1341         case XenbusStateReconfigured:
1342         case XenbusStateUnknown:
1343         case XenbusStateClosed:
1344                 break;
1345
1346         case XenbusStateConnected:
1347                 blkfront_connect(info);
1348                 break;
1349
1350         case XenbusStateClosing:
1351                 blkfront_closing(info);
1352                 break;
1353         }
1354 }
1355
1356 static int blkfront_remove(struct xenbus_device *xbdev)
1357 {
1358         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1359         struct block_device *bdev = NULL;
1360         struct gendisk *disk;
1361
1362         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1363
1364         blkif_free(info, 0);
1365
1366         mutex_lock(&info->mutex);
1367
1368         disk = info->gd;
1369         if (disk)
1370                 bdev = bdget_disk(disk, 0);
1371
1372         info->xbdev = NULL;
1373         mutex_unlock(&info->mutex);
1374
1375         if (!bdev) {
1376                 kfree(info);
1377                 return 0;
1378         }
1379
1380         /*
1381          * The xbdev was removed before we reached the Closed
1382          * state. See if it's safe to remove the disk. If the bdev
1383          * isn't closed yet, we let release take care of it.
1384          */
1385
1386         mutex_lock(&bdev->bd_mutex);
1387         info = disk->private_data;
1388
1389         dev_warn(disk_to_dev(disk),
1390                  "%s was hot-unplugged, %d stale handles\n",
1391                  xbdev->nodename, bdev->bd_openers);
1392
1393         if (info && !bdev->bd_openers) {
1394                 xlvbd_release_gendisk(info);
1395                 disk->private_data = NULL;
1396                 kfree(info);
1397         }
1398
1399         mutex_unlock(&bdev->bd_mutex);
1400         bdput(bdev);
1401
1402         return 0;
1403 }
1404
1405 static int blkfront_is_ready(struct xenbus_device *dev)
1406 {
1407         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1408
1409         return info->is_ready && info->xbdev;
1410 }
1411
1412 static int blkif_open(struct block_device *bdev, fmode_t mode)
1413 {
1414         struct gendisk *disk = bdev->bd_disk;
1415         struct blkfront_info *info;
1416         int err = 0;
1417
1418         mutex_lock(&blkfront_mutex);
1419
1420         info = disk->private_data;
1421         if (!info) {
1422                 /* xbdev gone */
1423                 err = -ERESTARTSYS;
1424                 goto out;
1425         }
1426
1427         mutex_lock(&info->mutex);
1428
1429         if (!info->gd)
1430                 /* xbdev is closed */
1431                 err = -ERESTARTSYS;
1432
1433         mutex_unlock(&info->mutex);
1434
1435 out:
1436         mutex_unlock(&blkfront_mutex);
1437         return err;
1438 }
1439
1440 static int blkif_release(struct gendisk *disk, fmode_t mode)
1441 {
1442         struct blkfront_info *info = disk->private_data;
1443         struct block_device *bdev;
1444         struct xenbus_device *xbdev;
1445
1446         mutex_lock(&blkfront_mutex);
1447
1448         bdev = bdget_disk(disk, 0);
1449
1450         if (bdev->bd_openers)
1451                 goto out;
1452
1453         /*
1454          * Check if we have been instructed to close. We will have
1455          * deferred this request, because the bdev was still open.
1456          */
1457
1458         mutex_lock(&info->mutex);
1459         xbdev = info->xbdev;
1460
1461         if (xbdev && xbdev->state == XenbusStateClosing) {
1462                 /* pending switch to state closed */
1463                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1464                 xlvbd_release_gendisk(info);
1465                 xenbus_frontend_closed(info->xbdev);
1466         }
1467
1468         mutex_unlock(&info->mutex);
1469
1470         if (!xbdev) {
1471                 /* sudden device removal */
1472                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1473                 xlvbd_release_gendisk(info);
1474                 disk->private_data = NULL;
1475                 kfree(info);
1476         }
1477
1478 out:
1479         bdput(bdev);
1480         mutex_unlock(&blkfront_mutex);
1481         return 0;
1482 }
1483
1484 static const struct block_device_operations xlvbd_block_fops =
1485 {
1486         .owner = THIS_MODULE,
1487         .open = blkif_open,
1488         .release = blkif_release,
1489         .getgeo = blkif_getgeo,
1490         .ioctl = blkif_ioctl,
1491 };
1492
1493
1494 static const struct xenbus_device_id blkfront_ids[] = {
1495         { "vbd" },
1496         { "" }
1497 };
1498
1499 static DEFINE_XENBUS_DRIVER(blkfront, ,
1500         .probe = blkfront_probe,
1501         .remove = blkfront_remove,
1502         .resume = blkfront_resume,
1503         .otherend_changed = blkback_changed,
1504         .is_ready = blkfront_is_ready,
1505 );
1506
1507 static int __init xlblk_init(void)
1508 {
1509         int ret;
1510
1511         if (!xen_domain())
1512                 return -ENODEV;
1513
1514         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1515                 return -ENODEV;
1516
1517         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1518                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1519                        XENVBD_MAJOR, DEV_NAME);
1520                 return -ENODEV;
1521         }
1522
1523         ret = xenbus_register_frontend(&blkfront_driver);
1524         if (ret) {
1525                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1526                 return ret;
1527         }
1528
1529         return 0;
1530 }
1531 module_init(xlblk_init);
1532
1533
1534 static void __exit xlblk_exit(void)
1535 {
1536         xenbus_unregister_driver(&blkfront_driver);
1537         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1538         kfree(minors);
1539 }
1540 module_exit(xlblk_exit);
1541
1542 MODULE_DESCRIPTION("Xen virtual block device frontend");
1543 MODULE_LICENSE("GPL");
1544 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1545 MODULE_ALIAS("xen:vbd");
1546 MODULE_ALIAS("xenblk");