Pileus Git - ~andy/linux/blob - 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 char *encode_disk_name(char *ptr, unsigned int n)
 530 {
 531         if (n >= 26)
 532                 ptr = encode_disk_name(ptr, n / 26 - 1);
 533         *ptr = 'a' + n % 26;
 534         return ptr + 1;
 535 }
 536
 537 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 538                                struct blkfront_info *info,
 539                                u16 vdisk_info, u16 sector_size)
 540 {
 541         struct gendisk *gd;
 542         int nr_minors = 1;
 543         int err;
 544         unsigned int offset;
 545         int minor;
 546         int nr_parts;
 547         char *ptr;
 548
 549         BUG_ON(info->gd != NULL);
 550         BUG_ON(info->rq != NULL);
 551
 552         if ((info->vdevice>>EXT_SHIFT) > 1) {
 553                 /* this is above the extended range; something is wrong */
 554                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 555                 return -ENODEV;
 556         }
 557
 558         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 559                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 560                 if (err)
 561                         return err;
 562                 nr_parts = PARTS_PER_DISK;
 563         } else {
 564                 minor = BLKIF_MINOR_EXT(info->vdevice);
 565                 nr_parts = PARTS_PER_EXT_DISK;
 566                 offset = minor / nr_parts;
 567                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 568                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 569                                         "emulated IDE disks,\n\t choose an xvd device name"
 570                                         "from xvde on\n", info->vdevice);
 571         }
 572         if (minor >> MINORBITS) {
 573                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 574                         info->vdevice, minor);
 575                 return -ENODEV;
 576         }
 577
 578         if ((minor % nr_parts) == 0)
 579                 nr_minors = nr_parts;
 580
 581         err = xlbd_reserve_minors(minor, nr_minors);
 582         if (err)
 583                 goto out;
 584         err = -ENODEV;
 585
 586         gd = alloc_disk(nr_minors);
 587         if (gd == NULL)
 588                 goto release;
 589
 590         strcpy(gd->disk_name, DEV_NAME);
 591         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 592         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 593         if (nr_minors > 1)
 594                 *ptr = 0;
 595         else
 596                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 597                          "%d", minor & (nr_parts - 1));
 598
 599         gd->major = XENVBD_MAJOR;
 600         gd->first_minor = minor;
 601         gd->fops = &xlvbd_block_fops;
 602         gd->private_data = info;
 603         gd->driverfs_dev = &(info->xbdev->dev);
 604         set_capacity(gd, capacity);
 605
 606         if (xlvbd_init_blk_queue(gd, sector_size)) {
 607                 del_gendisk(gd);
 608                 goto release;
 609         }
 610
 611         info->rq = gd->queue;
 612         info->gd = gd;
 613
 614         xlvbd_flush(info);
 615
 616         if (vdisk_info & VDISK_READONLY)
 617                 set_disk_ro(gd, 1);
 618
 619         if (vdisk_info & VDISK_REMOVABLE)
 620                 gd->flags |= GENHD_FL_REMOVABLE;
 621
 622         if (vdisk_info & VDISK_CDROM)
 623                 gd->flags |= GENHD_FL_CD;
 624
 625         return 0;
 626
 627  release:
 628         xlbd_release_minors(minor, nr_minors);
 629  out:
 630         return err;
 631 }
 632
 633 static void xlvbd_release_gendisk(struct blkfront_info *info)
 634 {
 635         unsigned int minor, nr_minors;
 636         unsigned long flags;
 637
 638         if (info->rq == NULL)
 639                 return;
 640
 641         spin_lock_irqsave(&info->io_lock, flags);
 642
 643         /* No more blkif_request(). */
 644         blk_stop_queue(info->rq);
 645
 646         /* No more gnttab callback work. */
 647         gnttab_cancel_free_callback(&info->callback);
 648         spin_unlock_irqrestore(&info->io_lock, flags);
 649
 650         /* Flush gnttab callback work. Must be done with no locks held. */
 651         flush_work_sync(&info->work);
 652
 653         del_gendisk(info->gd);
 654
 655         minor = info->gd->first_minor;
 656         nr_minors = info->gd->minors;
 657         xlbd_release_minors(minor, nr_minors);
 658
 659         blk_cleanup_queue(info->rq);
 660         info->rq = NULL;
 661
 662         put_disk(info->gd);
 663         info->gd = NULL;
 664 }
 665
 666 static void kick_pending_request_queues(struct blkfront_info *info)
 667 {
 668         if (!RING_FULL(&info->ring)) {
 669                 /* Re-enable calldowns. */
 670                 blk_start_queue(info->rq);
 671                 /* Kick things off immediately. */
 672                 do_blkif_request(info->rq);
 673         }
 674 }
 675
 676 static void blkif_restart_queue(struct work_struct *work)
 677 {
 678         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 679
 680         spin_lock_irq(&info->io_lock);
 681         if (info->connected == BLKIF_STATE_CONNECTED)
 682                 kick_pending_request_queues(info);
 683         spin_unlock_irq(&info->io_lock);
 684 }
 685
 686 static void blkif_free(struct blkfront_info *info, int suspend)
 687 {
 688         /* Prevent new requests being issued until we fix things up. */
 689         spin_lock_irq(&info->io_lock);
 690         info->connected = suspend ?
 691                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 692         /* No more blkif_request(). */
 693         if (info->rq)
 694                 blk_stop_queue(info->rq);
 695         /* No more gnttab callback work. */
 696         gnttab_cancel_free_callback(&info->callback);
 697         spin_unlock_irq(&info->io_lock);
 698
 699         /* Flush gnttab callback work. Must be done with no locks held. */
 700         flush_work_sync(&info->work);
 701
 702         /* Free resources associated with old device channel. */
 703         if (info->ring_ref != GRANT_INVALID_REF) {
 704                 gnttab_end_foreign_access(info->ring_ref, 0,
 705                                           (unsigned long)info->ring.sring);
 706                 info->ring_ref = GRANT_INVALID_REF;
 707                 info->ring.sring = NULL;
 708         }
 709         if (info->irq)
 710                 unbind_from_irqhandler(info->irq, info);
 711         info->evtchn = info->irq = 0;
 712
 713 }
 714
 715 static void blkif_completion(struct blk_shadow *s)
 716 {
 717         int i;
 718         /* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
 719          * flag. */
 720         for (i = 0; i < s->req.u.rw.nr_segments; i++)
 721                 gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
 722 }
 723
 724 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 725 {
 726         struct request *req;
 727         struct blkif_response *bret;
 728         RING_IDX i, rp;
 729         unsigned long flags;
 730         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 731         int error;
 732
 733         spin_lock_irqsave(&info->io_lock, flags);
 734
 735         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 736                 spin_unlock_irqrestore(&info->io_lock, flags);
 737                 return IRQ_HANDLED;
 738         }
 739
 740  again:
 741         rp = info->ring.sring->rsp_prod;
 742         rmb(); /* Ensure we see queued responses up to 'rp'. */
 743
 744         for (i = info->ring.rsp_cons; i != rp; i++) {
 745                 unsigned long id;
 746
 747                 bret = RING_GET_RESPONSE(&info->ring, i);
 748                 id   = bret->id;
 749                 req  = info->shadow[id].request;
 750
 751                 if (bret->operation != BLKIF_OP_DISCARD)
 752                         blkif_completion(&info->shadow[id]);
 753
 754                 add_id_to_freelist(info, id);
 755
 756                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 757                 switch (bret->operation) {
 758                 case BLKIF_OP_DISCARD:
 759                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 760                                 struct request_queue *rq = info->rq;
 761                                 printk(KERN_WARNING "blkfront: %s: discard op failed\n",
 762                                            info->gd->disk_name);
 763                                 error = -EOPNOTSUPP;
 764                                 info->feature_discard = 0;
 765                                 info->feature_secdiscard = 0;
 766                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 767                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 768                         }
 769                         __blk_end_request_all(req, error);
 770                         break;
 771                 case BLKIF_OP_FLUSH_DISKCACHE:
 772                 case BLKIF_OP_WRITE_BARRIER:
 773                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 774                                 printk(KERN_WARNING "blkfront: %s: write %s op failed\n",
 775                                        info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 776                                        "barrier" :  "flush disk cache",
 777                                        info->gd->disk_name);
 778                                 error = -EOPNOTSUPP;
 779                         }
 780                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
 781                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
 782                                 printk(KERN_WARNING "blkfront: %s: empty write %s op failed\n",
 783                                        info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 784                                        "barrier" :  "flush disk cache",
 785                                        info->gd->disk_name);
 786                                 error = -EOPNOTSUPP;
 787                         }
 788                         if (unlikely(error)) {
 789                                 if (error == -EOPNOTSUPP)
 790                                         error = 0;
 791                                 info->feature_flush = 0;
 792                                 info->flush_op = 0;
 793                                 xlvbd_flush(info);
 794                         }
 795                         /* fall through */
 796                 case BLKIF_OP_READ:
 797                 case BLKIF_OP_WRITE:
 798                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
 799                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
 800                                         "request: %x\n", bret->status);
 801
 802                         __blk_end_request_all(req, error);
 803                         break;
 804                 default:
 805                         BUG();
 806                 }
 807         }
 808
 809         info->ring.rsp_cons = i;
 810
 811         if (i != info->ring.req_prod_pvt) {
 812                 int more_to_do;
 813                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
 814                 if (more_to_do)
 815                         goto again;
 816         } else
 817                 info->ring.sring->rsp_event = i + 1;
 818
 819         kick_pending_request_queues(info);
 820
 821         spin_unlock_irqrestore(&info->io_lock, flags);
 822
 823         return IRQ_HANDLED;
 824 }
 825
 826
 827 static int setup_blkring(struct xenbus_device *dev,
 828                          struct blkfront_info *info)
 829 {
 830         struct blkif_sring *sring;
 831         int err;
 832
 833         info->ring_ref = GRANT_INVALID_REF;
 834
 835         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
 836         if (!sring) {
 837                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 838                 return -ENOMEM;
 839         }
 840         SHARED_RING_INIT(sring);
 841         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 842
 843         sg_init_table(info->sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 844
 845         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
 846         if (err < 0) {
 847                 free_page((unsigned long)sring);
 848                 info->ring.sring = NULL;
 849                 goto fail;
 850         }
 851         info->ring_ref = err;
 852
 853         err = xenbus_alloc_evtchn(dev, &info->evtchn);
 854         if (err)
 855                 goto fail;
 856
 857         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
 858                                         "blkif", info);
 859         if (err <= 0) {
 860                 xenbus_dev_fatal(dev, err,
 861                                  "bind_evtchn_to_irqhandler failed");
 862                 goto fail;
 863         }
 864         info->irq = err;
 865
 866         return 0;
 867 fail:
 868         blkif_free(info, 0);
 869         return err;
 870 }
 871
 872
 873 /* Common code used when first setting up, and when resuming. */
 874 static int talk_to_blkback(struct xenbus_device *dev,
 875                            struct blkfront_info *info)
 876 {
 877         const char *message = NULL;
 878         struct xenbus_transaction xbt;
 879         int err;
 880
 881         /* Create shared ring, alloc event channel. */
 882         err = setup_blkring(dev, info);
 883         if (err)
 884                 goto out;
 885
 886 again:
 887         err = xenbus_transaction_start(&xbt);
 888         if (err) {
 889                 xenbus_dev_fatal(dev, err, "starting transaction");
 890                 goto destroy_blkring;
 891         }
 892
 893         err = xenbus_printf(xbt, dev->nodename,
 894                             "ring-ref", "%u", info->ring_ref);
 895         if (err) {
 896                 message = "writing ring-ref";
 897                 goto abort_transaction;
 898         }
 899         err = xenbus_printf(xbt, dev->nodename,
 900                             "event-channel", "%u", info->evtchn);
 901         if (err) {
 902                 message = "writing event-channel";
 903                 goto abort_transaction;
 904         }
 905         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
 906                             XEN_IO_PROTO_ABI_NATIVE);
 907         if (err) {
 908                 message = "writing protocol";
 909                 goto abort_transaction;
 910         }
 911
 912         err = xenbus_transaction_end(xbt, 0);
 913         if (err) {
 914                 if (err == -EAGAIN)
 915                         goto again;
 916                 xenbus_dev_fatal(dev, err, "completing transaction");
 917                 goto destroy_blkring;
 918         }
 919
 920         xenbus_switch_state(dev, XenbusStateInitialised);
 921
 922         return 0;
 923
 924  abort_transaction:
 925         xenbus_transaction_end(xbt, 1);
 926         if (message)
 927                 xenbus_dev_fatal(dev, err, "%s", message);
 928  destroy_blkring:
 929         blkif_free(info, 0);
 930  out:
 931         return err;
 932 }
 933
 934 /**
 935  * Entry point to this code when a new device is created.  Allocate the basic
 936  * structures and the ring buffer for communication with the backend, and
 937  * inform the backend of the appropriate details for those.  Switch to
 938  * Initialised state.
 939  */
 940 static int blkfront_probe(struct xenbus_device *dev,
 941                           const struct xenbus_device_id *id)
 942 {
 943         int err, vdevice, i;
 944         struct blkfront_info *info;
 945
 946         /* FIXME: Use dynamic device id if this is not set. */
 947         err = xenbus_scanf(XBT_NIL, dev->nodename,
 948                            "virtual-device", "%i", &vdevice);
 949         if (err != 1) {
 950                 /* go looking in the extended area instead */
 951                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
 952                                    "%i", &vdevice);
 953                 if (err != 1) {
 954                         xenbus_dev_fatal(dev, err, "reading virtual-device");
 955                         return err;
 956                 }
 957         }
 958
 959         if (xen_hvm_domain()) {
 960                 char *type;
 961                 int len;
 962                 /* no unplug has been done: do not hook devices != xen vbds */
 963                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
 964                         int major;
 965
 966                         if (!VDEV_IS_EXTENDED(vdevice))
 967                                 major = BLKIF_MAJOR(vdevice);
 968                         else
 969                                 major = XENVBD_MAJOR;
 970
 971                         if (major != XENVBD_MAJOR) {
 972                                 printk(KERN_INFO
 973                                                 "%s: HVM does not support vbd %d as xen block device\n",
 974                                                 __FUNCTION__, vdevice);
 975                                 return -ENODEV;
 976                         }
 977                 }
 978                 /* do not create a PV cdrom device if we are an HVM guest */
 979                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
 980                 if (IS_ERR(type))
 981                         return -ENODEV;
 982                 if (strncmp(type, "cdrom", 5) == 0) {
 983                         kfree(type);
 984                         return -ENODEV;
 985                 }
 986                 kfree(type);
 987         }
 988         info = kzalloc(sizeof(*info), GFP_KERNEL);
 989         if (!info) {
 990                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
 991                 return -ENOMEM;
 992         }
 993
 994         mutex_init(&info->mutex);
 995         spin_lock_init(&info->io_lock);
 996         info->xbdev = dev;
 997         info->vdevice = vdevice;
 998         info->connected = BLKIF_STATE_DISCONNECTED;
 999         INIT_WORK(&info->work, blkif_restart_queue);
1000
1001         for (i = 0; i < BLK_RING_SIZE; i++)
1002                 info->shadow[i].req.u.rw.id = i+1;
1003         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1004
1005         /* Front end dir is a number, which is used as the id. */
1006         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1007         dev_set_drvdata(&dev->dev, info);
1008
1009         err = talk_to_blkback(dev, info);
1010         if (err) {
1011                 kfree(info);
1012                 dev_set_drvdata(&dev->dev, NULL);
1013                 return err;
1014         }
1015
1016         return 0;
1017 }
1018
1019
1020 static int blkif_recover(struct blkfront_info *info)
1021 {
1022         int i;
1023         struct blkif_request *req;
1024         struct blk_shadow *copy;
1025         int j;
1026
1027         /* Stage 1: Make a safe copy of the shadow state. */
1028         copy = kmalloc(sizeof(info->shadow),
1029                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1030         if (!copy)
1031                 return -ENOMEM;
1032         memcpy(copy, info->shadow, sizeof(info->shadow));
1033
1034         /* Stage 2: Set up free list. */
1035         memset(&info->shadow, 0, sizeof(info->shadow));
1036         for (i = 0; i < BLK_RING_SIZE; i++)
1037                 info->shadow[i].req.u.rw.id = i+1;
1038         info->shadow_free = info->ring.req_prod_pvt;
1039         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1040
1041         /* Stage 3: Find pending requests and requeue them. */
1042         for (i = 0; i < BLK_RING_SIZE; i++) {
1043                 /* Not in use? */
1044                 if (!copy[i].request)
1045                         continue;
1046
1047                 /* Grab a request slot and copy shadow state into it. */
1048                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1049                 *req = copy[i].req;
1050
1051                 /* We get a new request id, and must reset the shadow state. */
1052                 req->u.rw.id = get_id_from_freelist(info);
1053                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1054
1055                 if (req->operation != BLKIF_OP_DISCARD) {
1056                 /* Rewrite any grant references invalidated by susp/resume. */
1057                         for (j = 0; j < req->u.rw.nr_segments; j++)
1058                                 gnttab_grant_foreign_access_ref(
1059                                         req->u.rw.seg[j].gref,
1060                                         info->xbdev->otherend_id,
1061                                         pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
1062                                         rq_data_dir(info->shadow[req->u.rw.id].request));
1063                 }
1064                 info->shadow[req->u.rw.id].req = *req;
1065
1066                 info->ring.req_prod_pvt++;
1067         }
1068
1069         kfree(copy);
1070
1071         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1072
1073         spin_lock_irq(&info->io_lock);
1074
1075         /* Now safe for us to use the shared ring */
1076         info->connected = BLKIF_STATE_CONNECTED;
1077
1078         /* Send off requeued requests */
1079         flush_requests(info);
1080
1081         /* Kick any other new requests queued since we resumed */
1082         kick_pending_request_queues(info);
1083
1084         spin_unlock_irq(&info->io_lock);
1085
1086         return 0;
1087 }
1088
1089 /**
1090  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1091  * driver restart.  We tear down our blkif structure and recreate it, but
1092  * leave the device-layer structures intact so that this is transparent to the
1093  * rest of the kernel.
1094  */
1095 static int blkfront_resume(struct xenbus_device *dev)
1096 {
1097         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1098         int err;
1099
1100         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1101
1102         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1103
1104         err = talk_to_blkback(dev, info);
1105         if (info->connected == BLKIF_STATE_SUSPENDED && !err)
1106                 err = blkif_recover(info);
1107
1108         return err;
1109 }
1110
1111 static void
1112 blkfront_closing(struct blkfront_info *info)
1113 {
1114         struct xenbus_device *xbdev = info->xbdev;
1115         struct block_device *bdev = NULL;
1116
1117         mutex_lock(&info->mutex);
1118
1119         if (xbdev->state == XenbusStateClosing) {
1120                 mutex_unlock(&info->mutex);
1121                 return;
1122         }
1123
1124         if (info->gd)
1125                 bdev = bdget_disk(info->gd, 0);
1126
1127         mutex_unlock(&info->mutex);
1128
1129         if (!bdev) {
1130                 xenbus_frontend_closed(xbdev);
1131                 return;
1132         }
1133
1134         mutex_lock(&bdev->bd_mutex);
1135
1136         if (bdev->bd_openers) {
1137                 xenbus_dev_error(xbdev, -EBUSY,
1138                                  "Device in use; refusing to close");
1139                 xenbus_switch_state(xbdev, XenbusStateClosing);
1140         } else {
1141                 xlvbd_release_gendisk(info);
1142                 xenbus_frontend_closed(xbdev);
1143         }
1144
1145         mutex_unlock(&bdev->bd_mutex);
1146         bdput(bdev);
1147 }
1148
1149 static void blkfront_setup_discard(struct blkfront_info *info)
1150 {
1151         int err;
1152         char *type;
1153         unsigned int discard_granularity;
1154         unsigned int discard_alignment;
1155         unsigned int discard_secure;
1156
1157         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1158         if (IS_ERR(type))
1159                 return;
1160
1161         info->feature_secdiscard = 0;
1162         if (strncmp(type, "phy", 3) == 0) {
1163                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1164                         "discard-granularity", "%u", &discard_granularity,
1165                         "discard-alignment", "%u", &discard_alignment,
1166                         NULL);
1167                 if (!err) {
1168                         info->feature_discard = 1;
1169                         info->discard_granularity = discard_granularity;
1170                         info->discard_alignment = discard_alignment;
1171                 }
1172                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1173                             "discard-secure", "%d", &discard_secure,
1174                             NULL);
1175                 if (!err)
1176                         info->feature_secdiscard = discard_secure;
1177
1178         } else if (strncmp(type, "file", 4) == 0)
1179                 info->feature_discard = 1;
1180
1181         kfree(type);
1182 }
1183
1184 /*
1185  * Invoked when the backend is finally 'ready' (and has told produced
1186  * the details about the physical device - #sectors, size, etc).
1187  */
1188 static void blkfront_connect(struct blkfront_info *info)
1189 {
1190         unsigned long long sectors;
1191         unsigned long sector_size;
1192         unsigned int binfo;
1193         int err;
1194         int barrier, flush, discard;
1195
1196         switch (info->connected) {
1197         case BLKIF_STATE_CONNECTED:
1198                 /*
1199                  * Potentially, the back-end may be signalling
1200                  * a capacity change; update the capacity.
1201                  */
1202                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1203                                    "sectors", "%Lu", &sectors);
1204                 if (XENBUS_EXIST_ERR(err))
1205                         return;
1206                 printk(KERN_INFO "Setting capacity to %Lu\n",
1207                        sectors);
1208                 set_capacity(info->gd, sectors);
1209                 revalidate_disk(info->gd);
1210
1211                 /* fall through */
1212         case BLKIF_STATE_SUSPENDED:
1213                 return;
1214
1215         default:
1216                 break;
1217         }
1218
1219         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1220                 __func__, info->xbdev->otherend);
1221
1222         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1223                             "sectors", "%llu", &sectors,
1224                             "info", "%u", &binfo,
1225                             "sector-size", "%lu", &sector_size,
1226                             NULL);
1227         if (err) {
1228                 xenbus_dev_fatal(info->xbdev, err,
1229                                  "reading backend fields at %s",
1230                                  info->xbdev->otherend);
1231                 return;
1232         }
1233
1234         info->feature_flush = 0;
1235         info->flush_op = 0;
1236
1237         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1238                             "feature-barrier", "%d", &barrier,
1239                             NULL);
1240
1241         /*
1242          * If there's no "feature-barrier" defined, then it means
1243          * we're dealing with a very old backend which writes
1244          * synchronously; nothing to do.
1245          *
1246          * If there are barriers, then we use flush.
1247          */
1248         if (!err && barrier) {
1249                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1250                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1251         }
1252         /*
1253          * And if there is "feature-flush-cache" use that above
1254          * barriers.
1255          */
1256         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1257                             "feature-flush-cache", "%d", &flush,
1258                             NULL);
1259
1260         if (!err && flush) {
1261                 info->feature_flush = REQ_FLUSH;
1262                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1263         }
1264
1265         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1266                             "feature-discard", "%d", &discard,
1267                             NULL);
1268
1269         if (!err && discard)
1270                 blkfront_setup_discard(info);
1271
1272         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1273         if (err) {
1274                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1275                                  info->xbdev->otherend);
1276                 return;
1277         }
1278
1279         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1280
1281         /* Kick pending requests. */
1282         spin_lock_irq(&info->io_lock);
1283         info->connected = BLKIF_STATE_CONNECTED;
1284         kick_pending_request_queues(info);
1285         spin_unlock_irq(&info->io_lock);
1286
1287         add_disk(info->gd);
1288
1289         info->is_ready = 1;
1290 }
1291
1292 /**
1293  * Callback received when the backend's state changes.
1294  */
1295 static void blkback_changed(struct xenbus_device *dev,
1296                             enum xenbus_state backend_state)
1297 {
1298         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1299
1300         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1301
1302         switch (backend_state) {
1303         case XenbusStateInitialising:
1304         case XenbusStateInitWait:
1305         case XenbusStateInitialised:
1306         case XenbusStateReconfiguring:
1307         case XenbusStateReconfigured:
1308         case XenbusStateUnknown:
1309         case XenbusStateClosed:
1310                 break;
1311
1312         case XenbusStateConnected:
1313                 blkfront_connect(info);
1314                 break;
1315
1316         case XenbusStateClosing:
1317                 blkfront_closing(info);
1318                 break;
1319         }
1320 }
1321
1322 static int blkfront_remove(struct xenbus_device *xbdev)
1323 {
1324         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1325         struct block_device *bdev = NULL;
1326         struct gendisk *disk;
1327
1328         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1329
1330         blkif_free(info, 0);
1331
1332         mutex_lock(&info->mutex);
1333
1334         disk = info->gd;
1335         if (disk)
1336                 bdev = bdget_disk(disk, 0);
1337
1338         info->xbdev = NULL;
1339         mutex_unlock(&info->mutex);
1340
1341         if (!bdev) {
1342                 kfree(info);
1343                 return 0;
1344         }
1345
1346         /*
1347          * The xbdev was removed before we reached the Closed
1348          * state. See if it's safe to remove the disk. If the bdev
1349          * isn't closed yet, we let release take care of it.
1350          */
1351
1352         mutex_lock(&bdev->bd_mutex);
1353         info = disk->private_data;
1354
1355         dev_warn(disk_to_dev(disk),
1356                  "%s was hot-unplugged, %d stale handles\n",
1357                  xbdev->nodename, bdev->bd_openers);
1358
1359         if (info && !bdev->bd_openers) {
1360                 xlvbd_release_gendisk(info);
1361                 disk->private_data = NULL;
1362                 kfree(info);
1363         }
1364
1365         mutex_unlock(&bdev->bd_mutex);
1366         bdput(bdev);
1367
1368         return 0;
1369 }
1370
1371 static int blkfront_is_ready(struct xenbus_device *dev)
1372 {
1373         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1374
1375         return info->is_ready && info->xbdev;
1376 }
1377
1378 static int blkif_open(struct block_device *bdev, fmode_t mode)
1379 {
1380         struct gendisk *disk = bdev->bd_disk;
1381         struct blkfront_info *info;
1382         int err = 0;
1383
1384         mutex_lock(&blkfront_mutex);
1385
1386         info = disk->private_data;
1387         if (!info) {
1388                 /* xbdev gone */
1389                 err = -ERESTARTSYS;
1390                 goto out;
1391         }
1392
1393         mutex_lock(&info->mutex);
1394
1395         if (!info->gd)
1396                 /* xbdev is closed */
1397                 err = -ERESTARTSYS;
1398
1399         mutex_unlock(&info->mutex);
1400
1401 out:
1402         mutex_unlock(&blkfront_mutex);
1403         return err;
1404 }
1405
1406 static int blkif_release(struct gendisk *disk, fmode_t mode)
1407 {
1408         struct blkfront_info *info = disk->private_data;
1409         struct block_device *bdev;
1410         struct xenbus_device *xbdev;
1411
1412         mutex_lock(&blkfront_mutex);
1413
1414         bdev = bdget_disk(disk, 0);
1415
1416         if (bdev->bd_openers)
1417                 goto out;
1418
1419         /*
1420          * Check if we have been instructed to close. We will have
1421          * deferred this request, because the bdev was still open.
1422          */
1423
1424         mutex_lock(&info->mutex);
1425         xbdev = info->xbdev;
1426
1427         if (xbdev && xbdev->state == XenbusStateClosing) {
1428                 /* pending switch to state closed */
1429                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1430                 xlvbd_release_gendisk(info);
1431                 xenbus_frontend_closed(info->xbdev);
1432         }
1433
1434         mutex_unlock(&info->mutex);
1435
1436         if (!xbdev) {
1437                 /* sudden device removal */
1438                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1439                 xlvbd_release_gendisk(info);
1440                 disk->private_data = NULL;
1441                 kfree(info);
1442         }
1443
1444 out:
1445         bdput(bdev);
1446         mutex_unlock(&blkfront_mutex);
1447         return 0;
1448 }
1449
1450 static const struct block_device_operations xlvbd_block_fops =
1451 {
1452         .owner = THIS_MODULE,
1453         .open = blkif_open,
1454         .release = blkif_release,
1455         .getgeo = blkif_getgeo,
1456         .ioctl = blkif_ioctl,
1457 };
1458
1459
1460 static const struct xenbus_device_id blkfront_ids[] = {
1461         { "vbd" },
1462         { "" }
1463 };
1464
1465 static DEFINE_XENBUS_DRIVER(blkfront, ,
1466         .probe = blkfront_probe,
1467         .remove = blkfront_remove,
1468         .resume = blkfront_resume,
1469         .otherend_changed = blkback_changed,
1470         .is_ready = blkfront_is_ready,
1471 );
1472
1473 static int __init xlblk_init(void)
1474 {
1475         int ret;
1476
1477         if (!xen_domain())
1478                 return -ENODEV;
1479
1480         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1481                 return -ENODEV;
1482
1483         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1484                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1485                        XENVBD_MAJOR, DEV_NAME);
1486                 return -ENODEV;
1487         }
1488
1489         ret = xenbus_register_frontend(&blkfront_driver);
1490         if (ret) {
1491                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1492                 return ret;
1493         }
1494
1495         return 0;
1496 }
1497 module_init(xlblk_init);
1498
1499
1500 static void __exit xlblk_exit(void)
1501 {
1502         xenbus_unregister_driver(&blkfront_driver);
1503         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1504         kfree(minors);
1505 }
1506 module_exit(xlblk_exit);
1507
1508 MODULE_DESCRIPTION("Xen virtual block device frontend");
1509 MODULE_LICENSE("GPL");
1510 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1511 MODULE_ALIAS("xen:vbd");
1512 MODULE_ALIAS("xenblk");