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 #include <linux/list.h>
  48
  49 #include <xen/xen.h>
  50 #include <xen/xenbus.h>
  51 #include <xen/grant_table.h>
  52 #include <xen/events.h>
  53 #include <xen/page.h>
  54 #include <xen/platform_pci.h>
  55
  56 #include <xen/interface/grant_table.h>
  57 #include <xen/interface/io/blkif.h>
  58 #include <xen/interface/io/protocols.h>
  59
  60 #include <asm/xen/hypervisor.h>
  61
  62 enum blkif_state {
  63         BLKIF_STATE_DISCONNECTED,
  64         BLKIF_STATE_CONNECTED,
  65         BLKIF_STATE_SUSPENDED,
  66 };
  67
  68 struct grant {
  69         grant_ref_t gref;
  70         unsigned long pfn;
  71         struct list_head node;
  72 };
  73
  74 struct blk_shadow {
  75         struct blkif_request req;
  76         struct request *request;
  77         struct grant *grants_used[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  78         struct scatterlist sg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  79 };
  80
  81 static DEFINE_MUTEX(blkfront_mutex);
  82 static const struct block_device_operations xlvbd_block_fops;
  83
  84 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
  85
  86 /*
  87  * We have one of these per vbd, whether ide, scsi or 'other'.  They
  88  * hang in private_data off the gendisk structure. We may end up
  89  * putting all kinds of interesting stuff here :-)
  90  */
  91 struct blkfront_info
  92 {
  93         spinlock_t io_lock;
  94         struct mutex mutex;
  95         struct xenbus_device *xbdev;
  96         struct gendisk *gd;
  97         int vdevice;
  98         blkif_vdev_t handle;
  99         enum blkif_state connected;
 100         int ring_ref;
 101         struct blkif_front_ring ring;
 102         unsigned int evtchn, irq;
 103         struct request_queue *rq;
 104         struct work_struct work;
 105         struct gnttab_free_callback callback;
 106         struct blk_shadow shadow[BLK_RING_SIZE];
 107         struct list_head grants;
 108         unsigned int persistent_gnts_c;
 109         unsigned long shadow_free;
 110         unsigned int feature_flush;
 111         unsigned int flush_op;
 112         unsigned int feature_discard:1;
 113         unsigned int feature_secdiscard:1;
 114         unsigned int discard_granularity;
 115         unsigned int discard_alignment;
 116         unsigned int feature_persistent:1;
 117         int is_ready;
 118 };
 119
 120 static unsigned int nr_minors;
 121 static unsigned long *minors;
 122 static DEFINE_SPINLOCK(minor_lock);
 123
 124 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 125         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 126 #define GRANT_INVALID_REF       0
 127
 128 #define PARTS_PER_DISK          16
 129 #define PARTS_PER_EXT_DISK      256
 130
 131 #define BLKIF_MAJOR(dev) ((dev)>>8)
 132 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 133
 134 #define EXT_SHIFT 28
 135 #define EXTENDED (1<<EXT_SHIFT)
 136 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 137 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 138 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 139 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 140 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 141 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 142
 143 #define DEV_NAME        "xvd"   /* name in /dev */
 144
 145 static int get_id_from_freelist(struct blkfront_info *info)
 146 {
 147         unsigned long free = info->shadow_free;
 148         BUG_ON(free >= BLK_RING_SIZE);
 149         info->shadow_free = info->shadow[free].req.u.rw.id;
 150         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 151         return free;
 152 }
 153
 154 static int add_id_to_freelist(struct blkfront_info *info,
 155                                unsigned long id)
 156 {
 157         if (info->shadow[id].req.u.rw.id != id)
 158                 return -EINVAL;
 159         if (info->shadow[id].request == NULL)
 160                 return -EINVAL;
 161         info->shadow[id].req.u.rw.id  = info->shadow_free;
 162         info->shadow[id].request = NULL;
 163         info->shadow_free = id;
 164         return 0;
 165 }
 166
 167 static int fill_grant_buffer(struct blkfront_info *info, int num)
 168 {
 169         struct page *granted_page;
 170         struct grant *gnt_list_entry, *n;
 171         int i = 0;
 172
 173         while(i < num) {
 174                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 175                 if (!gnt_list_entry)
 176                         goto out_of_memory;
 177
 178                 if (info->feature_persistent) {
 179                         granted_page = alloc_page(GFP_NOIO);
 180                         if (!granted_page) {
 181                                 kfree(gnt_list_entry);
 182                                 goto out_of_memory;
 183                         }
 184                         gnt_list_entry->pfn = page_to_pfn(granted_page);
 185                 }
 186
 187                 gnt_list_entry->gref = GRANT_INVALID_REF;
 188                 list_add(&gnt_list_entry->node, &info->grants);
 189                 i++;
 190         }
 191
 192         return 0;
 193
 194 out_of_memory:
 195         list_for_each_entry_safe(gnt_list_entry, n,
 196                                  &info->grants, node) {
 197                 list_del(&gnt_list_entry->node);
 198                 if (info->feature_persistent)
 199                         __free_page(pfn_to_page(gnt_list_entry->pfn));
 200                 kfree(gnt_list_entry);
 201                 i--;
 202         }
 203         BUG_ON(i != 0);
 204         return -ENOMEM;
 205 }
 206
 207 static struct grant *get_grant(grant_ref_t *gref_head,
 208                                unsigned long pfn,
 209                                struct blkfront_info *info)
 210 {
 211         struct grant *gnt_list_entry;
 212         unsigned long buffer_mfn;
 213
 214         BUG_ON(list_empty(&info->grants));
 215         gnt_list_entry = list_first_entry(&info->grants, struct grant, node);
 216         list_del(&gnt_list_entry->node);
 217
 218         if (gnt_list_entry->gref != GRANT_INVALID_REF) {
 219                 info->persistent_gnts_c--;
 220                 return gnt_list_entry;
 221         }
 222
 223         /* Assign a gref to this page */
 224         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 225         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 226         if (!info->feature_persistent) {
 227                 BUG_ON(!pfn);
 228                 gnt_list_entry->pfn = pfn;
 229         }
 230         buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
 231         gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 232                                         info->xbdev->otherend_id,
 233                                         buffer_mfn, 0);
 234         return gnt_list_entry;
 235 }
 236
 237 static const char *op_name(int op)
 238 {
 239         static const char *const names[] = {
 240                 [BLKIF_OP_READ] = "read",
 241                 [BLKIF_OP_WRITE] = "write",
 242                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 243                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 244                 [BLKIF_OP_DISCARD] = "discard" };
 245
 246         if (op < 0 || op >= ARRAY_SIZE(names))
 247                 return "unknown";
 248
 249         if (!names[op])
 250                 return "reserved";
 251
 252         return names[op];
 253 }
 254 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 255 {
 256         unsigned int end = minor + nr;
 257         int rc;
 258
 259         if (end > nr_minors) {
 260                 unsigned long *bitmap, *old;
 261
 262                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 263                                  GFP_KERNEL);
 264                 if (bitmap == NULL)
 265                         return -ENOMEM;
 266
 267                 spin_lock(&minor_lock);
 268                 if (end > nr_minors) {
 269                         old = minors;
 270                         memcpy(bitmap, minors,
 271                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 272                         minors = bitmap;
 273                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 274                 } else
 275                         old = bitmap;
 276                 spin_unlock(&minor_lock);
 277                 kfree(old);
 278         }
 279
 280         spin_lock(&minor_lock);
 281         if (find_next_bit(minors, end, minor) >= end) {
 282                 bitmap_set(minors, minor, nr);
 283                 rc = 0;
 284         } else
 285                 rc = -EBUSY;
 286         spin_unlock(&minor_lock);
 287
 288         return rc;
 289 }
 290
 291 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 292 {
 293         unsigned int end = minor + nr;
 294
 295         BUG_ON(end > nr_minors);
 296         spin_lock(&minor_lock);
 297         bitmap_clear(minors,  minor, nr);
 298         spin_unlock(&minor_lock);
 299 }
 300
 301 static void blkif_restart_queue_callback(void *arg)
 302 {
 303         struct blkfront_info *info = (struct blkfront_info *)arg;
 304         schedule_work(&info->work);
 305 }
 306
 307 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 308 {
 309         /* We don't have real geometry info, but let's at least return
 310            values consistent with the size of the device */
 311         sector_t nsect = get_capacity(bd->bd_disk);
 312         sector_t cylinders = nsect;
 313
 314         hg->heads = 0xff;
 315         hg->sectors = 0x3f;
 316         sector_div(cylinders, hg->heads * hg->sectors);
 317         hg->cylinders = cylinders;
 318         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 319                 hg->cylinders = 0xffff;
 320         return 0;
 321 }
 322
 323 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 324                        unsigned command, unsigned long argument)
 325 {
 326         struct blkfront_info *info = bdev->bd_disk->private_data;
 327         int i;
 328
 329         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 330                 command, (long)argument);
 331
 332         switch (command) {
 333         case CDROMMULTISESSION:
 334                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 335                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 336                         if (put_user(0, (char __user *)(argument + i)))
 337                                 return -EFAULT;
 338                 return 0;
 339
 340         case CDROM_GET_CAPABILITY: {
 341                 struct gendisk *gd = info->gd;
 342                 if (gd->flags & GENHD_FL_CD)
 343                         return 0;
 344                 return -EINVAL;
 345         }
 346
 347         default:
 348                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 349                   command);*/
 350                 return -EINVAL; /* same return as native Linux */
 351         }
 352
 353         return 0;
 354 }
 355
 356 /*
 357  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 358  * and writes are handled as expected.
 359  *
 360  * @req: a request struct
 361  */
 362 static int blkif_queue_request(struct request *req)
 363 {
 364         struct blkfront_info *info = req->rq_disk->private_data;
 365         struct blkif_request *ring_req;
 366         unsigned long id;
 367         unsigned int fsect, lsect;
 368         int i, ref;
 369
 370         /*
 371          * Used to store if we are able to queue the request by just using
 372          * existing persistent grants, or if we have to get new grants,
 373          * as there are not sufficiently many free.
 374          */
 375         bool new_persistent_gnts;
 376         grant_ref_t gref_head;
 377         struct grant *gnt_list_entry = NULL;
 378         struct scatterlist *sg;
 379
 380         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 381                 return 1;
 382
 383         /* Check if we have enought grants to allocate a requests */
 384         if (info->persistent_gnts_c < BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 385                 new_persistent_gnts = 1;
 386                 if (gnttab_alloc_grant_references(
 387                     BLKIF_MAX_SEGMENTS_PER_REQUEST - info->persistent_gnts_c,
 388                     &gref_head) < 0) {
 389                         gnttab_request_free_callback(
 390                                 &info->callback,
 391                                 blkif_restart_queue_callback,
 392                                 info,
 393                                 BLKIF_MAX_SEGMENTS_PER_REQUEST);
 394                         return 1;
 395                 }
 396         } else
 397                 new_persistent_gnts = 0;
 398
 399         /* Fill out a communications ring structure. */
 400         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 401         id = get_id_from_freelist(info);
 402         info->shadow[id].request = req;
 403
 404         ring_req->u.rw.id = id;
 405         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 406         ring_req->u.rw.handle = info->handle;
 407
 408         ring_req->operation = rq_data_dir(req) ?
 409                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 410
 411         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 412                 /*
 413                  * Ideally we can do an unordered flush-to-disk. In case the
 414                  * backend onlysupports barriers, use that. A barrier request
 415                  * a superset of FUA, so we can implement it the same
 416                  * way.  (It's also a FLUSH+FUA, since it is
 417                  * guaranteed ordered WRT previous writes.)
 418                  */
 419                 ring_req->operation = info->flush_op;
 420         }
 421
 422         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 423                 /* id, sector_number and handle are set above. */
 424                 ring_req->operation = BLKIF_OP_DISCARD;
 425                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 426                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 427                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 428                 else
 429                         ring_req->u.discard.flag = 0;
 430         } else {
 431                 ring_req->u.rw.nr_segments = blk_rq_map_sg(req->q, req,
 432                                                            info->shadow[id].sg);
 433                 BUG_ON(ring_req->u.rw.nr_segments >
 434                        BLKIF_MAX_SEGMENTS_PER_REQUEST);
 435
 436                 for_each_sg(info->shadow[id].sg, sg, ring_req->u.rw.nr_segments, i) {
 437                         fsect = sg->offset >> 9;
 438                         lsect = fsect + (sg->length >> 9) - 1;
 439
 440                         gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
 441                         ref = gnt_list_entry->gref;
 442
 443                         info->shadow[id].grants_used[i] = gnt_list_entry;
 444
 445                         if (rq_data_dir(req) && info->feature_persistent) {
 446                                 char *bvec_data;
 447                                 void *shared_data;
 448
 449                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 450
 451                                 shared_data = kmap_atomic(
 452                                         pfn_to_page(gnt_list_entry->pfn));
 453                                 bvec_data = kmap_atomic(sg_page(sg));
 454
 455                                 /*
 456                                  * this does not wipe data stored outside the
 457                                  * range sg->offset..sg->offset+sg->length.
 458                                  * Therefore, blkback *could* see data from
 459                                  * previous requests. This is OK as long as
 460                                  * persistent grants are shared with just one
 461                                  * domain. It may need refactoring if this
 462                                  * changes
 463                                  */
 464                                 memcpy(shared_data + sg->offset,
 465                                        bvec_data   + sg->offset,
 466                                        sg->length);
 467
 468                                 kunmap_atomic(bvec_data);
 469                                 kunmap_atomic(shared_data);
 470                         }
 471
 472                         ring_req->u.rw.seg[i] =
 473                                         (struct blkif_request_segment) {
 474                                                 .gref       = ref,
 475                                                 .first_sect = fsect,
 476                                                 .last_sect  = lsect };
 477                 }
 478         }
 479
 480         info->ring.req_prod_pvt++;
 481
 482         /* Keep a private copy so we can reissue requests when recovering. */
 483         info->shadow[id].req = *ring_req;
 484
 485         if (new_persistent_gnts)
 486                 gnttab_free_grant_references(gref_head);
 487
 488         return 0;
 489 }
 490
 491
 492 static inline void flush_requests(struct blkfront_info *info)
 493 {
 494         int notify;
 495
 496         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 497
 498         if (notify)
 499                 notify_remote_via_irq(info->irq);
 500 }
 501
 502 /*
 503  * do_blkif_request
 504  *  read a block; request is in a request queue
 505  */
 506 static void do_blkif_request(struct request_queue *rq)
 507 {
 508         struct blkfront_info *info = NULL;
 509         struct request *req;
 510         int queued;
 511
 512         pr_debug("Entered do_blkif_request\n");
 513
 514         queued = 0;
 515
 516         while ((req = blk_peek_request(rq)) != NULL) {
 517                 info = req->rq_disk->private_data;
 518
 519                 if (RING_FULL(&info->ring))
 520                         goto wait;
 521
 522                 blk_start_request(req);
 523
 524                 if ((req->cmd_type != REQ_TYPE_FS) ||
 525                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 526                     !info->flush_op)) {
 527                         __blk_end_request_all(req, -EIO);
 528                         continue;
 529                 }
 530
 531                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 532                          "(%u/%u) buffer:%p [%s]\n",
 533                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 534                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 535                          req->buffer, rq_data_dir(req) ? "write" : "read");
 536
 537                 if (blkif_queue_request(req)) {
 538                         blk_requeue_request(rq, req);
 539 wait:
 540                         /* Avoid pointless unplugs. */
 541                         blk_stop_queue(rq);
 542                         break;
 543                 }
 544
 545                 queued++;
 546         }
 547
 548         if (queued != 0)
 549                 flush_requests(info);
 550 }
 551
 552 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 553 {
 554         struct request_queue *rq;
 555         struct blkfront_info *info = gd->private_data;
 556
 557         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 558         if (rq == NULL)
 559                 return -1;
 560
 561         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 562
 563         if (info->feature_discard) {
 564                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 565                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 566                 rq->limits.discard_granularity = info->discard_granularity;
 567                 rq->limits.discard_alignment = info->discard_alignment;
 568                 if (info->feature_secdiscard)
 569                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 570         }
 571
 572         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 573         blk_queue_logical_block_size(rq, sector_size);
 574         blk_queue_max_hw_sectors(rq, 512);
 575
 576         /* Each segment in a request is up to an aligned page in size. */
 577         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 578         blk_queue_max_segment_size(rq, PAGE_SIZE);
 579
 580         /* Ensure a merged request will fit in a single I/O ring slot. */
 581         blk_queue_max_segments(rq, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 582
 583         /* Make sure buffer addresses are sector-aligned. */
 584         blk_queue_dma_alignment(rq, 511);
 585
 586         /* Make sure we don't use bounce buffers. */
 587         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 588
 589         gd->queue = rq;
 590
 591         return 0;
 592 }
 593
 594
 595 static void xlvbd_flush(struct blkfront_info *info)
 596 {
 597         blk_queue_flush(info->rq, info->feature_flush);
 598         printk(KERN_INFO "blkfront: %s: %s: %s %s\n",
 599                info->gd->disk_name,
 600                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 601                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 602                 "flush diskcache" : "barrier or flush"),
 603                info->feature_flush ? "enabled" : "disabled",
 604                info->feature_persistent ? "using persistent grants" : "");
 605 }
 606
 607 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 608 {
 609         int major;
 610         major = BLKIF_MAJOR(vdevice);
 611         *minor = BLKIF_MINOR(vdevice);
 612         switch (major) {
 613                 case XEN_IDE0_MAJOR:
 614                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 615                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 616                                 EMULATED_HD_DISK_MINOR_OFFSET;
 617                         break;
 618                 case XEN_IDE1_MAJOR:
 619                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 620                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 621                                 EMULATED_HD_DISK_MINOR_OFFSET;
 622                         break;
 623                 case XEN_SCSI_DISK0_MAJOR:
 624                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 625                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 626                         break;
 627                 case XEN_SCSI_DISK1_MAJOR:
 628                 case XEN_SCSI_DISK2_MAJOR:
 629                 case XEN_SCSI_DISK3_MAJOR:
 630                 case XEN_SCSI_DISK4_MAJOR:
 631                 case XEN_SCSI_DISK5_MAJOR:
 632                 case XEN_SCSI_DISK6_MAJOR:
 633                 case XEN_SCSI_DISK7_MAJOR:
 634                         *offset = (*minor / PARTS_PER_DISK) +
 635                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 636                                 EMULATED_SD_DISK_NAME_OFFSET;
 637                         *minor = *minor +
 638                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 639                                 EMULATED_SD_DISK_MINOR_OFFSET;
 640                         break;
 641                 case XEN_SCSI_DISK8_MAJOR:
 642                 case XEN_SCSI_DISK9_MAJOR:
 643                 case XEN_SCSI_DISK10_MAJOR:
 644                 case XEN_SCSI_DISK11_MAJOR:
 645                 case XEN_SCSI_DISK12_MAJOR:
 646                 case XEN_SCSI_DISK13_MAJOR:
 647                 case XEN_SCSI_DISK14_MAJOR:
 648                 case XEN_SCSI_DISK15_MAJOR:
 649                         *offset = (*minor / PARTS_PER_DISK) +
 650                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 651                                 EMULATED_SD_DISK_NAME_OFFSET;
 652                         *minor = *minor +
 653                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 654                                 EMULATED_SD_DISK_MINOR_OFFSET;
 655                         break;
 656                 case XENVBD_MAJOR:
 657                         *offset = *minor / PARTS_PER_DISK;
 658                         break;
 659                 default:
 660                         printk(KERN_WARNING "blkfront: your disk configuration is "
 661                                         "incorrect, please use an xvd device instead\n");
 662                         return -ENODEV;
 663         }
 664         return 0;
 665 }
 666
 667 static char *encode_disk_name(char *ptr, unsigned int n)
 668 {
 669         if (n >= 26)
 670                 ptr = encode_disk_name(ptr, n / 26 - 1);
 671         *ptr = 'a' + n % 26;
 672         return ptr + 1;
 673 }
 674
 675 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 676                                struct blkfront_info *info,
 677                                u16 vdisk_info, u16 sector_size)
 678 {
 679         struct gendisk *gd;
 680         int nr_minors = 1;
 681         int err;
 682         unsigned int offset;
 683         int minor;
 684         int nr_parts;
 685         char *ptr;
 686
 687         BUG_ON(info->gd != NULL);
 688         BUG_ON(info->rq != NULL);
 689
 690         if ((info->vdevice>>EXT_SHIFT) > 1) {
 691                 /* this is above the extended range; something is wrong */
 692                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 693                 return -ENODEV;
 694         }
 695
 696         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 697                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 698                 if (err)
 699                         return err;
 700                 nr_parts = PARTS_PER_DISK;
 701         } else {
 702                 minor = BLKIF_MINOR_EXT(info->vdevice);
 703                 nr_parts = PARTS_PER_EXT_DISK;
 704                 offset = minor / nr_parts;
 705                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 706                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 707                                         "emulated IDE disks,\n\t choose an xvd device name"
 708                                         "from xvde on\n", info->vdevice);
 709         }
 710         if (minor >> MINORBITS) {
 711                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 712                         info->vdevice, minor);
 713                 return -ENODEV;
 714         }
 715
 716         if ((minor % nr_parts) == 0)
 717                 nr_minors = nr_parts;
 718
 719         err = xlbd_reserve_minors(minor, nr_minors);
 720         if (err)
 721                 goto out;
 722         err = -ENODEV;
 723
 724         gd = alloc_disk(nr_minors);
 725         if (gd == NULL)
 726                 goto release;
 727
 728         strcpy(gd->disk_name, DEV_NAME);
 729         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 730         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 731         if (nr_minors > 1)
 732                 *ptr = 0;
 733         else
 734                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 735                          "%d", minor & (nr_parts - 1));
 736
 737         gd->major = XENVBD_MAJOR;
 738         gd->first_minor = minor;
 739         gd->fops = &xlvbd_block_fops;
 740         gd->private_data = info;
 741         gd->driverfs_dev = &(info->xbdev->dev);
 742         set_capacity(gd, capacity);
 743
 744         if (xlvbd_init_blk_queue(gd, sector_size)) {
 745                 del_gendisk(gd);
 746                 goto release;
 747         }
 748
 749         info->rq = gd->queue;
 750         info->gd = gd;
 751
 752         xlvbd_flush(info);
 753
 754         if (vdisk_info & VDISK_READONLY)
 755                 set_disk_ro(gd, 1);
 756
 757         if (vdisk_info & VDISK_REMOVABLE)
 758                 gd->flags |= GENHD_FL_REMOVABLE;
 759
 760         if (vdisk_info & VDISK_CDROM)
 761                 gd->flags |= GENHD_FL_CD;
 762
 763         return 0;
 764
 765  release:
 766         xlbd_release_minors(minor, nr_minors);
 767  out:
 768         return err;
 769 }
 770
 771 static void xlvbd_release_gendisk(struct blkfront_info *info)
 772 {
 773         unsigned int minor, nr_minors;
 774         unsigned long flags;
 775
 776         if (info->rq == NULL)
 777                 return;
 778
 779         spin_lock_irqsave(&info->io_lock, flags);
 780
 781         /* No more blkif_request(). */
 782         blk_stop_queue(info->rq);
 783
 784         /* No more gnttab callback work. */
 785         gnttab_cancel_free_callback(&info->callback);
 786         spin_unlock_irqrestore(&info->io_lock, flags);
 787
 788         /* Flush gnttab callback work. Must be done with no locks held. */
 789         flush_work(&info->work);
 790
 791         del_gendisk(info->gd);
 792
 793         minor = info->gd->first_minor;
 794         nr_minors = info->gd->minors;
 795         xlbd_release_minors(minor, nr_minors);
 796
 797         blk_cleanup_queue(info->rq);
 798         info->rq = NULL;
 799
 800         put_disk(info->gd);
 801         info->gd = NULL;
 802 }
 803
 804 static void kick_pending_request_queues(struct blkfront_info *info)
 805 {
 806         if (!RING_FULL(&info->ring)) {
 807                 /* Re-enable calldowns. */
 808                 blk_start_queue(info->rq);
 809                 /* Kick things off immediately. */
 810                 do_blkif_request(info->rq);
 811         }
 812 }
 813
 814 static void blkif_restart_queue(struct work_struct *work)
 815 {
 816         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 817
 818         spin_lock_irq(&info->io_lock);
 819         if (info->connected == BLKIF_STATE_CONNECTED)
 820                 kick_pending_request_queues(info);
 821         spin_unlock_irq(&info->io_lock);
 822 }
 823
 824 static void blkif_free(struct blkfront_info *info, int suspend)
 825 {
 826         struct grant *persistent_gnt;
 827         struct grant *n;
 828
 829         /* Prevent new requests being issued until we fix things up. */
 830         spin_lock_irq(&info->io_lock);
 831         info->connected = suspend ?
 832                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 833         /* No more blkif_request(). */
 834         if (info->rq)
 835                 blk_stop_queue(info->rq);
 836
 837         /* Remove all persistent grants */
 838         if (!list_empty(&info->grants)) {
 839                 list_for_each_entry_safe(persistent_gnt, n,
 840                                          &info->grants, node) {
 841                         list_del(&persistent_gnt->node);
 842                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 843                                 gnttab_end_foreign_access(persistent_gnt->gref,
 844                                                           0, 0UL);
 845                                 info->persistent_gnts_c--;
 846                         }
 847                         if (info->feature_persistent)
 848                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 849                         kfree(persistent_gnt);
 850                 }
 851         }
 852         BUG_ON(info->persistent_gnts_c != 0);
 853
 854         /* No more gnttab callback work. */
 855         gnttab_cancel_free_callback(&info->callback);
 856         spin_unlock_irq(&info->io_lock);
 857
 858         /* Flush gnttab callback work. Must be done with no locks held. */
 859         flush_work(&info->work);
 860
 861         /* Free resources associated with old device channel. */
 862         if (info->ring_ref != GRANT_INVALID_REF) {
 863                 gnttab_end_foreign_access(info->ring_ref, 0,
 864                                           (unsigned long)info->ring.sring);
 865                 info->ring_ref = GRANT_INVALID_REF;
 866                 info->ring.sring = NULL;
 867         }
 868         if (info->irq)
 869                 unbind_from_irqhandler(info->irq, info);
 870         info->evtchn = info->irq = 0;
 871
 872 }
 873
 874 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
 875                              struct blkif_response *bret)
 876 {
 877         int i = 0;
 878         struct scatterlist *sg;
 879         char *bvec_data;
 880         void *shared_data;
 881         int nseg;
 882
 883         nseg = s->req.u.rw.nr_segments;
 884
 885         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
 886                 /*
 887                  * Copy the data received from the backend into the bvec.
 888                  * Since bv_offset can be different than 0, and bv_len different
 889                  * than PAGE_SIZE, we have to keep track of the current offset,
 890                  * to be sure we are copying the data from the right shared page.
 891                  */
 892                 for_each_sg(s->sg, sg, nseg, i) {
 893                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 894                         shared_data = kmap_atomic(
 895                                 pfn_to_page(s->grants_used[i]->pfn));
 896                         bvec_data = kmap_atomic(sg_page(sg));
 897                         memcpy(bvec_data   + sg->offset,
 898                                shared_data + sg->offset,
 899                                sg->length);
 900                         kunmap_atomic(bvec_data);
 901                         kunmap_atomic(shared_data);
 902                 }
 903         }
 904         /* Add the persistent grant into the list of free grants */
 905         for (i = 0; i < nseg; i++) {
 906                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
 907                         /*
 908                          * If the grant is still mapped by the backend (the
 909                          * backend has chosen to make this grant persistent)
 910                          * we add it at the head of the list, so it will be
 911                          * reused first.
 912                          */
 913                         if (!info->feature_persistent)
 914                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
 915                                                      s->grants_used[i]->gref);
 916                         list_add(&s->grants_used[i]->node, &info->grants);
 917                         info->persistent_gnts_c++;
 918                 } else {
 919                         /*
 920                          * If the grant is not mapped by the backend we end the
 921                          * foreign access and add it to the tail of the list,
 922                          * so it will not be picked again unless we run out of
 923                          * persistent grants.
 924                          */
 925                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
 926                         s->grants_used[i]->gref = GRANT_INVALID_REF;
 927                         list_add_tail(&s->grants_used[i]->node, &info->grants);
 928                 }
 929         }
 930 }
 931
 932 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 933 {
 934         struct request *req;
 935         struct blkif_response *bret;
 936         RING_IDX i, rp;
 937         unsigned long flags;
 938         struct blkfront_info *info = (struct blkfront_info *)dev_id;
 939         int error;
 940
 941         spin_lock_irqsave(&info->io_lock, flags);
 942
 943         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
 944                 spin_unlock_irqrestore(&info->io_lock, flags);
 945                 return IRQ_HANDLED;
 946         }
 947
 948  again:
 949         rp = info->ring.sring->rsp_prod;
 950         rmb(); /* Ensure we see queued responses up to 'rp'. */
 951
 952         for (i = info->ring.rsp_cons; i != rp; i++) {
 953                 unsigned long id;
 954
 955                 bret = RING_GET_RESPONSE(&info->ring, i);
 956                 id   = bret->id;
 957                 /*
 958                  * The backend has messed up and given us an id that we would
 959                  * never have given to it (we stamp it up to BLK_RING_SIZE -
 960                  * look in get_id_from_freelist.
 961                  */
 962                 if (id >= BLK_RING_SIZE) {
 963                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
 964                              info->gd->disk_name, op_name(bret->operation), id);
 965                         /* We can't safely get the 'struct request' as
 966                          * the id is busted. */
 967                         continue;
 968                 }
 969                 req  = info->shadow[id].request;
 970
 971                 if (bret->operation != BLKIF_OP_DISCARD)
 972                         blkif_completion(&info->shadow[id], info, bret);
 973
 974                 if (add_id_to_freelist(info, id)) {
 975                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
 976                              info->gd->disk_name, op_name(bret->operation), id);
 977                         continue;
 978                 }
 979
 980                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
 981                 switch (bret->operation) {
 982                 case BLKIF_OP_DISCARD:
 983                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 984                                 struct request_queue *rq = info->rq;
 985                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 986                                            info->gd->disk_name, op_name(bret->operation));
 987                                 error = -EOPNOTSUPP;
 988                                 info->feature_discard = 0;
 989                                 info->feature_secdiscard = 0;
 990                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
 991                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
 992                         }
 993                         __blk_end_request_all(req, error);
 994                         break;
 995                 case BLKIF_OP_FLUSH_DISKCACHE:
 996                 case BLKIF_OP_WRITE_BARRIER:
 997                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
 998                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
 999                                        info->gd->disk_name, op_name(bret->operation));
1000                                 error = -EOPNOTSUPP;
1001                         }
1002                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1003                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
1004                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1005                                        info->gd->disk_name, op_name(bret->operation));
1006                                 error = -EOPNOTSUPP;
1007                         }
1008                         if (unlikely(error)) {
1009                                 if (error == -EOPNOTSUPP)
1010                                         error = 0;
1011                                 info->feature_flush = 0;
1012                                 info->flush_op = 0;
1013                                 xlvbd_flush(info);
1014                         }
1015                         /* fall through */
1016                 case BLKIF_OP_READ:
1017                 case BLKIF_OP_WRITE:
1018                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1019                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1020                                         "request: %x\n", bret->status);
1021
1022                         __blk_end_request_all(req, error);
1023                         break;
1024                 default:
1025                         BUG();
1026                 }
1027         }
1028
1029         info->ring.rsp_cons = i;
1030
1031         if (i != info->ring.req_prod_pvt) {
1032                 int more_to_do;
1033                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1034                 if (more_to_do)
1035                         goto again;
1036         } else
1037                 info->ring.sring->rsp_event = i + 1;
1038
1039         kick_pending_request_queues(info);
1040
1041         spin_unlock_irqrestore(&info->io_lock, flags);
1042
1043         return IRQ_HANDLED;
1044 }
1045
1046
1047 static int setup_blkring(struct xenbus_device *dev,
1048                          struct blkfront_info *info)
1049 {
1050         struct blkif_sring *sring;
1051         int err, i;
1052
1053         info->ring_ref = GRANT_INVALID_REF;
1054
1055         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1056         if (!sring) {
1057                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1058                 return -ENOMEM;
1059         }
1060         SHARED_RING_INIT(sring);
1061         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1062
1063         for (i = 0; i < BLK_RING_SIZE; i++)
1064                 sg_init_table(info->shadow[i].sg, BLKIF_MAX_SEGMENTS_PER_REQUEST);
1065
1066         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1067         if (err < 0) {
1068                 free_page((unsigned long)sring);
1069                 info->ring.sring = NULL;
1070                 goto fail;
1071         }
1072         info->ring_ref = err;
1073
1074         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1075         if (err)
1076                 goto fail;
1077
1078         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1079                                         "blkif", info);
1080         if (err <= 0) {
1081                 xenbus_dev_fatal(dev, err,
1082                                  "bind_evtchn_to_irqhandler failed");
1083                 goto fail;
1084         }
1085         info->irq = err;
1086
1087         return 0;
1088 fail:
1089         blkif_free(info, 0);
1090         return err;
1091 }
1092
1093
1094 /* Common code used when first setting up, and when resuming. */
1095 static int talk_to_blkback(struct xenbus_device *dev,
1096                            struct blkfront_info *info)
1097 {
1098         const char *message = NULL;
1099         struct xenbus_transaction xbt;
1100         int err;
1101
1102         /* Create shared ring, alloc event channel. */
1103         err = setup_blkring(dev, info);
1104         if (err)
1105                 goto out;
1106
1107 again:
1108         err = xenbus_transaction_start(&xbt);
1109         if (err) {
1110                 xenbus_dev_fatal(dev, err, "starting transaction");
1111                 goto destroy_blkring;
1112         }
1113
1114         err = xenbus_printf(xbt, dev->nodename,
1115                             "ring-ref", "%u", info->ring_ref);
1116         if (err) {
1117                 message = "writing ring-ref";
1118                 goto abort_transaction;
1119         }
1120         err = xenbus_printf(xbt, dev->nodename,
1121                             "event-channel", "%u", info->evtchn);
1122         if (err) {
1123                 message = "writing event-channel";
1124                 goto abort_transaction;
1125         }
1126         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1127                             XEN_IO_PROTO_ABI_NATIVE);
1128         if (err) {
1129                 message = "writing protocol";
1130                 goto abort_transaction;
1131         }
1132         err = xenbus_printf(xbt, dev->nodename,
1133                             "feature-persistent", "%u", 1);
1134         if (err)
1135                 dev_warn(&dev->dev,
1136                          "writing persistent grants feature to xenbus");
1137
1138         err = xenbus_transaction_end(xbt, 0);
1139         if (err) {
1140                 if (err == -EAGAIN)
1141                         goto again;
1142                 xenbus_dev_fatal(dev, err, "completing transaction");
1143                 goto destroy_blkring;
1144         }
1145
1146         xenbus_switch_state(dev, XenbusStateInitialised);
1147
1148         return 0;
1149
1150  abort_transaction:
1151         xenbus_transaction_end(xbt, 1);
1152         if (message)
1153                 xenbus_dev_fatal(dev, err, "%s", message);
1154  destroy_blkring:
1155         blkif_free(info, 0);
1156  out:
1157         return err;
1158 }
1159
1160 /**
1161  * Entry point to this code when a new device is created.  Allocate the basic
1162  * structures and the ring buffer for communication with the backend, and
1163  * inform the backend of the appropriate details for those.  Switch to
1164  * Initialised state.
1165  */
1166 static int blkfront_probe(struct xenbus_device *dev,
1167                           const struct xenbus_device_id *id)
1168 {
1169         int err, vdevice, i;
1170         struct blkfront_info *info;
1171
1172         /* FIXME: Use dynamic device id if this is not set. */
1173         err = xenbus_scanf(XBT_NIL, dev->nodename,
1174                            "virtual-device", "%i", &vdevice);
1175         if (err != 1) {
1176                 /* go looking in the extended area instead */
1177                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1178                                    "%i", &vdevice);
1179                 if (err != 1) {
1180                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1181                         return err;
1182                 }
1183         }
1184
1185         if (xen_hvm_domain()) {
1186                 char *type;
1187                 int len;
1188                 /* no unplug has been done: do not hook devices != xen vbds */
1189                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1190                         int major;
1191
1192                         if (!VDEV_IS_EXTENDED(vdevice))
1193                                 major = BLKIF_MAJOR(vdevice);
1194                         else
1195                                 major = XENVBD_MAJOR;
1196
1197                         if (major != XENVBD_MAJOR) {
1198                                 printk(KERN_INFO
1199                                                 "%s: HVM does not support vbd %d as xen block device\n",
1200                                                 __FUNCTION__, vdevice);
1201                                 return -ENODEV;
1202                         }
1203                 }
1204                 /* do not create a PV cdrom device if we are an HVM guest */
1205                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1206                 if (IS_ERR(type))
1207                         return -ENODEV;
1208                 if (strncmp(type, "cdrom", 5) == 0) {
1209                         kfree(type);
1210                         return -ENODEV;
1211                 }
1212                 kfree(type);
1213         }
1214         info = kzalloc(sizeof(*info), GFP_KERNEL);
1215         if (!info) {
1216                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1217                 return -ENOMEM;
1218         }
1219
1220         mutex_init(&info->mutex);
1221         spin_lock_init(&info->io_lock);
1222         info->xbdev = dev;
1223         info->vdevice = vdevice;
1224         INIT_LIST_HEAD(&info->grants);
1225         info->persistent_gnts_c = 0;
1226         info->connected = BLKIF_STATE_DISCONNECTED;
1227         INIT_WORK(&info->work, blkif_restart_queue);
1228
1229         for (i = 0; i < BLK_RING_SIZE; i++)
1230                 info->shadow[i].req.u.rw.id = i+1;
1231         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1232
1233         /* Front end dir is a number, which is used as the id. */
1234         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1235         dev_set_drvdata(&dev->dev, info);
1236
1237         err = talk_to_blkback(dev, info);
1238         if (err) {
1239                 kfree(info);
1240                 dev_set_drvdata(&dev->dev, NULL);
1241                 return err;
1242         }
1243
1244         return 0;
1245 }
1246
1247
1248 static int blkif_recover(struct blkfront_info *info)
1249 {
1250         int i;
1251         struct blkif_request *req;
1252         struct blk_shadow *copy;
1253         unsigned int persistent;
1254         int j, rc;
1255
1256         /* Stage 1: Make a safe copy of the shadow state. */
1257         copy = kmemdup(info->shadow, sizeof(info->shadow),
1258                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1259         if (!copy)
1260                 return -ENOMEM;
1261
1262         /* Stage 2: Set up free list. */
1263         memset(&info->shadow, 0, sizeof(info->shadow));
1264         for (i = 0; i < BLK_RING_SIZE; i++)
1265                 info->shadow[i].req.u.rw.id = i+1;
1266         info->shadow_free = info->ring.req_prod_pvt;
1267         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1268
1269         /* Check if the backend supports persistent grants */
1270         rc = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1271                            "feature-persistent", "%u", &persistent,
1272                            NULL);
1273         if (rc)
1274                 info->feature_persistent = 0;
1275         else
1276                 info->feature_persistent = persistent;
1277
1278         /* Allocate memory for grants */
1279         rc = fill_grant_buffer(info, BLK_RING_SIZE *
1280                                      BLKIF_MAX_SEGMENTS_PER_REQUEST);
1281         if (rc) {
1282                 xenbus_dev_fatal(info->xbdev, rc, "setting grant buffer failed");
1283                 kfree(copy);
1284                 return rc;
1285         }
1286
1287         /* Stage 3: Find pending requests and requeue them. */
1288         for (i = 0; i < BLK_RING_SIZE; i++) {
1289                 /* Not in use? */
1290                 if (!copy[i].request)
1291                         continue;
1292
1293                 /* Grab a request slot and copy shadow state into it. */
1294                 req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
1295                 *req = copy[i].req;
1296
1297                 /* We get a new request id, and must reset the shadow state. */
1298                 req->u.rw.id = get_id_from_freelist(info);
1299                 memcpy(&info->shadow[req->u.rw.id], &copy[i], sizeof(copy[i]));
1300
1301                 if (req->operation != BLKIF_OP_DISCARD) {
1302                 /* Rewrite any grant references invalidated by susp/resume. */
1303                         for (j = 0; j < req->u.rw.nr_segments; j++)
1304                                 gnttab_grant_foreign_access_ref(
1305                                         req->u.rw.seg[j].gref,
1306                                         info->xbdev->otherend_id,
1307                                         pfn_to_mfn(copy[i].grants_used[j]->pfn),
1308                                         0);
1309                 }
1310                 info->shadow[req->u.rw.id].req = *req;
1311
1312                 info->ring.req_prod_pvt++;
1313         }
1314
1315         kfree(copy);
1316
1317         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1318
1319         spin_lock_irq(&info->io_lock);
1320
1321         /* Now safe for us to use the shared ring */
1322         info->connected = BLKIF_STATE_CONNECTED;
1323
1324         /* Send off requeued requests */
1325         flush_requests(info);
1326
1327         /* Kick any other new requests queued since we resumed */
1328         kick_pending_request_queues(info);
1329
1330         spin_unlock_irq(&info->io_lock);
1331
1332         return 0;
1333 }
1334
1335 /**
1336  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1337  * driver restart.  We tear down our blkif structure and recreate it, but
1338  * leave the device-layer structures intact so that this is transparent to the
1339  * rest of the kernel.
1340  */
1341 static int blkfront_resume(struct xenbus_device *dev)
1342 {
1343         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1344         int err;
1345
1346         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1347
1348         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1349
1350         err = talk_to_blkback(dev, info);
1351
1352         /*
1353          * We have to wait for the backend to switch to
1354          * connected state, since we want to read which
1355          * features it supports.
1356          */
1357
1358         return err;
1359 }
1360
1361 static void
1362 blkfront_closing(struct blkfront_info *info)
1363 {
1364         struct xenbus_device *xbdev = info->xbdev;
1365         struct block_device *bdev = NULL;
1366
1367         mutex_lock(&info->mutex);
1368
1369         if (xbdev->state == XenbusStateClosing) {
1370                 mutex_unlock(&info->mutex);
1371                 return;
1372         }
1373
1374         if (info->gd)
1375                 bdev = bdget_disk(info->gd, 0);
1376
1377         mutex_unlock(&info->mutex);
1378
1379         if (!bdev) {
1380                 xenbus_frontend_closed(xbdev);
1381                 return;
1382         }
1383
1384         mutex_lock(&bdev->bd_mutex);
1385
1386         if (bdev->bd_openers) {
1387                 xenbus_dev_error(xbdev, -EBUSY,
1388                                  "Device in use; refusing to close");
1389                 xenbus_switch_state(xbdev, XenbusStateClosing);
1390         } else {
1391                 xlvbd_release_gendisk(info);
1392                 xenbus_frontend_closed(xbdev);
1393         }
1394
1395         mutex_unlock(&bdev->bd_mutex);
1396         bdput(bdev);
1397 }
1398
1399 static void blkfront_setup_discard(struct blkfront_info *info)
1400 {
1401         int err;
1402         char *type;
1403         unsigned int discard_granularity;
1404         unsigned int discard_alignment;
1405         unsigned int discard_secure;
1406
1407         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1408         if (IS_ERR(type))
1409                 return;
1410
1411         info->feature_secdiscard = 0;
1412         if (strncmp(type, "phy", 3) == 0) {
1413                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1414                         "discard-granularity", "%u", &discard_granularity,
1415                         "discard-alignment", "%u", &discard_alignment,
1416                         NULL);
1417                 if (!err) {
1418                         info->feature_discard = 1;
1419                         info->discard_granularity = discard_granularity;
1420                         info->discard_alignment = discard_alignment;
1421                 }
1422                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1423                             "discard-secure", "%d", &discard_secure,
1424                             NULL);
1425                 if (!err)
1426                         info->feature_secdiscard = discard_secure;
1427
1428         } else if (strncmp(type, "file", 4) == 0)
1429                 info->feature_discard = 1;
1430
1431         kfree(type);
1432 }
1433
1434 /*
1435  * Invoked when the backend is finally 'ready' (and has told produced
1436  * the details about the physical device - #sectors, size, etc).
1437  */
1438 static void blkfront_connect(struct blkfront_info *info)
1439 {
1440         unsigned long long sectors;
1441         unsigned long sector_size;
1442         unsigned int binfo;
1443         int err;
1444         int barrier, flush, discard, persistent;
1445
1446         switch (info->connected) {
1447         case BLKIF_STATE_CONNECTED:
1448                 /*
1449                  * Potentially, the back-end may be signalling
1450                  * a capacity change; update the capacity.
1451                  */
1452                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1453                                    "sectors", "%Lu", &sectors);
1454                 if (XENBUS_EXIST_ERR(err))
1455                         return;
1456                 printk(KERN_INFO "Setting capacity to %Lu\n",
1457                        sectors);
1458                 set_capacity(info->gd, sectors);
1459                 revalidate_disk(info->gd);
1460                 return;
1461
1462         case BLKIF_STATE_SUSPENDED:
1463                 /*
1464                  * If we are recovering from suspension, we need to wait
1465                  * for the backend to announce it's features before
1466                  * reconnecting, we need to know if the backend supports
1467                  * persistent grants.
1468                  */
1469                 blkif_recover(info);
1470                 return;
1471
1472         default:
1473                 break;
1474         }
1475
1476         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1477                 __func__, info->xbdev->otherend);
1478
1479         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1480                             "sectors", "%llu", &sectors,
1481                             "info", "%u", &binfo,
1482                             "sector-size", "%lu", &sector_size,
1483                             NULL);
1484         if (err) {
1485                 xenbus_dev_fatal(info->xbdev, err,
1486                                  "reading backend fields at %s",
1487                                  info->xbdev->otherend);
1488                 return;
1489         }
1490
1491         info->feature_flush = 0;
1492         info->flush_op = 0;
1493
1494         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1495                             "feature-barrier", "%d", &barrier,
1496                             NULL);
1497
1498         /*
1499          * If there's no "feature-barrier" defined, then it means
1500          * we're dealing with a very old backend which writes
1501          * synchronously; nothing to do.
1502          *
1503          * If there are barriers, then we use flush.
1504          */
1505         if (!err && barrier) {
1506                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1507                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1508         }
1509         /*
1510          * And if there is "feature-flush-cache" use that above
1511          * barriers.
1512          */
1513         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1514                             "feature-flush-cache", "%d", &flush,
1515                             NULL);
1516
1517         if (!err && flush) {
1518                 info->feature_flush = REQ_FLUSH;
1519                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1520         }
1521
1522         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1523                             "feature-discard", "%d", &discard,
1524                             NULL);
1525
1526         if (!err && discard)
1527                 blkfront_setup_discard(info);
1528
1529         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1530                             "feature-persistent", "%u", &persistent,
1531                             NULL);
1532         if (err)
1533                 info->feature_persistent = 0;
1534         else
1535                 info->feature_persistent = persistent;
1536
1537         /* Allocate memory for grants */
1538         err = fill_grant_buffer(info, BLK_RING_SIZE *
1539                                       BLKIF_MAX_SEGMENTS_PER_REQUEST);
1540         if (err) {
1541                 xenbus_dev_fatal(info->xbdev, err, "setting grant buffer failed");
1542                 return;
1543         }
1544
1545         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
1546         if (err) {
1547                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1548                                  info->xbdev->otherend);
1549                 return;
1550         }
1551
1552         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1553
1554         /* Kick pending requests. */
1555         spin_lock_irq(&info->io_lock);
1556         info->connected = BLKIF_STATE_CONNECTED;
1557         kick_pending_request_queues(info);
1558         spin_unlock_irq(&info->io_lock);
1559
1560         add_disk(info->gd);
1561
1562         info->is_ready = 1;
1563 }
1564
1565 /**
1566  * Callback received when the backend's state changes.
1567  */
1568 static void blkback_changed(struct xenbus_device *dev,
1569                             enum xenbus_state backend_state)
1570 {
1571         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1572
1573         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1574
1575         switch (backend_state) {
1576         case XenbusStateInitialising:
1577         case XenbusStateInitWait:
1578         case XenbusStateInitialised:
1579         case XenbusStateReconfiguring:
1580         case XenbusStateReconfigured:
1581         case XenbusStateUnknown:
1582                 break;
1583
1584         case XenbusStateConnected:
1585                 blkfront_connect(info);
1586                 break;
1587
1588         case XenbusStateClosed:
1589                 if (dev->state == XenbusStateClosed)
1590                         break;
1591                 /* Missed the backend's Closing state -- fallthrough */
1592         case XenbusStateClosing:
1593                 blkfront_closing(info);
1594                 break;
1595         }
1596 }
1597
1598 static int blkfront_remove(struct xenbus_device *xbdev)
1599 {
1600         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1601         struct block_device *bdev = NULL;
1602         struct gendisk *disk;
1603
1604         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1605
1606         blkif_free(info, 0);
1607
1608         mutex_lock(&info->mutex);
1609
1610         disk = info->gd;
1611         if (disk)
1612                 bdev = bdget_disk(disk, 0);
1613
1614         info->xbdev = NULL;
1615         mutex_unlock(&info->mutex);
1616
1617         if (!bdev) {
1618                 kfree(info);
1619                 return 0;
1620         }
1621
1622         /*
1623          * The xbdev was removed before we reached the Closed
1624          * state. See if it's safe to remove the disk. If the bdev
1625          * isn't closed yet, we let release take care of it.
1626          */
1627
1628         mutex_lock(&bdev->bd_mutex);
1629         info = disk->private_data;
1630
1631         dev_warn(disk_to_dev(disk),
1632                  "%s was hot-unplugged, %d stale handles\n",
1633                  xbdev->nodename, bdev->bd_openers);
1634
1635         if (info && !bdev->bd_openers) {
1636                 xlvbd_release_gendisk(info);
1637                 disk->private_data = NULL;
1638                 kfree(info);
1639         }
1640
1641         mutex_unlock(&bdev->bd_mutex);
1642         bdput(bdev);
1643
1644         return 0;
1645 }
1646
1647 static int blkfront_is_ready(struct xenbus_device *dev)
1648 {
1649         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1650
1651         return info->is_ready && info->xbdev;
1652 }
1653
1654 static int blkif_open(struct block_device *bdev, fmode_t mode)
1655 {
1656         struct gendisk *disk = bdev->bd_disk;
1657         struct blkfront_info *info;
1658         int err = 0;
1659
1660         mutex_lock(&blkfront_mutex);
1661
1662         info = disk->private_data;
1663         if (!info) {
1664                 /* xbdev gone */
1665                 err = -ERESTARTSYS;
1666                 goto out;
1667         }
1668
1669         mutex_lock(&info->mutex);
1670
1671         if (!info->gd)
1672                 /* xbdev is closed */
1673                 err = -ERESTARTSYS;
1674
1675         mutex_unlock(&info->mutex);
1676
1677 out:
1678         mutex_unlock(&blkfront_mutex);
1679         return err;
1680 }
1681
1682 static void blkif_release(struct gendisk *disk, fmode_t mode)
1683 {
1684         struct blkfront_info *info = disk->private_data;
1685         struct block_device *bdev;
1686         struct xenbus_device *xbdev;
1687
1688         mutex_lock(&blkfront_mutex);
1689
1690         bdev = bdget_disk(disk, 0);
1691
1692         if (bdev->bd_openers)
1693                 goto out;
1694
1695         /*
1696          * Check if we have been instructed to close. We will have
1697          * deferred this request, because the bdev was still open.
1698          */
1699
1700         mutex_lock(&info->mutex);
1701         xbdev = info->xbdev;
1702
1703         if (xbdev && xbdev->state == XenbusStateClosing) {
1704                 /* pending switch to state closed */
1705                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1706                 xlvbd_release_gendisk(info);
1707                 xenbus_frontend_closed(info->xbdev);
1708         }
1709
1710         mutex_unlock(&info->mutex);
1711
1712         if (!xbdev) {
1713                 /* sudden device removal */
1714                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
1715                 xlvbd_release_gendisk(info);
1716                 disk->private_data = NULL;
1717                 kfree(info);
1718         }
1719
1720 out:
1721         bdput(bdev);
1722         mutex_unlock(&blkfront_mutex);
1723 }
1724
1725 static const struct block_device_operations xlvbd_block_fops =
1726 {
1727         .owner = THIS_MODULE,
1728         .open = blkif_open,
1729         .release = blkif_release,
1730         .getgeo = blkif_getgeo,
1731         .ioctl = blkif_ioctl,
1732 };
1733
1734
1735 static const struct xenbus_device_id blkfront_ids[] = {
1736         { "vbd" },
1737         { "" }
1738 };
1739
1740 static DEFINE_XENBUS_DRIVER(blkfront, ,
1741         .probe = blkfront_probe,
1742         .remove = blkfront_remove,
1743         .resume = blkfront_resume,
1744         .otherend_changed = blkback_changed,
1745         .is_ready = blkfront_is_ready,
1746 );
1747
1748 static int __init xlblk_init(void)
1749 {
1750         int ret;
1751
1752         if (!xen_domain())
1753                 return -ENODEV;
1754
1755         if (xen_hvm_domain() && !xen_platform_pci_unplug)
1756                 return -ENODEV;
1757
1758         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
1759                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
1760                        XENVBD_MAJOR, DEV_NAME);
1761                 return -ENODEV;
1762         }
1763
1764         ret = xenbus_register_frontend(&blkfront_driver);
1765         if (ret) {
1766                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1767                 return ret;
1768         }
1769
1770         return 0;
1771 }
1772 module_init(xlblk_init);
1773
1774
1775 static void __exit xlblk_exit(void)
1776 {
1777         xenbus_unregister_driver(&blkfront_driver);
1778         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
1779         kfree(minors);
1780 }
1781 module_exit(xlblk_exit);
1782
1783 MODULE_DESCRIPTION("Xen virtual block device frontend");
1784 MODULE_LICENSE("GPL");
1785 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
1786 MODULE_ALIAS("xen:vbd");
1787 MODULE_ALIAS("xenblk");