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;
  78         struct grant **indirect_grants;
  79         struct scatterlist *sg;
  80 };
  81
  82 struct split_bio {
  83         struct bio *bio;
  84         atomic_t pending;
  85         int err;
  86 };
  87
  88 static DEFINE_MUTEX(blkfront_mutex);
  89 static const struct block_device_operations xlvbd_block_fops;
  90
  91 /*
  92  * Maximum number of segments in indirect requests, the actual value used by
  93  * the frontend driver is the minimum of this value and the value provided
  94  * by the backend driver.
  95  */
  96
  97 static unsigned int xen_blkif_max_segments = 32;
  98 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
  99 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
 100
 101 #define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
 102
 103 /*
 104  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 105  * hang in private_data off the gendisk structure. We may end up
 106  * putting all kinds of interesting stuff here :-)
 107  */
 108 struct blkfront_info
 109 {
 110         spinlock_t io_lock;
 111         struct mutex mutex;
 112         struct xenbus_device *xbdev;
 113         struct gendisk *gd;
 114         int vdevice;
 115         blkif_vdev_t handle;
 116         enum blkif_state connected;
 117         int ring_ref;
 118         struct blkif_front_ring ring;
 119         unsigned int evtchn, irq;
 120         struct request_queue *rq;
 121         struct work_struct work;
 122         struct gnttab_free_callback callback;
 123         struct blk_shadow shadow[BLK_RING_SIZE];
 124         struct list_head grants;
 125         struct list_head indirect_pages;
 126         unsigned int persistent_gnts_c;
 127         unsigned long shadow_free;
 128         unsigned int feature_flush;
 129         unsigned int flush_op;
 130         unsigned int feature_discard:1;
 131         unsigned int feature_secdiscard:1;
 132         unsigned int discard_granularity;
 133         unsigned int discard_alignment;
 134         unsigned int feature_persistent:1;
 135         unsigned int max_indirect_segments;
 136         int is_ready;
 137 };
 138
 139 static unsigned int nr_minors;
 140 static unsigned long *minors;
 141 static DEFINE_SPINLOCK(minor_lock);
 142
 143 #define MAXIMUM_OUTSTANDING_BLOCK_REQS \
 144         (BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 145 #define GRANT_INVALID_REF       0
 146
 147 #define PARTS_PER_DISK          16
 148 #define PARTS_PER_EXT_DISK      256
 149
 150 #define BLKIF_MAJOR(dev) ((dev)>>8)
 151 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 152
 153 #define EXT_SHIFT 28
 154 #define EXTENDED (1<<EXT_SHIFT)
 155 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 156 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 157 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 158 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 159 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 160 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 161
 162 #define DEV_NAME        "xvd"   /* name in /dev */
 163
 164 #define SEGS_PER_INDIRECT_FRAME \
 165         (PAGE_SIZE/sizeof(struct blkif_request_segment))
 166 #define INDIRECT_GREFS(_segs) \
 167         ((_segs + SEGS_PER_INDIRECT_FRAME - 1)/SEGS_PER_INDIRECT_FRAME)
 168
 169 static int blkfront_setup_indirect(struct blkfront_info *info);
 170
 171 static int get_id_from_freelist(struct blkfront_info *info)
 172 {
 173         unsigned long free = info->shadow_free;
 174         BUG_ON(free >= BLK_RING_SIZE);
 175         info->shadow_free = info->shadow[free].req.u.rw.id;
 176         info->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 177         return free;
 178 }
 179
 180 static int add_id_to_freelist(struct blkfront_info *info,
 181                                unsigned long id)
 182 {
 183         if (info->shadow[id].req.u.rw.id != id)
 184                 return -EINVAL;
 185         if (info->shadow[id].request == NULL)
 186                 return -EINVAL;
 187         info->shadow[id].req.u.rw.id  = info->shadow_free;
 188         info->shadow[id].request = NULL;
 189         info->shadow_free = id;
 190         return 0;
 191 }
 192
 193 static int fill_grant_buffer(struct blkfront_info *info, int num)
 194 {
 195         struct page *granted_page;
 196         struct grant *gnt_list_entry, *n;
 197         int i = 0;
 198
 199         while(i < num) {
 200                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 201                 if (!gnt_list_entry)
 202                         goto out_of_memory;
 203
 204                 if (info->feature_persistent) {
 205                         granted_page = alloc_page(GFP_NOIO);
 206                         if (!granted_page) {
 207                                 kfree(gnt_list_entry);
 208                                 goto out_of_memory;
 209                         }
 210                         gnt_list_entry->pfn = page_to_pfn(granted_page);
 211                 }
 212
 213                 gnt_list_entry->gref = GRANT_INVALID_REF;
 214                 list_add(&gnt_list_entry->node, &info->grants);
 215                 i++;
 216         }
 217
 218         return 0;
 219
 220 out_of_memory:
 221         list_for_each_entry_safe(gnt_list_entry, n,
 222                                  &info->grants, node) {
 223                 list_del(&gnt_list_entry->node);
 224                 if (info->feature_persistent)
 225                         __free_page(pfn_to_page(gnt_list_entry->pfn));
 226                 kfree(gnt_list_entry);
 227                 i--;
 228         }
 229         BUG_ON(i != 0);
 230         return -ENOMEM;
 231 }
 232
 233 static struct grant *get_grant(grant_ref_t *gref_head,
 234                                unsigned long pfn,
 235                                struct blkfront_info *info)
 236 {
 237         struct grant *gnt_list_entry;
 238         unsigned long buffer_mfn;
 239
 240         BUG_ON(list_empty(&info->grants));
 241         gnt_list_entry = list_first_entry(&info->grants, struct grant,
 242                                           node);
 243         list_del(&gnt_list_entry->node);
 244
 245         if (gnt_list_entry->gref != GRANT_INVALID_REF) {
 246                 info->persistent_gnts_c--;
 247                 return gnt_list_entry;
 248         }
 249
 250         /* Assign a gref to this page */
 251         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 252         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 253         if (!info->feature_persistent) {
 254                 BUG_ON(!pfn);
 255                 gnt_list_entry->pfn = pfn;
 256         }
 257         buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
 258         gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 259                                         info->xbdev->otherend_id,
 260                                         buffer_mfn, 0);
 261         return gnt_list_entry;
 262 }
 263
 264 static const char *op_name(int op)
 265 {
 266         static const char *const names[] = {
 267                 [BLKIF_OP_READ] = "read",
 268                 [BLKIF_OP_WRITE] = "write",
 269                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 270                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 271                 [BLKIF_OP_DISCARD] = "discard" };
 272
 273         if (op < 0 || op >= ARRAY_SIZE(names))
 274                 return "unknown";
 275
 276         if (!names[op])
 277                 return "reserved";
 278
 279         return names[op];
 280 }
 281 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 282 {
 283         unsigned int end = minor + nr;
 284         int rc;
 285
 286         if (end > nr_minors) {
 287                 unsigned long *bitmap, *old;
 288
 289                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 290                                  GFP_KERNEL);
 291                 if (bitmap == NULL)
 292                         return -ENOMEM;
 293
 294                 spin_lock(&minor_lock);
 295                 if (end > nr_minors) {
 296                         old = minors;
 297                         memcpy(bitmap, minors,
 298                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 299                         minors = bitmap;
 300                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 301                 } else
 302                         old = bitmap;
 303                 spin_unlock(&minor_lock);
 304                 kfree(old);
 305         }
 306
 307         spin_lock(&minor_lock);
 308         if (find_next_bit(minors, end, minor) >= end) {
 309                 bitmap_set(minors, minor, nr);
 310                 rc = 0;
 311         } else
 312                 rc = -EBUSY;
 313         spin_unlock(&minor_lock);
 314
 315         return rc;
 316 }
 317
 318 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 319 {
 320         unsigned int end = minor + nr;
 321
 322         BUG_ON(end > nr_minors);
 323         spin_lock(&minor_lock);
 324         bitmap_clear(minors,  minor, nr);
 325         spin_unlock(&minor_lock);
 326 }
 327
 328 static void blkif_restart_queue_callback(void *arg)
 329 {
 330         struct blkfront_info *info = (struct blkfront_info *)arg;
 331         schedule_work(&info->work);
 332 }
 333
 334 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 335 {
 336         /* We don't have real geometry info, but let's at least return
 337            values consistent with the size of the device */
 338         sector_t nsect = get_capacity(bd->bd_disk);
 339         sector_t cylinders = nsect;
 340
 341         hg->heads = 0xff;
 342         hg->sectors = 0x3f;
 343         sector_div(cylinders, hg->heads * hg->sectors);
 344         hg->cylinders = cylinders;
 345         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 346                 hg->cylinders = 0xffff;
 347         return 0;
 348 }
 349
 350 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 351                        unsigned command, unsigned long argument)
 352 {
 353         struct blkfront_info *info = bdev->bd_disk->private_data;
 354         int i;
 355
 356         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 357                 command, (long)argument);
 358
 359         switch (command) {
 360         case CDROMMULTISESSION:
 361                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 362                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 363                         if (put_user(0, (char __user *)(argument + i)))
 364                                 return -EFAULT;
 365                 return 0;
 366
 367         case CDROM_GET_CAPABILITY: {
 368                 struct gendisk *gd = info->gd;
 369                 if (gd->flags & GENHD_FL_CD)
 370                         return 0;
 371                 return -EINVAL;
 372         }
 373
 374         default:
 375                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 376                   command);*/
 377                 return -EINVAL; /* same return as native Linux */
 378         }
 379
 380         return 0;
 381 }
 382
 383 /*
 384  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 385  * and writes are handled as expected.
 386  *
 387  * @req: a request struct
 388  */
 389 static int blkif_queue_request(struct request *req)
 390 {
 391         struct blkfront_info *info = req->rq_disk->private_data;
 392         struct blkif_request *ring_req;
 393         unsigned long id;
 394         unsigned int fsect, lsect;
 395         int i, ref, n;
 396         struct blkif_request_segment *segments = NULL;
 397
 398         /*
 399          * Used to store if we are able to queue the request by just using
 400          * existing persistent grants, or if we have to get new grants,
 401          * as there are not sufficiently many free.
 402          */
 403         bool new_persistent_gnts;
 404         grant_ref_t gref_head;
 405         struct grant *gnt_list_entry = NULL;
 406         struct scatterlist *sg;
 407         int nseg, max_grefs;
 408
 409         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 410                 return 1;
 411
 412         max_grefs = req->nr_phys_segments;
 413         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 414                 /*
 415                  * If we are using indirect segments we need to account
 416                  * for the indirect grefs used in the request.
 417                  */
 418                 max_grefs += INDIRECT_GREFS(req->nr_phys_segments);
 419
 420         /* Check if we have enough grants to allocate a requests */
 421         if (info->persistent_gnts_c < max_grefs) {
 422                 new_persistent_gnts = 1;
 423                 if (gnttab_alloc_grant_references(
 424                     max_grefs - info->persistent_gnts_c,
 425                     &gref_head) < 0) {
 426                         gnttab_request_free_callback(
 427                                 &info->callback,
 428                                 blkif_restart_queue_callback,
 429                                 info,
 430                                 max_grefs);
 431                         return 1;
 432                 }
 433         } else
 434                 new_persistent_gnts = 0;
 435
 436         /* Fill out a communications ring structure. */
 437         ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
 438         id = get_id_from_freelist(info);
 439         info->shadow[id].request = req;
 440
 441         if (unlikely(req->cmd_flags & (REQ_DISCARD | REQ_SECURE))) {
 442                 ring_req->operation = BLKIF_OP_DISCARD;
 443                 ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 444                 ring_req->u.discard.id = id;
 445                 ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 446                 if ((req->cmd_flags & REQ_SECURE) && info->feature_secdiscard)
 447                         ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 448                 else
 449                         ring_req->u.discard.flag = 0;
 450         } else {
 451                 BUG_ON(info->max_indirect_segments == 0 &&
 452                        req->nr_phys_segments > BLKIF_MAX_SEGMENTS_PER_REQUEST);
 453                 BUG_ON(info->max_indirect_segments &&
 454                        req->nr_phys_segments > info->max_indirect_segments);
 455                 nseg = blk_rq_map_sg(req->q, req, info->shadow[id].sg);
 456                 ring_req->u.rw.id = id;
 457                 if (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 458                         /*
 459                          * The indirect operation can only be a BLKIF_OP_READ or
 460                          * BLKIF_OP_WRITE
 461                          */
 462                         BUG_ON(req->cmd_flags & (REQ_FLUSH | REQ_FUA));
 463                         ring_req->operation = BLKIF_OP_INDIRECT;
 464                         ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 465                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 466                         ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 467                         ring_req->u.indirect.handle = info->handle;
 468                         ring_req->u.indirect.nr_segments = nseg;
 469                 } else {
 470                         ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 471                         ring_req->u.rw.handle = info->handle;
 472                         ring_req->operation = rq_data_dir(req) ?
 473                                 BLKIF_OP_WRITE : BLKIF_OP_READ;
 474                         if (req->cmd_flags & (REQ_FLUSH | REQ_FUA)) {
 475                                 /*
 476                                  * Ideally we can do an unordered flush-to-disk. In case the
 477                                  * backend onlysupports barriers, use that. A barrier request
 478                                  * a superset of FUA, so we can implement it the same
 479                                  * way.  (It's also a FLUSH+FUA, since it is
 480                                  * guaranteed ordered WRT previous writes.)
 481                                  */
 482                                 ring_req->operation = info->flush_op;
 483                         }
 484                         ring_req->u.rw.nr_segments = nseg;
 485                 }
 486                 for_each_sg(info->shadow[id].sg, sg, nseg, i) {
 487                         fsect = sg->offset >> 9;
 488                         lsect = fsect + (sg->length >> 9) - 1;
 489
 490                         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 491                             (i % SEGS_PER_INDIRECT_FRAME == 0)) {
 492                                 unsigned long uninitialized_var(pfn);
 493
 494                                 if (segments)
 495                                         kunmap_atomic(segments);
 496
 497                                 n = i / SEGS_PER_INDIRECT_FRAME;
 498                                 if (!info->feature_persistent) {
 499                                         struct page *indirect_page;
 500
 501                                         /* Fetch a pre-allocated page to use for indirect grefs */
 502                                         BUG_ON(list_empty(&info->indirect_pages));
 503                                         indirect_page = list_first_entry(&info->indirect_pages,
 504                                                                          struct page, lru);
 505                                         list_del(&indirect_page->lru);
 506                                         pfn = page_to_pfn(indirect_page);
 507                                 }
 508                                 gnt_list_entry = get_grant(&gref_head, pfn, info);
 509                                 info->shadow[id].indirect_grants[n] = gnt_list_entry;
 510                                 segments = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 511                                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 512                         }
 513
 514                         gnt_list_entry = get_grant(&gref_head, page_to_pfn(sg_page(sg)), info);
 515                         ref = gnt_list_entry->gref;
 516
 517                         info->shadow[id].grants_used[i] = gnt_list_entry;
 518
 519                         if (rq_data_dir(req) && info->feature_persistent) {
 520                                 char *bvec_data;
 521                                 void *shared_data;
 522
 523                                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 524
 525                                 shared_data = kmap_atomic(pfn_to_page(gnt_list_entry->pfn));
 526                                 bvec_data = kmap_atomic(sg_page(sg));
 527
 528                                 /*
 529                                  * this does not wipe data stored outside the
 530                                  * range sg->offset..sg->offset+sg->length.
 531                                  * Therefore, blkback *could* see data from
 532                                  * previous requests. This is OK as long as
 533                                  * persistent grants are shared with just one
 534                                  * domain. It may need refactoring if this
 535                                  * changes
 536                                  */
 537                                 memcpy(shared_data + sg->offset,
 538                                        bvec_data   + sg->offset,
 539                                        sg->length);
 540
 541                                 kunmap_atomic(bvec_data);
 542                                 kunmap_atomic(shared_data);
 543                         }
 544                         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 545                                 ring_req->u.rw.seg[i] =
 546                                                 (struct blkif_request_segment) {
 547                                                         .gref       = ref,
 548                                                         .first_sect = fsect,
 549                                                         .last_sect  = lsect };
 550                         } else {
 551                                 n = i % SEGS_PER_INDIRECT_FRAME;
 552                                 segments[n] =
 553                                         (struct blkif_request_segment) {
 554                                                         .gref       = ref,
 555                                                         .first_sect = fsect,
 556                                                         .last_sect  = lsect };
 557                         }
 558                 }
 559                 if (segments)
 560                         kunmap_atomic(segments);
 561         }
 562
 563         info->ring.req_prod_pvt++;
 564
 565         /* Keep a private copy so we can reissue requests when recovering. */
 566         info->shadow[id].req = *ring_req;
 567
 568         if (new_persistent_gnts)
 569                 gnttab_free_grant_references(gref_head);
 570
 571         return 0;
 572 }
 573
 574
 575 static inline void flush_requests(struct blkfront_info *info)
 576 {
 577         int notify;
 578
 579         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&info->ring, notify);
 580
 581         if (notify)
 582                 notify_remote_via_irq(info->irq);
 583 }
 584
 585 static inline bool blkif_request_flush_valid(struct request *req,
 586                                              struct blkfront_info *info)
 587 {
 588         return ((req->cmd_type != REQ_TYPE_FS) ||
 589                 ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 590                 !info->flush_op));
 591 }
 592
 593 /*
 594  * do_blkif_request
 595  *  read a block; request is in a request queue
 596  */
 597 static void do_blkif_request(struct request_queue *rq)
 598 {
 599         struct blkfront_info *info = NULL;
 600         struct request *req;
 601         int queued;
 602
 603         pr_debug("Entered do_blkif_request\n");
 604
 605         queued = 0;
 606
 607         while ((req = blk_peek_request(rq)) != NULL) {
 608                 info = req->rq_disk->private_data;
 609
 610                 if (RING_FULL(&info->ring))
 611                         goto wait;
 612
 613                 blk_start_request(req);
 614
 615                 if (blkif_request_flush_valid(req, info)) {
 616                         __blk_end_request_all(req, -EIO);
 617                         continue;
 618                 }
 619
 620                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 621                          "(%u/%u) [%s]\n",
 622                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 623                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 624                          rq_data_dir(req) ? "write" : "read");
 625
 626                 if (blkif_queue_request(req)) {
 627                         blk_requeue_request(rq, req);
 628 wait:
 629                         /* Avoid pointless unplugs. */
 630                         blk_stop_queue(rq);
 631                         break;
 632                 }
 633
 634                 queued++;
 635         }
 636
 637         if (queued != 0)
 638                 flush_requests(info);
 639 }
 640
 641 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 642                                 unsigned int physical_sector_size,
 643                                 unsigned int segments)
 644 {
 645         struct request_queue *rq;
 646         struct blkfront_info *info = gd->private_data;
 647
 648         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 649         if (rq == NULL)
 650                 return -1;
 651
 652         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 653
 654         if (info->feature_discard) {
 655                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 656                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 657                 rq->limits.discard_granularity = info->discard_granularity;
 658                 rq->limits.discard_alignment = info->discard_alignment;
 659                 if (info->feature_secdiscard)
 660                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 661         }
 662
 663         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 664         blk_queue_logical_block_size(rq, sector_size);
 665         blk_queue_physical_block_size(rq, physical_sector_size);
 666         blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
 667
 668         /* Each segment in a request is up to an aligned page in size. */
 669         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 670         blk_queue_max_segment_size(rq, PAGE_SIZE);
 671
 672         /* Ensure a merged request will fit in a single I/O ring slot. */
 673         blk_queue_max_segments(rq, segments);
 674
 675         /* Make sure buffer addresses are sector-aligned. */
 676         blk_queue_dma_alignment(rq, 511);
 677
 678         /* Make sure we don't use bounce buffers. */
 679         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 680
 681         gd->queue = rq;
 682
 683         return 0;
 684 }
 685
 686
 687 static void xlvbd_flush(struct blkfront_info *info)
 688 {
 689         blk_queue_flush(info->rq, info->feature_flush);
 690         printk(KERN_INFO "blkfront: %s: %s: %s %s %s %s %s\n",
 691                info->gd->disk_name,
 692                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 693                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 694                 "flush diskcache" : "barrier or flush"),
 695                info->feature_flush ? "enabled;" : "disabled;",
 696                "persistent grants:",
 697                info->feature_persistent ? "enabled;" : "disabled;",
 698                "indirect descriptors:",
 699                info->max_indirect_segments ? "enabled;" : "disabled;");
 700 }
 701
 702 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 703 {
 704         int major;
 705         major = BLKIF_MAJOR(vdevice);
 706         *minor = BLKIF_MINOR(vdevice);
 707         switch (major) {
 708                 case XEN_IDE0_MAJOR:
 709                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 710                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 711                                 EMULATED_HD_DISK_MINOR_OFFSET;
 712                         break;
 713                 case XEN_IDE1_MAJOR:
 714                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 715                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 716                                 EMULATED_HD_DISK_MINOR_OFFSET;
 717                         break;
 718                 case XEN_SCSI_DISK0_MAJOR:
 719                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 720                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 721                         break;
 722                 case XEN_SCSI_DISK1_MAJOR:
 723                 case XEN_SCSI_DISK2_MAJOR:
 724                 case XEN_SCSI_DISK3_MAJOR:
 725                 case XEN_SCSI_DISK4_MAJOR:
 726                 case XEN_SCSI_DISK5_MAJOR:
 727                 case XEN_SCSI_DISK6_MAJOR:
 728                 case XEN_SCSI_DISK7_MAJOR:
 729                         *offset = (*minor / PARTS_PER_DISK) +
 730                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 731                                 EMULATED_SD_DISK_NAME_OFFSET;
 732                         *minor = *minor +
 733                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 734                                 EMULATED_SD_DISK_MINOR_OFFSET;
 735                         break;
 736                 case XEN_SCSI_DISK8_MAJOR:
 737                 case XEN_SCSI_DISK9_MAJOR:
 738                 case XEN_SCSI_DISK10_MAJOR:
 739                 case XEN_SCSI_DISK11_MAJOR:
 740                 case XEN_SCSI_DISK12_MAJOR:
 741                 case XEN_SCSI_DISK13_MAJOR:
 742                 case XEN_SCSI_DISK14_MAJOR:
 743                 case XEN_SCSI_DISK15_MAJOR:
 744                         *offset = (*minor / PARTS_PER_DISK) +
 745                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 746                                 EMULATED_SD_DISK_NAME_OFFSET;
 747                         *minor = *minor +
 748                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 749                                 EMULATED_SD_DISK_MINOR_OFFSET;
 750                         break;
 751                 case XENVBD_MAJOR:
 752                         *offset = *minor / PARTS_PER_DISK;
 753                         break;
 754                 default:
 755                         printk(KERN_WARNING "blkfront: your disk configuration is "
 756                                         "incorrect, please use an xvd device instead\n");
 757                         return -ENODEV;
 758         }
 759         return 0;
 760 }
 761
 762 static char *encode_disk_name(char *ptr, unsigned int n)
 763 {
 764         if (n >= 26)
 765                 ptr = encode_disk_name(ptr, n / 26 - 1);
 766         *ptr = 'a' + n % 26;
 767         return ptr + 1;
 768 }
 769
 770 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 771                                struct blkfront_info *info,
 772                                u16 vdisk_info, u16 sector_size,
 773                                unsigned int physical_sector_size)
 774 {
 775         struct gendisk *gd;
 776         int nr_minors = 1;
 777         int err;
 778         unsigned int offset;
 779         int minor;
 780         int nr_parts;
 781         char *ptr;
 782
 783         BUG_ON(info->gd != NULL);
 784         BUG_ON(info->rq != NULL);
 785
 786         if ((info->vdevice>>EXT_SHIFT) > 1) {
 787                 /* this is above the extended range; something is wrong */
 788                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 789                 return -ENODEV;
 790         }
 791
 792         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 793                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 794                 if (err)
 795                         return err;
 796                 nr_parts = PARTS_PER_DISK;
 797         } else {
 798                 minor = BLKIF_MINOR_EXT(info->vdevice);
 799                 nr_parts = PARTS_PER_EXT_DISK;
 800                 offset = minor / nr_parts;
 801                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 802                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 803                                         "emulated IDE disks,\n\t choose an xvd device name"
 804                                         "from xvde on\n", info->vdevice);
 805         }
 806         if (minor >> MINORBITS) {
 807                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 808                         info->vdevice, minor);
 809                 return -ENODEV;
 810         }
 811
 812         if ((minor % nr_parts) == 0)
 813                 nr_minors = nr_parts;
 814
 815         err = xlbd_reserve_minors(minor, nr_minors);
 816         if (err)
 817                 goto out;
 818         err = -ENODEV;
 819
 820         gd = alloc_disk(nr_minors);
 821         if (gd == NULL)
 822                 goto release;
 823
 824         strcpy(gd->disk_name, DEV_NAME);
 825         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 826         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 827         if (nr_minors > 1)
 828                 *ptr = 0;
 829         else
 830                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 831                          "%d", minor & (nr_parts - 1));
 832
 833         gd->major = XENVBD_MAJOR;
 834         gd->first_minor = minor;
 835         gd->fops = &xlvbd_block_fops;
 836         gd->private_data = info;
 837         gd->driverfs_dev = &(info->xbdev->dev);
 838         set_capacity(gd, capacity);
 839
 840         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
 841                                  info->max_indirect_segments ? :
 842                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 843                 del_gendisk(gd);
 844                 goto release;
 845         }
 846
 847         info->rq = gd->queue;
 848         info->gd = gd;
 849
 850         xlvbd_flush(info);
 851
 852         if (vdisk_info & VDISK_READONLY)
 853                 set_disk_ro(gd, 1);
 854
 855         if (vdisk_info & VDISK_REMOVABLE)
 856                 gd->flags |= GENHD_FL_REMOVABLE;
 857
 858         if (vdisk_info & VDISK_CDROM)
 859                 gd->flags |= GENHD_FL_CD;
 860
 861         return 0;
 862
 863  release:
 864         xlbd_release_minors(minor, nr_minors);
 865  out:
 866         return err;
 867 }
 868
 869 static void xlvbd_release_gendisk(struct blkfront_info *info)
 870 {
 871         unsigned int minor, nr_minors;
 872         unsigned long flags;
 873
 874         if (info->rq == NULL)
 875                 return;
 876
 877         spin_lock_irqsave(&info->io_lock, flags);
 878
 879         /* No more blkif_request(). */
 880         blk_stop_queue(info->rq);
 881
 882         /* No more gnttab callback work. */
 883         gnttab_cancel_free_callback(&info->callback);
 884         spin_unlock_irqrestore(&info->io_lock, flags);
 885
 886         /* Flush gnttab callback work. Must be done with no locks held. */
 887         flush_work(&info->work);
 888
 889         del_gendisk(info->gd);
 890
 891         minor = info->gd->first_minor;
 892         nr_minors = info->gd->minors;
 893         xlbd_release_minors(minor, nr_minors);
 894
 895         blk_cleanup_queue(info->rq);
 896         info->rq = NULL;
 897
 898         put_disk(info->gd);
 899         info->gd = NULL;
 900 }
 901
 902 static void kick_pending_request_queues(struct blkfront_info *info)
 903 {
 904         if (!RING_FULL(&info->ring)) {
 905                 /* Re-enable calldowns. */
 906                 blk_start_queue(info->rq);
 907                 /* Kick things off immediately. */
 908                 do_blkif_request(info->rq);
 909         }
 910 }
 911
 912 static void blkif_restart_queue(struct work_struct *work)
 913 {
 914         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 915
 916         spin_lock_irq(&info->io_lock);
 917         if (info->connected == BLKIF_STATE_CONNECTED)
 918                 kick_pending_request_queues(info);
 919         spin_unlock_irq(&info->io_lock);
 920 }
 921
 922 static void blkif_free(struct blkfront_info *info, int suspend)
 923 {
 924         struct grant *persistent_gnt;
 925         struct grant *n;
 926         int i, j, segs;
 927
 928         /* Prevent new requests being issued until we fix things up. */
 929         spin_lock_irq(&info->io_lock);
 930         info->connected = suspend ?
 931                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 932         /* No more blkif_request(). */
 933         if (info->rq)
 934                 blk_stop_queue(info->rq);
 935
 936         /* Remove all persistent grants */
 937         if (!list_empty(&info->grants)) {
 938                 list_for_each_entry_safe(persistent_gnt, n,
 939                                          &info->grants, node) {
 940                         list_del(&persistent_gnt->node);
 941                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 942                                 gnttab_end_foreign_access(persistent_gnt->gref,
 943                                                           0, 0UL);
 944                                 info->persistent_gnts_c--;
 945                         }
 946                         if (info->feature_persistent)
 947                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 948                         kfree(persistent_gnt);
 949                 }
 950         }
 951         BUG_ON(info->persistent_gnts_c != 0);
 952
 953         /*
 954          * Remove indirect pages, this only happens when using indirect
 955          * descriptors but not persistent grants
 956          */
 957         if (!list_empty(&info->indirect_pages)) {
 958                 struct page *indirect_page, *n;
 959
 960                 BUG_ON(info->feature_persistent);
 961                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
 962                         list_del(&indirect_page->lru);
 963                         __free_page(indirect_page);
 964                 }
 965         }
 966
 967         for (i = 0; i < BLK_RING_SIZE; i++) {
 968                 /*
 969                  * Clear persistent grants present in requests already
 970                  * on the shared ring
 971                  */
 972                 if (!info->shadow[i].request)
 973                         goto free_shadow;
 974
 975                 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
 976                        info->shadow[i].req.u.indirect.nr_segments :
 977                        info->shadow[i].req.u.rw.nr_segments;
 978                 for (j = 0; j < segs; j++) {
 979                         persistent_gnt = info->shadow[i].grants_used[j];
 980                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
 981                         if (info->feature_persistent)
 982                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 983                         kfree(persistent_gnt);
 984                 }
 985
 986                 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
 987                         /*
 988                          * If this is not an indirect operation don't try to
 989                          * free indirect segments
 990                          */
 991                         goto free_shadow;
 992
 993                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
 994                         persistent_gnt = info->shadow[i].indirect_grants[j];
 995                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
 996                         __free_page(pfn_to_page(persistent_gnt->pfn));
 997                         kfree(persistent_gnt);
 998                 }
 999
1000 free_shadow:
1001                 kfree(info->shadow[i].grants_used);
1002                 info->shadow[i].grants_used = NULL;
1003                 kfree(info->shadow[i].indirect_grants);
1004                 info->shadow[i].indirect_grants = NULL;
1005                 kfree(info->shadow[i].sg);
1006                 info->shadow[i].sg = NULL;
1007         }
1008
1009         /* No more gnttab callback work. */
1010         gnttab_cancel_free_callback(&info->callback);
1011         spin_unlock_irq(&info->io_lock);
1012
1013         /* Flush gnttab callback work. Must be done with no locks held. */
1014         flush_work(&info->work);
1015
1016         /* Free resources associated with old device channel. */
1017         if (info->ring_ref != GRANT_INVALID_REF) {
1018                 gnttab_end_foreign_access(info->ring_ref, 0,
1019                                           (unsigned long)info->ring.sring);
1020                 info->ring_ref = GRANT_INVALID_REF;
1021                 info->ring.sring = NULL;
1022         }
1023         if (info->irq)
1024                 unbind_from_irqhandler(info->irq, info);
1025         info->evtchn = info->irq = 0;
1026
1027 }
1028
1029 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1030                              struct blkif_response *bret)
1031 {
1032         int i = 0;
1033         struct scatterlist *sg;
1034         char *bvec_data;
1035         void *shared_data;
1036         int nseg;
1037
1038         nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1039                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1040
1041         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1042                 /*
1043                  * Copy the data received from the backend into the bvec.
1044                  * Since bv_offset can be different than 0, and bv_len different
1045                  * than PAGE_SIZE, we have to keep track of the current offset,
1046                  * to be sure we are copying the data from the right shared page.
1047                  */
1048                 for_each_sg(s->sg, sg, nseg, i) {
1049                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1050                         shared_data = kmap_atomic(
1051                                 pfn_to_page(s->grants_used[i]->pfn));
1052                         bvec_data = kmap_atomic(sg_page(sg));
1053                         memcpy(bvec_data   + sg->offset,
1054                                shared_data + sg->offset,
1055                                sg->length);
1056                         kunmap_atomic(bvec_data);
1057                         kunmap_atomic(shared_data);
1058                 }
1059         }
1060         /* Add the persistent grant into the list of free grants */
1061         for (i = 0; i < nseg; i++) {
1062                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1063                         /*
1064                          * If the grant is still mapped by the backend (the
1065                          * backend has chosen to make this grant persistent)
1066                          * we add it at the head of the list, so it will be
1067                          * reused first.
1068                          */
1069                         if (!info->feature_persistent)
1070                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1071                                                      s->grants_used[i]->gref);
1072                         list_add(&s->grants_used[i]->node, &info->grants);
1073                         info->persistent_gnts_c++;
1074                 } else {
1075                         /*
1076                          * If the grant is not mapped by the backend we end the
1077                          * foreign access and add it to the tail of the list,
1078                          * so it will not be picked again unless we run out of
1079                          * persistent grants.
1080                          */
1081                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1082                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1083                         list_add_tail(&s->grants_used[i]->node, &info->grants);
1084                 }
1085         }
1086         if (s->req.operation == BLKIF_OP_INDIRECT) {
1087                 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1088                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1089                                 if (!info->feature_persistent)
1090                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1091                                                              s->indirect_grants[i]->gref);
1092                                 list_add(&s->indirect_grants[i]->node, &info->grants);
1093                                 info->persistent_gnts_c++;
1094                         } else {
1095                                 struct page *indirect_page;
1096
1097                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1098                                 /*
1099                                  * Add the used indirect page back to the list of
1100                                  * available pages for indirect grefs.
1101                                  */
1102                                 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1103                                 list_add(&indirect_page->lru, &info->indirect_pages);
1104                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1105                                 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1106                         }
1107                 }
1108         }
1109 }
1110
1111 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1112 {
1113         struct request *req;
1114         struct blkif_response *bret;
1115         RING_IDX i, rp;
1116         unsigned long flags;
1117         struct blkfront_info *info = (struct blkfront_info *)dev_id;
1118         int error;
1119
1120         spin_lock_irqsave(&info->io_lock, flags);
1121
1122         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1123                 spin_unlock_irqrestore(&info->io_lock, flags);
1124                 return IRQ_HANDLED;
1125         }
1126
1127  again:
1128         rp = info->ring.sring->rsp_prod;
1129         rmb(); /* Ensure we see queued responses up to 'rp'. */
1130
1131         for (i = info->ring.rsp_cons; i != rp; i++) {
1132                 unsigned long id;
1133
1134                 bret = RING_GET_RESPONSE(&info->ring, i);
1135                 id   = bret->id;
1136                 /*
1137                  * The backend has messed up and given us an id that we would
1138                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1139                  * look in get_id_from_freelist.
1140                  */
1141                 if (id >= BLK_RING_SIZE) {
1142                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1143                              info->gd->disk_name, op_name(bret->operation), id);
1144                         /* We can't safely get the 'struct request' as
1145                          * the id is busted. */
1146                         continue;
1147                 }
1148                 req  = info->shadow[id].request;
1149
1150                 if (bret->operation != BLKIF_OP_DISCARD)
1151                         blkif_completion(&info->shadow[id], info, bret);
1152
1153                 if (add_id_to_freelist(info, id)) {
1154                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1155                              info->gd->disk_name, op_name(bret->operation), id);
1156                         continue;
1157                 }
1158
1159                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1160                 switch (bret->operation) {
1161                 case BLKIF_OP_DISCARD:
1162                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1163                                 struct request_queue *rq = info->rq;
1164                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1165                                            info->gd->disk_name, op_name(bret->operation));
1166                                 error = -EOPNOTSUPP;
1167                                 info->feature_discard = 0;
1168                                 info->feature_secdiscard = 0;
1169                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1170                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1171                         }
1172                         __blk_end_request_all(req, error);
1173                         break;
1174                 case BLKIF_OP_FLUSH_DISKCACHE:
1175                 case BLKIF_OP_WRITE_BARRIER:
1176                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1177                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1178                                        info->gd->disk_name, op_name(bret->operation));
1179                                 error = -EOPNOTSUPP;
1180                         }
1181                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1182                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
1183                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1184                                        info->gd->disk_name, op_name(bret->operation));
1185                                 error = -EOPNOTSUPP;
1186                         }
1187                         if (unlikely(error)) {
1188                                 if (error == -EOPNOTSUPP)
1189                                         error = 0;
1190                                 info->feature_flush = 0;
1191                                 info->flush_op = 0;
1192                                 xlvbd_flush(info);
1193                         }
1194                         /* fall through */
1195                 case BLKIF_OP_READ:
1196                 case BLKIF_OP_WRITE:
1197                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1198                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1199                                         "request: %x\n", bret->status);
1200
1201                         __blk_end_request_all(req, error);
1202                         break;
1203                 default:
1204                         BUG();
1205                 }
1206         }
1207
1208         info->ring.rsp_cons = i;
1209
1210         if (i != info->ring.req_prod_pvt) {
1211                 int more_to_do;
1212                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1213                 if (more_to_do)
1214                         goto again;
1215         } else
1216                 info->ring.sring->rsp_event = i + 1;
1217
1218         kick_pending_request_queues(info);
1219
1220         spin_unlock_irqrestore(&info->io_lock, flags);
1221
1222         return IRQ_HANDLED;
1223 }
1224
1225
1226 static int setup_blkring(struct xenbus_device *dev,
1227                          struct blkfront_info *info)
1228 {
1229         struct blkif_sring *sring;
1230         int err;
1231
1232         info->ring_ref = GRANT_INVALID_REF;
1233
1234         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1235         if (!sring) {
1236                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1237                 return -ENOMEM;
1238         }
1239         SHARED_RING_INIT(sring);
1240         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1241
1242         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1243         if (err < 0) {
1244                 free_page((unsigned long)sring);
1245                 info->ring.sring = NULL;
1246                 goto fail;
1247         }
1248         info->ring_ref = err;
1249
1250         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1251         if (err)
1252                 goto fail;
1253
1254         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1255                                         "blkif", info);
1256         if (err <= 0) {
1257                 xenbus_dev_fatal(dev, err,
1258                                  "bind_evtchn_to_irqhandler failed");
1259                 goto fail;
1260         }
1261         info->irq = err;
1262
1263         return 0;
1264 fail:
1265         blkif_free(info, 0);
1266         return err;
1267 }
1268
1269
1270 /* Common code used when first setting up, and when resuming. */
1271 static int talk_to_blkback(struct xenbus_device *dev,
1272                            struct blkfront_info *info)
1273 {
1274         const char *message = NULL;
1275         struct xenbus_transaction xbt;
1276         int err;
1277
1278         /* Create shared ring, alloc event channel. */
1279         err = setup_blkring(dev, info);
1280         if (err)
1281                 goto out;
1282
1283 again:
1284         err = xenbus_transaction_start(&xbt);
1285         if (err) {
1286                 xenbus_dev_fatal(dev, err, "starting transaction");
1287                 goto destroy_blkring;
1288         }
1289
1290         err = xenbus_printf(xbt, dev->nodename,
1291                             "ring-ref", "%u", info->ring_ref);
1292         if (err) {
1293                 message = "writing ring-ref";
1294                 goto abort_transaction;
1295         }
1296         err = xenbus_printf(xbt, dev->nodename,
1297                             "event-channel", "%u", info->evtchn);
1298         if (err) {
1299                 message = "writing event-channel";
1300                 goto abort_transaction;
1301         }
1302         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1303                             XEN_IO_PROTO_ABI_NATIVE);
1304         if (err) {
1305                 message = "writing protocol";
1306                 goto abort_transaction;
1307         }
1308         err = xenbus_printf(xbt, dev->nodename,
1309                             "feature-persistent", "%u", 1);
1310         if (err)
1311                 dev_warn(&dev->dev,
1312                          "writing persistent grants feature to xenbus");
1313
1314         err = xenbus_transaction_end(xbt, 0);
1315         if (err) {
1316                 if (err == -EAGAIN)
1317                         goto again;
1318                 xenbus_dev_fatal(dev, err, "completing transaction");
1319                 goto destroy_blkring;
1320         }
1321
1322         xenbus_switch_state(dev, XenbusStateInitialised);
1323
1324         return 0;
1325
1326  abort_transaction:
1327         xenbus_transaction_end(xbt, 1);
1328         if (message)
1329                 xenbus_dev_fatal(dev, err, "%s", message);
1330  destroy_blkring:
1331         blkif_free(info, 0);
1332  out:
1333         return err;
1334 }
1335
1336 /**
1337  * Entry point to this code when a new device is created.  Allocate the basic
1338  * structures and the ring buffer for communication with the backend, and
1339  * inform the backend of the appropriate details for those.  Switch to
1340  * Initialised state.
1341  */
1342 static int blkfront_probe(struct xenbus_device *dev,
1343                           const struct xenbus_device_id *id)
1344 {
1345         int err, vdevice, i;
1346         struct blkfront_info *info;
1347
1348         /* FIXME: Use dynamic device id if this is not set. */
1349         err = xenbus_scanf(XBT_NIL, dev->nodename,
1350                            "virtual-device", "%i", &vdevice);
1351         if (err != 1) {
1352                 /* go looking in the extended area instead */
1353                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1354                                    "%i", &vdevice);
1355                 if (err != 1) {
1356                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1357                         return err;
1358                 }
1359         }
1360
1361         if (xen_hvm_domain()) {
1362                 char *type;
1363                 int len;
1364                 /* no unplug has been done: do not hook devices != xen vbds */
1365                 if (xen_has_pv_and_legacy_disk_devices()) {
1366                         int major;
1367
1368                         if (!VDEV_IS_EXTENDED(vdevice))
1369                                 major = BLKIF_MAJOR(vdevice);
1370                         else
1371                                 major = XENVBD_MAJOR;
1372
1373                         if (major != XENVBD_MAJOR) {
1374                                 printk(KERN_INFO
1375                                                 "%s: HVM does not support vbd %d as xen block device\n",
1376                                                 __FUNCTION__, vdevice);
1377                                 return -ENODEV;
1378                         }
1379                 }
1380                 /* do not create a PV cdrom device if we are an HVM guest */
1381                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1382                 if (IS_ERR(type))
1383                         return -ENODEV;
1384                 if (strncmp(type, "cdrom", 5) == 0) {
1385                         kfree(type);
1386                         return -ENODEV;
1387                 }
1388                 kfree(type);
1389         }
1390         info = kzalloc(sizeof(*info), GFP_KERNEL);
1391         if (!info) {
1392                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1393                 return -ENOMEM;
1394         }
1395
1396         mutex_init(&info->mutex);
1397         spin_lock_init(&info->io_lock);
1398         info->xbdev = dev;
1399         info->vdevice = vdevice;
1400         INIT_LIST_HEAD(&info->grants);
1401         INIT_LIST_HEAD(&info->indirect_pages);
1402         info->persistent_gnts_c = 0;
1403         info->connected = BLKIF_STATE_DISCONNECTED;
1404         INIT_WORK(&info->work, blkif_restart_queue);
1405
1406         for (i = 0; i < BLK_RING_SIZE; i++)
1407                 info->shadow[i].req.u.rw.id = i+1;
1408         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1409
1410         /* Front end dir is a number, which is used as the id. */
1411         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1412         dev_set_drvdata(&dev->dev, info);
1413
1414         err = talk_to_blkback(dev, info);
1415         if (err) {
1416                 kfree(info);
1417                 dev_set_drvdata(&dev->dev, NULL);
1418                 return err;
1419         }
1420
1421         return 0;
1422 }
1423
1424 static void split_bio_end(struct bio *bio, int error)
1425 {
1426         struct split_bio *split_bio = bio->bi_private;
1427
1428         if (error)
1429                 split_bio->err = error;
1430
1431         if (atomic_dec_and_test(&split_bio->pending)) {
1432                 split_bio->bio->bi_phys_segments = 0;
1433                 bio_endio(split_bio->bio, split_bio->err);
1434                 kfree(split_bio);
1435         }
1436         bio_put(bio);
1437 }
1438
1439 static int blkif_recover(struct blkfront_info *info)
1440 {
1441         int i;
1442         struct request *req, *n;
1443         struct blk_shadow *copy;
1444         int rc;
1445         struct bio *bio, *cloned_bio;
1446         struct bio_list bio_list, merge_bio;
1447         unsigned int segs, offset;
1448         int pending, size;
1449         struct split_bio *split_bio;
1450         struct list_head requests;
1451
1452         /* Stage 1: Make a safe copy of the shadow state. */
1453         copy = kmemdup(info->shadow, sizeof(info->shadow),
1454                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1455         if (!copy)
1456                 return -ENOMEM;
1457
1458         /* Stage 2: Set up free list. */
1459         memset(&info->shadow, 0, sizeof(info->shadow));
1460         for (i = 0; i < BLK_RING_SIZE; i++)
1461                 info->shadow[i].req.u.rw.id = i+1;
1462         info->shadow_free = info->ring.req_prod_pvt;
1463         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1464
1465         rc = blkfront_setup_indirect(info);
1466         if (rc) {
1467                 kfree(copy);
1468                 return rc;
1469         }
1470
1471         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1472         blk_queue_max_segments(info->rq, segs);
1473         bio_list_init(&bio_list);
1474         INIT_LIST_HEAD(&requests);
1475         for (i = 0; i < BLK_RING_SIZE; i++) {
1476                 /* Not in use? */
1477                 if (!copy[i].request)
1478                         continue;
1479
1480                 /*
1481                  * Get the bios in the request so we can re-queue them.
1482                  */
1483                 if (copy[i].request->cmd_flags &
1484                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1485                         /*
1486                          * Flush operations don't contain bios, so
1487                          * we need to requeue the whole request
1488                          */
1489                         list_add(&copy[i].request->queuelist, &requests);
1490                         continue;
1491                 }
1492                 merge_bio.head = copy[i].request->bio;
1493                 merge_bio.tail = copy[i].request->biotail;
1494                 bio_list_merge(&bio_list, &merge_bio);
1495                 copy[i].request->bio = NULL;
1496                 blk_put_request(copy[i].request);
1497         }
1498
1499         kfree(copy);
1500
1501         /*
1502          * Empty the queue, this is important because we might have
1503          * requests in the queue with more segments than what we
1504          * can handle now.
1505          */
1506         spin_lock_irq(&info->io_lock);
1507         while ((req = blk_fetch_request(info->rq)) != NULL) {
1508                 if (req->cmd_flags &
1509                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1510                         list_add(&req->queuelist, &requests);
1511                         continue;
1512                 }
1513                 merge_bio.head = req->bio;
1514                 merge_bio.tail = req->biotail;
1515                 bio_list_merge(&bio_list, &merge_bio);
1516                 req->bio = NULL;
1517                 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1518                         pr_alert("diskcache flush request found!\n");
1519                 __blk_put_request(info->rq, req);
1520         }
1521         spin_unlock_irq(&info->io_lock);
1522
1523         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1524
1525         spin_lock_irq(&info->io_lock);
1526
1527         /* Now safe for us to use the shared ring */
1528         info->connected = BLKIF_STATE_CONNECTED;
1529
1530         /* Kick any other new requests queued since we resumed */
1531         kick_pending_request_queues(info);
1532
1533         list_for_each_entry_safe(req, n, &requests, queuelist) {
1534                 /* Requeue pending requests (flush or discard) */
1535                 list_del_init(&req->queuelist);
1536                 BUG_ON(req->nr_phys_segments > segs);
1537                 blk_requeue_request(info->rq, req);
1538         }
1539         spin_unlock_irq(&info->io_lock);
1540
1541         while ((bio = bio_list_pop(&bio_list)) != NULL) {
1542                 /* Traverse the list of pending bios and re-queue them */
1543                 if (bio_segments(bio) > segs) {
1544                         /*
1545                          * This bio has more segments than what we can
1546                          * handle, we have to split it.
1547                          */
1548                         pending = (bio_segments(bio) + segs - 1) / segs;
1549                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1550                         BUG_ON(split_bio == NULL);
1551                         atomic_set(&split_bio->pending, pending);
1552                         split_bio->bio = bio;
1553                         for (i = 0; i < pending; i++) {
1554                                 offset = (i * segs * PAGE_SIZE) >> 9;
1555                                 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1556                                            (unsigned int)bio_sectors(bio) - offset);
1557                                 cloned_bio = bio_clone(bio, GFP_NOIO);
1558                                 BUG_ON(cloned_bio == NULL);
1559                                 bio_trim(cloned_bio, offset, size);
1560                                 cloned_bio->bi_private = split_bio;
1561                                 cloned_bio->bi_end_io = split_bio_end;
1562                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
1563                         }
1564                         /*
1565                          * Now we have to wait for all those smaller bios to
1566                          * end, so we can also end the "parent" bio.
1567                          */
1568                         continue;
1569                 }
1570                 /* We don't need to split this bio */
1571                 submit_bio(bio->bi_rw, bio);
1572         }
1573
1574         return 0;
1575 }
1576
1577 /**
1578  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1579  * driver restart.  We tear down our blkif structure and recreate it, but
1580  * leave the device-layer structures intact so that this is transparent to the
1581  * rest of the kernel.
1582  */
1583 static int blkfront_resume(struct xenbus_device *dev)
1584 {
1585         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1586         int err;
1587
1588         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1589
1590         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1591
1592         err = talk_to_blkback(dev, info);
1593
1594         /*
1595          * We have to wait for the backend to switch to
1596          * connected state, since we want to read which
1597          * features it supports.
1598          */
1599
1600         return err;
1601 }
1602
1603 static void
1604 blkfront_closing(struct blkfront_info *info)
1605 {
1606         struct xenbus_device *xbdev = info->xbdev;
1607         struct block_device *bdev = NULL;
1608
1609         mutex_lock(&info->mutex);
1610
1611         if (xbdev->state == XenbusStateClosing) {
1612                 mutex_unlock(&info->mutex);
1613                 return;
1614         }
1615
1616         if (info->gd)
1617                 bdev = bdget_disk(info->gd, 0);
1618
1619         mutex_unlock(&info->mutex);
1620
1621         if (!bdev) {
1622                 xenbus_frontend_closed(xbdev);
1623                 return;
1624         }
1625
1626         mutex_lock(&bdev->bd_mutex);
1627
1628         if (bdev->bd_openers) {
1629                 xenbus_dev_error(xbdev, -EBUSY,
1630                                  "Device in use; refusing to close");
1631                 xenbus_switch_state(xbdev, XenbusStateClosing);
1632         } else {
1633                 xlvbd_release_gendisk(info);
1634                 xenbus_frontend_closed(xbdev);
1635         }
1636
1637         mutex_unlock(&bdev->bd_mutex);
1638         bdput(bdev);
1639 }
1640
1641 static void blkfront_setup_discard(struct blkfront_info *info)
1642 {
1643         int err;
1644         unsigned int discard_granularity;
1645         unsigned int discard_alignment;
1646         unsigned int discard_secure;
1647
1648         info->feature_discard = 1;
1649         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1650                 "discard-granularity", "%u", &discard_granularity,
1651                 "discard-alignment", "%u", &discard_alignment,
1652                 NULL);
1653         if (!err) {
1654                 info->discard_granularity = discard_granularity;
1655                 info->discard_alignment = discard_alignment;
1656         }
1657         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1658                     "discard-secure", "%d", &discard_secure,
1659                     NULL);
1660         if (!err)
1661                 info->feature_secdiscard = !!discard_secure;
1662 }
1663
1664 static int blkfront_setup_indirect(struct blkfront_info *info)
1665 {
1666         unsigned int indirect_segments, segs;
1667         int err, i;
1668
1669         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1670                             "feature-max-indirect-segments", "%u", &indirect_segments,
1671                             NULL);
1672         if (err) {
1673                 info->max_indirect_segments = 0;
1674                 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1675         } else {
1676                 info->max_indirect_segments = min(indirect_segments,
1677                                                   xen_blkif_max_segments);
1678                 segs = info->max_indirect_segments;
1679         }
1680
1681         err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1682         if (err)
1683                 goto out_of_memory;
1684
1685         if (!info->feature_persistent && info->max_indirect_segments) {
1686                 /*
1687                  * We are using indirect descriptors but not persistent
1688                  * grants, we need to allocate a set of pages that can be
1689                  * used for mapping indirect grefs
1690                  */
1691                 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1692
1693                 BUG_ON(!list_empty(&info->indirect_pages));
1694                 for (i = 0; i < num; i++) {
1695                         struct page *indirect_page = alloc_page(GFP_NOIO);
1696                         if (!indirect_page)
1697                                 goto out_of_memory;
1698                         list_add(&indirect_page->lru, &info->indirect_pages);
1699                 }
1700         }
1701
1702         for (i = 0; i < BLK_RING_SIZE; i++) {
1703                 info->shadow[i].grants_used = kzalloc(
1704                         sizeof(info->shadow[i].grants_used[0]) * segs,
1705                         GFP_NOIO);
1706                 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1707                 if (info->max_indirect_segments)
1708                         info->shadow[i].indirect_grants = kzalloc(
1709                                 sizeof(info->shadow[i].indirect_grants[0]) *
1710                                 INDIRECT_GREFS(segs),
1711                                 GFP_NOIO);
1712                 if ((info->shadow[i].grants_used == NULL) ||
1713                         (info->shadow[i].sg == NULL) ||
1714                      (info->max_indirect_segments &&
1715                      (info->shadow[i].indirect_grants == NULL)))
1716                         goto out_of_memory;
1717                 sg_init_table(info->shadow[i].sg, segs);
1718         }
1719
1720
1721         return 0;
1722
1723 out_of_memory:
1724         for (i = 0; i < BLK_RING_SIZE; i++) {
1725                 kfree(info->shadow[i].grants_used);
1726                 info->shadow[i].grants_used = NULL;
1727                 kfree(info->shadow[i].sg);
1728                 info->shadow[i].sg = NULL;
1729                 kfree(info->shadow[i].indirect_grants);
1730                 info->shadow[i].indirect_grants = NULL;
1731         }
1732         if (!list_empty(&info->indirect_pages)) {
1733                 struct page *indirect_page, *n;
1734                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1735                         list_del(&indirect_page->lru);
1736                         __free_page(indirect_page);
1737                 }
1738         }
1739         return -ENOMEM;
1740 }
1741
1742 /*
1743  * Invoked when the backend is finally 'ready' (and has told produced
1744  * the details about the physical device - #sectors, size, etc).
1745  */
1746 static void blkfront_connect(struct blkfront_info *info)
1747 {
1748         unsigned long long sectors;
1749         unsigned long sector_size;
1750         unsigned int physical_sector_size;
1751         unsigned int binfo;
1752         int err;
1753         int barrier, flush, discard, persistent;
1754
1755         switch (info->connected) {
1756         case BLKIF_STATE_CONNECTED:
1757                 /*
1758                  * Potentially, the back-end may be signalling
1759                  * a capacity change; update the capacity.
1760                  */
1761                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1762                                    "sectors", "%Lu", &sectors);
1763                 if (XENBUS_EXIST_ERR(err))
1764                         return;
1765                 printk(KERN_INFO "Setting capacity to %Lu\n",
1766                        sectors);
1767                 set_capacity(info->gd, sectors);
1768                 revalidate_disk(info->gd);
1769
1770                 return;
1771         case BLKIF_STATE_SUSPENDED:
1772                 /*
1773                  * If we are recovering from suspension, we need to wait
1774                  * for the backend to announce it's features before
1775                  * reconnecting, at least we need to know if the backend
1776                  * supports indirect descriptors, and how many.
1777                  */
1778                 blkif_recover(info);
1779                 return;
1780
1781         default:
1782                 break;
1783         }
1784
1785         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1786                 __func__, info->xbdev->otherend);
1787
1788         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1789                             "sectors", "%llu", &sectors,
1790                             "info", "%u", &binfo,
1791                             "sector-size", "%lu", &sector_size,
1792                             NULL);
1793         if (err) {
1794                 xenbus_dev_fatal(info->xbdev, err,
1795                                  "reading backend fields at %s",
1796                                  info->xbdev->otherend);
1797                 return;
1798         }
1799
1800         /*
1801          * physcial-sector-size is a newer field, so old backends may not
1802          * provide this. Assume physical sector size to be the same as
1803          * sector_size in that case.
1804          */
1805         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1806                            "physical-sector-size", "%u", &physical_sector_size);
1807         if (err != 1)
1808                 physical_sector_size = sector_size;
1809
1810         info->feature_flush = 0;
1811         info->flush_op = 0;
1812
1813         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1814                             "feature-barrier", "%d", &barrier,
1815                             NULL);
1816
1817         /*
1818          * If there's no "feature-barrier" defined, then it means
1819          * we're dealing with a very old backend which writes
1820          * synchronously; nothing to do.
1821          *
1822          * If there are barriers, then we use flush.
1823          */
1824         if (!err && barrier) {
1825                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1826                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1827         }
1828         /*
1829          * And if there is "feature-flush-cache" use that above
1830          * barriers.
1831          */
1832         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1833                             "feature-flush-cache", "%d", &flush,
1834                             NULL);
1835
1836         if (!err && flush) {
1837                 info->feature_flush = REQ_FLUSH;
1838                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1839         }
1840
1841         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1842                             "feature-discard", "%d", &discard,
1843                             NULL);
1844
1845         if (!err && discard)
1846                 blkfront_setup_discard(info);
1847
1848         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1849                             "feature-persistent", "%u", &persistent,
1850                             NULL);
1851         if (err)
1852                 info->feature_persistent = 0;
1853         else
1854                 info->feature_persistent = persistent;
1855
1856         err = blkfront_setup_indirect(info);
1857         if (err) {
1858                 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1859                                  info->xbdev->otherend);
1860                 return;
1861         }
1862
1863         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1864                                   physical_sector_size);
1865         if (err) {
1866                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1867                                  info->xbdev->otherend);
1868                 return;
1869         }
1870
1871         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1872
1873         /* Kick pending requests. */
1874         spin_lock_irq(&info->io_lock);
1875         info->connected = BLKIF_STATE_CONNECTED;
1876         kick_pending_request_queues(info);
1877         spin_unlock_irq(&info->io_lock);
1878
1879         add_disk(info->gd);
1880
1881         info->is_ready = 1;
1882 }
1883
1884 /**
1885  * Callback received when the backend's state changes.
1886  */
1887 static void blkback_changed(struct xenbus_device *dev,
1888                             enum xenbus_state backend_state)
1889 {
1890         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1891
1892         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1893
1894         switch (backend_state) {
1895         case XenbusStateInitialising:
1896         case XenbusStateInitWait:
1897         case XenbusStateInitialised:
1898         case XenbusStateReconfiguring:
1899         case XenbusStateReconfigured:
1900         case XenbusStateUnknown:
1901                 break;
1902
1903         case XenbusStateConnected:
1904                 blkfront_connect(info);
1905                 break;
1906
1907         case XenbusStateClosed:
1908                 if (dev->state == XenbusStateClosed)
1909                         break;
1910                 /* Missed the backend's Closing state -- fallthrough */
1911         case XenbusStateClosing:
1912                 blkfront_closing(info);
1913                 break;
1914         }
1915 }
1916
1917 static int blkfront_remove(struct xenbus_device *xbdev)
1918 {
1919         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1920         struct block_device *bdev = NULL;
1921         struct gendisk *disk;
1922
1923         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1924
1925         blkif_free(info, 0);
1926
1927         mutex_lock(&info->mutex);
1928
1929         disk = info->gd;
1930         if (disk)
1931                 bdev = bdget_disk(disk, 0);
1932
1933         info->xbdev = NULL;
1934         mutex_unlock(&info->mutex);
1935
1936         if (!bdev) {
1937                 kfree(info);
1938                 return 0;
1939         }
1940
1941         /*
1942          * The xbdev was removed before we reached the Closed
1943          * state. See if it's safe to remove the disk. If the bdev
1944          * isn't closed yet, we let release take care of it.
1945          */
1946
1947         mutex_lock(&bdev->bd_mutex);
1948         info = disk->private_data;
1949
1950         dev_warn(disk_to_dev(disk),
1951                  "%s was hot-unplugged, %d stale handles\n",
1952                  xbdev->nodename, bdev->bd_openers);
1953
1954         if (info && !bdev->bd_openers) {
1955                 xlvbd_release_gendisk(info);
1956                 disk->private_data = NULL;
1957                 kfree(info);
1958         }
1959
1960         mutex_unlock(&bdev->bd_mutex);
1961         bdput(bdev);
1962
1963         return 0;
1964 }
1965
1966 static int blkfront_is_ready(struct xenbus_device *dev)
1967 {
1968         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1969
1970         return info->is_ready && info->xbdev;
1971 }
1972
1973 static int blkif_open(struct block_device *bdev, fmode_t mode)
1974 {
1975         struct gendisk *disk = bdev->bd_disk;
1976         struct blkfront_info *info;
1977         int err = 0;
1978
1979         mutex_lock(&blkfront_mutex);
1980
1981         info = disk->private_data;
1982         if (!info) {
1983                 /* xbdev gone */
1984                 err = -ERESTARTSYS;
1985                 goto out;
1986         }
1987
1988         mutex_lock(&info->mutex);
1989
1990         if (!info->gd)
1991                 /* xbdev is closed */
1992                 err = -ERESTARTSYS;
1993
1994         mutex_unlock(&info->mutex);
1995
1996 out:
1997         mutex_unlock(&blkfront_mutex);
1998         return err;
1999 }
2000
2001 static void blkif_release(struct gendisk *disk, fmode_t mode)
2002 {
2003         struct blkfront_info *info = disk->private_data;
2004         struct block_device *bdev;
2005         struct xenbus_device *xbdev;
2006
2007         mutex_lock(&blkfront_mutex);
2008
2009         bdev = bdget_disk(disk, 0);
2010
2011         if (!bdev) {
2012                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2013                 goto out_mutex;
2014         }
2015         if (bdev->bd_openers)
2016                 goto out;
2017
2018         /*
2019          * Check if we have been instructed to close. We will have
2020          * deferred this request, because the bdev was still open.
2021          */
2022
2023         mutex_lock(&info->mutex);
2024         xbdev = info->xbdev;
2025
2026         if (xbdev && xbdev->state == XenbusStateClosing) {
2027                 /* pending switch to state closed */
2028                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2029                 xlvbd_release_gendisk(info);
2030                 xenbus_frontend_closed(info->xbdev);
2031         }
2032
2033         mutex_unlock(&info->mutex);
2034
2035         if (!xbdev) {
2036                 /* sudden device removal */
2037                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2038                 xlvbd_release_gendisk(info);
2039                 disk->private_data = NULL;
2040                 kfree(info);
2041         }
2042
2043 out:
2044         bdput(bdev);
2045 out_mutex:
2046         mutex_unlock(&blkfront_mutex);
2047 }
2048
2049 static const struct block_device_operations xlvbd_block_fops =
2050 {
2051         .owner = THIS_MODULE,
2052         .open = blkif_open,
2053         .release = blkif_release,
2054         .getgeo = blkif_getgeo,
2055         .ioctl = blkif_ioctl,
2056 };
2057
2058
2059 static const struct xenbus_device_id blkfront_ids[] = {
2060         { "vbd" },
2061         { "" }
2062 };
2063
2064 static struct xenbus_driver blkfront_driver = {
2065         .ids  = blkfront_ids,
2066         .probe = blkfront_probe,
2067         .remove = blkfront_remove,
2068         .resume = blkfront_resume,
2069         .otherend_changed = blkback_changed,
2070         .is_ready = blkfront_is_ready,
2071 };
2072
2073 static int __init xlblk_init(void)
2074 {
2075         int ret;
2076
2077         if (!xen_domain())
2078                 return -ENODEV;
2079
2080         if (!xen_has_pv_disk_devices())
2081                 return -ENODEV;
2082
2083         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2084                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2085                        XENVBD_MAJOR, DEV_NAME);
2086                 return -ENODEV;
2087         }
2088
2089         ret = xenbus_register_frontend(&blkfront_driver);
2090         if (ret) {
2091                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2092                 return ret;
2093         }
2094
2095         return 0;
2096 }
2097 module_init(xlblk_init);
2098
2099
2100 static void __exit xlblk_exit(void)
2101 {
2102         xenbus_unregister_driver(&blkfront_driver);
2103         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2104         kfree(minors);
2105 }
2106 module_exit(xlblk_exit);
2107
2108 MODULE_DESCRIPTION("Xen virtual block device frontend");
2109 MODULE_LICENSE("GPL");
2110 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2111 MODULE_ALIAS("xen:vbd");
2112 MODULE_ALIAS("xenblk");