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_aligned))
 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_aligned *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_aligned) {
 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 /*
 586  * do_blkif_request
 587  *  read a block; request is in a request queue
 588  */
 589 static void do_blkif_request(struct request_queue *rq)
 590 {
 591         struct blkfront_info *info = NULL;
 592         struct request *req;
 593         int queued;
 594
 595         pr_debug("Entered do_blkif_request\n");
 596
 597         queued = 0;
 598
 599         while ((req = blk_peek_request(rq)) != NULL) {
 600                 info = req->rq_disk->private_data;
 601
 602                 if (RING_FULL(&info->ring))
 603                         goto wait;
 604
 605                 blk_start_request(req);
 606
 607                 if ((req->cmd_type != REQ_TYPE_FS) ||
 608                     ((req->cmd_flags & (REQ_FLUSH | REQ_FUA)) &&
 609                     !info->flush_op)) {
 610                         __blk_end_request_all(req, -EIO);
 611                         continue;
 612                 }
 613
 614                 pr_debug("do_blk_req %p: cmd %p, sec %lx, "
 615                          "(%u/%u) buffer:%p [%s]\n",
 616                          req, req->cmd, (unsigned long)blk_rq_pos(req),
 617                          blk_rq_cur_sectors(req), blk_rq_sectors(req),
 618                          req->buffer, rq_data_dir(req) ? "write" : "read");
 619
 620                 if (blkif_queue_request(req)) {
 621                         blk_requeue_request(rq, req);
 622 wait:
 623                         /* Avoid pointless unplugs. */
 624                         blk_stop_queue(rq);
 625                         break;
 626                 }
 627
 628                 queued++;
 629         }
 630
 631         if (queued != 0)
 632                 flush_requests(info);
 633 }
 634
 635 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 636                                 unsigned int physical_sector_size,
 637                                 unsigned int segments)
 638 {
 639         struct request_queue *rq;
 640         struct blkfront_info *info = gd->private_data;
 641
 642         rq = blk_init_queue(do_blkif_request, &info->io_lock);
 643         if (rq == NULL)
 644                 return -1;
 645
 646         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 647
 648         if (info->feature_discard) {
 649                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 650                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 651                 rq->limits.discard_granularity = info->discard_granularity;
 652                 rq->limits.discard_alignment = info->discard_alignment;
 653                 if (info->feature_secdiscard)
 654                         queue_flag_set_unlocked(QUEUE_FLAG_SECDISCARD, rq);
 655         }
 656
 657         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 658         blk_queue_logical_block_size(rq, sector_size);
 659         blk_queue_physical_block_size(rq, physical_sector_size);
 660         blk_queue_max_hw_sectors(rq, (segments * PAGE_SIZE) / 512);
 661
 662         /* Each segment in a request is up to an aligned page in size. */
 663         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 664         blk_queue_max_segment_size(rq, PAGE_SIZE);
 665
 666         /* Ensure a merged request will fit in a single I/O ring slot. */
 667         blk_queue_max_segments(rq, segments);
 668
 669         /* Make sure buffer addresses are sector-aligned. */
 670         blk_queue_dma_alignment(rq, 511);
 671
 672         /* Make sure we don't use bounce buffers. */
 673         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 674
 675         gd->queue = rq;
 676
 677         return 0;
 678 }
 679
 680
 681 static void xlvbd_flush(struct blkfront_info *info)
 682 {
 683         blk_queue_flush(info->rq, info->feature_flush);
 684         printk(KERN_INFO "blkfront: %s: %s: %s %s %s %s %s\n",
 685                info->gd->disk_name,
 686                info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 687                 "barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 688                 "flush diskcache" : "barrier or flush"),
 689                info->feature_flush ? "enabled;" : "disabled;",
 690                "persistent grants:",
 691                info->feature_persistent ? "enabled;" : "disabled;",
 692                "indirect descriptors:",
 693                info->max_indirect_segments ? "enabled;" : "disabled;");
 694 }
 695
 696 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
 697 {
 698         int major;
 699         major = BLKIF_MAJOR(vdevice);
 700         *minor = BLKIF_MINOR(vdevice);
 701         switch (major) {
 702                 case XEN_IDE0_MAJOR:
 703                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
 704                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
 705                                 EMULATED_HD_DISK_MINOR_OFFSET;
 706                         break;
 707                 case XEN_IDE1_MAJOR:
 708                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
 709                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
 710                                 EMULATED_HD_DISK_MINOR_OFFSET;
 711                         break;
 712                 case XEN_SCSI_DISK0_MAJOR:
 713                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
 714                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
 715                         break;
 716                 case XEN_SCSI_DISK1_MAJOR:
 717                 case XEN_SCSI_DISK2_MAJOR:
 718                 case XEN_SCSI_DISK3_MAJOR:
 719                 case XEN_SCSI_DISK4_MAJOR:
 720                 case XEN_SCSI_DISK5_MAJOR:
 721                 case XEN_SCSI_DISK6_MAJOR:
 722                 case XEN_SCSI_DISK7_MAJOR:
 723                         *offset = (*minor / PARTS_PER_DISK) +
 724                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
 725                                 EMULATED_SD_DISK_NAME_OFFSET;
 726                         *minor = *minor +
 727                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
 728                                 EMULATED_SD_DISK_MINOR_OFFSET;
 729                         break;
 730                 case XEN_SCSI_DISK8_MAJOR:
 731                 case XEN_SCSI_DISK9_MAJOR:
 732                 case XEN_SCSI_DISK10_MAJOR:
 733                 case XEN_SCSI_DISK11_MAJOR:
 734                 case XEN_SCSI_DISK12_MAJOR:
 735                 case XEN_SCSI_DISK13_MAJOR:
 736                 case XEN_SCSI_DISK14_MAJOR:
 737                 case XEN_SCSI_DISK15_MAJOR:
 738                         *offset = (*minor / PARTS_PER_DISK) +
 739                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
 740                                 EMULATED_SD_DISK_NAME_OFFSET;
 741                         *minor = *minor +
 742                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
 743                                 EMULATED_SD_DISK_MINOR_OFFSET;
 744                         break;
 745                 case XENVBD_MAJOR:
 746                         *offset = *minor / PARTS_PER_DISK;
 747                         break;
 748                 default:
 749                         printk(KERN_WARNING "blkfront: your disk configuration is "
 750                                         "incorrect, please use an xvd device instead\n");
 751                         return -ENODEV;
 752         }
 753         return 0;
 754 }
 755
 756 static char *encode_disk_name(char *ptr, unsigned int n)
 757 {
 758         if (n >= 26)
 759                 ptr = encode_disk_name(ptr, n / 26 - 1);
 760         *ptr = 'a' + n % 26;
 761         return ptr + 1;
 762 }
 763
 764 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
 765                                struct blkfront_info *info,
 766                                u16 vdisk_info, u16 sector_size,
 767                                unsigned int physical_sector_size)
 768 {
 769         struct gendisk *gd;
 770         int nr_minors = 1;
 771         int err;
 772         unsigned int offset;
 773         int minor;
 774         int nr_parts;
 775         char *ptr;
 776
 777         BUG_ON(info->gd != NULL);
 778         BUG_ON(info->rq != NULL);
 779
 780         if ((info->vdevice>>EXT_SHIFT) > 1) {
 781                 /* this is above the extended range; something is wrong */
 782                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
 783                 return -ENODEV;
 784         }
 785
 786         if (!VDEV_IS_EXTENDED(info->vdevice)) {
 787                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
 788                 if (err)
 789                         return err;
 790                 nr_parts = PARTS_PER_DISK;
 791         } else {
 792                 minor = BLKIF_MINOR_EXT(info->vdevice);
 793                 nr_parts = PARTS_PER_EXT_DISK;
 794                 offset = minor / nr_parts;
 795                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
 796                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
 797                                         "emulated IDE disks,\n\t choose an xvd device name"
 798                                         "from xvde on\n", info->vdevice);
 799         }
 800         if (minor >> MINORBITS) {
 801                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
 802                         info->vdevice, minor);
 803                 return -ENODEV;
 804         }
 805
 806         if ((minor % nr_parts) == 0)
 807                 nr_minors = nr_parts;
 808
 809         err = xlbd_reserve_minors(minor, nr_minors);
 810         if (err)
 811                 goto out;
 812         err = -ENODEV;
 813
 814         gd = alloc_disk(nr_minors);
 815         if (gd == NULL)
 816                 goto release;
 817
 818         strcpy(gd->disk_name, DEV_NAME);
 819         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
 820         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
 821         if (nr_minors > 1)
 822                 *ptr = 0;
 823         else
 824                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
 825                          "%d", minor & (nr_parts - 1));
 826
 827         gd->major = XENVBD_MAJOR;
 828         gd->first_minor = minor;
 829         gd->fops = &xlvbd_block_fops;
 830         gd->private_data = info;
 831         gd->driverfs_dev = &(info->xbdev->dev);
 832         set_capacity(gd, capacity);
 833
 834         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
 835                                  info->max_indirect_segments ? :
 836                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 837                 del_gendisk(gd);
 838                 goto release;
 839         }
 840
 841         info->rq = gd->queue;
 842         info->gd = gd;
 843
 844         xlvbd_flush(info);
 845
 846         if (vdisk_info & VDISK_READONLY)
 847                 set_disk_ro(gd, 1);
 848
 849         if (vdisk_info & VDISK_REMOVABLE)
 850                 gd->flags |= GENHD_FL_REMOVABLE;
 851
 852         if (vdisk_info & VDISK_CDROM)
 853                 gd->flags |= GENHD_FL_CD;
 854
 855         return 0;
 856
 857  release:
 858         xlbd_release_minors(minor, nr_minors);
 859  out:
 860         return err;
 861 }
 862
 863 static void xlvbd_release_gendisk(struct blkfront_info *info)
 864 {
 865         unsigned int minor, nr_minors;
 866         unsigned long flags;
 867
 868         if (info->rq == NULL)
 869                 return;
 870
 871         spin_lock_irqsave(&info->io_lock, flags);
 872
 873         /* No more blkif_request(). */
 874         blk_stop_queue(info->rq);
 875
 876         /* No more gnttab callback work. */
 877         gnttab_cancel_free_callback(&info->callback);
 878         spin_unlock_irqrestore(&info->io_lock, flags);
 879
 880         /* Flush gnttab callback work. Must be done with no locks held. */
 881         flush_work(&info->work);
 882
 883         del_gendisk(info->gd);
 884
 885         minor = info->gd->first_minor;
 886         nr_minors = info->gd->minors;
 887         xlbd_release_minors(minor, nr_minors);
 888
 889         blk_cleanup_queue(info->rq);
 890         info->rq = NULL;
 891
 892         put_disk(info->gd);
 893         info->gd = NULL;
 894 }
 895
 896 static void kick_pending_request_queues(struct blkfront_info *info)
 897 {
 898         if (!RING_FULL(&info->ring)) {
 899                 /* Re-enable calldowns. */
 900                 blk_start_queue(info->rq);
 901                 /* Kick things off immediately. */
 902                 do_blkif_request(info->rq);
 903         }
 904 }
 905
 906 static void blkif_restart_queue(struct work_struct *work)
 907 {
 908         struct blkfront_info *info = container_of(work, struct blkfront_info, work);
 909
 910         spin_lock_irq(&info->io_lock);
 911         if (info->connected == BLKIF_STATE_CONNECTED)
 912                 kick_pending_request_queues(info);
 913         spin_unlock_irq(&info->io_lock);
 914 }
 915
 916 static void blkif_free(struct blkfront_info *info, int suspend)
 917 {
 918         struct grant *persistent_gnt;
 919         struct grant *n;
 920         int i, j, segs;
 921
 922         /* Prevent new requests being issued until we fix things up. */
 923         spin_lock_irq(&info->io_lock);
 924         info->connected = suspend ?
 925                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
 926         /* No more blkif_request(). */
 927         if (info->rq)
 928                 blk_stop_queue(info->rq);
 929
 930         /* Remove all persistent grants */
 931         if (!list_empty(&info->grants)) {
 932                 list_for_each_entry_safe(persistent_gnt, n,
 933                                          &info->grants, node) {
 934                         list_del(&persistent_gnt->node);
 935                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
 936                                 gnttab_end_foreign_access(persistent_gnt->gref,
 937                                                           0, 0UL);
 938                                 info->persistent_gnts_c--;
 939                         }
 940                         if (info->feature_persistent)
 941                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 942                         kfree(persistent_gnt);
 943                 }
 944         }
 945         BUG_ON(info->persistent_gnts_c != 0);
 946
 947         /*
 948          * Remove indirect pages, this only happens when using indirect
 949          * descriptors but not persistent grants
 950          */
 951         if (!list_empty(&info->indirect_pages)) {
 952                 struct page *indirect_page, *n;
 953
 954                 BUG_ON(info->feature_persistent);
 955                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
 956                         list_del(&indirect_page->lru);
 957                         __free_page(indirect_page);
 958                 }
 959         }
 960
 961         for (i = 0; i < BLK_RING_SIZE; i++) {
 962                 /*
 963                  * Clear persistent grants present in requests already
 964                  * on the shared ring
 965                  */
 966                 if (!info->shadow[i].request)
 967                         goto free_shadow;
 968
 969                 segs = info->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
 970                        info->shadow[i].req.u.indirect.nr_segments :
 971                        info->shadow[i].req.u.rw.nr_segments;
 972                 for (j = 0; j < segs; j++) {
 973                         persistent_gnt = info->shadow[i].grants_used[j];
 974                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
 975                         if (info->feature_persistent)
 976                                 __free_page(pfn_to_page(persistent_gnt->pfn));
 977                         kfree(persistent_gnt);
 978                 }
 979
 980                 if (info->shadow[i].req.operation != BLKIF_OP_INDIRECT)
 981                         /*
 982                          * If this is not an indirect operation don't try to
 983                          * free indirect segments
 984                          */
 985                         goto free_shadow;
 986
 987                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
 988                         persistent_gnt = info->shadow[i].indirect_grants[j];
 989                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
 990                         __free_page(pfn_to_page(persistent_gnt->pfn));
 991                         kfree(persistent_gnt);
 992                 }
 993
 994 free_shadow:
 995                 kfree(info->shadow[i].grants_used);
 996                 info->shadow[i].grants_used = NULL;
 997                 kfree(info->shadow[i].indirect_grants);
 998                 info->shadow[i].indirect_grants = NULL;
 999                 kfree(info->shadow[i].sg);
1000                 info->shadow[i].sg = NULL;
1001         }
1002
1003         /* No more gnttab callback work. */
1004         gnttab_cancel_free_callback(&info->callback);
1005         spin_unlock_irq(&info->io_lock);
1006
1007         /* Flush gnttab callback work. Must be done with no locks held. */
1008         flush_work(&info->work);
1009
1010         /* Free resources associated with old device channel. */
1011         if (info->ring_ref != GRANT_INVALID_REF) {
1012                 gnttab_end_foreign_access(info->ring_ref, 0,
1013                                           (unsigned long)info->ring.sring);
1014                 info->ring_ref = GRANT_INVALID_REF;
1015                 info->ring.sring = NULL;
1016         }
1017         if (info->irq)
1018                 unbind_from_irqhandler(info->irq, info);
1019         info->evtchn = info->irq = 0;
1020
1021 }
1022
1023 static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
1024                              struct blkif_response *bret)
1025 {
1026         int i = 0;
1027         struct scatterlist *sg;
1028         char *bvec_data;
1029         void *shared_data;
1030         int nseg;
1031
1032         nseg = s->req.operation == BLKIF_OP_INDIRECT ?
1033                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1034
1035         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1036                 /*
1037                  * Copy the data received from the backend into the bvec.
1038                  * Since bv_offset can be different than 0, and bv_len different
1039                  * than PAGE_SIZE, we have to keep track of the current offset,
1040                  * to be sure we are copying the data from the right shared page.
1041                  */
1042                 for_each_sg(s->sg, sg, nseg, i) {
1043                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1044                         shared_data = kmap_atomic(
1045                                 pfn_to_page(s->grants_used[i]->pfn));
1046                         bvec_data = kmap_atomic(sg_page(sg));
1047                         memcpy(bvec_data   + sg->offset,
1048                                shared_data + sg->offset,
1049                                sg->length);
1050                         kunmap_atomic(bvec_data);
1051                         kunmap_atomic(shared_data);
1052                 }
1053         }
1054         /* Add the persistent grant into the list of free grants */
1055         for (i = 0; i < nseg; i++) {
1056                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1057                         /*
1058                          * If the grant is still mapped by the backend (the
1059                          * backend has chosen to make this grant persistent)
1060                          * we add it at the head of the list, so it will be
1061                          * reused first.
1062                          */
1063                         if (!info->feature_persistent)
1064                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1065                                                      s->grants_used[i]->gref);
1066                         list_add(&s->grants_used[i]->node, &info->grants);
1067                         info->persistent_gnts_c++;
1068                 } else {
1069                         /*
1070                          * If the grant is not mapped by the backend we end the
1071                          * foreign access and add it to the tail of the list,
1072                          * so it will not be picked again unless we run out of
1073                          * persistent grants.
1074                          */
1075                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1076                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1077                         list_add_tail(&s->grants_used[i]->node, &info->grants);
1078                 }
1079         }
1080         if (s->req.operation == BLKIF_OP_INDIRECT) {
1081                 for (i = 0; i < INDIRECT_GREFS(nseg); i++) {
1082                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1083                                 if (!info->feature_persistent)
1084                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1085                                                              s->indirect_grants[i]->gref);
1086                                 list_add(&s->indirect_grants[i]->node, &info->grants);
1087                                 info->persistent_gnts_c++;
1088                         } else {
1089                                 struct page *indirect_page;
1090
1091                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1092                                 /*
1093                                  * Add the used indirect page back to the list of
1094                                  * available pages for indirect grefs.
1095                                  */
1096                                 indirect_page = pfn_to_page(s->indirect_grants[i]->pfn);
1097                                 list_add(&indirect_page->lru, &info->indirect_pages);
1098                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1099                                 list_add_tail(&s->indirect_grants[i]->node, &info->grants);
1100                         }
1101                 }
1102         }
1103 }
1104
1105 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1106 {
1107         struct request *req;
1108         struct blkif_response *bret;
1109         RING_IDX i, rp;
1110         unsigned long flags;
1111         struct blkfront_info *info = (struct blkfront_info *)dev_id;
1112         int error;
1113
1114         spin_lock_irqsave(&info->io_lock, flags);
1115
1116         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1117                 spin_unlock_irqrestore(&info->io_lock, flags);
1118                 return IRQ_HANDLED;
1119         }
1120
1121  again:
1122         rp = info->ring.sring->rsp_prod;
1123         rmb(); /* Ensure we see queued responses up to 'rp'. */
1124
1125         for (i = info->ring.rsp_cons; i != rp; i++) {
1126                 unsigned long id;
1127
1128                 bret = RING_GET_RESPONSE(&info->ring, i);
1129                 id   = bret->id;
1130                 /*
1131                  * The backend has messed up and given us an id that we would
1132                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1133                  * look in get_id_from_freelist.
1134                  */
1135                 if (id >= BLK_RING_SIZE) {
1136                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1137                              info->gd->disk_name, op_name(bret->operation), id);
1138                         /* We can't safely get the 'struct request' as
1139                          * the id is busted. */
1140                         continue;
1141                 }
1142                 req  = info->shadow[id].request;
1143
1144                 if (bret->operation != BLKIF_OP_DISCARD)
1145                         blkif_completion(&info->shadow[id], info, bret);
1146
1147                 if (add_id_to_freelist(info, id)) {
1148                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1149                              info->gd->disk_name, op_name(bret->operation), id);
1150                         continue;
1151                 }
1152
1153                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1154                 switch (bret->operation) {
1155                 case BLKIF_OP_DISCARD:
1156                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1157                                 struct request_queue *rq = info->rq;
1158                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1159                                            info->gd->disk_name, op_name(bret->operation));
1160                                 error = -EOPNOTSUPP;
1161                                 info->feature_discard = 0;
1162                                 info->feature_secdiscard = 0;
1163                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1164                                 queue_flag_clear(QUEUE_FLAG_SECDISCARD, rq);
1165                         }
1166                         __blk_end_request_all(req, error);
1167                         break;
1168                 case BLKIF_OP_FLUSH_DISKCACHE:
1169                 case BLKIF_OP_WRITE_BARRIER:
1170                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1171                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1172                                        info->gd->disk_name, op_name(bret->operation));
1173                                 error = -EOPNOTSUPP;
1174                         }
1175                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1176                                      info->shadow[id].req.u.rw.nr_segments == 0)) {
1177                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1178                                        info->gd->disk_name, op_name(bret->operation));
1179                                 error = -EOPNOTSUPP;
1180                         }
1181                         if (unlikely(error)) {
1182                                 if (error == -EOPNOTSUPP)
1183                                         error = 0;
1184                                 info->feature_flush = 0;
1185                                 info->flush_op = 0;
1186                                 xlvbd_flush(info);
1187                         }
1188                         /* fall through */
1189                 case BLKIF_OP_READ:
1190                 case BLKIF_OP_WRITE:
1191                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1192                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1193                                         "request: %x\n", bret->status);
1194
1195                         __blk_end_request_all(req, error);
1196                         break;
1197                 default:
1198                         BUG();
1199                 }
1200         }
1201
1202         info->ring.rsp_cons = i;
1203
1204         if (i != info->ring.req_prod_pvt) {
1205                 int more_to_do;
1206                 RING_FINAL_CHECK_FOR_RESPONSES(&info->ring, more_to_do);
1207                 if (more_to_do)
1208                         goto again;
1209         } else
1210                 info->ring.sring->rsp_event = i + 1;
1211
1212         kick_pending_request_queues(info);
1213
1214         spin_unlock_irqrestore(&info->io_lock, flags);
1215
1216         return IRQ_HANDLED;
1217 }
1218
1219
1220 static int setup_blkring(struct xenbus_device *dev,
1221                          struct blkfront_info *info)
1222 {
1223         struct blkif_sring *sring;
1224         int err;
1225
1226         info->ring_ref = GRANT_INVALID_REF;
1227
1228         sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
1229         if (!sring) {
1230                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1231                 return -ENOMEM;
1232         }
1233         SHARED_RING_INIT(sring);
1234         FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
1235
1236         err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
1237         if (err < 0) {
1238                 free_page((unsigned long)sring);
1239                 info->ring.sring = NULL;
1240                 goto fail;
1241         }
1242         info->ring_ref = err;
1243
1244         err = xenbus_alloc_evtchn(dev, &info->evtchn);
1245         if (err)
1246                 goto fail;
1247
1248         err = bind_evtchn_to_irqhandler(info->evtchn, blkif_interrupt, 0,
1249                                         "blkif", info);
1250         if (err <= 0) {
1251                 xenbus_dev_fatal(dev, err,
1252                                  "bind_evtchn_to_irqhandler failed");
1253                 goto fail;
1254         }
1255         info->irq = err;
1256
1257         return 0;
1258 fail:
1259         blkif_free(info, 0);
1260         return err;
1261 }
1262
1263
1264 /* Common code used when first setting up, and when resuming. */
1265 static int talk_to_blkback(struct xenbus_device *dev,
1266                            struct blkfront_info *info)
1267 {
1268         const char *message = NULL;
1269         struct xenbus_transaction xbt;
1270         int err;
1271
1272         /* Create shared ring, alloc event channel. */
1273         err = setup_blkring(dev, info);
1274         if (err)
1275                 goto out;
1276
1277 again:
1278         err = xenbus_transaction_start(&xbt);
1279         if (err) {
1280                 xenbus_dev_fatal(dev, err, "starting transaction");
1281                 goto destroy_blkring;
1282         }
1283
1284         err = xenbus_printf(xbt, dev->nodename,
1285                             "ring-ref", "%u", info->ring_ref);
1286         if (err) {
1287                 message = "writing ring-ref";
1288                 goto abort_transaction;
1289         }
1290         err = xenbus_printf(xbt, dev->nodename,
1291                             "event-channel", "%u", info->evtchn);
1292         if (err) {
1293                 message = "writing event-channel";
1294                 goto abort_transaction;
1295         }
1296         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1297                             XEN_IO_PROTO_ABI_NATIVE);
1298         if (err) {
1299                 message = "writing protocol";
1300                 goto abort_transaction;
1301         }
1302         err = xenbus_printf(xbt, dev->nodename,
1303                             "feature-persistent", "%u", 1);
1304         if (err)
1305                 dev_warn(&dev->dev,
1306                          "writing persistent grants feature to xenbus");
1307
1308         err = xenbus_transaction_end(xbt, 0);
1309         if (err) {
1310                 if (err == -EAGAIN)
1311                         goto again;
1312                 xenbus_dev_fatal(dev, err, "completing transaction");
1313                 goto destroy_blkring;
1314         }
1315
1316         xenbus_switch_state(dev, XenbusStateInitialised);
1317
1318         return 0;
1319
1320  abort_transaction:
1321         xenbus_transaction_end(xbt, 1);
1322         if (message)
1323                 xenbus_dev_fatal(dev, err, "%s", message);
1324  destroy_blkring:
1325         blkif_free(info, 0);
1326  out:
1327         return err;
1328 }
1329
1330 /**
1331  * Entry point to this code when a new device is created.  Allocate the basic
1332  * structures and the ring buffer for communication with the backend, and
1333  * inform the backend of the appropriate details for those.  Switch to
1334  * Initialised state.
1335  */
1336 static int blkfront_probe(struct xenbus_device *dev,
1337                           const struct xenbus_device_id *id)
1338 {
1339         int err, vdevice, i;
1340         struct blkfront_info *info;
1341
1342         /* FIXME: Use dynamic device id if this is not set. */
1343         err = xenbus_scanf(XBT_NIL, dev->nodename,
1344                            "virtual-device", "%i", &vdevice);
1345         if (err != 1) {
1346                 /* go looking in the extended area instead */
1347                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1348                                    "%i", &vdevice);
1349                 if (err != 1) {
1350                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1351                         return err;
1352                 }
1353         }
1354
1355         if (xen_hvm_domain()) {
1356                 char *type;
1357                 int len;
1358                 /* no unplug has been done: do not hook devices != xen vbds */
1359                 if (xen_platform_pci_unplug & XEN_UNPLUG_UNNECESSARY) {
1360                         int major;
1361
1362                         if (!VDEV_IS_EXTENDED(vdevice))
1363                                 major = BLKIF_MAJOR(vdevice);
1364                         else
1365                                 major = XENVBD_MAJOR;
1366
1367                         if (major != XENVBD_MAJOR) {
1368                                 printk(KERN_INFO
1369                                                 "%s: HVM does not support vbd %d as xen block device\n",
1370                                                 __FUNCTION__, vdevice);
1371                                 return -ENODEV;
1372                         }
1373                 }
1374                 /* do not create a PV cdrom device if we are an HVM guest */
1375                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1376                 if (IS_ERR(type))
1377                         return -ENODEV;
1378                 if (strncmp(type, "cdrom", 5) == 0) {
1379                         kfree(type);
1380                         return -ENODEV;
1381                 }
1382                 kfree(type);
1383         }
1384         info = kzalloc(sizeof(*info), GFP_KERNEL);
1385         if (!info) {
1386                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1387                 return -ENOMEM;
1388         }
1389
1390         mutex_init(&info->mutex);
1391         spin_lock_init(&info->io_lock);
1392         info->xbdev = dev;
1393         info->vdevice = vdevice;
1394         INIT_LIST_HEAD(&info->grants);
1395         INIT_LIST_HEAD(&info->indirect_pages);
1396         info->persistent_gnts_c = 0;
1397         info->connected = BLKIF_STATE_DISCONNECTED;
1398         INIT_WORK(&info->work, blkif_restart_queue);
1399
1400         for (i = 0; i < BLK_RING_SIZE; i++)
1401                 info->shadow[i].req.u.rw.id = i+1;
1402         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1403
1404         /* Front end dir is a number, which is used as the id. */
1405         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1406         dev_set_drvdata(&dev->dev, info);
1407
1408         err = talk_to_blkback(dev, info);
1409         if (err) {
1410                 kfree(info);
1411                 dev_set_drvdata(&dev->dev, NULL);
1412                 return err;
1413         }
1414
1415         return 0;
1416 }
1417
1418 static void split_bio_end(struct bio *bio, int error)
1419 {
1420         struct split_bio *split_bio = bio->bi_private;
1421
1422         if (error)
1423                 split_bio->err = error;
1424
1425         if (atomic_dec_and_test(&split_bio->pending)) {
1426                 split_bio->bio->bi_phys_segments = 0;
1427                 bio_endio(split_bio->bio, split_bio->err);
1428                 kfree(split_bio);
1429         }
1430         bio_put(bio);
1431 }
1432
1433 static int blkif_recover(struct blkfront_info *info)
1434 {
1435         int i;
1436         struct request *req, *n;
1437         struct blk_shadow *copy;
1438         int rc;
1439         struct bio *bio, *cloned_bio;
1440         struct bio_list bio_list, merge_bio;
1441         unsigned int segs, offset;
1442         int pending, size;
1443         struct split_bio *split_bio;
1444         struct list_head requests;
1445
1446         /* Stage 1: Make a safe copy of the shadow state. */
1447         copy = kmemdup(info->shadow, sizeof(info->shadow),
1448                        GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
1449         if (!copy)
1450                 return -ENOMEM;
1451
1452         /* Stage 2: Set up free list. */
1453         memset(&info->shadow, 0, sizeof(info->shadow));
1454         for (i = 0; i < BLK_RING_SIZE; i++)
1455                 info->shadow[i].req.u.rw.id = i+1;
1456         info->shadow_free = info->ring.req_prod_pvt;
1457         info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
1458
1459         rc = blkfront_setup_indirect(info);
1460         if (rc) {
1461                 kfree(copy);
1462                 return rc;
1463         }
1464
1465         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
1466         blk_queue_max_segments(info->rq, segs);
1467         bio_list_init(&bio_list);
1468         INIT_LIST_HEAD(&requests);
1469         for (i = 0; i < BLK_RING_SIZE; i++) {
1470                 /* Not in use? */
1471                 if (!copy[i].request)
1472                         continue;
1473
1474                 /*
1475                  * Get the bios in the request so we can re-queue them.
1476                  */
1477                 if (copy[i].request->cmd_flags &
1478                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1479                         /*
1480                          * Flush operations don't contain bios, so
1481                          * we need to requeue the whole request
1482                          */
1483                         list_add(&copy[i].request->queuelist, &requests);
1484                         continue;
1485                 }
1486                 merge_bio.head = copy[i].request->bio;
1487                 merge_bio.tail = copy[i].request->biotail;
1488                 bio_list_merge(&bio_list, &merge_bio);
1489                 copy[i].request->bio = NULL;
1490                 blk_put_request(copy[i].request);
1491         }
1492
1493         kfree(copy);
1494
1495         /*
1496          * Empty the queue, this is important because we might have
1497          * requests in the queue with more segments than what we
1498          * can handle now.
1499          */
1500         spin_lock_irq(&info->io_lock);
1501         while ((req = blk_fetch_request(info->rq)) != NULL) {
1502                 if (req->cmd_flags &
1503                     (REQ_FLUSH | REQ_FUA | REQ_DISCARD | REQ_SECURE)) {
1504                         list_add(&req->queuelist, &requests);
1505                         continue;
1506                 }
1507                 merge_bio.head = req->bio;
1508                 merge_bio.tail = req->biotail;
1509                 bio_list_merge(&bio_list, &merge_bio);
1510                 req->bio = NULL;
1511                 if (req->cmd_flags & (REQ_FLUSH | REQ_FUA))
1512                         pr_alert("diskcache flush request found!\n");
1513                 __blk_put_request(info->rq, req);
1514         }
1515         spin_unlock_irq(&info->io_lock);
1516
1517         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1518
1519         spin_lock_irq(&info->io_lock);
1520
1521         /* Now safe for us to use the shared ring */
1522         info->connected = BLKIF_STATE_CONNECTED;
1523
1524         /* Kick any other new requests queued since we resumed */
1525         kick_pending_request_queues(info);
1526
1527         list_for_each_entry_safe(req, n, &requests, queuelist) {
1528                 /* Requeue pending requests (flush or discard) */
1529                 list_del_init(&req->queuelist);
1530                 BUG_ON(req->nr_phys_segments > segs);
1531                 blk_requeue_request(info->rq, req);
1532         }
1533         spin_unlock_irq(&info->io_lock);
1534
1535         while ((bio = bio_list_pop(&bio_list)) != NULL) {
1536                 /* Traverse the list of pending bios and re-queue them */
1537                 if (bio_segments(bio) > segs) {
1538                         /*
1539                          * This bio has more segments than what we can
1540                          * handle, we have to split it.
1541                          */
1542                         pending = (bio_segments(bio) + segs - 1) / segs;
1543                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
1544                         BUG_ON(split_bio == NULL);
1545                         atomic_set(&split_bio->pending, pending);
1546                         split_bio->bio = bio;
1547                         for (i = 0; i < pending; i++) {
1548                                 offset = (i * segs * PAGE_SIZE) >> 9;
1549                                 size = min((unsigned int)(segs * PAGE_SIZE) >> 9,
1550                                            (unsigned int)(bio->bi_size >> 9) - offset);
1551                                 cloned_bio = bio_clone(bio, GFP_NOIO);
1552                                 BUG_ON(cloned_bio == NULL);
1553                                 bio_trim(cloned_bio, offset, size);
1554                                 cloned_bio->bi_private = split_bio;
1555                                 cloned_bio->bi_end_io = split_bio_end;
1556                                 submit_bio(cloned_bio->bi_rw, cloned_bio);
1557                         }
1558                         /*
1559                          * Now we have to wait for all those smaller bios to
1560                          * end, so we can also end the "parent" bio.
1561                          */
1562                         continue;
1563                 }
1564                 /* We don't need to split this bio */
1565                 submit_bio(bio->bi_rw, bio);
1566         }
1567
1568         return 0;
1569 }
1570
1571 /**
1572  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1573  * driver restart.  We tear down our blkif structure and recreate it, but
1574  * leave the device-layer structures intact so that this is transparent to the
1575  * rest of the kernel.
1576  */
1577 static int blkfront_resume(struct xenbus_device *dev)
1578 {
1579         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1580         int err;
1581
1582         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
1583
1584         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
1585
1586         err = talk_to_blkback(dev, info);
1587
1588         /*
1589          * We have to wait for the backend to switch to
1590          * connected state, since we want to read which
1591          * features it supports.
1592          */
1593
1594         return err;
1595 }
1596
1597 static void
1598 blkfront_closing(struct blkfront_info *info)
1599 {
1600         struct xenbus_device *xbdev = info->xbdev;
1601         struct block_device *bdev = NULL;
1602
1603         mutex_lock(&info->mutex);
1604
1605         if (xbdev->state == XenbusStateClosing) {
1606                 mutex_unlock(&info->mutex);
1607                 return;
1608         }
1609
1610         if (info->gd)
1611                 bdev = bdget_disk(info->gd, 0);
1612
1613         mutex_unlock(&info->mutex);
1614
1615         if (!bdev) {
1616                 xenbus_frontend_closed(xbdev);
1617                 return;
1618         }
1619
1620         mutex_lock(&bdev->bd_mutex);
1621
1622         if (bdev->bd_openers) {
1623                 xenbus_dev_error(xbdev, -EBUSY,
1624                                  "Device in use; refusing to close");
1625                 xenbus_switch_state(xbdev, XenbusStateClosing);
1626         } else {
1627                 xlvbd_release_gendisk(info);
1628                 xenbus_frontend_closed(xbdev);
1629         }
1630
1631         mutex_unlock(&bdev->bd_mutex);
1632         bdput(bdev);
1633 }
1634
1635 static void blkfront_setup_discard(struct blkfront_info *info)
1636 {
1637         int err;
1638         char *type;
1639         unsigned int discard_granularity;
1640         unsigned int discard_alignment;
1641         unsigned int discard_secure;
1642
1643         type = xenbus_read(XBT_NIL, info->xbdev->otherend, "type", NULL);
1644         if (IS_ERR(type))
1645                 return;
1646
1647         info->feature_secdiscard = 0;
1648         if (strncmp(type, "phy", 3) == 0) {
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->feature_discard = 1;
1655                         info->discard_granularity = discard_granularity;
1656                         info->discard_alignment = discard_alignment;
1657                 }
1658                 err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1659                             "discard-secure", "%d", &discard_secure,
1660                             NULL);
1661                 if (!err)
1662                         info->feature_secdiscard = discard_secure;
1663
1664         } else if (strncmp(type, "file", 4) == 0)
1665                 info->feature_discard = 1;
1666
1667         kfree(type);
1668 }
1669
1670 static int blkfront_setup_indirect(struct blkfront_info *info)
1671 {
1672         unsigned int indirect_segments, segs;
1673         int err, i;
1674
1675         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1676                             "feature-max-indirect-segments", "%u", &indirect_segments,
1677                             NULL);
1678         if (err) {
1679                 info->max_indirect_segments = 0;
1680                 segs = BLKIF_MAX_SEGMENTS_PER_REQUEST;
1681         } else {
1682                 info->max_indirect_segments = min(indirect_segments,
1683                                                   xen_blkif_max_segments);
1684                 segs = info->max_indirect_segments;
1685         }
1686
1687         err = fill_grant_buffer(info, (segs + INDIRECT_GREFS(segs)) * BLK_RING_SIZE);
1688         if (err)
1689                 goto out_of_memory;
1690
1691         if (!info->feature_persistent && info->max_indirect_segments) {
1692                 /*
1693                  * We are using indirect descriptors but not persistent
1694                  * grants, we need to allocate a set of pages that can be
1695                  * used for mapping indirect grefs
1696                  */
1697                 int num = INDIRECT_GREFS(segs) * BLK_RING_SIZE;
1698
1699                 BUG_ON(!list_empty(&info->indirect_pages));
1700                 for (i = 0; i < num; i++) {
1701                         struct page *indirect_page = alloc_page(GFP_NOIO);
1702                         if (!indirect_page)
1703                                 goto out_of_memory;
1704                         list_add(&indirect_page->lru, &info->indirect_pages);
1705                 }
1706         }
1707
1708         for (i = 0; i < BLK_RING_SIZE; i++) {
1709                 info->shadow[i].grants_used = kzalloc(
1710                         sizeof(info->shadow[i].grants_used[0]) * segs,
1711                         GFP_NOIO);
1712                 info->shadow[i].sg = kzalloc(sizeof(info->shadow[i].sg[0]) * segs, GFP_NOIO);
1713                 if (info->max_indirect_segments)
1714                         info->shadow[i].indirect_grants = kzalloc(
1715                                 sizeof(info->shadow[i].indirect_grants[0]) *
1716                                 INDIRECT_GREFS(segs),
1717                                 GFP_NOIO);
1718                 if ((info->shadow[i].grants_used == NULL) ||
1719                         (info->shadow[i].sg == NULL) ||
1720                      (info->max_indirect_segments &&
1721                      (info->shadow[i].indirect_grants == NULL)))
1722                         goto out_of_memory;
1723                 sg_init_table(info->shadow[i].sg, segs);
1724         }
1725
1726
1727         return 0;
1728
1729 out_of_memory:
1730         for (i = 0; i < BLK_RING_SIZE; i++) {
1731                 kfree(info->shadow[i].grants_used);
1732                 info->shadow[i].grants_used = NULL;
1733                 kfree(info->shadow[i].sg);
1734                 info->shadow[i].sg = NULL;
1735                 kfree(info->shadow[i].indirect_grants);
1736                 info->shadow[i].indirect_grants = NULL;
1737         }
1738         if (!list_empty(&info->indirect_pages)) {
1739                 struct page *indirect_page, *n;
1740                 list_for_each_entry_safe(indirect_page, n, &info->indirect_pages, lru) {
1741                         list_del(&indirect_page->lru);
1742                         __free_page(indirect_page);
1743                 }
1744         }
1745         return -ENOMEM;
1746 }
1747
1748 /*
1749  * Invoked when the backend is finally 'ready' (and has told produced
1750  * the details about the physical device - #sectors, size, etc).
1751  */
1752 static void blkfront_connect(struct blkfront_info *info)
1753 {
1754         unsigned long long sectors;
1755         unsigned long sector_size;
1756         unsigned int physical_sector_size;
1757         unsigned int binfo;
1758         int err;
1759         int barrier, flush, discard, persistent;
1760
1761         switch (info->connected) {
1762         case BLKIF_STATE_CONNECTED:
1763                 /*
1764                  * Potentially, the back-end may be signalling
1765                  * a capacity change; update the capacity.
1766                  */
1767                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1768                                    "sectors", "%Lu", &sectors);
1769                 if (XENBUS_EXIST_ERR(err))
1770                         return;
1771                 printk(KERN_INFO "Setting capacity to %Lu\n",
1772                        sectors);
1773                 set_capacity(info->gd, sectors);
1774                 revalidate_disk(info->gd);
1775
1776                 return;
1777         case BLKIF_STATE_SUSPENDED:
1778                 /*
1779                  * If we are recovering from suspension, we need to wait
1780                  * for the backend to announce it's features before
1781                  * reconnecting, at least we need to know if the backend
1782                  * supports indirect descriptors, and how many.
1783                  */
1784                 blkif_recover(info);
1785                 return;
1786
1787         default:
1788                 break;
1789         }
1790
1791         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
1792                 __func__, info->xbdev->otherend);
1793
1794         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1795                             "sectors", "%llu", &sectors,
1796                             "info", "%u", &binfo,
1797                             "sector-size", "%lu", &sector_size,
1798                             NULL);
1799         if (err) {
1800                 xenbus_dev_fatal(info->xbdev, err,
1801                                  "reading backend fields at %s",
1802                                  info->xbdev->otherend);
1803                 return;
1804         }
1805
1806         /*
1807          * physcial-sector-size is a newer field, so old backends may not
1808          * provide this. Assume physical sector size to be the same as
1809          * sector_size in that case.
1810          */
1811         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1812                            "physical-sector-size", "%u", &physical_sector_size);
1813         if (err != 1)
1814                 physical_sector_size = sector_size;
1815
1816         info->feature_flush = 0;
1817         info->flush_op = 0;
1818
1819         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1820                             "feature-barrier", "%d", &barrier,
1821                             NULL);
1822
1823         /*
1824          * If there's no "feature-barrier" defined, then it means
1825          * we're dealing with a very old backend which writes
1826          * synchronously; nothing to do.
1827          *
1828          * If there are barriers, then we use flush.
1829          */
1830         if (!err && barrier) {
1831                 info->feature_flush = REQ_FLUSH | REQ_FUA;
1832                 info->flush_op = BLKIF_OP_WRITE_BARRIER;
1833         }
1834         /*
1835          * And if there is "feature-flush-cache" use that above
1836          * barriers.
1837          */
1838         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1839                             "feature-flush-cache", "%d", &flush,
1840                             NULL);
1841
1842         if (!err && flush) {
1843                 info->feature_flush = REQ_FLUSH;
1844                 info->flush_op = BLKIF_OP_FLUSH_DISKCACHE;
1845         }
1846
1847         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1848                             "feature-discard", "%d", &discard,
1849                             NULL);
1850
1851         if (!err && discard)
1852                 blkfront_setup_discard(info);
1853
1854         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
1855                             "feature-persistent", "%u", &persistent,
1856                             NULL);
1857         if (err)
1858                 info->feature_persistent = 0;
1859         else
1860                 info->feature_persistent = persistent;
1861
1862         err = blkfront_setup_indirect(info);
1863         if (err) {
1864                 xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
1865                                  info->xbdev->otherend);
1866                 return;
1867         }
1868
1869         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
1870                                   physical_sector_size);
1871         if (err) {
1872                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
1873                                  info->xbdev->otherend);
1874                 return;
1875         }
1876
1877         xenbus_switch_state(info->xbdev, XenbusStateConnected);
1878
1879         /* Kick pending requests. */
1880         spin_lock_irq(&info->io_lock);
1881         info->connected = BLKIF_STATE_CONNECTED;
1882         kick_pending_request_queues(info);
1883         spin_unlock_irq(&info->io_lock);
1884
1885         add_disk(info->gd);
1886
1887         info->is_ready = 1;
1888 }
1889
1890 /**
1891  * Callback received when the backend's state changes.
1892  */
1893 static void blkback_changed(struct xenbus_device *dev,
1894                             enum xenbus_state backend_state)
1895 {
1896         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1897
1898         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
1899
1900         switch (backend_state) {
1901         case XenbusStateInitialising:
1902         case XenbusStateInitWait:
1903         case XenbusStateInitialised:
1904         case XenbusStateReconfiguring:
1905         case XenbusStateReconfigured:
1906         case XenbusStateUnknown:
1907         case XenbusStateClosed:
1908                 break;
1909
1910         case XenbusStateConnected:
1911                 blkfront_connect(info);
1912                 break;
1913
1914         case XenbusStateClosing:
1915                 blkfront_closing(info);
1916                 break;
1917         }
1918 }
1919
1920 static int blkfront_remove(struct xenbus_device *xbdev)
1921 {
1922         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
1923         struct block_device *bdev = NULL;
1924         struct gendisk *disk;
1925
1926         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
1927
1928         blkif_free(info, 0);
1929
1930         mutex_lock(&info->mutex);
1931
1932         disk = info->gd;
1933         if (disk)
1934                 bdev = bdget_disk(disk, 0);
1935
1936         info->xbdev = NULL;
1937         mutex_unlock(&info->mutex);
1938
1939         if (!bdev) {
1940                 kfree(info);
1941                 return 0;
1942         }
1943
1944         /*
1945          * The xbdev was removed before we reached the Closed
1946          * state. See if it's safe to remove the disk. If the bdev
1947          * isn't closed yet, we let release take care of it.
1948          */
1949
1950         mutex_lock(&bdev->bd_mutex);
1951         info = disk->private_data;
1952
1953         dev_warn(disk_to_dev(disk),
1954                  "%s was hot-unplugged, %d stale handles\n",
1955                  xbdev->nodename, bdev->bd_openers);
1956
1957         if (info && !bdev->bd_openers) {
1958                 xlvbd_release_gendisk(info);
1959                 disk->private_data = NULL;
1960                 kfree(info);
1961         }
1962
1963         mutex_unlock(&bdev->bd_mutex);
1964         bdput(bdev);
1965
1966         return 0;
1967 }
1968
1969 static int blkfront_is_ready(struct xenbus_device *dev)
1970 {
1971         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
1972
1973         return info->is_ready && info->xbdev;
1974 }
1975
1976 static int blkif_open(struct block_device *bdev, fmode_t mode)
1977 {
1978         struct gendisk *disk = bdev->bd_disk;
1979         struct blkfront_info *info;
1980         int err = 0;
1981
1982         mutex_lock(&blkfront_mutex);
1983
1984         info = disk->private_data;
1985         if (!info) {
1986                 /* xbdev gone */
1987                 err = -ERESTARTSYS;
1988                 goto out;
1989         }
1990
1991         mutex_lock(&info->mutex);
1992
1993         if (!info->gd)
1994                 /* xbdev is closed */
1995                 err = -ERESTARTSYS;
1996
1997         mutex_unlock(&info->mutex);
1998
1999 out:
2000         mutex_unlock(&blkfront_mutex);
2001         return err;
2002 }
2003
2004 static void blkif_release(struct gendisk *disk, fmode_t mode)
2005 {
2006         struct blkfront_info *info = disk->private_data;
2007         struct block_device *bdev;
2008         struct xenbus_device *xbdev;
2009
2010         mutex_lock(&blkfront_mutex);
2011
2012         bdev = bdget_disk(disk, 0);
2013
2014         if (!bdev) {
2015                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2016                 goto out_mutex;
2017         }
2018         if (bdev->bd_openers)
2019                 goto out;
2020
2021         /*
2022          * Check if we have been instructed to close. We will have
2023          * deferred this request, because the bdev was still open.
2024          */
2025
2026         mutex_lock(&info->mutex);
2027         xbdev = info->xbdev;
2028
2029         if (xbdev && xbdev->state == XenbusStateClosing) {
2030                 /* pending switch to state closed */
2031                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2032                 xlvbd_release_gendisk(info);
2033                 xenbus_frontend_closed(info->xbdev);
2034         }
2035
2036         mutex_unlock(&info->mutex);
2037
2038         if (!xbdev) {
2039                 /* sudden device removal */
2040                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2041                 xlvbd_release_gendisk(info);
2042                 disk->private_data = NULL;
2043                 kfree(info);
2044         }
2045
2046 out:
2047         bdput(bdev);
2048 out_mutex:
2049         mutex_unlock(&blkfront_mutex);
2050 }
2051
2052 static const struct block_device_operations xlvbd_block_fops =
2053 {
2054         .owner = THIS_MODULE,
2055         .open = blkif_open,
2056         .release = blkif_release,
2057         .getgeo = blkif_getgeo,
2058         .ioctl = blkif_ioctl,
2059 };
2060
2061
2062 static const struct xenbus_device_id blkfront_ids[] = {
2063         { "vbd" },
2064         { "" }
2065 };
2066
2067 static DEFINE_XENBUS_DRIVER(blkfront, ,
2068         .probe = blkfront_probe,
2069         .remove = blkfront_remove,
2070         .resume = blkfront_resume,
2071         .otherend_changed = blkback_changed,
2072         .is_ready = blkfront_is_ready,
2073 );
2074
2075 static int __init xlblk_init(void)
2076 {
2077         int ret;
2078
2079         if (!xen_domain())
2080                 return -ENODEV;
2081
2082         if (xen_hvm_domain() && !xen_platform_pci_unplug)
2083                 return -ENODEV;
2084
2085         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2086                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2087                        XENVBD_MAJOR, DEV_NAME);
2088                 return -ENODEV;
2089         }
2090
2091         ret = xenbus_register_frontend(&blkfront_driver);
2092         if (ret) {
2093                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2094                 return ret;
2095         }
2096
2097         return 0;
2098 }
2099 module_init(xlblk_init);
2100
2101
2102 static void __exit xlblk_exit(void)
2103 {
2104         xenbus_unregister_driver(&blkfront_driver);
2105         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2106         kfree(minors);
2107 }
2108 module_exit(xlblk_exit);
2109
2110 MODULE_DESCRIPTION("Xen virtual block device frontend");
2111 MODULE_LICENSE("GPL");
2112 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2113 MODULE_ALIAS("xen:vbd");
2114 MODULE_ALIAS("xenblk");