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