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/blk-mq.h>
  41 #include <linux/hdreg.h>
  42 #include <linux/cdrom.h>
  43 #include <linux/module.h>
  44 #include <linux/slab.h>
  45 #include <linux/mutex.h>
  46 #include <linux/scatterlist.h>
  47 #include <linux/bitmap.h>
  48 #include <linux/list.h>
  49
  50 #include <xen/xen.h>
  51 #include <xen/xenbus.h>
  52 #include <xen/grant_table.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56
  57 #include <xen/interface/grant_table.h>
  58 #include <xen/interface/io/blkif.h>
  59 #include <xen/interface/io/protocols.h>
  60
  61 #include <asm/xen/hypervisor.h>
  62
  63 /*
  64  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  65  * When Linux is using a different page size than Xen, it may not be possible
  66  * to put all the data in a single segment.
  67  * This can happen when the backend doesn't support indirect descriptor and
  68  * therefore the maximum amount of data that a request can carry is
  69  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  70  *
  71  * Note that we only support one extra request. So the Linux page size
  72  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  73  * 88KB.
  74  */
  75 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  76
  77 enum blkif_state {
  78         BLKIF_STATE_DISCONNECTED,
  79         BLKIF_STATE_CONNECTED,
  80         BLKIF_STATE_SUSPENDED,
  81 };
  82
  83 struct grant {
  84         grant_ref_t gref;
  85         struct page *page;
  86         struct list_head node;
  87 };
  88
  89 enum blk_req_status {
  90         REQ_WAITING,
  91         REQ_DONE,
  92         REQ_ERROR,
  93         REQ_EOPNOTSUPP,
  94 };
  95
  96 struct blk_shadow {
  97         struct blkif_request req;
  98         struct request *request;
  99         struct grant **grants_used;
 100         struct grant **indirect_grants;
 101         struct scatterlist *sg;
 102         unsigned int num_sg;
 103         enum blk_req_status status;
 104
 105         #define NO_ASSOCIATED_ID ~0UL
 106         /*
 107          * Id of the sibling if we ever need 2 requests when handling a
 108          * block I/O request
 109          */
 110         unsigned long associated_id;
 111 };
 112
 113 struct split_bio {
 114         struct bio *bio;
 115         atomic_t pending;
 116 };
 117
 118 static DEFINE_MUTEX(blkfront_mutex);
 119 static const struct block_device_operations xlvbd_block_fops;
 120
 121 /*
 122  * Maximum number of segments in indirect requests, the actual value used by
 123  * the frontend driver is the minimum of this value and the value provided
 124  * by the backend driver.
 125  */
 126
 127 static unsigned int xen_blkif_max_segments = 32;
 128 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint,
 129                    S_IRUGO);
 130 MODULE_PARM_DESC(max_indirect_segments,
 131                  "Maximum amount of segments in indirect requests (default is 32)");
 132
 133 static unsigned int xen_blkif_max_queues = 4;
 134 module_param_named(max_queues, xen_blkif_max_queues, uint, S_IRUGO);
 135 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 136
 137 /*
 138  * Maximum order of pages to be used for the shared ring between front and
 139  * backend, 4KB page granularity is used.
 140  */
 141 static unsigned int xen_blkif_max_ring_order;
 142 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
 143 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 144
 145 #define BLK_RING_SIZE(info)     \
 146         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 147
 148 #define BLK_MAX_RING_SIZE       \
 149         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
 150
 151 /*
 152  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 153  * characters are enough. Define to 20 to keep consistent with backend.
 154  */
 155 #define RINGREF_NAME_LEN (20)
 156 /*
 157  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 158  */
 159 #define QUEUE_NAME_LEN (17)
 160
 161 /*
 162  *  Per-ring info.
 163  *  Every blkfront device can associate with one or more blkfront_ring_info,
 164  *  depending on how many hardware queues/rings to be used.
 165  */
 166 struct blkfront_ring_info {
 167         /* Lock to protect data in every ring buffer. */
 168         spinlock_t ring_lock;
 169         struct blkif_front_ring ring;
 170         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 171         unsigned int evtchn, irq;
 172         struct work_struct work;
 173         struct gnttab_free_callback callback;
 174         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 175         struct list_head indirect_pages;
 176         struct list_head grants;
 177         unsigned int persistent_gnts_c;
 178         unsigned long shadow_free;
 179         struct blkfront_info *dev_info;
 180 };
 181
 182 /*
 183  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 184  * hang in private_data off the gendisk structure. We may end up
 185  * putting all kinds of interesting stuff here :-)
 186  */
 187 struct blkfront_info
 188 {
 189         struct mutex mutex;
 190         struct xenbus_device *xbdev;
 191         struct gendisk *gd;
 192         int vdevice;
 193         blkif_vdev_t handle;
 194         enum blkif_state connected;
 195         /* Number of pages per ring buffer. */
 196         unsigned int nr_ring_pages;
 197         struct request_queue *rq;
 198         unsigned int feature_flush;
 199         unsigned int feature_fua;
 200         unsigned int feature_discard:1;
 201         unsigned int feature_secdiscard:1;
 202         unsigned int discard_granularity;
 203         unsigned int discard_alignment;
 204         unsigned int feature_persistent:1;
 205         /* Number of 4KB segments handled */
 206         unsigned int max_indirect_segments;
 207         int is_ready;
 208         struct blk_mq_tag_set tag_set;
 209         struct blkfront_ring_info *rinfo;
 210         unsigned int nr_rings;
 211 };
 212
 213 static unsigned int nr_minors;
 214 static unsigned long *minors;
 215 static DEFINE_SPINLOCK(minor_lock);
 216
 217 #define GRANT_INVALID_REF       0
 218
 219 #define PARTS_PER_DISK          16
 220 #define PARTS_PER_EXT_DISK      256
 221
 222 #define BLKIF_MAJOR(dev) ((dev)>>8)
 223 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 224
 225 #define EXT_SHIFT 28
 226 #define EXTENDED (1<<EXT_SHIFT)
 227 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 228 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 229 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 230 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 231 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 232 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 233
 234 #define DEV_NAME        "xvd"   /* name in /dev */
 235
 236 /*
 237  * Grants are always the same size as a Xen page (i.e 4KB).
 238  * A physical segment is always the same size as a Linux page.
 239  * Number of grants per physical segment
 240  */
 241 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 242
 243 #define GRANTS_PER_INDIRECT_FRAME \
 244         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 245
 246 #define PSEGS_PER_INDIRECT_FRAME        \
 247         (GRANTS_INDIRECT_FRAME / GRANTS_PSEGS)
 248
 249 #define INDIRECT_GREFS(_grants)         \
 250         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 251
 252 #define GREFS(_psegs)   ((_psegs) * GRANTS_PER_PSEG)
 253
 254 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 255 static void blkfront_gather_backend_features(struct blkfront_info *info);
 256
 257 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 258 {
 259         unsigned long free = rinfo->shadow_free;
 260
 261         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 262         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 263         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 264         return free;
 265 }
 266
 267 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 268                               unsigned long id)
 269 {
 270         if (rinfo->shadow[id].req.u.rw.id != id)
 271                 return -EINVAL;
 272         if (rinfo->shadow[id].request == NULL)
 273                 return -EINVAL;
 274         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 275         rinfo->shadow[id].request = NULL;
 276         rinfo->shadow_free = id;
 277         return 0;
 278 }
 279
 280 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 281 {
 282         struct blkfront_info *info = rinfo->dev_info;
 283         struct page *granted_page;
 284         struct grant *gnt_list_entry, *n;
 285         int i = 0;
 286
 287         while (i < num) {
 288                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 289                 if (!gnt_list_entry)
 290                         goto out_of_memory;
 291
 292                 if (info->feature_persistent) {
 293                         granted_page = alloc_page(GFP_NOIO);
 294                         if (!granted_page) {
 295                                 kfree(gnt_list_entry);
 296                                 goto out_of_memory;
 297                         }
 298                         gnt_list_entry->page = granted_page;
 299                 }
 300
 301                 gnt_list_entry->gref = GRANT_INVALID_REF;
 302                 list_add(&gnt_list_entry->node, &rinfo->grants);
 303                 i++;
 304         }
 305
 306         return 0;
 307
 308 out_of_memory:
 309         list_for_each_entry_safe(gnt_list_entry, n,
 310                                  &rinfo->grants, node) {
 311                 list_del(&gnt_list_entry->node);
 312                 if (info->feature_persistent)
 313                         __free_page(gnt_list_entry->page);
 314                 kfree(gnt_list_entry);
 315                 i--;
 316         }
 317         BUG_ON(i != 0);
 318         return -ENOMEM;
 319 }
 320
 321 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 322 {
 323         struct grant *gnt_list_entry;
 324
 325         BUG_ON(list_empty(&rinfo->grants));
 326         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 327                                           node);
 328         list_del(&gnt_list_entry->node);
 329
 330         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 331                 rinfo->persistent_gnts_c--;
 332
 333         return gnt_list_entry;
 334 }
 335
 336 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 337                                         const struct blkfront_info *info)
 338 {
 339         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 340                                                  info->xbdev->otherend_id,
 341                                                  gnt_list_entry->page,
 342                                                  0);
 343 }
 344
 345 static struct grant *get_grant(grant_ref_t *gref_head,
 346                                unsigned long gfn,
 347                                struct blkfront_ring_info *rinfo)
 348 {
 349         struct grant *gnt_list_entry = get_free_grant(rinfo);
 350         struct blkfront_info *info = rinfo->dev_info;
 351
 352         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 353                 return gnt_list_entry;
 354
 355         /* Assign a gref to this page */
 356         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 357         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 358         if (info->feature_persistent)
 359                 grant_foreign_access(gnt_list_entry, info);
 360         else {
 361                 /* Grant access to the GFN passed by the caller */
 362                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 363                                                 info->xbdev->otherend_id,
 364                                                 gfn, 0);
 365         }
 366
 367         return gnt_list_entry;
 368 }
 369
 370 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 371                                         struct blkfront_ring_info *rinfo)
 372 {
 373         struct grant *gnt_list_entry = get_free_grant(rinfo);
 374         struct blkfront_info *info = rinfo->dev_info;
 375
 376         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 377                 return gnt_list_entry;
 378
 379         /* Assign a gref to this page */
 380         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 381         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 382         if (!info->feature_persistent) {
 383                 struct page *indirect_page;
 384
 385                 /* Fetch a pre-allocated page to use for indirect grefs */
 386                 BUG_ON(list_empty(&rinfo->indirect_pages));
 387                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 388                                                  struct page, lru);
 389                 list_del(&indirect_page->lru);
 390                 gnt_list_entry->page = indirect_page;
 391         }
 392         grant_foreign_access(gnt_list_entry, info);
 393
 394         return gnt_list_entry;
 395 }
 396
 397 static const char *op_name(int op)
 398 {
 399         static const char *const names[] = {
 400                 [BLKIF_OP_READ] = "read",
 401                 [BLKIF_OP_WRITE] = "write",
 402                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 403                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 404                 [BLKIF_OP_DISCARD] = "discard" };
 405
 406         if (op < 0 || op >= ARRAY_SIZE(names))
 407                 return "unknown";
 408
 409         if (!names[op])
 410                 return "reserved";
 411
 412         return names[op];
 413 }
 414 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 415 {
 416         unsigned int end = minor + nr;
 417         int rc;
 418
 419         if (end > nr_minors) {
 420                 unsigned long *bitmap, *old;
 421
 422                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 423                                  GFP_KERNEL);
 424                 if (bitmap == NULL)
 425                         return -ENOMEM;
 426
 427                 spin_lock(&minor_lock);
 428                 if (end > nr_minors) {
 429                         old = minors;
 430                         memcpy(bitmap, minors,
 431                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 432                         minors = bitmap;
 433                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 434                 } else
 435                         old = bitmap;
 436                 spin_unlock(&minor_lock);
 437                 kfree(old);
 438         }
 439
 440         spin_lock(&minor_lock);
 441         if (find_next_bit(minors, end, minor) >= end) {
 442                 bitmap_set(minors, minor, nr);
 443                 rc = 0;
 444         } else
 445                 rc = -EBUSY;
 446         spin_unlock(&minor_lock);
 447
 448         return rc;
 449 }
 450
 451 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 452 {
 453         unsigned int end = minor + nr;
 454
 455         BUG_ON(end > nr_minors);
 456         spin_lock(&minor_lock);
 457         bitmap_clear(minors,  minor, nr);
 458         spin_unlock(&minor_lock);
 459 }
 460
 461 static void blkif_restart_queue_callback(void *arg)
 462 {
 463         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 464         schedule_work(&rinfo->work);
 465 }
 466
 467 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 468 {
 469         /* We don't have real geometry info, but let's at least return
 470            values consistent with the size of the device */
 471         sector_t nsect = get_capacity(bd->bd_disk);
 472         sector_t cylinders = nsect;
 473
 474         hg->heads = 0xff;
 475         hg->sectors = 0x3f;
 476         sector_div(cylinders, hg->heads * hg->sectors);
 477         hg->cylinders = cylinders;
 478         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 479                 hg->cylinders = 0xffff;
 480         return 0;
 481 }
 482
 483 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 484                        unsigned command, unsigned long argument)
 485 {
 486         struct blkfront_info *info = bdev->bd_disk->private_data;
 487         int i;
 488
 489         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 490                 command, (long)argument);
 491
 492         switch (command) {
 493         case CDROMMULTISESSION:
 494                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 495                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 496                         if (put_user(0, (char __user *)(argument + i)))
 497                                 return -EFAULT;
 498                 return 0;
 499
 500         case CDROM_GET_CAPABILITY: {
 501                 struct gendisk *gd = info->gd;
 502                 if (gd->flags & GENHD_FL_CD)
 503                         return 0;
 504                 return -EINVAL;
 505         }
 506
 507         default:
 508                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 509                   command);*/
 510                 return -EINVAL; /* same return as native Linux */
 511         }
 512
 513         return 0;
 514 }
 515
 516 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 517                                             struct request *req,
 518                                             struct blkif_request **ring_req)
 519 {
 520         unsigned long id;
 521
 522         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 523         rinfo->ring.req_prod_pvt++;
 524
 525         id = get_id_from_freelist(rinfo);
 526         rinfo->shadow[id].request = req;
 527         rinfo->shadow[id].status = REQ_WAITING;
 528         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 529
 530         (*ring_req)->u.rw.id = id;
 531
 532         return id;
 533 }
 534
 535 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 536 {
 537         struct blkfront_info *info = rinfo->dev_info;
 538         struct blkif_request *ring_req;
 539         unsigned long id;
 540
 541         /* Fill out a communications ring structure. */
 542         id = blkif_ring_get_request(rinfo, req, &ring_req);
 543
 544         ring_req->operation = BLKIF_OP_DISCARD;
 545         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 546         ring_req->u.discard.id = id;
 547         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 548         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 549                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 550         else
 551                 ring_req->u.discard.flag = 0;
 552
 553         /* Keep a private copy so we can reissue requests when recovering. */
 554         rinfo->shadow[id].req = *ring_req;
 555
 556         return 0;
 557 }
 558
 559 struct setup_rw_req {
 560         unsigned int grant_idx;
 561         struct blkif_request_segment *segments;
 562         struct blkfront_ring_info *rinfo;
 563         struct blkif_request *ring_req;
 564         grant_ref_t gref_head;
 565         unsigned int id;
 566         /* Only used when persistent grant is used and it's a read request */
 567         bool need_copy;
 568         unsigned int bvec_off;
 569         char *bvec_data;
 570
 571         bool require_extra_req;
 572         struct blkif_request *extra_ring_req;
 573 };
 574
 575 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 576                                      unsigned int len, void *data)
 577 {
 578         struct setup_rw_req *setup = data;
 579         int n, ref;
 580         struct grant *gnt_list_entry;
 581         unsigned int fsect, lsect;
 582         /* Convenient aliases */
 583         unsigned int grant_idx = setup->grant_idx;
 584         struct blkif_request *ring_req = setup->ring_req;
 585         struct blkfront_ring_info *rinfo = setup->rinfo;
 586         /*
 587          * We always use the shadow of the first request to store the list
 588          * of grant associated to the block I/O request. This made the
 589          * completion more easy to handle even if the block I/O request is
 590          * split.
 591          */
 592         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 593
 594         if (unlikely(setup->require_extra_req &&
 595                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 596                 /*
 597                  * We are using the second request, setup grant_idx
 598                  * to be the index of the segment array.
 599                  */
 600                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 601                 ring_req = setup->extra_ring_req;
 602         }
 603
 604         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 605             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 606                 if (setup->segments)
 607                         kunmap_atomic(setup->segments);
 608
 609                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 610                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 611                 shadow->indirect_grants[n] = gnt_list_entry;
 612                 setup->segments = kmap_atomic(gnt_list_entry->page);
 613                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 614         }
 615
 616         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 617         ref = gnt_list_entry->gref;
 618         /*
 619          * All the grants are stored in the shadow of the first
 620          * request. Therefore we have to use the global index.
 621          */
 622         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 623
 624         if (setup->need_copy) {
 625                 void *shared_data;
 626
 627                 shared_data = kmap_atomic(gnt_list_entry->page);
 628                 /*
 629                  * this does not wipe data stored outside the
 630                  * range sg->offset..sg->offset+sg->length.
 631                  * Therefore, blkback *could* see data from
 632                  * previous requests. This is OK as long as
 633                  * persistent grants are shared with just one
 634                  * domain. It may need refactoring if this
 635                  * changes
 636                  */
 637                 memcpy(shared_data + offset,
 638                        setup->bvec_data + setup->bvec_off,
 639                        len);
 640
 641                 kunmap_atomic(shared_data);
 642                 setup->bvec_off += len;
 643         }
 644
 645         fsect = offset >> 9;
 646         lsect = fsect + (len >> 9) - 1;
 647         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 648                 ring_req->u.rw.seg[grant_idx] =
 649                         (struct blkif_request_segment) {
 650                                 .gref       = ref,
 651                                 .first_sect = fsect,
 652                                 .last_sect  = lsect };
 653         } else {
 654                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 655                         (struct blkif_request_segment) {
 656                                 .gref       = ref,
 657                                 .first_sect = fsect,
 658                                 .last_sect  = lsect };
 659         }
 660
 661         (setup->grant_idx)++;
 662 }
 663
 664 static void blkif_setup_extra_req(struct blkif_request *first,
 665                                   struct blkif_request *second)
 666 {
 667         uint16_t nr_segments = first->u.rw.nr_segments;
 668
 669         /*
 670          * The second request is only present when the first request uses
 671          * all its segments. It's always the continuity of the first one.
 672          */
 673         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 674
 675         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 676         second->u.rw.sector_number = first->u.rw.sector_number +
 677                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 678
 679         second->u.rw.handle = first->u.rw.handle;
 680         second->operation = first->operation;
 681 }
 682
 683 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 684 {
 685         struct blkfront_info *info = rinfo->dev_info;
 686         struct blkif_request *ring_req, *extra_ring_req = NULL;
 687         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 688         bool require_extra_req = false;
 689         int i;
 690         struct setup_rw_req setup = {
 691                 .grant_idx = 0,
 692                 .segments = NULL,
 693                 .rinfo = rinfo,
 694                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 695         };
 696
 697         /*
 698          * Used to store if we are able to queue the request by just using
 699          * existing persistent grants, or if we have to get new grants,
 700          * as there are not sufficiently many free.
 701          */
 702         struct scatterlist *sg;
 703         int num_sg, max_grefs, num_grant;
 704
 705         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 706         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 707                 /*
 708                  * If we are using indirect segments we need to account
 709                  * for the indirect grefs used in the request.
 710                  */
 711                 max_grefs += INDIRECT_GREFS(max_grefs);
 712
 713         /*
 714          * We have to reserve 'max_grefs' grants because persistent
 715          * grants are shared by all rings.
 716          */
 717         if (max_grefs > 0)
 718                 if (gnttab_alloc_grant_references(max_grefs, &setup.gref_head) < 0) {
 719                         gnttab_request_free_callback(
 720                                 &rinfo->callback,
 721                                 blkif_restart_queue_callback,
 722                                 rinfo,
 723                                 max_grefs);
 724                         return 1;
 725                 }
 726
 727         /* Fill out a communications ring structure. */
 728         id = blkif_ring_get_request(rinfo, req, &ring_req);
 729
 730         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 731         num_grant = 0;
 732         /* Calculate the number of grant used */
 733         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 734                num_grant += gnttab_count_grant(sg->offset, sg->length);
 735
 736         require_extra_req = info->max_indirect_segments == 0 &&
 737                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 738         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 739
 740         rinfo->shadow[id].num_sg = num_sg;
 741         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 742             likely(!require_extra_req)) {
 743                 /*
 744                  * The indirect operation can only be a BLKIF_OP_READ or
 745                  * BLKIF_OP_WRITE
 746                  */
 747                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 748                 ring_req->operation = BLKIF_OP_INDIRECT;
 749                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 750                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 751                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 752                 ring_req->u.indirect.handle = info->handle;
 753                 ring_req->u.indirect.nr_segments = num_grant;
 754         } else {
 755                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 756                 ring_req->u.rw.handle = info->handle;
 757                 ring_req->operation = rq_data_dir(req) ?
 758                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 759                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 760                         /*
 761                          * Ideally we can do an unordered flush-to-disk.
 762                          * In case the backend onlysupports barriers, use that.
 763                          * A barrier request a superset of FUA, so we can
 764                          * implement it the same way.  (It's also a FLUSH+FUA,
 765                          * since it is guaranteed ordered WRT previous writes.)
 766                          */
 767                         if (info->feature_flush && info->feature_fua)
 768                                 ring_req->operation =
 769                                         BLKIF_OP_WRITE_BARRIER;
 770                         else if (info->feature_flush)
 771                                 ring_req->operation =
 772                                         BLKIF_OP_FLUSH_DISKCACHE;
 773                         else
 774                                 ring_req->operation = 0;
 775                 }
 776                 ring_req->u.rw.nr_segments = num_grant;
 777                 if (unlikely(require_extra_req)) {
 778                         extra_id = blkif_ring_get_request(rinfo, req,
 779                                                           &extra_ring_req);
 780                         /*
 781                          * Only the first request contains the scatter-gather
 782                          * list.
 783                          */
 784                         rinfo->shadow[extra_id].num_sg = 0;
 785
 786                         blkif_setup_extra_req(ring_req, extra_ring_req);
 787
 788                         /* Link the 2 requests together */
 789                         rinfo->shadow[extra_id].associated_id = id;
 790                         rinfo->shadow[id].associated_id = extra_id;
 791                 }
 792         }
 793
 794         setup.ring_req = ring_req;
 795         setup.id = id;
 796
 797         setup.require_extra_req = require_extra_req;
 798         if (unlikely(require_extra_req))
 799                 setup.extra_ring_req = extra_ring_req;
 800
 801         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 802                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 803
 804                 if (setup.need_copy) {
 805                         setup.bvec_off = sg->offset;
 806                         setup.bvec_data = kmap_atomic(sg_page(sg));
 807                 }
 808
 809                 gnttab_foreach_grant_in_range(sg_page(sg),
 810                                               sg->offset,
 811                                               sg->length,
 812                                               blkif_setup_rw_req_grant,
 813                                               &setup);
 814
 815                 if (setup.need_copy)
 816                         kunmap_atomic(setup.bvec_data);
 817         }
 818         if (setup.segments)
 819                 kunmap_atomic(setup.segments);
 820
 821         /* Keep a private copy so we can reissue requests when recovering. */
 822         rinfo->shadow[id].req = *ring_req;
 823         if (unlikely(require_extra_req))
 824                 rinfo->shadow[extra_id].req = *extra_ring_req;
 825
 826         if (max_grefs > 0)
 827                 gnttab_free_grant_references(setup.gref_head);
 828
 829         return 0;
 830 }
 831
 832 /*
 833  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 834  * and writes are handled as expected.
 835  *
 836  * @req: a request struct
 837  */
 838 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 839 {
 840         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 841                 return 1;
 842
 843         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 844                      req_op(req) == REQ_OP_SECURE_ERASE))
 845                 return blkif_queue_discard_req(req, rinfo);
 846         else
 847                 return blkif_queue_rw_req(req, rinfo);
 848 }
 849
 850 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 851 {
 852         int notify;
 853
 854         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 855
 856         if (notify)
 857                 notify_remote_via_irq(rinfo->irq);
 858 }
 859
 860 static inline bool blkif_request_flush_invalid(struct request *req,
 861                                                struct blkfront_info *info)
 862 {
 863         return ((req->cmd_type != REQ_TYPE_FS) ||
 864                 ((req_op(req) == REQ_OP_FLUSH) &&
 865                  !info->feature_flush) ||
 866                 ((req->cmd_flags & REQ_FUA) &&
 867                  !info->feature_fua));
 868 }
 869
 870 static int blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 871                           const struct blk_mq_queue_data *qd)
 872 {
 873         unsigned long flags;
 874         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)hctx->driver_data;
 875
 876         blk_mq_start_request(qd->rq);
 877         spin_lock_irqsave(&rinfo->ring_lock, flags);
 878         if (RING_FULL(&rinfo->ring))
 879                 goto out_busy;
 880
 881         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 882                 goto out_err;
 883
 884         if (blkif_queue_request(qd->rq, rinfo))
 885                 goto out_busy;
 886
 887         flush_requests(rinfo);
 888         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 889         return BLK_MQ_RQ_QUEUE_OK;
 890
 891 out_err:
 892         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 893         return BLK_MQ_RQ_QUEUE_ERROR;
 894
 895 out_busy:
 896         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 897         blk_mq_stop_hw_queue(hctx);
 898         return BLK_MQ_RQ_QUEUE_BUSY;
 899 }
 900
 901 static int blk_mq_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
 902                             unsigned int index)
 903 {
 904         struct blkfront_info *info = (struct blkfront_info *)data;
 905
 906         BUG_ON(info->nr_rings <= index);
 907         hctx->driver_data = &info->rinfo[index];
 908         return 0;
 909 }
 910
 911 static struct blk_mq_ops blkfront_mq_ops = {
 912         .queue_rq = blkif_queue_rq,
 913         .map_queue = blk_mq_map_queue,
 914         .init_hctx = blk_mq_init_hctx,
 915 };
 916
 917 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 918                                 unsigned int physical_sector_size,
 919                                 unsigned int segments)
 920 {
 921         struct request_queue *rq;
 922         struct blkfront_info *info = gd->private_data;
 923
 924         memset(&info->tag_set, 0, sizeof(info->tag_set));
 925         info->tag_set.ops = &blkfront_mq_ops;
 926         info->tag_set.nr_hw_queues = info->nr_rings;
 927         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 928                 /*
 929                  * When indirect descriptior is not supported, the I/O request
 930                  * will be split between multiple request in the ring.
 931                  * To avoid problems when sending the request, divide by
 932                  * 2 the depth of the queue.
 933                  */
 934                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 935         } else
 936                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 937         info->tag_set.numa_node = NUMA_NO_NODE;
 938         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 939         info->tag_set.cmd_size = 0;
 940         info->tag_set.driver_data = info;
 941
 942         if (blk_mq_alloc_tag_set(&info->tag_set))
 943                 return -EINVAL;
 944         rq = blk_mq_init_queue(&info->tag_set);
 945         if (IS_ERR(rq)) {
 946                 blk_mq_free_tag_set(&info->tag_set);
 947                 return PTR_ERR(rq);
 948         }
 949
 950         queue_flag_set_unlocked(QUEUE_FLAG_VIRT, rq);
 951
 952         if (info->feature_discard) {
 953                 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, rq);
 954                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 955                 rq->limits.discard_granularity = info->discard_granularity;
 956                 rq->limits.discard_alignment = info->discard_alignment;
 957                 if (info->feature_secdiscard)
 958                         queue_flag_set_unlocked(QUEUE_FLAG_SECERASE, rq);
 959         }
 960
 961         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 962         blk_queue_logical_block_size(rq, sector_size);
 963         blk_queue_physical_block_size(rq, physical_sector_size);
 964         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 965
 966         /* Each segment in a request is up to an aligned page in size. */
 967         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 968         blk_queue_max_segment_size(rq, PAGE_SIZE);
 969
 970         /* Ensure a merged request will fit in a single I/O ring slot. */
 971         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 972
 973         /* Make sure buffer addresses are sector-aligned. */
 974         blk_queue_dma_alignment(rq, 511);
 975
 976         /* Make sure we don't use bounce buffers. */
 977         blk_queue_bounce_limit(rq, BLK_BOUNCE_ANY);
 978
 979         gd->queue = rq;
 980
 981         return 0;
 982 }
 983
 984 static const char *flush_info(struct blkfront_info *info)
 985 {
 986         if (info->feature_flush && info->feature_fua)
 987                 return "barrier: enabled;";
 988         else if (info->feature_flush)
 989                 return "flush diskcache: enabled;";
 990         else
 991                 return "barrier or flush: disabled;";
 992 }
 993
 994 static void xlvbd_flush(struct blkfront_info *info)
 995 {
 996         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 997                               info->feature_fua ? true : false);
 998         pr_info("blkfront: %s: %s %s %s %s %s\n",
 999                 info->gd->disk_name, flush_info(info),
1000                 "persistent grants:", info->feature_persistent ?
1001                 "enabled;" : "disabled;", "indirect descriptors:",
1002                 info->max_indirect_segments ? "enabled;" : "disabled;");
1003 }
1004
1005 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1006 {
1007         int major;
1008         major = BLKIF_MAJOR(vdevice);
1009         *minor = BLKIF_MINOR(vdevice);
1010         switch (major) {
1011                 case XEN_IDE0_MAJOR:
1012                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1013                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1014                                 EMULATED_HD_DISK_MINOR_OFFSET;
1015                         break;
1016                 case XEN_IDE1_MAJOR:
1017                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1018                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1019                                 EMULATED_HD_DISK_MINOR_OFFSET;
1020                         break;
1021                 case XEN_SCSI_DISK0_MAJOR:
1022                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1023                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1024                         break;
1025                 case XEN_SCSI_DISK1_MAJOR:
1026                 case XEN_SCSI_DISK2_MAJOR:
1027                 case XEN_SCSI_DISK3_MAJOR:
1028                 case XEN_SCSI_DISK4_MAJOR:
1029                 case XEN_SCSI_DISK5_MAJOR:
1030                 case XEN_SCSI_DISK6_MAJOR:
1031                 case XEN_SCSI_DISK7_MAJOR:
1032                         *offset = (*minor / PARTS_PER_DISK) +
1033                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1034                                 EMULATED_SD_DISK_NAME_OFFSET;
1035                         *minor = *minor +
1036                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1037                                 EMULATED_SD_DISK_MINOR_OFFSET;
1038                         break;
1039                 case XEN_SCSI_DISK8_MAJOR:
1040                 case XEN_SCSI_DISK9_MAJOR:
1041                 case XEN_SCSI_DISK10_MAJOR:
1042                 case XEN_SCSI_DISK11_MAJOR:
1043                 case XEN_SCSI_DISK12_MAJOR:
1044                 case XEN_SCSI_DISK13_MAJOR:
1045                 case XEN_SCSI_DISK14_MAJOR:
1046                 case XEN_SCSI_DISK15_MAJOR:
1047                         *offset = (*minor / PARTS_PER_DISK) +
1048                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1049                                 EMULATED_SD_DISK_NAME_OFFSET;
1050                         *minor = *minor +
1051                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1052                                 EMULATED_SD_DISK_MINOR_OFFSET;
1053                         break;
1054                 case XENVBD_MAJOR:
1055                         *offset = *minor / PARTS_PER_DISK;
1056                         break;
1057                 default:
1058                         printk(KERN_WARNING "blkfront: your disk configuration is "
1059                                         "incorrect, please use an xvd device instead\n");
1060                         return -ENODEV;
1061         }
1062         return 0;
1063 }
1064
1065 static char *encode_disk_name(char *ptr, unsigned int n)
1066 {
1067         if (n >= 26)
1068                 ptr = encode_disk_name(ptr, n / 26 - 1);
1069         *ptr = 'a' + n % 26;
1070         return ptr + 1;
1071 }
1072
1073 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1074                                struct blkfront_info *info,
1075                                u16 vdisk_info, u16 sector_size,
1076                                unsigned int physical_sector_size)
1077 {
1078         struct gendisk *gd;
1079         int nr_minors = 1;
1080         int err;
1081         unsigned int offset;
1082         int minor;
1083         int nr_parts;
1084         char *ptr;
1085
1086         BUG_ON(info->gd != NULL);
1087         BUG_ON(info->rq != NULL);
1088
1089         if ((info->vdevice>>EXT_SHIFT) > 1) {
1090                 /* this is above the extended range; something is wrong */
1091                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1092                 return -ENODEV;
1093         }
1094
1095         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1096                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1097                 if (err)
1098                         return err;
1099                 nr_parts = PARTS_PER_DISK;
1100         } else {
1101                 minor = BLKIF_MINOR_EXT(info->vdevice);
1102                 nr_parts = PARTS_PER_EXT_DISK;
1103                 offset = minor / nr_parts;
1104                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1105                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1106                                         "emulated IDE disks,\n\t choose an xvd device name"
1107                                         "from xvde on\n", info->vdevice);
1108         }
1109         if (minor >> MINORBITS) {
1110                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1111                         info->vdevice, minor);
1112                 return -ENODEV;
1113         }
1114
1115         if ((minor % nr_parts) == 0)
1116                 nr_minors = nr_parts;
1117
1118         err = xlbd_reserve_minors(minor, nr_minors);
1119         if (err)
1120                 goto out;
1121         err = -ENODEV;
1122
1123         gd = alloc_disk(nr_minors);
1124         if (gd == NULL)
1125                 goto release;
1126
1127         strcpy(gd->disk_name, DEV_NAME);
1128         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1129         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1130         if (nr_minors > 1)
1131                 *ptr = 0;
1132         else
1133                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1134                          "%d", minor & (nr_parts - 1));
1135
1136         gd->major = XENVBD_MAJOR;
1137         gd->first_minor = minor;
1138         gd->fops = &xlvbd_block_fops;
1139         gd->private_data = info;
1140         gd->driverfs_dev = &(info->xbdev->dev);
1141         set_capacity(gd, capacity);
1142
1143         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size,
1144                                  info->max_indirect_segments ? :
1145                                  BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
1146                 del_gendisk(gd);
1147                 goto release;
1148         }
1149
1150         info->rq = gd->queue;
1151         info->gd = gd;
1152
1153         xlvbd_flush(info);
1154
1155         if (vdisk_info & VDISK_READONLY)
1156                 set_disk_ro(gd, 1);
1157
1158         if (vdisk_info & VDISK_REMOVABLE)
1159                 gd->flags |= GENHD_FL_REMOVABLE;
1160
1161         if (vdisk_info & VDISK_CDROM)
1162                 gd->flags |= GENHD_FL_CD;
1163
1164         return 0;
1165
1166  release:
1167         xlbd_release_minors(minor, nr_minors);
1168  out:
1169         return err;
1170 }
1171
1172 static void xlvbd_release_gendisk(struct blkfront_info *info)
1173 {
1174         unsigned int minor, nr_minors, i;
1175
1176         if (info->rq == NULL)
1177                 return;
1178
1179         /* No more blkif_request(). */
1180         blk_mq_stop_hw_queues(info->rq);
1181
1182         for (i = 0; i < info->nr_rings; i++) {
1183                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1184
1185                 /* No more gnttab callback work. */
1186                 gnttab_cancel_free_callback(&rinfo->callback);
1187
1188                 /* Flush gnttab callback work. Must be done with no locks held. */
1189                 flush_work(&rinfo->work);
1190         }
1191
1192         del_gendisk(info->gd);
1193
1194         minor = info->gd->first_minor;
1195         nr_minors = info->gd->minors;
1196         xlbd_release_minors(minor, nr_minors);
1197
1198         blk_cleanup_queue(info->rq);
1199         blk_mq_free_tag_set(&info->tag_set);
1200         info->rq = NULL;
1201
1202         put_disk(info->gd);
1203         info->gd = NULL;
1204 }
1205
1206 /* Already hold rinfo->ring_lock. */
1207 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1208 {
1209         if (!RING_FULL(&rinfo->ring))
1210                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1211 }
1212
1213 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1214 {
1215         unsigned long flags;
1216
1217         spin_lock_irqsave(&rinfo->ring_lock, flags);
1218         kick_pending_request_queues_locked(rinfo);
1219         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1220 }
1221
1222 static void blkif_restart_queue(struct work_struct *work)
1223 {
1224         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1225
1226         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1227                 kick_pending_request_queues(rinfo);
1228 }
1229
1230 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1231 {
1232         struct grant *persistent_gnt, *n;
1233         struct blkfront_info *info = rinfo->dev_info;
1234         int i, j, segs;
1235
1236         /*
1237          * Remove indirect pages, this only happens when using indirect
1238          * descriptors but not persistent grants
1239          */
1240         if (!list_empty(&rinfo->indirect_pages)) {
1241                 struct page *indirect_page, *n;
1242
1243                 BUG_ON(info->feature_persistent);
1244                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1245                         list_del(&indirect_page->lru);
1246                         __free_page(indirect_page);
1247                 }
1248         }
1249
1250         /* Remove all persistent grants. */
1251         if (!list_empty(&rinfo->grants)) {
1252                 list_for_each_entry_safe(persistent_gnt, n,
1253                                          &rinfo->grants, node) {
1254                         list_del(&persistent_gnt->node);
1255                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1256                                 gnttab_end_foreign_access(persistent_gnt->gref,
1257                                                           0, 0UL);
1258                                 rinfo->persistent_gnts_c--;
1259                         }
1260                         if (info->feature_persistent)
1261                                 __free_page(persistent_gnt->page);
1262                         kfree(persistent_gnt);
1263                 }
1264         }
1265         BUG_ON(rinfo->persistent_gnts_c != 0);
1266
1267         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1268                 /*
1269                  * Clear persistent grants present in requests already
1270                  * on the shared ring
1271                  */
1272                 if (!rinfo->shadow[i].request)
1273                         goto free_shadow;
1274
1275                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1276                        rinfo->shadow[i].req.u.indirect.nr_segments :
1277                        rinfo->shadow[i].req.u.rw.nr_segments;
1278                 for (j = 0; j < segs; j++) {
1279                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1280                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1281                         if (info->feature_persistent)
1282                                 __free_page(persistent_gnt->page);
1283                         kfree(persistent_gnt);
1284                 }
1285
1286                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1287                         /*
1288                          * If this is not an indirect operation don't try to
1289                          * free indirect segments
1290                          */
1291                         goto free_shadow;
1292
1293                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1294                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1295                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1296                         __free_page(persistent_gnt->page);
1297                         kfree(persistent_gnt);
1298                 }
1299
1300 free_shadow:
1301                 kfree(rinfo->shadow[i].grants_used);
1302                 rinfo->shadow[i].grants_used = NULL;
1303                 kfree(rinfo->shadow[i].indirect_grants);
1304                 rinfo->shadow[i].indirect_grants = NULL;
1305                 kfree(rinfo->shadow[i].sg);
1306                 rinfo->shadow[i].sg = NULL;
1307         }
1308
1309         /* No more gnttab callback work. */
1310         gnttab_cancel_free_callback(&rinfo->callback);
1311
1312         /* Flush gnttab callback work. Must be done with no locks held. */
1313         flush_work(&rinfo->work);
1314
1315         /* Free resources associated with old device channel. */
1316         for (i = 0; i < info->nr_ring_pages; i++) {
1317                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1318                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1319                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1320                 }
1321         }
1322         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
1323         rinfo->ring.sring = NULL;
1324
1325         if (rinfo->irq)
1326                 unbind_from_irqhandler(rinfo->irq, rinfo);
1327         rinfo->evtchn = rinfo->irq = 0;
1328 }
1329
1330 static void blkif_free(struct blkfront_info *info, int suspend)
1331 {
1332         unsigned int i;
1333
1334         /* Prevent new requests being issued until we fix things up. */
1335         info->connected = suspend ?
1336                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1337         /* No more blkif_request(). */
1338         if (info->rq)
1339                 blk_mq_stop_hw_queues(info->rq);
1340
1341         for (i = 0; i < info->nr_rings; i++)
1342                 blkif_free_ring(&info->rinfo[i]);
1343
1344         kfree(info->rinfo);
1345         info->rinfo = NULL;
1346         info->nr_rings = 0;
1347 }
1348
1349 struct copy_from_grant {
1350         const struct blk_shadow *s;
1351         unsigned int grant_idx;
1352         unsigned int bvec_offset;
1353         char *bvec_data;
1354 };
1355
1356 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1357                                   unsigned int len, void *data)
1358 {
1359         struct copy_from_grant *info = data;
1360         char *shared_data;
1361         /* Convenient aliases */
1362         const struct blk_shadow *s = info->s;
1363
1364         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1365
1366         memcpy(info->bvec_data + info->bvec_offset,
1367                shared_data + offset, len);
1368
1369         info->bvec_offset += len;
1370         info->grant_idx++;
1371
1372         kunmap_atomic(shared_data);
1373 }
1374
1375 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1376 {
1377         switch (rsp)
1378         {
1379         case BLKIF_RSP_OKAY:
1380                 return REQ_DONE;
1381         case BLKIF_RSP_EOPNOTSUPP:
1382                 return REQ_EOPNOTSUPP;
1383         case BLKIF_RSP_ERROR:
1384                 /* Fallthrough. */
1385         default:
1386                 return REQ_ERROR;
1387         }
1388 }
1389
1390 /*
1391  * Get the final status of the block request based on two ring response
1392  */
1393 static int blkif_get_final_status(enum blk_req_status s1,
1394                                   enum blk_req_status s2)
1395 {
1396         BUG_ON(s1 == REQ_WAITING);
1397         BUG_ON(s2 == REQ_WAITING);
1398
1399         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1400                 return BLKIF_RSP_ERROR;
1401         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1402                 return BLKIF_RSP_EOPNOTSUPP;
1403         return BLKIF_RSP_OKAY;
1404 }
1405
1406 static bool blkif_completion(unsigned long *id,
1407                              struct blkfront_ring_info *rinfo,
1408                              struct blkif_response *bret)
1409 {
1410         int i = 0;
1411         struct scatterlist *sg;
1412         int num_sg, num_grant;
1413         struct blkfront_info *info = rinfo->dev_info;
1414         struct blk_shadow *s = &rinfo->shadow[*id];
1415         struct copy_from_grant data = {
1416                 .grant_idx = 0,
1417         };
1418
1419         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1420                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1421
1422         /* The I/O request may be split in two. */
1423         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1424                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1425
1426                 /* Keep the status of the current response in shadow. */
1427                 s->status = blkif_rsp_to_req_status(bret->status);
1428
1429                 /* Wait the second response if not yet here. */
1430                 if (s2->status == REQ_WAITING)
1431                         return 0;
1432
1433                 bret->status = blkif_get_final_status(s->status,
1434                                                       s2->status);
1435
1436                 /*
1437                  * All the grants is stored in the first shadow in order
1438                  * to make the completion code simpler.
1439                  */
1440                 num_grant += s2->req.u.rw.nr_segments;
1441
1442                 /*
1443                  * The two responses may not come in order. Only the
1444                  * first request will store the scatter-gather list.
1445                  */
1446                 if (s2->num_sg != 0) {
1447                         /* Update "id" with the ID of the first response. */
1448                         *id = s->associated_id;
1449                         s = s2;
1450                 }
1451
1452                 /*
1453                  * We don't need anymore the second request, so recycling
1454                  * it now.
1455                  */
1456                 if (add_id_to_freelist(rinfo, s->associated_id))
1457                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1458                              info->gd->disk_name, s->associated_id);
1459         }
1460
1461         data.s = s;
1462         num_sg = s->num_sg;
1463
1464         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1465                 for_each_sg(s->sg, sg, num_sg, i) {
1466                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1467
1468                         data.bvec_offset = sg->offset;
1469                         data.bvec_data = kmap_atomic(sg_page(sg));
1470
1471                         gnttab_foreach_grant_in_range(sg_page(sg),
1472                                                       sg->offset,
1473                                                       sg->length,
1474                                                       blkif_copy_from_grant,
1475                                                       &data);
1476
1477                         kunmap_atomic(data.bvec_data);
1478                 }
1479         }
1480         /* Add the persistent grant into the list of free grants */
1481         for (i = 0; i < num_grant; i++) {
1482                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1483                         /*
1484                          * If the grant is still mapped by the backend (the
1485                          * backend has chosen to make this grant persistent)
1486                          * we add it at the head of the list, so it will be
1487                          * reused first.
1488                          */
1489                         if (!info->feature_persistent)
1490                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1491                                                      s->grants_used[i]->gref);
1492                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1493                         rinfo->persistent_gnts_c++;
1494                 } else {
1495                         /*
1496                          * If the grant is not mapped by the backend we end the
1497                          * foreign access and add it to the tail of the list,
1498                          * so it will not be picked again unless we run out of
1499                          * persistent grants.
1500                          */
1501                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1502                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1503                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1504                 }
1505         }
1506         if (s->req.operation == BLKIF_OP_INDIRECT) {
1507                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1508                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1509                                 if (!info->feature_persistent)
1510                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1511                                                              s->indirect_grants[i]->gref);
1512                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1513                                 rinfo->persistent_gnts_c++;
1514                         } else {
1515                                 struct page *indirect_page;
1516
1517                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1518                                 /*
1519                                  * Add the used indirect page back to the list of
1520                                  * available pages for indirect grefs.
1521                                  */
1522                                 if (!info->feature_persistent) {
1523                                         indirect_page = s->indirect_grants[i]->page;
1524                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1525                                 }
1526                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1527                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1528                         }
1529                 }
1530         }
1531
1532         return 1;
1533 }
1534
1535 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1536 {
1537         struct request *req;
1538         struct blkif_response *bret;
1539         RING_IDX i, rp;
1540         unsigned long flags;
1541         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1542         struct blkfront_info *info = rinfo->dev_info;
1543         int error;
1544
1545         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1546                 return IRQ_HANDLED;
1547
1548         spin_lock_irqsave(&rinfo->ring_lock, flags);
1549  again:
1550         rp = rinfo->ring.sring->rsp_prod;
1551         rmb(); /* Ensure we see queued responses up to 'rp'. */
1552
1553         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1554                 unsigned long id;
1555
1556                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1557                 id   = bret->id;
1558                 /*
1559                  * The backend has messed up and given us an id that we would
1560                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1561                  * look in get_id_from_freelist.
1562                  */
1563                 if (id >= BLK_RING_SIZE(info)) {
1564                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1565                              info->gd->disk_name, op_name(bret->operation), id);
1566                         /* We can't safely get the 'struct request' as
1567                          * the id is busted. */
1568                         continue;
1569                 }
1570                 req  = rinfo->shadow[id].request;
1571
1572                 if (bret->operation != BLKIF_OP_DISCARD) {
1573                         /*
1574                          * We may need to wait for an extra response if the
1575                          * I/O request is split in 2
1576                          */
1577                         if (!blkif_completion(&id, rinfo, bret))
1578                                 continue;
1579                 }
1580
1581                 if (add_id_to_freelist(rinfo, id)) {
1582                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1583                              info->gd->disk_name, op_name(bret->operation), id);
1584                         continue;
1585                 }
1586
1587                 error = (bret->status == BLKIF_RSP_OKAY) ? 0 : -EIO;
1588                 switch (bret->operation) {
1589                 case BLKIF_OP_DISCARD:
1590                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1591                                 struct request_queue *rq = info->rq;
1592                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1593                                            info->gd->disk_name, op_name(bret->operation));
1594                                 error = -EOPNOTSUPP;
1595                                 info->feature_discard = 0;
1596                                 info->feature_secdiscard = 0;
1597                                 queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1598                                 queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1599                         }
1600                         blk_mq_complete_request(req, error);
1601                         break;
1602                 case BLKIF_OP_FLUSH_DISKCACHE:
1603                 case BLKIF_OP_WRITE_BARRIER:
1604                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1605                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1606                                        info->gd->disk_name, op_name(bret->operation));
1607                                 error = -EOPNOTSUPP;
1608                         }
1609                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1610                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1611                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1612                                        info->gd->disk_name, op_name(bret->operation));
1613                                 error = -EOPNOTSUPP;
1614                         }
1615                         if (unlikely(error)) {
1616                                 if (error == -EOPNOTSUPP)
1617                                         error = 0;
1618                                 info->feature_fua = 0;
1619                                 info->feature_flush = 0;
1620                                 xlvbd_flush(info);
1621                         }
1622                         /* fall through */
1623                 case BLKIF_OP_READ:
1624                 case BLKIF_OP_WRITE:
1625                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1626                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1627                                         "request: %x\n", bret->status);
1628
1629                         blk_mq_complete_request(req, error);
1630                         break;
1631                 default:
1632                         BUG();
1633                 }
1634         }
1635
1636         rinfo->ring.rsp_cons = i;
1637
1638         if (i != rinfo->ring.req_prod_pvt) {
1639                 int more_to_do;
1640                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1641                 if (more_to_do)
1642                         goto again;
1643         } else
1644                 rinfo->ring.sring->rsp_event = i + 1;
1645
1646         kick_pending_request_queues_locked(rinfo);
1647
1648         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1649
1650         return IRQ_HANDLED;
1651 }
1652
1653
1654 static int setup_blkring(struct xenbus_device *dev,
1655                          struct blkfront_ring_info *rinfo)
1656 {
1657         struct blkif_sring *sring;
1658         int err, i;
1659         struct blkfront_info *info = rinfo->dev_info;
1660         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1661         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1662
1663         for (i = 0; i < info->nr_ring_pages; i++)
1664                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1665
1666         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1667                                                        get_order(ring_size));
1668         if (!sring) {
1669                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1670                 return -ENOMEM;
1671         }
1672         SHARED_RING_INIT(sring);
1673         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1674
1675         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1676         if (err < 0) {
1677                 free_pages((unsigned long)sring, get_order(ring_size));
1678                 rinfo->ring.sring = NULL;
1679                 goto fail;
1680         }
1681         for (i = 0; i < info->nr_ring_pages; i++)
1682                 rinfo->ring_ref[i] = gref[i];
1683
1684         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1685         if (err)
1686                 goto fail;
1687
1688         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1689                                         "blkif", rinfo);
1690         if (err <= 0) {
1691                 xenbus_dev_fatal(dev, err,
1692                                  "bind_evtchn_to_irqhandler failed");
1693                 goto fail;
1694         }
1695         rinfo->irq = err;
1696
1697         return 0;
1698 fail:
1699         blkif_free(info, 0);
1700         return err;
1701 }
1702
1703 /*
1704  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1705  * ring buffer may have multi pages depending on ->nr_ring_pages.
1706  */
1707 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1708                                 struct blkfront_ring_info *rinfo, const char *dir)
1709 {
1710         int err;
1711         unsigned int i;
1712         const char *message = NULL;
1713         struct blkfront_info *info = rinfo->dev_info;
1714
1715         if (info->nr_ring_pages == 1) {
1716                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1717                 if (err) {
1718                         message = "writing ring-ref";
1719                         goto abort_transaction;
1720                 }
1721         } else {
1722                 for (i = 0; i < info->nr_ring_pages; i++) {
1723                         char ring_ref_name[RINGREF_NAME_LEN];
1724
1725                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1726                         err = xenbus_printf(xbt, dir, ring_ref_name,
1727                                             "%u", rinfo->ring_ref[i]);
1728                         if (err) {
1729                                 message = "writing ring-ref";
1730                                 goto abort_transaction;
1731                         }
1732                 }
1733         }
1734
1735         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1736         if (err) {
1737                 message = "writing event-channel";
1738                 goto abort_transaction;
1739         }
1740
1741         return 0;
1742
1743 abort_transaction:
1744         xenbus_transaction_end(xbt, 1);
1745         if (message)
1746                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1747
1748         return err;
1749 }
1750
1751 /* Common code used when first setting up, and when resuming. */
1752 static int talk_to_blkback(struct xenbus_device *dev,
1753                            struct blkfront_info *info)
1754 {
1755         const char *message = NULL;
1756         struct xenbus_transaction xbt;
1757         int err;
1758         unsigned int i, max_page_order = 0;
1759         unsigned int ring_page_order = 0;
1760
1761         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1762                            "max-ring-page-order", "%u", &max_page_order);
1763         if (err != 1)
1764                 info->nr_ring_pages = 1;
1765         else {
1766                 ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1767                 info->nr_ring_pages = 1 << ring_page_order;
1768         }
1769
1770         for (i = 0; i < info->nr_rings; i++) {
1771                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1772
1773                 /* Create shared ring, alloc event channel. */
1774                 err = setup_blkring(dev, rinfo);
1775                 if (err)
1776                         goto destroy_blkring;
1777         }
1778
1779 again:
1780         err = xenbus_transaction_start(&xbt);
1781         if (err) {
1782                 xenbus_dev_fatal(dev, err, "starting transaction");
1783                 goto destroy_blkring;
1784         }
1785
1786         if (info->nr_ring_pages > 1) {
1787                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1788                                     ring_page_order);
1789                 if (err) {
1790                         message = "writing ring-page-order";
1791                         goto abort_transaction;
1792                 }
1793         }
1794
1795         /* We already got the number of queues/rings in _probe */
1796         if (info->nr_rings == 1) {
1797                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1798                 if (err)
1799                         goto destroy_blkring;
1800         } else {
1801                 char *path;
1802                 size_t pathsize;
1803
1804                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1805                                     info->nr_rings);
1806                 if (err) {
1807                         message = "writing multi-queue-num-queues";
1808                         goto abort_transaction;
1809                 }
1810
1811                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1812                 path = kmalloc(pathsize, GFP_KERNEL);
1813                 if (!path) {
1814                         err = -ENOMEM;
1815                         message = "ENOMEM while writing ring references";
1816                         goto abort_transaction;
1817                 }
1818
1819                 for (i = 0; i < info->nr_rings; i++) {
1820                         memset(path, 0, pathsize);
1821                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1822                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1823                         if (err) {
1824                                 kfree(path);
1825                                 goto destroy_blkring;
1826                         }
1827                 }
1828                 kfree(path);
1829         }
1830         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1831                             XEN_IO_PROTO_ABI_NATIVE);
1832         if (err) {
1833                 message = "writing protocol";
1834                 goto abort_transaction;
1835         }
1836         err = xenbus_printf(xbt, dev->nodename,
1837                             "feature-persistent", "%u", 1);
1838         if (err)
1839                 dev_warn(&dev->dev,
1840                          "writing persistent grants feature to xenbus");
1841
1842         err = xenbus_transaction_end(xbt, 0);
1843         if (err) {
1844                 if (err == -EAGAIN)
1845                         goto again;
1846                 xenbus_dev_fatal(dev, err, "completing transaction");
1847                 goto destroy_blkring;
1848         }
1849
1850         for (i = 0; i < info->nr_rings; i++) {
1851                 unsigned int j;
1852                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1853
1854                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1855                         rinfo->shadow[j].req.u.rw.id = j + 1;
1856                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1857         }
1858         xenbus_switch_state(dev, XenbusStateInitialised);
1859
1860         return 0;
1861
1862  abort_transaction:
1863         xenbus_transaction_end(xbt, 1);
1864         if (message)
1865                 xenbus_dev_fatal(dev, err, "%s", message);
1866  destroy_blkring:
1867         blkif_free(info, 0);
1868
1869         kfree(info);
1870         dev_set_drvdata(&dev->dev, NULL);
1871
1872         return err;
1873 }
1874
1875 static int negotiate_mq(struct blkfront_info *info)
1876 {
1877         unsigned int backend_max_queues = 0;
1878         int err;
1879         unsigned int i;
1880
1881         BUG_ON(info->nr_rings);
1882
1883         /* Check if backend supports multiple queues. */
1884         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1885                            "multi-queue-max-queues", "%u", &backend_max_queues);
1886         if (err < 0)
1887                 backend_max_queues = 1;
1888
1889         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1890         /* We need at least one ring. */
1891         if (!info->nr_rings)
1892                 info->nr_rings = 1;
1893
1894         info->rinfo = kzalloc(sizeof(struct blkfront_ring_info) * info->nr_rings, GFP_KERNEL);
1895         if (!info->rinfo) {
1896                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1897                 return -ENOMEM;
1898         }
1899
1900         for (i = 0; i < info->nr_rings; i++) {
1901                 struct blkfront_ring_info *rinfo;
1902
1903                 rinfo = &info->rinfo[i];
1904                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1905                 INIT_LIST_HEAD(&rinfo->grants);
1906                 rinfo->dev_info = info;
1907                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1908                 spin_lock_init(&rinfo->ring_lock);
1909         }
1910         return 0;
1911 }
1912 /**
1913  * Entry point to this code when a new device is created.  Allocate the basic
1914  * structures and the ring buffer for communication with the backend, and
1915  * inform the backend of the appropriate details for those.  Switch to
1916  * Initialised state.
1917  */
1918 static int blkfront_probe(struct xenbus_device *dev,
1919                           const struct xenbus_device_id *id)
1920 {
1921         int err, vdevice;
1922         struct blkfront_info *info;
1923
1924         /* FIXME: Use dynamic device id if this is not set. */
1925         err = xenbus_scanf(XBT_NIL, dev->nodename,
1926                            "virtual-device", "%i", &vdevice);
1927         if (err != 1) {
1928                 /* go looking in the extended area instead */
1929                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1930                                    "%i", &vdevice);
1931                 if (err != 1) {
1932                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1933                         return err;
1934                 }
1935         }
1936
1937         if (xen_hvm_domain()) {
1938                 char *type;
1939                 int len;
1940                 /* no unplug has been done: do not hook devices != xen vbds */
1941                 if (xen_has_pv_and_legacy_disk_devices()) {
1942                         int major;
1943
1944                         if (!VDEV_IS_EXTENDED(vdevice))
1945                                 major = BLKIF_MAJOR(vdevice);
1946                         else
1947                                 major = XENVBD_MAJOR;
1948
1949                         if (major != XENVBD_MAJOR) {
1950                                 printk(KERN_INFO
1951                                                 "%s: HVM does not support vbd %d as xen block device\n",
1952                                                 __func__, vdevice);
1953                                 return -ENODEV;
1954                         }
1955                 }
1956                 /* do not create a PV cdrom device if we are an HVM guest */
1957                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1958                 if (IS_ERR(type))
1959                         return -ENODEV;
1960                 if (strncmp(type, "cdrom", 5) == 0) {
1961                         kfree(type);
1962                         return -ENODEV;
1963                 }
1964                 kfree(type);
1965         }
1966         info = kzalloc(sizeof(*info), GFP_KERNEL);
1967         if (!info) {
1968                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
1969                 return -ENOMEM;
1970         }
1971
1972         info->xbdev = dev;
1973         err = negotiate_mq(info);
1974         if (err) {
1975                 kfree(info);
1976                 return err;
1977         }
1978
1979         mutex_init(&info->mutex);
1980         info->vdevice = vdevice;
1981         info->connected = BLKIF_STATE_DISCONNECTED;
1982
1983         /* Front end dir is a number, which is used as the id. */
1984         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
1985         dev_set_drvdata(&dev->dev, info);
1986
1987         return 0;
1988 }
1989
1990 static void split_bio_end(struct bio *bio)
1991 {
1992         struct split_bio *split_bio = bio->bi_private;
1993
1994         if (atomic_dec_and_test(&split_bio->pending)) {
1995                 split_bio->bio->bi_phys_segments = 0;
1996                 split_bio->bio->bi_error = bio->bi_error;
1997                 bio_endio(split_bio->bio);
1998                 kfree(split_bio);
1999         }
2000         bio_put(bio);
2001 }
2002
2003 static int blkif_recover(struct blkfront_info *info)
2004 {
2005         unsigned int i, r_index;
2006         struct request *req, *n;
2007         struct blk_shadow *copy;
2008         int rc;
2009         struct bio *bio, *cloned_bio;
2010         struct bio_list bio_list, merge_bio;
2011         unsigned int segs, offset;
2012         int pending, size;
2013         struct split_bio *split_bio;
2014         struct list_head requests;
2015
2016         blkfront_gather_backend_features(info);
2017         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2018         blk_queue_max_segments(info->rq, segs);
2019         bio_list_init(&bio_list);
2020         INIT_LIST_HEAD(&requests);
2021
2022         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2023                 struct blkfront_ring_info *rinfo;
2024
2025                 rinfo = &info->rinfo[r_index];
2026                 /* Stage 1: Make a safe copy of the shadow state. */
2027                 copy = kmemdup(rinfo->shadow, sizeof(rinfo->shadow),
2028                                GFP_NOIO | __GFP_REPEAT | __GFP_HIGH);
2029                 if (!copy)
2030                         return -ENOMEM;
2031
2032                 /* Stage 2: Set up free list. */
2033                 memset(&rinfo->shadow, 0, sizeof(rinfo->shadow));
2034                 for (i = 0; i < BLK_RING_SIZE(info); i++)
2035                         rinfo->shadow[i].req.u.rw.id = i+1;
2036                 rinfo->shadow_free = rinfo->ring.req_prod_pvt;
2037                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
2038
2039                 rc = blkfront_setup_indirect(rinfo);
2040                 if (rc) {
2041                         kfree(copy);
2042                         return rc;
2043                 }
2044
2045                 for (i = 0; i < BLK_RING_SIZE(info); i++) {
2046                         /* Not in use? */
2047                         if (!copy[i].request)
2048                                 continue;
2049
2050                         /*
2051                          * Get the bios in the request so we can re-queue them.
2052                          */
2053                         if (req_op(copy[i].request) == REQ_OP_FLUSH ||
2054                             req_op(copy[i].request) == REQ_OP_DISCARD ||
2055                             req_op(copy[i].request) == REQ_OP_SECURE_ERASE ||
2056                             copy[i].request->cmd_flags & REQ_FUA) {
2057                                 /*
2058                                  * Flush operations don't contain bios, so
2059                                  * we need to requeue the whole request
2060                                  *
2061                                  * XXX: but this doesn't make any sense for a
2062                                  * write with the FUA flag set..
2063                                  */
2064                                 list_add(&copy[i].request->queuelist, &requests);
2065                                 continue;
2066                         }
2067                         merge_bio.head = copy[i].request->bio;
2068                         merge_bio.tail = copy[i].request->biotail;
2069                         bio_list_merge(&bio_list, &merge_bio);
2070                         copy[i].request->bio = NULL;
2071                         blk_end_request_all(copy[i].request, 0);
2072                 }
2073
2074                 kfree(copy);
2075         }
2076         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2077
2078         /* Now safe for us to use the shared ring */
2079         info->connected = BLKIF_STATE_CONNECTED;
2080
2081         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2082                 struct blkfront_ring_info *rinfo;
2083
2084                 rinfo = &info->rinfo[r_index];
2085                 /* Kick any other new requests queued since we resumed */
2086                 kick_pending_request_queues(rinfo);
2087         }
2088
2089         list_for_each_entry_safe(req, n, &requests, queuelist) {
2090                 /* Requeue pending requests (flush or discard) */
2091                 list_del_init(&req->queuelist);
2092                 BUG_ON(req->nr_phys_segments > segs);
2093                 blk_mq_requeue_request(req);
2094         }
2095         blk_mq_kick_requeue_list(info->rq);
2096
2097         while ((bio = bio_list_pop(&bio_list)) != NULL) {
2098                 /* Traverse the list of pending bios and re-queue them */
2099                 if (bio_segments(bio) > segs) {
2100                         /*
2101                          * This bio has more segments than what we can
2102                          * handle, we have to split it.
2103                          */
2104                         pending = (bio_segments(bio) + segs - 1) / segs;
2105                         split_bio = kzalloc(sizeof(*split_bio), GFP_NOIO);
2106                         BUG_ON(split_bio == NULL);
2107                         atomic_set(&split_bio->pending, pending);
2108                         split_bio->bio = bio;
2109                         for (i = 0; i < pending; i++) {
2110                                 offset = (i * segs * XEN_PAGE_SIZE) >> 9;
2111                                 size = min((unsigned int)(segs * XEN_PAGE_SIZE) >> 9,
2112                                            (unsigned int)bio_sectors(bio) - offset);
2113                                 cloned_bio = bio_clone(bio, GFP_NOIO);
2114                                 BUG_ON(cloned_bio == NULL);
2115                                 bio_trim(cloned_bio, offset, size);
2116                                 cloned_bio->bi_private = split_bio;
2117                                 cloned_bio->bi_end_io = split_bio_end;
2118                                 submit_bio(cloned_bio);
2119                         }
2120                         /*
2121                          * Now we have to wait for all those smaller bios to
2122                          * end, so we can also end the "parent" bio.
2123                          */
2124                         continue;
2125                 }
2126                 /* We don't need to split this bio */
2127                 submit_bio(bio);
2128         }
2129
2130         return 0;
2131 }
2132
2133 /**
2134  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2135  * driver restart.  We tear down our blkif structure and recreate it, but
2136  * leave the device-layer structures intact so that this is transparent to the
2137  * rest of the kernel.
2138  */
2139 static int blkfront_resume(struct xenbus_device *dev)
2140 {
2141         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2142         int err = 0;
2143
2144         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2145
2146         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2147
2148         err = negotiate_mq(info);
2149         if (err)
2150                 return err;
2151
2152         err = talk_to_blkback(dev, info);
2153
2154         /*
2155          * We have to wait for the backend to switch to
2156          * connected state, since we want to read which
2157          * features it supports.
2158          */
2159
2160         return err;
2161 }
2162
2163 static void blkfront_closing(struct blkfront_info *info)
2164 {
2165         struct xenbus_device *xbdev = info->xbdev;
2166         struct block_device *bdev = NULL;
2167
2168         mutex_lock(&info->mutex);
2169
2170         if (xbdev->state == XenbusStateClosing) {
2171                 mutex_unlock(&info->mutex);
2172                 return;
2173         }
2174
2175         if (info->gd)
2176                 bdev = bdget_disk(info->gd, 0);
2177
2178         mutex_unlock(&info->mutex);
2179
2180         if (!bdev) {
2181                 xenbus_frontend_closed(xbdev);
2182                 return;
2183         }
2184
2185         mutex_lock(&bdev->bd_mutex);
2186
2187         if (bdev->bd_openers) {
2188                 xenbus_dev_error(xbdev, -EBUSY,
2189                                  "Device in use; refusing to close");
2190                 xenbus_switch_state(xbdev, XenbusStateClosing);
2191         } else {
2192                 xlvbd_release_gendisk(info);
2193                 xenbus_frontend_closed(xbdev);
2194         }
2195
2196         mutex_unlock(&bdev->bd_mutex);
2197         bdput(bdev);
2198 }
2199
2200 static void blkfront_setup_discard(struct blkfront_info *info)
2201 {
2202         int err;
2203         unsigned int discard_granularity;
2204         unsigned int discard_alignment;
2205         unsigned int discard_secure;
2206
2207         info->feature_discard = 1;
2208         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2209                 "discard-granularity", "%u", &discard_granularity,
2210                 "discard-alignment", "%u", &discard_alignment,
2211                 NULL);
2212         if (!err) {
2213                 info->discard_granularity = discard_granularity;
2214                 info->discard_alignment = discard_alignment;
2215         }
2216         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2217                     "discard-secure", "%d", &discard_secure,
2218                     NULL);
2219         if (!err)
2220                 info->feature_secdiscard = !!discard_secure;
2221 }
2222
2223 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2224 {
2225         unsigned int psegs, grants;
2226         int err, i;
2227         struct blkfront_info *info = rinfo->dev_info;
2228
2229         if (info->max_indirect_segments == 0) {
2230                 if (!HAS_EXTRA_REQ)
2231                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2232                 else {
2233                         /*
2234                          * When an extra req is required, the maximum
2235                          * grants supported is related to the size of the
2236                          * Linux block segment.
2237                          */
2238                         grants = GRANTS_PER_PSEG;
2239                 }
2240         }
2241         else
2242                 grants = info->max_indirect_segments;
2243         psegs = grants / GRANTS_PER_PSEG;
2244
2245         err = fill_grant_buffer(rinfo,
2246                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2247         if (err)
2248                 goto out_of_memory;
2249
2250         if (!info->feature_persistent && info->max_indirect_segments) {
2251                 /*
2252                  * We are using indirect descriptors but not persistent
2253                  * grants, we need to allocate a set of pages that can be
2254                  * used for mapping indirect grefs
2255                  */
2256                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2257
2258                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2259                 for (i = 0; i < num; i++) {
2260                         struct page *indirect_page = alloc_page(GFP_NOIO);
2261                         if (!indirect_page)
2262                                 goto out_of_memory;
2263                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2264                 }
2265         }
2266
2267         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2268                 rinfo->shadow[i].grants_used = kzalloc(
2269                         sizeof(rinfo->shadow[i].grants_used[0]) * grants,
2270                         GFP_NOIO);
2271                 rinfo->shadow[i].sg = kzalloc(sizeof(rinfo->shadow[i].sg[0]) * psegs, GFP_NOIO);
2272                 if (info->max_indirect_segments)
2273                         rinfo->shadow[i].indirect_grants = kzalloc(
2274                                 sizeof(rinfo->shadow[i].indirect_grants[0]) *
2275                                 INDIRECT_GREFS(grants),
2276                                 GFP_NOIO);
2277                 if ((rinfo->shadow[i].grants_used == NULL) ||
2278                         (rinfo->shadow[i].sg == NULL) ||
2279                      (info->max_indirect_segments &&
2280                      (rinfo->shadow[i].indirect_grants == NULL)))
2281                         goto out_of_memory;
2282                 sg_init_table(rinfo->shadow[i].sg, psegs);
2283         }
2284
2285
2286         return 0;
2287
2288 out_of_memory:
2289         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2290                 kfree(rinfo->shadow[i].grants_used);
2291                 rinfo->shadow[i].grants_used = NULL;
2292                 kfree(rinfo->shadow[i].sg);
2293                 rinfo->shadow[i].sg = NULL;
2294                 kfree(rinfo->shadow[i].indirect_grants);
2295                 rinfo->shadow[i].indirect_grants = NULL;
2296         }
2297         if (!list_empty(&rinfo->indirect_pages)) {
2298                 struct page *indirect_page, *n;
2299                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2300                         list_del(&indirect_page->lru);
2301                         __free_page(indirect_page);
2302                 }
2303         }
2304         return -ENOMEM;
2305 }
2306
2307 /*
2308  * Gather all backend feature-*
2309  */
2310 static void blkfront_gather_backend_features(struct blkfront_info *info)
2311 {
2312         int err;
2313         int barrier, flush, discard, persistent;
2314         unsigned int indirect_segments;
2315
2316         info->feature_flush = 0;
2317         info->feature_fua = 0;
2318
2319         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2320                         "feature-barrier", "%d", &barrier,
2321                         NULL);
2322
2323         /*
2324          * If there's no "feature-barrier" defined, then it means
2325          * we're dealing with a very old backend which writes
2326          * synchronously; nothing to do.
2327          *
2328          * If there are barriers, then we use flush.
2329          */
2330         if (!err && barrier) {
2331                 info->feature_flush = 1;
2332                 info->feature_fua = 1;
2333         }
2334
2335         /*
2336          * And if there is "feature-flush-cache" use that above
2337          * barriers.
2338          */
2339         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2340                         "feature-flush-cache", "%d", &flush,
2341                         NULL);
2342
2343         if (!err && flush) {
2344                 info->feature_flush = 1;
2345                 info->feature_fua = 0;
2346         }
2347
2348         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2349                         "feature-discard", "%d", &discard,
2350                         NULL);
2351
2352         if (!err && discard)
2353                 blkfront_setup_discard(info);
2354
2355         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2356                         "feature-persistent", "%u", &persistent,
2357                         NULL);
2358         if (err)
2359                 info->feature_persistent = 0;
2360         else
2361                 info->feature_persistent = persistent;
2362
2363         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2364                             "feature-max-indirect-segments", "%u", &indirect_segments,
2365                             NULL);
2366         if (err)
2367                 info->max_indirect_segments = 0;
2368         else
2369                 info->max_indirect_segments = min(indirect_segments,
2370                                                   xen_blkif_max_segments);
2371 }
2372
2373 /*
2374  * Invoked when the backend is finally 'ready' (and has told produced
2375  * the details about the physical device - #sectors, size, etc).
2376  */
2377 static void blkfront_connect(struct blkfront_info *info)
2378 {
2379         unsigned long long sectors;
2380         unsigned long sector_size;
2381         unsigned int physical_sector_size;
2382         unsigned int binfo;
2383         int err, i;
2384
2385         switch (info->connected) {
2386         case BLKIF_STATE_CONNECTED:
2387                 /*
2388                  * Potentially, the back-end may be signalling
2389                  * a capacity change; update the capacity.
2390                  */
2391                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2392                                    "sectors", "%Lu", &sectors);
2393                 if (XENBUS_EXIST_ERR(err))
2394                         return;
2395                 printk(KERN_INFO "Setting capacity to %Lu\n",
2396                        sectors);
2397                 set_capacity(info->gd, sectors);
2398                 revalidate_disk(info->gd);
2399
2400                 return;
2401         case BLKIF_STATE_SUSPENDED:
2402                 /*
2403                  * If we are recovering from suspension, we need to wait
2404                  * for the backend to announce it's features before
2405                  * reconnecting, at least we need to know if the backend
2406                  * supports indirect descriptors, and how many.
2407                  */
2408                 blkif_recover(info);
2409                 return;
2410
2411         default:
2412                 break;
2413         }
2414
2415         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2416                 __func__, info->xbdev->otherend);
2417
2418         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2419                             "sectors", "%llu", &sectors,
2420                             "info", "%u", &binfo,
2421                             "sector-size", "%lu", &sector_size,
2422                             NULL);
2423         if (err) {
2424                 xenbus_dev_fatal(info->xbdev, err,
2425                                  "reading backend fields at %s",
2426                                  info->xbdev->otherend);
2427                 return;
2428         }
2429
2430         /*
2431          * physcial-sector-size is a newer field, so old backends may not
2432          * provide this. Assume physical sector size to be the same as
2433          * sector_size in that case.
2434          */
2435         err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2436                            "physical-sector-size", "%u", &physical_sector_size);
2437         if (err != 1)
2438                 physical_sector_size = sector_size;
2439
2440         blkfront_gather_backend_features(info);
2441         for (i = 0; i < info->nr_rings; i++) {
2442                 err = blkfront_setup_indirect(&info->rinfo[i]);
2443                 if (err) {
2444                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2445                                          info->xbdev->otherend);
2446                         blkif_free(info, 0);
2447                         break;
2448                 }
2449         }
2450
2451         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2452                                   physical_sector_size);
2453         if (err) {
2454                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2455                                  info->xbdev->otherend);
2456                 return;
2457         }
2458
2459         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2460
2461         /* Kick pending requests. */
2462         info->connected = BLKIF_STATE_CONNECTED;
2463         for (i = 0; i < info->nr_rings; i++)
2464                 kick_pending_request_queues(&info->rinfo[i]);
2465
2466         add_disk(info->gd);
2467
2468         info->is_ready = 1;
2469 }
2470
2471 /**
2472  * Callback received when the backend's state changes.
2473  */
2474 static void blkback_changed(struct xenbus_device *dev,
2475                             enum xenbus_state backend_state)
2476 {
2477         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2478
2479         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2480
2481         switch (backend_state) {
2482         case XenbusStateInitWait:
2483                 if (dev->state != XenbusStateInitialising)
2484                         break;
2485                 if (talk_to_blkback(dev, info))
2486                         break;
2487         case XenbusStateInitialising:
2488         case XenbusStateInitialised:
2489         case XenbusStateReconfiguring:
2490         case XenbusStateReconfigured:
2491         case XenbusStateUnknown:
2492                 break;
2493
2494         case XenbusStateConnected:
2495                 if (dev->state != XenbusStateInitialised) {
2496                         if (talk_to_blkback(dev, info))
2497                                 break;
2498                 }
2499                 blkfront_connect(info);
2500                 break;
2501
2502         case XenbusStateClosed:
2503                 if (dev->state == XenbusStateClosed)
2504                         break;
2505                 /* Missed the backend's Closing state -- fallthrough */
2506         case XenbusStateClosing:
2507                 if (info)
2508                         blkfront_closing(info);
2509                 break;
2510         }
2511 }
2512
2513 static int blkfront_remove(struct xenbus_device *xbdev)
2514 {
2515         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2516         struct block_device *bdev = NULL;
2517         struct gendisk *disk;
2518
2519         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2520
2521         blkif_free(info, 0);
2522
2523         mutex_lock(&info->mutex);
2524
2525         disk = info->gd;
2526         if (disk)
2527                 bdev = bdget_disk(disk, 0);
2528
2529         info->xbdev = NULL;
2530         mutex_unlock(&info->mutex);
2531
2532         if (!bdev) {
2533                 kfree(info);
2534                 return 0;
2535         }
2536
2537         /*
2538          * The xbdev was removed before we reached the Closed
2539          * state. See if it's safe to remove the disk. If the bdev
2540          * isn't closed yet, we let release take care of it.
2541          */
2542
2543         mutex_lock(&bdev->bd_mutex);
2544         info = disk->private_data;
2545
2546         dev_warn(disk_to_dev(disk),
2547                  "%s was hot-unplugged, %d stale handles\n",
2548                  xbdev->nodename, bdev->bd_openers);
2549
2550         if (info && !bdev->bd_openers) {
2551                 xlvbd_release_gendisk(info);
2552                 disk->private_data = NULL;
2553                 kfree(info);
2554         }
2555
2556         mutex_unlock(&bdev->bd_mutex);
2557         bdput(bdev);
2558
2559         return 0;
2560 }
2561
2562 static int blkfront_is_ready(struct xenbus_device *dev)
2563 {
2564         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2565
2566         return info->is_ready && info->xbdev;
2567 }
2568
2569 static int blkif_open(struct block_device *bdev, fmode_t mode)
2570 {
2571         struct gendisk *disk = bdev->bd_disk;
2572         struct blkfront_info *info;
2573         int err = 0;
2574
2575         mutex_lock(&blkfront_mutex);
2576
2577         info = disk->private_data;
2578         if (!info) {
2579                 /* xbdev gone */
2580                 err = -ERESTARTSYS;
2581                 goto out;
2582         }
2583
2584         mutex_lock(&info->mutex);
2585
2586         if (!info->gd)
2587                 /* xbdev is closed */
2588                 err = -ERESTARTSYS;
2589
2590         mutex_unlock(&info->mutex);
2591
2592 out:
2593         mutex_unlock(&blkfront_mutex);
2594         return err;
2595 }
2596
2597 static void blkif_release(struct gendisk *disk, fmode_t mode)
2598 {
2599         struct blkfront_info *info = disk->private_data;
2600         struct block_device *bdev;
2601         struct xenbus_device *xbdev;
2602
2603         mutex_lock(&blkfront_mutex);
2604
2605         bdev = bdget_disk(disk, 0);
2606
2607         if (!bdev) {
2608                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2609                 goto out_mutex;
2610         }
2611         if (bdev->bd_openers)
2612                 goto out;
2613
2614         /*
2615          * Check if we have been instructed to close. We will have
2616          * deferred this request, because the bdev was still open.
2617          */
2618
2619         mutex_lock(&info->mutex);
2620         xbdev = info->xbdev;
2621
2622         if (xbdev && xbdev->state == XenbusStateClosing) {
2623                 /* pending switch to state closed */
2624                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2625                 xlvbd_release_gendisk(info);
2626                 xenbus_frontend_closed(info->xbdev);
2627         }
2628
2629         mutex_unlock(&info->mutex);
2630
2631         if (!xbdev) {
2632                 /* sudden device removal */
2633                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2634                 xlvbd_release_gendisk(info);
2635                 disk->private_data = NULL;
2636                 kfree(info);
2637         }
2638
2639 out:
2640         bdput(bdev);
2641 out_mutex:
2642         mutex_unlock(&blkfront_mutex);
2643 }
2644
2645 static const struct block_device_operations xlvbd_block_fops =
2646 {
2647         .owner = THIS_MODULE,
2648         .open = blkif_open,
2649         .release = blkif_release,
2650         .getgeo = blkif_getgeo,
2651         .ioctl = blkif_ioctl,
2652 };
2653
2654
2655 static const struct xenbus_device_id blkfront_ids[] = {
2656         { "vbd" },
2657         { "" }
2658 };
2659
2660 static struct xenbus_driver blkfront_driver = {
2661         .ids  = blkfront_ids,
2662         .probe = blkfront_probe,
2663         .remove = blkfront_remove,
2664         .resume = blkfront_resume,
2665         .otherend_changed = blkback_changed,
2666         .is_ready = blkfront_is_ready,
2667 };
2668
2669 static int __init xlblk_init(void)
2670 {
2671         int ret;
2672         int nr_cpus = num_online_cpus();
2673
2674         if (!xen_domain())
2675                 return -ENODEV;
2676
2677         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2678                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2679                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2680                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2681         }
2682
2683         if (xen_blkif_max_queues > nr_cpus) {
2684                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2685                         xen_blkif_max_queues, nr_cpus);
2686                 xen_blkif_max_queues = nr_cpus;
2687         }
2688
2689         if (!xen_has_pv_disk_devices())
2690                 return -ENODEV;
2691
2692         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2693                 printk(KERN_WARNING "xen_blk: can't get major %d with name %s\n",
2694                        XENVBD_MAJOR, DEV_NAME);
2695                 return -ENODEV;
2696         }
2697
2698         ret = xenbus_register_frontend(&blkfront_driver);
2699         if (ret) {
2700                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2701                 return ret;
2702         }
2703
2704         return 0;
2705 }
2706 module_init(xlblk_init);
2707
2708
2709 static void __exit xlblk_exit(void)
2710 {
2711         xenbus_unregister_driver(&blkfront_driver);
2712         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2713         kfree(minors);
2714 }
2715 module_exit(xlblk_exit);
2716
2717 MODULE_DESCRIPTION("Xen virtual block device frontend");
2718 MODULE_LICENSE("GPL");
2719 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2720 MODULE_ALIAS("xen:vbd");
2721 MODULE_ALIAS("xenblk");