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 #include <linux/workqueue.h>
  50
  51 #include <xen/xen.h>
  52 #include <xen/xenbus.h>
  53 #include <xen/grant_table.h>
  54 #include <xen/events.h>
  55 #include <xen/page.h>
  56 #include <xen/platform_pci.h>
  57
  58 #include <xen/interface/grant_table.h>
  59 #include <xen/interface/io/blkif.h>
  60 #include <xen/interface/io/protocols.h>
  61
  62 #include <asm/xen/hypervisor.h>
  63
  64 /*
  65  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  66  * When Linux is using a different page size than Xen, it may not be possible
  67  * to put all the data in a single segment.
  68  * This can happen when the backend doesn't support indirect descriptor and
  69  * therefore the maximum amount of data that a request can carry is
  70  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  71  *
  72  * Note that we only support one extra request. So the Linux page size
  73  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  74  * 88KB.
  75  */
  76 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  77
  78 enum blkif_state {
  79         BLKIF_STATE_DISCONNECTED,
  80         BLKIF_STATE_CONNECTED,
  81         BLKIF_STATE_SUSPENDED,
  82 };
  83
  84 struct grant {
  85         grant_ref_t gref;
  86         struct page *page;
  87         struct list_head node;
  88 };
  89
  90 enum blk_req_status {
  91         REQ_WAITING,
  92         REQ_DONE,
  93         REQ_ERROR,
  94         REQ_EOPNOTSUPP,
  95 };
  96
  97 struct blk_shadow {
  98         struct blkif_request req;
  99         struct request *request;
 100         struct grant **grants_used;
 101         struct grant **indirect_grants;
 102         struct scatterlist *sg;
 103         unsigned int num_sg;
 104         enum blk_req_status status;
 105
 106         #define NO_ASSOCIATED_ID ~0UL
 107         /*
 108          * Id of the sibling if we ever need 2 requests when handling a
 109          * block I/O request
 110          */
 111         unsigned long associated_id;
 112 };
 113
 114 struct blkif_req {
 115         blk_status_t    error;
 116 };
 117
 118 static inline struct blkif_req *blkif_req(struct request *rq)
 119 {
 120         return blk_mq_rq_to_pdu(rq);
 121 }
 122
 123 static DEFINE_MUTEX(blkfront_mutex);
 124 static const struct block_device_operations xlvbd_block_fops;
 125 static struct delayed_work blkfront_work;
 126 static LIST_HEAD(info_list);
 127
 128 /*
 129  * Maximum number of segments in indirect requests, the actual value used by
 130  * the frontend driver is the minimum of this value and the value provided
 131  * by the backend driver.
 132  */
 133
 134 static unsigned int xen_blkif_max_segments = 32;
 135 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 136 MODULE_PARM_DESC(max_indirect_segments,
 137                  "Maximum amount of segments in indirect requests (default is 32)");
 138
 139 static unsigned int xen_blkif_max_queues = 4;
 140 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 141 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 142
 143 /*
 144  * Maximum order of pages to be used for the shared ring between front and
 145  * backend, 4KB page granularity is used.
 146  */
 147 static unsigned int xen_blkif_max_ring_order;
 148 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 149 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 150
 151 #define BLK_RING_SIZE(info)     \
 152         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 153
 154 /*
 155  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 156  * characters are enough. Define to 20 to keep consistent with backend.
 157  */
 158 #define RINGREF_NAME_LEN (20)
 159 /*
 160  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 161  */
 162 #define QUEUE_NAME_LEN (17)
 163
 164 /*
 165  *  Per-ring info.
 166  *  Every blkfront device can associate with one or more blkfront_ring_info,
 167  *  depending on how many hardware queues/rings to be used.
 168  */
 169 struct blkfront_ring_info {
 170         /* Lock to protect data in every ring buffer. */
 171         spinlock_t ring_lock;
 172         struct blkif_front_ring ring;
 173         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 174         unsigned int evtchn, irq;
 175         struct work_struct work;
 176         struct gnttab_free_callback callback;
 177         struct list_head indirect_pages;
 178         struct list_head grants;
 179         unsigned int persistent_gnts_c;
 180         unsigned long shadow_free;
 181         struct blkfront_info *dev_info;
 182         struct blk_shadow shadow[];
 183 };
 184
 185 /*
 186  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 187  * hang in private_data off the gendisk structure. We may end up
 188  * putting all kinds of interesting stuff here :-)
 189  */
 190 struct blkfront_info
 191 {
 192         struct mutex mutex;
 193         struct xenbus_device *xbdev;
 194         struct gendisk *gd;
 195         u16 sector_size;
 196         unsigned int physical_sector_size;
 197         int vdevice;
 198         blkif_vdev_t handle;
 199         enum blkif_state connected;
 200         /* Number of pages per ring buffer. */
 201         unsigned int nr_ring_pages;
 202         struct request_queue *rq;
 203         unsigned int feature_flush:1;
 204         unsigned int feature_fua:1;
 205         unsigned int feature_discard:1;
 206         unsigned int feature_secdiscard:1;
 207         unsigned int feature_persistent:1;
 208         unsigned int discard_granularity;
 209         unsigned int discard_alignment;
 210         /* Number of 4KB segments handled */
 211         unsigned int max_indirect_segments;
 212         int is_ready;
 213         struct blk_mq_tag_set tag_set;
 214         struct blkfront_ring_info *rinfo;
 215         unsigned int nr_rings;
 216         unsigned int rinfo_size;
 217         /* Save uncomplete reqs and bios for migration. */
 218         struct list_head requests;
 219         struct bio_list bio_list;
 220         struct list_head info_list;
 221 };
 222
 223 static unsigned int nr_minors;
 224 static unsigned long *minors;
 225 static DEFINE_SPINLOCK(minor_lock);
 226
 227 #define GRANT_INVALID_REF       0
 228
 229 #define PARTS_PER_DISK          16
 230 #define PARTS_PER_EXT_DISK      256
 231
 232 #define BLKIF_MAJOR(dev) ((dev)>>8)
 233 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 234
 235 #define EXT_SHIFT 28
 236 #define EXTENDED (1<<EXT_SHIFT)
 237 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 238 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 239 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 240 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 241 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 242 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 243
 244 #define DEV_NAME        "xvd"   /* name in /dev */
 245
 246 /*
 247  * Grants are always the same size as a Xen page (i.e 4KB).
 248  * A physical segment is always the same size as a Linux page.
 249  * Number of grants per physical segment
 250  */
 251 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 252
 253 #define GRANTS_PER_INDIRECT_FRAME \
 254         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 255
 256 #define INDIRECT_GREFS(_grants)         \
 257         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 258
 259 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 260 static void blkfront_gather_backend_features(struct blkfront_info *info);
 261 static int negotiate_mq(struct blkfront_info *info);
 262
 263 #define for_each_rinfo(info, ptr, idx)                          \
 264         for ((ptr) = (info)->rinfo, (idx) = 0;                  \
 265              (idx) < (info)->nr_rings;                          \
 266              (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 267
 268 static inline struct blkfront_ring_info *
 269 get_rinfo(const struct blkfront_info *info, unsigned int i)
 270 {
 271         BUG_ON(i >= info->nr_rings);
 272         return (void *)info->rinfo + i * info->rinfo_size;
 273 }
 274
 275 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 276 {
 277         unsigned long free = rinfo->shadow_free;
 278
 279         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 280         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 281         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 282         return free;
 283 }
 284
 285 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 286                               unsigned long id)
 287 {
 288         if (rinfo->shadow[id].req.u.rw.id != id)
 289                 return -EINVAL;
 290         if (rinfo->shadow[id].request == NULL)
 291                 return -EINVAL;
 292         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 293         rinfo->shadow[id].request = NULL;
 294         rinfo->shadow_free = id;
 295         return 0;
 296 }
 297
 298 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 299 {
 300         struct blkfront_info *info = rinfo->dev_info;
 301         struct page *granted_page;
 302         struct grant *gnt_list_entry, *n;
 303         int i = 0;
 304
 305         while (i < num) {
 306                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 307                 if (!gnt_list_entry)
 308                         goto out_of_memory;
 309
 310                 if (info->feature_persistent) {
 311                         granted_page = alloc_page(GFP_NOIO);
 312                         if (!granted_page) {
 313                                 kfree(gnt_list_entry);
 314                                 goto out_of_memory;
 315                         }
 316                         gnt_list_entry->page = granted_page;
 317                 }
 318
 319                 gnt_list_entry->gref = GRANT_INVALID_REF;
 320                 list_add(&gnt_list_entry->node, &rinfo->grants);
 321                 i++;
 322         }
 323
 324         return 0;
 325
 326 out_of_memory:
 327         list_for_each_entry_safe(gnt_list_entry, n,
 328                                  &rinfo->grants, node) {
 329                 list_del(&gnt_list_entry->node);
 330                 if (info->feature_persistent)
 331                         __free_page(gnt_list_entry->page);
 332                 kfree(gnt_list_entry);
 333                 i--;
 334         }
 335         BUG_ON(i != 0);
 336         return -ENOMEM;
 337 }
 338
 339 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 340 {
 341         struct grant *gnt_list_entry;
 342
 343         BUG_ON(list_empty(&rinfo->grants));
 344         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 345                                           node);
 346         list_del(&gnt_list_entry->node);
 347
 348         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 349                 rinfo->persistent_gnts_c--;
 350
 351         return gnt_list_entry;
 352 }
 353
 354 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 355                                         const struct blkfront_info *info)
 356 {
 357         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 358                                                  info->xbdev->otherend_id,
 359                                                  gnt_list_entry->page,
 360                                                  0);
 361 }
 362
 363 static struct grant *get_grant(grant_ref_t *gref_head,
 364                                unsigned long gfn,
 365                                struct blkfront_ring_info *rinfo)
 366 {
 367         struct grant *gnt_list_entry = get_free_grant(rinfo);
 368         struct blkfront_info *info = rinfo->dev_info;
 369
 370         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 371                 return gnt_list_entry;
 372
 373         /* Assign a gref to this page */
 374         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 375         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 376         if (info->feature_persistent)
 377                 grant_foreign_access(gnt_list_entry, info);
 378         else {
 379                 /* Grant access to the GFN passed by the caller */
 380                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 381                                                 info->xbdev->otherend_id,
 382                                                 gfn, 0);
 383         }
 384
 385         return gnt_list_entry;
 386 }
 387
 388 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 389                                         struct blkfront_ring_info *rinfo)
 390 {
 391         struct grant *gnt_list_entry = get_free_grant(rinfo);
 392         struct blkfront_info *info = rinfo->dev_info;
 393
 394         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 395                 return gnt_list_entry;
 396
 397         /* Assign a gref to this page */
 398         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 399         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 400         if (!info->feature_persistent) {
 401                 struct page *indirect_page;
 402
 403                 /* Fetch a pre-allocated page to use for indirect grefs */
 404                 BUG_ON(list_empty(&rinfo->indirect_pages));
 405                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 406                                                  struct page, lru);
 407                 list_del(&indirect_page->lru);
 408                 gnt_list_entry->page = indirect_page;
 409         }
 410         grant_foreign_access(gnt_list_entry, info);
 411
 412         return gnt_list_entry;
 413 }
 414
 415 static const char *op_name(int op)
 416 {
 417         static const char *const names[] = {
 418                 [BLKIF_OP_READ] = "read",
 419                 [BLKIF_OP_WRITE] = "write",
 420                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 421                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 422                 [BLKIF_OP_DISCARD] = "discard" };
 423
 424         if (op < 0 || op >= ARRAY_SIZE(names))
 425                 return "unknown";
 426
 427         if (!names[op])
 428                 return "reserved";
 429
 430         return names[op];
 431 }
 432 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 433 {
 434         unsigned int end = minor + nr;
 435         int rc;
 436
 437         if (end > nr_minors) {
 438                 unsigned long *bitmap, *old;
 439
 440                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 441                                  GFP_KERNEL);
 442                 if (bitmap == NULL)
 443                         return -ENOMEM;
 444
 445                 spin_lock(&minor_lock);
 446                 if (end > nr_minors) {
 447                         old = minors;
 448                         memcpy(bitmap, minors,
 449                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 450                         minors = bitmap;
 451                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 452                 } else
 453                         old = bitmap;
 454                 spin_unlock(&minor_lock);
 455                 kfree(old);
 456         }
 457
 458         spin_lock(&minor_lock);
 459         if (find_next_bit(minors, end, minor) >= end) {
 460                 bitmap_set(minors, minor, nr);
 461                 rc = 0;
 462         } else
 463                 rc = -EBUSY;
 464         spin_unlock(&minor_lock);
 465
 466         return rc;
 467 }
 468
 469 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 470 {
 471         unsigned int end = minor + nr;
 472
 473         BUG_ON(end > nr_minors);
 474         spin_lock(&minor_lock);
 475         bitmap_clear(minors,  minor, nr);
 476         spin_unlock(&minor_lock);
 477 }
 478
 479 static void blkif_restart_queue_callback(void *arg)
 480 {
 481         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 482         schedule_work(&rinfo->work);
 483 }
 484
 485 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 486 {
 487         /* We don't have real geometry info, but let's at least return
 488            values consistent with the size of the device */
 489         sector_t nsect = get_capacity(bd->bd_disk);
 490         sector_t cylinders = nsect;
 491
 492         hg->heads = 0xff;
 493         hg->sectors = 0x3f;
 494         sector_div(cylinders, hg->heads * hg->sectors);
 495         hg->cylinders = cylinders;
 496         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 497                 hg->cylinders = 0xffff;
 498         return 0;
 499 }
 500
 501 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 502                        unsigned command, unsigned long argument)
 503 {
 504         struct blkfront_info *info = bdev->bd_disk->private_data;
 505         int i;
 506
 507         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 508                 command, (long)argument);
 509
 510         switch (command) {
 511         case CDROMMULTISESSION:
 512                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 513                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 514                         if (put_user(0, (char __user *)(argument + i)))
 515                                 return -EFAULT;
 516                 return 0;
 517
 518         case CDROM_GET_CAPABILITY: {
 519                 struct gendisk *gd = info->gd;
 520                 if (gd->flags & GENHD_FL_CD)
 521                         return 0;
 522                 return -EINVAL;
 523         }
 524
 525         default:
 526                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 527                   command);*/
 528                 return -EINVAL; /* same return as native Linux */
 529         }
 530
 531         return 0;
 532 }
 533
 534 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 535                                             struct request *req,
 536                                             struct blkif_request **ring_req)
 537 {
 538         unsigned long id;
 539
 540         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 541         rinfo->ring.req_prod_pvt++;
 542
 543         id = get_id_from_freelist(rinfo);
 544         rinfo->shadow[id].request = req;
 545         rinfo->shadow[id].status = REQ_WAITING;
 546         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 547
 548         (*ring_req)->u.rw.id = id;
 549
 550         return id;
 551 }
 552
 553 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 554 {
 555         struct blkfront_info *info = rinfo->dev_info;
 556         struct blkif_request *ring_req;
 557         unsigned long id;
 558
 559         /* Fill out a communications ring structure. */
 560         id = blkif_ring_get_request(rinfo, req, &ring_req);
 561
 562         ring_req->operation = BLKIF_OP_DISCARD;
 563         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 564         ring_req->u.discard.id = id;
 565         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 566         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 567                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 568         else
 569                 ring_req->u.discard.flag = 0;
 570
 571         /* Keep a private copy so we can reissue requests when recovering. */
 572         rinfo->shadow[id].req = *ring_req;
 573
 574         return 0;
 575 }
 576
 577 struct setup_rw_req {
 578         unsigned int grant_idx;
 579         struct blkif_request_segment *segments;
 580         struct blkfront_ring_info *rinfo;
 581         struct blkif_request *ring_req;
 582         grant_ref_t gref_head;
 583         unsigned int id;
 584         /* Only used when persistent grant is used and it's a read request */
 585         bool need_copy;
 586         unsigned int bvec_off;
 587         char *bvec_data;
 588
 589         bool require_extra_req;
 590         struct blkif_request *extra_ring_req;
 591 };
 592
 593 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 594                                      unsigned int len, void *data)
 595 {
 596         struct setup_rw_req *setup = data;
 597         int n, ref;
 598         struct grant *gnt_list_entry;
 599         unsigned int fsect, lsect;
 600         /* Convenient aliases */
 601         unsigned int grant_idx = setup->grant_idx;
 602         struct blkif_request *ring_req = setup->ring_req;
 603         struct blkfront_ring_info *rinfo = setup->rinfo;
 604         /*
 605          * We always use the shadow of the first request to store the list
 606          * of grant associated to the block I/O request. This made the
 607          * completion more easy to handle even if the block I/O request is
 608          * split.
 609          */
 610         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 611
 612         if (unlikely(setup->require_extra_req &&
 613                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 614                 /*
 615                  * We are using the second request, setup grant_idx
 616                  * to be the index of the segment array.
 617                  */
 618                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 619                 ring_req = setup->extra_ring_req;
 620         }
 621
 622         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 623             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 624                 if (setup->segments)
 625                         kunmap_atomic(setup->segments);
 626
 627                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 628                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 629                 shadow->indirect_grants[n] = gnt_list_entry;
 630                 setup->segments = kmap_atomic(gnt_list_entry->page);
 631                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 632         }
 633
 634         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 635         ref = gnt_list_entry->gref;
 636         /*
 637          * All the grants are stored in the shadow of the first
 638          * request. Therefore we have to use the global index.
 639          */
 640         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 641
 642         if (setup->need_copy) {
 643                 void *shared_data;
 644
 645                 shared_data = kmap_atomic(gnt_list_entry->page);
 646                 /*
 647                  * this does not wipe data stored outside the
 648                  * range sg->offset..sg->offset+sg->length.
 649                  * Therefore, blkback *could* see data from
 650                  * previous requests. This is OK as long as
 651                  * persistent grants are shared with just one
 652                  * domain. It may need refactoring if this
 653                  * changes
 654                  */
 655                 memcpy(shared_data + offset,
 656                        setup->bvec_data + setup->bvec_off,
 657                        len);
 658
 659                 kunmap_atomic(shared_data);
 660                 setup->bvec_off += len;
 661         }
 662
 663         fsect = offset >> 9;
 664         lsect = fsect + (len >> 9) - 1;
 665         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 666                 ring_req->u.rw.seg[grant_idx] =
 667                         (struct blkif_request_segment) {
 668                                 .gref       = ref,
 669                                 .first_sect = fsect,
 670                                 .last_sect  = lsect };
 671         } else {
 672                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 673                         (struct blkif_request_segment) {
 674                                 .gref       = ref,
 675                                 .first_sect = fsect,
 676                                 .last_sect  = lsect };
 677         }
 678
 679         (setup->grant_idx)++;
 680 }
 681
 682 static void blkif_setup_extra_req(struct blkif_request *first,
 683                                   struct blkif_request *second)
 684 {
 685         uint16_t nr_segments = first->u.rw.nr_segments;
 686
 687         /*
 688          * The second request is only present when the first request uses
 689          * all its segments. It's always the continuity of the first one.
 690          */
 691         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 692
 693         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 694         second->u.rw.sector_number = first->u.rw.sector_number +
 695                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 696
 697         second->u.rw.handle = first->u.rw.handle;
 698         second->operation = first->operation;
 699 }
 700
 701 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 702 {
 703         struct blkfront_info *info = rinfo->dev_info;
 704         struct blkif_request *ring_req, *extra_ring_req = NULL;
 705         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 706         bool require_extra_req = false;
 707         int i;
 708         struct setup_rw_req setup = {
 709                 .grant_idx = 0,
 710                 .segments = NULL,
 711                 .rinfo = rinfo,
 712                 .need_copy = rq_data_dir(req) && info->feature_persistent,
 713         };
 714
 715         /*
 716          * Used to store if we are able to queue the request by just using
 717          * existing persistent grants, or if we have to get new grants,
 718          * as there are not sufficiently many free.
 719          */
 720         bool new_persistent_gnts = false;
 721         struct scatterlist *sg;
 722         int num_sg, max_grefs, num_grant;
 723
 724         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 725         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 726                 /*
 727                  * If we are using indirect segments we need to account
 728                  * for the indirect grefs used in the request.
 729                  */
 730                 max_grefs += INDIRECT_GREFS(max_grefs);
 731
 732         /* Check if we have enough persistent grants to allocate a requests */
 733         if (rinfo->persistent_gnts_c < max_grefs) {
 734                 new_persistent_gnts = true;
 735
 736                 if (gnttab_alloc_grant_references(
 737                     max_grefs - rinfo->persistent_gnts_c,
 738                     &setup.gref_head) < 0) {
 739                         gnttab_request_free_callback(
 740                                 &rinfo->callback,
 741                                 blkif_restart_queue_callback,
 742                                 rinfo,
 743                                 max_grefs - rinfo->persistent_gnts_c);
 744                         return 1;
 745                 }
 746         }
 747
 748         /* Fill out a communications ring structure. */
 749         id = blkif_ring_get_request(rinfo, req, &ring_req);
 750
 751         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 752         num_grant = 0;
 753         /* Calculate the number of grant used */
 754         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 755                num_grant += gnttab_count_grant(sg->offset, sg->length);
 756
 757         require_extra_req = info->max_indirect_segments == 0 &&
 758                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 759         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 760
 761         rinfo->shadow[id].num_sg = num_sg;
 762         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 763             likely(!require_extra_req)) {
 764                 /*
 765                  * The indirect operation can only be a BLKIF_OP_READ or
 766                  * BLKIF_OP_WRITE
 767                  */
 768                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 769                 ring_req->operation = BLKIF_OP_INDIRECT;
 770                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 771                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 772                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 773                 ring_req->u.indirect.handle = info->handle;
 774                 ring_req->u.indirect.nr_segments = num_grant;
 775         } else {
 776                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 777                 ring_req->u.rw.handle = info->handle;
 778                 ring_req->operation = rq_data_dir(req) ?
 779                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 780                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 781                         /*
 782                          * Ideally we can do an unordered flush-to-disk.
 783                          * In case the backend onlysupports barriers, use that.
 784                          * A barrier request a superset of FUA, so we can
 785                          * implement it the same way.  (It's also a FLUSH+FUA,
 786                          * since it is guaranteed ordered WRT previous writes.)
 787                          */
 788                         if (info->feature_flush && info->feature_fua)
 789                                 ring_req->operation =
 790                                         BLKIF_OP_WRITE_BARRIER;
 791                         else if (info->feature_flush)
 792                                 ring_req->operation =
 793                                         BLKIF_OP_FLUSH_DISKCACHE;
 794                         else
 795                                 ring_req->operation = 0;
 796                 }
 797                 ring_req->u.rw.nr_segments = num_grant;
 798                 if (unlikely(require_extra_req)) {
 799                         extra_id = blkif_ring_get_request(rinfo, req,
 800                                                           &extra_ring_req);
 801                         /*
 802                          * Only the first request contains the scatter-gather
 803                          * list.
 804                          */
 805                         rinfo->shadow[extra_id].num_sg = 0;
 806
 807                         blkif_setup_extra_req(ring_req, extra_ring_req);
 808
 809                         /* Link the 2 requests together */
 810                         rinfo->shadow[extra_id].associated_id = id;
 811                         rinfo->shadow[id].associated_id = extra_id;
 812                 }
 813         }
 814
 815         setup.ring_req = ring_req;
 816         setup.id = id;
 817
 818         setup.require_extra_req = require_extra_req;
 819         if (unlikely(require_extra_req))
 820                 setup.extra_ring_req = extra_ring_req;
 821
 822         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 823                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 824
 825                 if (setup.need_copy) {
 826                         setup.bvec_off = sg->offset;
 827                         setup.bvec_data = kmap_atomic(sg_page(sg));
 828                 }
 829
 830                 gnttab_foreach_grant_in_range(sg_page(sg),
 831                                               sg->offset,
 832                                               sg->length,
 833                                               blkif_setup_rw_req_grant,
 834                                               &setup);
 835
 836                 if (setup.need_copy)
 837                         kunmap_atomic(setup.bvec_data);
 838         }
 839         if (setup.segments)
 840                 kunmap_atomic(setup.segments);
 841
 842         /* Keep a private copy so we can reissue requests when recovering. */
 843         rinfo->shadow[id].req = *ring_req;
 844         if (unlikely(require_extra_req))
 845                 rinfo->shadow[extra_id].req = *extra_ring_req;
 846
 847         if (new_persistent_gnts)
 848                 gnttab_free_grant_references(setup.gref_head);
 849
 850         return 0;
 851 }
 852
 853 /*
 854  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 855  * and writes are handled as expected.
 856  *
 857  * @req: a request struct
 858  */
 859 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 860 {
 861         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 862                 return 1;
 863
 864         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 865                      req_op(req) == REQ_OP_SECURE_ERASE))
 866                 return blkif_queue_discard_req(req, rinfo);
 867         else
 868                 return blkif_queue_rw_req(req, rinfo);
 869 }
 870
 871 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 872 {
 873         int notify;
 874
 875         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 876
 877         if (notify)
 878                 notify_remote_via_irq(rinfo->irq);
 879 }
 880
 881 static inline bool blkif_request_flush_invalid(struct request *req,
 882                                                struct blkfront_info *info)
 883 {
 884         return (blk_rq_is_passthrough(req) ||
 885                 ((req_op(req) == REQ_OP_FLUSH) &&
 886                  !info->feature_flush) ||
 887                 ((req->cmd_flags & REQ_FUA) &&
 888                  !info->feature_fua));
 889 }
 890
 891 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 892                           const struct blk_mq_queue_data *qd)
 893 {
 894         unsigned long flags;
 895         int qid = hctx->queue_num;
 896         struct blkfront_info *info = hctx->queue->queuedata;
 897         struct blkfront_ring_info *rinfo = NULL;
 898
 899         rinfo = get_rinfo(info, qid);
 900         blk_mq_start_request(qd->rq);
 901         spin_lock_irqsave(&rinfo->ring_lock, flags);
 902         if (RING_FULL(&rinfo->ring))
 903                 goto out_busy;
 904
 905         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 906                 goto out_err;
 907
 908         if (blkif_queue_request(qd->rq, rinfo))
 909                 goto out_busy;
 910
 911         flush_requests(rinfo);
 912         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 913         return BLK_STS_OK;
 914
 915 out_err:
 916         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 917         return BLK_STS_IOERR;
 918
 919 out_busy:
 920         blk_mq_stop_hw_queue(hctx);
 921         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 922         return BLK_STS_DEV_RESOURCE;
 923 }
 924
 925 static void blkif_complete_rq(struct request *rq)
 926 {
 927         blk_mq_end_request(rq, blkif_req(rq)->error);
 928 }
 929
 930 static const struct blk_mq_ops blkfront_mq_ops = {
 931         .queue_rq = blkif_queue_rq,
 932         .complete = blkif_complete_rq,
 933 };
 934
 935 static void blkif_set_queue_limits(struct blkfront_info *info)
 936 {
 937         struct request_queue *rq = info->rq;
 938         struct gendisk *gd = info->gd;
 939         unsigned int segments = info->max_indirect_segments ? :
 940                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 941
 942         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 943
 944         if (info->feature_discard) {
 945                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 946                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 947                 rq->limits.discard_granularity = info->discard_granularity;
 948                 rq->limits.discard_alignment = info->discard_alignment;
 949                 if (info->feature_secdiscard)
 950                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 951         }
 952
 953         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 954         blk_queue_logical_block_size(rq, info->sector_size);
 955         blk_queue_physical_block_size(rq, info->physical_sector_size);
 956         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 957
 958         /* Each segment in a request is up to an aligned page in size. */
 959         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 960         blk_queue_max_segment_size(rq, PAGE_SIZE);
 961
 962         /* Ensure a merged request will fit in a single I/O ring slot. */
 963         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 964
 965         /* Make sure buffer addresses are sector-aligned. */
 966         blk_queue_dma_alignment(rq, 511);
 967 }
 968
 969 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 970                                 unsigned int physical_sector_size)
 971 {
 972         struct request_queue *rq;
 973         struct blkfront_info *info = gd->private_data;
 974
 975         memset(&info->tag_set, 0, sizeof(info->tag_set));
 976         info->tag_set.ops = &blkfront_mq_ops;
 977         info->tag_set.nr_hw_queues = info->nr_rings;
 978         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 979                 /*
 980                  * When indirect descriptior is not supported, the I/O request
 981                  * will be split between multiple request in the ring.
 982                  * To avoid problems when sending the request, divide by
 983                  * 2 the depth of the queue.
 984                  */
 985                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 986         } else
 987                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 988         info->tag_set.numa_node = NUMA_NO_NODE;
 989         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 990         info->tag_set.cmd_size = sizeof(struct blkif_req);
 991         info->tag_set.driver_data = info;
 992
 993         if (blk_mq_alloc_tag_set(&info->tag_set))
 994                 return -EINVAL;
 995         rq = blk_mq_init_queue(&info->tag_set);
 996         if (IS_ERR(rq)) {
 997                 blk_mq_free_tag_set(&info->tag_set);
 998                 return PTR_ERR(rq);
 999         }
1000
1001         rq->queuedata = info;
1002         info->rq = gd->queue = rq;
1003         info->gd = gd;
1004         info->sector_size = sector_size;
1005         info->physical_sector_size = physical_sector_size;
1006         blkif_set_queue_limits(info);
1007
1008         return 0;
1009 }
1010
1011 static const char *flush_info(struct blkfront_info *info)
1012 {
1013         if (info->feature_flush && info->feature_fua)
1014                 return "barrier: enabled;";
1015         else if (info->feature_flush)
1016                 return "flush diskcache: enabled;";
1017         else
1018                 return "barrier or flush: disabled;";
1019 }
1020
1021 static void xlvbd_flush(struct blkfront_info *info)
1022 {
1023         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1024                               info->feature_fua ? true : false);
1025         pr_info("blkfront: %s: %s %s %s %s %s\n",
1026                 info->gd->disk_name, flush_info(info),
1027                 "persistent grants:", info->feature_persistent ?
1028                 "enabled;" : "disabled;", "indirect descriptors:",
1029                 info->max_indirect_segments ? "enabled;" : "disabled;");
1030 }
1031
1032 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1033 {
1034         int major;
1035         major = BLKIF_MAJOR(vdevice);
1036         *minor = BLKIF_MINOR(vdevice);
1037         switch (major) {
1038                 case XEN_IDE0_MAJOR:
1039                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1040                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1041                                 EMULATED_HD_DISK_MINOR_OFFSET;
1042                         break;
1043                 case XEN_IDE1_MAJOR:
1044                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1045                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1046                                 EMULATED_HD_DISK_MINOR_OFFSET;
1047                         break;
1048                 case XEN_SCSI_DISK0_MAJOR:
1049                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1050                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1051                         break;
1052                 case XEN_SCSI_DISK1_MAJOR:
1053                 case XEN_SCSI_DISK2_MAJOR:
1054                 case XEN_SCSI_DISK3_MAJOR:
1055                 case XEN_SCSI_DISK4_MAJOR:
1056                 case XEN_SCSI_DISK5_MAJOR:
1057                 case XEN_SCSI_DISK6_MAJOR:
1058                 case XEN_SCSI_DISK7_MAJOR:
1059                         *offset = (*minor / PARTS_PER_DISK) +
1060                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1061                                 EMULATED_SD_DISK_NAME_OFFSET;
1062                         *minor = *minor +
1063                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1064                                 EMULATED_SD_DISK_MINOR_OFFSET;
1065                         break;
1066                 case XEN_SCSI_DISK8_MAJOR:
1067                 case XEN_SCSI_DISK9_MAJOR:
1068                 case XEN_SCSI_DISK10_MAJOR:
1069                 case XEN_SCSI_DISK11_MAJOR:
1070                 case XEN_SCSI_DISK12_MAJOR:
1071                 case XEN_SCSI_DISK13_MAJOR:
1072                 case XEN_SCSI_DISK14_MAJOR:
1073                 case XEN_SCSI_DISK15_MAJOR:
1074                         *offset = (*minor / PARTS_PER_DISK) +
1075                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1076                                 EMULATED_SD_DISK_NAME_OFFSET;
1077                         *minor = *minor +
1078                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1079                                 EMULATED_SD_DISK_MINOR_OFFSET;
1080                         break;
1081                 case XENVBD_MAJOR:
1082                         *offset = *minor / PARTS_PER_DISK;
1083                         break;
1084                 default:
1085                         printk(KERN_WARNING "blkfront: your disk configuration is "
1086                                         "incorrect, please use an xvd device instead\n");
1087                         return -ENODEV;
1088         }
1089         return 0;
1090 }
1091
1092 static char *encode_disk_name(char *ptr, unsigned int n)
1093 {
1094         if (n >= 26)
1095                 ptr = encode_disk_name(ptr, n / 26 - 1);
1096         *ptr = 'a' + n % 26;
1097         return ptr + 1;
1098 }
1099
1100 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1101                                struct blkfront_info *info,
1102                                u16 vdisk_info, u16 sector_size,
1103                                unsigned int physical_sector_size)
1104 {
1105         struct gendisk *gd;
1106         int nr_minors = 1;
1107         int err;
1108         unsigned int offset;
1109         int minor;
1110         int nr_parts;
1111         char *ptr;
1112
1113         BUG_ON(info->gd != NULL);
1114         BUG_ON(info->rq != NULL);
1115
1116         if ((info->vdevice>>EXT_SHIFT) > 1) {
1117                 /* this is above the extended range; something is wrong */
1118                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1119                 return -ENODEV;
1120         }
1121
1122         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1123                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1124                 if (err)
1125                         return err;
1126                 nr_parts = PARTS_PER_DISK;
1127         } else {
1128                 minor = BLKIF_MINOR_EXT(info->vdevice);
1129                 nr_parts = PARTS_PER_EXT_DISK;
1130                 offset = minor / nr_parts;
1131                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1132                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1133                                         "emulated IDE disks,\n\t choose an xvd device name"
1134                                         "from xvde on\n", info->vdevice);
1135         }
1136         if (minor >> MINORBITS) {
1137                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1138                         info->vdevice, minor);
1139                 return -ENODEV;
1140         }
1141
1142         if ((minor % nr_parts) == 0)
1143                 nr_minors = nr_parts;
1144
1145         err = xlbd_reserve_minors(minor, nr_minors);
1146         if (err)
1147                 goto out;
1148         err = -ENODEV;
1149
1150         gd = alloc_disk(nr_minors);
1151         if (gd == NULL)
1152                 goto release;
1153
1154         strcpy(gd->disk_name, DEV_NAME);
1155         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1156         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1157         if (nr_minors > 1)
1158                 *ptr = 0;
1159         else
1160                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1161                          "%d", minor & (nr_parts - 1));
1162
1163         gd->major = XENVBD_MAJOR;
1164         gd->first_minor = minor;
1165         gd->fops = &xlvbd_block_fops;
1166         gd->private_data = info;
1167         set_capacity(gd, capacity);
1168
1169         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1170                 del_gendisk(gd);
1171                 goto release;
1172         }
1173
1174         xlvbd_flush(info);
1175
1176         if (vdisk_info & VDISK_READONLY)
1177                 set_disk_ro(gd, 1);
1178
1179         if (vdisk_info & VDISK_REMOVABLE)
1180                 gd->flags |= GENHD_FL_REMOVABLE;
1181
1182         if (vdisk_info & VDISK_CDROM)
1183                 gd->flags |= GENHD_FL_CD;
1184
1185         return 0;
1186
1187  release:
1188         xlbd_release_minors(minor, nr_minors);
1189  out:
1190         return err;
1191 }
1192
1193 static void xlvbd_release_gendisk(struct blkfront_info *info)
1194 {
1195         unsigned int minor, nr_minors, i;
1196         struct blkfront_ring_info *rinfo;
1197
1198         if (info->rq == NULL)
1199                 return;
1200
1201         /* No more blkif_request(). */
1202         blk_mq_stop_hw_queues(info->rq);
1203
1204         for_each_rinfo(info, rinfo, i) {
1205                 /* No more gnttab callback work. */
1206                 gnttab_cancel_free_callback(&rinfo->callback);
1207
1208                 /* Flush gnttab callback work. Must be done with no locks held. */
1209                 flush_work(&rinfo->work);
1210         }
1211
1212         del_gendisk(info->gd);
1213
1214         minor = info->gd->first_minor;
1215         nr_minors = info->gd->minors;
1216         xlbd_release_minors(minor, nr_minors);
1217
1218         blk_cleanup_queue(info->rq);
1219         blk_mq_free_tag_set(&info->tag_set);
1220         info->rq = NULL;
1221
1222         put_disk(info->gd);
1223         info->gd = NULL;
1224 }
1225
1226 /* Already hold rinfo->ring_lock. */
1227 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1228 {
1229         if (!RING_FULL(&rinfo->ring))
1230                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1231 }
1232
1233 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1234 {
1235         unsigned long flags;
1236
1237         spin_lock_irqsave(&rinfo->ring_lock, flags);
1238         kick_pending_request_queues_locked(rinfo);
1239         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1240 }
1241
1242 static void blkif_restart_queue(struct work_struct *work)
1243 {
1244         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1245
1246         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1247                 kick_pending_request_queues(rinfo);
1248 }
1249
1250 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1251 {
1252         struct grant *persistent_gnt, *n;
1253         struct blkfront_info *info = rinfo->dev_info;
1254         int i, j, segs;
1255
1256         /*
1257          * Remove indirect pages, this only happens when using indirect
1258          * descriptors but not persistent grants
1259          */
1260         if (!list_empty(&rinfo->indirect_pages)) {
1261                 struct page *indirect_page, *n;
1262
1263                 BUG_ON(info->feature_persistent);
1264                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1265                         list_del(&indirect_page->lru);
1266                         __free_page(indirect_page);
1267                 }
1268         }
1269
1270         /* Remove all persistent grants. */
1271         if (!list_empty(&rinfo->grants)) {
1272                 list_for_each_entry_safe(persistent_gnt, n,
1273                                          &rinfo->grants, node) {
1274                         list_del(&persistent_gnt->node);
1275                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1276                                 gnttab_end_foreign_access(persistent_gnt->gref,
1277                                                           0, 0UL);
1278                                 rinfo->persistent_gnts_c--;
1279                         }
1280                         if (info->feature_persistent)
1281                                 __free_page(persistent_gnt->page);
1282                         kfree(persistent_gnt);
1283                 }
1284         }
1285         BUG_ON(rinfo->persistent_gnts_c != 0);
1286
1287         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1288                 /*
1289                  * Clear persistent grants present in requests already
1290                  * on the shared ring
1291                  */
1292                 if (!rinfo->shadow[i].request)
1293                         goto free_shadow;
1294
1295                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1296                        rinfo->shadow[i].req.u.indirect.nr_segments :
1297                        rinfo->shadow[i].req.u.rw.nr_segments;
1298                 for (j = 0; j < segs; j++) {
1299                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1300                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1301                         if (info->feature_persistent)
1302                                 __free_page(persistent_gnt->page);
1303                         kfree(persistent_gnt);
1304                 }
1305
1306                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1307                         /*
1308                          * If this is not an indirect operation don't try to
1309                          * free indirect segments
1310                          */
1311                         goto free_shadow;
1312
1313                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1314                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1315                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1316                         __free_page(persistent_gnt->page);
1317                         kfree(persistent_gnt);
1318                 }
1319
1320 free_shadow:
1321                 kvfree(rinfo->shadow[i].grants_used);
1322                 rinfo->shadow[i].grants_used = NULL;
1323                 kvfree(rinfo->shadow[i].indirect_grants);
1324                 rinfo->shadow[i].indirect_grants = NULL;
1325                 kvfree(rinfo->shadow[i].sg);
1326                 rinfo->shadow[i].sg = NULL;
1327         }
1328
1329         /* No more gnttab callback work. */
1330         gnttab_cancel_free_callback(&rinfo->callback);
1331
1332         /* Flush gnttab callback work. Must be done with no locks held. */
1333         flush_work(&rinfo->work);
1334
1335         /* Free resources associated with old device channel. */
1336         for (i = 0; i < info->nr_ring_pages; i++) {
1337                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1338                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1339                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1340                 }
1341         }
1342         free_pages((unsigned long)rinfo->ring.sring, get_order(info->nr_ring_pages * XEN_PAGE_SIZE));
1343         rinfo->ring.sring = NULL;
1344
1345         if (rinfo->irq)
1346                 unbind_from_irqhandler(rinfo->irq, rinfo);
1347         rinfo->evtchn = rinfo->irq = 0;
1348 }
1349
1350 static void blkif_free(struct blkfront_info *info, int suspend)
1351 {
1352         unsigned int i;
1353         struct blkfront_ring_info *rinfo;
1354
1355         /* Prevent new requests being issued until we fix things up. */
1356         info->connected = suspend ?
1357                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1358         /* No more blkif_request(). */
1359         if (info->rq)
1360                 blk_mq_stop_hw_queues(info->rq);
1361
1362         for_each_rinfo(info, rinfo, i)
1363                 blkif_free_ring(rinfo);
1364
1365         kvfree(info->rinfo);
1366         info->rinfo = NULL;
1367         info->nr_rings = 0;
1368 }
1369
1370 struct copy_from_grant {
1371         const struct blk_shadow *s;
1372         unsigned int grant_idx;
1373         unsigned int bvec_offset;
1374         char *bvec_data;
1375 };
1376
1377 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1378                                   unsigned int len, void *data)
1379 {
1380         struct copy_from_grant *info = data;
1381         char *shared_data;
1382         /* Convenient aliases */
1383         const struct blk_shadow *s = info->s;
1384
1385         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1386
1387         memcpy(info->bvec_data + info->bvec_offset,
1388                shared_data + offset, len);
1389
1390         info->bvec_offset += len;
1391         info->grant_idx++;
1392
1393         kunmap_atomic(shared_data);
1394 }
1395
1396 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1397 {
1398         switch (rsp)
1399         {
1400         case BLKIF_RSP_OKAY:
1401                 return REQ_DONE;
1402         case BLKIF_RSP_EOPNOTSUPP:
1403                 return REQ_EOPNOTSUPP;
1404         case BLKIF_RSP_ERROR:
1405                 /* Fallthrough. */
1406         default:
1407                 return REQ_ERROR;
1408         }
1409 }
1410
1411 /*
1412  * Get the final status of the block request based on two ring response
1413  */
1414 static int blkif_get_final_status(enum blk_req_status s1,
1415                                   enum blk_req_status s2)
1416 {
1417         BUG_ON(s1 == REQ_WAITING);
1418         BUG_ON(s2 == REQ_WAITING);
1419
1420         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1421                 return BLKIF_RSP_ERROR;
1422         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1423                 return BLKIF_RSP_EOPNOTSUPP;
1424         return BLKIF_RSP_OKAY;
1425 }
1426
1427 static bool blkif_completion(unsigned long *id,
1428                              struct blkfront_ring_info *rinfo,
1429                              struct blkif_response *bret)
1430 {
1431         int i = 0;
1432         struct scatterlist *sg;
1433         int num_sg, num_grant;
1434         struct blkfront_info *info = rinfo->dev_info;
1435         struct blk_shadow *s = &rinfo->shadow[*id];
1436         struct copy_from_grant data = {
1437                 .grant_idx = 0,
1438         };
1439
1440         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1441                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1442
1443         /* The I/O request may be split in two. */
1444         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1445                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1446
1447                 /* Keep the status of the current response in shadow. */
1448                 s->status = blkif_rsp_to_req_status(bret->status);
1449
1450                 /* Wait the second response if not yet here. */
1451                 if (s2->status == REQ_WAITING)
1452                         return false;
1453
1454                 bret->status = blkif_get_final_status(s->status,
1455                                                       s2->status);
1456
1457                 /*
1458                  * All the grants is stored in the first shadow in order
1459                  * to make the completion code simpler.
1460                  */
1461                 num_grant += s2->req.u.rw.nr_segments;
1462
1463                 /*
1464                  * The two responses may not come in order. Only the
1465                  * first request will store the scatter-gather list.
1466                  */
1467                 if (s2->num_sg != 0) {
1468                         /* Update "id" with the ID of the first response. */
1469                         *id = s->associated_id;
1470                         s = s2;
1471                 }
1472
1473                 /*
1474                  * We don't need anymore the second request, so recycling
1475                  * it now.
1476                  */
1477                 if (add_id_to_freelist(rinfo, s->associated_id))
1478                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1479                              info->gd->disk_name, s->associated_id);
1480         }
1481
1482         data.s = s;
1483         num_sg = s->num_sg;
1484
1485         if (bret->operation == BLKIF_OP_READ && info->feature_persistent) {
1486                 for_each_sg(s->sg, sg, num_sg, i) {
1487                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1488
1489                         data.bvec_offset = sg->offset;
1490                         data.bvec_data = kmap_atomic(sg_page(sg));
1491
1492                         gnttab_foreach_grant_in_range(sg_page(sg),
1493                                                       sg->offset,
1494                                                       sg->length,
1495                                                       blkif_copy_from_grant,
1496                                                       &data);
1497
1498                         kunmap_atomic(data.bvec_data);
1499                 }
1500         }
1501         /* Add the persistent grant into the list of free grants */
1502         for (i = 0; i < num_grant; i++) {
1503                 if (gnttab_query_foreign_access(s->grants_used[i]->gref)) {
1504                         /*
1505                          * If the grant is still mapped by the backend (the
1506                          * backend has chosen to make this grant persistent)
1507                          * we add it at the head of the list, so it will be
1508                          * reused first.
1509                          */
1510                         if (!info->feature_persistent)
1511                                 pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1512                                                      s->grants_used[i]->gref);
1513                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1514                         rinfo->persistent_gnts_c++;
1515                 } else {
1516                         /*
1517                          * If the grant is not mapped by the backend we end the
1518                          * foreign access and add it to the tail of the list,
1519                          * so it will not be picked again unless we run out of
1520                          * persistent grants.
1521                          */
1522                         gnttab_end_foreign_access(s->grants_used[i]->gref, 0, 0UL);
1523                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1524                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1525                 }
1526         }
1527         if (s->req.operation == BLKIF_OP_INDIRECT) {
1528                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1529                         if (gnttab_query_foreign_access(s->indirect_grants[i]->gref)) {
1530                                 if (!info->feature_persistent)
1531                                         pr_alert_ratelimited("backed has not unmapped grant: %u\n",
1532                                                              s->indirect_grants[i]->gref);
1533                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1534                                 rinfo->persistent_gnts_c++;
1535                         } else {
1536                                 struct page *indirect_page;
1537
1538                                 gnttab_end_foreign_access(s->indirect_grants[i]->gref, 0, 0UL);
1539                                 /*
1540                                  * Add the used indirect page back to the list of
1541                                  * available pages for indirect grefs.
1542                                  */
1543                                 if (!info->feature_persistent) {
1544                                         indirect_page = s->indirect_grants[i]->page;
1545                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1546                                 }
1547                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1548                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1549                         }
1550                 }
1551         }
1552
1553         return true;
1554 }
1555
1556 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1557 {
1558         struct request *req;
1559         struct blkif_response *bret;
1560         RING_IDX i, rp;
1561         unsigned long flags;
1562         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1563         struct blkfront_info *info = rinfo->dev_info;
1564
1565         if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
1566                 return IRQ_HANDLED;
1567
1568         spin_lock_irqsave(&rinfo->ring_lock, flags);
1569  again:
1570         rp = rinfo->ring.sring->rsp_prod;
1571         rmb(); /* Ensure we see queued responses up to 'rp'. */
1572
1573         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1574                 unsigned long id;
1575
1576                 bret = RING_GET_RESPONSE(&rinfo->ring, i);
1577                 id   = bret->id;
1578                 /*
1579                  * The backend has messed up and given us an id that we would
1580                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1581                  * look in get_id_from_freelist.
1582                  */
1583                 if (id >= BLK_RING_SIZE(info)) {
1584                         WARN(1, "%s: response to %s has incorrect id (%ld)\n",
1585                              info->gd->disk_name, op_name(bret->operation), id);
1586                         /* We can't safely get the 'struct request' as
1587                          * the id is busted. */
1588                         continue;
1589                 }
1590                 req  = rinfo->shadow[id].request;
1591
1592                 if (bret->operation != BLKIF_OP_DISCARD) {
1593                         /*
1594                          * We may need to wait for an extra response if the
1595                          * I/O request is split in 2
1596                          */
1597                         if (!blkif_completion(&id, rinfo, bret))
1598                                 continue;
1599                 }
1600
1601                 if (add_id_to_freelist(rinfo, id)) {
1602                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1603                              info->gd->disk_name, op_name(bret->operation), id);
1604                         continue;
1605                 }
1606
1607                 if (bret->status == BLKIF_RSP_OKAY)
1608                         blkif_req(req)->error = BLK_STS_OK;
1609                 else
1610                         blkif_req(req)->error = BLK_STS_IOERR;
1611
1612                 switch (bret->operation) {
1613                 case BLKIF_OP_DISCARD:
1614                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1615                                 struct request_queue *rq = info->rq;
1616                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1617                                            info->gd->disk_name, op_name(bret->operation));
1618                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619                                 info->feature_discard = 0;
1620                                 info->feature_secdiscard = 0;
1621                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1622                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1623                         }
1624                         break;
1625                 case BLKIF_OP_FLUSH_DISKCACHE:
1626                 case BLKIF_OP_WRITE_BARRIER:
1627                         if (unlikely(bret->status == BLKIF_RSP_EOPNOTSUPP)) {
1628                                 printk(KERN_WARNING "blkfront: %s: %s op failed\n",
1629                                        info->gd->disk_name, op_name(bret->operation));
1630                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1631                         }
1632                         if (unlikely(bret->status == BLKIF_RSP_ERROR &&
1633                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1634                                 printk(KERN_WARNING "blkfront: %s: empty %s op failed\n",
1635                                        info->gd->disk_name, op_name(bret->operation));
1636                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1637                         }
1638                         if (unlikely(blkif_req(req)->error)) {
1639                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1640                                         blkif_req(req)->error = BLK_STS_OK;
1641                                 info->feature_fua = 0;
1642                                 info->feature_flush = 0;
1643                                 xlvbd_flush(info);
1644                         }
1645                         /* fall through */
1646                 case BLKIF_OP_READ:
1647                 case BLKIF_OP_WRITE:
1648                         if (unlikely(bret->status != BLKIF_RSP_OKAY))
1649                                 dev_dbg(&info->xbdev->dev, "Bad return from blkdev data "
1650                                         "request: %x\n", bret->status);
1651
1652                         break;
1653                 default:
1654                         BUG();
1655                 }
1656
1657                 blk_mq_complete_request(req);
1658         }
1659
1660         rinfo->ring.rsp_cons = i;
1661
1662         if (i != rinfo->ring.req_prod_pvt) {
1663                 int more_to_do;
1664                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1665                 if (more_to_do)
1666                         goto again;
1667         } else
1668                 rinfo->ring.sring->rsp_event = i + 1;
1669
1670         kick_pending_request_queues_locked(rinfo);
1671
1672         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1673
1674         return IRQ_HANDLED;
1675 }
1676
1677
1678 static int setup_blkring(struct xenbus_device *dev,
1679                          struct blkfront_ring_info *rinfo)
1680 {
1681         struct blkif_sring *sring;
1682         int err, i;
1683         struct blkfront_info *info = rinfo->dev_info;
1684         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1685         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1686
1687         for (i = 0; i < info->nr_ring_pages; i++)
1688                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1689
1690         sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
1691                                                        get_order(ring_size));
1692         if (!sring) {
1693                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1694                 return -ENOMEM;
1695         }
1696         SHARED_RING_INIT(sring);
1697         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1698
1699         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1700         if (err < 0) {
1701                 free_pages((unsigned long)sring, get_order(ring_size));
1702                 rinfo->ring.sring = NULL;
1703                 goto fail;
1704         }
1705         for (i = 0; i < info->nr_ring_pages; i++)
1706                 rinfo->ring_ref[i] = gref[i];
1707
1708         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1709         if (err)
1710                 goto fail;
1711
1712         err = bind_evtchn_to_irqhandler(rinfo->evtchn, blkif_interrupt, 0,
1713                                         "blkif", rinfo);
1714         if (err <= 0) {
1715                 xenbus_dev_fatal(dev, err,
1716                                  "bind_evtchn_to_irqhandler failed");
1717                 goto fail;
1718         }
1719         rinfo->irq = err;
1720
1721         return 0;
1722 fail:
1723         blkif_free(info, 0);
1724         return err;
1725 }
1726
1727 /*
1728  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1729  * ring buffer may have multi pages depending on ->nr_ring_pages.
1730  */
1731 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1732                                 struct blkfront_ring_info *rinfo, const char *dir)
1733 {
1734         int err;
1735         unsigned int i;
1736         const char *message = NULL;
1737         struct blkfront_info *info = rinfo->dev_info;
1738
1739         if (info->nr_ring_pages == 1) {
1740                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1741                 if (err) {
1742                         message = "writing ring-ref";
1743                         goto abort_transaction;
1744                 }
1745         } else {
1746                 for (i = 0; i < info->nr_ring_pages; i++) {
1747                         char ring_ref_name[RINGREF_NAME_LEN];
1748
1749                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1750                         err = xenbus_printf(xbt, dir, ring_ref_name,
1751                                             "%u", rinfo->ring_ref[i]);
1752                         if (err) {
1753                                 message = "writing ring-ref";
1754                                 goto abort_transaction;
1755                         }
1756                 }
1757         }
1758
1759         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1760         if (err) {
1761                 message = "writing event-channel";
1762                 goto abort_transaction;
1763         }
1764
1765         return 0;
1766
1767 abort_transaction:
1768         xenbus_transaction_end(xbt, 1);
1769         if (message)
1770                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1771
1772         return err;
1773 }
1774
1775 static void free_info(struct blkfront_info *info)
1776 {
1777         list_del(&info->info_list);
1778         kfree(info);
1779 }
1780
1781 /* Common code used when first setting up, and when resuming. */
1782 static int talk_to_blkback(struct xenbus_device *dev,
1783                            struct blkfront_info *info)
1784 {
1785         const char *message = NULL;
1786         struct xenbus_transaction xbt;
1787         int err;
1788         unsigned int i, max_page_order;
1789         unsigned int ring_page_order;
1790         struct blkfront_ring_info *rinfo;
1791
1792         if (!info)
1793                 return -ENODEV;
1794
1795         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1796                                               "max-ring-page-order", 0);
1797         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1798         info->nr_ring_pages = 1 << ring_page_order;
1799
1800         err = negotiate_mq(info);
1801         if (err)
1802                 goto destroy_blkring;
1803
1804         for_each_rinfo(info, rinfo, i) {
1805                 /* Create shared ring, alloc event channel. */
1806                 err = setup_blkring(dev, rinfo);
1807                 if (err)
1808                         goto destroy_blkring;
1809         }
1810
1811 again:
1812         err = xenbus_transaction_start(&xbt);
1813         if (err) {
1814                 xenbus_dev_fatal(dev, err, "starting transaction");
1815                 goto destroy_blkring;
1816         }
1817
1818         if (info->nr_ring_pages > 1) {
1819                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1820                                     ring_page_order);
1821                 if (err) {
1822                         message = "writing ring-page-order";
1823                         goto abort_transaction;
1824                 }
1825         }
1826
1827         /* We already got the number of queues/rings in _probe */
1828         if (info->nr_rings == 1) {
1829                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1830                 if (err)
1831                         goto destroy_blkring;
1832         } else {
1833                 char *path;
1834                 size_t pathsize;
1835
1836                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1837                                     info->nr_rings);
1838                 if (err) {
1839                         message = "writing multi-queue-num-queues";
1840                         goto abort_transaction;
1841                 }
1842
1843                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1844                 path = kmalloc(pathsize, GFP_KERNEL);
1845                 if (!path) {
1846                         err = -ENOMEM;
1847                         message = "ENOMEM while writing ring references";
1848                         goto abort_transaction;
1849                 }
1850
1851                 for_each_rinfo(info, rinfo, i) {
1852                         memset(path, 0, pathsize);
1853                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1854                         err = write_per_ring_nodes(xbt, rinfo, path);
1855                         if (err) {
1856                                 kfree(path);
1857                                 goto destroy_blkring;
1858                         }
1859                 }
1860                 kfree(path);
1861         }
1862         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1863                             XEN_IO_PROTO_ABI_NATIVE);
1864         if (err) {
1865                 message = "writing protocol";
1866                 goto abort_transaction;
1867         }
1868         err = xenbus_printf(xbt, dev->nodename,
1869                             "feature-persistent", "%u", 1);
1870         if (err)
1871                 dev_warn(&dev->dev,
1872                          "writing persistent grants feature to xenbus");
1873
1874         err = xenbus_transaction_end(xbt, 0);
1875         if (err) {
1876                 if (err == -EAGAIN)
1877                         goto again;
1878                 xenbus_dev_fatal(dev, err, "completing transaction");
1879                 goto destroy_blkring;
1880         }
1881
1882         for_each_rinfo(info, rinfo, i) {
1883                 unsigned int j;
1884
1885                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1886                         rinfo->shadow[j].req.u.rw.id = j + 1;
1887                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1888         }
1889         xenbus_switch_state(dev, XenbusStateInitialised);
1890
1891         return 0;
1892
1893  abort_transaction:
1894         xenbus_transaction_end(xbt, 1);
1895         if (message)
1896                 xenbus_dev_fatal(dev, err, "%s", message);
1897  destroy_blkring:
1898         blkif_free(info, 0);
1899
1900         mutex_lock(&blkfront_mutex);
1901         free_info(info);
1902         mutex_unlock(&blkfront_mutex);
1903
1904         dev_set_drvdata(&dev->dev, NULL);
1905
1906         return err;
1907 }
1908
1909 static int negotiate_mq(struct blkfront_info *info)
1910 {
1911         unsigned int backend_max_queues;
1912         unsigned int i;
1913         struct blkfront_ring_info *rinfo;
1914
1915         BUG_ON(info->nr_rings);
1916
1917         /* Check if backend supports multiple queues. */
1918         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1919                                                   "multi-queue-max-queues", 1);
1920         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1921         /* We need at least one ring. */
1922         if (!info->nr_rings)
1923                 info->nr_rings = 1;
1924
1925         info->rinfo_size = struct_size(info->rinfo, shadow,
1926                                        BLK_RING_SIZE(info));
1927         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1928         if (!info->rinfo) {
1929                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1930                 info->nr_rings = 0;
1931                 return -ENOMEM;
1932         }
1933
1934         for_each_rinfo(info, rinfo, i) {
1935                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1936                 INIT_LIST_HEAD(&rinfo->grants);
1937                 rinfo->dev_info = info;
1938                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1939                 spin_lock_init(&rinfo->ring_lock);
1940         }
1941         return 0;
1942 }
1943 /**
1944  * Entry point to this code when a new device is created.  Allocate the basic
1945  * structures and the ring buffer for communication with the backend, and
1946  * inform the backend of the appropriate details for those.  Switch to
1947  * Initialised state.
1948  */
1949 static int blkfront_probe(struct xenbus_device *dev,
1950                           const struct xenbus_device_id *id)
1951 {
1952         int err, vdevice;
1953         struct blkfront_info *info;
1954
1955         /* FIXME: Use dynamic device id if this is not set. */
1956         err = xenbus_scanf(XBT_NIL, dev->nodename,
1957                            "virtual-device", "%i", &vdevice);
1958         if (err != 1) {
1959                 /* go looking in the extended area instead */
1960                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1961                                    "%i", &vdevice);
1962                 if (err != 1) {
1963                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1964                         return err;
1965                 }
1966         }
1967
1968         if (xen_hvm_domain()) {
1969                 char *type;
1970                 int len;
1971                 /* no unplug has been done: do not hook devices != xen vbds */
1972                 if (xen_has_pv_and_legacy_disk_devices()) {
1973                         int major;
1974
1975                         if (!VDEV_IS_EXTENDED(vdevice))
1976                                 major = BLKIF_MAJOR(vdevice);
1977                         else
1978                                 major = XENVBD_MAJOR;
1979
1980                         if (major != XENVBD_MAJOR) {
1981                                 printk(KERN_INFO
1982                                                 "%s: HVM does not support vbd %d as xen block device\n",
1983                                                 __func__, vdevice);
1984                                 return -ENODEV;
1985                         }
1986                 }
1987                 /* do not create a PV cdrom device if we are an HVM guest */
1988                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
1989                 if (IS_ERR(type))
1990                         return -ENODEV;
1991                 if (strncmp(type, "cdrom", 5) == 0) {
1992                         kfree(type);
1993                         return -ENODEV;
1994                 }
1995                 kfree(type);
1996         }
1997         info = kzalloc(sizeof(*info), GFP_KERNEL);
1998         if (!info) {
1999                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2000                 return -ENOMEM;
2001         }
2002
2003         info->xbdev = dev;
2004
2005         mutex_init(&info->mutex);
2006         info->vdevice = vdevice;
2007         info->connected = BLKIF_STATE_DISCONNECTED;
2008
2009         /* Front end dir is a number, which is used as the id. */
2010         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2011         dev_set_drvdata(&dev->dev, info);
2012
2013         mutex_lock(&blkfront_mutex);
2014         list_add(&info->info_list, &info_list);
2015         mutex_unlock(&blkfront_mutex);
2016
2017         return 0;
2018 }
2019
2020 static int blkif_recover(struct blkfront_info *info)
2021 {
2022         unsigned int r_index;
2023         struct request *req, *n;
2024         int rc;
2025         struct bio *bio;
2026         unsigned int segs;
2027         struct blkfront_ring_info *rinfo;
2028
2029         blkfront_gather_backend_features(info);
2030         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2031         blkif_set_queue_limits(info);
2032         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2033         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2034
2035         for_each_rinfo(info, rinfo, r_index) {
2036                 rc = blkfront_setup_indirect(rinfo);
2037                 if (rc)
2038                         return rc;
2039         }
2040         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2041
2042         /* Now safe for us to use the shared ring */
2043         info->connected = BLKIF_STATE_CONNECTED;
2044
2045         for_each_rinfo(info, rinfo, r_index) {
2046                 /* Kick any other new requests queued since we resumed */
2047                 kick_pending_request_queues(rinfo);
2048         }
2049
2050         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2051                 /* Requeue pending requests (flush or discard) */
2052                 list_del_init(&req->queuelist);
2053                 BUG_ON(req->nr_phys_segments > segs);
2054                 blk_mq_requeue_request(req, false);
2055         }
2056         blk_mq_start_stopped_hw_queues(info->rq, true);
2057         blk_mq_kick_requeue_list(info->rq);
2058
2059         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2060                 /* Traverse the list of pending bios and re-queue them */
2061                 submit_bio(bio);
2062         }
2063
2064         return 0;
2065 }
2066
2067 /**
2068  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2069  * driver restart.  We tear down our blkif structure and recreate it, but
2070  * leave the device-layer structures intact so that this is transparent to the
2071  * rest of the kernel.
2072  */
2073 static int blkfront_resume(struct xenbus_device *dev)
2074 {
2075         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2076         int err = 0;
2077         unsigned int i, j;
2078         struct blkfront_ring_info *rinfo;
2079
2080         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2081
2082         bio_list_init(&info->bio_list);
2083         INIT_LIST_HEAD(&info->requests);
2084         for_each_rinfo(info, rinfo, i) {
2085                 struct bio_list merge_bio;
2086                 struct blk_shadow *shadow = rinfo->shadow;
2087
2088                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2089                         /* Not in use? */
2090                         if (!shadow[j].request)
2091                                 continue;
2092
2093                         /*
2094                          * Get the bios in the request so we can re-queue them.
2095                          */
2096                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2097                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2098                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2099                             shadow[j].request->cmd_flags & REQ_FUA) {
2100                                 /*
2101                                  * Flush operations don't contain bios, so
2102                                  * we need to requeue the whole request
2103                                  *
2104                                  * XXX: but this doesn't make any sense for a
2105                                  * write with the FUA flag set..
2106                                  */
2107                                 list_add(&shadow[j].request->queuelist, &info->requests);
2108                                 continue;
2109                         }
2110                         merge_bio.head = shadow[j].request->bio;
2111                         merge_bio.tail = shadow[j].request->biotail;
2112                         bio_list_merge(&info->bio_list, &merge_bio);
2113                         shadow[j].request->bio = NULL;
2114                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2115                 }
2116         }
2117
2118         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2119
2120         err = talk_to_blkback(dev, info);
2121         if (!err)
2122                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2123
2124         /*
2125          * We have to wait for the backend to switch to
2126          * connected state, since we want to read which
2127          * features it supports.
2128          */
2129
2130         return err;
2131 }
2132
2133 static void blkfront_closing(struct blkfront_info *info)
2134 {
2135         struct xenbus_device *xbdev = info->xbdev;
2136         struct block_device *bdev = NULL;
2137
2138         mutex_lock(&info->mutex);
2139
2140         if (xbdev->state == XenbusStateClosing) {
2141                 mutex_unlock(&info->mutex);
2142                 return;
2143         }
2144
2145         if (info->gd)
2146                 bdev = bdget_disk(info->gd, 0);
2147
2148         mutex_unlock(&info->mutex);
2149
2150         if (!bdev) {
2151                 xenbus_frontend_closed(xbdev);
2152                 return;
2153         }
2154
2155         mutex_lock(&bdev->bd_mutex);
2156
2157         if (bdev->bd_openers) {
2158                 xenbus_dev_error(xbdev, -EBUSY,
2159                                  "Device in use; refusing to close");
2160                 xenbus_switch_state(xbdev, XenbusStateClosing);
2161         } else {
2162                 xlvbd_release_gendisk(info);
2163                 xenbus_frontend_closed(xbdev);
2164         }
2165
2166         mutex_unlock(&bdev->bd_mutex);
2167         bdput(bdev);
2168 }
2169
2170 static void blkfront_setup_discard(struct blkfront_info *info)
2171 {
2172         int err;
2173         unsigned int discard_granularity;
2174         unsigned int discard_alignment;
2175
2176         info->feature_discard = 1;
2177         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2178                 "discard-granularity", "%u", &discard_granularity,
2179                 "discard-alignment", "%u", &discard_alignment,
2180                 NULL);
2181         if (!err) {
2182                 info->discard_granularity = discard_granularity;
2183                 info->discard_alignment = discard_alignment;
2184         }
2185         info->feature_secdiscard =
2186                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2187                                        0);
2188 }
2189
2190 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2191 {
2192         unsigned int psegs, grants;
2193         int err, i;
2194         struct blkfront_info *info = rinfo->dev_info;
2195
2196         if (info->max_indirect_segments == 0) {
2197                 if (!HAS_EXTRA_REQ)
2198                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2199                 else {
2200                         /*
2201                          * When an extra req is required, the maximum
2202                          * grants supported is related to the size of the
2203                          * Linux block segment.
2204                          */
2205                         grants = GRANTS_PER_PSEG;
2206                 }
2207         }
2208         else
2209                 grants = info->max_indirect_segments;
2210         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2211
2212         err = fill_grant_buffer(rinfo,
2213                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2214         if (err)
2215                 goto out_of_memory;
2216
2217         if (!info->feature_persistent && info->max_indirect_segments) {
2218                 /*
2219                  * We are using indirect descriptors but not persistent
2220                  * grants, we need to allocate a set of pages that can be
2221                  * used for mapping indirect grefs
2222                  */
2223                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2224
2225                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2226                 for (i = 0; i < num; i++) {
2227                         struct page *indirect_page = alloc_page(GFP_NOIO);
2228                         if (!indirect_page)
2229                                 goto out_of_memory;
2230                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2231                 }
2232         }
2233
2234         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2235                 rinfo->shadow[i].grants_used =
2236                         kvcalloc(grants,
2237                                  sizeof(rinfo->shadow[i].grants_used[0]),
2238                                  GFP_NOIO);
2239                 rinfo->shadow[i].sg = kvcalloc(psegs,
2240                                                sizeof(rinfo->shadow[i].sg[0]),
2241                                                GFP_NOIO);
2242                 if (info->max_indirect_segments)
2243                         rinfo->shadow[i].indirect_grants =
2244                                 kvcalloc(INDIRECT_GREFS(grants),
2245                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2246                                          GFP_NOIO);
2247                 if ((rinfo->shadow[i].grants_used == NULL) ||
2248                         (rinfo->shadow[i].sg == NULL) ||
2249                      (info->max_indirect_segments &&
2250                      (rinfo->shadow[i].indirect_grants == NULL)))
2251                         goto out_of_memory;
2252                 sg_init_table(rinfo->shadow[i].sg, psegs);
2253         }
2254
2255
2256         return 0;
2257
2258 out_of_memory:
2259         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2260                 kvfree(rinfo->shadow[i].grants_used);
2261                 rinfo->shadow[i].grants_used = NULL;
2262                 kvfree(rinfo->shadow[i].sg);
2263                 rinfo->shadow[i].sg = NULL;
2264                 kvfree(rinfo->shadow[i].indirect_grants);
2265                 rinfo->shadow[i].indirect_grants = NULL;
2266         }
2267         if (!list_empty(&rinfo->indirect_pages)) {
2268                 struct page *indirect_page, *n;
2269                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2270                         list_del(&indirect_page->lru);
2271                         __free_page(indirect_page);
2272                 }
2273         }
2274         return -ENOMEM;
2275 }
2276
2277 /*
2278  * Gather all backend feature-*
2279  */
2280 static void blkfront_gather_backend_features(struct blkfront_info *info)
2281 {
2282         unsigned int indirect_segments;
2283
2284         info->feature_flush = 0;
2285         info->feature_fua = 0;
2286
2287         /*
2288          * If there's no "feature-barrier" defined, then it means
2289          * we're dealing with a very old backend which writes
2290          * synchronously; nothing to do.
2291          *
2292          * If there are barriers, then we use flush.
2293          */
2294         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2295                 info->feature_flush = 1;
2296                 info->feature_fua = 1;
2297         }
2298
2299         /*
2300          * And if there is "feature-flush-cache" use that above
2301          * barriers.
2302          */
2303         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2304                                  0)) {
2305                 info->feature_flush = 1;
2306                 info->feature_fua = 0;
2307         }
2308
2309         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2310                 blkfront_setup_discard(info);
2311
2312         info->feature_persistent =
2313                 !!xenbus_read_unsigned(info->xbdev->otherend,
2314                                        "feature-persistent", 0);
2315
2316         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2317                                         "feature-max-indirect-segments", 0);
2318         if (indirect_segments > xen_blkif_max_segments)
2319                 indirect_segments = xen_blkif_max_segments;
2320         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2321                 indirect_segments = 0;
2322         info->max_indirect_segments = indirect_segments;
2323
2324         if (info->feature_persistent) {
2325                 mutex_lock(&blkfront_mutex);
2326                 schedule_delayed_work(&blkfront_work, HZ * 10);
2327                 mutex_unlock(&blkfront_mutex);
2328         }
2329 }
2330
2331 /*
2332  * Invoked when the backend is finally 'ready' (and has told produced
2333  * the details about the physical device - #sectors, size, etc).
2334  */
2335 static void blkfront_connect(struct blkfront_info *info)
2336 {
2337         unsigned long long sectors;
2338         unsigned long sector_size;
2339         unsigned int physical_sector_size;
2340         unsigned int binfo;
2341         int err, i;
2342         struct blkfront_ring_info *rinfo;
2343
2344         switch (info->connected) {
2345         case BLKIF_STATE_CONNECTED:
2346                 /*
2347                  * Potentially, the back-end may be signalling
2348                  * a capacity change; update the capacity.
2349                  */
2350                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2351                                    "sectors", "%Lu", &sectors);
2352                 if (XENBUS_EXIST_ERR(err))
2353                         return;
2354                 printk(KERN_INFO "Setting capacity to %Lu\n",
2355                        sectors);
2356                 set_capacity_revalidate_and_notify(info->gd, sectors, true);
2357
2358                 return;
2359         case BLKIF_STATE_SUSPENDED:
2360                 /*
2361                  * If we are recovering from suspension, we need to wait
2362                  * for the backend to announce it's features before
2363                  * reconnecting, at least we need to know if the backend
2364                  * supports indirect descriptors, and how many.
2365                  */
2366                 blkif_recover(info);
2367                 return;
2368
2369         default:
2370                 break;
2371         }
2372
2373         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2374                 __func__, info->xbdev->otherend);
2375
2376         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2377                             "sectors", "%llu", &sectors,
2378                             "info", "%u", &binfo,
2379                             "sector-size", "%lu", &sector_size,
2380                             NULL);
2381         if (err) {
2382                 xenbus_dev_fatal(info->xbdev, err,
2383                                  "reading backend fields at %s",
2384                                  info->xbdev->otherend);
2385                 return;
2386         }
2387
2388         /*
2389          * physcial-sector-size is a newer field, so old backends may not
2390          * provide this. Assume physical sector size to be the same as
2391          * sector_size in that case.
2392          */
2393         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2394                                                     "physical-sector-size",
2395                                                     sector_size);
2396         blkfront_gather_backend_features(info);
2397         for_each_rinfo(info, rinfo, i) {
2398                 err = blkfront_setup_indirect(rinfo);
2399                 if (err) {
2400                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2401                                          info->xbdev->otherend);
2402                         blkif_free(info, 0);
2403                         break;
2404                 }
2405         }
2406
2407         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2408                                   physical_sector_size);
2409         if (err) {
2410                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2411                                  info->xbdev->otherend);
2412                 goto fail;
2413         }
2414
2415         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2416
2417         /* Kick pending requests. */
2418         info->connected = BLKIF_STATE_CONNECTED;
2419         for_each_rinfo(info, rinfo, i)
2420                 kick_pending_request_queues(rinfo);
2421
2422         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2423
2424         info->is_ready = 1;
2425         return;
2426
2427 fail:
2428         blkif_free(info, 0);
2429         return;
2430 }
2431
2432 /**
2433  * Callback received when the backend's state changes.
2434  */
2435 static void blkback_changed(struct xenbus_device *dev,
2436                             enum xenbus_state backend_state)
2437 {
2438         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2439
2440         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2441
2442         switch (backend_state) {
2443         case XenbusStateInitWait:
2444                 if (dev->state != XenbusStateInitialising)
2445                         break;
2446                 if (talk_to_blkback(dev, info))
2447                         break;
2448         case XenbusStateInitialising:
2449         case XenbusStateInitialised:
2450         case XenbusStateReconfiguring:
2451         case XenbusStateReconfigured:
2452         case XenbusStateUnknown:
2453                 break;
2454
2455         case XenbusStateConnected:
2456                 /*
2457                  * talk_to_blkback sets state to XenbusStateInitialised
2458                  * and blkfront_connect sets it to XenbusStateConnected
2459                  * (if connection went OK).
2460                  *
2461                  * If the backend (or toolstack) decides to poke at backend
2462                  * state (and re-trigger the watch by setting the state repeatedly
2463                  * to XenbusStateConnected (4)) we need to deal with this.
2464                  * This is allowed as this is used to communicate to the guest
2465                  * that the size of disk has changed!
2466                  */
2467                 if ((dev->state != XenbusStateInitialised) &&
2468                     (dev->state != XenbusStateConnected)) {
2469                         if (talk_to_blkback(dev, info))
2470                                 break;
2471                 }
2472
2473                 blkfront_connect(info);
2474                 break;
2475
2476         case XenbusStateClosed:
2477                 if (dev->state == XenbusStateClosed)
2478                         break;
2479                 /* fall through */
2480         case XenbusStateClosing:
2481                 if (info)
2482                         blkfront_closing(info);
2483                 break;
2484         }
2485 }
2486
2487 static int blkfront_remove(struct xenbus_device *xbdev)
2488 {
2489         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2490         struct block_device *bdev = NULL;
2491         struct gendisk *disk;
2492
2493         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2494
2495         if (!info)
2496                 return 0;
2497
2498         blkif_free(info, 0);
2499
2500         mutex_lock(&info->mutex);
2501
2502         disk = info->gd;
2503         if (disk)
2504                 bdev = bdget_disk(disk, 0);
2505
2506         info->xbdev = NULL;
2507         mutex_unlock(&info->mutex);
2508
2509         if (!bdev) {
2510                 mutex_lock(&blkfront_mutex);
2511                 free_info(info);
2512                 mutex_unlock(&blkfront_mutex);
2513                 return 0;
2514         }
2515
2516         /*
2517          * The xbdev was removed before we reached the Closed
2518          * state. See if it's safe to remove the disk. If the bdev
2519          * isn't closed yet, we let release take care of it.
2520          */
2521
2522         mutex_lock(&bdev->bd_mutex);
2523         info = disk->private_data;
2524
2525         dev_warn(disk_to_dev(disk),
2526                  "%s was hot-unplugged, %d stale handles\n",
2527                  xbdev->nodename, bdev->bd_openers);
2528
2529         if (info && !bdev->bd_openers) {
2530                 xlvbd_release_gendisk(info);
2531                 disk->private_data = NULL;
2532                 mutex_lock(&blkfront_mutex);
2533                 free_info(info);
2534                 mutex_unlock(&blkfront_mutex);
2535         }
2536
2537         mutex_unlock(&bdev->bd_mutex);
2538         bdput(bdev);
2539
2540         return 0;
2541 }
2542
2543 static int blkfront_is_ready(struct xenbus_device *dev)
2544 {
2545         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2546
2547         return info->is_ready && info->xbdev;
2548 }
2549
2550 static int blkif_open(struct block_device *bdev, fmode_t mode)
2551 {
2552         struct gendisk *disk = bdev->bd_disk;
2553         struct blkfront_info *info;
2554         int err = 0;
2555
2556         mutex_lock(&blkfront_mutex);
2557
2558         info = disk->private_data;
2559         if (!info) {
2560                 /* xbdev gone */
2561                 err = -ERESTARTSYS;
2562                 goto out;
2563         }
2564
2565         mutex_lock(&info->mutex);
2566
2567         if (!info->gd)
2568                 /* xbdev is closed */
2569                 err = -ERESTARTSYS;
2570
2571         mutex_unlock(&info->mutex);
2572
2573 out:
2574         mutex_unlock(&blkfront_mutex);
2575         return err;
2576 }
2577
2578 static void blkif_release(struct gendisk *disk, fmode_t mode)
2579 {
2580         struct blkfront_info *info = disk->private_data;
2581         struct block_device *bdev;
2582         struct xenbus_device *xbdev;
2583
2584         mutex_lock(&blkfront_mutex);
2585
2586         bdev = bdget_disk(disk, 0);
2587
2588         if (!bdev) {
2589                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2590                 goto out_mutex;
2591         }
2592         if (bdev->bd_openers)
2593                 goto out;
2594
2595         /*
2596          * Check if we have been instructed to close. We will have
2597          * deferred this request, because the bdev was still open.
2598          */
2599
2600         mutex_lock(&info->mutex);
2601         xbdev = info->xbdev;
2602
2603         if (xbdev && xbdev->state == XenbusStateClosing) {
2604                 /* pending switch to state closed */
2605                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2606                 xlvbd_release_gendisk(info);
2607                 xenbus_frontend_closed(info->xbdev);
2608         }
2609
2610         mutex_unlock(&info->mutex);
2611
2612         if (!xbdev) {
2613                 /* sudden device removal */
2614                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2615                 xlvbd_release_gendisk(info);
2616                 disk->private_data = NULL;
2617                 free_info(info);
2618         }
2619
2620 out:
2621         bdput(bdev);
2622 out_mutex:
2623         mutex_unlock(&blkfront_mutex);
2624 }
2625
2626 static const struct block_device_operations xlvbd_block_fops =
2627 {
2628         .owner = THIS_MODULE,
2629         .open = blkif_open,
2630         .release = blkif_release,
2631         .getgeo = blkif_getgeo,
2632         .ioctl = blkif_ioctl,
2633         .compat_ioctl = blkdev_compat_ptr_ioctl,
2634 };
2635
2636
2637 static const struct xenbus_device_id blkfront_ids[] = {
2638         { "vbd" },
2639         { "" }
2640 };
2641
2642 static struct xenbus_driver blkfront_driver = {
2643         .ids  = blkfront_ids,
2644         .probe = blkfront_probe,
2645         .remove = blkfront_remove,
2646         .resume = blkfront_resume,
2647         .otherend_changed = blkback_changed,
2648         .is_ready = blkfront_is_ready,
2649 };
2650
2651 static void purge_persistent_grants(struct blkfront_info *info)
2652 {
2653         unsigned int i;
2654         unsigned long flags;
2655         struct blkfront_ring_info *rinfo;
2656
2657         for_each_rinfo(info, rinfo, i) {
2658                 struct grant *gnt_list_entry, *tmp;
2659
2660                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2661
2662                 if (rinfo->persistent_gnts_c == 0) {
2663                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2664                         continue;
2665                 }
2666
2667                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2668                                          node) {
2669                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2670                             gnttab_query_foreign_access(gnt_list_entry->gref))
2671                                 continue;
2672
2673                         list_del(&gnt_list_entry->node);
2674                         gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
2675                         rinfo->persistent_gnts_c--;
2676                         gnt_list_entry->gref = GRANT_INVALID_REF;
2677                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2678                 }
2679
2680                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2681         }
2682 }
2683
2684 static void blkfront_delay_work(struct work_struct *work)
2685 {
2686         struct blkfront_info *info;
2687         bool need_schedule_work = false;
2688
2689         mutex_lock(&blkfront_mutex);
2690
2691         list_for_each_entry(info, &info_list, info_list) {
2692                 if (info->feature_persistent) {
2693                         need_schedule_work = true;
2694                         mutex_lock(&info->mutex);
2695                         purge_persistent_grants(info);
2696                         mutex_unlock(&info->mutex);
2697                 }
2698         }
2699
2700         if (need_schedule_work)
2701                 schedule_delayed_work(&blkfront_work, HZ * 10);
2702
2703         mutex_unlock(&blkfront_mutex);
2704 }
2705
2706 static int __init xlblk_init(void)
2707 {
2708         int ret;
2709         int nr_cpus = num_online_cpus();
2710
2711         if (!xen_domain())
2712                 return -ENODEV;
2713
2714         if (!xen_has_pv_disk_devices())
2715                 return -ENODEV;
2716
2717         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2718                 pr_warn("xen_blk: can't get major %d with name %s\n",
2719                         XENVBD_MAJOR, DEV_NAME);
2720                 return -ENODEV;
2721         }
2722
2723         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2724                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2725
2726         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2727                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2728                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2729                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2730         }
2731
2732         if (xen_blkif_max_queues > nr_cpus) {
2733                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2734                         xen_blkif_max_queues, nr_cpus);
2735                 xen_blkif_max_queues = nr_cpus;
2736         }
2737
2738         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2739
2740         ret = xenbus_register_frontend(&blkfront_driver);
2741         if (ret) {
2742                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2743                 return ret;
2744         }
2745
2746         return 0;
2747 }
2748 module_init(xlblk_init);
2749
2750
2751 static void __exit xlblk_exit(void)
2752 {
2753         cancel_delayed_work_sync(&blkfront_work);
2754
2755         xenbus_unregister_driver(&blkfront_driver);
2756         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2757         kfree(minors);
2758 }
2759 module_exit(xlblk_exit);
2760
2761 MODULE_DESCRIPTION("Xen virtual block device frontend");
2762 MODULE_LICENSE("GPL");
2763 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2764 MODULE_ALIAS("xen:vbd");
2765 MODULE_ALIAS("xenblk");