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 #include <linux/sched/mm.h>
  51
  52 #include <xen/xen.h>
  53 #include <xen/xenbus.h>
  54 #include <xen/grant_table.h>
  55 #include <xen/events.h>
  56 #include <xen/page.h>
  57 #include <xen/platform_pci.h>
  58
  59 #include <xen/interface/grant_table.h>
  60 #include <xen/interface/io/blkif.h>
  61 #include <xen/interface/io/protocols.h>
  62
  63 #include <asm/xen/hypervisor.h>
  64
  65 /*
  66  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  67  * When Linux is using a different page size than Xen, it may not be possible
  68  * to put all the data in a single segment.
  69  * This can happen when the backend doesn't support indirect descriptor and
  70  * therefore the maximum amount of data that a request can carry is
  71  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  72  *
  73  * Note that we only support one extra request. So the Linux page size
  74  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  75  * 88KB.
  76  */
  77 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  78
  79 enum blkif_state {
  80         BLKIF_STATE_DISCONNECTED,
  81         BLKIF_STATE_CONNECTED,
  82         BLKIF_STATE_SUSPENDED,
  83         BLKIF_STATE_ERROR,
  84 };
  85
  86 struct grant {
  87         grant_ref_t gref;
  88         struct page *page;
  89         struct list_head node;
  90 };
  91
  92 enum blk_req_status {
  93         REQ_PROCESSING,
  94         REQ_WAITING,
  95         REQ_DONE,
  96         REQ_ERROR,
  97         REQ_EOPNOTSUPP,
  98 };
  99
 100 struct blk_shadow {
 101         struct blkif_request req;
 102         struct request *request;
 103         struct grant **grants_used;
 104         struct grant **indirect_grants;
 105         struct scatterlist *sg;
 106         unsigned int num_sg;
 107         enum blk_req_status status;
 108
 109         #define NO_ASSOCIATED_ID ~0UL
 110         /*
 111          * Id of the sibling if we ever need 2 requests when handling a
 112          * block I/O request
 113          */
 114         unsigned long associated_id;
 115 };
 116
 117 struct blkif_req {
 118         blk_status_t    error;
 119 };
 120
 121 static inline struct blkif_req *blkif_req(struct request *rq)
 122 {
 123         return blk_mq_rq_to_pdu(rq);
 124 }
 125
 126 static DEFINE_MUTEX(blkfront_mutex);
 127 static const struct block_device_operations xlvbd_block_fops;
 128 static struct delayed_work blkfront_work;
 129 static LIST_HEAD(info_list);
 130
 131 /*
 132  * Maximum number of segments in indirect requests, the actual value used by
 133  * the frontend driver is the minimum of this value and the value provided
 134  * by the backend driver.
 135  */
 136
 137 static unsigned int xen_blkif_max_segments = 32;
 138 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 139 MODULE_PARM_DESC(max_indirect_segments,
 140                  "Maximum amount of segments in indirect requests (default is 32)");
 141
 142 static unsigned int xen_blkif_max_queues = 4;
 143 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 144 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 145
 146 /*
 147  * Maximum order of pages to be used for the shared ring between front and
 148  * backend, 4KB page granularity is used.
 149  */
 150 static unsigned int xen_blkif_max_ring_order;
 151 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 152 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 153
 154 static bool __read_mostly xen_blkif_trusted = true;
 155 module_param_named(trusted, xen_blkif_trusted, bool, 0644);
 156 MODULE_PARM_DESC(trusted, "Is the backend trusted");
 157
 158 #define BLK_RING_SIZE(info)     \
 159         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 160
 161 #define BLK_MAX_RING_SIZE       \
 162         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * XENBUS_MAX_RING_GRANTS)
 163
 164 /*
 165  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 166  * characters are enough. Define to 20 to keep consistent with backend.
 167  */
 168 #define RINGREF_NAME_LEN (20)
 169 /*
 170  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 171  */
 172 #define QUEUE_NAME_LEN (17)
 173
 174 /*
 175  *  Per-ring info.
 176  *  Every blkfront device can associate with one or more blkfront_ring_info,
 177  *  depending on how many hardware queues/rings to be used.
 178  */
 179 struct blkfront_ring_info {
 180         /* Lock to protect data in every ring buffer. */
 181         spinlock_t ring_lock;
 182         struct blkif_front_ring ring;
 183         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 184         unsigned int evtchn, irq;
 185         struct work_struct work;
 186         struct gnttab_free_callback callback;
 187         struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 188         struct list_head indirect_pages;
 189         struct list_head grants;
 190         unsigned int persistent_gnts_c;
 191         unsigned long shadow_free;
 192         struct blkfront_info *dev_info;
 193 };
 194
 195 /*
 196  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 197  * hang in private_data off the gendisk structure. We may end up
 198  * putting all kinds of interesting stuff here :-)
 199  */
 200 struct blkfront_info
 201 {
 202         struct mutex mutex;
 203         struct xenbus_device *xbdev;
 204         struct gendisk *gd;
 205         u16 sector_size;
 206         unsigned int physical_sector_size;
 207         int vdevice;
 208         blkif_vdev_t handle;
 209         enum blkif_state connected;
 210         /* Number of pages per ring buffer. */
 211         unsigned int nr_ring_pages;
 212         struct request_queue *rq;
 213         unsigned int feature_flush:1;
 214         unsigned int feature_fua:1;
 215         unsigned int feature_discard:1;
 216         unsigned int feature_secdiscard:1;
 217         unsigned int feature_persistent:1;
 218         unsigned int bounce:1;
 219         unsigned int discard_granularity;
 220         unsigned int discard_alignment;
 221         /* Number of 4KB segments handled */
 222         unsigned int max_indirect_segments;
 223         int is_ready;
 224         struct blk_mq_tag_set tag_set;
 225         struct blkfront_ring_info *rinfo;
 226         unsigned int nr_rings;
 227         /* Save uncomplete reqs and bios for migration. */
 228         struct list_head requests;
 229         struct bio_list bio_list;
 230         struct list_head info_list;
 231 };
 232
 233 static unsigned int nr_minors;
 234 static unsigned long *minors;
 235 static DEFINE_SPINLOCK(minor_lock);
 236
 237 #define GRANT_INVALID_REF       0
 238
 239 #define PARTS_PER_DISK          16
 240 #define PARTS_PER_EXT_DISK      256
 241
 242 #define BLKIF_MAJOR(dev) ((dev)>>8)
 243 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 244
 245 #define EXT_SHIFT 28
 246 #define EXTENDED (1<<EXT_SHIFT)
 247 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 248 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 249 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 250 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 251 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 252 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 253
 254 #define DEV_NAME        "xvd"   /* name in /dev */
 255
 256 /*
 257  * Grants are always the same size as a Xen page (i.e 4KB).
 258  * A physical segment is always the same size as a Linux page.
 259  * Number of grants per physical segment
 260  */
 261 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 262
 263 #define GRANTS_PER_INDIRECT_FRAME \
 264         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 265
 266 #define INDIRECT_GREFS(_grants)         \
 267         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 268
 269 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 270 static void blkfront_gather_backend_features(struct blkfront_info *info);
 271 static int negotiate_mq(struct blkfront_info *info);
 272
 273 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 274 {
 275         unsigned long free = rinfo->shadow_free;
 276
 277         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 278         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 279         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 280         return free;
 281 }
 282
 283 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 284                               unsigned long id)
 285 {
 286         if (rinfo->shadow[id].req.u.rw.id != id)
 287                 return -EINVAL;
 288         if (rinfo->shadow[id].request == NULL)
 289                 return -EINVAL;
 290         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 291         rinfo->shadow[id].request = NULL;
 292         rinfo->shadow_free = id;
 293         return 0;
 294 }
 295
 296 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 297 {
 298         struct blkfront_info *info = rinfo->dev_info;
 299         struct page *granted_page;
 300         struct grant *gnt_list_entry, *n;
 301         int i = 0;
 302
 303         while (i < num) {
 304                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 305                 if (!gnt_list_entry)
 306                         goto out_of_memory;
 307
 308                 if (info->bounce) {
 309                         granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
 310                         if (!granted_page) {
 311                                 kfree(gnt_list_entry);
 312                                 goto out_of_memory;
 313                         }
 314                         gnt_list_entry->page = granted_page;
 315                 }
 316
 317                 gnt_list_entry->gref = GRANT_INVALID_REF;
 318                 list_add(&gnt_list_entry->node, &rinfo->grants);
 319                 i++;
 320         }
 321
 322         return 0;
 323
 324 out_of_memory:
 325         list_for_each_entry_safe(gnt_list_entry, n,
 326                                  &rinfo->grants, node) {
 327                 list_del(&gnt_list_entry->node);
 328                 if (info->bounce)
 329                         __free_page(gnt_list_entry->page);
 330                 kfree(gnt_list_entry);
 331                 i--;
 332         }
 333         BUG_ON(i != 0);
 334         return -ENOMEM;
 335 }
 336
 337 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 338 {
 339         struct grant *gnt_list_entry;
 340
 341         BUG_ON(list_empty(&rinfo->grants));
 342         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 343                                           node);
 344         list_del(&gnt_list_entry->node);
 345
 346         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 347                 rinfo->persistent_gnts_c--;
 348
 349         return gnt_list_entry;
 350 }
 351
 352 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 353                                         const struct blkfront_info *info)
 354 {
 355         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 356                                                  info->xbdev->otherend_id,
 357                                                  gnt_list_entry->page,
 358                                                  0);
 359 }
 360
 361 static struct grant *get_grant(grant_ref_t *gref_head,
 362                                unsigned long gfn,
 363                                struct blkfront_ring_info *rinfo)
 364 {
 365         struct grant *gnt_list_entry = get_free_grant(rinfo);
 366         struct blkfront_info *info = rinfo->dev_info;
 367
 368         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 369                 return gnt_list_entry;
 370
 371         /* Assign a gref to this page */
 372         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 373         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 374         if (info->bounce)
 375                 grant_foreign_access(gnt_list_entry, info);
 376         else {
 377                 /* Grant access to the GFN passed by the caller */
 378                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 379                                                 info->xbdev->otherend_id,
 380                                                 gfn, 0);
 381         }
 382
 383         return gnt_list_entry;
 384 }
 385
 386 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 387                                         struct blkfront_ring_info *rinfo)
 388 {
 389         struct grant *gnt_list_entry = get_free_grant(rinfo);
 390         struct blkfront_info *info = rinfo->dev_info;
 391
 392         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 393                 return gnt_list_entry;
 394
 395         /* Assign a gref to this page */
 396         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 397         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 398         if (!info->bounce) {
 399                 struct page *indirect_page;
 400
 401                 /* Fetch a pre-allocated page to use for indirect grefs */
 402                 BUG_ON(list_empty(&rinfo->indirect_pages));
 403                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 404                                                  struct page, lru);
 405                 list_del(&indirect_page->lru);
 406                 gnt_list_entry->page = indirect_page;
 407         }
 408         grant_foreign_access(gnt_list_entry, info);
 409
 410         return gnt_list_entry;
 411 }
 412
 413 static const char *op_name(int op)
 414 {
 415         static const char *const names[] = {
 416                 [BLKIF_OP_READ] = "read",
 417                 [BLKIF_OP_WRITE] = "write",
 418                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 419                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 420                 [BLKIF_OP_DISCARD] = "discard" };
 421
 422         if (op < 0 || op >= ARRAY_SIZE(names))
 423                 return "unknown";
 424
 425         if (!names[op])
 426                 return "reserved";
 427
 428         return names[op];
 429 }
 430 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 431 {
 432         unsigned int end = minor + nr;
 433         int rc;
 434
 435         if (end > nr_minors) {
 436                 unsigned long *bitmap, *old;
 437
 438                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 439                                  GFP_KERNEL);
 440                 if (bitmap == NULL)
 441                         return -ENOMEM;
 442
 443                 spin_lock(&minor_lock);
 444                 if (end > nr_minors) {
 445                         old = minors;
 446                         memcpy(bitmap, minors,
 447                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 448                         minors = bitmap;
 449                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 450                 } else
 451                         old = bitmap;
 452                 spin_unlock(&minor_lock);
 453                 kfree(old);
 454         }
 455
 456         spin_lock(&minor_lock);
 457         if (find_next_bit(minors, end, minor) >= end) {
 458                 bitmap_set(minors, minor, nr);
 459                 rc = 0;
 460         } else
 461                 rc = -EBUSY;
 462         spin_unlock(&minor_lock);
 463
 464         return rc;
 465 }
 466
 467 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 468 {
 469         unsigned int end = minor + nr;
 470
 471         BUG_ON(end > nr_minors);
 472         spin_lock(&minor_lock);
 473         bitmap_clear(minors,  minor, nr);
 474         spin_unlock(&minor_lock);
 475 }
 476
 477 static void blkif_restart_queue_callback(void *arg)
 478 {
 479         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 480         schedule_work(&rinfo->work);
 481 }
 482
 483 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 484 {
 485         /* We don't have real geometry info, but let's at least return
 486            values consistent with the size of the device */
 487         sector_t nsect = get_capacity(bd->bd_disk);
 488         sector_t cylinders = nsect;
 489
 490         hg->heads = 0xff;
 491         hg->sectors = 0x3f;
 492         sector_div(cylinders, hg->heads * hg->sectors);
 493         hg->cylinders = cylinders;
 494         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 495                 hg->cylinders = 0xffff;
 496         return 0;
 497 }
 498
 499 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 500                        unsigned command, unsigned long argument)
 501 {
 502         struct blkfront_info *info = bdev->bd_disk->private_data;
 503         int i;
 504
 505         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 506                 command, (long)argument);
 507
 508         switch (command) {
 509         case CDROMMULTISESSION:
 510                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 511                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 512                         if (put_user(0, (char __user *)(argument + i)))
 513                                 return -EFAULT;
 514                 return 0;
 515
 516         case CDROM_GET_CAPABILITY: {
 517                 struct gendisk *gd = info->gd;
 518                 if (gd->flags & GENHD_FL_CD)
 519                         return 0;
 520                 return -EINVAL;
 521         }
 522
 523         default:
 524                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 525                   command);*/
 526                 return -EINVAL; /* same return as native Linux */
 527         }
 528
 529         return 0;
 530 }
 531
 532 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 533                                             struct request *req,
 534                                             struct blkif_request **ring_req)
 535 {
 536         unsigned long id;
 537
 538         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 539         rinfo->ring.req_prod_pvt++;
 540
 541         id = get_id_from_freelist(rinfo);
 542         rinfo->shadow[id].request = req;
 543         rinfo->shadow[id].status = REQ_PROCESSING;
 544         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 545
 546         rinfo->shadow[id].req.u.rw.id = id;
 547
 548         return id;
 549 }
 550
 551 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 552 {
 553         struct blkfront_info *info = rinfo->dev_info;
 554         struct blkif_request *ring_req, *final_ring_req;
 555         unsigned long id;
 556
 557         /* Fill out a communications ring structure. */
 558         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 559         ring_req = &rinfo->shadow[id].req;
 560
 561         ring_req->operation = BLKIF_OP_DISCARD;
 562         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 563         ring_req->u.discard.id = id;
 564         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 565         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 566                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 567         else
 568                 ring_req->u.discard.flag = 0;
 569
 570         /* Copy the request to the ring page. */
 571         *final_ring_req = *ring_req;
 572         rinfo->shadow[id].status = REQ_WAITING;
 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         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 706         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 707         bool require_extra_req = false;
 708         int i;
 709         struct setup_rw_req setup = {
 710                 .grant_idx = 0,
 711                 .segments = NULL,
 712                 .rinfo = rinfo,
 713                 .need_copy = rq_data_dir(req) && info->bounce,
 714         };
 715
 716         /*
 717          * Used to store if we are able to queue the request by just using
 718          * existing persistent grants, or if we have to get new grants,
 719          * as there are not sufficiently many free.
 720          */
 721         bool new_persistent_gnts = false;
 722         struct scatterlist *sg;
 723         int num_sg, max_grefs, num_grant;
 724
 725         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 726         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 727                 /*
 728                  * If we are using indirect segments we need to account
 729                  * for the indirect grefs used in the request.
 730                  */
 731                 max_grefs += INDIRECT_GREFS(max_grefs);
 732
 733         /* Check if we have enough persistent grants to allocate a requests */
 734         if (rinfo->persistent_gnts_c < max_grefs) {
 735                 new_persistent_gnts = true;
 736
 737                 if (gnttab_alloc_grant_references(
 738                     max_grefs - rinfo->persistent_gnts_c,
 739                     &setup.gref_head) < 0) {
 740                         gnttab_request_free_callback(
 741                                 &rinfo->callback,
 742                                 blkif_restart_queue_callback,
 743                                 rinfo,
 744                                 max_grefs - rinfo->persistent_gnts_c);
 745                         return 1;
 746                 }
 747         }
 748
 749         /* Fill out a communications ring structure. */
 750         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 751         ring_req = &rinfo->shadow[id].req;
 752
 753         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 754         num_grant = 0;
 755         /* Calculate the number of grant used */
 756         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 757                num_grant += gnttab_count_grant(sg->offset, sg->length);
 758
 759         require_extra_req = info->max_indirect_segments == 0 &&
 760                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 761         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 762
 763         rinfo->shadow[id].num_sg = num_sg;
 764         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 765             likely(!require_extra_req)) {
 766                 /*
 767                  * The indirect operation can only be a BLKIF_OP_READ or
 768                  * BLKIF_OP_WRITE
 769                  */
 770                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 771                 ring_req->operation = BLKIF_OP_INDIRECT;
 772                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 773                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 774                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 775                 ring_req->u.indirect.handle = info->handle;
 776                 ring_req->u.indirect.nr_segments = num_grant;
 777         } else {
 778                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 779                 ring_req->u.rw.handle = info->handle;
 780                 ring_req->operation = rq_data_dir(req) ?
 781                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 782                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 783                         /*
 784                          * Ideally we can do an unordered flush-to-disk.
 785                          * In case the backend onlysupports barriers, use that.
 786                          * A barrier request a superset of FUA, so we can
 787                          * implement it the same way.  (It's also a FLUSH+FUA,
 788                          * since it is guaranteed ordered WRT previous writes.)
 789                          */
 790                         if (info->feature_flush && info->feature_fua)
 791                                 ring_req->operation =
 792                                         BLKIF_OP_WRITE_BARRIER;
 793                         else if (info->feature_flush)
 794                                 ring_req->operation =
 795                                         BLKIF_OP_FLUSH_DISKCACHE;
 796                         else
 797                                 ring_req->operation = 0;
 798                 }
 799                 ring_req->u.rw.nr_segments = num_grant;
 800                 if (unlikely(require_extra_req)) {
 801                         extra_id = blkif_ring_get_request(rinfo, req,
 802                                                           &final_extra_ring_req);
 803                         extra_ring_req = &rinfo->shadow[extra_id].req;
 804
 805                         /*
 806                          * Only the first request contains the scatter-gather
 807                          * list.
 808                          */
 809                         rinfo->shadow[extra_id].num_sg = 0;
 810
 811                         blkif_setup_extra_req(ring_req, extra_ring_req);
 812
 813                         /* Link the 2 requests together */
 814                         rinfo->shadow[extra_id].associated_id = id;
 815                         rinfo->shadow[id].associated_id = extra_id;
 816                 }
 817         }
 818
 819         setup.ring_req = ring_req;
 820         setup.id = id;
 821
 822         setup.require_extra_req = require_extra_req;
 823         if (unlikely(require_extra_req))
 824                 setup.extra_ring_req = extra_ring_req;
 825
 826         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 827                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 828
 829                 if (setup.need_copy) {
 830                         setup.bvec_off = sg->offset;
 831                         setup.bvec_data = kmap_atomic(sg_page(sg));
 832                 }
 833
 834                 gnttab_foreach_grant_in_range(sg_page(sg),
 835                                               sg->offset,
 836                                               sg->length,
 837                                               blkif_setup_rw_req_grant,
 838                                               &setup);
 839
 840                 if (setup.need_copy)
 841                         kunmap_atomic(setup.bvec_data);
 842         }
 843         if (setup.segments)
 844                 kunmap_atomic(setup.segments);
 845
 846         /* Copy request(s) to the ring page. */
 847         *final_ring_req = *ring_req;
 848         rinfo->shadow[id].status = REQ_WAITING;
 849         if (unlikely(require_extra_req)) {
 850                 *final_extra_ring_req = *extra_ring_req;
 851                 rinfo->shadow[extra_id].status = REQ_WAITING;
 852         }
 853
 854         if (new_persistent_gnts)
 855                 gnttab_free_grant_references(setup.gref_head);
 856
 857         return 0;
 858 }
 859
 860 /*
 861  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 862  * and writes are handled as expected.
 863  *
 864  * @req: a request struct
 865  */
 866 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 867 {
 868         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 869                 return 1;
 870
 871         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 872                      req_op(req) == REQ_OP_SECURE_ERASE))
 873                 return blkif_queue_discard_req(req, rinfo);
 874         else
 875                 return blkif_queue_rw_req(req, rinfo);
 876 }
 877
 878 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 879 {
 880         int notify;
 881
 882         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 883
 884         if (notify)
 885                 notify_remote_via_irq(rinfo->irq);
 886 }
 887
 888 static inline bool blkif_request_flush_invalid(struct request *req,
 889                                                struct blkfront_info *info)
 890 {
 891         return (blk_rq_is_passthrough(req) ||
 892                 ((req_op(req) == REQ_OP_FLUSH) &&
 893                  !info->feature_flush) ||
 894                 ((req->cmd_flags & REQ_FUA) &&
 895                  !info->feature_fua));
 896 }
 897
 898 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 899                           const struct blk_mq_queue_data *qd)
 900 {
 901         unsigned long flags;
 902         int qid = hctx->queue_num;
 903         struct blkfront_info *info = hctx->queue->queuedata;
 904         struct blkfront_ring_info *rinfo = NULL;
 905
 906         BUG_ON(info->nr_rings <= qid);
 907         rinfo = &info->rinfo[qid];
 908         blk_mq_start_request(qd->rq);
 909         spin_lock_irqsave(&rinfo->ring_lock, flags);
 910         if (RING_FULL(&rinfo->ring))
 911                 goto out_busy;
 912
 913         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 914                 goto out_err;
 915
 916         if (blkif_queue_request(qd->rq, rinfo))
 917                 goto out_busy;
 918
 919         flush_requests(rinfo);
 920         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 921         return BLK_STS_OK;
 922
 923 out_err:
 924         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 925         return BLK_STS_IOERR;
 926
 927 out_busy:
 928         blk_mq_stop_hw_queue(hctx);
 929         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 930         return BLK_STS_DEV_RESOURCE;
 931 }
 932
 933 static void blkif_complete_rq(struct request *rq)
 934 {
 935         blk_mq_end_request(rq, blkif_req(rq)->error);
 936 }
 937
 938 static const struct blk_mq_ops blkfront_mq_ops = {
 939         .queue_rq = blkif_queue_rq,
 940         .complete = blkif_complete_rq,
 941 };
 942
 943 static void blkif_set_queue_limits(struct blkfront_info *info)
 944 {
 945         struct request_queue *rq = info->rq;
 946         struct gendisk *gd = info->gd;
 947         unsigned int segments = info->max_indirect_segments ? :
 948                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 949
 950         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 951
 952         if (info->feature_discard) {
 953                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 954                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 955                 rq->limits.discard_granularity = info->discard_granularity ?:
 956                                                  info->physical_sector_size;
 957                 rq->limits.discard_alignment = info->discard_alignment;
 958                 if (info->feature_secdiscard)
 959                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 960         }
 961
 962         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 963         blk_queue_logical_block_size(rq, info->sector_size);
 964         blk_queue_physical_block_size(rq, info->physical_sector_size);
 965         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 966
 967         /* Each segment in a request is up to an aligned page in size. */
 968         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 969         blk_queue_max_segment_size(rq, PAGE_SIZE);
 970
 971         /* Ensure a merged request will fit in a single I/O ring slot. */
 972         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 973
 974         /* Make sure buffer addresses are sector-aligned. */
 975         blk_queue_dma_alignment(rq, 511);
 976 }
 977
 978 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 979                                 unsigned int physical_sector_size)
 980 {
 981         struct request_queue *rq;
 982         struct blkfront_info *info = gd->private_data;
 983
 984         memset(&info->tag_set, 0, sizeof(info->tag_set));
 985         info->tag_set.ops = &blkfront_mq_ops;
 986         info->tag_set.nr_hw_queues = info->nr_rings;
 987         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
 988                 /*
 989                  * When indirect descriptior is not supported, the I/O request
 990                  * will be split between multiple request in the ring.
 991                  * To avoid problems when sending the request, divide by
 992                  * 2 the depth of the queue.
 993                  */
 994                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
 995         } else
 996                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
 997         info->tag_set.numa_node = NUMA_NO_NODE;
 998         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
 999         info->tag_set.cmd_size = sizeof(struct blkif_req);
1000         info->tag_set.driver_data = info;
1001
1002         if (blk_mq_alloc_tag_set(&info->tag_set))
1003                 return -EINVAL;
1004         rq = blk_mq_init_queue(&info->tag_set);
1005         if (IS_ERR(rq)) {
1006                 blk_mq_free_tag_set(&info->tag_set);
1007                 return PTR_ERR(rq);
1008         }
1009
1010         rq->queuedata = info;
1011         info->rq = gd->queue = rq;
1012         info->gd = gd;
1013         info->sector_size = sector_size;
1014         info->physical_sector_size = physical_sector_size;
1015         blkif_set_queue_limits(info);
1016
1017         return 0;
1018 }
1019
1020 static const char *flush_info(struct blkfront_info *info)
1021 {
1022         if (info->feature_flush && info->feature_fua)
1023                 return "barrier: enabled;";
1024         else if (info->feature_flush)
1025                 return "flush diskcache: enabled;";
1026         else
1027                 return "barrier or flush: disabled;";
1028 }
1029
1030 static void xlvbd_flush(struct blkfront_info *info)
1031 {
1032         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1033                               info->feature_fua ? true : false);
1034         pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
1035                 info->gd->disk_name, flush_info(info),
1036                 "persistent grants:", info->feature_persistent ?
1037                 "enabled;" : "disabled;", "indirect descriptors:",
1038                 info->max_indirect_segments ? "enabled;" : "disabled;",
1039                 "bounce buffer:", info->bounce ? "enabled" : "disabled;");
1040 }
1041
1042 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1043 {
1044         int major;
1045         major = BLKIF_MAJOR(vdevice);
1046         *minor = BLKIF_MINOR(vdevice);
1047         switch (major) {
1048                 case XEN_IDE0_MAJOR:
1049                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1050                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1051                                 EMULATED_HD_DISK_MINOR_OFFSET;
1052                         break;
1053                 case XEN_IDE1_MAJOR:
1054                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1055                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1056                                 EMULATED_HD_DISK_MINOR_OFFSET;
1057                         break;
1058                 case XEN_SCSI_DISK0_MAJOR:
1059                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1060                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1061                         break;
1062                 case XEN_SCSI_DISK1_MAJOR:
1063                 case XEN_SCSI_DISK2_MAJOR:
1064                 case XEN_SCSI_DISK3_MAJOR:
1065                 case XEN_SCSI_DISK4_MAJOR:
1066                 case XEN_SCSI_DISK5_MAJOR:
1067                 case XEN_SCSI_DISK6_MAJOR:
1068                 case XEN_SCSI_DISK7_MAJOR:
1069                         *offset = (*minor / PARTS_PER_DISK) +
1070                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1071                                 EMULATED_SD_DISK_NAME_OFFSET;
1072                         *minor = *minor +
1073                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1074                                 EMULATED_SD_DISK_MINOR_OFFSET;
1075                         break;
1076                 case XEN_SCSI_DISK8_MAJOR:
1077                 case XEN_SCSI_DISK9_MAJOR:
1078                 case XEN_SCSI_DISK10_MAJOR:
1079                 case XEN_SCSI_DISK11_MAJOR:
1080                 case XEN_SCSI_DISK12_MAJOR:
1081                 case XEN_SCSI_DISK13_MAJOR:
1082                 case XEN_SCSI_DISK14_MAJOR:
1083                 case XEN_SCSI_DISK15_MAJOR:
1084                         *offset = (*minor / PARTS_PER_DISK) +
1085                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1086                                 EMULATED_SD_DISK_NAME_OFFSET;
1087                         *minor = *minor +
1088                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1089                                 EMULATED_SD_DISK_MINOR_OFFSET;
1090                         break;
1091                 case XENVBD_MAJOR:
1092                         *offset = *minor / PARTS_PER_DISK;
1093                         break;
1094                 default:
1095                         printk(KERN_WARNING "blkfront: your disk configuration is "
1096                                         "incorrect, please use an xvd device instead\n");
1097                         return -ENODEV;
1098         }
1099         return 0;
1100 }
1101
1102 static char *encode_disk_name(char *ptr, unsigned int n)
1103 {
1104         if (n >= 26)
1105                 ptr = encode_disk_name(ptr, n / 26 - 1);
1106         *ptr = 'a' + n % 26;
1107         return ptr + 1;
1108 }
1109
1110 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1111                                struct blkfront_info *info,
1112                                u16 vdisk_info, u16 sector_size,
1113                                unsigned int physical_sector_size)
1114 {
1115         struct gendisk *gd;
1116         int nr_minors = 1;
1117         int err;
1118         unsigned int offset;
1119         int minor;
1120         int nr_parts;
1121         char *ptr;
1122
1123         BUG_ON(info->gd != NULL);
1124         BUG_ON(info->rq != NULL);
1125
1126         if ((info->vdevice>>EXT_SHIFT) > 1) {
1127                 /* this is above the extended range; something is wrong */
1128                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1129                 return -ENODEV;
1130         }
1131
1132         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1133                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1134                 if (err)
1135                         return err;
1136                 nr_parts = PARTS_PER_DISK;
1137         } else {
1138                 minor = BLKIF_MINOR_EXT(info->vdevice);
1139                 nr_parts = PARTS_PER_EXT_DISK;
1140                 offset = minor / nr_parts;
1141                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1142                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1143                                         "emulated IDE disks,\n\t choose an xvd device name"
1144                                         "from xvde on\n", info->vdevice);
1145         }
1146         if (minor >> MINORBITS) {
1147                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1148                         info->vdevice, minor);
1149                 return -ENODEV;
1150         }
1151
1152         if ((minor % nr_parts) == 0)
1153                 nr_minors = nr_parts;
1154
1155         err = xlbd_reserve_minors(minor, nr_minors);
1156         if (err)
1157                 goto out;
1158         err = -ENODEV;
1159
1160         gd = alloc_disk(nr_minors);
1161         if (gd == NULL)
1162                 goto release;
1163
1164         strcpy(gd->disk_name, DEV_NAME);
1165         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1166         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1167         if (nr_minors > 1)
1168                 *ptr = 0;
1169         else
1170                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1171                          "%d", minor & (nr_parts - 1));
1172
1173         gd->major = XENVBD_MAJOR;
1174         gd->first_minor = minor;
1175         gd->fops = &xlvbd_block_fops;
1176         gd->private_data = info;
1177         set_capacity(gd, capacity);
1178
1179         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1180                 del_gendisk(gd);
1181                 goto release;
1182         }
1183
1184         xlvbd_flush(info);
1185
1186         if (vdisk_info & VDISK_READONLY)
1187                 set_disk_ro(gd, 1);
1188
1189         if (vdisk_info & VDISK_REMOVABLE)
1190                 gd->flags |= GENHD_FL_REMOVABLE;
1191
1192         if (vdisk_info & VDISK_CDROM)
1193                 gd->flags |= GENHD_FL_CD;
1194
1195         return 0;
1196
1197  release:
1198         xlbd_release_minors(minor, nr_minors);
1199  out:
1200         return err;
1201 }
1202
1203 static void xlvbd_release_gendisk(struct blkfront_info *info)
1204 {
1205         unsigned int minor, nr_minors, i;
1206
1207         if (info->rq == NULL)
1208                 return;
1209
1210         /* No more blkif_request(). */
1211         blk_mq_stop_hw_queues(info->rq);
1212
1213         for (i = 0; i < info->nr_rings; i++) {
1214                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1215
1216                 /* No more gnttab callback work. */
1217                 gnttab_cancel_free_callback(&rinfo->callback);
1218
1219                 /* Flush gnttab callback work. Must be done with no locks held. */
1220                 flush_work(&rinfo->work);
1221         }
1222
1223         del_gendisk(info->gd);
1224
1225         minor = info->gd->first_minor;
1226         nr_minors = info->gd->minors;
1227         xlbd_release_minors(minor, nr_minors);
1228
1229         blk_cleanup_queue(info->rq);
1230         blk_mq_free_tag_set(&info->tag_set);
1231         info->rq = NULL;
1232
1233         put_disk(info->gd);
1234         info->gd = NULL;
1235 }
1236
1237 /* Already hold rinfo->ring_lock. */
1238 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1239 {
1240         if (!RING_FULL(&rinfo->ring))
1241                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1242 }
1243
1244 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1245 {
1246         unsigned long flags;
1247
1248         spin_lock_irqsave(&rinfo->ring_lock, flags);
1249         kick_pending_request_queues_locked(rinfo);
1250         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1251 }
1252
1253 static void blkif_restart_queue(struct work_struct *work)
1254 {
1255         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1256
1257         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1258                 kick_pending_request_queues(rinfo);
1259 }
1260
1261 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1262 {
1263         struct grant *persistent_gnt, *n;
1264         struct blkfront_info *info = rinfo->dev_info;
1265         int i, j, segs;
1266
1267         /*
1268          * Remove indirect pages, this only happens when using indirect
1269          * descriptors but not persistent grants
1270          */
1271         if (!list_empty(&rinfo->indirect_pages)) {
1272                 struct page *indirect_page, *n;
1273
1274                 BUG_ON(info->bounce);
1275                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1276                         list_del(&indirect_page->lru);
1277                         __free_page(indirect_page);
1278                 }
1279         }
1280
1281         /* Remove all persistent grants. */
1282         if (!list_empty(&rinfo->grants)) {
1283                 list_for_each_entry_safe(persistent_gnt, n,
1284                                          &rinfo->grants, node) {
1285                         list_del(&persistent_gnt->node);
1286                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1287                                 gnttab_end_foreign_access(persistent_gnt->gref,
1288                                                           0, 0UL);
1289                                 rinfo->persistent_gnts_c--;
1290                         }
1291                         if (info->bounce)
1292                                 __free_page(persistent_gnt->page);
1293                         kfree(persistent_gnt);
1294                 }
1295         }
1296         BUG_ON(rinfo->persistent_gnts_c != 0);
1297
1298         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1299                 /*
1300                  * Clear persistent grants present in requests already
1301                  * on the shared ring
1302                  */
1303                 if (!rinfo->shadow[i].request)
1304                         goto free_shadow;
1305
1306                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1307                        rinfo->shadow[i].req.u.indirect.nr_segments :
1308                        rinfo->shadow[i].req.u.rw.nr_segments;
1309                 for (j = 0; j < segs; j++) {
1310                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1311                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1312                         if (info->bounce)
1313                                 __free_page(persistent_gnt->page);
1314                         kfree(persistent_gnt);
1315                 }
1316
1317                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1318                         /*
1319                          * If this is not an indirect operation don't try to
1320                          * free indirect segments
1321                          */
1322                         goto free_shadow;
1323
1324                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1325                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1326                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1327                         __free_page(persistent_gnt->page);
1328                         kfree(persistent_gnt);
1329                 }
1330
1331 free_shadow:
1332                 kvfree(rinfo->shadow[i].grants_used);
1333                 rinfo->shadow[i].grants_used = NULL;
1334                 kvfree(rinfo->shadow[i].indirect_grants);
1335                 rinfo->shadow[i].indirect_grants = NULL;
1336                 kvfree(rinfo->shadow[i].sg);
1337                 rinfo->shadow[i].sg = NULL;
1338         }
1339
1340         /* No more gnttab callback work. */
1341         gnttab_cancel_free_callback(&rinfo->callback);
1342
1343         /* Flush gnttab callback work. Must be done with no locks held. */
1344         flush_work(&rinfo->work);
1345
1346         /* Free resources associated with old device channel. */
1347         for (i = 0; i < info->nr_ring_pages; i++) {
1348                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1349                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1350                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1351                 }
1352         }
1353         free_pages_exact(rinfo->ring.sring,
1354                          info->nr_ring_pages * XEN_PAGE_SIZE);
1355         rinfo->ring.sring = NULL;
1356
1357         if (rinfo->irq)
1358                 unbind_from_irqhandler(rinfo->irq, rinfo);
1359         rinfo->evtchn = rinfo->irq = 0;
1360 }
1361
1362 static void blkif_free(struct blkfront_info *info, int suspend)
1363 {
1364         unsigned int i;
1365
1366         /* Prevent new requests being issued until we fix things up. */
1367         info->connected = suspend ?
1368                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1369         /* No more blkif_request(). */
1370         if (info->rq)
1371                 blk_mq_stop_hw_queues(info->rq);
1372
1373         for (i = 0; i < info->nr_rings; i++)
1374                 blkif_free_ring(&info->rinfo[i]);
1375
1376         kvfree(info->rinfo);
1377         info->rinfo = NULL;
1378         info->nr_rings = 0;
1379 }
1380
1381 struct copy_from_grant {
1382         const struct blk_shadow *s;
1383         unsigned int grant_idx;
1384         unsigned int bvec_offset;
1385         char *bvec_data;
1386 };
1387
1388 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1389                                   unsigned int len, void *data)
1390 {
1391         struct copy_from_grant *info = data;
1392         char *shared_data;
1393         /* Convenient aliases */
1394         const struct blk_shadow *s = info->s;
1395
1396         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1397
1398         memcpy(info->bvec_data + info->bvec_offset,
1399                shared_data + offset, len);
1400
1401         info->bvec_offset += len;
1402         info->grant_idx++;
1403
1404         kunmap_atomic(shared_data);
1405 }
1406
1407 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1408 {
1409         switch (rsp)
1410         {
1411         case BLKIF_RSP_OKAY:
1412                 return REQ_DONE;
1413         case BLKIF_RSP_EOPNOTSUPP:
1414                 return REQ_EOPNOTSUPP;
1415         case BLKIF_RSP_ERROR:
1416                 /* Fallthrough. */
1417         default:
1418                 return REQ_ERROR;
1419         }
1420 }
1421
1422 /*
1423  * Get the final status of the block request based on two ring response
1424  */
1425 static int blkif_get_final_status(enum blk_req_status s1,
1426                                   enum blk_req_status s2)
1427 {
1428         BUG_ON(s1 < REQ_DONE);
1429         BUG_ON(s2 < REQ_DONE);
1430
1431         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1432                 return BLKIF_RSP_ERROR;
1433         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1434                 return BLKIF_RSP_EOPNOTSUPP;
1435         return BLKIF_RSP_OKAY;
1436 }
1437
1438 /*
1439  * Return values:
1440  *  1 response processed.
1441  *  0 missing further responses.
1442  * -1 error while processing.
1443  */
1444 static int blkif_completion(unsigned long *id,
1445                             struct blkfront_ring_info *rinfo,
1446                             struct blkif_response *bret)
1447 {
1448         int i = 0;
1449         struct scatterlist *sg;
1450         int num_sg, num_grant;
1451         struct blkfront_info *info = rinfo->dev_info;
1452         struct blk_shadow *s = &rinfo->shadow[*id];
1453         struct copy_from_grant data = {
1454                 .grant_idx = 0,
1455         };
1456
1457         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1458                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1459
1460         /* The I/O request may be split in two. */
1461         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1462                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1463
1464                 /* Keep the status of the current response in shadow. */
1465                 s->status = blkif_rsp_to_req_status(bret->status);
1466
1467                 /* Wait the second response if not yet here. */
1468                 if (s2->status < REQ_DONE)
1469                         return 0;
1470
1471                 bret->status = blkif_get_final_status(s->status,
1472                                                       s2->status);
1473
1474                 /*
1475                  * All the grants is stored in the first shadow in order
1476                  * to make the completion code simpler.
1477                  */
1478                 num_grant += s2->req.u.rw.nr_segments;
1479
1480                 /*
1481                  * The two responses may not come in order. Only the
1482                  * first request will store the scatter-gather list.
1483                  */
1484                 if (s2->num_sg != 0) {
1485                         /* Update "id" with the ID of the first response. */
1486                         *id = s->associated_id;
1487                         s = s2;
1488                 }
1489
1490                 /*
1491                  * We don't need anymore the second request, so recycling
1492                  * it now.
1493                  */
1494                 if (add_id_to_freelist(rinfo, s->associated_id))
1495                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1496                              info->gd->disk_name, s->associated_id);
1497         }
1498
1499         data.s = s;
1500         num_sg = s->num_sg;
1501
1502         if (bret->operation == BLKIF_OP_READ && info->bounce) {
1503                 for_each_sg(s->sg, sg, num_sg, i) {
1504                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1505
1506                         data.bvec_offset = sg->offset;
1507                         data.bvec_data = kmap_atomic(sg_page(sg));
1508
1509                         gnttab_foreach_grant_in_range(sg_page(sg),
1510                                                       sg->offset,
1511                                                       sg->length,
1512                                                       blkif_copy_from_grant,
1513                                                       &data);
1514
1515                         kunmap_atomic(data.bvec_data);
1516                 }
1517         }
1518         /* Add the persistent grant into the list of free grants */
1519         for (i = 0; i < num_grant; i++) {
1520                 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1521                         /*
1522                          * If the grant is still mapped by the backend (the
1523                          * backend has chosen to make this grant persistent)
1524                          * we add it at the head of the list, so it will be
1525                          * reused first.
1526                          */
1527                         if (!info->feature_persistent) {
1528                                 pr_alert("backed has not unmapped grant: %u\n",
1529                                          s->grants_used[i]->gref);
1530                                 return -1;
1531                         }
1532                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1533                         rinfo->persistent_gnts_c++;
1534                 } else {
1535                         /*
1536                          * If the grant is not mapped by the backend we add it
1537                          * to the tail of the list, so it will not be picked
1538                          * again unless we run out of persistent grants.
1539                          */
1540                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1541                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1542                 }
1543         }
1544         if (s->req.operation == BLKIF_OP_INDIRECT) {
1545                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1546                         if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1547                                 if (!info->feature_persistent) {
1548                                         pr_alert("backed has not unmapped grant: %u\n",
1549                                                  s->indirect_grants[i]->gref);
1550                                         return -1;
1551                                 }
1552                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1553                                 rinfo->persistent_gnts_c++;
1554                         } else {
1555                                 struct page *indirect_page;
1556
1557                                 /*
1558                                  * Add the used indirect page back to the list of
1559                                  * available pages for indirect grefs.
1560                                  */
1561                                 if (!info->bounce) {
1562                                         indirect_page = s->indirect_grants[i]->page;
1563                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1564                                 }
1565                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1566                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1567                         }
1568                 }
1569         }
1570
1571         return 1;
1572 }
1573
1574 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1575 {
1576         struct request *req;
1577         struct blkif_response bret;
1578         RING_IDX i, rp;
1579         unsigned long flags;
1580         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1581         struct blkfront_info *info = rinfo->dev_info;
1582         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1583
1584         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1585                 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1586                 return IRQ_HANDLED;
1587         }
1588
1589         spin_lock_irqsave(&rinfo->ring_lock, flags);
1590  again:
1591         rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1592         virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1593         if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1594                 pr_alert("%s: illegal number of responses %u\n",
1595                          info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1596                 goto err;
1597         }
1598
1599         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1600                 unsigned long id;
1601                 unsigned int op;
1602
1603                 eoiflag = 0;
1604
1605                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1606                 id = bret.id;
1607
1608                 /*
1609                  * The backend has messed up and given us an id that we would
1610                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1611                  * look in get_id_from_freelist.
1612                  */
1613                 if (id >= BLK_RING_SIZE(info)) {
1614                         pr_alert("%s: response has incorrect id (%ld)\n",
1615                                  info->gd->disk_name, id);
1616                         goto err;
1617                 }
1618                 if (rinfo->shadow[id].status != REQ_WAITING) {
1619                         pr_alert("%s: response references no pending request\n",
1620                                  info->gd->disk_name);
1621                         goto err;
1622                 }
1623
1624                 rinfo->shadow[id].status = REQ_PROCESSING;
1625                 req  = rinfo->shadow[id].request;
1626
1627                 op = rinfo->shadow[id].req.operation;
1628                 if (op == BLKIF_OP_INDIRECT)
1629                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1630                 if (bret.operation != op) {
1631                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1632                                  info->gd->disk_name, bret.operation, op);
1633                         goto err;
1634                 }
1635
1636                 if (bret.operation != BLKIF_OP_DISCARD) {
1637                         int ret;
1638
1639                         /*
1640                          * We may need to wait for an extra response if the
1641                          * I/O request is split in 2
1642                          */
1643                         ret = blkif_completion(&id, rinfo, &bret);
1644                         if (!ret)
1645                                 continue;
1646                         if (unlikely(ret < 0))
1647                                 goto err;
1648                 }
1649
1650                 if (add_id_to_freelist(rinfo, id)) {
1651                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1652                              info->gd->disk_name, op_name(bret.operation), id);
1653                         continue;
1654                 }
1655
1656                 if (bret.status == BLKIF_RSP_OKAY)
1657                         blkif_req(req)->error = BLK_STS_OK;
1658                 else
1659                         blkif_req(req)->error = BLK_STS_IOERR;
1660
1661                 switch (bret.operation) {
1662                 case BLKIF_OP_DISCARD:
1663                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1664                                 struct request_queue *rq = info->rq;
1665
1666                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1667                                            info->gd->disk_name, op_name(bret.operation));
1668                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1669                                 info->feature_discard = 0;
1670                                 info->feature_secdiscard = 0;
1671                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1672                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1673                         }
1674                         break;
1675                 case BLKIF_OP_FLUSH_DISKCACHE:
1676                 case BLKIF_OP_WRITE_BARRIER:
1677                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1678                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1679                                        info->gd->disk_name, op_name(bret.operation));
1680                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1681                         }
1682                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1683                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1684                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1685                                        info->gd->disk_name, op_name(bret.operation));
1686                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1687                         }
1688                         if (unlikely(blkif_req(req)->error)) {
1689                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1690                                         blkif_req(req)->error = BLK_STS_OK;
1691                                 info->feature_fua = 0;
1692                                 info->feature_flush = 0;
1693                                 xlvbd_flush(info);
1694                         }
1695                         /* fall through */
1696                 case BLKIF_OP_READ:
1697                 case BLKIF_OP_WRITE:
1698                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1699                                 dev_dbg_ratelimited(&info->xbdev->dev,
1700                                         "Bad return from blkdev data request: %#x\n",
1701                                         bret.status);
1702
1703                         break;
1704                 default:
1705                         BUG();
1706                 }
1707
1708                 blk_mq_complete_request(req);
1709         }
1710
1711         rinfo->ring.rsp_cons = i;
1712
1713         if (i != rinfo->ring.req_prod_pvt) {
1714                 int more_to_do;
1715                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1716                 if (more_to_do)
1717                         goto again;
1718         } else
1719                 rinfo->ring.sring->rsp_event = i + 1;
1720
1721         kick_pending_request_queues_locked(rinfo);
1722
1723         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1724
1725         xen_irq_lateeoi(irq, eoiflag);
1726
1727         return IRQ_HANDLED;
1728
1729  err:
1730         info->connected = BLKIF_STATE_ERROR;
1731
1732         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1733
1734         /* No EOI in order to avoid further interrupts. */
1735
1736         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1737         return IRQ_HANDLED;
1738 }
1739
1740
1741 static int setup_blkring(struct xenbus_device *dev,
1742                          struct blkfront_ring_info *rinfo)
1743 {
1744         struct blkif_sring *sring;
1745         int err, i;
1746         struct blkfront_info *info = rinfo->dev_info;
1747         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1748         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1749
1750         for (i = 0; i < info->nr_ring_pages; i++)
1751                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1752
1753         sring = alloc_pages_exact(ring_size, GFP_NOIO | __GFP_ZERO);
1754         if (!sring) {
1755                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1756                 return -ENOMEM;
1757         }
1758         SHARED_RING_INIT(sring);
1759         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1760
1761         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1762         if (err < 0) {
1763                 free_pages_exact(sring, ring_size);
1764                 rinfo->ring.sring = NULL;
1765                 goto fail;
1766         }
1767         for (i = 0; i < info->nr_ring_pages; i++)
1768                 rinfo->ring_ref[i] = gref[i];
1769
1770         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1771         if (err)
1772                 goto fail;
1773
1774         err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1775                                                 0, "blkif", rinfo);
1776         if (err <= 0) {
1777                 xenbus_dev_fatal(dev, err,
1778                                  "bind_evtchn_to_irqhandler failed");
1779                 goto fail;
1780         }
1781         rinfo->irq = err;
1782
1783         return 0;
1784 fail:
1785         blkif_free(info, 0);
1786         return err;
1787 }
1788
1789 /*
1790  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1791  * ring buffer may have multi pages depending on ->nr_ring_pages.
1792  */
1793 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1794                                 struct blkfront_ring_info *rinfo, const char *dir)
1795 {
1796         int err;
1797         unsigned int i;
1798         const char *message = NULL;
1799         struct blkfront_info *info = rinfo->dev_info;
1800
1801         if (info->nr_ring_pages == 1) {
1802                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1803                 if (err) {
1804                         message = "writing ring-ref";
1805                         goto abort_transaction;
1806                 }
1807         } else {
1808                 for (i = 0; i < info->nr_ring_pages; i++) {
1809                         char ring_ref_name[RINGREF_NAME_LEN];
1810
1811                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1812                         err = xenbus_printf(xbt, dir, ring_ref_name,
1813                                             "%u", rinfo->ring_ref[i]);
1814                         if (err) {
1815                                 message = "writing ring-ref";
1816                                 goto abort_transaction;
1817                         }
1818                 }
1819         }
1820
1821         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1822         if (err) {
1823                 message = "writing event-channel";
1824                 goto abort_transaction;
1825         }
1826
1827         return 0;
1828
1829 abort_transaction:
1830         xenbus_transaction_end(xbt, 1);
1831         if (message)
1832                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1833
1834         return err;
1835 }
1836
1837 static void free_info(struct blkfront_info *info)
1838 {
1839         list_del(&info->info_list);
1840         kfree(info);
1841 }
1842
1843 /* Common code used when first setting up, and when resuming. */
1844 static int talk_to_blkback(struct xenbus_device *dev,
1845                            struct blkfront_info *info)
1846 {
1847         const char *message = NULL;
1848         struct xenbus_transaction xbt;
1849         int err;
1850         unsigned int i, max_page_order;
1851         unsigned int ring_page_order;
1852
1853         if (!info)
1854                 return -ENODEV;
1855
1856         /* Check if backend is trusted. */
1857         info->bounce = !xen_blkif_trusted ||
1858                        !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1859
1860         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1861                                               "max-ring-page-order", 0);
1862         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1863         info->nr_ring_pages = 1 << ring_page_order;
1864
1865         err = negotiate_mq(info);
1866         if (err)
1867                 goto destroy_blkring;
1868
1869         for (i = 0; i < info->nr_rings; i++) {
1870                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1871
1872                 /* Create shared ring, alloc event channel. */
1873                 err = setup_blkring(dev, rinfo);
1874                 if (err)
1875                         goto destroy_blkring;
1876         }
1877
1878 again:
1879         err = xenbus_transaction_start(&xbt);
1880         if (err) {
1881                 xenbus_dev_fatal(dev, err, "starting transaction");
1882                 goto destroy_blkring;
1883         }
1884
1885         if (info->nr_ring_pages > 1) {
1886                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1887                                     ring_page_order);
1888                 if (err) {
1889                         message = "writing ring-page-order";
1890                         goto abort_transaction;
1891                 }
1892         }
1893
1894         /* We already got the number of queues/rings in _probe */
1895         if (info->nr_rings == 1) {
1896                 err = write_per_ring_nodes(xbt, &info->rinfo[0], dev->nodename);
1897                 if (err)
1898                         goto destroy_blkring;
1899         } else {
1900                 char *path;
1901                 size_t pathsize;
1902
1903                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1904                                     info->nr_rings);
1905                 if (err) {
1906                         message = "writing multi-queue-num-queues";
1907                         goto abort_transaction;
1908                 }
1909
1910                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1911                 path = kmalloc(pathsize, GFP_KERNEL);
1912                 if (!path) {
1913                         err = -ENOMEM;
1914                         message = "ENOMEM while writing ring references";
1915                         goto abort_transaction;
1916                 }
1917
1918                 for (i = 0; i < info->nr_rings; i++) {
1919                         memset(path, 0, pathsize);
1920                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1921                         err = write_per_ring_nodes(xbt, &info->rinfo[i], path);
1922                         if (err) {
1923                                 kfree(path);
1924                                 goto destroy_blkring;
1925                         }
1926                 }
1927                 kfree(path);
1928         }
1929         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1930                             XEN_IO_PROTO_ABI_NATIVE);
1931         if (err) {
1932                 message = "writing protocol";
1933                 goto abort_transaction;
1934         }
1935         err = xenbus_printf(xbt, dev->nodename,
1936                             "feature-persistent", "%u", 1);
1937         if (err)
1938                 dev_warn(&dev->dev,
1939                          "writing persistent grants feature to xenbus");
1940
1941         err = xenbus_transaction_end(xbt, 0);
1942         if (err) {
1943                 if (err == -EAGAIN)
1944                         goto again;
1945                 xenbus_dev_fatal(dev, err, "completing transaction");
1946                 goto destroy_blkring;
1947         }
1948
1949         for (i = 0; i < info->nr_rings; i++) {
1950                 unsigned int j;
1951                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
1952
1953                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1954                         rinfo->shadow[j].req.u.rw.id = j + 1;
1955                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1956         }
1957         xenbus_switch_state(dev, XenbusStateInitialised);
1958
1959         return 0;
1960
1961  abort_transaction:
1962         xenbus_transaction_end(xbt, 1);
1963         if (message)
1964                 xenbus_dev_fatal(dev, err, "%s", message);
1965  destroy_blkring:
1966         blkif_free(info, 0);
1967
1968         mutex_lock(&blkfront_mutex);
1969         free_info(info);
1970         mutex_unlock(&blkfront_mutex);
1971
1972         dev_set_drvdata(&dev->dev, NULL);
1973
1974         return err;
1975 }
1976
1977 static int negotiate_mq(struct blkfront_info *info)
1978 {
1979         unsigned int backend_max_queues;
1980         unsigned int i;
1981
1982         BUG_ON(info->nr_rings);
1983
1984         /* Check if backend supports multiple queues. */
1985         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1986                                                   "multi-queue-max-queues", 1);
1987         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1988         /* We need at least one ring. */
1989         if (!info->nr_rings)
1990                 info->nr_rings = 1;
1991
1992         info->rinfo = kvcalloc(info->nr_rings,
1993                                sizeof(struct blkfront_ring_info),
1994                                GFP_KERNEL);
1995         if (!info->rinfo) {
1996                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1997                 info->nr_rings = 0;
1998                 return -ENOMEM;
1999         }
2000
2001         for (i = 0; i < info->nr_rings; i++) {
2002                 struct blkfront_ring_info *rinfo;
2003
2004                 rinfo = &info->rinfo[i];
2005                 INIT_LIST_HEAD(&rinfo->indirect_pages);
2006                 INIT_LIST_HEAD(&rinfo->grants);
2007                 rinfo->dev_info = info;
2008                 INIT_WORK(&rinfo->work, blkif_restart_queue);
2009                 spin_lock_init(&rinfo->ring_lock);
2010         }
2011         return 0;
2012 }
2013 /**
2014  * Entry point to this code when a new device is created.  Allocate the basic
2015  * structures and the ring buffer for communication with the backend, and
2016  * inform the backend of the appropriate details for those.  Switch to
2017  * Initialised state.
2018  */
2019 static int blkfront_probe(struct xenbus_device *dev,
2020                           const struct xenbus_device_id *id)
2021 {
2022         int err, vdevice;
2023         struct blkfront_info *info;
2024
2025         /* FIXME: Use dynamic device id if this is not set. */
2026         err = xenbus_scanf(XBT_NIL, dev->nodename,
2027                            "virtual-device", "%i", &vdevice);
2028         if (err != 1) {
2029                 /* go looking in the extended area instead */
2030                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
2031                                    "%i", &vdevice);
2032                 if (err != 1) {
2033                         xenbus_dev_fatal(dev, err, "reading virtual-device");
2034                         return err;
2035                 }
2036         }
2037
2038         if (xen_hvm_domain()) {
2039                 char *type;
2040                 int len;
2041                 /* no unplug has been done: do not hook devices != xen vbds */
2042                 if (xen_has_pv_and_legacy_disk_devices()) {
2043                         int major;
2044
2045                         if (!VDEV_IS_EXTENDED(vdevice))
2046                                 major = BLKIF_MAJOR(vdevice);
2047                         else
2048                                 major = XENVBD_MAJOR;
2049
2050                         if (major != XENVBD_MAJOR) {
2051                                 printk(KERN_INFO
2052                                                 "%s: HVM does not support vbd %d as xen block device\n",
2053                                                 __func__, vdevice);
2054                                 return -ENODEV;
2055                         }
2056                 }
2057                 /* do not create a PV cdrom device if we are an HVM guest */
2058                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
2059                 if (IS_ERR(type))
2060                         return -ENODEV;
2061                 if (strncmp(type, "cdrom", 5) == 0) {
2062                         kfree(type);
2063                         return -ENODEV;
2064                 }
2065                 kfree(type);
2066         }
2067         info = kzalloc(sizeof(*info), GFP_KERNEL);
2068         if (!info) {
2069                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2070                 return -ENOMEM;
2071         }
2072
2073         info->xbdev = dev;
2074
2075         mutex_init(&info->mutex);
2076         info->vdevice = vdevice;
2077         info->connected = BLKIF_STATE_DISCONNECTED;
2078
2079         /* Front end dir is a number, which is used as the id. */
2080         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2081         dev_set_drvdata(&dev->dev, info);
2082
2083         mutex_lock(&blkfront_mutex);
2084         list_add(&info->info_list, &info_list);
2085         mutex_unlock(&blkfront_mutex);
2086
2087         return 0;
2088 }
2089
2090 static int blkif_recover(struct blkfront_info *info)
2091 {
2092         unsigned int r_index;
2093         struct request *req, *n;
2094         int rc;
2095         struct bio *bio;
2096         unsigned int segs;
2097
2098         blkfront_gather_backend_features(info);
2099         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2100         blkif_set_queue_limits(info);
2101         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2102         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2103
2104         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2105                 struct blkfront_ring_info *rinfo = &info->rinfo[r_index];
2106
2107                 rc = blkfront_setup_indirect(rinfo);
2108                 if (rc)
2109                         return rc;
2110         }
2111         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2112
2113         /* Now safe for us to use the shared ring */
2114         info->connected = BLKIF_STATE_CONNECTED;
2115
2116         for (r_index = 0; r_index < info->nr_rings; r_index++) {
2117                 struct blkfront_ring_info *rinfo;
2118
2119                 rinfo = &info->rinfo[r_index];
2120                 /* Kick any other new requests queued since we resumed */
2121                 kick_pending_request_queues(rinfo);
2122         }
2123
2124         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2125                 /* Requeue pending requests (flush or discard) */
2126                 list_del_init(&req->queuelist);
2127                 BUG_ON(req->nr_phys_segments > segs);
2128                 blk_mq_requeue_request(req, false);
2129         }
2130         blk_mq_start_stopped_hw_queues(info->rq, true);
2131         blk_mq_kick_requeue_list(info->rq);
2132
2133         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2134                 /* Traverse the list of pending bios and re-queue them */
2135                 submit_bio(bio);
2136         }
2137
2138         return 0;
2139 }
2140
2141 /**
2142  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2143  * driver restart.  We tear down our blkif structure and recreate it, but
2144  * leave the device-layer structures intact so that this is transparent to the
2145  * rest of the kernel.
2146  */
2147 static int blkfront_resume(struct xenbus_device *dev)
2148 {
2149         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2150         int err = 0;
2151         unsigned int i, j;
2152
2153         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2154
2155         bio_list_init(&info->bio_list);
2156         INIT_LIST_HEAD(&info->requests);
2157         for (i = 0; i < info->nr_rings; i++) {
2158                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2159                 struct bio_list merge_bio;
2160                 struct blk_shadow *shadow = rinfo->shadow;
2161
2162                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2163                         /* Not in use? */
2164                         if (!shadow[j].request)
2165                                 continue;
2166
2167                         /*
2168                          * Get the bios in the request so we can re-queue them.
2169                          */
2170                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2171                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2172                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2173                             shadow[j].request->cmd_flags & REQ_FUA) {
2174                                 /*
2175                                  * Flush operations don't contain bios, so
2176                                  * we need to requeue the whole request
2177                                  *
2178                                  * XXX: but this doesn't make any sense for a
2179                                  * write with the FUA flag set..
2180                                  */
2181                                 list_add(&shadow[j].request->queuelist, &info->requests);
2182                                 continue;
2183                         }
2184                         merge_bio.head = shadow[j].request->bio;
2185                         merge_bio.tail = shadow[j].request->biotail;
2186                         bio_list_merge(&info->bio_list, &merge_bio);
2187                         shadow[j].request->bio = NULL;
2188                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2189                 }
2190         }
2191
2192         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2193
2194         err = talk_to_blkback(dev, info);
2195         if (!err)
2196                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2197
2198         /*
2199          * We have to wait for the backend to switch to
2200          * connected state, since we want to read which
2201          * features it supports.
2202          */
2203
2204         return err;
2205 }
2206
2207 static void blkfront_closing(struct blkfront_info *info)
2208 {
2209         struct xenbus_device *xbdev = info->xbdev;
2210         struct block_device *bdev = NULL;
2211
2212         mutex_lock(&info->mutex);
2213
2214         if (xbdev->state == XenbusStateClosing) {
2215                 mutex_unlock(&info->mutex);
2216                 return;
2217         }
2218
2219         if (info->gd)
2220                 bdev = bdget_disk(info->gd, 0);
2221
2222         mutex_unlock(&info->mutex);
2223
2224         if (!bdev) {
2225                 xenbus_frontend_closed(xbdev);
2226                 return;
2227         }
2228
2229         mutex_lock(&bdev->bd_mutex);
2230
2231         if (bdev->bd_openers) {
2232                 xenbus_dev_error(xbdev, -EBUSY,
2233                                  "Device in use; refusing to close");
2234                 xenbus_switch_state(xbdev, XenbusStateClosing);
2235         } else {
2236                 xlvbd_release_gendisk(info);
2237                 xenbus_frontend_closed(xbdev);
2238         }
2239
2240         mutex_unlock(&bdev->bd_mutex);
2241         bdput(bdev);
2242 }
2243
2244 static void blkfront_setup_discard(struct blkfront_info *info)
2245 {
2246         info->feature_discard = 1;
2247         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2248                                                          "discard-granularity",
2249                                                          0);
2250         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2251                                                        "discard-alignment", 0);
2252         info->feature_secdiscard =
2253                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2254                                        0);
2255 }
2256
2257 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2258 {
2259         unsigned int psegs, grants, memflags;
2260         int err, i;
2261         struct blkfront_info *info = rinfo->dev_info;
2262
2263         memflags = memalloc_noio_save();
2264
2265         if (info->max_indirect_segments == 0) {
2266                 if (!HAS_EXTRA_REQ)
2267                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2268                 else {
2269                         /*
2270                          * When an extra req is required, the maximum
2271                          * grants supported is related to the size of the
2272                          * Linux block segment.
2273                          */
2274                         grants = GRANTS_PER_PSEG;
2275                 }
2276         }
2277         else
2278                 grants = info->max_indirect_segments;
2279         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2280
2281         err = fill_grant_buffer(rinfo,
2282                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2283         if (err)
2284                 goto out_of_memory;
2285
2286         if (!info->bounce && info->max_indirect_segments) {
2287                 /*
2288                  * We are using indirect descriptors but don't have a bounce
2289                  * buffer, we need to allocate a set of pages that can be
2290                  * used for mapping indirect grefs
2291                  */
2292                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2293
2294                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2295                 for (i = 0; i < num; i++) {
2296                         struct page *indirect_page = alloc_page(GFP_KERNEL |
2297                                                                 __GFP_ZERO);
2298                         if (!indirect_page)
2299                                 goto out_of_memory;
2300                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2301                 }
2302         }
2303
2304         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2305                 rinfo->shadow[i].grants_used =
2306                         kvcalloc(grants,
2307                                  sizeof(rinfo->shadow[i].grants_used[0]),
2308                                  GFP_KERNEL);
2309                 rinfo->shadow[i].sg = kvcalloc(psegs,
2310                                                sizeof(rinfo->shadow[i].sg[0]),
2311                                                GFP_KERNEL);
2312                 if (info->max_indirect_segments)
2313                         rinfo->shadow[i].indirect_grants =
2314                                 kvcalloc(INDIRECT_GREFS(grants),
2315                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2316                                          GFP_KERNEL);
2317                 if ((rinfo->shadow[i].grants_used == NULL) ||
2318                         (rinfo->shadow[i].sg == NULL) ||
2319                      (info->max_indirect_segments &&
2320                      (rinfo->shadow[i].indirect_grants == NULL)))
2321                         goto out_of_memory;
2322                 sg_init_table(rinfo->shadow[i].sg, psegs);
2323         }
2324
2325         memalloc_noio_restore(memflags);
2326
2327         return 0;
2328
2329 out_of_memory:
2330         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2331                 kvfree(rinfo->shadow[i].grants_used);
2332                 rinfo->shadow[i].grants_used = NULL;
2333                 kvfree(rinfo->shadow[i].sg);
2334                 rinfo->shadow[i].sg = NULL;
2335                 kvfree(rinfo->shadow[i].indirect_grants);
2336                 rinfo->shadow[i].indirect_grants = NULL;
2337         }
2338         if (!list_empty(&rinfo->indirect_pages)) {
2339                 struct page *indirect_page, *n;
2340                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2341                         list_del(&indirect_page->lru);
2342                         __free_page(indirect_page);
2343                 }
2344         }
2345
2346         memalloc_noio_restore(memflags);
2347
2348         return -ENOMEM;
2349 }
2350
2351 /*
2352  * Gather all backend feature-*
2353  */
2354 static void blkfront_gather_backend_features(struct blkfront_info *info)
2355 {
2356         unsigned int indirect_segments;
2357
2358         info->feature_flush = 0;
2359         info->feature_fua = 0;
2360
2361         /*
2362          * If there's no "feature-barrier" defined, then it means
2363          * we're dealing with a very old backend which writes
2364          * synchronously; nothing to do.
2365          *
2366          * If there are barriers, then we use flush.
2367          */
2368         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2369                 info->feature_flush = 1;
2370                 info->feature_fua = 1;
2371         }
2372
2373         /*
2374          * And if there is "feature-flush-cache" use that above
2375          * barriers.
2376          */
2377         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2378                                  0)) {
2379                 info->feature_flush = 1;
2380                 info->feature_fua = 0;
2381         }
2382
2383         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2384                 blkfront_setup_discard(info);
2385
2386         info->feature_persistent =
2387                 !!xenbus_read_unsigned(info->xbdev->otherend,
2388                                        "feature-persistent", 0);
2389         if (info->feature_persistent)
2390                 info->bounce = true;
2391
2392         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2393                                         "feature-max-indirect-segments", 0);
2394         if (indirect_segments > xen_blkif_max_segments)
2395                 indirect_segments = xen_blkif_max_segments;
2396         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2397                 indirect_segments = 0;
2398         info->max_indirect_segments = indirect_segments;
2399
2400         if (info->feature_persistent) {
2401                 mutex_lock(&blkfront_mutex);
2402                 schedule_delayed_work(&blkfront_work, HZ * 10);
2403                 mutex_unlock(&blkfront_mutex);
2404         }
2405 }
2406
2407 /*
2408  * Invoked when the backend is finally 'ready' (and has told produced
2409  * the details about the physical device - #sectors, size, etc).
2410  */
2411 static void blkfront_connect(struct blkfront_info *info)
2412 {
2413         unsigned long long sectors;
2414         unsigned long sector_size;
2415         unsigned int physical_sector_size;
2416         unsigned int binfo;
2417         char *envp[] = { "RESIZE=1", NULL };
2418         int err, i;
2419
2420         switch (info->connected) {
2421         case BLKIF_STATE_CONNECTED:
2422                 /*
2423                  * Potentially, the back-end may be signalling
2424                  * a capacity change; update the capacity.
2425                  */
2426                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2427                                    "sectors", "%Lu", &sectors);
2428                 if (XENBUS_EXIST_ERR(err))
2429                         return;
2430                 printk(KERN_INFO "Setting capacity to %Lu\n",
2431                        sectors);
2432                 set_capacity(info->gd, sectors);
2433                 revalidate_disk(info->gd);
2434                 kobject_uevent_env(&disk_to_dev(info->gd)->kobj,
2435                                    KOBJ_CHANGE, envp);
2436
2437                 return;
2438         case BLKIF_STATE_SUSPENDED:
2439                 /*
2440                  * If we are recovering from suspension, we need to wait
2441                  * for the backend to announce it's features before
2442                  * reconnecting, at least we need to know if the backend
2443                  * supports indirect descriptors, and how many.
2444                  */
2445                 blkif_recover(info);
2446                 return;
2447
2448         default:
2449                 break;
2450         }
2451
2452         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2453                 __func__, info->xbdev->otherend);
2454
2455         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2456                             "sectors", "%llu", &sectors,
2457                             "info", "%u", &binfo,
2458                             "sector-size", "%lu", &sector_size,
2459                             NULL);
2460         if (err) {
2461                 xenbus_dev_fatal(info->xbdev, err,
2462                                  "reading backend fields at %s",
2463                                  info->xbdev->otherend);
2464                 return;
2465         }
2466
2467         /*
2468          * physcial-sector-size is a newer field, so old backends may not
2469          * provide this. Assume physical sector size to be the same as
2470          * sector_size in that case.
2471          */
2472         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2473                                                     "physical-sector-size",
2474                                                     sector_size);
2475         blkfront_gather_backend_features(info);
2476         for (i = 0; i < info->nr_rings; i++) {
2477                 err = blkfront_setup_indirect(&info->rinfo[i]);
2478                 if (err) {
2479                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2480                                          info->xbdev->otherend);
2481                         blkif_free(info, 0);
2482                         break;
2483                 }
2484         }
2485
2486         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2487                                   physical_sector_size);
2488         if (err) {
2489                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2490                                  info->xbdev->otherend);
2491                 goto fail;
2492         }
2493
2494         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2495
2496         /* Kick pending requests. */
2497         info->connected = BLKIF_STATE_CONNECTED;
2498         for (i = 0; i < info->nr_rings; i++)
2499                 kick_pending_request_queues(&info->rinfo[i]);
2500
2501         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2502
2503         info->is_ready = 1;
2504         return;
2505
2506 fail:
2507         blkif_free(info, 0);
2508         return;
2509 }
2510
2511 /**
2512  * Callback received when the backend's state changes.
2513  */
2514 static void blkback_changed(struct xenbus_device *dev,
2515                             enum xenbus_state backend_state)
2516 {
2517         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2518
2519         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2520
2521         switch (backend_state) {
2522         case XenbusStateInitWait:
2523                 if (dev->state != XenbusStateInitialising)
2524                         break;
2525                 if (talk_to_blkback(dev, info))
2526                         break;
2527         case XenbusStateInitialising:
2528         case XenbusStateInitialised:
2529         case XenbusStateReconfiguring:
2530         case XenbusStateReconfigured:
2531         case XenbusStateUnknown:
2532                 break;
2533
2534         case XenbusStateConnected:
2535                 /*
2536                  * talk_to_blkback sets state to XenbusStateInitialised
2537                  * and blkfront_connect sets it to XenbusStateConnected
2538                  * (if connection went OK).
2539                  *
2540                  * If the backend (or toolstack) decides to poke at backend
2541                  * state (and re-trigger the watch by setting the state repeatedly
2542                  * to XenbusStateConnected (4)) we need to deal with this.
2543                  * This is allowed as this is used to communicate to the guest
2544                  * that the size of disk has changed!
2545                  */
2546                 if ((dev->state != XenbusStateInitialised) &&
2547                     (dev->state != XenbusStateConnected)) {
2548                         if (talk_to_blkback(dev, info))
2549                                 break;
2550                 }
2551
2552                 blkfront_connect(info);
2553                 break;
2554
2555         case XenbusStateClosed:
2556                 if (dev->state == XenbusStateClosed)
2557                         break;
2558                 /* fall through */
2559         case XenbusStateClosing:
2560                 if (info)
2561                         blkfront_closing(info);
2562                 break;
2563         }
2564 }
2565
2566 static int blkfront_remove(struct xenbus_device *xbdev)
2567 {
2568         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2569         struct block_device *bdev = NULL;
2570         struct gendisk *disk;
2571
2572         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2573
2574         if (!info)
2575                 return 0;
2576
2577         blkif_free(info, 0);
2578
2579         mutex_lock(&info->mutex);
2580
2581         disk = info->gd;
2582         if (disk)
2583                 bdev = bdget_disk(disk, 0);
2584
2585         info->xbdev = NULL;
2586         mutex_unlock(&info->mutex);
2587
2588         if (!bdev) {
2589                 mutex_lock(&blkfront_mutex);
2590                 free_info(info);
2591                 mutex_unlock(&blkfront_mutex);
2592                 return 0;
2593         }
2594
2595         /*
2596          * The xbdev was removed before we reached the Closed
2597          * state. See if it's safe to remove the disk. If the bdev
2598          * isn't closed yet, we let release take care of it.
2599          */
2600
2601         mutex_lock(&bdev->bd_mutex);
2602         info = disk->private_data;
2603
2604         dev_warn(disk_to_dev(disk),
2605                  "%s was hot-unplugged, %d stale handles\n",
2606                  xbdev->nodename, bdev->bd_openers);
2607
2608         if (info && !bdev->bd_openers) {
2609                 xlvbd_release_gendisk(info);
2610                 disk->private_data = NULL;
2611                 mutex_lock(&blkfront_mutex);
2612                 free_info(info);
2613                 mutex_unlock(&blkfront_mutex);
2614         }
2615
2616         mutex_unlock(&bdev->bd_mutex);
2617         bdput(bdev);
2618
2619         return 0;
2620 }
2621
2622 static int blkfront_is_ready(struct xenbus_device *dev)
2623 {
2624         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2625
2626         return info->is_ready && info->xbdev;
2627 }
2628
2629 static int blkif_open(struct block_device *bdev, fmode_t mode)
2630 {
2631         struct gendisk *disk = bdev->bd_disk;
2632         struct blkfront_info *info;
2633         int err = 0;
2634
2635         mutex_lock(&blkfront_mutex);
2636
2637         info = disk->private_data;
2638         if (!info) {
2639                 /* xbdev gone */
2640                 err = -ERESTARTSYS;
2641                 goto out;
2642         }
2643
2644         mutex_lock(&info->mutex);
2645
2646         if (!info->gd)
2647                 /* xbdev is closed */
2648                 err = -ERESTARTSYS;
2649
2650         mutex_unlock(&info->mutex);
2651
2652 out:
2653         mutex_unlock(&blkfront_mutex);
2654         return err;
2655 }
2656
2657 static void blkif_release(struct gendisk *disk, fmode_t mode)
2658 {
2659         struct blkfront_info *info = disk->private_data;
2660         struct block_device *bdev;
2661         struct xenbus_device *xbdev;
2662
2663         mutex_lock(&blkfront_mutex);
2664
2665         bdev = bdget_disk(disk, 0);
2666
2667         if (!bdev) {
2668                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2669                 goto out_mutex;
2670         }
2671         if (bdev->bd_openers)
2672                 goto out;
2673
2674         /*
2675          * Check if we have been instructed to close. We will have
2676          * deferred this request, because the bdev was still open.
2677          */
2678
2679         mutex_lock(&info->mutex);
2680         xbdev = info->xbdev;
2681
2682         if (xbdev && xbdev->state == XenbusStateClosing) {
2683                 /* pending switch to state closed */
2684                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2685                 xlvbd_release_gendisk(info);
2686                 xenbus_frontend_closed(info->xbdev);
2687         }
2688
2689         mutex_unlock(&info->mutex);
2690
2691         if (!xbdev) {
2692                 /* sudden device removal */
2693                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2694                 xlvbd_release_gendisk(info);
2695                 disk->private_data = NULL;
2696                 free_info(info);
2697         }
2698
2699 out:
2700         bdput(bdev);
2701 out_mutex:
2702         mutex_unlock(&blkfront_mutex);
2703 }
2704
2705 static const struct block_device_operations xlvbd_block_fops =
2706 {
2707         .owner = THIS_MODULE,
2708         .open = blkif_open,
2709         .release = blkif_release,
2710         .getgeo = blkif_getgeo,
2711         .ioctl = blkif_ioctl,
2712 };
2713
2714
2715 static const struct xenbus_device_id blkfront_ids[] = {
2716         { "vbd" },
2717         { "" }
2718 };
2719
2720 static struct xenbus_driver blkfront_driver = {
2721         .ids  = blkfront_ids,
2722         .probe = blkfront_probe,
2723         .remove = blkfront_remove,
2724         .resume = blkfront_resume,
2725         .otherend_changed = blkback_changed,
2726         .is_ready = blkfront_is_ready,
2727 };
2728
2729 static void purge_persistent_grants(struct blkfront_info *info)
2730 {
2731         unsigned int i;
2732         unsigned long flags;
2733
2734         for (i = 0; i < info->nr_rings; i++) {
2735                 struct blkfront_ring_info *rinfo = &info->rinfo[i];
2736                 struct grant *gnt_list_entry, *tmp;
2737
2738                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2739
2740                 if (rinfo->persistent_gnts_c == 0) {
2741                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2742                         continue;
2743                 }
2744
2745                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2746                                          node) {
2747                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2748                             !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2749                                 continue;
2750
2751                         list_del(&gnt_list_entry->node);
2752                         rinfo->persistent_gnts_c--;
2753                         gnt_list_entry->gref = GRANT_INVALID_REF;
2754                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2755                 }
2756
2757                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2758         }
2759 }
2760
2761 static void blkfront_delay_work(struct work_struct *work)
2762 {
2763         struct blkfront_info *info;
2764         bool need_schedule_work = false;
2765
2766         /*
2767          * Note that when using bounce buffers but not persistent grants
2768          * there's no need to run blkfront_delay_work because grants are
2769          * revoked in blkif_completion or else an error is reported and the
2770          * connection is closed.
2771          */
2772
2773         mutex_lock(&blkfront_mutex);
2774
2775         list_for_each_entry(info, &info_list, info_list) {
2776                 if (info->feature_persistent) {
2777                         need_schedule_work = true;
2778                         mutex_lock(&info->mutex);
2779                         purge_persistent_grants(info);
2780                         mutex_unlock(&info->mutex);
2781                 }
2782         }
2783
2784         if (need_schedule_work)
2785                 schedule_delayed_work(&blkfront_work, HZ * 10);
2786
2787         mutex_unlock(&blkfront_mutex);
2788 }
2789
2790 static int __init xlblk_init(void)
2791 {
2792         int ret;
2793         int nr_cpus = num_online_cpus();
2794
2795         if (!xen_domain())
2796                 return -ENODEV;
2797
2798         if (!xen_has_pv_disk_devices())
2799                 return -ENODEV;
2800
2801         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2802                 pr_warn("xen_blk: can't get major %d with name %s\n",
2803                         XENVBD_MAJOR, DEV_NAME);
2804                 return -ENODEV;
2805         }
2806
2807         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2808                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2809
2810         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2811                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2812                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2813                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2814         }
2815
2816         if (xen_blkif_max_queues > nr_cpus) {
2817                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2818                         xen_blkif_max_queues, nr_cpus);
2819                 xen_blkif_max_queues = nr_cpus;
2820         }
2821
2822         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2823
2824         ret = xenbus_register_frontend(&blkfront_driver);
2825         if (ret) {
2826                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2827                 return ret;
2828         }
2829
2830         return 0;
2831 }
2832 module_init(xlblk_init);
2833
2834
2835 static void __exit xlblk_exit(void)
2836 {
2837         cancel_delayed_work_sync(&blkfront_work);
2838
2839         xenbus_unregister_driver(&blkfront_driver);
2840         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2841         kfree(minors);
2842 }
2843 module_exit(xlblk_exit);
2844
2845 MODULE_DESCRIPTION("Xen virtual block device frontend");
2846 MODULE_LICENSE("GPL");
2847 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2848 MODULE_ALIAS("xen:vbd");
2849 MODULE_ALIAS("xenblk");