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 /*
 162  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 163  * characters are enough. Define to 20 to keep consistent with backend.
 164  */
 165 #define RINGREF_NAME_LEN (20)
 166 /*
 167  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 168  */
 169 #define QUEUE_NAME_LEN (17)
 170
 171 /*
 172  *  Per-ring info.
 173  *  Every blkfront device can associate with one or more blkfront_ring_info,
 174  *  depending on how many hardware queues/rings to be used.
 175  */
 176 struct blkfront_ring_info {
 177         /* Lock to protect data in every ring buffer. */
 178         spinlock_t ring_lock;
 179         struct blkif_front_ring ring;
 180         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 181         unsigned int evtchn, irq;
 182         struct work_struct work;
 183         struct gnttab_free_callback callback;
 184         struct list_head indirect_pages;
 185         struct list_head grants;
 186         unsigned int persistent_gnts_c;
 187         unsigned long shadow_free;
 188         struct blkfront_info *dev_info;
 189         struct blk_shadow shadow[];
 190 };
 191
 192 /*
 193  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 194  * hang in private_data off the gendisk structure. We may end up
 195  * putting all kinds of interesting stuff here :-)
 196  */
 197 struct blkfront_info
 198 {
 199         struct mutex mutex;
 200         struct xenbus_device *xbdev;
 201         struct gendisk *gd;
 202         u16 sector_size;
 203         unsigned int physical_sector_size;
 204         int vdevice;
 205         blkif_vdev_t handle;
 206         enum blkif_state connected;
 207         /* Number of pages per ring buffer. */
 208         unsigned int nr_ring_pages;
 209         struct request_queue *rq;
 210         unsigned int feature_flush:1;
 211         unsigned int feature_fua:1;
 212         unsigned int feature_discard:1;
 213         unsigned int feature_secdiscard:1;
 214         /* Connect-time cached feature_persistent parameter */
 215         unsigned int feature_persistent_parm:1;
 216         /* Persistent grants feature negotiation result */
 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         unsigned int rinfo_size;
 228         /* Save uncomplete reqs and bios for migration. */
 229         struct list_head requests;
 230         struct bio_list bio_list;
 231         struct list_head info_list;
 232 };
 233
 234 static unsigned int nr_minors;
 235 static unsigned long *minors;
 236 static DEFINE_SPINLOCK(minor_lock);
 237
 238 #define GRANT_INVALID_REF       0
 239
 240 #define PARTS_PER_DISK          16
 241 #define PARTS_PER_EXT_DISK      256
 242
 243 #define BLKIF_MAJOR(dev) ((dev)>>8)
 244 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 245
 246 #define EXT_SHIFT 28
 247 #define EXTENDED (1<<EXT_SHIFT)
 248 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 249 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 250 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 251 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 252 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 253 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 254
 255 #define DEV_NAME        "xvd"   /* name in /dev */
 256
 257 /*
 258  * Grants are always the same size as a Xen page (i.e 4KB).
 259  * A physical segment is always the same size as a Linux page.
 260  * Number of grants per physical segment
 261  */
 262 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 263
 264 #define GRANTS_PER_INDIRECT_FRAME \
 265         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 266
 267 #define INDIRECT_GREFS(_grants)         \
 268         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 269
 270 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 271 static void blkfront_gather_backend_features(struct blkfront_info *info);
 272 static int negotiate_mq(struct blkfront_info *info);
 273
 274 #define for_each_rinfo(info, ptr, idx)                          \
 275         for ((ptr) = (info)->rinfo, (idx) = 0;                  \
 276              (idx) < (info)->nr_rings;                          \
 277              (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 278
 279 static inline struct blkfront_ring_info *
 280 get_rinfo(const struct blkfront_info *info, unsigned int i)
 281 {
 282         BUG_ON(i >= info->nr_rings);
 283         return (void *)info->rinfo + i * info->rinfo_size;
 284 }
 285
 286 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 287 {
 288         unsigned long free = rinfo->shadow_free;
 289
 290         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 291         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 292         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 293         return free;
 294 }
 295
 296 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 297                               unsigned long id)
 298 {
 299         if (rinfo->shadow[id].req.u.rw.id != id)
 300                 return -EINVAL;
 301         if (rinfo->shadow[id].request == NULL)
 302                 return -EINVAL;
 303         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 304         rinfo->shadow[id].request = NULL;
 305         rinfo->shadow_free = id;
 306         return 0;
 307 }
 308
 309 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 310 {
 311         struct blkfront_info *info = rinfo->dev_info;
 312         struct page *granted_page;
 313         struct grant *gnt_list_entry, *n;
 314         int i = 0;
 315
 316         while (i < num) {
 317                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 318                 if (!gnt_list_entry)
 319                         goto out_of_memory;
 320
 321                 if (info->bounce) {
 322                         granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
 323                         if (!granted_page) {
 324                                 kfree(gnt_list_entry);
 325                                 goto out_of_memory;
 326                         }
 327                         gnt_list_entry->page = granted_page;
 328                 }
 329
 330                 gnt_list_entry->gref = GRANT_INVALID_REF;
 331                 list_add(&gnt_list_entry->node, &rinfo->grants);
 332                 i++;
 333         }
 334
 335         return 0;
 336
 337 out_of_memory:
 338         list_for_each_entry_safe(gnt_list_entry, n,
 339                                  &rinfo->grants, node) {
 340                 list_del(&gnt_list_entry->node);
 341                 if (info->bounce)
 342                         __free_page(gnt_list_entry->page);
 343                 kfree(gnt_list_entry);
 344                 i--;
 345         }
 346         BUG_ON(i != 0);
 347         return -ENOMEM;
 348 }
 349
 350 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 351 {
 352         struct grant *gnt_list_entry;
 353
 354         BUG_ON(list_empty(&rinfo->grants));
 355         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 356                                           node);
 357         list_del(&gnt_list_entry->node);
 358
 359         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 360                 rinfo->persistent_gnts_c--;
 361
 362         return gnt_list_entry;
 363 }
 364
 365 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 366                                         const struct blkfront_info *info)
 367 {
 368         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 369                                                  info->xbdev->otherend_id,
 370                                                  gnt_list_entry->page,
 371                                                  0);
 372 }
 373
 374 static struct grant *get_grant(grant_ref_t *gref_head,
 375                                unsigned long gfn,
 376                                struct blkfront_ring_info *rinfo)
 377 {
 378         struct grant *gnt_list_entry = get_free_grant(rinfo);
 379         struct blkfront_info *info = rinfo->dev_info;
 380
 381         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 382                 return gnt_list_entry;
 383
 384         /* Assign a gref to this page */
 385         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 386         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 387         if (info->bounce)
 388                 grant_foreign_access(gnt_list_entry, info);
 389         else {
 390                 /* Grant access to the GFN passed by the caller */
 391                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 392                                                 info->xbdev->otherend_id,
 393                                                 gfn, 0);
 394         }
 395
 396         return gnt_list_entry;
 397 }
 398
 399 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 400                                         struct blkfront_ring_info *rinfo)
 401 {
 402         struct grant *gnt_list_entry = get_free_grant(rinfo);
 403         struct blkfront_info *info = rinfo->dev_info;
 404
 405         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 406                 return gnt_list_entry;
 407
 408         /* Assign a gref to this page */
 409         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 410         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 411         if (!info->bounce) {
 412                 struct page *indirect_page;
 413
 414                 /* Fetch a pre-allocated page to use for indirect grefs */
 415                 BUG_ON(list_empty(&rinfo->indirect_pages));
 416                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 417                                                  struct page, lru);
 418                 list_del(&indirect_page->lru);
 419                 gnt_list_entry->page = indirect_page;
 420         }
 421         grant_foreign_access(gnt_list_entry, info);
 422
 423         return gnt_list_entry;
 424 }
 425
 426 static const char *op_name(int op)
 427 {
 428         static const char *const names[] = {
 429                 [BLKIF_OP_READ] = "read",
 430                 [BLKIF_OP_WRITE] = "write",
 431                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 432                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 433                 [BLKIF_OP_DISCARD] = "discard" };
 434
 435         if (op < 0 || op >= ARRAY_SIZE(names))
 436                 return "unknown";
 437
 438         if (!names[op])
 439                 return "reserved";
 440
 441         return names[op];
 442 }
 443 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 444 {
 445         unsigned int end = minor + nr;
 446         int rc;
 447
 448         if (end > nr_minors) {
 449                 unsigned long *bitmap, *old;
 450
 451                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 452                                  GFP_KERNEL);
 453                 if (bitmap == NULL)
 454                         return -ENOMEM;
 455
 456                 spin_lock(&minor_lock);
 457                 if (end > nr_minors) {
 458                         old = minors;
 459                         memcpy(bitmap, minors,
 460                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 461                         minors = bitmap;
 462                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 463                 } else
 464                         old = bitmap;
 465                 spin_unlock(&minor_lock);
 466                 kfree(old);
 467         }
 468
 469         spin_lock(&minor_lock);
 470         if (find_next_bit(minors, end, minor) >= end) {
 471                 bitmap_set(minors, minor, nr);
 472                 rc = 0;
 473         } else
 474                 rc = -EBUSY;
 475         spin_unlock(&minor_lock);
 476
 477         return rc;
 478 }
 479
 480 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 481 {
 482         unsigned int end = minor + nr;
 483
 484         BUG_ON(end > nr_minors);
 485         spin_lock(&minor_lock);
 486         bitmap_clear(minors,  minor, nr);
 487         spin_unlock(&minor_lock);
 488 }
 489
 490 static void blkif_restart_queue_callback(void *arg)
 491 {
 492         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 493         schedule_work(&rinfo->work);
 494 }
 495
 496 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 497 {
 498         /* We don't have real geometry info, but let's at least return
 499            values consistent with the size of the device */
 500         sector_t nsect = get_capacity(bd->bd_disk);
 501         sector_t cylinders = nsect;
 502
 503         hg->heads = 0xff;
 504         hg->sectors = 0x3f;
 505         sector_div(cylinders, hg->heads * hg->sectors);
 506         hg->cylinders = cylinders;
 507         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 508                 hg->cylinders = 0xffff;
 509         return 0;
 510 }
 511
 512 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 513                        unsigned command, unsigned long argument)
 514 {
 515         struct blkfront_info *info = bdev->bd_disk->private_data;
 516         int i;
 517
 518         dev_dbg(&info->xbdev->dev, "command: 0x%x, argument: 0x%lx\n",
 519                 command, (long)argument);
 520
 521         switch (command) {
 522         case CDROMMULTISESSION:
 523                 dev_dbg(&info->xbdev->dev, "FIXME: support multisession CDs later\n");
 524                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 525                         if (put_user(0, (char __user *)(argument + i)))
 526                                 return -EFAULT;
 527                 return 0;
 528
 529         case CDROM_GET_CAPABILITY: {
 530                 struct gendisk *gd = info->gd;
 531                 if (gd->flags & GENHD_FL_CD)
 532                         return 0;
 533                 return -EINVAL;
 534         }
 535
 536         default:
 537                 /*printk(KERN_ALERT "ioctl %08x not supported by Xen blkdev\n",
 538                   command);*/
 539                 return -EINVAL; /* same return as native Linux */
 540         }
 541
 542         return 0;
 543 }
 544
 545 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 546                                             struct request *req,
 547                                             struct blkif_request **ring_req)
 548 {
 549         unsigned long id;
 550
 551         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 552         rinfo->ring.req_prod_pvt++;
 553
 554         id = get_id_from_freelist(rinfo);
 555         rinfo->shadow[id].request = req;
 556         rinfo->shadow[id].status = REQ_PROCESSING;
 557         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 558
 559         rinfo->shadow[id].req.u.rw.id = id;
 560
 561         return id;
 562 }
 563
 564 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 565 {
 566         struct blkfront_info *info = rinfo->dev_info;
 567         struct blkif_request *ring_req, *final_ring_req;
 568         unsigned long id;
 569
 570         /* Fill out a communications ring structure. */
 571         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 572         ring_req = &rinfo->shadow[id].req;
 573
 574         ring_req->operation = BLKIF_OP_DISCARD;
 575         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 576         ring_req->u.discard.id = id;
 577         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 578         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 579                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 580         else
 581                 ring_req->u.discard.flag = 0;
 582
 583         /* Copy the request to the ring page. */
 584         *final_ring_req = *ring_req;
 585         rinfo->shadow[id].status = REQ_WAITING;
 586
 587         return 0;
 588 }
 589
 590 struct setup_rw_req {
 591         unsigned int grant_idx;
 592         struct blkif_request_segment *segments;
 593         struct blkfront_ring_info *rinfo;
 594         struct blkif_request *ring_req;
 595         grant_ref_t gref_head;
 596         unsigned int id;
 597         /* Only used when persistent grant is used and it's a read request */
 598         bool need_copy;
 599         unsigned int bvec_off;
 600         char *bvec_data;
 601
 602         bool require_extra_req;
 603         struct blkif_request *extra_ring_req;
 604 };
 605
 606 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 607                                      unsigned int len, void *data)
 608 {
 609         struct setup_rw_req *setup = data;
 610         int n, ref;
 611         struct grant *gnt_list_entry;
 612         unsigned int fsect, lsect;
 613         /* Convenient aliases */
 614         unsigned int grant_idx = setup->grant_idx;
 615         struct blkif_request *ring_req = setup->ring_req;
 616         struct blkfront_ring_info *rinfo = setup->rinfo;
 617         /*
 618          * We always use the shadow of the first request to store the list
 619          * of grant associated to the block I/O request. This made the
 620          * completion more easy to handle even if the block I/O request is
 621          * split.
 622          */
 623         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 624
 625         if (unlikely(setup->require_extra_req &&
 626                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 627                 /*
 628                  * We are using the second request, setup grant_idx
 629                  * to be the index of the segment array.
 630                  */
 631                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 632                 ring_req = setup->extra_ring_req;
 633         }
 634
 635         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 636             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 637                 if (setup->segments)
 638                         kunmap_atomic(setup->segments);
 639
 640                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 641                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 642                 shadow->indirect_grants[n] = gnt_list_entry;
 643                 setup->segments = kmap_atomic(gnt_list_entry->page);
 644                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 645         }
 646
 647         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 648         ref = gnt_list_entry->gref;
 649         /*
 650          * All the grants are stored in the shadow of the first
 651          * request. Therefore we have to use the global index.
 652          */
 653         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 654
 655         if (setup->need_copy) {
 656                 void *shared_data;
 657
 658                 shared_data = kmap_atomic(gnt_list_entry->page);
 659                 /*
 660                  * this does not wipe data stored outside the
 661                  * range sg->offset..sg->offset+sg->length.
 662                  * Therefore, blkback *could* see data from
 663                  * previous requests. This is OK as long as
 664                  * persistent grants are shared with just one
 665                  * domain. It may need refactoring if this
 666                  * changes
 667                  */
 668                 memcpy(shared_data + offset,
 669                        setup->bvec_data + setup->bvec_off,
 670                        len);
 671
 672                 kunmap_atomic(shared_data);
 673                 setup->bvec_off += len;
 674         }
 675
 676         fsect = offset >> 9;
 677         lsect = fsect + (len >> 9) - 1;
 678         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 679                 ring_req->u.rw.seg[grant_idx] =
 680                         (struct blkif_request_segment) {
 681                                 .gref       = ref,
 682                                 .first_sect = fsect,
 683                                 .last_sect  = lsect };
 684         } else {
 685                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 686                         (struct blkif_request_segment) {
 687                                 .gref       = ref,
 688                                 .first_sect = fsect,
 689                                 .last_sect  = lsect };
 690         }
 691
 692         (setup->grant_idx)++;
 693 }
 694
 695 static void blkif_setup_extra_req(struct blkif_request *first,
 696                                   struct blkif_request *second)
 697 {
 698         uint16_t nr_segments = first->u.rw.nr_segments;
 699
 700         /*
 701          * The second request is only present when the first request uses
 702          * all its segments. It's always the continuity of the first one.
 703          */
 704         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 705
 706         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 707         second->u.rw.sector_number = first->u.rw.sector_number +
 708                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 709
 710         second->u.rw.handle = first->u.rw.handle;
 711         second->operation = first->operation;
 712 }
 713
 714 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 715 {
 716         struct blkfront_info *info = rinfo->dev_info;
 717         struct blkif_request *ring_req, *extra_ring_req = NULL;
 718         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 719         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 720         bool require_extra_req = false;
 721         int i;
 722         struct setup_rw_req setup = {
 723                 .grant_idx = 0,
 724                 .segments = NULL,
 725                 .rinfo = rinfo,
 726                 .need_copy = rq_data_dir(req) && info->bounce,
 727         };
 728
 729         /*
 730          * Used to store if we are able to queue the request by just using
 731          * existing persistent grants, or if we have to get new grants,
 732          * as there are not sufficiently many free.
 733          */
 734         bool new_persistent_gnts = false;
 735         struct scatterlist *sg;
 736         int num_sg, max_grefs, num_grant;
 737
 738         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 739         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 740                 /*
 741                  * If we are using indirect segments we need to account
 742                  * for the indirect grefs used in the request.
 743                  */
 744                 max_grefs += INDIRECT_GREFS(max_grefs);
 745
 746         /* Check if we have enough persistent grants to allocate a requests */
 747         if (rinfo->persistent_gnts_c < max_grefs) {
 748                 new_persistent_gnts = true;
 749
 750                 if (gnttab_alloc_grant_references(
 751                     max_grefs - rinfo->persistent_gnts_c,
 752                     &setup.gref_head) < 0) {
 753                         gnttab_request_free_callback(
 754                                 &rinfo->callback,
 755                                 blkif_restart_queue_callback,
 756                                 rinfo,
 757                                 max_grefs - rinfo->persistent_gnts_c);
 758                         return 1;
 759                 }
 760         }
 761
 762         /* Fill out a communications ring structure. */
 763         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 764         ring_req = &rinfo->shadow[id].req;
 765
 766         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 767         num_grant = 0;
 768         /* Calculate the number of grant used */
 769         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 770                num_grant += gnttab_count_grant(sg->offset, sg->length);
 771
 772         require_extra_req = info->max_indirect_segments == 0 &&
 773                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 774         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 775
 776         rinfo->shadow[id].num_sg = num_sg;
 777         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 778             likely(!require_extra_req)) {
 779                 /*
 780                  * The indirect operation can only be a BLKIF_OP_READ or
 781                  * BLKIF_OP_WRITE
 782                  */
 783                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 784                 ring_req->operation = BLKIF_OP_INDIRECT;
 785                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 786                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 787                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 788                 ring_req->u.indirect.handle = info->handle;
 789                 ring_req->u.indirect.nr_segments = num_grant;
 790         } else {
 791                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 792                 ring_req->u.rw.handle = info->handle;
 793                 ring_req->operation = rq_data_dir(req) ?
 794                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 795                 if (req_op(req) == REQ_OP_FLUSH ||
 796                     (req_op(req) == REQ_OP_WRITE && (req->cmd_flags & REQ_FUA))) {
 797                         /*
 798                          * Ideally we can do an unordered flush-to-disk.
 799                          * In case the backend onlysupports barriers, use that.
 800                          * A barrier request a superset of FUA, so we can
 801                          * implement it the same way.  (It's also a FLUSH+FUA,
 802                          * since it is guaranteed ordered WRT previous writes.)
 803                          */
 804                         if (info->feature_flush && info->feature_fua)
 805                                 ring_req->operation =
 806                                         BLKIF_OP_WRITE_BARRIER;
 807                         else if (info->feature_flush)
 808                                 ring_req->operation =
 809                                         BLKIF_OP_FLUSH_DISKCACHE;
 810                         else
 811                                 ring_req->operation = 0;
 812                 }
 813                 ring_req->u.rw.nr_segments = num_grant;
 814                 if (unlikely(require_extra_req)) {
 815                         extra_id = blkif_ring_get_request(rinfo, req,
 816                                                           &final_extra_ring_req);
 817                         extra_ring_req = &rinfo->shadow[extra_id].req;
 818
 819                         /*
 820                          * Only the first request contains the scatter-gather
 821                          * list.
 822                          */
 823                         rinfo->shadow[extra_id].num_sg = 0;
 824
 825                         blkif_setup_extra_req(ring_req, extra_ring_req);
 826
 827                         /* Link the 2 requests together */
 828                         rinfo->shadow[extra_id].associated_id = id;
 829                         rinfo->shadow[id].associated_id = extra_id;
 830                 }
 831         }
 832
 833         setup.ring_req = ring_req;
 834         setup.id = id;
 835
 836         setup.require_extra_req = require_extra_req;
 837         if (unlikely(require_extra_req))
 838                 setup.extra_ring_req = extra_ring_req;
 839
 840         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 841                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 842
 843                 if (setup.need_copy) {
 844                         setup.bvec_off = sg->offset;
 845                         setup.bvec_data = kmap_atomic(sg_page(sg));
 846                 }
 847
 848                 gnttab_foreach_grant_in_range(sg_page(sg),
 849                                               sg->offset,
 850                                               sg->length,
 851                                               blkif_setup_rw_req_grant,
 852                                               &setup);
 853
 854                 if (setup.need_copy)
 855                         kunmap_atomic(setup.bvec_data);
 856         }
 857         if (setup.segments)
 858                 kunmap_atomic(setup.segments);
 859
 860         /* Copy request(s) to the ring page. */
 861         *final_ring_req = *ring_req;
 862         rinfo->shadow[id].status = REQ_WAITING;
 863         if (unlikely(require_extra_req)) {
 864                 *final_extra_ring_req = *extra_ring_req;
 865                 rinfo->shadow[extra_id].status = REQ_WAITING;
 866         }
 867
 868         if (new_persistent_gnts)
 869                 gnttab_free_grant_references(setup.gref_head);
 870
 871         return 0;
 872 }
 873
 874 /*
 875  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 876  * and writes are handled as expected.
 877  *
 878  * @req: a request struct
 879  */
 880 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 881 {
 882         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 883                 return 1;
 884
 885         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 886                      req_op(req) == REQ_OP_SECURE_ERASE))
 887                 return blkif_queue_discard_req(req, rinfo);
 888         else
 889                 return blkif_queue_rw_req(req, rinfo);
 890 }
 891
 892 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 893 {
 894         int notify;
 895
 896         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 897
 898         if (notify)
 899                 notify_remote_via_irq(rinfo->irq);
 900 }
 901
 902 static inline bool blkif_request_flush_invalid(struct request *req,
 903                                                struct blkfront_info *info)
 904 {
 905         return (blk_rq_is_passthrough(req) ||
 906                 ((req_op(req) == REQ_OP_FLUSH) &&
 907                  !info->feature_flush) ||
 908                 ((req->cmd_flags & REQ_FUA) &&
 909                  !info->feature_fua));
 910 }
 911
 912 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 913                           const struct blk_mq_queue_data *qd)
 914 {
 915         unsigned long flags;
 916         int qid = hctx->queue_num;
 917         struct blkfront_info *info = hctx->queue->queuedata;
 918         struct blkfront_ring_info *rinfo = NULL;
 919
 920         rinfo = get_rinfo(info, qid);
 921         blk_mq_start_request(qd->rq);
 922         spin_lock_irqsave(&rinfo->ring_lock, flags);
 923         if (RING_FULL(&rinfo->ring))
 924                 goto out_busy;
 925
 926         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 927                 goto out_err;
 928
 929         if (blkif_queue_request(qd->rq, rinfo))
 930                 goto out_busy;
 931
 932         flush_requests(rinfo);
 933         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 934         return BLK_STS_OK;
 935
 936 out_err:
 937         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 938         return BLK_STS_IOERR;
 939
 940 out_busy:
 941         blk_mq_stop_hw_queue(hctx);
 942         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 943         return BLK_STS_DEV_RESOURCE;
 944 }
 945
 946 static void blkif_complete_rq(struct request *rq)
 947 {
 948         blk_mq_end_request(rq, blkif_req(rq)->error);
 949 }
 950
 951 static const struct blk_mq_ops blkfront_mq_ops = {
 952         .queue_rq = blkif_queue_rq,
 953         .complete = blkif_complete_rq,
 954 };
 955
 956 static void blkif_set_queue_limits(struct blkfront_info *info)
 957 {
 958         struct request_queue *rq = info->rq;
 959         struct gendisk *gd = info->gd;
 960         unsigned int segments = info->max_indirect_segments ? :
 961                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 962
 963         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 964
 965         if (info->feature_discard) {
 966                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 967                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 968                 rq->limits.discard_granularity = info->discard_granularity ?:
 969                                                  info->physical_sector_size;
 970                 rq->limits.discard_alignment = info->discard_alignment;
 971                 if (info->feature_secdiscard)
 972                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 973         }
 974
 975         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 976         blk_queue_logical_block_size(rq, info->sector_size);
 977         blk_queue_physical_block_size(rq, info->physical_sector_size);
 978         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 979
 980         /* Each segment in a request is up to an aligned page in size. */
 981         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 982         blk_queue_max_segment_size(rq, PAGE_SIZE);
 983
 984         /* Ensure a merged request will fit in a single I/O ring slot. */
 985         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 986
 987         /* Make sure buffer addresses are sector-aligned. */
 988         blk_queue_dma_alignment(rq, 511);
 989 }
 990
 991 static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size,
 992                                 unsigned int physical_sector_size)
 993 {
 994         struct request_queue *rq;
 995         struct blkfront_info *info = gd->private_data;
 996
 997         memset(&info->tag_set, 0, sizeof(info->tag_set));
 998         info->tag_set.ops = &blkfront_mq_ops;
 999         info->tag_set.nr_hw_queues = info->nr_rings;
1000         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1001                 /*
1002                  * When indirect descriptior is not supported, the I/O request
1003                  * will be split between multiple request in the ring.
1004                  * To avoid problems when sending the request, divide by
1005                  * 2 the depth of the queue.
1006                  */
1007                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1008         } else
1009                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1010         info->tag_set.numa_node = NUMA_NO_NODE;
1011         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1012         info->tag_set.cmd_size = sizeof(struct blkif_req);
1013         info->tag_set.driver_data = info;
1014
1015         if (blk_mq_alloc_tag_set(&info->tag_set))
1016                 return -EINVAL;
1017         rq = blk_mq_init_queue(&info->tag_set);
1018         if (IS_ERR(rq)) {
1019                 blk_mq_free_tag_set(&info->tag_set);
1020                 return PTR_ERR(rq);
1021         }
1022
1023         rq->queuedata = info;
1024         info->rq = gd->queue = rq;
1025         info->gd = gd;
1026         info->sector_size = sector_size;
1027         info->physical_sector_size = physical_sector_size;
1028         blkif_set_queue_limits(info);
1029
1030         return 0;
1031 }
1032
1033 static const char *flush_info(struct blkfront_info *info)
1034 {
1035         if (info->feature_flush && info->feature_fua)
1036                 return "barrier: enabled;";
1037         else if (info->feature_flush)
1038                 return "flush diskcache: enabled;";
1039         else
1040                 return "barrier or flush: disabled;";
1041 }
1042
1043 static void xlvbd_flush(struct blkfront_info *info)
1044 {
1045         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
1046                               info->feature_fua ? true : false);
1047         pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
1048                 info->gd->disk_name, flush_info(info),
1049                 "persistent grants:", info->feature_persistent ?
1050                 "enabled;" : "disabled;", "indirect descriptors:",
1051                 info->max_indirect_segments ? "enabled;" : "disabled;",
1052                 "bounce buffer:", info->bounce ? "enabled" : "disabled;");
1053 }
1054
1055 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1056 {
1057         int major;
1058         major = BLKIF_MAJOR(vdevice);
1059         *minor = BLKIF_MINOR(vdevice);
1060         switch (major) {
1061                 case XEN_IDE0_MAJOR:
1062                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1063                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1064                                 EMULATED_HD_DISK_MINOR_OFFSET;
1065                         break;
1066                 case XEN_IDE1_MAJOR:
1067                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1068                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1069                                 EMULATED_HD_DISK_MINOR_OFFSET;
1070                         break;
1071                 case XEN_SCSI_DISK0_MAJOR:
1072                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1073                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1074                         break;
1075                 case XEN_SCSI_DISK1_MAJOR:
1076                 case XEN_SCSI_DISK2_MAJOR:
1077                 case XEN_SCSI_DISK3_MAJOR:
1078                 case XEN_SCSI_DISK4_MAJOR:
1079                 case XEN_SCSI_DISK5_MAJOR:
1080                 case XEN_SCSI_DISK6_MAJOR:
1081                 case XEN_SCSI_DISK7_MAJOR:
1082                         *offset = (*minor / PARTS_PER_DISK) +
1083                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1084                                 EMULATED_SD_DISK_NAME_OFFSET;
1085                         *minor = *minor +
1086                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1087                                 EMULATED_SD_DISK_MINOR_OFFSET;
1088                         break;
1089                 case XEN_SCSI_DISK8_MAJOR:
1090                 case XEN_SCSI_DISK9_MAJOR:
1091                 case XEN_SCSI_DISK10_MAJOR:
1092                 case XEN_SCSI_DISK11_MAJOR:
1093                 case XEN_SCSI_DISK12_MAJOR:
1094                 case XEN_SCSI_DISK13_MAJOR:
1095                 case XEN_SCSI_DISK14_MAJOR:
1096                 case XEN_SCSI_DISK15_MAJOR:
1097                         *offset = (*minor / PARTS_PER_DISK) +
1098                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1099                                 EMULATED_SD_DISK_NAME_OFFSET;
1100                         *minor = *minor +
1101                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1102                                 EMULATED_SD_DISK_MINOR_OFFSET;
1103                         break;
1104                 case XENVBD_MAJOR:
1105                         *offset = *minor / PARTS_PER_DISK;
1106                         break;
1107                 default:
1108                         printk(KERN_WARNING "blkfront: your disk configuration is "
1109                                         "incorrect, please use an xvd device instead\n");
1110                         return -ENODEV;
1111         }
1112         return 0;
1113 }
1114
1115 static char *encode_disk_name(char *ptr, unsigned int n)
1116 {
1117         if (n >= 26)
1118                 ptr = encode_disk_name(ptr, n / 26 - 1);
1119         *ptr = 'a' + n % 26;
1120         return ptr + 1;
1121 }
1122
1123 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1124                                struct blkfront_info *info,
1125                                u16 vdisk_info, u16 sector_size,
1126                                unsigned int physical_sector_size)
1127 {
1128         struct gendisk *gd;
1129         int nr_minors = 1;
1130         int err;
1131         unsigned int offset;
1132         int minor;
1133         int nr_parts;
1134         char *ptr;
1135
1136         BUG_ON(info->gd != NULL);
1137         BUG_ON(info->rq != NULL);
1138
1139         if ((info->vdevice>>EXT_SHIFT) > 1) {
1140                 /* this is above the extended range; something is wrong */
1141                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1142                 return -ENODEV;
1143         }
1144
1145         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1146                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1147                 if (err)
1148                         return err;
1149                 nr_parts = PARTS_PER_DISK;
1150         } else {
1151                 minor = BLKIF_MINOR_EXT(info->vdevice);
1152                 nr_parts = PARTS_PER_EXT_DISK;
1153                 offset = minor / nr_parts;
1154                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1155                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1156                                         "emulated IDE disks,\n\t choose an xvd device name"
1157                                         "from xvde on\n", info->vdevice);
1158         }
1159         if (minor >> MINORBITS) {
1160                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1161                         info->vdevice, minor);
1162                 return -ENODEV;
1163         }
1164
1165         if ((minor % nr_parts) == 0)
1166                 nr_minors = nr_parts;
1167
1168         err = xlbd_reserve_minors(minor, nr_minors);
1169         if (err)
1170                 goto out;
1171         err = -ENODEV;
1172
1173         gd = alloc_disk(nr_minors);
1174         if (gd == NULL)
1175                 goto release;
1176
1177         strcpy(gd->disk_name, DEV_NAME);
1178         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1179         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1180         if (nr_minors > 1)
1181                 *ptr = 0;
1182         else
1183                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1184                          "%d", minor & (nr_parts - 1));
1185
1186         gd->major = XENVBD_MAJOR;
1187         gd->first_minor = minor;
1188         gd->fops = &xlvbd_block_fops;
1189         gd->private_data = info;
1190         set_capacity(gd, capacity);
1191
1192         if (xlvbd_init_blk_queue(gd, sector_size, physical_sector_size)) {
1193                 del_gendisk(gd);
1194                 goto release;
1195         }
1196
1197         xlvbd_flush(info);
1198
1199         if (vdisk_info & VDISK_READONLY)
1200                 set_disk_ro(gd, 1);
1201
1202         if (vdisk_info & VDISK_REMOVABLE)
1203                 gd->flags |= GENHD_FL_REMOVABLE;
1204
1205         if (vdisk_info & VDISK_CDROM)
1206                 gd->flags |= GENHD_FL_CD;
1207
1208         return 0;
1209
1210  release:
1211         xlbd_release_minors(minor, nr_minors);
1212  out:
1213         return err;
1214 }
1215
1216 static void xlvbd_release_gendisk(struct blkfront_info *info)
1217 {
1218         unsigned int minor, nr_minors, i;
1219         struct blkfront_ring_info *rinfo;
1220
1221         if (info->rq == NULL)
1222                 return;
1223
1224         /* No more blkif_request(). */
1225         blk_mq_stop_hw_queues(info->rq);
1226
1227         for_each_rinfo(info, rinfo, i) {
1228                 /* No more gnttab callback work. */
1229                 gnttab_cancel_free_callback(&rinfo->callback);
1230
1231                 /* Flush gnttab callback work. Must be done with no locks held. */
1232                 flush_work(&rinfo->work);
1233         }
1234
1235         del_gendisk(info->gd);
1236
1237         minor = info->gd->first_minor;
1238         nr_minors = info->gd->minors;
1239         xlbd_release_minors(minor, nr_minors);
1240
1241         blk_cleanup_queue(info->rq);
1242         blk_mq_free_tag_set(&info->tag_set);
1243         info->rq = NULL;
1244
1245         put_disk(info->gd);
1246         info->gd = NULL;
1247 }
1248
1249 /* Already hold rinfo->ring_lock. */
1250 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1251 {
1252         if (!RING_FULL(&rinfo->ring))
1253                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1254 }
1255
1256 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1257 {
1258         unsigned long flags;
1259
1260         spin_lock_irqsave(&rinfo->ring_lock, flags);
1261         kick_pending_request_queues_locked(rinfo);
1262         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1263 }
1264
1265 static void blkif_restart_queue(struct work_struct *work)
1266 {
1267         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1268
1269         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1270                 kick_pending_request_queues(rinfo);
1271 }
1272
1273 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1274 {
1275         struct grant *persistent_gnt, *n;
1276         struct blkfront_info *info = rinfo->dev_info;
1277         int i, j, segs;
1278
1279         /*
1280          * Remove indirect pages, this only happens when using indirect
1281          * descriptors but not persistent grants
1282          */
1283         if (!list_empty(&rinfo->indirect_pages)) {
1284                 struct page *indirect_page, *n;
1285
1286                 BUG_ON(info->bounce);
1287                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1288                         list_del(&indirect_page->lru);
1289                         __free_page(indirect_page);
1290                 }
1291         }
1292
1293         /* Remove all persistent grants. */
1294         if (!list_empty(&rinfo->grants)) {
1295                 list_for_each_entry_safe(persistent_gnt, n,
1296                                          &rinfo->grants, node) {
1297                         list_del(&persistent_gnt->node);
1298                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1299                                 gnttab_end_foreign_access(persistent_gnt->gref,
1300                                                           0, 0UL);
1301                                 rinfo->persistent_gnts_c--;
1302                         }
1303                         if (info->bounce)
1304                                 __free_page(persistent_gnt->page);
1305                         kfree(persistent_gnt);
1306                 }
1307         }
1308         BUG_ON(rinfo->persistent_gnts_c != 0);
1309
1310         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1311                 /*
1312                  * Clear persistent grants present in requests already
1313                  * on the shared ring
1314                  */
1315                 if (!rinfo->shadow[i].request)
1316                         goto free_shadow;
1317
1318                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1319                        rinfo->shadow[i].req.u.indirect.nr_segments :
1320                        rinfo->shadow[i].req.u.rw.nr_segments;
1321                 for (j = 0; j < segs; j++) {
1322                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1323                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1324                         if (info->bounce)
1325                                 __free_page(persistent_gnt->page);
1326                         kfree(persistent_gnt);
1327                 }
1328
1329                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1330                         /*
1331                          * If this is not an indirect operation don't try to
1332                          * free indirect segments
1333                          */
1334                         goto free_shadow;
1335
1336                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1337                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1338                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1339                         __free_page(persistent_gnt->page);
1340                         kfree(persistent_gnt);
1341                 }
1342
1343 free_shadow:
1344                 kvfree(rinfo->shadow[i].grants_used);
1345                 rinfo->shadow[i].grants_used = NULL;
1346                 kvfree(rinfo->shadow[i].indirect_grants);
1347                 rinfo->shadow[i].indirect_grants = NULL;
1348                 kvfree(rinfo->shadow[i].sg);
1349                 rinfo->shadow[i].sg = NULL;
1350         }
1351
1352         /* No more gnttab callback work. */
1353         gnttab_cancel_free_callback(&rinfo->callback);
1354
1355         /* Flush gnttab callback work. Must be done with no locks held. */
1356         flush_work(&rinfo->work);
1357
1358         /* Free resources associated with old device channel. */
1359         for (i = 0; i < info->nr_ring_pages; i++) {
1360                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1361                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1362                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1363                 }
1364         }
1365         free_pages_exact(rinfo->ring.sring,
1366                          info->nr_ring_pages * XEN_PAGE_SIZE);
1367         rinfo->ring.sring = NULL;
1368
1369         if (rinfo->irq)
1370                 unbind_from_irqhandler(rinfo->irq, rinfo);
1371         rinfo->evtchn = rinfo->irq = 0;
1372 }
1373
1374 static void blkif_free(struct blkfront_info *info, int suspend)
1375 {
1376         unsigned int i;
1377         struct blkfront_ring_info *rinfo;
1378
1379         /* Prevent new requests being issued until we fix things up. */
1380         info->connected = suspend ?
1381                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1382         /* No more blkif_request(). */
1383         if (info->rq)
1384                 blk_mq_stop_hw_queues(info->rq);
1385
1386         for_each_rinfo(info, rinfo, i)
1387                 blkif_free_ring(rinfo);
1388
1389         kvfree(info->rinfo);
1390         info->rinfo = NULL;
1391         info->nr_rings = 0;
1392 }
1393
1394 struct copy_from_grant {
1395         const struct blk_shadow *s;
1396         unsigned int grant_idx;
1397         unsigned int bvec_offset;
1398         char *bvec_data;
1399 };
1400
1401 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1402                                   unsigned int len, void *data)
1403 {
1404         struct copy_from_grant *info = data;
1405         char *shared_data;
1406         /* Convenient aliases */
1407         const struct blk_shadow *s = info->s;
1408
1409         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1410
1411         memcpy(info->bvec_data + info->bvec_offset,
1412                shared_data + offset, len);
1413
1414         info->bvec_offset += len;
1415         info->grant_idx++;
1416
1417         kunmap_atomic(shared_data);
1418 }
1419
1420 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1421 {
1422         switch (rsp)
1423         {
1424         case BLKIF_RSP_OKAY:
1425                 return REQ_DONE;
1426         case BLKIF_RSP_EOPNOTSUPP:
1427                 return REQ_EOPNOTSUPP;
1428         case BLKIF_RSP_ERROR:
1429         default:
1430                 return REQ_ERROR;
1431         }
1432 }
1433
1434 /*
1435  * Get the final status of the block request based on two ring response
1436  */
1437 static int blkif_get_final_status(enum blk_req_status s1,
1438                                   enum blk_req_status s2)
1439 {
1440         BUG_ON(s1 < REQ_DONE);
1441         BUG_ON(s2 < REQ_DONE);
1442
1443         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1444                 return BLKIF_RSP_ERROR;
1445         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1446                 return BLKIF_RSP_EOPNOTSUPP;
1447         return BLKIF_RSP_OKAY;
1448 }
1449
1450 /*
1451  * Return values:
1452  *  1 response processed.
1453  *  0 missing further responses.
1454  * -1 error while processing.
1455  */
1456 static int blkif_completion(unsigned long *id,
1457                             struct blkfront_ring_info *rinfo,
1458                             struct blkif_response *bret)
1459 {
1460         int i = 0;
1461         struct scatterlist *sg;
1462         int num_sg, num_grant;
1463         struct blkfront_info *info = rinfo->dev_info;
1464         struct blk_shadow *s = &rinfo->shadow[*id];
1465         struct copy_from_grant data = {
1466                 .grant_idx = 0,
1467         };
1468
1469         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1470                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1471
1472         /* The I/O request may be split in two. */
1473         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1474                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1475
1476                 /* Keep the status of the current response in shadow. */
1477                 s->status = blkif_rsp_to_req_status(bret->status);
1478
1479                 /* Wait the second response if not yet here. */
1480                 if (s2->status < REQ_DONE)
1481                         return 0;
1482
1483                 bret->status = blkif_get_final_status(s->status,
1484                                                       s2->status);
1485
1486                 /*
1487                  * All the grants is stored in the first shadow in order
1488                  * to make the completion code simpler.
1489                  */
1490                 num_grant += s2->req.u.rw.nr_segments;
1491
1492                 /*
1493                  * The two responses may not come in order. Only the
1494                  * first request will store the scatter-gather list.
1495                  */
1496                 if (s2->num_sg != 0) {
1497                         /* Update "id" with the ID of the first response. */
1498                         *id = s->associated_id;
1499                         s = s2;
1500                 }
1501
1502                 /*
1503                  * We don't need anymore the second request, so recycling
1504                  * it now.
1505                  */
1506                 if (add_id_to_freelist(rinfo, s->associated_id))
1507                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1508                              info->gd->disk_name, s->associated_id);
1509         }
1510
1511         data.s = s;
1512         num_sg = s->num_sg;
1513
1514         if (bret->operation == BLKIF_OP_READ && info->bounce) {
1515                 for_each_sg(s->sg, sg, num_sg, i) {
1516                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1517
1518                         data.bvec_offset = sg->offset;
1519                         data.bvec_data = kmap_atomic(sg_page(sg));
1520
1521                         gnttab_foreach_grant_in_range(sg_page(sg),
1522                                                       sg->offset,
1523                                                       sg->length,
1524                                                       blkif_copy_from_grant,
1525                                                       &data);
1526
1527                         kunmap_atomic(data.bvec_data);
1528                 }
1529         }
1530         /* Add the persistent grant into the list of free grants */
1531         for (i = 0; i < num_grant; i++) {
1532                 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1533                         /*
1534                          * If the grant is still mapped by the backend (the
1535                          * backend has chosen to make this grant persistent)
1536                          * we add it at the head of the list, so it will be
1537                          * reused first.
1538                          */
1539                         if (!info->feature_persistent) {
1540                                 pr_alert("backed has not unmapped grant: %u\n",
1541                                          s->grants_used[i]->gref);
1542                                 return -1;
1543                         }
1544                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1545                         rinfo->persistent_gnts_c++;
1546                 } else {
1547                         /*
1548                          * If the grant is not mapped by the backend we add it
1549                          * to the tail of the list, so it will not be picked
1550                          * again unless we run out of persistent grants.
1551                          */
1552                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1553                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1554                 }
1555         }
1556         if (s->req.operation == BLKIF_OP_INDIRECT) {
1557                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1558                         if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1559                                 if (!info->feature_persistent) {
1560                                         pr_alert("backed has not unmapped grant: %u\n",
1561                                                  s->indirect_grants[i]->gref);
1562                                         return -1;
1563                                 }
1564                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1565                                 rinfo->persistent_gnts_c++;
1566                         } else {
1567                                 struct page *indirect_page;
1568
1569                                 /*
1570                                  * Add the used indirect page back to the list of
1571                                  * available pages for indirect grefs.
1572                                  */
1573                                 if (!info->bounce) {
1574                                         indirect_page = s->indirect_grants[i]->page;
1575                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1576                                 }
1577                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1578                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1579                         }
1580                 }
1581         }
1582
1583         return 1;
1584 }
1585
1586 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1587 {
1588         struct request *req;
1589         struct blkif_response bret;
1590         RING_IDX i, rp;
1591         unsigned long flags;
1592         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1593         struct blkfront_info *info = rinfo->dev_info;
1594         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1595
1596         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1597                 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1598                 return IRQ_HANDLED;
1599         }
1600
1601         spin_lock_irqsave(&rinfo->ring_lock, flags);
1602  again:
1603         rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1604         virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1605         if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1606                 pr_alert("%s: illegal number of responses %u\n",
1607                          info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1608                 goto err;
1609         }
1610
1611         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1612                 unsigned long id;
1613                 unsigned int op;
1614
1615                 eoiflag = 0;
1616
1617                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1618                 id = bret.id;
1619
1620                 /*
1621                  * The backend has messed up and given us an id that we would
1622                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1623                  * look in get_id_from_freelist.
1624                  */
1625                 if (id >= BLK_RING_SIZE(info)) {
1626                         pr_alert("%s: response has incorrect id (%ld)\n",
1627                                  info->gd->disk_name, id);
1628                         goto err;
1629                 }
1630                 if (rinfo->shadow[id].status != REQ_WAITING) {
1631                         pr_alert("%s: response references no pending request\n",
1632                                  info->gd->disk_name);
1633                         goto err;
1634                 }
1635
1636                 rinfo->shadow[id].status = REQ_PROCESSING;
1637                 req  = rinfo->shadow[id].request;
1638
1639                 op = rinfo->shadow[id].req.operation;
1640                 if (op == BLKIF_OP_INDIRECT)
1641                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1642                 if (bret.operation != op) {
1643                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1644                                  info->gd->disk_name, bret.operation, op);
1645                         goto err;
1646                 }
1647
1648                 if (bret.operation != BLKIF_OP_DISCARD) {
1649                         int ret;
1650
1651                         /*
1652                          * We may need to wait for an extra response if the
1653                          * I/O request is split in 2
1654                          */
1655                         ret = blkif_completion(&id, rinfo, &bret);
1656                         if (!ret)
1657                                 continue;
1658                         if (unlikely(ret < 0))
1659                                 goto err;
1660                 }
1661
1662                 if (add_id_to_freelist(rinfo, id)) {
1663                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1664                              info->gd->disk_name, op_name(bret.operation), id);
1665                         continue;
1666                 }
1667
1668                 if (bret.status == BLKIF_RSP_OKAY)
1669                         blkif_req(req)->error = BLK_STS_OK;
1670                 else
1671                         blkif_req(req)->error = BLK_STS_IOERR;
1672
1673                 switch (bret.operation) {
1674                 case BLKIF_OP_DISCARD:
1675                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1676                                 struct request_queue *rq = info->rq;
1677
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                                 info->feature_discard = 0;
1682                                 info->feature_secdiscard = 0;
1683                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1684                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1685                         }
1686                         break;
1687                 case BLKIF_OP_FLUSH_DISKCACHE:
1688                 case BLKIF_OP_WRITE_BARRIER:
1689                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1690                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1691                                        info->gd->disk_name, op_name(bret.operation));
1692                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1693                         }
1694                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1695                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1696                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1697                                        info->gd->disk_name, op_name(bret.operation));
1698                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1699                         }
1700                         if (unlikely(blkif_req(req)->error)) {
1701                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1702                                         blkif_req(req)->error = BLK_STS_OK;
1703                                 info->feature_fua = 0;
1704                                 info->feature_flush = 0;
1705                                 xlvbd_flush(info);
1706                         }
1707                         fallthrough;
1708                 case BLKIF_OP_READ:
1709                 case BLKIF_OP_WRITE:
1710                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1711                                 dev_dbg_ratelimited(&info->xbdev->dev,
1712                                         "Bad return from blkdev data request: %#x\n",
1713                                         bret.status);
1714
1715                         break;
1716                 default:
1717                         BUG();
1718                 }
1719
1720                 if (likely(!blk_should_fake_timeout(req->q)))
1721                         blk_mq_complete_request(req);
1722         }
1723
1724         rinfo->ring.rsp_cons = i;
1725
1726         if (i != rinfo->ring.req_prod_pvt) {
1727                 int more_to_do;
1728                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1729                 if (more_to_do)
1730                         goto again;
1731         } else
1732                 rinfo->ring.sring->rsp_event = i + 1;
1733
1734         kick_pending_request_queues_locked(rinfo);
1735
1736         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1737
1738         xen_irq_lateeoi(irq, eoiflag);
1739
1740         return IRQ_HANDLED;
1741
1742  err:
1743         info->connected = BLKIF_STATE_ERROR;
1744
1745         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1746
1747         /* No EOI in order to avoid further interrupts. */
1748
1749         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1750         return IRQ_HANDLED;
1751 }
1752
1753
1754 static int setup_blkring(struct xenbus_device *dev,
1755                          struct blkfront_ring_info *rinfo)
1756 {
1757         struct blkif_sring *sring;
1758         int err, i;
1759         struct blkfront_info *info = rinfo->dev_info;
1760         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1761         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1762
1763         for (i = 0; i < info->nr_ring_pages; i++)
1764                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1765
1766         sring = alloc_pages_exact(ring_size, GFP_NOIO | __GFP_ZERO);
1767         if (!sring) {
1768                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1769                 return -ENOMEM;
1770         }
1771         SHARED_RING_INIT(sring);
1772         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1773
1774         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1775         if (err < 0) {
1776                 free_pages_exact(sring, ring_size);
1777                 rinfo->ring.sring = NULL;
1778                 goto fail;
1779         }
1780         for (i = 0; i < info->nr_ring_pages; i++)
1781                 rinfo->ring_ref[i] = gref[i];
1782
1783         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1784         if (err)
1785                 goto fail;
1786
1787         err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1788                                                 0, "blkif", rinfo);
1789         if (err <= 0) {
1790                 xenbus_dev_fatal(dev, err,
1791                                  "bind_evtchn_to_irqhandler failed");
1792                 goto fail;
1793         }
1794         rinfo->irq = err;
1795
1796         return 0;
1797 fail:
1798         blkif_free(info, 0);
1799         return err;
1800 }
1801
1802 /*
1803  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1804  * ring buffer may have multi pages depending on ->nr_ring_pages.
1805  */
1806 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1807                                 struct blkfront_ring_info *rinfo, const char *dir)
1808 {
1809         int err;
1810         unsigned int i;
1811         const char *message = NULL;
1812         struct blkfront_info *info = rinfo->dev_info;
1813
1814         if (info->nr_ring_pages == 1) {
1815                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1816                 if (err) {
1817                         message = "writing ring-ref";
1818                         goto abort_transaction;
1819                 }
1820         } else {
1821                 for (i = 0; i < info->nr_ring_pages; i++) {
1822                         char ring_ref_name[RINGREF_NAME_LEN];
1823
1824                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1825                         err = xenbus_printf(xbt, dir, ring_ref_name,
1826                                             "%u", rinfo->ring_ref[i]);
1827                         if (err) {
1828                                 message = "writing ring-ref";
1829                                 goto abort_transaction;
1830                         }
1831                 }
1832         }
1833
1834         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1835         if (err) {
1836                 message = "writing event-channel";
1837                 goto abort_transaction;
1838         }
1839
1840         return 0;
1841
1842 abort_transaction:
1843         xenbus_transaction_end(xbt, 1);
1844         if (message)
1845                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1846
1847         return err;
1848 }
1849
1850 static void free_info(struct blkfront_info *info)
1851 {
1852         list_del(&info->info_list);
1853         kfree(info);
1854 }
1855
1856 /* Enable the persistent grants feature. */
1857 static bool feature_persistent = true;
1858 module_param(feature_persistent, bool, 0644);
1859 MODULE_PARM_DESC(feature_persistent,
1860                 "Enables the persistent grants feature");
1861
1862 /* Common code used when first setting up, and when resuming. */
1863 static int talk_to_blkback(struct xenbus_device *dev,
1864                            struct blkfront_info *info)
1865 {
1866         const char *message = NULL;
1867         struct xenbus_transaction xbt;
1868         int err;
1869         unsigned int i, max_page_order;
1870         unsigned int ring_page_order;
1871         struct blkfront_ring_info *rinfo;
1872
1873         if (!info)
1874                 return -ENODEV;
1875
1876         /* Check if backend is trusted. */
1877         info->bounce = !xen_blkif_trusted ||
1878                        !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1879
1880         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1881                                               "max-ring-page-order", 0);
1882         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1883         info->nr_ring_pages = 1 << ring_page_order;
1884
1885         err = negotiate_mq(info);
1886         if (err)
1887                 goto destroy_blkring;
1888
1889         for_each_rinfo(info, rinfo, i) {
1890                 /* Create shared ring, alloc event channel. */
1891                 err = setup_blkring(dev, rinfo);
1892                 if (err)
1893                         goto destroy_blkring;
1894         }
1895
1896 again:
1897         err = xenbus_transaction_start(&xbt);
1898         if (err) {
1899                 xenbus_dev_fatal(dev, err, "starting transaction");
1900                 goto destroy_blkring;
1901         }
1902
1903         if (info->nr_ring_pages > 1) {
1904                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1905                                     ring_page_order);
1906                 if (err) {
1907                         message = "writing ring-page-order";
1908                         goto abort_transaction;
1909                 }
1910         }
1911
1912         /* We already got the number of queues/rings in _probe */
1913         if (info->nr_rings == 1) {
1914                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1915                 if (err)
1916                         goto destroy_blkring;
1917         } else {
1918                 char *path;
1919                 size_t pathsize;
1920
1921                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1922                                     info->nr_rings);
1923                 if (err) {
1924                         message = "writing multi-queue-num-queues";
1925                         goto abort_transaction;
1926                 }
1927
1928                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1929                 path = kmalloc(pathsize, GFP_KERNEL);
1930                 if (!path) {
1931                         err = -ENOMEM;
1932                         message = "ENOMEM while writing ring references";
1933                         goto abort_transaction;
1934                 }
1935
1936                 for_each_rinfo(info, rinfo, i) {
1937                         memset(path, 0, pathsize);
1938                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1939                         err = write_per_ring_nodes(xbt, rinfo, path);
1940                         if (err) {
1941                                 kfree(path);
1942                                 goto destroy_blkring;
1943                         }
1944                 }
1945                 kfree(path);
1946         }
1947         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1948                             XEN_IO_PROTO_ABI_NATIVE);
1949         if (err) {
1950                 message = "writing protocol";
1951                 goto abort_transaction;
1952         }
1953         info->feature_persistent_parm = feature_persistent;
1954         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1955                         info->feature_persistent_parm);
1956         if (err)
1957                 dev_warn(&dev->dev,
1958                          "writing persistent grants feature to xenbus");
1959
1960         err = xenbus_transaction_end(xbt, 0);
1961         if (err) {
1962                 if (err == -EAGAIN)
1963                         goto again;
1964                 xenbus_dev_fatal(dev, err, "completing transaction");
1965                 goto destroy_blkring;
1966         }
1967
1968         for_each_rinfo(info, rinfo, i) {
1969                 unsigned int j;
1970
1971                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1972                         rinfo->shadow[j].req.u.rw.id = j + 1;
1973                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1974         }
1975         xenbus_switch_state(dev, XenbusStateInitialised);
1976
1977         return 0;
1978
1979  abort_transaction:
1980         xenbus_transaction_end(xbt, 1);
1981         if (message)
1982                 xenbus_dev_fatal(dev, err, "%s", message);
1983  destroy_blkring:
1984         blkif_free(info, 0);
1985
1986         mutex_lock(&blkfront_mutex);
1987         free_info(info);
1988         mutex_unlock(&blkfront_mutex);
1989
1990         dev_set_drvdata(&dev->dev, NULL);
1991
1992         return err;
1993 }
1994
1995 static int negotiate_mq(struct blkfront_info *info)
1996 {
1997         unsigned int backend_max_queues;
1998         unsigned int i;
1999         struct blkfront_ring_info *rinfo;
2000
2001         BUG_ON(info->nr_rings);
2002
2003         /* Check if backend supports multiple queues. */
2004         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2005                                                   "multi-queue-max-queues", 1);
2006         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
2007         /* We need at least one ring. */
2008         if (!info->nr_rings)
2009                 info->nr_rings = 1;
2010
2011         info->rinfo_size = struct_size(info->rinfo, shadow,
2012                                        BLK_RING_SIZE(info));
2013         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
2014         if (!info->rinfo) {
2015                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
2016                 info->nr_rings = 0;
2017                 return -ENOMEM;
2018         }
2019
2020         for_each_rinfo(info, rinfo, i) {
2021                 INIT_LIST_HEAD(&rinfo->indirect_pages);
2022                 INIT_LIST_HEAD(&rinfo->grants);
2023                 rinfo->dev_info = info;
2024                 INIT_WORK(&rinfo->work, blkif_restart_queue);
2025                 spin_lock_init(&rinfo->ring_lock);
2026         }
2027         return 0;
2028 }
2029
2030 /**
2031  * Entry point to this code when a new device is created.  Allocate the basic
2032  * structures and the ring buffer for communication with the backend, and
2033  * inform the backend of the appropriate details for those.  Switch to
2034  * Initialised state.
2035  */
2036 static int blkfront_probe(struct xenbus_device *dev,
2037                           const struct xenbus_device_id *id)
2038 {
2039         int err, vdevice;
2040         struct blkfront_info *info;
2041
2042         /* FIXME: Use dynamic device id if this is not set. */
2043         err = xenbus_scanf(XBT_NIL, dev->nodename,
2044                            "virtual-device", "%i", &vdevice);
2045         if (err != 1) {
2046                 /* go looking in the extended area instead */
2047                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
2048                                    "%i", &vdevice);
2049                 if (err != 1) {
2050                         xenbus_dev_fatal(dev, err, "reading virtual-device");
2051                         return err;
2052                 }
2053         }
2054
2055         if (xen_hvm_domain()) {
2056                 char *type;
2057                 int len;
2058                 /* no unplug has been done: do not hook devices != xen vbds */
2059                 if (xen_has_pv_and_legacy_disk_devices()) {
2060                         int major;
2061
2062                         if (!VDEV_IS_EXTENDED(vdevice))
2063                                 major = BLKIF_MAJOR(vdevice);
2064                         else
2065                                 major = XENVBD_MAJOR;
2066
2067                         if (major != XENVBD_MAJOR) {
2068                                 printk(KERN_INFO
2069                                                 "%s: HVM does not support vbd %d as xen block device\n",
2070                                                 __func__, vdevice);
2071                                 return -ENODEV;
2072                         }
2073                 }
2074                 /* do not create a PV cdrom device if we are an HVM guest */
2075                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
2076                 if (IS_ERR(type))
2077                         return -ENODEV;
2078                 if (strncmp(type, "cdrom", 5) == 0) {
2079                         kfree(type);
2080                         return -ENODEV;
2081                 }
2082                 kfree(type);
2083         }
2084         info = kzalloc(sizeof(*info), GFP_KERNEL);
2085         if (!info) {
2086                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2087                 return -ENOMEM;
2088         }
2089
2090         info->xbdev = dev;
2091
2092         mutex_init(&info->mutex);
2093         info->vdevice = vdevice;
2094         info->connected = BLKIF_STATE_DISCONNECTED;
2095
2096         /* Front end dir is a number, which is used as the id. */
2097         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2098         dev_set_drvdata(&dev->dev, info);
2099
2100         mutex_lock(&blkfront_mutex);
2101         list_add(&info->info_list, &info_list);
2102         mutex_unlock(&blkfront_mutex);
2103
2104         return 0;
2105 }
2106
2107 static int blkif_recover(struct blkfront_info *info)
2108 {
2109         unsigned int r_index;
2110         struct request *req, *n;
2111         int rc;
2112         struct bio *bio;
2113         unsigned int segs;
2114         struct blkfront_ring_info *rinfo;
2115
2116         blkfront_gather_backend_features(info);
2117         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2118         blkif_set_queue_limits(info);
2119         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2120         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2121
2122         for_each_rinfo(info, rinfo, r_index) {
2123                 rc = blkfront_setup_indirect(rinfo);
2124                 if (rc)
2125                         return rc;
2126         }
2127         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2128
2129         /* Now safe for us to use the shared ring */
2130         info->connected = BLKIF_STATE_CONNECTED;
2131
2132         for_each_rinfo(info, rinfo, r_index) {
2133                 /* Kick any other new requests queued since we resumed */
2134                 kick_pending_request_queues(rinfo);
2135         }
2136
2137         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2138                 /* Requeue pending requests (flush or discard) */
2139                 list_del_init(&req->queuelist);
2140                 BUG_ON(req->nr_phys_segments > segs);
2141                 blk_mq_requeue_request(req, false);
2142         }
2143         blk_mq_start_stopped_hw_queues(info->rq, true);
2144         blk_mq_kick_requeue_list(info->rq);
2145
2146         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2147                 /* Traverse the list of pending bios and re-queue them */
2148                 submit_bio(bio);
2149         }
2150
2151         return 0;
2152 }
2153
2154 /**
2155  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2156  * driver restart.  We tear down our blkif structure and recreate it, but
2157  * leave the device-layer structures intact so that this is transparent to the
2158  * rest of the kernel.
2159  */
2160 static int blkfront_resume(struct xenbus_device *dev)
2161 {
2162         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2163         int err = 0;
2164         unsigned int i, j;
2165         struct blkfront_ring_info *rinfo;
2166
2167         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2168
2169         bio_list_init(&info->bio_list);
2170         INIT_LIST_HEAD(&info->requests);
2171         for_each_rinfo(info, rinfo, i) {
2172                 struct bio_list merge_bio;
2173                 struct blk_shadow *shadow = rinfo->shadow;
2174
2175                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2176                         /* Not in use? */
2177                         if (!shadow[j].request)
2178                                 continue;
2179
2180                         /*
2181                          * Get the bios in the request so we can re-queue them.
2182                          */
2183                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2184                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2185                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2186                             shadow[j].request->cmd_flags & REQ_FUA) {
2187                                 /*
2188                                  * Flush operations don't contain bios, so
2189                                  * we need to requeue the whole request
2190                                  *
2191                                  * XXX: but this doesn't make any sense for a
2192                                  * write with the FUA flag set..
2193                                  */
2194                                 list_add(&shadow[j].request->queuelist, &info->requests);
2195                                 continue;
2196                         }
2197                         merge_bio.head = shadow[j].request->bio;
2198                         merge_bio.tail = shadow[j].request->biotail;
2199                         bio_list_merge(&info->bio_list, &merge_bio);
2200                         shadow[j].request->bio = NULL;
2201                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2202                 }
2203         }
2204
2205         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2206
2207         err = talk_to_blkback(dev, info);
2208         if (!err)
2209                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2210
2211         /*
2212          * We have to wait for the backend to switch to
2213          * connected state, since we want to read which
2214          * features it supports.
2215          */
2216
2217         return err;
2218 }
2219
2220 static void blkfront_closing(struct blkfront_info *info)
2221 {
2222         struct xenbus_device *xbdev = info->xbdev;
2223         struct block_device *bdev = NULL;
2224
2225         mutex_lock(&info->mutex);
2226
2227         if (xbdev->state == XenbusStateClosing) {
2228                 mutex_unlock(&info->mutex);
2229                 return;
2230         }
2231
2232         if (info->gd)
2233                 bdev = bdget_disk(info->gd, 0);
2234
2235         mutex_unlock(&info->mutex);
2236
2237         if (!bdev) {
2238                 xenbus_frontend_closed(xbdev);
2239                 return;
2240         }
2241
2242         mutex_lock(&bdev->bd_mutex);
2243
2244         if (bdev->bd_openers) {
2245                 xenbus_dev_error(xbdev, -EBUSY,
2246                                  "Device in use; refusing to close");
2247                 xenbus_switch_state(xbdev, XenbusStateClosing);
2248         } else {
2249                 xlvbd_release_gendisk(info);
2250                 xenbus_frontend_closed(xbdev);
2251         }
2252
2253         mutex_unlock(&bdev->bd_mutex);
2254         bdput(bdev);
2255 }
2256
2257 static void blkfront_setup_discard(struct blkfront_info *info)
2258 {
2259         info->feature_discard = 1;
2260         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2261                                                          "discard-granularity",
2262                                                          0);
2263         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2264                                                        "discard-alignment", 0);
2265         info->feature_secdiscard =
2266                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2267                                        0);
2268 }
2269
2270 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2271 {
2272         unsigned int psegs, grants, memflags;
2273         int err, i;
2274         struct blkfront_info *info = rinfo->dev_info;
2275
2276         memflags = memalloc_noio_save();
2277
2278         if (info->max_indirect_segments == 0) {
2279                 if (!HAS_EXTRA_REQ)
2280                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2281                 else {
2282                         /*
2283                          * When an extra req is required, the maximum
2284                          * grants supported is related to the size of the
2285                          * Linux block segment.
2286                          */
2287                         grants = GRANTS_PER_PSEG;
2288                 }
2289         }
2290         else
2291                 grants = info->max_indirect_segments;
2292         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2293
2294         err = fill_grant_buffer(rinfo,
2295                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2296         if (err)
2297                 goto out_of_memory;
2298
2299         if (!info->bounce && info->max_indirect_segments) {
2300                 /*
2301                  * We are using indirect descriptors but don't have a bounce
2302                  * buffer, we need to allocate a set of pages that can be
2303                  * used for mapping indirect grefs
2304                  */
2305                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2306
2307                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2308                 for (i = 0; i < num; i++) {
2309                         struct page *indirect_page = alloc_page(GFP_KERNEL |
2310                                                                 __GFP_ZERO);
2311                         if (!indirect_page)
2312                                 goto out_of_memory;
2313                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2314                 }
2315         }
2316
2317         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2318                 rinfo->shadow[i].grants_used =
2319                         kvcalloc(grants,
2320                                  sizeof(rinfo->shadow[i].grants_used[0]),
2321                                  GFP_KERNEL);
2322                 rinfo->shadow[i].sg = kvcalloc(psegs,
2323                                                sizeof(rinfo->shadow[i].sg[0]),
2324                                                GFP_KERNEL);
2325                 if (info->max_indirect_segments)
2326                         rinfo->shadow[i].indirect_grants =
2327                                 kvcalloc(INDIRECT_GREFS(grants),
2328                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2329                                          GFP_KERNEL);
2330                 if ((rinfo->shadow[i].grants_used == NULL) ||
2331                         (rinfo->shadow[i].sg == NULL) ||
2332                      (info->max_indirect_segments &&
2333                      (rinfo->shadow[i].indirect_grants == NULL)))
2334                         goto out_of_memory;
2335                 sg_init_table(rinfo->shadow[i].sg, psegs);
2336         }
2337
2338         memalloc_noio_restore(memflags);
2339
2340         return 0;
2341
2342 out_of_memory:
2343         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2344                 kvfree(rinfo->shadow[i].grants_used);
2345                 rinfo->shadow[i].grants_used = NULL;
2346                 kvfree(rinfo->shadow[i].sg);
2347                 rinfo->shadow[i].sg = NULL;
2348                 kvfree(rinfo->shadow[i].indirect_grants);
2349                 rinfo->shadow[i].indirect_grants = NULL;
2350         }
2351         if (!list_empty(&rinfo->indirect_pages)) {
2352                 struct page *indirect_page, *n;
2353                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2354                         list_del(&indirect_page->lru);
2355                         __free_page(indirect_page);
2356                 }
2357         }
2358
2359         memalloc_noio_restore(memflags);
2360
2361         return -ENOMEM;
2362 }
2363
2364 /*
2365  * Gather all backend feature-*
2366  */
2367 static void blkfront_gather_backend_features(struct blkfront_info *info)
2368 {
2369         unsigned int indirect_segments;
2370
2371         info->feature_flush = 0;
2372         info->feature_fua = 0;
2373
2374         /*
2375          * If there's no "feature-barrier" defined, then it means
2376          * we're dealing with a very old backend which writes
2377          * synchronously; nothing to do.
2378          *
2379          * If there are barriers, then we use flush.
2380          */
2381         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2382                 info->feature_flush = 1;
2383                 info->feature_fua = 1;
2384         }
2385
2386         /*
2387          * And if there is "feature-flush-cache" use that above
2388          * barriers.
2389          */
2390         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2391                                  0)) {
2392                 info->feature_flush = 1;
2393                 info->feature_fua = 0;
2394         }
2395
2396         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2397                 blkfront_setup_discard(info);
2398
2399         if (info->feature_persistent_parm)
2400                 info->feature_persistent =
2401                         !!xenbus_read_unsigned(info->xbdev->otherend,
2402                                                "feature-persistent", 0);
2403         if (info->feature_persistent)
2404                 info->bounce = true;
2405
2406         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2407                                         "feature-max-indirect-segments", 0);
2408         if (indirect_segments > xen_blkif_max_segments)
2409                 indirect_segments = xen_blkif_max_segments;
2410         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2411                 indirect_segments = 0;
2412         info->max_indirect_segments = indirect_segments;
2413
2414         if (info->feature_persistent) {
2415                 mutex_lock(&blkfront_mutex);
2416                 schedule_delayed_work(&blkfront_work, HZ * 10);
2417                 mutex_unlock(&blkfront_mutex);
2418         }
2419 }
2420
2421 /*
2422  * Invoked when the backend is finally 'ready' (and has told produced
2423  * the details about the physical device - #sectors, size, etc).
2424  */
2425 static void blkfront_connect(struct blkfront_info *info)
2426 {
2427         unsigned long long sectors;
2428         unsigned long sector_size;
2429         unsigned int physical_sector_size;
2430         unsigned int binfo;
2431         int err, i;
2432         struct blkfront_ring_info *rinfo;
2433
2434         switch (info->connected) {
2435         case BLKIF_STATE_CONNECTED:
2436                 /*
2437                  * Potentially, the back-end may be signalling
2438                  * a capacity change; update the capacity.
2439                  */
2440                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2441                                    "sectors", "%Lu", &sectors);
2442                 if (XENBUS_EXIST_ERR(err))
2443                         return;
2444                 printk(KERN_INFO "Setting capacity to %Lu\n",
2445                        sectors);
2446                 set_capacity_revalidate_and_notify(info->gd, sectors, true);
2447
2448                 return;
2449         case BLKIF_STATE_SUSPENDED:
2450                 /*
2451                  * If we are recovering from suspension, we need to wait
2452                  * for the backend to announce it's features before
2453                  * reconnecting, at least we need to know if the backend
2454                  * supports indirect descriptors, and how many.
2455                  */
2456                 blkif_recover(info);
2457                 return;
2458
2459         default:
2460                 break;
2461         }
2462
2463         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2464                 __func__, info->xbdev->otherend);
2465
2466         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2467                             "sectors", "%llu", &sectors,
2468                             "info", "%u", &binfo,
2469                             "sector-size", "%lu", &sector_size,
2470                             NULL);
2471         if (err) {
2472                 xenbus_dev_fatal(info->xbdev, err,
2473                                  "reading backend fields at %s",
2474                                  info->xbdev->otherend);
2475                 return;
2476         }
2477
2478         /*
2479          * physcial-sector-size is a newer field, so old backends may not
2480          * provide this. Assume physical sector size to be the same as
2481          * sector_size in that case.
2482          */
2483         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2484                                                     "physical-sector-size",
2485                                                     sector_size);
2486         blkfront_gather_backend_features(info);
2487         for_each_rinfo(info, rinfo, i) {
2488                 err = blkfront_setup_indirect(rinfo);
2489                 if (err) {
2490                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2491                                          info->xbdev->otherend);
2492                         blkif_free(info, 0);
2493                         break;
2494                 }
2495         }
2496
2497         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2498                                   physical_sector_size);
2499         if (err) {
2500                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2501                                  info->xbdev->otherend);
2502                 goto fail;
2503         }
2504
2505         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2506
2507         /* Kick pending requests. */
2508         info->connected = BLKIF_STATE_CONNECTED;
2509         for_each_rinfo(info, rinfo, i)
2510                 kick_pending_request_queues(rinfo);
2511
2512         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2513
2514         info->is_ready = 1;
2515         return;
2516
2517 fail:
2518         blkif_free(info, 0);
2519         return;
2520 }
2521
2522 /**
2523  * Callback received when the backend's state changes.
2524  */
2525 static void blkback_changed(struct xenbus_device *dev,
2526                             enum xenbus_state backend_state)
2527 {
2528         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2529
2530         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2531
2532         switch (backend_state) {
2533         case XenbusStateInitWait:
2534                 if (dev->state != XenbusStateInitialising)
2535                         break;
2536                 if (talk_to_blkback(dev, info))
2537                         break;
2538         case XenbusStateInitialising:
2539         case XenbusStateInitialised:
2540         case XenbusStateReconfiguring:
2541         case XenbusStateReconfigured:
2542         case XenbusStateUnknown:
2543                 break;
2544
2545         case XenbusStateConnected:
2546                 /*
2547                  * talk_to_blkback sets state to XenbusStateInitialised
2548                  * and blkfront_connect sets it to XenbusStateConnected
2549                  * (if connection went OK).
2550                  *
2551                  * If the backend (or toolstack) decides to poke at backend
2552                  * state (and re-trigger the watch by setting the state repeatedly
2553                  * to XenbusStateConnected (4)) we need to deal with this.
2554                  * This is allowed as this is used to communicate to the guest
2555                  * that the size of disk has changed!
2556                  */
2557                 if ((dev->state != XenbusStateInitialised) &&
2558                     (dev->state != XenbusStateConnected)) {
2559                         if (talk_to_blkback(dev, info))
2560                                 break;
2561                 }
2562
2563                 blkfront_connect(info);
2564                 break;
2565
2566         case XenbusStateClosed:
2567                 if (dev->state == XenbusStateClosed)
2568                         break;
2569                 fallthrough;
2570         case XenbusStateClosing:
2571                 if (info)
2572                         blkfront_closing(info);
2573                 break;
2574         }
2575 }
2576
2577 static int blkfront_remove(struct xenbus_device *xbdev)
2578 {
2579         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2580         struct block_device *bdev = NULL;
2581         struct gendisk *disk;
2582
2583         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2584
2585         if (!info)
2586                 return 0;
2587
2588         blkif_free(info, 0);
2589
2590         mutex_lock(&info->mutex);
2591
2592         disk = info->gd;
2593         if (disk)
2594                 bdev = bdget_disk(disk, 0);
2595
2596         info->xbdev = NULL;
2597         mutex_unlock(&info->mutex);
2598
2599         if (!bdev) {
2600                 mutex_lock(&blkfront_mutex);
2601                 free_info(info);
2602                 mutex_unlock(&blkfront_mutex);
2603                 return 0;
2604         }
2605
2606         /*
2607          * The xbdev was removed before we reached the Closed
2608          * state. See if it's safe to remove the disk. If the bdev
2609          * isn't closed yet, we let release take care of it.
2610          */
2611
2612         mutex_lock(&bdev->bd_mutex);
2613         info = disk->private_data;
2614
2615         dev_warn(disk_to_dev(disk),
2616                  "%s was hot-unplugged, %d stale handles\n",
2617                  xbdev->nodename, bdev->bd_openers);
2618
2619         if (info && !bdev->bd_openers) {
2620                 xlvbd_release_gendisk(info);
2621                 disk->private_data = NULL;
2622                 mutex_lock(&blkfront_mutex);
2623                 free_info(info);
2624                 mutex_unlock(&blkfront_mutex);
2625         }
2626
2627         mutex_unlock(&bdev->bd_mutex);
2628         bdput(bdev);
2629
2630         return 0;
2631 }
2632
2633 static int blkfront_is_ready(struct xenbus_device *dev)
2634 {
2635         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2636
2637         return info->is_ready && info->xbdev;
2638 }
2639
2640 static int blkif_open(struct block_device *bdev, fmode_t mode)
2641 {
2642         struct gendisk *disk = bdev->bd_disk;
2643         struct blkfront_info *info;
2644         int err = 0;
2645
2646         mutex_lock(&blkfront_mutex);
2647
2648         info = disk->private_data;
2649         if (!info) {
2650                 /* xbdev gone */
2651                 err = -ERESTARTSYS;
2652                 goto out;
2653         }
2654
2655         mutex_lock(&info->mutex);
2656
2657         if (!info->gd)
2658                 /* xbdev is closed */
2659                 err = -ERESTARTSYS;
2660
2661         mutex_unlock(&info->mutex);
2662
2663 out:
2664         mutex_unlock(&blkfront_mutex);
2665         return err;
2666 }
2667
2668 static void blkif_release(struct gendisk *disk, fmode_t mode)
2669 {
2670         struct blkfront_info *info = disk->private_data;
2671         struct block_device *bdev;
2672         struct xenbus_device *xbdev;
2673
2674         mutex_lock(&blkfront_mutex);
2675
2676         bdev = bdget_disk(disk, 0);
2677
2678         if (!bdev) {
2679                 WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
2680                 goto out_mutex;
2681         }
2682         if (bdev->bd_openers)
2683                 goto out;
2684
2685         /*
2686          * Check if we have been instructed to close. We will have
2687          * deferred this request, because the bdev was still open.
2688          */
2689
2690         mutex_lock(&info->mutex);
2691         xbdev = info->xbdev;
2692
2693         if (xbdev && xbdev->state == XenbusStateClosing) {
2694                 /* pending switch to state closed */
2695                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2696                 xlvbd_release_gendisk(info);
2697                 xenbus_frontend_closed(info->xbdev);
2698         }
2699
2700         mutex_unlock(&info->mutex);
2701
2702         if (!xbdev) {
2703                 /* sudden device removal */
2704                 dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
2705                 xlvbd_release_gendisk(info);
2706                 disk->private_data = NULL;
2707                 free_info(info);
2708         }
2709
2710 out:
2711         bdput(bdev);
2712 out_mutex:
2713         mutex_unlock(&blkfront_mutex);
2714 }
2715
2716 static const struct block_device_operations xlvbd_block_fops =
2717 {
2718         .owner = THIS_MODULE,
2719         .open = blkif_open,
2720         .release = blkif_release,
2721         .getgeo = blkif_getgeo,
2722         .ioctl = blkif_ioctl,
2723         .compat_ioctl = blkdev_compat_ptr_ioctl,
2724 };
2725
2726
2727 static const struct xenbus_device_id blkfront_ids[] = {
2728         { "vbd" },
2729         { "" }
2730 };
2731
2732 static struct xenbus_driver blkfront_driver = {
2733         .ids  = blkfront_ids,
2734         .probe = blkfront_probe,
2735         .remove = blkfront_remove,
2736         .resume = blkfront_resume,
2737         .otherend_changed = blkback_changed,
2738         .is_ready = blkfront_is_ready,
2739 };
2740
2741 static void purge_persistent_grants(struct blkfront_info *info)
2742 {
2743         unsigned int i;
2744         unsigned long flags;
2745         struct blkfront_ring_info *rinfo;
2746
2747         for_each_rinfo(info, rinfo, i) {
2748                 struct grant *gnt_list_entry, *tmp;
2749
2750                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2751
2752                 if (rinfo->persistent_gnts_c == 0) {
2753                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2754                         continue;
2755                 }
2756
2757                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2758                                          node) {
2759                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2760                             !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2761                                 continue;
2762
2763                         list_del(&gnt_list_entry->node);
2764                         rinfo->persistent_gnts_c--;
2765                         gnt_list_entry->gref = GRANT_INVALID_REF;
2766                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2767                 }
2768
2769                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2770         }
2771 }
2772
2773 static void blkfront_delay_work(struct work_struct *work)
2774 {
2775         struct blkfront_info *info;
2776         bool need_schedule_work = false;
2777
2778         /*
2779          * Note that when using bounce buffers but not persistent grants
2780          * there's no need to run blkfront_delay_work because grants are
2781          * revoked in blkif_completion or else an error is reported and the
2782          * connection is closed.
2783          */
2784
2785         mutex_lock(&blkfront_mutex);
2786
2787         list_for_each_entry(info, &info_list, info_list) {
2788                 if (info->feature_persistent) {
2789                         need_schedule_work = true;
2790                         mutex_lock(&info->mutex);
2791                         purge_persistent_grants(info);
2792                         mutex_unlock(&info->mutex);
2793                 }
2794         }
2795
2796         if (need_schedule_work)
2797                 schedule_delayed_work(&blkfront_work, HZ * 10);
2798
2799         mutex_unlock(&blkfront_mutex);
2800 }
2801
2802 static int __init xlblk_init(void)
2803 {
2804         int ret;
2805         int nr_cpus = num_online_cpus();
2806
2807         if (!xen_domain())
2808                 return -ENODEV;
2809
2810         if (!xen_has_pv_disk_devices())
2811                 return -ENODEV;
2812
2813         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2814                 pr_warn("xen_blk: can't get major %d with name %s\n",
2815                         XENVBD_MAJOR, DEV_NAME);
2816                 return -ENODEV;
2817         }
2818
2819         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2820                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2821
2822         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2823                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2824                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2825                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2826         }
2827
2828         if (xen_blkif_max_queues > nr_cpus) {
2829                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2830                         xen_blkif_max_queues, nr_cpus);
2831                 xen_blkif_max_queues = nr_cpus;
2832         }
2833
2834         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2835
2836         ret = xenbus_register_frontend(&blkfront_driver);
2837         if (ret) {
2838                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2839                 return ret;
2840         }
2841
2842         return 0;
2843 }
2844 module_init(xlblk_init);
2845
2846
2847 static void __exit xlblk_exit(void)
2848 {
2849         cancel_delayed_work_sync(&blkfront_work);
2850
2851         xenbus_unregister_driver(&blkfront_driver);
2852         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2853         kfree(minors);
2854 }
2855 module_exit(xlblk_exit);
2856
2857 MODULE_DESCRIPTION("Xen virtual block device frontend");
2858 MODULE_LICENSE("GPL");
2859 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2860 MODULE_ALIAS("xen:vbd");
2861 MODULE_ALIAS("xenblk");