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/major.h>
  46 #include <linux/mutex.h>
  47 #include <linux/scatterlist.h>
  48 #include <linux/bitmap.h>
  49 #include <linux/list.h>
  50 #include <linux/workqueue.h>
  51 #include <linux/sched/mm.h>
  52
  53 #include <xen/xen.h>
  54 #include <xen/xenbus.h>
  55 #include <xen/grant_table.h>
  56 #include <xen/events.h>
  57 #include <xen/page.h>
  58 #include <xen/platform_pci.h>
  59
  60 #include <xen/interface/grant_table.h>
  61 #include <xen/interface/io/blkif.h>
  62 #include <xen/interface/io/protocols.h>
  63
  64 #include <asm/xen/hypervisor.h>
  65
  66 /*
  67  * The minimal size of segment supported by the block framework is PAGE_SIZE.
  68  * When Linux is using a different page size than Xen, it may not be possible
  69  * to put all the data in a single segment.
  70  * This can happen when the backend doesn't support indirect descriptor and
  71  * therefore the maximum amount of data that a request can carry is
  72  * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE = 44KB
  73  *
  74  * Note that we only support one extra request. So the Linux page size
  75  * should be <= ( 2 * BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) =
  76  * 88KB.
  77  */
  78 #define HAS_EXTRA_REQ (BLKIF_MAX_SEGMENTS_PER_REQUEST < XEN_PFN_PER_PAGE)
  79
  80 enum blkif_state {
  81         BLKIF_STATE_DISCONNECTED,
  82         BLKIF_STATE_CONNECTED,
  83         BLKIF_STATE_SUSPENDED,
  84         BLKIF_STATE_ERROR,
  85 };
  86
  87 struct grant {
  88         grant_ref_t gref;
  89         struct page *page;
  90         struct list_head node;
  91 };
  92
  93 enum blk_req_status {
  94         REQ_PROCESSING,
  95         REQ_WAITING,
  96         REQ_DONE,
  97         REQ_ERROR,
  98         REQ_EOPNOTSUPP,
  99 };
 100
 101 struct blk_shadow {
 102         struct blkif_request req;
 103         struct request *request;
 104         struct grant **grants_used;
 105         struct grant **indirect_grants;
 106         struct scatterlist *sg;
 107         unsigned int num_sg;
 108         enum blk_req_status status;
 109
 110         #define NO_ASSOCIATED_ID ~0UL
 111         /*
 112          * Id of the sibling if we ever need 2 requests when handling a
 113          * block I/O request
 114          */
 115         unsigned long associated_id;
 116 };
 117
 118 struct blkif_req {
 119         blk_status_t    error;
 120 };
 121
 122 static inline struct blkif_req *blkif_req(struct request *rq)
 123 {
 124         return blk_mq_rq_to_pdu(rq);
 125 }
 126
 127 static DEFINE_MUTEX(blkfront_mutex);
 128 static const struct block_device_operations xlvbd_block_fops;
 129 static struct delayed_work blkfront_work;
 130 static LIST_HEAD(info_list);
 131
 132 /*
 133  * Maximum number of segments in indirect requests, the actual value used by
 134  * the frontend driver is the minimum of this value and the value provided
 135  * by the backend driver.
 136  */
 137
 138 static unsigned int xen_blkif_max_segments = 32;
 139 module_param_named(max_indirect_segments, xen_blkif_max_segments, uint, 0444);
 140 MODULE_PARM_DESC(max_indirect_segments,
 141                  "Maximum amount of segments in indirect requests (default is 32)");
 142
 143 static unsigned int xen_blkif_max_queues = 4;
 144 module_param_named(max_queues, xen_blkif_max_queues, uint, 0444);
 145 MODULE_PARM_DESC(max_queues, "Maximum number of hardware queues/rings used per virtual disk");
 146
 147 /*
 148  * Maximum order of pages to be used for the shared ring between front and
 149  * backend, 4KB page granularity is used.
 150  */
 151 static unsigned int xen_blkif_max_ring_order;
 152 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, 0444);
 153 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
 154
 155 static bool __read_mostly xen_blkif_trusted = true;
 156 module_param_named(trusted, xen_blkif_trusted, bool, 0644);
 157 MODULE_PARM_DESC(trusted, "Is the backend trusted");
 158
 159 #define BLK_RING_SIZE(info)     \
 160         __CONST_RING_SIZE(blkif, XEN_PAGE_SIZE * (info)->nr_ring_pages)
 161
 162 /*
 163  * ring-ref%u i=(-1UL) would take 11 characters + 'ring-ref' is 8, so 19
 164  * characters are enough. Define to 20 to keep consistent with backend.
 165  */
 166 #define RINGREF_NAME_LEN (20)
 167 /*
 168  * queue-%u would take 7 + 10(UINT_MAX) = 17 characters.
 169  */
 170 #define QUEUE_NAME_LEN (17)
 171
 172 /*
 173  *  Per-ring info.
 174  *  Every blkfront device can associate with one or more blkfront_ring_info,
 175  *  depending on how many hardware queues/rings to be used.
 176  */
 177 struct blkfront_ring_info {
 178         /* Lock to protect data in every ring buffer. */
 179         spinlock_t ring_lock;
 180         struct blkif_front_ring ring;
 181         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 182         unsigned int evtchn, irq;
 183         struct work_struct work;
 184         struct gnttab_free_callback callback;
 185         struct list_head indirect_pages;
 186         struct list_head grants;
 187         unsigned int persistent_gnts_c;
 188         unsigned long shadow_free;
 189         struct blkfront_info *dev_info;
 190         struct blk_shadow shadow[];
 191 };
 192
 193 /*
 194  * We have one of these per vbd, whether ide, scsi or 'other'.  They
 195  * hang in private_data off the gendisk structure. We may end up
 196  * putting all kinds of interesting stuff here :-)
 197  */
 198 struct blkfront_info
 199 {
 200         struct mutex mutex;
 201         struct xenbus_device *xbdev;
 202         struct gendisk *gd;
 203         u16 sector_size;
 204         unsigned int physical_sector_size;
 205         int vdevice;
 206         blkif_vdev_t handle;
 207         enum blkif_state connected;
 208         /* Number of pages per ring buffer. */
 209         unsigned int nr_ring_pages;
 210         struct request_queue *rq;
 211         unsigned int feature_flush:1;
 212         unsigned int feature_fua:1;
 213         unsigned int feature_discard:1;
 214         unsigned int feature_secdiscard:1;
 215         /* Connect-time cached feature_persistent parameter */
 216         unsigned int feature_persistent_parm:1;
 217         /* Persistent grants feature negotiation result */
 218         unsigned int feature_persistent:1;
 219         unsigned int bounce:1;
 220         unsigned int discard_granularity;
 221         unsigned int discard_alignment;
 222         /* Number of 4KB segments handled */
 223         unsigned int max_indirect_segments;
 224         int is_ready;
 225         struct blk_mq_tag_set tag_set;
 226         struct blkfront_ring_info *rinfo;
 227         unsigned int nr_rings;
 228         unsigned int rinfo_size;
 229         /* Save uncomplete reqs and bios for migration. */
 230         struct list_head requests;
 231         struct bio_list bio_list;
 232         struct list_head info_list;
 233 };
 234
 235 static unsigned int nr_minors;
 236 static unsigned long *minors;
 237 static DEFINE_SPINLOCK(minor_lock);
 238
 239 #define GRANT_INVALID_REF       0
 240
 241 #define PARTS_PER_DISK          16
 242 #define PARTS_PER_EXT_DISK      256
 243
 244 #define BLKIF_MAJOR(dev) ((dev)>>8)
 245 #define BLKIF_MINOR(dev) ((dev) & 0xff)
 246
 247 #define EXT_SHIFT 28
 248 #define EXTENDED (1<<EXT_SHIFT)
 249 #define VDEV_IS_EXTENDED(dev) ((dev)&(EXTENDED))
 250 #define BLKIF_MINOR_EXT(dev) ((dev)&(~EXTENDED))
 251 #define EMULATED_HD_DISK_MINOR_OFFSET (0)
 252 #define EMULATED_HD_DISK_NAME_OFFSET (EMULATED_HD_DISK_MINOR_OFFSET / 256)
 253 #define EMULATED_SD_DISK_MINOR_OFFSET (0)
 254 #define EMULATED_SD_DISK_NAME_OFFSET (EMULATED_SD_DISK_MINOR_OFFSET / 256)
 255
 256 #define DEV_NAME        "xvd"   /* name in /dev */
 257
 258 /*
 259  * Grants are always the same size as a Xen page (i.e 4KB).
 260  * A physical segment is always the same size as a Linux page.
 261  * Number of grants per physical segment
 262  */
 263 #define GRANTS_PER_PSEG (PAGE_SIZE / XEN_PAGE_SIZE)
 264
 265 #define GRANTS_PER_INDIRECT_FRAME \
 266         (XEN_PAGE_SIZE / sizeof(struct blkif_request_segment))
 267
 268 #define INDIRECT_GREFS(_grants)         \
 269         DIV_ROUND_UP(_grants, GRANTS_PER_INDIRECT_FRAME)
 270
 271 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo);
 272 static void blkfront_gather_backend_features(struct blkfront_info *info);
 273 static int negotiate_mq(struct blkfront_info *info);
 274
 275 #define for_each_rinfo(info, ptr, idx)                          \
 276         for ((ptr) = (info)->rinfo, (idx) = 0;                  \
 277              (idx) < (info)->nr_rings;                          \
 278              (idx)++, (ptr) = (void *)(ptr) + (info)->rinfo_size)
 279
 280 static inline struct blkfront_ring_info *
 281 get_rinfo(const struct blkfront_info *info, unsigned int i)
 282 {
 283         BUG_ON(i >= info->nr_rings);
 284         return (void *)info->rinfo + i * info->rinfo_size;
 285 }
 286
 287 static int get_id_from_freelist(struct blkfront_ring_info *rinfo)
 288 {
 289         unsigned long free = rinfo->shadow_free;
 290
 291         BUG_ON(free >= BLK_RING_SIZE(rinfo->dev_info));
 292         rinfo->shadow_free = rinfo->shadow[free].req.u.rw.id;
 293         rinfo->shadow[free].req.u.rw.id = 0x0fffffee; /* debug */
 294         return free;
 295 }
 296
 297 static int add_id_to_freelist(struct blkfront_ring_info *rinfo,
 298                               unsigned long id)
 299 {
 300         if (rinfo->shadow[id].req.u.rw.id != id)
 301                 return -EINVAL;
 302         if (rinfo->shadow[id].request == NULL)
 303                 return -EINVAL;
 304         rinfo->shadow[id].req.u.rw.id  = rinfo->shadow_free;
 305         rinfo->shadow[id].request = NULL;
 306         rinfo->shadow_free = id;
 307         return 0;
 308 }
 309
 310 static int fill_grant_buffer(struct blkfront_ring_info *rinfo, int num)
 311 {
 312         struct blkfront_info *info = rinfo->dev_info;
 313         struct page *granted_page;
 314         struct grant *gnt_list_entry, *n;
 315         int i = 0;
 316
 317         while (i < num) {
 318                 gnt_list_entry = kzalloc(sizeof(struct grant), GFP_NOIO);
 319                 if (!gnt_list_entry)
 320                         goto out_of_memory;
 321
 322                 if (info->bounce) {
 323                         granted_page = alloc_page(GFP_NOIO | __GFP_ZERO);
 324                         if (!granted_page) {
 325                                 kfree(gnt_list_entry);
 326                                 goto out_of_memory;
 327                         }
 328                         gnt_list_entry->page = granted_page;
 329                 }
 330
 331                 gnt_list_entry->gref = GRANT_INVALID_REF;
 332                 list_add(&gnt_list_entry->node, &rinfo->grants);
 333                 i++;
 334         }
 335
 336         return 0;
 337
 338 out_of_memory:
 339         list_for_each_entry_safe(gnt_list_entry, n,
 340                                  &rinfo->grants, node) {
 341                 list_del(&gnt_list_entry->node);
 342                 if (info->bounce)
 343                         __free_page(gnt_list_entry->page);
 344                 kfree(gnt_list_entry);
 345                 i--;
 346         }
 347         BUG_ON(i != 0);
 348         return -ENOMEM;
 349 }
 350
 351 static struct grant *get_free_grant(struct blkfront_ring_info *rinfo)
 352 {
 353         struct grant *gnt_list_entry;
 354
 355         BUG_ON(list_empty(&rinfo->grants));
 356         gnt_list_entry = list_first_entry(&rinfo->grants, struct grant,
 357                                           node);
 358         list_del(&gnt_list_entry->node);
 359
 360         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 361                 rinfo->persistent_gnts_c--;
 362
 363         return gnt_list_entry;
 364 }
 365
 366 static inline void grant_foreign_access(const struct grant *gnt_list_entry,
 367                                         const struct blkfront_info *info)
 368 {
 369         gnttab_page_grant_foreign_access_ref_one(gnt_list_entry->gref,
 370                                                  info->xbdev->otherend_id,
 371                                                  gnt_list_entry->page,
 372                                                  0);
 373 }
 374
 375 static struct grant *get_grant(grant_ref_t *gref_head,
 376                                unsigned long gfn,
 377                                struct blkfront_ring_info *rinfo)
 378 {
 379         struct grant *gnt_list_entry = get_free_grant(rinfo);
 380         struct blkfront_info *info = rinfo->dev_info;
 381
 382         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 383                 return gnt_list_entry;
 384
 385         /* Assign a gref to this page */
 386         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 387         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 388         if (info->bounce)
 389                 grant_foreign_access(gnt_list_entry, info);
 390         else {
 391                 /* Grant access to the GFN passed by the caller */
 392                 gnttab_grant_foreign_access_ref(gnt_list_entry->gref,
 393                                                 info->xbdev->otherend_id,
 394                                                 gfn, 0);
 395         }
 396
 397         return gnt_list_entry;
 398 }
 399
 400 static struct grant *get_indirect_grant(grant_ref_t *gref_head,
 401                                         struct blkfront_ring_info *rinfo)
 402 {
 403         struct grant *gnt_list_entry = get_free_grant(rinfo);
 404         struct blkfront_info *info = rinfo->dev_info;
 405
 406         if (gnt_list_entry->gref != GRANT_INVALID_REF)
 407                 return gnt_list_entry;
 408
 409         /* Assign a gref to this page */
 410         gnt_list_entry->gref = gnttab_claim_grant_reference(gref_head);
 411         BUG_ON(gnt_list_entry->gref == -ENOSPC);
 412         if (!info->bounce) {
 413                 struct page *indirect_page;
 414
 415                 /* Fetch a pre-allocated page to use for indirect grefs */
 416                 BUG_ON(list_empty(&rinfo->indirect_pages));
 417                 indirect_page = list_first_entry(&rinfo->indirect_pages,
 418                                                  struct page, lru);
 419                 list_del(&indirect_page->lru);
 420                 gnt_list_entry->page = indirect_page;
 421         }
 422         grant_foreign_access(gnt_list_entry, info);
 423
 424         return gnt_list_entry;
 425 }
 426
 427 static const char *op_name(int op)
 428 {
 429         static const char *const names[] = {
 430                 [BLKIF_OP_READ] = "read",
 431                 [BLKIF_OP_WRITE] = "write",
 432                 [BLKIF_OP_WRITE_BARRIER] = "barrier",
 433                 [BLKIF_OP_FLUSH_DISKCACHE] = "flush",
 434                 [BLKIF_OP_DISCARD] = "discard" };
 435
 436         if (op < 0 || op >= ARRAY_SIZE(names))
 437                 return "unknown";
 438
 439         if (!names[op])
 440                 return "reserved";
 441
 442         return names[op];
 443 }
 444 static int xlbd_reserve_minors(unsigned int minor, unsigned int nr)
 445 {
 446         unsigned int end = minor + nr;
 447         int rc;
 448
 449         if (end > nr_minors) {
 450                 unsigned long *bitmap, *old;
 451
 452                 bitmap = kcalloc(BITS_TO_LONGS(end), sizeof(*bitmap),
 453                                  GFP_KERNEL);
 454                 if (bitmap == NULL)
 455                         return -ENOMEM;
 456
 457                 spin_lock(&minor_lock);
 458                 if (end > nr_minors) {
 459                         old = minors;
 460                         memcpy(bitmap, minors,
 461                                BITS_TO_LONGS(nr_minors) * sizeof(*bitmap));
 462                         minors = bitmap;
 463                         nr_minors = BITS_TO_LONGS(end) * BITS_PER_LONG;
 464                 } else
 465                         old = bitmap;
 466                 spin_unlock(&minor_lock);
 467                 kfree(old);
 468         }
 469
 470         spin_lock(&minor_lock);
 471         if (find_next_bit(minors, end, minor) >= end) {
 472                 bitmap_set(minors, minor, nr);
 473                 rc = 0;
 474         } else
 475                 rc = -EBUSY;
 476         spin_unlock(&minor_lock);
 477
 478         return rc;
 479 }
 480
 481 static void xlbd_release_minors(unsigned int minor, unsigned int nr)
 482 {
 483         unsigned int end = minor + nr;
 484
 485         BUG_ON(end > nr_minors);
 486         spin_lock(&minor_lock);
 487         bitmap_clear(minors,  minor, nr);
 488         spin_unlock(&minor_lock);
 489 }
 490
 491 static void blkif_restart_queue_callback(void *arg)
 492 {
 493         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)arg;
 494         schedule_work(&rinfo->work);
 495 }
 496
 497 static int blkif_getgeo(struct block_device *bd, struct hd_geometry *hg)
 498 {
 499         /* We don't have real geometry info, but let's at least return
 500            values consistent with the size of the device */
 501         sector_t nsect = get_capacity(bd->bd_disk);
 502         sector_t cylinders = nsect;
 503
 504         hg->heads = 0xff;
 505         hg->sectors = 0x3f;
 506         sector_div(cylinders, hg->heads * hg->sectors);
 507         hg->cylinders = cylinders;
 508         if ((sector_t)(hg->cylinders + 1) * hg->heads * hg->sectors < nsect)
 509                 hg->cylinders = 0xffff;
 510         return 0;
 511 }
 512
 513 static int blkif_ioctl(struct block_device *bdev, fmode_t mode,
 514                        unsigned command, unsigned long argument)
 515 {
 516         int i;
 517
 518         switch (command) {
 519         case CDROMMULTISESSION:
 520                 for (i = 0; i < sizeof(struct cdrom_multisession); i++)
 521                         if (put_user(0, (char __user *)(argument + i)))
 522                                 return -EFAULT;
 523                 return 0;
 524         case CDROM_GET_CAPABILITY:
 525                 if (bdev->bd_disk->flags & GENHD_FL_CD)
 526                         return 0;
 527                 return -EINVAL;
 528         default:
 529                 return -EINVAL;
 530         }
 531 }
 532
 533 static unsigned long blkif_ring_get_request(struct blkfront_ring_info *rinfo,
 534                                             struct request *req,
 535                                             struct blkif_request **ring_req)
 536 {
 537         unsigned long id;
 538
 539         *ring_req = RING_GET_REQUEST(&rinfo->ring, rinfo->ring.req_prod_pvt);
 540         rinfo->ring.req_prod_pvt++;
 541
 542         id = get_id_from_freelist(rinfo);
 543         rinfo->shadow[id].request = req;
 544         rinfo->shadow[id].status = REQ_PROCESSING;
 545         rinfo->shadow[id].associated_id = NO_ASSOCIATED_ID;
 546
 547         rinfo->shadow[id].req.u.rw.id = id;
 548
 549         return id;
 550 }
 551
 552 static int blkif_queue_discard_req(struct request *req, struct blkfront_ring_info *rinfo)
 553 {
 554         struct blkfront_info *info = rinfo->dev_info;
 555         struct blkif_request *ring_req, *final_ring_req;
 556         unsigned long id;
 557
 558         /* Fill out a communications ring structure. */
 559         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 560         ring_req = &rinfo->shadow[id].req;
 561
 562         ring_req->operation = BLKIF_OP_DISCARD;
 563         ring_req->u.discard.nr_sectors = blk_rq_sectors(req);
 564         ring_req->u.discard.id = id;
 565         ring_req->u.discard.sector_number = (blkif_sector_t)blk_rq_pos(req);
 566         if (req_op(req) == REQ_OP_SECURE_ERASE && info->feature_secdiscard)
 567                 ring_req->u.discard.flag = BLKIF_DISCARD_SECURE;
 568         else
 569                 ring_req->u.discard.flag = 0;
 570
 571         /* Copy the request to the ring page. */
 572         *final_ring_req = *ring_req;
 573         rinfo->shadow[id].status = REQ_WAITING;
 574
 575         return 0;
 576 }
 577
 578 struct setup_rw_req {
 579         unsigned int grant_idx;
 580         struct blkif_request_segment *segments;
 581         struct blkfront_ring_info *rinfo;
 582         struct blkif_request *ring_req;
 583         grant_ref_t gref_head;
 584         unsigned int id;
 585         /* Only used when persistent grant is used and it's a read request */
 586         bool need_copy;
 587         unsigned int bvec_off;
 588         char *bvec_data;
 589
 590         bool require_extra_req;
 591         struct blkif_request *extra_ring_req;
 592 };
 593
 594 static void blkif_setup_rw_req_grant(unsigned long gfn, unsigned int offset,
 595                                      unsigned int len, void *data)
 596 {
 597         struct setup_rw_req *setup = data;
 598         int n, ref;
 599         struct grant *gnt_list_entry;
 600         unsigned int fsect, lsect;
 601         /* Convenient aliases */
 602         unsigned int grant_idx = setup->grant_idx;
 603         struct blkif_request *ring_req = setup->ring_req;
 604         struct blkfront_ring_info *rinfo = setup->rinfo;
 605         /*
 606          * We always use the shadow of the first request to store the list
 607          * of grant associated to the block I/O request. This made the
 608          * completion more easy to handle even if the block I/O request is
 609          * split.
 610          */
 611         struct blk_shadow *shadow = &rinfo->shadow[setup->id];
 612
 613         if (unlikely(setup->require_extra_req &&
 614                      grant_idx >= BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 615                 /*
 616                  * We are using the second request, setup grant_idx
 617                  * to be the index of the segment array.
 618                  */
 619                 grant_idx -= BLKIF_MAX_SEGMENTS_PER_REQUEST;
 620                 ring_req = setup->extra_ring_req;
 621         }
 622
 623         if ((ring_req->operation == BLKIF_OP_INDIRECT) &&
 624             (grant_idx % GRANTS_PER_INDIRECT_FRAME == 0)) {
 625                 if (setup->segments)
 626                         kunmap_atomic(setup->segments);
 627
 628                 n = grant_idx / GRANTS_PER_INDIRECT_FRAME;
 629                 gnt_list_entry = get_indirect_grant(&setup->gref_head, rinfo);
 630                 shadow->indirect_grants[n] = gnt_list_entry;
 631                 setup->segments = kmap_atomic(gnt_list_entry->page);
 632                 ring_req->u.indirect.indirect_grefs[n] = gnt_list_entry->gref;
 633         }
 634
 635         gnt_list_entry = get_grant(&setup->gref_head, gfn, rinfo);
 636         ref = gnt_list_entry->gref;
 637         /*
 638          * All the grants are stored in the shadow of the first
 639          * request. Therefore we have to use the global index.
 640          */
 641         shadow->grants_used[setup->grant_idx] = gnt_list_entry;
 642
 643         if (setup->need_copy) {
 644                 void *shared_data;
 645
 646                 shared_data = kmap_atomic(gnt_list_entry->page);
 647                 /*
 648                  * this does not wipe data stored outside the
 649                  * range sg->offset..sg->offset+sg->length.
 650                  * Therefore, blkback *could* see data from
 651                  * previous requests. This is OK as long as
 652                  * persistent grants are shared with just one
 653                  * domain. It may need refactoring if this
 654                  * changes
 655                  */
 656                 memcpy(shared_data + offset,
 657                        setup->bvec_data + setup->bvec_off,
 658                        len);
 659
 660                 kunmap_atomic(shared_data);
 661                 setup->bvec_off += len;
 662         }
 663
 664         fsect = offset >> 9;
 665         lsect = fsect + (len >> 9) - 1;
 666         if (ring_req->operation != BLKIF_OP_INDIRECT) {
 667                 ring_req->u.rw.seg[grant_idx] =
 668                         (struct blkif_request_segment) {
 669                                 .gref       = ref,
 670                                 .first_sect = fsect,
 671                                 .last_sect  = lsect };
 672         } else {
 673                 setup->segments[grant_idx % GRANTS_PER_INDIRECT_FRAME] =
 674                         (struct blkif_request_segment) {
 675                                 .gref       = ref,
 676                                 .first_sect = fsect,
 677                                 .last_sect  = lsect };
 678         }
 679
 680         (setup->grant_idx)++;
 681 }
 682
 683 static void blkif_setup_extra_req(struct blkif_request *first,
 684                                   struct blkif_request *second)
 685 {
 686         uint16_t nr_segments = first->u.rw.nr_segments;
 687
 688         /*
 689          * The second request is only present when the first request uses
 690          * all its segments. It's always the continuity of the first one.
 691          */
 692         first->u.rw.nr_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
 693
 694         second->u.rw.nr_segments = nr_segments - BLKIF_MAX_SEGMENTS_PER_REQUEST;
 695         second->u.rw.sector_number = first->u.rw.sector_number +
 696                 (BLKIF_MAX_SEGMENTS_PER_REQUEST * XEN_PAGE_SIZE) / 512;
 697
 698         second->u.rw.handle = first->u.rw.handle;
 699         second->operation = first->operation;
 700 }
 701
 702 static int blkif_queue_rw_req(struct request *req, struct blkfront_ring_info *rinfo)
 703 {
 704         struct blkfront_info *info = rinfo->dev_info;
 705         struct blkif_request *ring_req, *extra_ring_req = NULL;
 706         struct blkif_request *final_ring_req, *final_extra_ring_req = NULL;
 707         unsigned long id, extra_id = NO_ASSOCIATED_ID;
 708         bool require_extra_req = false;
 709         int i;
 710         struct setup_rw_req setup = {
 711                 .grant_idx = 0,
 712                 .segments = NULL,
 713                 .rinfo = rinfo,
 714                 .need_copy = rq_data_dir(req) && info->bounce,
 715         };
 716
 717         /*
 718          * Used to store if we are able to queue the request by just using
 719          * existing persistent grants, or if we have to get new grants,
 720          * as there are not sufficiently many free.
 721          */
 722         bool new_persistent_gnts = false;
 723         struct scatterlist *sg;
 724         int num_sg, max_grefs, num_grant;
 725
 726         max_grefs = req->nr_phys_segments * GRANTS_PER_PSEG;
 727         if (max_grefs > BLKIF_MAX_SEGMENTS_PER_REQUEST)
 728                 /*
 729                  * If we are using indirect segments we need to account
 730                  * for the indirect grefs used in the request.
 731                  */
 732                 max_grefs += INDIRECT_GREFS(max_grefs);
 733
 734         /* Check if we have enough persistent grants to allocate a requests */
 735         if (rinfo->persistent_gnts_c < max_grefs) {
 736                 new_persistent_gnts = true;
 737
 738                 if (gnttab_alloc_grant_references(
 739                     max_grefs - rinfo->persistent_gnts_c,
 740                     &setup.gref_head) < 0) {
 741                         gnttab_request_free_callback(
 742                                 &rinfo->callback,
 743                                 blkif_restart_queue_callback,
 744                                 rinfo,
 745                                 max_grefs - rinfo->persistent_gnts_c);
 746                         return 1;
 747                 }
 748         }
 749
 750         /* Fill out a communications ring structure. */
 751         id = blkif_ring_get_request(rinfo, req, &final_ring_req);
 752         ring_req = &rinfo->shadow[id].req;
 753
 754         num_sg = blk_rq_map_sg(req->q, req, rinfo->shadow[id].sg);
 755         num_grant = 0;
 756         /* Calculate the number of grant used */
 757         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i)
 758                num_grant += gnttab_count_grant(sg->offset, sg->length);
 759
 760         require_extra_req = info->max_indirect_segments == 0 &&
 761                 num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST;
 762         BUG_ON(!HAS_EXTRA_REQ && require_extra_req);
 763
 764         rinfo->shadow[id].num_sg = num_sg;
 765         if (num_grant > BLKIF_MAX_SEGMENTS_PER_REQUEST &&
 766             likely(!require_extra_req)) {
 767                 /*
 768                  * The indirect operation can only be a BLKIF_OP_READ or
 769                  * BLKIF_OP_WRITE
 770                  */
 771                 BUG_ON(req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA);
 772                 ring_req->operation = BLKIF_OP_INDIRECT;
 773                 ring_req->u.indirect.indirect_op = rq_data_dir(req) ?
 774                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 775                 ring_req->u.indirect.sector_number = (blkif_sector_t)blk_rq_pos(req);
 776                 ring_req->u.indirect.handle = info->handle;
 777                 ring_req->u.indirect.nr_segments = num_grant;
 778         } else {
 779                 ring_req->u.rw.sector_number = (blkif_sector_t)blk_rq_pos(req);
 780                 ring_req->u.rw.handle = info->handle;
 781                 ring_req->operation = rq_data_dir(req) ?
 782                         BLKIF_OP_WRITE : BLKIF_OP_READ;
 783                 if (req_op(req) == REQ_OP_FLUSH || req->cmd_flags & REQ_FUA) {
 784                         /*
 785                          * Ideally we can do an unordered flush-to-disk.
 786                          * In case the backend onlysupports barriers, use that.
 787                          * A barrier request a superset of FUA, so we can
 788                          * implement it the same way.  (It's also a FLUSH+FUA,
 789                          * since it is guaranteed ordered WRT previous writes.)
 790                          */
 791                         if (info->feature_flush && info->feature_fua)
 792                                 ring_req->operation =
 793                                         BLKIF_OP_WRITE_BARRIER;
 794                         else if (info->feature_flush)
 795                                 ring_req->operation =
 796                                         BLKIF_OP_FLUSH_DISKCACHE;
 797                         else
 798                                 ring_req->operation = 0;
 799                 }
 800                 ring_req->u.rw.nr_segments = num_grant;
 801                 if (unlikely(require_extra_req)) {
 802                         extra_id = blkif_ring_get_request(rinfo, req,
 803                                                           &final_extra_ring_req);
 804                         extra_ring_req = &rinfo->shadow[extra_id].req;
 805
 806                         /*
 807                          * Only the first request contains the scatter-gather
 808                          * list.
 809                          */
 810                         rinfo->shadow[extra_id].num_sg = 0;
 811
 812                         blkif_setup_extra_req(ring_req, extra_ring_req);
 813
 814                         /* Link the 2 requests together */
 815                         rinfo->shadow[extra_id].associated_id = id;
 816                         rinfo->shadow[id].associated_id = extra_id;
 817                 }
 818         }
 819
 820         setup.ring_req = ring_req;
 821         setup.id = id;
 822
 823         setup.require_extra_req = require_extra_req;
 824         if (unlikely(require_extra_req))
 825                 setup.extra_ring_req = extra_ring_req;
 826
 827         for_each_sg(rinfo->shadow[id].sg, sg, num_sg, i) {
 828                 BUG_ON(sg->offset + sg->length > PAGE_SIZE);
 829
 830                 if (setup.need_copy) {
 831                         setup.bvec_off = sg->offset;
 832                         setup.bvec_data = kmap_atomic(sg_page(sg));
 833                 }
 834
 835                 gnttab_foreach_grant_in_range(sg_page(sg),
 836                                               sg->offset,
 837                                               sg->length,
 838                                               blkif_setup_rw_req_grant,
 839                                               &setup);
 840
 841                 if (setup.need_copy)
 842                         kunmap_atomic(setup.bvec_data);
 843         }
 844         if (setup.segments)
 845                 kunmap_atomic(setup.segments);
 846
 847         /* Copy request(s) to the ring page. */
 848         *final_ring_req = *ring_req;
 849         rinfo->shadow[id].status = REQ_WAITING;
 850         if (unlikely(require_extra_req)) {
 851                 *final_extra_ring_req = *extra_ring_req;
 852                 rinfo->shadow[extra_id].status = REQ_WAITING;
 853         }
 854
 855         if (new_persistent_gnts)
 856                 gnttab_free_grant_references(setup.gref_head);
 857
 858         return 0;
 859 }
 860
 861 /*
 862  * Generate a Xen blkfront IO request from a blk layer request.  Reads
 863  * and writes are handled as expected.
 864  *
 865  * @req: a request struct
 866  */
 867 static int blkif_queue_request(struct request *req, struct blkfront_ring_info *rinfo)
 868 {
 869         if (unlikely(rinfo->dev_info->connected != BLKIF_STATE_CONNECTED))
 870                 return 1;
 871
 872         if (unlikely(req_op(req) == REQ_OP_DISCARD ||
 873                      req_op(req) == REQ_OP_SECURE_ERASE))
 874                 return blkif_queue_discard_req(req, rinfo);
 875         else
 876                 return blkif_queue_rw_req(req, rinfo);
 877 }
 878
 879 static inline void flush_requests(struct blkfront_ring_info *rinfo)
 880 {
 881         int notify;
 882
 883         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&rinfo->ring, notify);
 884
 885         if (notify)
 886                 notify_remote_via_irq(rinfo->irq);
 887 }
 888
 889 static inline bool blkif_request_flush_invalid(struct request *req,
 890                                                struct blkfront_info *info)
 891 {
 892         return (blk_rq_is_passthrough(req) ||
 893                 ((req_op(req) == REQ_OP_FLUSH) &&
 894                  !info->feature_flush) ||
 895                 ((req->cmd_flags & REQ_FUA) &&
 896                  !info->feature_fua));
 897 }
 898
 899 static blk_status_t blkif_queue_rq(struct blk_mq_hw_ctx *hctx,
 900                           const struct blk_mq_queue_data *qd)
 901 {
 902         unsigned long flags;
 903         int qid = hctx->queue_num;
 904         struct blkfront_info *info = hctx->queue->queuedata;
 905         struct blkfront_ring_info *rinfo = NULL;
 906
 907         rinfo = get_rinfo(info, qid);
 908         blk_mq_start_request(qd->rq);
 909         spin_lock_irqsave(&rinfo->ring_lock, flags);
 910         if (RING_FULL(&rinfo->ring))
 911                 goto out_busy;
 912
 913         if (blkif_request_flush_invalid(qd->rq, rinfo->dev_info))
 914                 goto out_err;
 915
 916         if (blkif_queue_request(qd->rq, rinfo))
 917                 goto out_busy;
 918
 919         flush_requests(rinfo);
 920         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 921         return BLK_STS_OK;
 922
 923 out_err:
 924         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 925         return BLK_STS_IOERR;
 926
 927 out_busy:
 928         blk_mq_stop_hw_queue(hctx);
 929         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
 930         return BLK_STS_DEV_RESOURCE;
 931 }
 932
 933 static void blkif_complete_rq(struct request *rq)
 934 {
 935         blk_mq_end_request(rq, blkif_req(rq)->error);
 936 }
 937
 938 static const struct blk_mq_ops blkfront_mq_ops = {
 939         .queue_rq = blkif_queue_rq,
 940         .complete = blkif_complete_rq,
 941 };
 942
 943 static void blkif_set_queue_limits(struct blkfront_info *info)
 944 {
 945         struct request_queue *rq = info->rq;
 946         struct gendisk *gd = info->gd;
 947         unsigned int segments = info->max_indirect_segments ? :
 948                                 BLKIF_MAX_SEGMENTS_PER_REQUEST;
 949
 950         blk_queue_flag_set(QUEUE_FLAG_VIRT, rq);
 951
 952         if (info->feature_discard) {
 953                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, rq);
 954                 blk_queue_max_discard_sectors(rq, get_capacity(gd));
 955                 rq->limits.discard_granularity = info->discard_granularity ?:
 956                                                  info->physical_sector_size;
 957                 rq->limits.discard_alignment = info->discard_alignment;
 958                 if (info->feature_secdiscard)
 959                         blk_queue_flag_set(QUEUE_FLAG_SECERASE, rq);
 960         }
 961
 962         /* Hard sector size and max sectors impersonate the equiv. hardware. */
 963         blk_queue_logical_block_size(rq, info->sector_size);
 964         blk_queue_physical_block_size(rq, info->physical_sector_size);
 965         blk_queue_max_hw_sectors(rq, (segments * XEN_PAGE_SIZE) / 512);
 966
 967         /* Each segment in a request is up to an aligned page in size. */
 968         blk_queue_segment_boundary(rq, PAGE_SIZE - 1);
 969         blk_queue_max_segment_size(rq, PAGE_SIZE);
 970
 971         /* Ensure a merged request will fit in a single I/O ring slot. */
 972         blk_queue_max_segments(rq, segments / GRANTS_PER_PSEG);
 973
 974         /* Make sure buffer addresses are sector-aligned. */
 975         blk_queue_dma_alignment(rq, 511);
 976 }
 977
 978 static const char *flush_info(struct blkfront_info *info)
 979 {
 980         if (info->feature_flush && info->feature_fua)
 981                 return "barrier: enabled;";
 982         else if (info->feature_flush)
 983                 return "flush diskcache: enabled;";
 984         else
 985                 return "barrier or flush: disabled;";
 986 }
 987
 988 static void xlvbd_flush(struct blkfront_info *info)
 989 {
 990         blk_queue_write_cache(info->rq, info->feature_flush ? true : false,
 991                               info->feature_fua ? true : false);
 992         pr_info("blkfront: %s: %s %s %s %s %s %s %s\n",
 993                 info->gd->disk_name, flush_info(info),
 994                 "persistent grants:", info->feature_persistent ?
 995                 "enabled;" : "disabled;", "indirect descriptors:",
 996                 info->max_indirect_segments ? "enabled;" : "disabled;",
 997                 "bounce buffer:", info->bounce ? "enabled" : "disabled;");
 998 }
 999
1000 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
1001 {
1002         int major;
1003         major = BLKIF_MAJOR(vdevice);
1004         *minor = BLKIF_MINOR(vdevice);
1005         switch (major) {
1006                 case XEN_IDE0_MAJOR:
1007                         *offset = (*minor / 64) + EMULATED_HD_DISK_NAME_OFFSET;
1008                         *minor = ((*minor / 64) * PARTS_PER_DISK) +
1009                                 EMULATED_HD_DISK_MINOR_OFFSET;
1010                         break;
1011                 case XEN_IDE1_MAJOR:
1012                         *offset = (*minor / 64) + 2 + EMULATED_HD_DISK_NAME_OFFSET;
1013                         *minor = (((*minor / 64) + 2) * PARTS_PER_DISK) +
1014                                 EMULATED_HD_DISK_MINOR_OFFSET;
1015                         break;
1016                 case XEN_SCSI_DISK0_MAJOR:
1017                         *offset = (*minor / PARTS_PER_DISK) + EMULATED_SD_DISK_NAME_OFFSET;
1018                         *minor = *minor + EMULATED_SD_DISK_MINOR_OFFSET;
1019                         break;
1020                 case XEN_SCSI_DISK1_MAJOR:
1021                 case XEN_SCSI_DISK2_MAJOR:
1022                 case XEN_SCSI_DISK3_MAJOR:
1023                 case XEN_SCSI_DISK4_MAJOR:
1024                 case XEN_SCSI_DISK5_MAJOR:
1025                 case XEN_SCSI_DISK6_MAJOR:
1026                 case XEN_SCSI_DISK7_MAJOR:
1027                         *offset = (*minor / PARTS_PER_DISK) +
1028                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16) +
1029                                 EMULATED_SD_DISK_NAME_OFFSET;
1030                         *minor = *minor +
1031                                 ((major - XEN_SCSI_DISK1_MAJOR + 1) * 16 * PARTS_PER_DISK) +
1032                                 EMULATED_SD_DISK_MINOR_OFFSET;
1033                         break;
1034                 case XEN_SCSI_DISK8_MAJOR:
1035                 case XEN_SCSI_DISK9_MAJOR:
1036                 case XEN_SCSI_DISK10_MAJOR:
1037                 case XEN_SCSI_DISK11_MAJOR:
1038                 case XEN_SCSI_DISK12_MAJOR:
1039                 case XEN_SCSI_DISK13_MAJOR:
1040                 case XEN_SCSI_DISK14_MAJOR:
1041                 case XEN_SCSI_DISK15_MAJOR:
1042                         *offset = (*minor / PARTS_PER_DISK) +
1043                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16) +
1044                                 EMULATED_SD_DISK_NAME_OFFSET;
1045                         *minor = *minor +
1046                                 ((major - XEN_SCSI_DISK8_MAJOR + 8) * 16 * PARTS_PER_DISK) +
1047                                 EMULATED_SD_DISK_MINOR_OFFSET;
1048                         break;
1049                 case XENVBD_MAJOR:
1050                         *offset = *minor / PARTS_PER_DISK;
1051                         break;
1052                 default:
1053                         printk(KERN_WARNING "blkfront: your disk configuration is "
1054                                         "incorrect, please use an xvd device instead\n");
1055                         return -ENODEV;
1056         }
1057         return 0;
1058 }
1059
1060 static char *encode_disk_name(char *ptr, unsigned int n)
1061 {
1062         if (n >= 26)
1063                 ptr = encode_disk_name(ptr, n / 26 - 1);
1064         *ptr = 'a' + n % 26;
1065         return ptr + 1;
1066 }
1067
1068 static int xlvbd_alloc_gendisk(blkif_sector_t capacity,
1069                                struct blkfront_info *info,
1070                                u16 vdisk_info, u16 sector_size,
1071                                unsigned int physical_sector_size)
1072 {
1073         struct gendisk *gd;
1074         int nr_minors = 1;
1075         int err;
1076         unsigned int offset;
1077         int minor;
1078         int nr_parts;
1079         char *ptr;
1080
1081         BUG_ON(info->gd != NULL);
1082         BUG_ON(info->rq != NULL);
1083
1084         if ((info->vdevice>>EXT_SHIFT) > 1) {
1085                 /* this is above the extended range; something is wrong */
1086                 printk(KERN_WARNING "blkfront: vdevice 0x%x is above the extended range; ignoring\n", info->vdevice);
1087                 return -ENODEV;
1088         }
1089
1090         if (!VDEV_IS_EXTENDED(info->vdevice)) {
1091                 err = xen_translate_vdev(info->vdevice, &minor, &offset);
1092                 if (err)
1093                         return err;
1094                 nr_parts = PARTS_PER_DISK;
1095         } else {
1096                 minor = BLKIF_MINOR_EXT(info->vdevice);
1097                 nr_parts = PARTS_PER_EXT_DISK;
1098                 offset = minor / nr_parts;
1099                 if (xen_hvm_domain() && offset < EMULATED_HD_DISK_NAME_OFFSET + 4)
1100                         printk(KERN_WARNING "blkfront: vdevice 0x%x might conflict with "
1101                                         "emulated IDE disks,\n\t choose an xvd device name"
1102                                         "from xvde on\n", info->vdevice);
1103         }
1104         if (minor >> MINORBITS) {
1105                 pr_warn("blkfront: %#x's minor (%#x) out of range; ignoring\n",
1106                         info->vdevice, minor);
1107                 return -ENODEV;
1108         }
1109
1110         if ((minor % nr_parts) == 0)
1111                 nr_minors = nr_parts;
1112
1113         err = xlbd_reserve_minors(minor, nr_minors);
1114         if (err)
1115                 return err;
1116
1117         memset(&info->tag_set, 0, sizeof(info->tag_set));
1118         info->tag_set.ops = &blkfront_mq_ops;
1119         info->tag_set.nr_hw_queues = info->nr_rings;
1120         if (HAS_EXTRA_REQ && info->max_indirect_segments == 0) {
1121                 /*
1122                  * When indirect descriptior is not supported, the I/O request
1123                  * will be split between multiple request in the ring.
1124                  * To avoid problems when sending the request, divide by
1125                  * 2 the depth of the queue.
1126                  */
1127                 info->tag_set.queue_depth =  BLK_RING_SIZE(info) / 2;
1128         } else
1129                 info->tag_set.queue_depth = BLK_RING_SIZE(info);
1130         info->tag_set.numa_node = NUMA_NO_NODE;
1131         info->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1132         info->tag_set.cmd_size = sizeof(struct blkif_req);
1133         info->tag_set.driver_data = info;
1134
1135         err = blk_mq_alloc_tag_set(&info->tag_set);
1136         if (err)
1137                 goto out_release_minors;
1138
1139         gd = blk_mq_alloc_disk(&info->tag_set, info);
1140         if (IS_ERR(gd)) {
1141                 err = PTR_ERR(gd);
1142                 goto out_free_tag_set;
1143         }
1144
1145         strcpy(gd->disk_name, DEV_NAME);
1146         ptr = encode_disk_name(gd->disk_name + sizeof(DEV_NAME) - 1, offset);
1147         BUG_ON(ptr >= gd->disk_name + DISK_NAME_LEN);
1148         if (nr_minors > 1)
1149                 *ptr = 0;
1150         else
1151                 snprintf(ptr, gd->disk_name + DISK_NAME_LEN - ptr,
1152                          "%d", minor & (nr_parts - 1));
1153
1154         gd->major = XENVBD_MAJOR;
1155         gd->first_minor = minor;
1156         gd->minors = nr_minors;
1157         gd->fops = &xlvbd_block_fops;
1158         gd->private_data = info;
1159         set_capacity(gd, capacity);
1160
1161         info->rq = gd->queue;
1162         info->gd = gd;
1163         info->sector_size = sector_size;
1164         info->physical_sector_size = physical_sector_size;
1165         blkif_set_queue_limits(info);
1166
1167         xlvbd_flush(info);
1168
1169         if (vdisk_info & VDISK_READONLY)
1170                 set_disk_ro(gd, 1);
1171
1172         if (vdisk_info & VDISK_REMOVABLE)
1173                 gd->flags |= GENHD_FL_REMOVABLE;
1174
1175         if (vdisk_info & VDISK_CDROM)
1176                 gd->flags |= GENHD_FL_CD;
1177
1178         return 0;
1179
1180 out_free_tag_set:
1181         blk_mq_free_tag_set(&info->tag_set);
1182 out_release_minors:
1183         xlbd_release_minors(minor, nr_minors);
1184         return err;
1185 }
1186
1187 /* Already hold rinfo->ring_lock. */
1188 static inline void kick_pending_request_queues_locked(struct blkfront_ring_info *rinfo)
1189 {
1190         if (!RING_FULL(&rinfo->ring))
1191                 blk_mq_start_stopped_hw_queues(rinfo->dev_info->rq, true);
1192 }
1193
1194 static void kick_pending_request_queues(struct blkfront_ring_info *rinfo)
1195 {
1196         unsigned long flags;
1197
1198         spin_lock_irqsave(&rinfo->ring_lock, flags);
1199         kick_pending_request_queues_locked(rinfo);
1200         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1201 }
1202
1203 static void blkif_restart_queue(struct work_struct *work)
1204 {
1205         struct blkfront_ring_info *rinfo = container_of(work, struct blkfront_ring_info, work);
1206
1207         if (rinfo->dev_info->connected == BLKIF_STATE_CONNECTED)
1208                 kick_pending_request_queues(rinfo);
1209 }
1210
1211 static void blkif_free_ring(struct blkfront_ring_info *rinfo)
1212 {
1213         struct grant *persistent_gnt, *n;
1214         struct blkfront_info *info = rinfo->dev_info;
1215         int i, j, segs;
1216
1217         /*
1218          * Remove indirect pages, this only happens when using indirect
1219          * descriptors but not persistent grants
1220          */
1221         if (!list_empty(&rinfo->indirect_pages)) {
1222                 struct page *indirect_page, *n;
1223
1224                 BUG_ON(info->bounce);
1225                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
1226                         list_del(&indirect_page->lru);
1227                         __free_page(indirect_page);
1228                 }
1229         }
1230
1231         /* Remove all persistent grants. */
1232         if (!list_empty(&rinfo->grants)) {
1233                 list_for_each_entry_safe(persistent_gnt, n,
1234                                          &rinfo->grants, node) {
1235                         list_del(&persistent_gnt->node);
1236                         if (persistent_gnt->gref != GRANT_INVALID_REF) {
1237                                 gnttab_end_foreign_access(persistent_gnt->gref,
1238                                                           0, 0UL);
1239                                 rinfo->persistent_gnts_c--;
1240                         }
1241                         if (info->bounce)
1242                                 __free_page(persistent_gnt->page);
1243                         kfree(persistent_gnt);
1244                 }
1245         }
1246         BUG_ON(rinfo->persistent_gnts_c != 0);
1247
1248         for (i = 0; i < BLK_RING_SIZE(info); i++) {
1249                 /*
1250                  * Clear persistent grants present in requests already
1251                  * on the shared ring
1252                  */
1253                 if (!rinfo->shadow[i].request)
1254                         goto free_shadow;
1255
1256                 segs = rinfo->shadow[i].req.operation == BLKIF_OP_INDIRECT ?
1257                        rinfo->shadow[i].req.u.indirect.nr_segments :
1258                        rinfo->shadow[i].req.u.rw.nr_segments;
1259                 for (j = 0; j < segs; j++) {
1260                         persistent_gnt = rinfo->shadow[i].grants_used[j];
1261                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1262                         if (info->bounce)
1263                                 __free_page(persistent_gnt->page);
1264                         kfree(persistent_gnt);
1265                 }
1266
1267                 if (rinfo->shadow[i].req.operation != BLKIF_OP_INDIRECT)
1268                         /*
1269                          * If this is not an indirect operation don't try to
1270                          * free indirect segments
1271                          */
1272                         goto free_shadow;
1273
1274                 for (j = 0; j < INDIRECT_GREFS(segs); j++) {
1275                         persistent_gnt = rinfo->shadow[i].indirect_grants[j];
1276                         gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
1277                         __free_page(persistent_gnt->page);
1278                         kfree(persistent_gnt);
1279                 }
1280
1281 free_shadow:
1282                 kvfree(rinfo->shadow[i].grants_used);
1283                 rinfo->shadow[i].grants_used = NULL;
1284                 kvfree(rinfo->shadow[i].indirect_grants);
1285                 rinfo->shadow[i].indirect_grants = NULL;
1286                 kvfree(rinfo->shadow[i].sg);
1287                 rinfo->shadow[i].sg = NULL;
1288         }
1289
1290         /* No more gnttab callback work. */
1291         gnttab_cancel_free_callback(&rinfo->callback);
1292
1293         /* Flush gnttab callback work. Must be done with no locks held. */
1294         flush_work(&rinfo->work);
1295
1296         /* Free resources associated with old device channel. */
1297         for (i = 0; i < info->nr_ring_pages; i++) {
1298                 if (rinfo->ring_ref[i] != GRANT_INVALID_REF) {
1299                         gnttab_end_foreign_access(rinfo->ring_ref[i], 0, 0);
1300                         rinfo->ring_ref[i] = GRANT_INVALID_REF;
1301                 }
1302         }
1303         free_pages_exact(rinfo->ring.sring,
1304                          info->nr_ring_pages * XEN_PAGE_SIZE);
1305         rinfo->ring.sring = NULL;
1306
1307         if (rinfo->irq)
1308                 unbind_from_irqhandler(rinfo->irq, rinfo);
1309         rinfo->evtchn = rinfo->irq = 0;
1310 }
1311
1312 static void blkif_free(struct blkfront_info *info, int suspend)
1313 {
1314         unsigned int i;
1315         struct blkfront_ring_info *rinfo;
1316
1317         /* Prevent new requests being issued until we fix things up. */
1318         info->connected = suspend ?
1319                 BLKIF_STATE_SUSPENDED : BLKIF_STATE_DISCONNECTED;
1320         /* No more blkif_request(). */
1321         if (info->rq)
1322                 blk_mq_stop_hw_queues(info->rq);
1323
1324         for_each_rinfo(info, rinfo, i)
1325                 blkif_free_ring(rinfo);
1326
1327         kvfree(info->rinfo);
1328         info->rinfo = NULL;
1329         info->nr_rings = 0;
1330 }
1331
1332 struct copy_from_grant {
1333         const struct blk_shadow *s;
1334         unsigned int grant_idx;
1335         unsigned int bvec_offset;
1336         char *bvec_data;
1337 };
1338
1339 static void blkif_copy_from_grant(unsigned long gfn, unsigned int offset,
1340                                   unsigned int len, void *data)
1341 {
1342         struct copy_from_grant *info = data;
1343         char *shared_data;
1344         /* Convenient aliases */
1345         const struct blk_shadow *s = info->s;
1346
1347         shared_data = kmap_atomic(s->grants_used[info->grant_idx]->page);
1348
1349         memcpy(info->bvec_data + info->bvec_offset,
1350                shared_data + offset, len);
1351
1352         info->bvec_offset += len;
1353         info->grant_idx++;
1354
1355         kunmap_atomic(shared_data);
1356 }
1357
1358 static enum blk_req_status blkif_rsp_to_req_status(int rsp)
1359 {
1360         switch (rsp)
1361         {
1362         case BLKIF_RSP_OKAY:
1363                 return REQ_DONE;
1364         case BLKIF_RSP_EOPNOTSUPP:
1365                 return REQ_EOPNOTSUPP;
1366         case BLKIF_RSP_ERROR:
1367         default:
1368                 return REQ_ERROR;
1369         }
1370 }
1371
1372 /*
1373  * Get the final status of the block request based on two ring response
1374  */
1375 static int blkif_get_final_status(enum blk_req_status s1,
1376                                   enum blk_req_status s2)
1377 {
1378         BUG_ON(s1 < REQ_DONE);
1379         BUG_ON(s2 < REQ_DONE);
1380
1381         if (s1 == REQ_ERROR || s2 == REQ_ERROR)
1382                 return BLKIF_RSP_ERROR;
1383         else if (s1 == REQ_EOPNOTSUPP || s2 == REQ_EOPNOTSUPP)
1384                 return BLKIF_RSP_EOPNOTSUPP;
1385         return BLKIF_RSP_OKAY;
1386 }
1387
1388 /*
1389  * Return values:
1390  *  1 response processed.
1391  *  0 missing further responses.
1392  * -1 error while processing.
1393  */
1394 static int blkif_completion(unsigned long *id,
1395                             struct blkfront_ring_info *rinfo,
1396                             struct blkif_response *bret)
1397 {
1398         int i = 0;
1399         struct scatterlist *sg;
1400         int num_sg, num_grant;
1401         struct blkfront_info *info = rinfo->dev_info;
1402         struct blk_shadow *s = &rinfo->shadow[*id];
1403         struct copy_from_grant data = {
1404                 .grant_idx = 0,
1405         };
1406
1407         num_grant = s->req.operation == BLKIF_OP_INDIRECT ?
1408                 s->req.u.indirect.nr_segments : s->req.u.rw.nr_segments;
1409
1410         /* The I/O request may be split in two. */
1411         if (unlikely(s->associated_id != NO_ASSOCIATED_ID)) {
1412                 struct blk_shadow *s2 = &rinfo->shadow[s->associated_id];
1413
1414                 /* Keep the status of the current response in shadow. */
1415                 s->status = blkif_rsp_to_req_status(bret->status);
1416
1417                 /* Wait the second response if not yet here. */
1418                 if (s2->status < REQ_DONE)
1419                         return 0;
1420
1421                 bret->status = blkif_get_final_status(s->status,
1422                                                       s2->status);
1423
1424                 /*
1425                  * All the grants is stored in the first shadow in order
1426                  * to make the completion code simpler.
1427                  */
1428                 num_grant += s2->req.u.rw.nr_segments;
1429
1430                 /*
1431                  * The two responses may not come in order. Only the
1432                  * first request will store the scatter-gather list.
1433                  */
1434                 if (s2->num_sg != 0) {
1435                         /* Update "id" with the ID of the first response. */
1436                         *id = s->associated_id;
1437                         s = s2;
1438                 }
1439
1440                 /*
1441                  * We don't need anymore the second request, so recycling
1442                  * it now.
1443                  */
1444                 if (add_id_to_freelist(rinfo, s->associated_id))
1445                         WARN(1, "%s: can't recycle the second part (id = %ld) of the request\n",
1446                              info->gd->disk_name, s->associated_id);
1447         }
1448
1449         data.s = s;
1450         num_sg = s->num_sg;
1451
1452         if (bret->operation == BLKIF_OP_READ && info->bounce) {
1453                 for_each_sg(s->sg, sg, num_sg, i) {
1454                         BUG_ON(sg->offset + sg->length > PAGE_SIZE);
1455
1456                         data.bvec_offset = sg->offset;
1457                         data.bvec_data = kmap_atomic(sg_page(sg));
1458
1459                         gnttab_foreach_grant_in_range(sg_page(sg),
1460                                                       sg->offset,
1461                                                       sg->length,
1462                                                       blkif_copy_from_grant,
1463                                                       &data);
1464
1465                         kunmap_atomic(data.bvec_data);
1466                 }
1467         }
1468         /* Add the persistent grant into the list of free grants */
1469         for (i = 0; i < num_grant; i++) {
1470                 if (!gnttab_try_end_foreign_access(s->grants_used[i]->gref)) {
1471                         /*
1472                          * If the grant is still mapped by the backend (the
1473                          * backend has chosen to make this grant persistent)
1474                          * we add it at the head of the list, so it will be
1475                          * reused first.
1476                          */
1477                         if (!info->feature_persistent) {
1478                                 pr_alert("backed has not unmapped grant: %u\n",
1479                                          s->grants_used[i]->gref);
1480                                 return -1;
1481                         }
1482                         list_add(&s->grants_used[i]->node, &rinfo->grants);
1483                         rinfo->persistent_gnts_c++;
1484                 } else {
1485                         /*
1486                          * If the grant is not mapped by the backend we add it
1487                          * to the tail of the list, so it will not be picked
1488                          * again unless we run out of persistent grants.
1489                          */
1490                         s->grants_used[i]->gref = GRANT_INVALID_REF;
1491                         list_add_tail(&s->grants_used[i]->node, &rinfo->grants);
1492                 }
1493         }
1494         if (s->req.operation == BLKIF_OP_INDIRECT) {
1495                 for (i = 0; i < INDIRECT_GREFS(num_grant); i++) {
1496                         if (!gnttab_try_end_foreign_access(s->indirect_grants[i]->gref)) {
1497                                 if (!info->feature_persistent) {
1498                                         pr_alert("backed has not unmapped grant: %u\n",
1499                                                  s->indirect_grants[i]->gref);
1500                                         return -1;
1501                                 }
1502                                 list_add(&s->indirect_grants[i]->node, &rinfo->grants);
1503                                 rinfo->persistent_gnts_c++;
1504                         } else {
1505                                 struct page *indirect_page;
1506
1507                                 /*
1508                                  * Add the used indirect page back to the list of
1509                                  * available pages for indirect grefs.
1510                                  */
1511                                 if (!info->bounce) {
1512                                         indirect_page = s->indirect_grants[i]->page;
1513                                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
1514                                 }
1515                                 s->indirect_grants[i]->gref = GRANT_INVALID_REF;
1516                                 list_add_tail(&s->indirect_grants[i]->node, &rinfo->grants);
1517                         }
1518                 }
1519         }
1520
1521         return 1;
1522 }
1523
1524 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
1525 {
1526         struct request *req;
1527         struct blkif_response bret;
1528         RING_IDX i, rp;
1529         unsigned long flags;
1530         struct blkfront_ring_info *rinfo = (struct blkfront_ring_info *)dev_id;
1531         struct blkfront_info *info = rinfo->dev_info;
1532         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1533
1534         if (unlikely(info->connected != BLKIF_STATE_CONNECTED)) {
1535                 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
1536                 return IRQ_HANDLED;
1537         }
1538
1539         spin_lock_irqsave(&rinfo->ring_lock, flags);
1540  again:
1541         rp = READ_ONCE(rinfo->ring.sring->rsp_prod);
1542         virt_rmb(); /* Ensure we see queued responses up to 'rp'. */
1543         if (RING_RESPONSE_PROD_OVERFLOW(&rinfo->ring, rp)) {
1544                 pr_alert("%s: illegal number of responses %u\n",
1545                          info->gd->disk_name, rp - rinfo->ring.rsp_cons);
1546                 goto err;
1547         }
1548
1549         for (i = rinfo->ring.rsp_cons; i != rp; i++) {
1550                 unsigned long id;
1551                 unsigned int op;
1552
1553                 eoiflag = 0;
1554
1555                 RING_COPY_RESPONSE(&rinfo->ring, i, &bret);
1556                 id = bret.id;
1557
1558                 /*
1559                  * The backend has messed up and given us an id that we would
1560                  * never have given to it (we stamp it up to BLK_RING_SIZE -
1561                  * look in get_id_from_freelist.
1562                  */
1563                 if (id >= BLK_RING_SIZE(info)) {
1564                         pr_alert("%s: response has incorrect id (%ld)\n",
1565                                  info->gd->disk_name, id);
1566                         goto err;
1567                 }
1568                 if (rinfo->shadow[id].status != REQ_WAITING) {
1569                         pr_alert("%s: response references no pending request\n",
1570                                  info->gd->disk_name);
1571                         goto err;
1572                 }
1573
1574                 rinfo->shadow[id].status = REQ_PROCESSING;
1575                 req  = rinfo->shadow[id].request;
1576
1577                 op = rinfo->shadow[id].req.operation;
1578                 if (op == BLKIF_OP_INDIRECT)
1579                         op = rinfo->shadow[id].req.u.indirect.indirect_op;
1580                 if (bret.operation != op) {
1581                         pr_alert("%s: response has wrong operation (%u instead of %u)\n",
1582                                  info->gd->disk_name, bret.operation, op);
1583                         goto err;
1584                 }
1585
1586                 if (bret.operation != BLKIF_OP_DISCARD) {
1587                         int ret;
1588
1589                         /*
1590                          * We may need to wait for an extra response if the
1591                          * I/O request is split in 2
1592                          */
1593                         ret = blkif_completion(&id, rinfo, &bret);
1594                         if (!ret)
1595                                 continue;
1596                         if (unlikely(ret < 0))
1597                                 goto err;
1598                 }
1599
1600                 if (add_id_to_freelist(rinfo, id)) {
1601                         WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
1602                              info->gd->disk_name, op_name(bret.operation), id);
1603                         continue;
1604                 }
1605
1606                 if (bret.status == BLKIF_RSP_OKAY)
1607                         blkif_req(req)->error = BLK_STS_OK;
1608                 else
1609                         blkif_req(req)->error = BLK_STS_IOERR;
1610
1611                 switch (bret.operation) {
1612                 case BLKIF_OP_DISCARD:
1613                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1614                                 struct request_queue *rq = info->rq;
1615
1616                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1617                                            info->gd->disk_name, op_name(bret.operation));
1618                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1619                                 info->feature_discard = 0;
1620                                 info->feature_secdiscard = 0;
1621                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, rq);
1622                                 blk_queue_flag_clear(QUEUE_FLAG_SECERASE, rq);
1623                         }
1624                         break;
1625                 case BLKIF_OP_FLUSH_DISKCACHE:
1626                 case BLKIF_OP_WRITE_BARRIER:
1627                         if (unlikely(bret.status == BLKIF_RSP_EOPNOTSUPP)) {
1628                                 pr_warn_ratelimited("blkfront: %s: %s op failed\n",
1629                                        info->gd->disk_name, op_name(bret.operation));
1630                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1631                         }
1632                         if (unlikely(bret.status == BLKIF_RSP_ERROR &&
1633                                      rinfo->shadow[id].req.u.rw.nr_segments == 0)) {
1634                                 pr_warn_ratelimited("blkfront: %s: empty %s op failed\n",
1635                                        info->gd->disk_name, op_name(bret.operation));
1636                                 blkif_req(req)->error = BLK_STS_NOTSUPP;
1637                         }
1638                         if (unlikely(blkif_req(req)->error)) {
1639                                 if (blkif_req(req)->error == BLK_STS_NOTSUPP)
1640                                         blkif_req(req)->error = BLK_STS_OK;
1641                                 info->feature_fua = 0;
1642                                 info->feature_flush = 0;
1643                                 xlvbd_flush(info);
1644                         }
1645                         fallthrough;
1646                 case BLKIF_OP_READ:
1647                 case BLKIF_OP_WRITE:
1648                         if (unlikely(bret.status != BLKIF_RSP_OKAY))
1649                                 dev_dbg_ratelimited(&info->xbdev->dev,
1650                                         "Bad return from blkdev data request: %#x\n",
1651                                         bret.status);
1652
1653                         break;
1654                 default:
1655                         BUG();
1656                 }
1657
1658                 if (likely(!blk_should_fake_timeout(req->q)))
1659                         blk_mq_complete_request(req);
1660         }
1661
1662         rinfo->ring.rsp_cons = i;
1663
1664         if (i != rinfo->ring.req_prod_pvt) {
1665                 int more_to_do;
1666                 RING_FINAL_CHECK_FOR_RESPONSES(&rinfo->ring, more_to_do);
1667                 if (more_to_do)
1668                         goto again;
1669         } else
1670                 rinfo->ring.sring->rsp_event = i + 1;
1671
1672         kick_pending_request_queues_locked(rinfo);
1673
1674         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1675
1676         xen_irq_lateeoi(irq, eoiflag);
1677
1678         return IRQ_HANDLED;
1679
1680  err:
1681         info->connected = BLKIF_STATE_ERROR;
1682
1683         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
1684
1685         /* No EOI in order to avoid further interrupts. */
1686
1687         pr_alert("%s disabled for further use\n", info->gd->disk_name);
1688         return IRQ_HANDLED;
1689 }
1690
1691
1692 static int setup_blkring(struct xenbus_device *dev,
1693                          struct blkfront_ring_info *rinfo)
1694 {
1695         struct blkif_sring *sring;
1696         int err, i;
1697         struct blkfront_info *info = rinfo->dev_info;
1698         unsigned long ring_size = info->nr_ring_pages * XEN_PAGE_SIZE;
1699         grant_ref_t gref[XENBUS_MAX_RING_GRANTS];
1700
1701         for (i = 0; i < info->nr_ring_pages; i++)
1702                 rinfo->ring_ref[i] = GRANT_INVALID_REF;
1703
1704         sring = alloc_pages_exact(ring_size, GFP_NOIO | __GFP_ZERO);
1705         if (!sring) {
1706                 xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
1707                 return -ENOMEM;
1708         }
1709         SHARED_RING_INIT(sring);
1710         FRONT_RING_INIT(&rinfo->ring, sring, ring_size);
1711
1712         err = xenbus_grant_ring(dev, rinfo->ring.sring, info->nr_ring_pages, gref);
1713         if (err < 0) {
1714                 free_pages_exact(sring, ring_size);
1715                 rinfo->ring.sring = NULL;
1716                 goto fail;
1717         }
1718         for (i = 0; i < info->nr_ring_pages; i++)
1719                 rinfo->ring_ref[i] = gref[i];
1720
1721         err = xenbus_alloc_evtchn(dev, &rinfo->evtchn);
1722         if (err)
1723                 goto fail;
1724
1725         err = bind_evtchn_to_irqhandler_lateeoi(rinfo->evtchn, blkif_interrupt,
1726                                                 0, "blkif", rinfo);
1727         if (err <= 0) {
1728                 xenbus_dev_fatal(dev, err,
1729                                  "bind_evtchn_to_irqhandler failed");
1730                 goto fail;
1731         }
1732         rinfo->irq = err;
1733
1734         return 0;
1735 fail:
1736         blkif_free(info, 0);
1737         return err;
1738 }
1739
1740 /*
1741  * Write out per-ring/queue nodes including ring-ref and event-channel, and each
1742  * ring buffer may have multi pages depending on ->nr_ring_pages.
1743  */
1744 static int write_per_ring_nodes(struct xenbus_transaction xbt,
1745                                 struct blkfront_ring_info *rinfo, const char *dir)
1746 {
1747         int err;
1748         unsigned int i;
1749         const char *message = NULL;
1750         struct blkfront_info *info = rinfo->dev_info;
1751
1752         if (info->nr_ring_pages == 1) {
1753                 err = xenbus_printf(xbt, dir, "ring-ref", "%u", rinfo->ring_ref[0]);
1754                 if (err) {
1755                         message = "writing ring-ref";
1756                         goto abort_transaction;
1757                 }
1758         } else {
1759                 for (i = 0; i < info->nr_ring_pages; i++) {
1760                         char ring_ref_name[RINGREF_NAME_LEN];
1761
1762                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1763                         err = xenbus_printf(xbt, dir, ring_ref_name,
1764                                             "%u", rinfo->ring_ref[i]);
1765                         if (err) {
1766                                 message = "writing ring-ref";
1767                                 goto abort_transaction;
1768                         }
1769                 }
1770         }
1771
1772         err = xenbus_printf(xbt, dir, "event-channel", "%u", rinfo->evtchn);
1773         if (err) {
1774                 message = "writing event-channel";
1775                 goto abort_transaction;
1776         }
1777
1778         return 0;
1779
1780 abort_transaction:
1781         xenbus_transaction_end(xbt, 1);
1782         if (message)
1783                 xenbus_dev_fatal(info->xbdev, err, "%s", message);
1784
1785         return err;
1786 }
1787
1788 /* Enable the persistent grants feature. */
1789 static bool feature_persistent = true;
1790 module_param(feature_persistent, bool, 0644);
1791 MODULE_PARM_DESC(feature_persistent,
1792                 "Enables the persistent grants feature");
1793
1794 /* Common code used when first setting up, and when resuming. */
1795 static int talk_to_blkback(struct xenbus_device *dev,
1796                            struct blkfront_info *info)
1797 {
1798         const char *message = NULL;
1799         struct xenbus_transaction xbt;
1800         int err;
1801         unsigned int i, max_page_order;
1802         unsigned int ring_page_order;
1803         struct blkfront_ring_info *rinfo;
1804
1805         if (!info)
1806                 return -ENODEV;
1807
1808         /* Check if backend is trusted. */
1809         info->bounce = !xen_blkif_trusted ||
1810                        !xenbus_read_unsigned(dev->nodename, "trusted", 1);
1811
1812         max_page_order = xenbus_read_unsigned(info->xbdev->otherend,
1813                                               "max-ring-page-order", 0);
1814         ring_page_order = min(xen_blkif_max_ring_order, max_page_order);
1815         info->nr_ring_pages = 1 << ring_page_order;
1816
1817         err = negotiate_mq(info);
1818         if (err)
1819                 goto destroy_blkring;
1820
1821         for_each_rinfo(info, rinfo, i) {
1822                 /* Create shared ring, alloc event channel. */
1823                 err = setup_blkring(dev, rinfo);
1824                 if (err)
1825                         goto destroy_blkring;
1826         }
1827
1828 again:
1829         err = xenbus_transaction_start(&xbt);
1830         if (err) {
1831                 xenbus_dev_fatal(dev, err, "starting transaction");
1832                 goto destroy_blkring;
1833         }
1834
1835         if (info->nr_ring_pages > 1) {
1836                 err = xenbus_printf(xbt, dev->nodename, "ring-page-order", "%u",
1837                                     ring_page_order);
1838                 if (err) {
1839                         message = "writing ring-page-order";
1840                         goto abort_transaction;
1841                 }
1842         }
1843
1844         /* We already got the number of queues/rings in _probe */
1845         if (info->nr_rings == 1) {
1846                 err = write_per_ring_nodes(xbt, info->rinfo, dev->nodename);
1847                 if (err)
1848                         goto destroy_blkring;
1849         } else {
1850                 char *path;
1851                 size_t pathsize;
1852
1853                 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", "%u",
1854                                     info->nr_rings);
1855                 if (err) {
1856                         message = "writing multi-queue-num-queues";
1857                         goto abort_transaction;
1858                 }
1859
1860                 pathsize = strlen(dev->nodename) + QUEUE_NAME_LEN;
1861                 path = kmalloc(pathsize, GFP_KERNEL);
1862                 if (!path) {
1863                         err = -ENOMEM;
1864                         message = "ENOMEM while writing ring references";
1865                         goto abort_transaction;
1866                 }
1867
1868                 for_each_rinfo(info, rinfo, i) {
1869                         memset(path, 0, pathsize);
1870                         snprintf(path, pathsize, "%s/queue-%u", dev->nodename, i);
1871                         err = write_per_ring_nodes(xbt, rinfo, path);
1872                         if (err) {
1873                                 kfree(path);
1874                                 goto destroy_blkring;
1875                         }
1876                 }
1877                 kfree(path);
1878         }
1879         err = xenbus_printf(xbt, dev->nodename, "protocol", "%s",
1880                             XEN_IO_PROTO_ABI_NATIVE);
1881         if (err) {
1882                 message = "writing protocol";
1883                 goto abort_transaction;
1884         }
1885         info->feature_persistent_parm = feature_persistent;
1886         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
1887                         info->feature_persistent_parm);
1888         if (err)
1889                 dev_warn(&dev->dev,
1890                          "writing persistent grants feature to xenbus");
1891
1892         err = xenbus_transaction_end(xbt, 0);
1893         if (err) {
1894                 if (err == -EAGAIN)
1895                         goto again;
1896                 xenbus_dev_fatal(dev, err, "completing transaction");
1897                 goto destroy_blkring;
1898         }
1899
1900         for_each_rinfo(info, rinfo, i) {
1901                 unsigned int j;
1902
1903                 for (j = 0; j < BLK_RING_SIZE(info); j++)
1904                         rinfo->shadow[j].req.u.rw.id = j + 1;
1905                 rinfo->shadow[BLK_RING_SIZE(info)-1].req.u.rw.id = 0x0fffffff;
1906         }
1907         xenbus_switch_state(dev, XenbusStateInitialised);
1908
1909         return 0;
1910
1911  abort_transaction:
1912         xenbus_transaction_end(xbt, 1);
1913         if (message)
1914                 xenbus_dev_fatal(dev, err, "%s", message);
1915  destroy_blkring:
1916         blkif_free(info, 0);
1917         return err;
1918 }
1919
1920 static int negotiate_mq(struct blkfront_info *info)
1921 {
1922         unsigned int backend_max_queues;
1923         unsigned int i;
1924         struct blkfront_ring_info *rinfo;
1925
1926         BUG_ON(info->nr_rings);
1927
1928         /* Check if backend supports multiple queues. */
1929         backend_max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1930                                                   "multi-queue-max-queues", 1);
1931         info->nr_rings = min(backend_max_queues, xen_blkif_max_queues);
1932         /* We need at least one ring. */
1933         if (!info->nr_rings)
1934                 info->nr_rings = 1;
1935
1936         info->rinfo_size = struct_size(info->rinfo, shadow,
1937                                        BLK_RING_SIZE(info));
1938         info->rinfo = kvcalloc(info->nr_rings, info->rinfo_size, GFP_KERNEL);
1939         if (!info->rinfo) {
1940                 xenbus_dev_fatal(info->xbdev, -ENOMEM, "allocating ring_info structure");
1941                 info->nr_rings = 0;
1942                 return -ENOMEM;
1943         }
1944
1945         for_each_rinfo(info, rinfo, i) {
1946                 INIT_LIST_HEAD(&rinfo->indirect_pages);
1947                 INIT_LIST_HEAD(&rinfo->grants);
1948                 rinfo->dev_info = info;
1949                 INIT_WORK(&rinfo->work, blkif_restart_queue);
1950                 spin_lock_init(&rinfo->ring_lock);
1951         }
1952         return 0;
1953 }
1954
1955 /*
1956  * Entry point to this code when a new device is created.  Allocate the basic
1957  * structures and the ring buffer for communication with the backend, and
1958  * inform the backend of the appropriate details for those.  Switch to
1959  * Initialised state.
1960  */
1961 static int blkfront_probe(struct xenbus_device *dev,
1962                           const struct xenbus_device_id *id)
1963 {
1964         int err, vdevice;
1965         struct blkfront_info *info;
1966
1967         /* FIXME: Use dynamic device id if this is not set. */
1968         err = xenbus_scanf(XBT_NIL, dev->nodename,
1969                            "virtual-device", "%i", &vdevice);
1970         if (err != 1) {
1971                 /* go looking in the extended area instead */
1972                 err = xenbus_scanf(XBT_NIL, dev->nodename, "virtual-device-ext",
1973                                    "%i", &vdevice);
1974                 if (err != 1) {
1975                         xenbus_dev_fatal(dev, err, "reading virtual-device");
1976                         return err;
1977                 }
1978         }
1979
1980         if (xen_hvm_domain()) {
1981                 char *type;
1982                 int len;
1983                 /* no unplug has been done: do not hook devices != xen vbds */
1984                 if (xen_has_pv_and_legacy_disk_devices()) {
1985                         int major;
1986
1987                         if (!VDEV_IS_EXTENDED(vdevice))
1988                                 major = BLKIF_MAJOR(vdevice);
1989                         else
1990                                 major = XENVBD_MAJOR;
1991
1992                         if (major != XENVBD_MAJOR) {
1993                                 printk(KERN_INFO
1994                                                 "%s: HVM does not support vbd %d as xen block device\n",
1995                                                 __func__, vdevice);
1996                                 return -ENODEV;
1997                         }
1998                 }
1999                 /* do not create a PV cdrom device if we are an HVM guest */
2000                 type = xenbus_read(XBT_NIL, dev->nodename, "device-type", &len);
2001                 if (IS_ERR(type))
2002                         return -ENODEV;
2003                 if (strncmp(type, "cdrom", 5) == 0) {
2004                         kfree(type);
2005                         return -ENODEV;
2006                 }
2007                 kfree(type);
2008         }
2009         info = kzalloc(sizeof(*info), GFP_KERNEL);
2010         if (!info) {
2011                 xenbus_dev_fatal(dev, -ENOMEM, "allocating info structure");
2012                 return -ENOMEM;
2013         }
2014
2015         info->xbdev = dev;
2016
2017         mutex_init(&info->mutex);
2018         info->vdevice = vdevice;
2019         info->connected = BLKIF_STATE_DISCONNECTED;
2020
2021         /* Front end dir is a number, which is used as the id. */
2022         info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
2023         dev_set_drvdata(&dev->dev, info);
2024
2025         mutex_lock(&blkfront_mutex);
2026         list_add(&info->info_list, &info_list);
2027         mutex_unlock(&blkfront_mutex);
2028
2029         return 0;
2030 }
2031
2032 static int blkif_recover(struct blkfront_info *info)
2033 {
2034         unsigned int r_index;
2035         struct request *req, *n;
2036         int rc;
2037         struct bio *bio;
2038         unsigned int segs;
2039         struct blkfront_ring_info *rinfo;
2040
2041         blkfront_gather_backend_features(info);
2042         /* Reset limits changed by blk_mq_update_nr_hw_queues(). */
2043         blkif_set_queue_limits(info);
2044         segs = info->max_indirect_segments ? : BLKIF_MAX_SEGMENTS_PER_REQUEST;
2045         blk_queue_max_segments(info->rq, segs / GRANTS_PER_PSEG);
2046
2047         for_each_rinfo(info, rinfo, r_index) {
2048                 rc = blkfront_setup_indirect(rinfo);
2049                 if (rc)
2050                         return rc;
2051         }
2052         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2053
2054         /* Now safe for us to use the shared ring */
2055         info->connected = BLKIF_STATE_CONNECTED;
2056
2057         for_each_rinfo(info, rinfo, r_index) {
2058                 /* Kick any other new requests queued since we resumed */
2059                 kick_pending_request_queues(rinfo);
2060         }
2061
2062         list_for_each_entry_safe(req, n, &info->requests, queuelist) {
2063                 /* Requeue pending requests (flush or discard) */
2064                 list_del_init(&req->queuelist);
2065                 BUG_ON(req->nr_phys_segments > segs);
2066                 blk_mq_requeue_request(req, false);
2067         }
2068         blk_mq_start_stopped_hw_queues(info->rq, true);
2069         blk_mq_kick_requeue_list(info->rq);
2070
2071         while ((bio = bio_list_pop(&info->bio_list)) != NULL) {
2072                 /* Traverse the list of pending bios and re-queue them */
2073                 submit_bio(bio);
2074         }
2075
2076         return 0;
2077 }
2078
2079 /*
2080  * We are reconnecting to the backend, due to a suspend/resume, or a backend
2081  * driver restart.  We tear down our blkif structure and recreate it, but
2082  * leave the device-layer structures intact so that this is transparent to the
2083  * rest of the kernel.
2084  */
2085 static int blkfront_resume(struct xenbus_device *dev)
2086 {
2087         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2088         int err = 0;
2089         unsigned int i, j;
2090         struct blkfront_ring_info *rinfo;
2091
2092         dev_dbg(&dev->dev, "blkfront_resume: %s\n", dev->nodename);
2093
2094         bio_list_init(&info->bio_list);
2095         INIT_LIST_HEAD(&info->requests);
2096         for_each_rinfo(info, rinfo, i) {
2097                 struct bio_list merge_bio;
2098                 struct blk_shadow *shadow = rinfo->shadow;
2099
2100                 for (j = 0; j < BLK_RING_SIZE(info); j++) {
2101                         /* Not in use? */
2102                         if (!shadow[j].request)
2103                                 continue;
2104
2105                         /*
2106                          * Get the bios in the request so we can re-queue them.
2107                          */
2108                         if (req_op(shadow[j].request) == REQ_OP_FLUSH ||
2109                             req_op(shadow[j].request) == REQ_OP_DISCARD ||
2110                             req_op(shadow[j].request) == REQ_OP_SECURE_ERASE ||
2111                             shadow[j].request->cmd_flags & REQ_FUA) {
2112                                 /*
2113                                  * Flush operations don't contain bios, so
2114                                  * we need to requeue the whole request
2115                                  *
2116                                  * XXX: but this doesn't make any sense for a
2117                                  * write with the FUA flag set..
2118                                  */
2119                                 list_add(&shadow[j].request->queuelist, &info->requests);
2120                                 continue;
2121                         }
2122                         merge_bio.head = shadow[j].request->bio;
2123                         merge_bio.tail = shadow[j].request->biotail;
2124                         bio_list_merge(&info->bio_list, &merge_bio);
2125                         shadow[j].request->bio = NULL;
2126                         blk_mq_end_request(shadow[j].request, BLK_STS_OK);
2127                 }
2128         }
2129
2130         blkif_free(info, info->connected == BLKIF_STATE_CONNECTED);
2131
2132         err = talk_to_blkback(dev, info);
2133         if (!err)
2134                 blk_mq_update_nr_hw_queues(&info->tag_set, info->nr_rings);
2135
2136         /*
2137          * We have to wait for the backend to switch to
2138          * connected state, since we want to read which
2139          * features it supports.
2140          */
2141
2142         return err;
2143 }
2144
2145 static void blkfront_closing(struct blkfront_info *info)
2146 {
2147         struct xenbus_device *xbdev = info->xbdev;
2148         struct blkfront_ring_info *rinfo;
2149         unsigned int i;
2150
2151         if (xbdev->state == XenbusStateClosing)
2152                 return;
2153
2154         /* No more blkif_request(). */
2155         if (info->rq && info->gd) {
2156                 blk_mq_stop_hw_queues(info->rq);
2157                 blk_mark_disk_dead(info->gd);
2158                 set_capacity(info->gd, 0);
2159         }
2160
2161         for_each_rinfo(info, rinfo, i) {
2162                 /* No more gnttab callback work. */
2163                 gnttab_cancel_free_callback(&rinfo->callback);
2164
2165                 /* Flush gnttab callback work. Must be done with no locks held. */
2166                 flush_work(&rinfo->work);
2167         }
2168
2169         xenbus_frontend_closed(xbdev);
2170 }
2171
2172 static void blkfront_setup_discard(struct blkfront_info *info)
2173 {
2174         info->feature_discard = 1;
2175         info->discard_granularity = xenbus_read_unsigned(info->xbdev->otherend,
2176                                                          "discard-granularity",
2177                                                          0);
2178         info->discard_alignment = xenbus_read_unsigned(info->xbdev->otherend,
2179                                                        "discard-alignment", 0);
2180         info->feature_secdiscard =
2181                 !!xenbus_read_unsigned(info->xbdev->otherend, "discard-secure",
2182                                        0);
2183 }
2184
2185 static int blkfront_setup_indirect(struct blkfront_ring_info *rinfo)
2186 {
2187         unsigned int psegs, grants, memflags;
2188         int err, i;
2189         struct blkfront_info *info = rinfo->dev_info;
2190
2191         memflags = memalloc_noio_save();
2192
2193         if (info->max_indirect_segments == 0) {
2194                 if (!HAS_EXTRA_REQ)
2195                         grants = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2196                 else {
2197                         /*
2198                          * When an extra req is required, the maximum
2199                          * grants supported is related to the size of the
2200                          * Linux block segment.
2201                          */
2202                         grants = GRANTS_PER_PSEG;
2203                 }
2204         }
2205         else
2206                 grants = info->max_indirect_segments;
2207         psegs = DIV_ROUND_UP(grants, GRANTS_PER_PSEG);
2208
2209         err = fill_grant_buffer(rinfo,
2210                                 (grants + INDIRECT_GREFS(grants)) * BLK_RING_SIZE(info));
2211         if (err)
2212                 goto out_of_memory;
2213
2214         if (!info->bounce && info->max_indirect_segments) {
2215                 /*
2216                  * We are using indirect descriptors but don't have a bounce
2217                  * buffer, we need to allocate a set of pages that can be
2218                  * used for mapping indirect grefs
2219                  */
2220                 int num = INDIRECT_GREFS(grants) * BLK_RING_SIZE(info);
2221
2222                 BUG_ON(!list_empty(&rinfo->indirect_pages));
2223                 for (i = 0; i < num; i++) {
2224                         struct page *indirect_page = alloc_page(GFP_KERNEL |
2225                                                                 __GFP_ZERO);
2226                         if (!indirect_page)
2227                                 goto out_of_memory;
2228                         list_add(&indirect_page->lru, &rinfo->indirect_pages);
2229                 }
2230         }
2231
2232         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2233                 rinfo->shadow[i].grants_used =
2234                         kvcalloc(grants,
2235                                  sizeof(rinfo->shadow[i].grants_used[0]),
2236                                  GFP_KERNEL);
2237                 rinfo->shadow[i].sg = kvcalloc(psegs,
2238                                                sizeof(rinfo->shadow[i].sg[0]),
2239                                                GFP_KERNEL);
2240                 if (info->max_indirect_segments)
2241                         rinfo->shadow[i].indirect_grants =
2242                                 kvcalloc(INDIRECT_GREFS(grants),
2243                                          sizeof(rinfo->shadow[i].indirect_grants[0]),
2244                                          GFP_KERNEL);
2245                 if ((rinfo->shadow[i].grants_used == NULL) ||
2246                         (rinfo->shadow[i].sg == NULL) ||
2247                      (info->max_indirect_segments &&
2248                      (rinfo->shadow[i].indirect_grants == NULL)))
2249                         goto out_of_memory;
2250                 sg_init_table(rinfo->shadow[i].sg, psegs);
2251         }
2252
2253         memalloc_noio_restore(memflags);
2254
2255         return 0;
2256
2257 out_of_memory:
2258         for (i = 0; i < BLK_RING_SIZE(info); i++) {
2259                 kvfree(rinfo->shadow[i].grants_used);
2260                 rinfo->shadow[i].grants_used = NULL;
2261                 kvfree(rinfo->shadow[i].sg);
2262                 rinfo->shadow[i].sg = NULL;
2263                 kvfree(rinfo->shadow[i].indirect_grants);
2264                 rinfo->shadow[i].indirect_grants = NULL;
2265         }
2266         if (!list_empty(&rinfo->indirect_pages)) {
2267                 struct page *indirect_page, *n;
2268                 list_for_each_entry_safe(indirect_page, n, &rinfo->indirect_pages, lru) {
2269                         list_del(&indirect_page->lru);
2270                         __free_page(indirect_page);
2271                 }
2272         }
2273
2274         memalloc_noio_restore(memflags);
2275
2276         return -ENOMEM;
2277 }
2278
2279 /*
2280  * Gather all backend feature-*
2281  */
2282 static void blkfront_gather_backend_features(struct blkfront_info *info)
2283 {
2284         unsigned int indirect_segments;
2285
2286         info->feature_flush = 0;
2287         info->feature_fua = 0;
2288
2289         /*
2290          * If there's no "feature-barrier" defined, then it means
2291          * we're dealing with a very old backend which writes
2292          * synchronously; nothing to do.
2293          *
2294          * If there are barriers, then we use flush.
2295          */
2296         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-barrier", 0)) {
2297                 info->feature_flush = 1;
2298                 info->feature_fua = 1;
2299         }
2300
2301         /*
2302          * And if there is "feature-flush-cache" use that above
2303          * barriers.
2304          */
2305         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-flush-cache",
2306                                  0)) {
2307                 info->feature_flush = 1;
2308                 info->feature_fua = 0;
2309         }
2310
2311         if (xenbus_read_unsigned(info->xbdev->otherend, "feature-discard", 0))
2312                 blkfront_setup_discard(info);
2313
2314         if (info->feature_persistent_parm)
2315                 info->feature_persistent =
2316                         !!xenbus_read_unsigned(info->xbdev->otherend,
2317                                                "feature-persistent", 0);
2318         if (info->feature_persistent)
2319                 info->bounce = true;
2320
2321         indirect_segments = xenbus_read_unsigned(info->xbdev->otherend,
2322                                         "feature-max-indirect-segments", 0);
2323         if (indirect_segments > xen_blkif_max_segments)
2324                 indirect_segments = xen_blkif_max_segments;
2325         if (indirect_segments <= BLKIF_MAX_SEGMENTS_PER_REQUEST)
2326                 indirect_segments = 0;
2327         info->max_indirect_segments = indirect_segments;
2328
2329         if (info->feature_persistent) {
2330                 mutex_lock(&blkfront_mutex);
2331                 schedule_delayed_work(&blkfront_work, HZ * 10);
2332                 mutex_unlock(&blkfront_mutex);
2333         }
2334 }
2335
2336 /*
2337  * Invoked when the backend is finally 'ready' (and has told produced
2338  * the details about the physical device - #sectors, size, etc).
2339  */
2340 static void blkfront_connect(struct blkfront_info *info)
2341 {
2342         unsigned long long sectors;
2343         unsigned long sector_size;
2344         unsigned int physical_sector_size;
2345         unsigned int binfo;
2346         int err, i;
2347         struct blkfront_ring_info *rinfo;
2348
2349         switch (info->connected) {
2350         case BLKIF_STATE_CONNECTED:
2351                 /*
2352                  * Potentially, the back-end may be signalling
2353                  * a capacity change; update the capacity.
2354                  */
2355                 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
2356                                    "sectors", "%Lu", &sectors);
2357                 if (XENBUS_EXIST_ERR(err))
2358                         return;
2359                 printk(KERN_INFO "Setting capacity to %Lu\n",
2360                        sectors);
2361                 set_capacity_and_notify(info->gd, sectors);
2362
2363                 return;
2364         case BLKIF_STATE_SUSPENDED:
2365                 /*
2366                  * If we are recovering from suspension, we need to wait
2367                  * for the backend to announce it's features before
2368                  * reconnecting, at least we need to know if the backend
2369                  * supports indirect descriptors, and how many.
2370                  */
2371                 blkif_recover(info);
2372                 return;
2373
2374         default:
2375                 break;
2376         }
2377
2378         dev_dbg(&info->xbdev->dev, "%s:%s.\n",
2379                 __func__, info->xbdev->otherend);
2380
2381         err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
2382                             "sectors", "%llu", &sectors,
2383                             "info", "%u", &binfo,
2384                             "sector-size", "%lu", &sector_size,
2385                             NULL);
2386         if (err) {
2387                 xenbus_dev_fatal(info->xbdev, err,
2388                                  "reading backend fields at %s",
2389                                  info->xbdev->otherend);
2390                 return;
2391         }
2392
2393         /*
2394          * physical-sector-size is a newer field, so old backends may not
2395          * provide this. Assume physical sector size to be the same as
2396          * sector_size in that case.
2397          */
2398         physical_sector_size = xenbus_read_unsigned(info->xbdev->otherend,
2399                                                     "physical-sector-size",
2400                                                     sector_size);
2401         blkfront_gather_backend_features(info);
2402         for_each_rinfo(info, rinfo, i) {
2403                 err = blkfront_setup_indirect(rinfo);
2404                 if (err) {
2405                         xenbus_dev_fatal(info->xbdev, err, "setup_indirect at %s",
2406                                          info->xbdev->otherend);
2407                         blkif_free(info, 0);
2408                         break;
2409                 }
2410         }
2411
2412         err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size,
2413                                   physical_sector_size);
2414         if (err) {
2415                 xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",
2416                                  info->xbdev->otherend);
2417                 goto fail;
2418         }
2419
2420         xenbus_switch_state(info->xbdev, XenbusStateConnected);
2421
2422         /* Kick pending requests. */
2423         info->connected = BLKIF_STATE_CONNECTED;
2424         for_each_rinfo(info, rinfo, i)
2425                 kick_pending_request_queues(rinfo);
2426
2427         device_add_disk(&info->xbdev->dev, info->gd, NULL);
2428
2429         info->is_ready = 1;
2430         return;
2431
2432 fail:
2433         blkif_free(info, 0);
2434         return;
2435 }
2436
2437 /*
2438  * Callback received when the backend's state changes.
2439  */
2440 static void blkback_changed(struct xenbus_device *dev,
2441                             enum xenbus_state backend_state)
2442 {
2443         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2444
2445         dev_dbg(&dev->dev, "blkfront:blkback_changed to state %d.\n", backend_state);
2446
2447         switch (backend_state) {
2448         case XenbusStateInitWait:
2449                 if (dev->state != XenbusStateInitialising)
2450                         break;
2451                 if (talk_to_blkback(dev, info))
2452                         break;
2453                 break;
2454         case XenbusStateInitialising:
2455         case XenbusStateInitialised:
2456         case XenbusStateReconfiguring:
2457         case XenbusStateReconfigured:
2458         case XenbusStateUnknown:
2459                 break;
2460
2461         case XenbusStateConnected:
2462                 /*
2463                  * talk_to_blkback sets state to XenbusStateInitialised
2464                  * and blkfront_connect sets it to XenbusStateConnected
2465                  * (if connection went OK).
2466                  *
2467                  * If the backend (or toolstack) decides to poke at backend
2468                  * state (and re-trigger the watch by setting the state repeatedly
2469                  * to XenbusStateConnected (4)) we need to deal with this.
2470                  * This is allowed as this is used to communicate to the guest
2471                  * that the size of disk has changed!
2472                  */
2473                 if ((dev->state != XenbusStateInitialised) &&
2474                     (dev->state != XenbusStateConnected)) {
2475                         if (talk_to_blkback(dev, info))
2476                                 break;
2477                 }
2478
2479                 blkfront_connect(info);
2480                 break;
2481
2482         case XenbusStateClosed:
2483                 if (dev->state == XenbusStateClosed)
2484                         break;
2485                 fallthrough;
2486         case XenbusStateClosing:
2487                 blkfront_closing(info);
2488                 break;
2489         }
2490 }
2491
2492 static int blkfront_remove(struct xenbus_device *xbdev)
2493 {
2494         struct blkfront_info *info = dev_get_drvdata(&xbdev->dev);
2495
2496         dev_dbg(&xbdev->dev, "%s removed", xbdev->nodename);
2497
2498         if (info->gd)
2499                 del_gendisk(info->gd);
2500
2501         mutex_lock(&blkfront_mutex);
2502         list_del(&info->info_list);
2503         mutex_unlock(&blkfront_mutex);
2504
2505         blkif_free(info, 0);
2506         if (info->gd) {
2507                 xlbd_release_minors(info->gd->first_minor, info->gd->minors);
2508                 blk_cleanup_disk(info->gd);
2509                 blk_mq_free_tag_set(&info->tag_set);
2510         }
2511
2512         kfree(info);
2513         return 0;
2514 }
2515
2516 static int blkfront_is_ready(struct xenbus_device *dev)
2517 {
2518         struct blkfront_info *info = dev_get_drvdata(&dev->dev);
2519
2520         return info->is_ready && info->xbdev;
2521 }
2522
2523 static const struct block_device_operations xlvbd_block_fops =
2524 {
2525         .owner = THIS_MODULE,
2526         .getgeo = blkif_getgeo,
2527         .ioctl = blkif_ioctl,
2528         .compat_ioctl = blkdev_compat_ptr_ioctl,
2529 };
2530
2531
2532 static const struct xenbus_device_id blkfront_ids[] = {
2533         { "vbd" },
2534         { "" }
2535 };
2536
2537 static struct xenbus_driver blkfront_driver = {
2538         .ids  = blkfront_ids,
2539         .probe = blkfront_probe,
2540         .remove = blkfront_remove,
2541         .resume = blkfront_resume,
2542         .otherend_changed = blkback_changed,
2543         .is_ready = blkfront_is_ready,
2544 };
2545
2546 static void purge_persistent_grants(struct blkfront_info *info)
2547 {
2548         unsigned int i;
2549         unsigned long flags;
2550         struct blkfront_ring_info *rinfo;
2551
2552         for_each_rinfo(info, rinfo, i) {
2553                 struct grant *gnt_list_entry, *tmp;
2554
2555                 spin_lock_irqsave(&rinfo->ring_lock, flags);
2556
2557                 if (rinfo->persistent_gnts_c == 0) {
2558                         spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2559                         continue;
2560                 }
2561
2562                 list_for_each_entry_safe(gnt_list_entry, tmp, &rinfo->grants,
2563                                          node) {
2564                         if (gnt_list_entry->gref == GRANT_INVALID_REF ||
2565                             !gnttab_try_end_foreign_access(gnt_list_entry->gref))
2566                                 continue;
2567
2568                         list_del(&gnt_list_entry->node);
2569                         rinfo->persistent_gnts_c--;
2570                         gnt_list_entry->gref = GRANT_INVALID_REF;
2571                         list_add_tail(&gnt_list_entry->node, &rinfo->grants);
2572                 }
2573
2574                 spin_unlock_irqrestore(&rinfo->ring_lock, flags);
2575         }
2576 }
2577
2578 static void blkfront_delay_work(struct work_struct *work)
2579 {
2580         struct blkfront_info *info;
2581         bool need_schedule_work = false;
2582
2583         /*
2584          * Note that when using bounce buffers but not persistent grants
2585          * there's no need to run blkfront_delay_work because grants are
2586          * revoked in blkif_completion or else an error is reported and the
2587          * connection is closed.
2588          */
2589
2590         mutex_lock(&blkfront_mutex);
2591
2592         list_for_each_entry(info, &info_list, info_list) {
2593                 if (info->feature_persistent) {
2594                         need_schedule_work = true;
2595                         mutex_lock(&info->mutex);
2596                         purge_persistent_grants(info);
2597                         mutex_unlock(&info->mutex);
2598                 }
2599         }
2600
2601         if (need_schedule_work)
2602                 schedule_delayed_work(&blkfront_work, HZ * 10);
2603
2604         mutex_unlock(&blkfront_mutex);
2605 }
2606
2607 static int __init xlblk_init(void)
2608 {
2609         int ret;
2610         int nr_cpus = num_online_cpus();
2611
2612         if (!xen_domain())
2613                 return -ENODEV;
2614
2615         if (!xen_has_pv_disk_devices())
2616                 return -ENODEV;
2617
2618         if (register_blkdev(XENVBD_MAJOR, DEV_NAME)) {
2619                 pr_warn("xen_blk: can't get major %d with name %s\n",
2620                         XENVBD_MAJOR, DEV_NAME);
2621                 return -ENODEV;
2622         }
2623
2624         if (xen_blkif_max_segments < BLKIF_MAX_SEGMENTS_PER_REQUEST)
2625                 xen_blkif_max_segments = BLKIF_MAX_SEGMENTS_PER_REQUEST;
2626
2627         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_GRANT_ORDER) {
2628                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
2629                         xen_blkif_max_ring_order, XENBUS_MAX_RING_GRANT_ORDER);
2630                 xen_blkif_max_ring_order = XENBUS_MAX_RING_GRANT_ORDER;
2631         }
2632
2633         if (xen_blkif_max_queues > nr_cpus) {
2634                 pr_info("Invalid max_queues (%d), will use default max: %d.\n",
2635                         xen_blkif_max_queues, nr_cpus);
2636                 xen_blkif_max_queues = nr_cpus;
2637         }
2638
2639         INIT_DELAYED_WORK(&blkfront_work, blkfront_delay_work);
2640
2641         ret = xenbus_register_frontend(&blkfront_driver);
2642         if (ret) {
2643                 unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2644                 return ret;
2645         }
2646
2647         return 0;
2648 }
2649 module_init(xlblk_init);
2650
2651
2652 static void __exit xlblk_exit(void)
2653 {
2654         cancel_delayed_work_sync(&blkfront_work);
2655
2656         xenbus_unregister_driver(&blkfront_driver);
2657         unregister_blkdev(XENVBD_MAJOR, DEV_NAME);
2658         kfree(minors);
2659 }
2660 module_exit(xlblk_exit);
2661
2662 MODULE_DESCRIPTION("Xen virtual block device frontend");
2663 MODULE_LICENSE("GPL");
2664 MODULE_ALIAS_BLOCKDEV_MAJOR(XENVBD_MAJOR);
2665 MODULE_ALIAS("xen:vbd");
2666 MODULE_ALIAS("xenblk");