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