drivers/block/xen-blkback/blkback.c

   1 /******************************************************************************
   2  *
   3  * Back-end of the driver for virtual block devices. This portion of the
   4  * driver exports a 'unified' block-device interface that can be accessed
   5  * by any operating system that implements a compatible front end. A
   6  * reference front-end implementation can be found in:
   7  *  drivers/block/xen-blkfront.c
   8  *
   9  * Copyright (c) 2003-2004, Keir Fraser & Steve Hand
  10  * Copyright (c) 2005, Christopher Clark
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public License version 2
  14  * as published by the Free Software Foundation; or, when distributed
  15  * separately from the Linux kernel or incorporated into other
  16  * software packages, subject to the following license:
  17  *
  18  * Permission is hereby granted, free of charge, to any person obtaining a copy
  19  * of this source file (the "Software"), to deal in the Software without
  20  * restriction, including without limitation the rights to use, copy, modify,
  21  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  22  * and to permit persons to whom the Software is furnished to do so, subject to
  23  * the following conditions:
  24  *
  25  * The above copyright notice and this permission notice shall be included in
  26  * all copies or substantial portions of the Software.
  27  *
  28  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  29  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  30  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  31  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  32  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  33  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  34  * IN THE SOFTWARE.
  35  */
  36
  37 #define pr_fmt(fmt) "xen-blkback: " fmt
  38
  39 #include <linux/spinlock.h>
  40 #include <linux/kthread.h>
  41 #include <linux/list.h>
  42 #include <linux/delay.h>
  43 #include <linux/freezer.h>
  44 #include <linux/bitmap.h>
  45
  46 #include <xen/events.h>
  47 #include <xen/page.h>
  48 #include <xen/xen.h>
  49 #include <asm/xen/hypervisor.h>
  50 #include <asm/xen/hypercall.h>
  51 #include <xen/balloon.h>
  52 #include <xen/grant_table.h>
  53 #include "common.h"
  54
  55 /*
  56  * Maximum number of unused free pages to keep in the internal buffer.
  57  * Setting this to a value too low will reduce memory used in each backend,
  58  * but can have a performance penalty.
  59  *
  60  * A sane value is xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST, but can
  61  * be set to a lower value that might degrade performance on some intensive
  62  * IO workloads.
  63  */
  64
  65 static int xen_blkif_max_buffer_pages = 1024;
  66 module_param_named(max_buffer_pages, xen_blkif_max_buffer_pages, int, 0644);
  67 MODULE_PARM_DESC(max_buffer_pages,
  68 "Maximum number of free pages to keep in each block backend buffer");
  69
  70 /*
  71  * Maximum number of grants to map persistently in blkback. For maximum
  72  * performance this should be the total numbers of grants that can be used
  73  * to fill the ring, but since this might become too high, specially with
  74  * the use of indirect descriptors, we set it to a value that provides good
  75  * performance without using too much memory.
  76  *
  77  * When the list of persistent grants is full we clean it up using a LRU
  78  * algorithm.
  79  */
  80
  81 static int xen_blkif_max_pgrants = 1056;
  82 module_param_named(max_persistent_grants, xen_blkif_max_pgrants, int, 0644);
  83 MODULE_PARM_DESC(max_persistent_grants,
  84                  "Maximum number of grants to map persistently");
  85
  86 /*
  87  * Maximum order of pages to be used for the shared ring between front and
  88  * backend, 4KB page granularity is used.
  89  */
  90 unsigned int xen_blkif_max_ring_order = XENBUS_MAX_RING_PAGE_ORDER;
  91 module_param_named(max_ring_page_order, xen_blkif_max_ring_order, int, S_IRUGO);
  92 MODULE_PARM_DESC(max_ring_page_order, "Maximum order of pages to be used for the shared ring");
  93 /*
  94  * The LRU mechanism to clean the lists of persistent grants needs to
  95  * be executed periodically. The time interval between consecutive executions
  96  * of the purge mechanism is set in ms.
  97  */
  98 #define LRU_INTERVAL 100
  99
 100 /*
 101  * When the persistent grants list is full we will remove unused grants
 102  * from the list. The percent number of grants to be removed at each LRU
 103  * execution.
 104  */
 105 #define LRU_PERCENT_CLEAN 5
 106
 107 /* Run-time switchable: /sys/module/blkback/parameters/ */
 108 static unsigned int log_stats;
 109 module_param(log_stats, int, 0644);
 110
 111 #define BLKBACK_INVALID_HANDLE (~0)
 112
 113 /* Number of free pages to remove on each call to gnttab_free_pages */
 114 #define NUM_BATCH_FREE_PAGES 10
 115
 116 static inline int get_free_page(struct xen_blkif *blkif, struct page **page)
 117 {
 118         unsigned long flags;
 119
 120         spin_lock_irqsave(&blkif->free_pages_lock, flags);
 121         if (list_empty(&blkif->free_pages)) {
 122                 BUG_ON(blkif->free_pages_num != 0);
 123                 spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
 124                 return gnttab_alloc_pages(1, page);
 125         }
 126         BUG_ON(blkif->free_pages_num == 0);
 127         page[0] = list_first_entry(&blkif->free_pages, struct page, lru);
 128         list_del(&page[0]->lru);
 129         blkif->free_pages_num--;
 130         spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
 131
 132         return 0;
 133 }
 134
 135 static inline void put_free_pages(struct xen_blkif *blkif, struct page **page,
 136                                   int num)
 137 {
 138         unsigned long flags;
 139         int i;
 140
 141         spin_lock_irqsave(&blkif->free_pages_lock, flags);
 142         for (i = 0; i < num; i++)
 143                 list_add(&page[i]->lru, &blkif->free_pages);
 144         blkif->free_pages_num += num;
 145         spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
 146 }
 147
 148 static inline void shrink_free_pagepool(struct xen_blkif *blkif, int num)
 149 {
 150         /* Remove requested pages in batches of NUM_BATCH_FREE_PAGES */
 151         struct page *page[NUM_BATCH_FREE_PAGES];
 152         unsigned int num_pages = 0;
 153         unsigned long flags;
 154
 155         spin_lock_irqsave(&blkif->free_pages_lock, flags);
 156         while (blkif->free_pages_num > num) {
 157                 BUG_ON(list_empty(&blkif->free_pages));
 158                 page[num_pages] = list_first_entry(&blkif->free_pages,
 159                                                    struct page, lru);
 160                 list_del(&page[num_pages]->lru);
 161                 blkif->free_pages_num--;
 162                 if (++num_pages == NUM_BATCH_FREE_PAGES) {
 163                         spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
 164                         gnttab_free_pages(num_pages, page);
 165                         spin_lock_irqsave(&blkif->free_pages_lock, flags);
 166                         num_pages = 0;
 167                 }
 168         }
 169         spin_unlock_irqrestore(&blkif->free_pages_lock, flags);
 170         if (num_pages != 0)
 171                 gnttab_free_pages(num_pages, page);
 172 }
 173
 174 #define vaddr(page) ((unsigned long)pfn_to_kaddr(page_to_pfn(page)))
 175
 176 static int do_block_io_op(struct xen_blkif *blkif);
 177 static int dispatch_rw_block_io(struct xen_blkif *blkif,
 178                                 struct blkif_request *req,
 179                                 struct pending_req *pending_req);
 180 static void make_response(struct xen_blkif *blkif, u64 id,
 181                           unsigned short op, int st);
 182
 183 #define foreach_grant_safe(pos, n, rbtree, node) \
 184         for ((pos) = container_of(rb_first((rbtree)), typeof(*(pos)), node), \
 185              (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL; \
 186              &(pos)->node != NULL; \
 187              (pos) = container_of(n, typeof(*(pos)), node), \
 188              (n) = (&(pos)->node != NULL) ? rb_next(&(pos)->node) : NULL)
 189
 190
 191 /*
 192  * We don't need locking around the persistent grant helpers
 193  * because blkback uses a single-thread for each backed, so we
 194  * can be sure that this functions will never be called recursively.
 195  *
 196  * The only exception to that is put_persistent_grant, that can be called
 197  * from interrupt context (by xen_blkbk_unmap), so we have to use atomic
 198  * bit operations to modify the flags of a persistent grant and to count
 199  * the number of used grants.
 200  */
 201 static int add_persistent_gnt(struct xen_blkif *blkif,
 202                                struct persistent_gnt *persistent_gnt)
 203 {
 204         struct rb_node **new = NULL, *parent = NULL;
 205         struct persistent_gnt *this;
 206
 207         if (blkif->persistent_gnt_c >= xen_blkif_max_pgrants) {
 208                 if (!blkif->vbd.overflow_max_grants)
 209                         blkif->vbd.overflow_max_grants = 1;
 210                 return -EBUSY;
 211         }
 212         /* Figure out where to put new node */
 213         new = &blkif->persistent_gnts.rb_node;
 214         while (*new) {
 215                 this = container_of(*new, struct persistent_gnt, node);
 216
 217                 parent = *new;
 218                 if (persistent_gnt->gnt < this->gnt)
 219                         new = &((*new)->rb_left);
 220                 else if (persistent_gnt->gnt > this->gnt)
 221                         new = &((*new)->rb_right);
 222                 else {
 223                         pr_alert_ratelimited("trying to add a gref that's already in the tree\n");
 224                         return -EINVAL;
 225                 }
 226         }
 227
 228         bitmap_zero(persistent_gnt->flags, PERSISTENT_GNT_FLAGS_SIZE);
 229         set_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags);
 230         /* Add new node and rebalance tree. */
 231         rb_link_node(&(persistent_gnt->node), parent, new);
 232         rb_insert_color(&(persistent_gnt->node), &blkif->persistent_gnts);
 233         blkif->persistent_gnt_c++;
 234         atomic_inc(&blkif->persistent_gnt_in_use);
 235         return 0;
 236 }
 237
 238 static struct persistent_gnt *get_persistent_gnt(struct xen_blkif *blkif,
 239                                                  grant_ref_t gref)
 240 {
 241         struct persistent_gnt *data;
 242         struct rb_node *node = NULL;
 243
 244         node = blkif->persistent_gnts.rb_node;
 245         while (node) {
 246                 data = container_of(node, struct persistent_gnt, node);
 247
 248                 if (gref < data->gnt)
 249                         node = node->rb_left;
 250                 else if (gref > data->gnt)
 251                         node = node->rb_right;
 252                 else {
 253                         if(test_bit(PERSISTENT_GNT_ACTIVE, data->flags)) {
 254                                 pr_alert_ratelimited("requesting a grant already in use\n");
 255                                 return NULL;
 256                         }
 257                         set_bit(PERSISTENT_GNT_ACTIVE, data->flags);
 258                         atomic_inc(&blkif->persistent_gnt_in_use);
 259                         return data;
 260                 }
 261         }
 262         return NULL;
 263 }
 264
 265 static void put_persistent_gnt(struct xen_blkif *blkif,
 266                                struct persistent_gnt *persistent_gnt)
 267 {
 268         if(!test_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags))
 269                 pr_alert_ratelimited("freeing a grant already unused\n");
 270         set_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags);
 271         clear_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags);
 272         atomic_dec(&blkif->persistent_gnt_in_use);
 273 }
 274
 275 static void free_persistent_gnts(struct xen_blkif *blkif, struct rb_root *root,
 276                                  unsigned int num)
 277 {
 278         struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 279         struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 280         struct persistent_gnt *persistent_gnt;
 281         struct rb_node *n;
 282         int segs_to_unmap = 0;
 283         struct gntab_unmap_queue_data unmap_data;
 284
 285         unmap_data.pages = pages;
 286         unmap_data.unmap_ops = unmap;
 287         unmap_data.kunmap_ops = NULL;
 288
 289         foreach_grant_safe(persistent_gnt, n, root, node) {
 290                 BUG_ON(persistent_gnt->handle ==
 291                         BLKBACK_INVALID_HANDLE);
 292                 gnttab_set_unmap_op(&unmap[segs_to_unmap],
 293                         (unsigned long) pfn_to_kaddr(page_to_pfn(
 294                                 persistent_gnt->page)),
 295                         GNTMAP_host_map,
 296                         persistent_gnt->handle);
 297
 298                 pages[segs_to_unmap] = persistent_gnt->page;
 299
 300                 if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST ||
 301                         !rb_next(&persistent_gnt->node)) {
 302
 303                         unmap_data.count = segs_to_unmap;
 304                         BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
 305
 306                         put_free_pages(blkif, pages, segs_to_unmap);
 307                         segs_to_unmap = 0;
 308                 }
 309
 310                 rb_erase(&persistent_gnt->node, root);
 311                 kfree(persistent_gnt);
 312                 num--;
 313         }
 314         BUG_ON(num != 0);
 315 }
 316
 317 void xen_blkbk_unmap_purged_grants(struct work_struct *work)
 318 {
 319         struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 320         struct page *pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 321         struct persistent_gnt *persistent_gnt;
 322         int segs_to_unmap = 0;
 323         struct xen_blkif *blkif = container_of(work, typeof(*blkif), persistent_purge_work);
 324         struct gntab_unmap_queue_data unmap_data;
 325
 326         unmap_data.pages = pages;
 327         unmap_data.unmap_ops = unmap;
 328         unmap_data.kunmap_ops = NULL;
 329
 330         while(!list_empty(&blkif->persistent_purge_list)) {
 331                 persistent_gnt = list_first_entry(&blkif->persistent_purge_list,
 332                                                   struct persistent_gnt,
 333                                                   remove_node);
 334                 list_del(&persistent_gnt->remove_node);
 335
 336                 gnttab_set_unmap_op(&unmap[segs_to_unmap],
 337                         vaddr(persistent_gnt->page),
 338                         GNTMAP_host_map,
 339                         persistent_gnt->handle);
 340
 341                 pages[segs_to_unmap] = persistent_gnt->page;
 342
 343                 if (++segs_to_unmap == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 344                         unmap_data.count = segs_to_unmap;
 345                         BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
 346                         put_free_pages(blkif, pages, segs_to_unmap);
 347                         segs_to_unmap = 0;
 348                 }
 349                 kfree(persistent_gnt);
 350         }
 351         if (segs_to_unmap > 0) {
 352                 unmap_data.count = segs_to_unmap;
 353                 BUG_ON(gnttab_unmap_refs_sync(&unmap_data));
 354                 put_free_pages(blkif, pages, segs_to_unmap);
 355         }
 356 }
 357
 358 static void purge_persistent_gnt(struct xen_blkif *blkif)
 359 {
 360         struct persistent_gnt *persistent_gnt;
 361         struct rb_node *n;
 362         unsigned int num_clean, total;
 363         bool scan_used = false, clean_used = false;
 364         struct rb_root *root;
 365
 366         if (blkif->persistent_gnt_c < xen_blkif_max_pgrants ||
 367             (blkif->persistent_gnt_c == xen_blkif_max_pgrants &&
 368             !blkif->vbd.overflow_max_grants)) {
 369                 return;
 370         }
 371
 372         if (work_busy(&blkif->persistent_purge_work)) {
 373                 pr_alert_ratelimited("Scheduled work from previous purge is still busy, cannot purge list\n");
 374                 return;
 375         }
 376
 377         num_clean = (xen_blkif_max_pgrants / 100) * LRU_PERCENT_CLEAN;
 378         num_clean = blkif->persistent_gnt_c - xen_blkif_max_pgrants + num_clean;
 379         num_clean = min(blkif->persistent_gnt_c, num_clean);
 380         if ((num_clean == 0) ||
 381             (num_clean > (blkif->persistent_gnt_c - atomic_read(&blkif->persistent_gnt_in_use))))
 382                 return;
 383
 384         /*
 385          * At this point, we can assure that there will be no calls
 386          * to get_persistent_grant (because we are executing this code from
 387          * xen_blkif_schedule), there can only be calls to put_persistent_gnt,
 388          * which means that the number of currently used grants will go down,
 389          * but never up, so we will always be able to remove the requested
 390          * number of grants.
 391          */
 392
 393         total = num_clean;
 394
 395         pr_debug("Going to purge %u persistent grants\n", num_clean);
 396
 397         BUG_ON(!list_empty(&blkif->persistent_purge_list));
 398         root = &blkif->persistent_gnts;
 399 purge_list:
 400         foreach_grant_safe(persistent_gnt, n, root, node) {
 401                 BUG_ON(persistent_gnt->handle ==
 402                         BLKBACK_INVALID_HANDLE);
 403
 404                 if (clean_used) {
 405                         clear_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags);
 406                         continue;
 407                 }
 408
 409                 if (test_bit(PERSISTENT_GNT_ACTIVE, persistent_gnt->flags))
 410                         continue;
 411                 if (!scan_used &&
 412                     (test_bit(PERSISTENT_GNT_WAS_ACTIVE, persistent_gnt->flags)))
 413                         continue;
 414
 415                 rb_erase(&persistent_gnt->node, root);
 416                 list_add(&persistent_gnt->remove_node,
 417                          &blkif->persistent_purge_list);
 418                 if (--num_clean == 0)
 419                         goto finished;
 420         }
 421         /*
 422          * If we get here it means we also need to start cleaning
 423          * grants that were used since last purge in order to cope
 424          * with the requested num
 425          */
 426         if (!scan_used && !clean_used) {
 427                 pr_debug("Still missing %u purged frames\n", num_clean);
 428                 scan_used = true;
 429                 goto purge_list;
 430         }
 431 finished:
 432         if (!clean_used) {
 433                 pr_debug("Finished scanning for grants to clean, removing used flag\n");
 434                 clean_used = true;
 435                 goto purge_list;
 436         }
 437
 438         blkif->persistent_gnt_c -= (total - num_clean);
 439         blkif->vbd.overflow_max_grants = 0;
 440
 441         /* We can defer this work */
 442         schedule_work(&blkif->persistent_purge_work);
 443         pr_debug("Purged %u/%u\n", (total - num_clean), total);
 444         return;
 445 }
 446
 447 /*
 448  * Retrieve from the 'pending_reqs' a free pending_req structure to be used.
 449  */
 450 static struct pending_req *alloc_req(struct xen_blkif *blkif)
 451 {
 452         struct pending_req *req = NULL;
 453         unsigned long flags;
 454
 455         spin_lock_irqsave(&blkif->pending_free_lock, flags);
 456         if (!list_empty(&blkif->pending_free)) {
 457                 req = list_entry(blkif->pending_free.next, struct pending_req,
 458                                  free_list);
 459                 list_del(&req->free_list);
 460         }
 461         spin_unlock_irqrestore(&blkif->pending_free_lock, flags);
 462         return req;
 463 }
 464
 465 /*
 466  * Return the 'pending_req' structure back to the freepool. We also
 467  * wake up the thread if it was waiting for a free page.
 468  */
 469 static void free_req(struct xen_blkif *blkif, struct pending_req *req)
 470 {
 471         unsigned long flags;
 472         int was_empty;
 473
 474         spin_lock_irqsave(&blkif->pending_free_lock, flags);
 475         was_empty = list_empty(&blkif->pending_free);
 476         list_add(&req->free_list, &blkif->pending_free);
 477         spin_unlock_irqrestore(&blkif->pending_free_lock, flags);
 478         if (was_empty)
 479                 wake_up(&blkif->pending_free_wq);
 480 }
 481
 482 /*
 483  * Routines for managing virtual block devices (vbds).
 484  */
 485 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
 486                              int operation)
 487 {
 488         struct xen_vbd *vbd = &blkif->vbd;
 489         int rc = -EACCES;
 490
 491         if ((operation != READ) && vbd->readonly)
 492                 goto out;
 493
 494         if (likely(req->nr_sects)) {
 495                 blkif_sector_t end = req->sector_number + req->nr_sects;
 496
 497                 if (unlikely(end < req->sector_number))
 498                         goto out;
 499                 if (unlikely(end > vbd_sz(vbd)))
 500                         goto out;
 501         }
 502
 503         req->dev  = vbd->pdevice;
 504         req->bdev = vbd->bdev;
 505         rc = 0;
 506
 507  out:
 508         return rc;
 509 }
 510
 511 static void xen_vbd_resize(struct xen_blkif *blkif)
 512 {
 513         struct xen_vbd *vbd = &blkif->vbd;
 514         struct xenbus_transaction xbt;
 515         int err;
 516         struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
 517         unsigned long long new_size = vbd_sz(vbd);
 518
 519         pr_info("VBD Resize: Domid: %d, Device: (%d, %d)\n",
 520                 blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
 521         pr_info("VBD Resize: new size %llu\n", new_size);
 522         vbd->size = new_size;
 523 again:
 524         err = xenbus_transaction_start(&xbt);
 525         if (err) {
 526                 pr_warn("Error starting transaction\n");
 527                 return;
 528         }
 529         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 530                             (unsigned long long)vbd_sz(vbd));
 531         if (err) {
 532                 pr_warn("Error writing new size\n");
 533                 goto abort;
 534         }
 535         /*
 536          * Write the current state; we will use this to synchronize
 537          * the front-end. If the current state is "connected" the
 538          * front-end will get the new size information online.
 539          */
 540         err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
 541         if (err) {
 542                 pr_warn("Error writing the state\n");
 543                 goto abort;
 544         }
 545
 546         err = xenbus_transaction_end(xbt, 0);
 547         if (err == -EAGAIN)
 548                 goto again;
 549         if (err)
 550                 pr_warn("Error ending transaction\n");
 551         return;
 552 abort:
 553         xenbus_transaction_end(xbt, 1);
 554 }
 555
 556 /*
 557  * Notification from the guest OS.
 558  */
 559 static void blkif_notify_work(struct xen_blkif *blkif)
 560 {
 561         blkif->waiting_reqs = 1;
 562         wake_up(&blkif->wq);
 563 }
 564
 565 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
 566 {
 567         blkif_notify_work(dev_id);
 568         return IRQ_HANDLED;
 569 }
 570
 571 /*
 572  * SCHEDULER FUNCTIONS
 573  */
 574
 575 static void print_stats(struct xen_blkif *blkif)
 576 {
 577         pr_info("(%s): oo %3llu  |  rd %4llu  |  wr %4llu  |  f %4llu"
 578                  "  |  ds %4llu | pg: %4u/%4d\n",
 579                  current->comm, blkif->st_oo_req,
 580                  blkif->st_rd_req, blkif->st_wr_req,
 581                  blkif->st_f_req, blkif->st_ds_req,
 582                  blkif->persistent_gnt_c,
 583                  xen_blkif_max_pgrants);
 584         blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000);
 585         blkif->st_rd_req = 0;
 586         blkif->st_wr_req = 0;
 587         blkif->st_oo_req = 0;
 588         blkif->st_ds_req = 0;
 589 }
 590
 591 int xen_blkif_schedule(void *arg)
 592 {
 593         struct xen_blkif *blkif = arg;
 594         struct xen_vbd *vbd = &blkif->vbd;
 595         unsigned long timeout;
 596         int ret;
 597
 598         xen_blkif_get(blkif);
 599
 600         while (!kthread_should_stop()) {
 601                 if (try_to_freeze())
 602                         continue;
 603                 if (unlikely(vbd->size != vbd_sz(vbd)))
 604                         xen_vbd_resize(blkif);
 605
 606                 timeout = msecs_to_jiffies(LRU_INTERVAL);
 607
 608                 timeout = wait_event_interruptible_timeout(
 609                         blkif->wq,
 610                         blkif->waiting_reqs || kthread_should_stop(),
 611                         timeout);
 612                 if (timeout == 0)
 613                         goto purge_gnt_list;
 614                 timeout = wait_event_interruptible_timeout(
 615                         blkif->pending_free_wq,
 616                         !list_empty(&blkif->pending_free) ||
 617                         kthread_should_stop(),
 618                         timeout);
 619                 if (timeout == 0)
 620                         goto purge_gnt_list;
 621
 622                 blkif->waiting_reqs = 0;
 623                 smp_mb(); /* clear flag *before* checking for work */
 624
 625                 ret = do_block_io_op(blkif);
 626                 if (ret > 0)
 627                         blkif->waiting_reqs = 1;
 628                 if (ret == -EACCES)
 629                         wait_event_interruptible(blkif->shutdown_wq,
 630                                                  kthread_should_stop());
 631
 632 purge_gnt_list:
 633                 if (blkif->vbd.feature_gnt_persistent &&
 634                     time_after(jiffies, blkif->next_lru)) {
 635                         purge_persistent_gnt(blkif);
 636                         blkif->next_lru = jiffies + msecs_to_jiffies(LRU_INTERVAL);
 637                 }
 638
 639                 /* Shrink if we have more than xen_blkif_max_buffer_pages */
 640                 shrink_free_pagepool(blkif, xen_blkif_max_buffer_pages);
 641
 642                 if (log_stats && time_after(jiffies, blkif->st_print))
 643                         print_stats(blkif);
 644         }
 645
 646         /* Drain pending purge work */
 647         flush_work(&blkif->persistent_purge_work);
 648
 649         if (log_stats)
 650                 print_stats(blkif);
 651
 652         blkif->xenblkd = NULL;
 653         xen_blkif_put(blkif);
 654
 655         return 0;
 656 }
 657
 658 /*
 659  * Remove persistent grants and empty the pool of free pages
 660  */
 661 void xen_blkbk_free_caches(struct xen_blkif *blkif)
 662 {
 663         /* Free all persistent grant pages */
 664         if (!RB_EMPTY_ROOT(&blkif->persistent_gnts))
 665                 free_persistent_gnts(blkif, &blkif->persistent_gnts,
 666                         blkif->persistent_gnt_c);
 667
 668         BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
 669         blkif->persistent_gnt_c = 0;
 670
 671         /* Since we are shutting down remove all pages from the buffer */
 672         shrink_free_pagepool(blkif, 0 /* All */);
 673 }
 674
 675 static unsigned int xen_blkbk_unmap_prepare(
 676         struct xen_blkif *blkif,
 677         struct grant_page **pages,
 678         unsigned int num,
 679         struct gnttab_unmap_grant_ref *unmap_ops,
 680         struct page **unmap_pages)
 681 {
 682         unsigned int i, invcount = 0;
 683
 684         for (i = 0; i < num; i++) {
 685                 if (pages[i]->persistent_gnt != NULL) {
 686                         put_persistent_gnt(blkif, pages[i]->persistent_gnt);
 687                         continue;
 688                 }
 689                 if (pages[i]->handle == BLKBACK_INVALID_HANDLE)
 690                         continue;
 691                 unmap_pages[invcount] = pages[i]->page;
 692                 gnttab_set_unmap_op(&unmap_ops[invcount], vaddr(pages[i]->page),
 693                                     GNTMAP_host_map, pages[i]->handle);
 694                 pages[i]->handle = BLKBACK_INVALID_HANDLE;
 695                 invcount++;
 696        }
 697
 698        return invcount;
 699 }
 700
 701 static void xen_blkbk_unmap_and_respond_callback(int result, struct gntab_unmap_queue_data *data)
 702 {
 703         struct pending_req* pending_req = (struct pending_req*) (data->data);
 704         struct xen_blkif *blkif = pending_req->blkif;
 705
 706         /* BUG_ON used to reproduce existing behaviour,
 707            but is this the best way to deal with this? */
 708         BUG_ON(result);
 709
 710         put_free_pages(blkif, data->pages, data->count);
 711         make_response(blkif, pending_req->id,
 712                       pending_req->operation, pending_req->status);
 713         free_req(blkif, pending_req);
 714         /*
 715          * Make sure the request is freed before releasing blkif,
 716          * or there could be a race between free_req and the
 717          * cleanup done in xen_blkif_free during shutdown.
 718          *
 719          * NB: The fact that we might try to wake up pending_free_wq
 720          * before drain_complete (in case there's a drain going on)
 721          * it's not a problem with our current implementation
 722          * because we can assure there's no thread waiting on
 723          * pending_free_wq if there's a drain going on, but it has
 724          * to be taken into account if the current model is changed.
 725          */
 726         if (atomic_dec_and_test(&blkif->inflight) && atomic_read(&blkif->drain)) {
 727                 complete(&blkif->drain_complete);
 728         }
 729         xen_blkif_put(blkif);
 730 }
 731
 732 static void xen_blkbk_unmap_and_respond(struct pending_req *req)
 733 {
 734         struct gntab_unmap_queue_data* work = &req->gnttab_unmap_data;
 735         struct xen_blkif *blkif = req->blkif;
 736         struct grant_page **pages = req->segments;
 737         unsigned int invcount;
 738
 739         invcount = xen_blkbk_unmap_prepare(blkif, pages, req->nr_segs,
 740                                            req->unmap, req->unmap_pages);
 741
 742         work->data = req;
 743         work->done = xen_blkbk_unmap_and_respond_callback;
 744         work->unmap_ops = req->unmap;
 745         work->kunmap_ops = NULL;
 746         work->pages = req->unmap_pages;
 747         work->count = invcount;
 748
 749         gnttab_unmap_refs_async(&req->gnttab_unmap_data);
 750 }
 751
 752
 753 /*
 754  * Unmap the grant references.
 755  *
 756  * This could accumulate ops up to the batch size to reduce the number
 757  * of hypercalls, but since this is only used in error paths there's
 758  * no real need.
 759  */
 760 static void xen_blkbk_unmap(struct xen_blkif *blkif,
 761                             struct grant_page *pages[],
 762                             int num)
 763 {
 764         struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 765         struct page *unmap_pages[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 766         unsigned int invcount = 0;
 767         int ret;
 768
 769         while (num) {
 770                 unsigned int batch = min(num, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 771
 772                 invcount = xen_blkbk_unmap_prepare(blkif, pages, batch,
 773                                                    unmap, unmap_pages);
 774                 if (invcount) {
 775                         ret = gnttab_unmap_refs(unmap, NULL, unmap_pages, invcount);
 776                         BUG_ON(ret);
 777                         put_free_pages(blkif, unmap_pages, invcount);
 778                 }
 779                 pages += batch;
 780                 num -= batch;
 781         }
 782 }
 783
 784 static int xen_blkbk_map(struct xen_blkif *blkif,
 785                          struct grant_page *pages[],
 786                          int num, bool ro)
 787 {
 788         struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 789         struct page *pages_to_gnt[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 790         struct persistent_gnt *persistent_gnt = NULL;
 791         phys_addr_t addr = 0;
 792         int i, seg_idx, new_map_idx;
 793         int segs_to_map = 0;
 794         int ret = 0;
 795         int last_map = 0, map_until = 0;
 796         int use_persistent_gnts;
 797
 798         use_persistent_gnts = (blkif->vbd.feature_gnt_persistent);
 799
 800         /*
 801          * Fill out preq.nr_sects with proper amount of sectors, and setup
 802          * assign map[..] with the PFN of the page in our domain with the
 803          * corresponding grant reference for each page.
 804          */
 805 again:
 806         for (i = map_until; i < num; i++) {
 807                 uint32_t flags;
 808
 809                 if (use_persistent_gnts)
 810                         persistent_gnt = get_persistent_gnt(
 811                                 blkif,
 812                                 pages[i]->gref);
 813
 814                 if (persistent_gnt) {
 815                         /*
 816                          * We are using persistent grants and
 817                          * the grant is already mapped
 818                          */
 819                         pages[i]->page = persistent_gnt->page;
 820                         pages[i]->persistent_gnt = persistent_gnt;
 821                 } else {
 822                         if (get_free_page(blkif, &pages[i]->page))
 823                                 goto out_of_memory;
 824                         addr = vaddr(pages[i]->page);
 825                         pages_to_gnt[segs_to_map] = pages[i]->page;
 826                         pages[i]->persistent_gnt = NULL;
 827                         flags = GNTMAP_host_map;
 828                         if (!use_persistent_gnts && ro)
 829                                 flags |= GNTMAP_readonly;
 830                         gnttab_set_map_op(&map[segs_to_map++], addr,
 831                                           flags, pages[i]->gref,
 832                                           blkif->domid);
 833                 }
 834                 map_until = i + 1;
 835                 if (segs_to_map == BLKIF_MAX_SEGMENTS_PER_REQUEST)
 836                         break;
 837         }
 838
 839         if (segs_to_map) {
 840                 ret = gnttab_map_refs(map, NULL, pages_to_gnt, segs_to_map);
 841                 BUG_ON(ret);
 842         }
 843
 844         /*
 845          * Now swizzle the MFN in our domain with the MFN from the other domain
 846          * so that when we access vaddr(pending_req,i) it has the contents of
 847          * the page from the other domain.
 848          */
 849         for (seg_idx = last_map, new_map_idx = 0; seg_idx < map_until; seg_idx++) {
 850                 if (!pages[seg_idx]->persistent_gnt) {
 851                         /* This is a newly mapped grant */
 852                         BUG_ON(new_map_idx >= segs_to_map);
 853                         if (unlikely(map[new_map_idx].status != 0)) {
 854                                 pr_debug("invalid buffer -- could not remap it\n");
 855                                 put_free_pages(blkif, &pages[seg_idx]->page, 1);
 856                                 pages[seg_idx]->handle = BLKBACK_INVALID_HANDLE;
 857                                 ret |= 1;
 858                                 goto next;
 859                         }
 860                         pages[seg_idx]->handle = map[new_map_idx].handle;
 861                 } else {
 862                         continue;
 863                 }
 864                 if (use_persistent_gnts &&
 865                     blkif->persistent_gnt_c < xen_blkif_max_pgrants) {
 866                         /*
 867                          * We are using persistent grants, the grant is
 868                          * not mapped but we might have room for it.
 869                          */
 870                         persistent_gnt = kmalloc(sizeof(struct persistent_gnt),
 871                                                  GFP_KERNEL);
 872                         if (!persistent_gnt) {
 873                                 /*
 874                                  * If we don't have enough memory to
 875                                  * allocate the persistent_gnt struct
 876                                  * map this grant non-persistenly
 877                                  */
 878                                 goto next;
 879                         }
 880                         persistent_gnt->gnt = map[new_map_idx].ref;
 881                         persistent_gnt->handle = map[new_map_idx].handle;
 882                         persistent_gnt->page = pages[seg_idx]->page;
 883                         if (add_persistent_gnt(blkif,
 884                                                persistent_gnt)) {
 885                                 kfree(persistent_gnt);
 886                                 persistent_gnt = NULL;
 887                                 goto next;
 888                         }
 889                         pages[seg_idx]->persistent_gnt = persistent_gnt;
 890                         pr_debug("grant %u added to the tree of persistent grants, using %u/%u\n",
 891                                  persistent_gnt->gnt, blkif->persistent_gnt_c,
 892                                  xen_blkif_max_pgrants);
 893                         goto next;
 894                 }
 895                 if (use_persistent_gnts && !blkif->vbd.overflow_max_grants) {
 896                         blkif->vbd.overflow_max_grants = 1;
 897                         pr_debug("domain %u, device %#x is using maximum number of persistent grants\n",
 898                                  blkif->domid, blkif->vbd.handle);
 899                 }
 900                 /*
 901                  * We could not map this grant persistently, so use it as
 902                  * a non-persistent grant.
 903                  */
 904 next:
 905                 new_map_idx++;
 906         }
 907         segs_to_map = 0;
 908         last_map = map_until;
 909         if (map_until != num)
 910                 goto again;
 911
 912         return ret;
 913
 914 out_of_memory:
 915         pr_alert("%s: out of memory\n", __func__);
 916         put_free_pages(blkif, pages_to_gnt, segs_to_map);
 917         return -ENOMEM;
 918 }
 919
 920 static int xen_blkbk_map_seg(struct pending_req *pending_req)
 921 {
 922         int rc;
 923
 924         rc = xen_blkbk_map(pending_req->blkif, pending_req->segments,
 925                            pending_req->nr_segs,
 926                            (pending_req->operation != BLKIF_OP_READ));
 927
 928         return rc;
 929 }
 930
 931 static int xen_blkbk_parse_indirect(struct blkif_request *req,
 932                                     struct pending_req *pending_req,
 933                                     struct seg_buf seg[],
 934                                     struct phys_req *preq)
 935 {
 936         struct grant_page **pages = pending_req->indirect_pages;
 937         struct xen_blkif *blkif = pending_req->blkif;
 938         int indirect_grefs, rc, n, nseg, i;
 939         struct blkif_request_segment *segments = NULL;
 940
 941         nseg = pending_req->nr_segs;
 942         indirect_grefs = INDIRECT_PAGES(nseg);
 943         BUG_ON(indirect_grefs > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 944
 945         for (i = 0; i < indirect_grefs; i++)
 946                 pages[i]->gref = req->u.indirect.indirect_grefs[i];
 947
 948         rc = xen_blkbk_map(blkif, pages, indirect_grefs, true);
 949         if (rc)
 950                 goto unmap;
 951
 952         for (n = 0, i = 0; n < nseg; n++) {
 953                 if ((n % SEGS_PER_INDIRECT_FRAME) == 0) {
 954                         /* Map indirect segments */
 955                         if (segments)
 956                                 kunmap_atomic(segments);
 957                         segments = kmap_atomic(pages[n/SEGS_PER_INDIRECT_FRAME]->page);
 958                 }
 959                 i = n % SEGS_PER_INDIRECT_FRAME;
 960                 pending_req->segments[n]->gref = segments[i].gref;
 961                 seg[n].nsec = segments[i].last_sect -
 962                         segments[i].first_sect + 1;
 963                 seg[n].offset = (segments[i].first_sect << 9);
 964                 if ((segments[i].last_sect >= (PAGE_SIZE >> 9)) ||
 965                     (segments[i].last_sect < segments[i].first_sect)) {
 966                         rc = -EINVAL;
 967                         goto unmap;
 968                 }
 969                 preq->nr_sects += seg[n].nsec;
 970         }
 971
 972 unmap:
 973         if (segments)
 974                 kunmap_atomic(segments);
 975         xen_blkbk_unmap(blkif, pages, indirect_grefs);
 976         return rc;
 977 }
 978
 979 static int dispatch_discard_io(struct xen_blkif *blkif,
 980                                 struct blkif_request *req)
 981 {
 982         int err = 0;
 983         int status = BLKIF_RSP_OKAY;
 984         struct block_device *bdev = blkif->vbd.bdev;
 985         unsigned long secure;
 986         struct phys_req preq;
 987
 988         xen_blkif_get(blkif);
 989
 990         preq.sector_number = req->u.discard.sector_number;
 991         preq.nr_sects      = req->u.discard.nr_sectors;
 992
 993         err = xen_vbd_translate(&preq, blkif, WRITE);
 994         if (err) {
 995                 pr_warn("access denied: DISCARD [%llu->%llu] on dev=%04x\n",
 996                         preq.sector_number,
 997                         preq.sector_number + preq.nr_sects, blkif->vbd.pdevice);
 998                 goto fail_response;
 999         }
1000         blkif->st_ds_req++;
1001
1002         secure = (blkif->vbd.discard_secure &&
1003                  (req->u.discard.flag & BLKIF_DISCARD_SECURE)) ?
1004                  BLKDEV_DISCARD_SECURE : 0;
1005
1006         err = blkdev_issue_discard(bdev, req->u.discard.sector_number,
1007                                    req->u.discard.nr_sectors,
1008                                    GFP_KERNEL, secure);
1009 fail_response:
1010         if (err == -EOPNOTSUPP) {
1011                 pr_debug("discard op failed, not supported\n");
1012                 status = BLKIF_RSP_EOPNOTSUPP;
1013         } else if (err)
1014                 status = BLKIF_RSP_ERROR;
1015
1016         make_response(blkif, req->u.discard.id, req->operation, status);
1017         xen_blkif_put(blkif);
1018         return err;
1019 }
1020
1021 static int dispatch_other_io(struct xen_blkif *blkif,
1022                              struct blkif_request *req,
1023                              struct pending_req *pending_req)
1024 {
1025         free_req(blkif, pending_req);
1026         make_response(blkif, req->u.other.id, req->operation,
1027                       BLKIF_RSP_EOPNOTSUPP);
1028         return -EIO;
1029 }
1030
1031 static void xen_blk_drain_io(struct xen_blkif *blkif)
1032 {
1033         atomic_set(&blkif->drain, 1);
1034         do {
1035                 if (atomic_read(&blkif->inflight) == 0)
1036                         break;
1037                 wait_for_completion_interruptible_timeout(
1038                                 &blkif->drain_complete, HZ);
1039
1040                 if (!atomic_read(&blkif->drain))
1041                         break;
1042         } while (!kthread_should_stop());
1043         atomic_set(&blkif->drain, 0);
1044 }
1045
1046 /*
1047  * Completion callback on the bio's. Called as bh->b_end_io()
1048  */
1049
1050 static void __end_block_io_op(struct pending_req *pending_req, int error)
1051 {
1052         /* An error fails the entire request. */
1053         if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) &&
1054             (error == -EOPNOTSUPP)) {
1055                 pr_debug("flush diskcache op failed, not supported\n");
1056                 xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0);
1057                 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
1058         } else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) &&
1059                     (error == -EOPNOTSUPP)) {
1060                 pr_debug("write barrier op failed, not supported\n");
1061                 xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0);
1062                 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
1063         } else if (error) {
1064                 pr_debug("Buffer not up-to-date at end of operation,"
1065                          " error=%d\n", error);
1066                 pending_req->status = BLKIF_RSP_ERROR;
1067         }
1068
1069         /*
1070          * If all of the bio's have completed it is time to unmap
1071          * the grant references associated with 'request' and provide
1072          * the proper response on the ring.
1073          */
1074         if (atomic_dec_and_test(&pending_req->pendcnt))
1075                 xen_blkbk_unmap_and_respond(pending_req);
1076 }
1077
1078 /*
1079  * bio callback.
1080  */
1081 static void end_block_io_op(struct bio *bio, int error)
1082 {
1083         __end_block_io_op(bio->bi_private, error);
1084         bio_put(bio);
1085 }
1086
1087
1088
1089 /*
1090  * Function to copy the from the ring buffer the 'struct blkif_request'
1091  * (which has the sectors we want, number of them, grant references, etc),
1092  * and transmute  it to the block API to hand it over to the proper block disk.
1093  */
1094 static int
1095 __do_block_io_op(struct xen_blkif *blkif)
1096 {
1097         union blkif_back_rings *blk_rings = &blkif->blk_rings;
1098         struct blkif_request req;
1099         struct pending_req *pending_req;
1100         RING_IDX rc, rp;
1101         int more_to_do = 0;
1102
1103         rc = blk_rings->common.req_cons;
1104         rp = blk_rings->common.sring->req_prod;
1105         rmb(); /* Ensure we see queued requests up to 'rp'. */
1106
1107         if (RING_REQUEST_PROD_OVERFLOW(&blk_rings->common, rp)) {
1108                 rc = blk_rings->common.rsp_prod_pvt;
1109                 pr_warn("Frontend provided bogus ring requests (%d - %d = %d). Halting ring processing on dev=%04x\n",
1110                         rp, rc, rp - rc, blkif->vbd.pdevice);
1111                 return -EACCES;
1112         }
1113         while (rc != rp) {
1114
1115                 if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
1116                         break;
1117
1118                 if (kthread_should_stop()) {
1119                         more_to_do = 1;
1120                         break;
1121                 }
1122
1123                 pending_req = alloc_req(blkif);
1124                 if (NULL == pending_req) {
1125                         blkif->st_oo_req++;
1126                         more_to_do = 1;
1127                         break;
1128                 }
1129
1130                 switch (blkif->blk_protocol) {
1131                 case BLKIF_PROTOCOL_NATIVE:
1132                         memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
1133                         break;
1134                 case BLKIF_PROTOCOL_X86_32:
1135                         blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
1136                         break;
1137                 case BLKIF_PROTOCOL_X86_64:
1138                         blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
1139                         break;
1140                 default:
1141                         BUG();
1142                 }
1143                 blk_rings->common.req_cons = ++rc; /* before make_response() */
1144
1145                 /* Apply all sanity checks to /private copy/ of request. */
1146                 barrier();
1147
1148                 switch (req.operation) {
1149                 case BLKIF_OP_READ:
1150                 case BLKIF_OP_WRITE:
1151                 case BLKIF_OP_WRITE_BARRIER:
1152                 case BLKIF_OP_FLUSH_DISKCACHE:
1153                 case BLKIF_OP_INDIRECT:
1154                         if (dispatch_rw_block_io(blkif, &req, pending_req))
1155                                 goto done;
1156                         break;
1157                 case BLKIF_OP_DISCARD:
1158                         free_req(blkif, pending_req);
1159                         if (dispatch_discard_io(blkif, &req))
1160                                 goto done;
1161                         break;
1162                 default:
1163                         if (dispatch_other_io(blkif, &req, pending_req))
1164                                 goto done;
1165                         break;
1166                 }
1167
1168                 /* Yield point for this unbounded loop. */
1169                 cond_resched();
1170         }
1171 done:
1172         return more_to_do;
1173 }
1174
1175 static int
1176 do_block_io_op(struct xen_blkif *blkif)
1177 {
1178         union blkif_back_rings *blk_rings = &blkif->blk_rings;
1179         int more_to_do;
1180
1181         do {
1182                 more_to_do = __do_block_io_op(blkif);
1183                 if (more_to_do)
1184                         break;
1185
1186                 RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
1187         } while (more_to_do);
1188
1189         return more_to_do;
1190 }
1191 /*
1192  * Transmutation of the 'struct blkif_request' to a proper 'struct bio'
1193  * and call the 'submit_bio' to pass it to the underlying storage.
1194  */
1195 static int dispatch_rw_block_io(struct xen_blkif *blkif,
1196                                 struct blkif_request *req,
1197                                 struct pending_req *pending_req)
1198 {
1199         struct phys_req preq;
1200         struct seg_buf *seg = pending_req->seg;
1201         unsigned int nseg;
1202         struct bio *bio = NULL;
1203         struct bio **biolist = pending_req->biolist;
1204         int i, nbio = 0;
1205         int operation;
1206         struct blk_plug plug;
1207         bool drain = false;
1208         struct grant_page **pages = pending_req->segments;
1209         unsigned short req_operation;
1210
1211         req_operation = req->operation == BLKIF_OP_INDIRECT ?
1212                         req->u.indirect.indirect_op : req->operation;
1213         if ((req->operation == BLKIF_OP_INDIRECT) &&
1214             (req_operation != BLKIF_OP_READ) &&
1215             (req_operation != BLKIF_OP_WRITE)) {
1216                 pr_debug("Invalid indirect operation (%u)\n", req_operation);
1217                 goto fail_response;
1218         }
1219
1220         switch (req_operation) {
1221         case BLKIF_OP_READ:
1222                 blkif->st_rd_req++;
1223                 operation = READ;
1224                 break;
1225         case BLKIF_OP_WRITE:
1226                 blkif->st_wr_req++;
1227                 operation = WRITE_ODIRECT;
1228                 break;
1229         case BLKIF_OP_WRITE_BARRIER:
1230                 drain = true;
1231         case BLKIF_OP_FLUSH_DISKCACHE:
1232                 blkif->st_f_req++;
1233                 operation = WRITE_FLUSH;
1234                 break;
1235         default:
1236                 operation = 0; /* make gcc happy */
1237                 goto fail_response;
1238                 break;
1239         }
1240
1241         /* Check that the number of segments is sane. */
1242         nseg = req->operation == BLKIF_OP_INDIRECT ?
1243                req->u.indirect.nr_segments : req->u.rw.nr_segments;
1244
1245         if (unlikely(nseg == 0 && operation != WRITE_FLUSH) ||
1246             unlikely((req->operation != BLKIF_OP_INDIRECT) &&
1247                      (nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) ||
1248             unlikely((req->operation == BLKIF_OP_INDIRECT) &&
1249                      (nseg > MAX_INDIRECT_SEGMENTS))) {
1250                 pr_debug("Bad number of segments in request (%d)\n", nseg);
1251                 /* Haven't submitted any bio's yet. */
1252                 goto fail_response;
1253         }
1254
1255         preq.nr_sects      = 0;
1256
1257         pending_req->blkif     = blkif;
1258         pending_req->id        = req->u.rw.id;
1259         pending_req->operation = req_operation;
1260         pending_req->status    = BLKIF_RSP_OKAY;
1261         pending_req->nr_segs   = nseg;
1262
1263         if (req->operation != BLKIF_OP_INDIRECT) {
1264                 preq.dev               = req->u.rw.handle;
1265                 preq.sector_number     = req->u.rw.sector_number;
1266                 for (i = 0; i < nseg; i++) {
1267                         pages[i]->gref = req->u.rw.seg[i].gref;
1268                         seg[i].nsec = req->u.rw.seg[i].last_sect -
1269                                 req->u.rw.seg[i].first_sect + 1;
1270                         seg[i].offset = (req->u.rw.seg[i].first_sect << 9);
1271                         if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) ||
1272                             (req->u.rw.seg[i].last_sect <
1273                              req->u.rw.seg[i].first_sect))
1274                                 goto fail_response;
1275                         preq.nr_sects += seg[i].nsec;
1276                 }
1277         } else {
1278                 preq.dev               = req->u.indirect.handle;
1279                 preq.sector_number     = req->u.indirect.sector_number;
1280                 if (xen_blkbk_parse_indirect(req, pending_req, seg, &preq))
1281                         goto fail_response;
1282         }
1283
1284         if (xen_vbd_translate(&preq, blkif, operation) != 0) {
1285                 pr_debug("access denied: %s of [%llu,%llu] on dev=%04x\n",
1286                          operation == READ ? "read" : "write",
1287                          preq.sector_number,
1288                          preq.sector_number + preq.nr_sects,
1289                          blkif->vbd.pdevice);
1290                 goto fail_response;
1291         }
1292
1293         /*
1294          * This check _MUST_ be done after xen_vbd_translate as the preq.bdev
1295          * is set there.
1296          */
1297         for (i = 0; i < nseg; i++) {
1298                 if (((int)preq.sector_number|(int)seg[i].nsec) &
1299                     ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
1300                         pr_debug("Misaligned I/O request from domain %d\n",
1301                                  blkif->domid);
1302                         goto fail_response;
1303                 }
1304         }
1305
1306         /* Wait on all outstanding I/O's and once that has been completed
1307          * issue the WRITE_FLUSH.
1308          */
1309         if (drain)
1310                 xen_blk_drain_io(pending_req->blkif);
1311
1312         /*
1313          * If we have failed at this point, we need to undo the M2P override,
1314          * set gnttab_set_unmap_op on all of the grant references and perform
1315          * the hypercall to unmap the grants - that is all done in
1316          * xen_blkbk_unmap.
1317          */
1318         if (xen_blkbk_map_seg(pending_req))
1319                 goto fail_flush;
1320
1321         /*
1322          * This corresponding xen_blkif_put is done in __end_block_io_op, or
1323          * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
1324          */
1325         xen_blkif_get(blkif);
1326         atomic_inc(&blkif->inflight);
1327
1328         for (i = 0; i < nseg; i++) {
1329                 while ((bio == NULL) ||
1330                        (bio_add_page(bio,
1331                                      pages[i]->page,
1332                                      seg[i].nsec << 9,
1333                                      seg[i].offset) == 0)) {
1334
1335                         int nr_iovecs = min_t(int, (nseg-i), BIO_MAX_PAGES);
1336                         bio = bio_alloc(GFP_KERNEL, nr_iovecs);
1337                         if (unlikely(bio == NULL))
1338                                 goto fail_put_bio;
1339
1340                         biolist[nbio++] = bio;
1341                         bio->bi_bdev    = preq.bdev;
1342                         bio->bi_private = pending_req;
1343                         bio->bi_end_io  = end_block_io_op;
1344                         bio->bi_iter.bi_sector  = preq.sector_number;
1345                 }
1346
1347                 preq.sector_number += seg[i].nsec;
1348         }
1349
1350         /* This will be hit if the operation was a flush or discard. */
1351         if (!bio) {
1352                 BUG_ON(operation != WRITE_FLUSH);
1353
1354                 bio = bio_alloc(GFP_KERNEL, 0);
1355                 if (unlikely(bio == NULL))
1356                         goto fail_put_bio;
1357
1358                 biolist[nbio++] = bio;
1359                 bio->bi_bdev    = preq.bdev;
1360                 bio->bi_private = pending_req;
1361                 bio->bi_end_io  = end_block_io_op;
1362         }
1363
1364         atomic_set(&pending_req->pendcnt, nbio);
1365         blk_start_plug(&plug);
1366
1367         for (i = 0; i < nbio; i++)
1368                 submit_bio(operation, biolist[i]);
1369
1370         /* Let the I/Os go.. */
1371         blk_finish_plug(&plug);
1372
1373         if (operation == READ)
1374                 blkif->st_rd_sect += preq.nr_sects;
1375         else if (operation & WRITE)
1376                 blkif->st_wr_sect += preq.nr_sects;
1377
1378         return 0;
1379
1380  fail_flush:
1381         xen_blkbk_unmap(blkif, pending_req->segments,
1382                         pending_req->nr_segs);
1383  fail_response:
1384         /* Haven't submitted any bio's yet. */
1385         make_response(blkif, req->u.rw.id, req_operation, BLKIF_RSP_ERROR);
1386         free_req(blkif, pending_req);
1387         msleep(1); /* back off a bit */
1388         return -EIO;
1389
1390  fail_put_bio:
1391         for (i = 0; i < nbio; i++)
1392                 bio_put(biolist[i]);
1393         atomic_set(&pending_req->pendcnt, 1);
1394         __end_block_io_op(pending_req, -EINVAL);
1395         msleep(1); /* back off a bit */
1396         return -EIO;
1397 }
1398
1399
1400
1401 /*
1402  * Put a response on the ring on how the operation fared.
1403  */
1404 static void make_response(struct xen_blkif *blkif, u64 id,
1405                           unsigned short op, int st)
1406 {
1407         struct blkif_response  resp;
1408         unsigned long     flags;
1409         union blkif_back_rings *blk_rings = &blkif->blk_rings;
1410         int notify;
1411
1412         resp.id        = id;
1413         resp.operation = op;
1414         resp.status    = st;
1415
1416         spin_lock_irqsave(&blkif->blk_ring_lock, flags);
1417         /* Place on the response ring for the relevant domain. */
1418         switch (blkif->blk_protocol) {
1419         case BLKIF_PROTOCOL_NATIVE:
1420                 memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt),
1421                        &resp, sizeof(resp));
1422                 break;
1423         case BLKIF_PROTOCOL_X86_32:
1424                 memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt),
1425                        &resp, sizeof(resp));
1426                 break;
1427         case BLKIF_PROTOCOL_X86_64:
1428                 memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt),
1429                        &resp, sizeof(resp));
1430                 break;
1431         default:
1432                 BUG();
1433         }
1434         blk_rings->common.rsp_prod_pvt++;
1435         RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
1436         spin_unlock_irqrestore(&blkif->blk_ring_lock, flags);
1437         if (notify)
1438                 notify_remote_via_irq(blkif->irq);
1439 }
1440
1441 static int __init xen_blkif_init(void)
1442 {
1443         int rc = 0;
1444
1445         if (!xen_domain())
1446                 return -ENODEV;
1447
1448         if (xen_blkif_max_ring_order > XENBUS_MAX_RING_PAGE_ORDER) {
1449                 pr_info("Invalid max_ring_order (%d), will use default max: %d.\n",
1450                         xen_blkif_max_ring_order, XENBUS_MAX_RING_PAGE_ORDER);
1451                 xen_blkif_max_ring_order = XENBUS_MAX_RING_PAGE_ORDER;
1452         }
1453
1454         rc = xen_blkif_interface_init();
1455         if (rc)
1456                 goto failed_init;
1457
1458         rc = xen_blkif_xenbus_init();
1459         if (rc)
1460                 goto failed_init;
1461
1462  failed_init:
1463         return rc;
1464 }
1465
1466 module_init(xen_blkif_init);
1467
1468 MODULE_LICENSE("Dual BSD/GPL");
1469 MODULE_ALIAS("xen-backend:vbd");