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 #include <linux/spinlock.h>
  38 #include <linux/kthread.h>
  39 #include <linux/list.h>
  40 #include <linux/delay.h>
  41 #include <linux/freezer.h>
  42 #include <linux/loop.h>
  43 #include <linux/falloc.h>
  44 #include <linux/fs.h>
  45
  46 #include <xen/events.h>
  47 #include <xen/page.h>
  48 #include <asm/xen/hypervisor.h>
  49 #include <asm/xen/hypercall.h>
  50 #include "common.h"
  51
  52 /*
  53  * These are rather arbitrary. They are fairly large because adjacent requests
  54  * pulled from a communication ring are quite likely to end up being part of
  55  * the same scatter/gather request at the disc.
  56  *
  57  * ** TRY INCREASING 'xen_blkif_reqs' IF WRITE SPEEDS SEEM TOO LOW **
  58  *
  59  * This will increase the chances of being able to write whole tracks.
  60  * 64 should be enough to keep us competitive with Linux.
  61  */
  62 static int xen_blkif_reqs = 64;
  63 module_param_named(reqs, xen_blkif_reqs, int, 0);
  64 MODULE_PARM_DESC(reqs, "Number of blkback requests to allocate");
  65
  66 /* Run-time switchable: /sys/module/blkback/parameters/ */
  67 static unsigned int log_stats;
  68 module_param(log_stats, int, 0644);
  69
  70 /*
  71  * Each outstanding request that we've passed to the lower device layers has a
  72  * 'pending_req' allocated to it. Each buffer_head that completes decrements
  73  * the pendcnt towards zero. When it hits zero, the specified domain has a
  74  * response queued for it, with the saved 'id' passed back.
  75  */
  76 struct pending_req {
  77         struct xen_blkif        *blkif;
  78         u64                     id;
  79         int                     nr_pages;
  80         atomic_t                pendcnt;
  81         unsigned short          operation;
  82         int                     status;
  83         struct list_head        free_list;
  84 };
  85
  86 #define BLKBACK_INVALID_HANDLE (~0)
  87
  88 struct xen_blkbk {
  89         struct pending_req      *pending_reqs;
  90         /* List of all 'pending_req' available */
  91         struct list_head        pending_free;
  92         /* And its spinlock. */
  93         spinlock_t              pending_free_lock;
  94         wait_queue_head_t       pending_free_wq;
  95         /* The list of all pages that are available. */
  96         struct page             **pending_pages;
  97         /* And the grant handles that are available. */
  98         grant_handle_t          *pending_grant_handles;
  99 };
 100
 101 static struct xen_blkbk *blkbk;
 102
 103 /*
 104  * Little helpful macro to figure out the index and virtual address of the
 105  * pending_pages[..]. For each 'pending_req' we have have up to
 106  * BLKIF_MAX_SEGMENTS_PER_REQUEST (11) pages. The seg would be from 0 through
 107  * 10 and would index in the pending_pages[..].
 108  */
 109 static inline int vaddr_pagenr(struct pending_req *req, int seg)
 110 {
 111         return (req - blkbk->pending_reqs) *
 112                 BLKIF_MAX_SEGMENTS_PER_REQUEST + seg;
 113 }
 114
 115 #define pending_page(req, seg) pending_pages[vaddr_pagenr(req, seg)]
 116
 117 static inline unsigned long vaddr(struct pending_req *req, int seg)
 118 {
 119         unsigned long pfn = page_to_pfn(blkbk->pending_page(req, seg));
 120         return (unsigned long)pfn_to_kaddr(pfn);
 121 }
 122
 123 #define pending_handle(_req, _seg) \
 124         (blkbk->pending_grant_handles[vaddr_pagenr(_req, _seg)])
 125
 126
 127 static int do_block_io_op(struct xen_blkif *blkif);
 128 static int dispatch_rw_block_io(struct xen_blkif *blkif,
 129                                 struct blkif_request *req,
 130                                 struct pending_req *pending_req);
 131 static void make_response(struct xen_blkif *blkif, u64 id,
 132                           unsigned short op, int st);
 133
 134 /*
 135  * Retrieve from the 'pending_reqs' a free pending_req structure to be used.
 136  */
 137 static struct pending_req *alloc_req(void)
 138 {
 139         struct pending_req *req = NULL;
 140         unsigned long flags;
 141
 142         spin_lock_irqsave(&blkbk->pending_free_lock, flags);
 143         if (!list_empty(&blkbk->pending_free)) {
 144                 req = list_entry(blkbk->pending_free.next, struct pending_req,
 145                                  free_list);
 146                 list_del(&req->free_list);
 147         }
 148         spin_unlock_irqrestore(&blkbk->pending_free_lock, flags);
 149         return req;
 150 }
 151
 152 /*
 153  * Return the 'pending_req' structure back to the freepool. We also
 154  * wake up the thread if it was waiting for a free page.
 155  */
 156 static void free_req(struct pending_req *req)
 157 {
 158         unsigned long flags;
 159         int was_empty;
 160
 161         spin_lock_irqsave(&blkbk->pending_free_lock, flags);
 162         was_empty = list_empty(&blkbk->pending_free);
 163         list_add(&req->free_list, &blkbk->pending_free);
 164         spin_unlock_irqrestore(&blkbk->pending_free_lock, flags);
 165         if (was_empty)
 166                 wake_up(&blkbk->pending_free_wq);
 167 }
 168
 169 /*
 170  * Routines for managing virtual block devices (vbds).
 171  */
 172 static int xen_vbd_translate(struct phys_req *req, struct xen_blkif *blkif,
 173                              int operation)
 174 {
 175         struct xen_vbd *vbd = &blkif->vbd;
 176         int rc = -EACCES;
 177
 178         if ((operation != READ) && vbd->readonly)
 179                 goto out;
 180
 181         if (likely(req->nr_sects)) {
 182                 blkif_sector_t end = req->sector_number + req->nr_sects;
 183
 184                 if (unlikely(end < req->sector_number))
 185                         goto out;
 186                 if (unlikely(end > vbd_sz(vbd)))
 187                         goto out;
 188         }
 189
 190         req->dev  = vbd->pdevice;
 191         req->bdev = vbd->bdev;
 192         rc = 0;
 193
 194  out:
 195         return rc;
 196 }
 197
 198 static void xen_vbd_resize(struct xen_blkif *blkif)
 199 {
 200         struct xen_vbd *vbd = &blkif->vbd;
 201         struct xenbus_transaction xbt;
 202         int err;
 203         struct xenbus_device *dev = xen_blkbk_xenbus(blkif->be);
 204         unsigned long long new_size = vbd_sz(vbd);
 205
 206         pr_info(DRV_PFX "VBD Resize: Domid: %d, Device: (%d, %d)\n",
 207                 blkif->domid, MAJOR(vbd->pdevice), MINOR(vbd->pdevice));
 208         pr_info(DRV_PFX "VBD Resize: new size %llu\n", new_size);
 209         vbd->size = new_size;
 210 again:
 211         err = xenbus_transaction_start(&xbt);
 212         if (err) {
 213                 pr_warn(DRV_PFX "Error starting transaction");
 214                 return;
 215         }
 216         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 217                             (unsigned long long)vbd_sz(vbd));
 218         if (err) {
 219                 pr_warn(DRV_PFX "Error writing new size");
 220                 goto abort;
 221         }
 222         /*
 223          * Write the current state; we will use this to synchronize
 224          * the front-end. If the current state is "connected" the
 225          * front-end will get the new size information online.
 226          */
 227         err = xenbus_printf(xbt, dev->nodename, "state", "%d", dev->state);
 228         if (err) {
 229                 pr_warn(DRV_PFX "Error writing the state");
 230                 goto abort;
 231         }
 232
 233         err = xenbus_transaction_end(xbt, 0);
 234         if (err == -EAGAIN)
 235                 goto again;
 236         if (err)
 237                 pr_warn(DRV_PFX "Error ending transaction");
 238         return;
 239 abort:
 240         xenbus_transaction_end(xbt, 1);
 241 }
 242
 243 /*
 244  * Notification from the guest OS.
 245  */
 246 static void blkif_notify_work(struct xen_blkif *blkif)
 247 {
 248         blkif->waiting_reqs = 1;
 249         wake_up(&blkif->wq);
 250 }
 251
 252 irqreturn_t xen_blkif_be_int(int irq, void *dev_id)
 253 {
 254         blkif_notify_work(dev_id);
 255         return IRQ_HANDLED;
 256 }
 257
 258 /*
 259  * SCHEDULER FUNCTIONS
 260  */
 261
 262 static void print_stats(struct xen_blkif *blkif)
 263 {
 264         pr_info("xen-blkback (%s): oo %3d  |  rd %4d  |  wr %4d  |  f %4d"
 265                  "  |  ds %4d\n",
 266                  current->comm, blkif->st_oo_req,
 267                  blkif->st_rd_req, blkif->st_wr_req,
 268                  blkif->st_f_req, blkif->st_ds_req);
 269         blkif->st_print = jiffies + msecs_to_jiffies(10 * 1000);
 270         blkif->st_rd_req = 0;
 271         blkif->st_wr_req = 0;
 272         blkif->st_oo_req = 0;
 273         blkif->st_ds_req = 0;
 274 }
 275
 276 int xen_blkif_schedule(void *arg)
 277 {
 278         struct xen_blkif *blkif = arg;
 279         struct xen_vbd *vbd = &blkif->vbd;
 280
 281         xen_blkif_get(blkif);
 282
 283         while (!kthread_should_stop()) {
 284                 if (try_to_freeze())
 285                         continue;
 286                 if (unlikely(vbd->size != vbd_sz(vbd)))
 287                         xen_vbd_resize(blkif);
 288
 289                 wait_event_interruptible(
 290                         blkif->wq,
 291                         blkif->waiting_reqs || kthread_should_stop());
 292                 wait_event_interruptible(
 293                         blkbk->pending_free_wq,
 294                         !list_empty(&blkbk->pending_free) ||
 295                         kthread_should_stop());
 296
 297                 blkif->waiting_reqs = 0;
 298                 smp_mb(); /* clear flag *before* checking for work */
 299
 300                 if (do_block_io_op(blkif))
 301                         blkif->waiting_reqs = 1;
 302
 303                 if (log_stats && time_after(jiffies, blkif->st_print))
 304                         print_stats(blkif);
 305         }
 306
 307         if (log_stats)
 308                 print_stats(blkif);
 309
 310         blkif->xenblkd = NULL;
 311         xen_blkif_put(blkif);
 312
 313         return 0;
 314 }
 315
 316 struct seg_buf {
 317         unsigned long buf;
 318         unsigned int nsec;
 319 };
 320 /*
 321  * Unmap the grant references, and also remove the M2P over-rides
 322  * used in the 'pending_req'.
 323  */
 324 static void xen_blkbk_unmap(struct pending_req *req)
 325 {
 326         struct gnttab_unmap_grant_ref unmap[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 327         unsigned int i, invcount = 0;
 328         grant_handle_t handle;
 329         int ret;
 330
 331         for (i = 0; i < req->nr_pages; i++) {
 332                 handle = pending_handle(req, i);
 333                 if (handle == BLKBACK_INVALID_HANDLE)
 334                         continue;
 335                 gnttab_set_unmap_op(&unmap[invcount], vaddr(req, i),
 336                                     GNTMAP_host_map, handle);
 337                 pending_handle(req, i) = BLKBACK_INVALID_HANDLE;
 338                 invcount++;
 339         }
 340
 341         ret = HYPERVISOR_grant_table_op(
 342                 GNTTABOP_unmap_grant_ref, unmap, invcount);
 343         BUG_ON(ret);
 344         /*
 345          * Note, we use invcount, so nr->pages, so we can't index
 346          * using vaddr(req, i).
 347          */
 348         for (i = 0; i < invcount; i++) {
 349                 ret = m2p_remove_override(
 350                         virt_to_page(unmap[i].host_addr), false);
 351                 if (ret) {
 352                         pr_alert(DRV_PFX "Failed to remove M2P override for %lx\n",
 353                                  (unsigned long)unmap[i].host_addr);
 354                         continue;
 355                 }
 356         }
 357 }
 358
 359 static int xen_blkbk_map(struct blkif_request *req,
 360                          struct pending_req *pending_req,
 361                          struct seg_buf seg[])
 362 {
 363         struct gnttab_map_grant_ref map[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 364         int i;
 365         int nseg = req->nr_segments;
 366         int ret = 0;
 367
 368         /*
 369          * Fill out preq.nr_sects with proper amount of sectors, and setup
 370          * assign map[..] with the PFN of the page in our domain with the
 371          * corresponding grant reference for each page.
 372          */
 373         for (i = 0; i < nseg; i++) {
 374                 uint32_t flags;
 375
 376                 flags = GNTMAP_host_map;
 377                 if (pending_req->operation != BLKIF_OP_READ)
 378                         flags |= GNTMAP_readonly;
 379                 gnttab_set_map_op(&map[i], vaddr(pending_req, i), flags,
 380                                   req->u.rw.seg[i].gref,
 381                                   pending_req->blkif->domid);
 382         }
 383
 384         ret = HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, map, nseg);
 385         BUG_ON(ret);
 386
 387         /*
 388          * Now swizzle the MFN in our domain with the MFN from the other domain
 389          * so that when we access vaddr(pending_req,i) it has the contents of
 390          * the page from the other domain.
 391          */
 392         for (i = 0; i < nseg; i++) {
 393                 if (unlikely(map[i].status != 0)) {
 394                         pr_debug(DRV_PFX "invalid buffer -- could not remap it\n");
 395                         map[i].handle = BLKBACK_INVALID_HANDLE;
 396                         ret |= 1;
 397                 }
 398
 399                 pending_handle(pending_req, i) = map[i].handle;
 400
 401                 if (ret)
 402                         continue;
 403
 404                 ret = m2p_add_override(PFN_DOWN(map[i].dev_bus_addr),
 405                         blkbk->pending_page(pending_req, i), NULL);
 406                 if (ret) {
 407                         pr_alert(DRV_PFX "Failed to install M2P override for %lx (ret: %d)\n",
 408                                  (unsigned long)map[i].dev_bus_addr, ret);
 409                         /* We could switch over to GNTTABOP_copy */
 410                         continue;
 411                 }
 412
 413                 seg[i].buf  = map[i].dev_bus_addr |
 414                         (req->u.rw.seg[i].first_sect << 9);
 415         }
 416         return ret;
 417 }
 418
 419 static void xen_blk_discard(struct xen_blkif *blkif, struct blkif_request *req)
 420 {
 421         int err = 0;
 422         int status = BLKIF_RSP_OKAY;
 423         struct block_device *bdev = blkif->vbd.bdev;
 424
 425         if (blkif->blk_backend_type == BLKIF_BACKEND_PHY)
 426                 /* just forward the discard request */
 427                 err = blkdev_issue_discard(bdev,
 428                                 req->u.discard.sector_number,
 429                                 req->u.discard.nr_sectors,
 430                                 GFP_KERNEL, 0);
 431         else if (blkif->blk_backend_type == BLKIF_BACKEND_FILE) {
 432                 /* punch a hole in the backing file */
 433                 struct loop_device *lo = bdev->bd_disk->private_data;
 434                 struct file *file = lo->lo_backing_file;
 435
 436                 if (file->f_op->fallocate)
 437                         err = file->f_op->fallocate(file,
 438                                 FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
 439                                 req->u.discard.sector_number << 9,
 440                                 req->u.discard.nr_sectors << 9);
 441                 else
 442                         err = -EOPNOTSUPP;
 443         } else
 444                 err = -EOPNOTSUPP;
 445
 446         if (err == -EOPNOTSUPP) {
 447                 pr_debug(DRV_PFX "discard op failed, not supported\n");
 448                 status = BLKIF_RSP_EOPNOTSUPP;
 449         } else if (err)
 450                 status = BLKIF_RSP_ERROR;
 451
 452         make_response(blkif, req->id, req->operation, status);
 453 }
 454
 455 static void xen_blk_drain_io(struct xen_blkif *blkif)
 456 {
 457         atomic_set(&blkif->drain, 1);
 458         do {
 459                 /* The initial value is one, and one refcnt taken at the
 460                  * start of the xen_blkif_schedule thread. */
 461                 if (atomic_read(&blkif->refcnt) <= 2)
 462                         break;
 463                 wait_for_completion_interruptible_timeout(
 464                                 &blkif->drain_complete, HZ);
 465
 466                 if (!atomic_read(&blkif->drain))
 467                         break;
 468         } while (!kthread_should_stop());
 469         atomic_set(&blkif->drain, 0);
 470 }
 471
 472 /*
 473  * Completion callback on the bio's. Called as bh->b_end_io()
 474  */
 475
 476 static void __end_block_io_op(struct pending_req *pending_req, int error)
 477 {
 478         /* An error fails the entire request. */
 479         if ((pending_req->operation == BLKIF_OP_FLUSH_DISKCACHE) &&
 480             (error == -EOPNOTSUPP)) {
 481                 pr_debug(DRV_PFX "flush diskcache op failed, not supported\n");
 482                 xen_blkbk_flush_diskcache(XBT_NIL, pending_req->blkif->be, 0);
 483                 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
 484         } else if ((pending_req->operation == BLKIF_OP_WRITE_BARRIER) &&
 485                     (error == -EOPNOTSUPP)) {
 486                 pr_debug(DRV_PFX "write barrier op failed, not supported\n");
 487                 xen_blkbk_barrier(XBT_NIL, pending_req->blkif->be, 0);
 488                 pending_req->status = BLKIF_RSP_EOPNOTSUPP;
 489         } else if (error) {
 490                 pr_debug(DRV_PFX "Buffer not up-to-date at end of operation,"
 491                          " error=%d\n", error);
 492                 pending_req->status = BLKIF_RSP_ERROR;
 493         }
 494
 495         /*
 496          * If all of the bio's have completed it is time to unmap
 497          * the grant references associated with 'request' and provide
 498          * the proper response on the ring.
 499          */
 500         if (atomic_dec_and_test(&pending_req->pendcnt)) {
 501                 xen_blkbk_unmap(pending_req);
 502                 make_response(pending_req->blkif, pending_req->id,
 503                               pending_req->operation, pending_req->status);
 504                 xen_blkif_put(pending_req->blkif);
 505                 if (atomic_read(&pending_req->blkif->refcnt) <= 2) {
 506                         if (atomic_read(&pending_req->blkif->drain))
 507                                 complete(&pending_req->blkif->drain_complete);
 508                 }
 509                 free_req(pending_req);
 510         }
 511 }
 512
 513 /*
 514  * bio callback.
 515  */
 516 static void end_block_io_op(struct bio *bio, int error)
 517 {
 518         __end_block_io_op(bio->bi_private, error);
 519         bio_put(bio);
 520 }
 521
 522
 523
 524 /*
 525  * Function to copy the from the ring buffer the 'struct blkif_request'
 526  * (which has the sectors we want, number of them, grant references, etc),
 527  * and transmute  it to the block API to hand it over to the proper block disk.
 528  */
 529 static int
 530 __do_block_io_op(struct xen_blkif *blkif)
 531 {
 532         union blkif_back_rings *blk_rings = &blkif->blk_rings;
 533         struct blkif_request req;
 534         struct pending_req *pending_req;
 535         RING_IDX rc, rp;
 536         int more_to_do = 0;
 537
 538         rc = blk_rings->common.req_cons;
 539         rp = blk_rings->common.sring->req_prod;
 540         rmb(); /* Ensure we see queued requests up to 'rp'. */
 541
 542         while (rc != rp) {
 543
 544                 if (RING_REQUEST_CONS_OVERFLOW(&blk_rings->common, rc))
 545                         break;
 546
 547                 if (kthread_should_stop()) {
 548                         more_to_do = 1;
 549                         break;
 550                 }
 551
 552                 pending_req = alloc_req();
 553                 if (NULL == pending_req) {
 554                         blkif->st_oo_req++;
 555                         more_to_do = 1;
 556                         break;
 557                 }
 558
 559                 switch (blkif->blk_protocol) {
 560                 case BLKIF_PROTOCOL_NATIVE:
 561                         memcpy(&req, RING_GET_REQUEST(&blk_rings->native, rc), sizeof(req));
 562                         break;
 563                 case BLKIF_PROTOCOL_X86_32:
 564                         blkif_get_x86_32_req(&req, RING_GET_REQUEST(&blk_rings->x86_32, rc));
 565                         break;
 566                 case BLKIF_PROTOCOL_X86_64:
 567                         blkif_get_x86_64_req(&req, RING_GET_REQUEST(&blk_rings->x86_64, rc));
 568                         break;
 569                 default:
 570                         BUG();
 571                 }
 572                 blk_rings->common.req_cons = ++rc; /* before make_response() */
 573
 574                 /* Apply all sanity checks to /private copy/ of request. */
 575                 barrier();
 576
 577                 if (dispatch_rw_block_io(blkif, &req, pending_req))
 578                         break;
 579
 580                 /* Yield point for this unbounded loop. */
 581                 cond_resched();
 582         }
 583
 584         return more_to_do;
 585 }
 586
 587 static int
 588 do_block_io_op(struct xen_blkif *blkif)
 589 {
 590         union blkif_back_rings *blk_rings = &blkif->blk_rings;
 591         int more_to_do;
 592
 593         do {
 594                 more_to_do = __do_block_io_op(blkif);
 595                 if (more_to_do)
 596                         break;
 597
 598                 RING_FINAL_CHECK_FOR_REQUESTS(&blk_rings->common, more_to_do);
 599         } while (more_to_do);
 600
 601         return more_to_do;
 602 }
 603 /*
 604  * Transmutation of the 'struct blkif_request' to a proper 'struct bio'
 605  * and call the 'submit_bio' to pass it to the underlying storage.
 606  */
 607 static int dispatch_rw_block_io(struct xen_blkif *blkif,
 608                                 struct blkif_request *req,
 609                                 struct pending_req *pending_req)
 610 {
 611         struct phys_req preq;
 612         struct seg_buf seg[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 613         unsigned int nseg;
 614         struct bio *bio = NULL;
 615         struct bio *biolist[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 616         int i, nbio = 0;
 617         int operation;
 618         struct blk_plug plug;
 619         bool drain = false;
 620
 621         switch (req->operation) {
 622         case BLKIF_OP_READ:
 623                 blkif->st_rd_req++;
 624                 operation = READ;
 625                 break;
 626         case BLKIF_OP_WRITE:
 627                 blkif->st_wr_req++;
 628                 operation = WRITE_ODIRECT;
 629                 break;
 630         case BLKIF_OP_WRITE_BARRIER:
 631                 drain = true;
 632         case BLKIF_OP_FLUSH_DISKCACHE:
 633                 blkif->st_f_req++;
 634                 operation = WRITE_FLUSH;
 635                 break;
 636         case BLKIF_OP_DISCARD:
 637                 blkif->st_ds_req++;
 638                 operation = REQ_DISCARD;
 639                 break;
 640         default:
 641                 operation = 0; /* make gcc happy */
 642                 goto fail_response;
 643                 break;
 644         }
 645
 646         /* Check that the number of segments is sane. */
 647         nseg = req->nr_segments;
 648         if (unlikely(nseg == 0 && operation != WRITE_FLUSH &&
 649                                 operation != REQ_DISCARD) ||
 650             unlikely(nseg > BLKIF_MAX_SEGMENTS_PER_REQUEST)) {
 651                 pr_debug(DRV_PFX "Bad number of segments in request (%d)\n",
 652                          nseg);
 653                 /* Haven't submitted any bio's yet. */
 654                 goto fail_response;
 655         }
 656
 657         preq.dev           = req->handle;
 658         preq.sector_number = req->u.rw.sector_number;
 659         preq.nr_sects      = 0;
 660
 661         pending_req->blkif     = blkif;
 662         pending_req->id        = req->id;
 663         pending_req->operation = req->operation;
 664         pending_req->status    = BLKIF_RSP_OKAY;
 665         pending_req->nr_pages  = nseg;
 666
 667         for (i = 0; i < nseg; i++) {
 668                 seg[i].nsec = req->u.rw.seg[i].last_sect -
 669                         req->u.rw.seg[i].first_sect + 1;
 670                 if ((req->u.rw.seg[i].last_sect >= (PAGE_SIZE >> 9)) ||
 671                     (req->u.rw.seg[i].last_sect < req->u.rw.seg[i].first_sect))
 672                         goto fail_response;
 673                 preq.nr_sects += seg[i].nsec;
 674
 675         }
 676
 677         if (xen_vbd_translate(&preq, blkif, operation) != 0) {
 678                 pr_debug(DRV_PFX "access denied: %s of [%llu,%llu] on dev=%04x\n",
 679                          operation == READ ? "read" : "write",
 680                          preq.sector_number,
 681                          preq.sector_number + preq.nr_sects, preq.dev);
 682                 goto fail_response;
 683         }
 684
 685         /*
 686          * This check _MUST_ be done after xen_vbd_translate as the preq.bdev
 687          * is set there.
 688          */
 689         for (i = 0; i < nseg; i++) {
 690                 if (((int)preq.sector_number|(int)seg[i].nsec) &
 691                     ((bdev_logical_block_size(preq.bdev) >> 9) - 1)) {
 692                         pr_debug(DRV_PFX "Misaligned I/O request from domain %d",
 693                                  blkif->domid);
 694                         goto fail_response;
 695                 }
 696         }
 697
 698         /* Wait on all outstanding I/O's and once that has been completed
 699          * issue the WRITE_FLUSH.
 700          */
 701         if (drain)
 702                 xen_blk_drain_io(pending_req->blkif);
 703
 704         /*
 705          * If we have failed at this point, we need to undo the M2P override,
 706          * set gnttab_set_unmap_op on all of the grant references and perform
 707          * the hypercall to unmap the grants - that is all done in
 708          * xen_blkbk_unmap.
 709          */
 710         if (operation != REQ_DISCARD && xen_blkbk_map(req, pending_req, seg))
 711                 goto fail_flush;
 712
 713         /*
 714          * This corresponding xen_blkif_put is done in __end_block_io_op, or
 715          * below (in "!bio") if we are handling a BLKIF_OP_DISCARD.
 716          */
 717         xen_blkif_get(blkif);
 718
 719         for (i = 0; i < nseg; i++) {
 720                 while ((bio == NULL) ||
 721                        (bio_add_page(bio,
 722                                      blkbk->pending_page(pending_req, i),
 723                                      seg[i].nsec << 9,
 724                                      seg[i].buf & ~PAGE_MASK) == 0)) {
 725
 726                         bio = bio_alloc(GFP_KERNEL, nseg-i);
 727                         if (unlikely(bio == NULL))
 728                                 goto fail_put_bio;
 729
 730                         biolist[nbio++] = bio;
 731                         bio->bi_bdev    = preq.bdev;
 732                         bio->bi_private = pending_req;
 733                         bio->bi_end_io  = end_block_io_op;
 734                         bio->bi_sector  = preq.sector_number;
 735                 }
 736
 737                 preq.sector_number += seg[i].nsec;
 738         }
 739
 740         /* This will be hit if the operation was a flush or discard. */
 741         if (!bio) {
 742                 BUG_ON(operation != WRITE_FLUSH && operation != REQ_DISCARD);
 743
 744                 if (operation == WRITE_FLUSH) {
 745                         bio = bio_alloc(GFP_KERNEL, 0);
 746                         if (unlikely(bio == NULL))
 747                                 goto fail_put_bio;
 748
 749                         biolist[nbio++] = bio;
 750                         bio->bi_bdev    = preq.bdev;
 751                         bio->bi_private = pending_req;
 752                         bio->bi_end_io  = end_block_io_op;
 753                 } else if (operation == REQ_DISCARD) {
 754                         xen_blk_discard(blkif, req);
 755                         xen_blkif_put(blkif);
 756                         free_req(pending_req);
 757                         return 0;
 758                 }
 759         }
 760
 761         /*
 762          * We set it one so that the last submit_bio does not have to call
 763          * atomic_inc.
 764          */
 765         atomic_set(&pending_req->pendcnt, nbio);
 766
 767         /* Get a reference count for the disk queue and start sending I/O */
 768         blk_start_plug(&plug);
 769
 770         for (i = 0; i < nbio; i++)
 771                 submit_bio(operation, biolist[i]);
 772
 773         /* Let the I/Os go.. */
 774         blk_finish_plug(&plug);
 775
 776         if (operation == READ)
 777                 blkif->st_rd_sect += preq.nr_sects;
 778         else if (operation & WRITE)
 779                 blkif->st_wr_sect += preq.nr_sects;
 780
 781         return 0;
 782
 783  fail_flush:
 784         xen_blkbk_unmap(pending_req);
 785  fail_response:
 786         /* Haven't submitted any bio's yet. */
 787         make_response(blkif, req->id, req->operation, BLKIF_RSP_ERROR);
 788         free_req(pending_req);
 789         msleep(1); /* back off a bit */
 790         return -EIO;
 791
 792  fail_put_bio:
 793         for (i = 0; i < nbio; i++)
 794                 bio_put(biolist[i]);
 795         __end_block_io_op(pending_req, -EINVAL);
 796         msleep(1); /* back off a bit */
 797         return -EIO;
 798 }
 799
 800
 801
 802 /*
 803  * Put a response on the ring on how the operation fared.
 804  */
 805 static void make_response(struct xen_blkif *blkif, u64 id,
 806                           unsigned short op, int st)
 807 {
 808         struct blkif_response  resp;
 809         unsigned long     flags;
 810         union blkif_back_rings *blk_rings = &blkif->blk_rings;
 811         int notify;
 812
 813         resp.id        = id;
 814         resp.operation = op;
 815         resp.status    = st;
 816
 817         spin_lock_irqsave(&blkif->blk_ring_lock, flags);
 818         /* Place on the response ring for the relevant domain. */
 819         switch (blkif->blk_protocol) {
 820         case BLKIF_PROTOCOL_NATIVE:
 821                 memcpy(RING_GET_RESPONSE(&blk_rings->native, blk_rings->native.rsp_prod_pvt),
 822                        &resp, sizeof(resp));
 823                 break;
 824         case BLKIF_PROTOCOL_X86_32:
 825                 memcpy(RING_GET_RESPONSE(&blk_rings->x86_32, blk_rings->x86_32.rsp_prod_pvt),
 826                        &resp, sizeof(resp));
 827                 break;
 828         case BLKIF_PROTOCOL_X86_64:
 829                 memcpy(RING_GET_RESPONSE(&blk_rings->x86_64, blk_rings->x86_64.rsp_prod_pvt),
 830                        &resp, sizeof(resp));
 831                 break;
 832         default:
 833                 BUG();
 834         }
 835         blk_rings->common.rsp_prod_pvt++;
 836         RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blk_rings->common, notify);
 837         spin_unlock_irqrestore(&blkif->blk_ring_lock, flags);
 838         if (notify)
 839                 notify_remote_via_irq(blkif->irq);
 840 }
 841
 842 static int __init xen_blkif_init(void)
 843 {
 844         int i, mmap_pages;
 845         int rc = 0;
 846
 847         if (!xen_pv_domain())
 848                 return -ENODEV;
 849
 850         blkbk = kzalloc(sizeof(struct xen_blkbk), GFP_KERNEL);
 851         if (!blkbk) {
 852                 pr_alert(DRV_PFX "%s: out of memory!\n", __func__);
 853                 return -ENOMEM;
 854         }
 855
 856         mmap_pages = xen_blkif_reqs * BLKIF_MAX_SEGMENTS_PER_REQUEST;
 857
 858         blkbk->pending_reqs          = kzalloc(sizeof(blkbk->pending_reqs[0]) *
 859                                         xen_blkif_reqs, GFP_KERNEL);
 860         blkbk->pending_grant_handles = kmalloc(sizeof(blkbk->pending_grant_handles[0]) *
 861                                         mmap_pages, GFP_KERNEL);
 862         blkbk->pending_pages         = kzalloc(sizeof(blkbk->pending_pages[0]) *
 863                                         mmap_pages, GFP_KERNEL);
 864
 865         if (!blkbk->pending_reqs || !blkbk->pending_grant_handles ||
 866             !blkbk->pending_pages) {
 867                 rc = -ENOMEM;
 868                 goto out_of_memory;
 869         }
 870
 871         for (i = 0; i < mmap_pages; i++) {
 872                 blkbk->pending_grant_handles[i] = BLKBACK_INVALID_HANDLE;
 873                 blkbk->pending_pages[i] = alloc_page(GFP_KERNEL);
 874                 if (blkbk->pending_pages[i] == NULL) {
 875                         rc = -ENOMEM;
 876                         goto out_of_memory;
 877                 }
 878         }
 879         rc = xen_blkif_interface_init();
 880         if (rc)
 881                 goto failed_init;
 882
 883         INIT_LIST_HEAD(&blkbk->pending_free);
 884         spin_lock_init(&blkbk->pending_free_lock);
 885         init_waitqueue_head(&blkbk->pending_free_wq);
 886
 887         for (i = 0; i < xen_blkif_reqs; i++)
 888                 list_add_tail(&blkbk->pending_reqs[i].free_list,
 889                               &blkbk->pending_free);
 890
 891         rc = xen_blkif_xenbus_init();
 892         if (rc)
 893                 goto failed_init;
 894
 895         return 0;
 896
 897  out_of_memory:
 898         pr_alert(DRV_PFX "%s: out of memory\n", __func__);
 899  failed_init:
 900         kfree(blkbk->pending_reqs);
 901         kfree(blkbk->pending_grant_handles);
 902         if (blkbk->pending_pages) {
 903                 for (i = 0; i < mmap_pages; i++) {
 904                         if (blkbk->pending_pages[i])
 905                                 __free_page(blkbk->pending_pages[i]);
 906                 }
 907                 kfree(blkbk->pending_pages);
 908         }
 909         kfree(blkbk);
 910         blkbk = NULL;
 911         return rc;
 912 }
 913
 914 module_init(xen_blkif_init);
 915
 916 MODULE_LICENSE("Dual BSD/GPL");
 917 MODULE_ALIAS("xen-backend:vbd");