drivers/xen/xenbus/xenbus_client.c

   1 /******************************************************************************
   2  * Client-facing interface for the Xenbus driver.  In other words, the
   3  * interface between the Xenbus and the device-specific code, be it the
   4  * frontend or the backend of that driver.
   5  *
   6  * Copyright (C) 2005 XenSource Ltd
   7  *
   8  * This program is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU General Public License version 2
  10  * as published by the Free Software Foundation; or, when distributed
  11  * separately from the Linux kernel or incorporated into other
  12  * software packages, subject to the following license:
  13  *
  14  * Permission is hereby granted, free of charge, to any person obtaining a copy
  15  * of this source file (the "Software"), to deal in the Software without
  16  * restriction, including without limitation the rights to use, copy, modify,
  17  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  18  * and to permit persons to whom the Software is furnished to do so, subject to
  19  * the following conditions:
  20  *
  21  * The above copyright notice and this permission notice shall be included in
  22  * all copies or substantial portions of the Software.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  25  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  26  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  27  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  28  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  29  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  30  * IN THE SOFTWARE.
  31  */
  32
  33 #include <linux/mm.h>
  34 #include <linux/slab.h>
  35 #include <linux/types.h>
  36 #include <linux/spinlock.h>
  37 #include <linux/vmalloc.h>
  38 #include <linux/export.h>
  39 #include <asm/xen/hypervisor.h>
  40 #include <xen/page.h>
  41 #include <xen/interface/xen.h>
  42 #include <xen/interface/event_channel.h>
  43 #include <xen/balloon.h>
  44 #include <xen/events.h>
  45 #include <xen/grant_table.h>
  46 #include <xen/xenbus.h>
  47 #include <xen/xen.h>
  48 #include <xen/features.h>
  49
  50 #include "xenbus.h"
  51
  52 #define XENBUS_PAGES(_grants)   (DIV_ROUND_UP(_grants, XEN_PFN_PER_PAGE))
  53
  54 #define XENBUS_MAX_RING_PAGES   (XENBUS_PAGES(XENBUS_MAX_RING_GRANTS))
  55
  56 struct xenbus_map_node {
  57         struct list_head next;
  58         union {
  59                 struct {
  60                         struct vm_struct *area;
  61                 } pv;
  62                 struct {
  63                         struct page *pages[XENBUS_MAX_RING_PAGES];
  64                         unsigned long addrs[XENBUS_MAX_RING_GRANTS];
  65                         void *addr;
  66                 } hvm;
  67         };
  68         grant_handle_t handles[XENBUS_MAX_RING_GRANTS];
  69         unsigned int   nr_handles;
  70 };
  71
  72 static DEFINE_SPINLOCK(xenbus_valloc_lock);
  73 static LIST_HEAD(xenbus_valloc_pages);
  74
  75 struct xenbus_ring_ops {
  76         int (*map)(struct xenbus_device *dev,
  77                    grant_ref_t *gnt_refs, unsigned int nr_grefs,
  78                    void **vaddr);
  79         int (*unmap)(struct xenbus_device *dev, void *vaddr);
  80 };
  81
  82 static const struct xenbus_ring_ops *ring_ops __read_mostly;
  83
  84 const char *xenbus_strstate(enum xenbus_state state)
  85 {
  86         static const char *const name[] = {
  87                 [ XenbusStateUnknown      ] = "Unknown",
  88                 [ XenbusStateInitialising ] = "Initialising",
  89                 [ XenbusStateInitWait     ] = "InitWait",
  90                 [ XenbusStateInitialised  ] = "Initialised",
  91                 [ XenbusStateConnected    ] = "Connected",
  92                 [ XenbusStateClosing      ] = "Closing",
  93                 [ XenbusStateClosed       ] = "Closed",
  94                 [XenbusStateReconfiguring] = "Reconfiguring",
  95                 [XenbusStateReconfigured] = "Reconfigured",
  96         };
  97         return (state < ARRAY_SIZE(name)) ? name[state] : "INVALID";
  98 }
  99 EXPORT_SYMBOL_GPL(xenbus_strstate);
 100
 101 /**
 102  * xenbus_watch_path - register a watch
 103  * @dev: xenbus device
 104  * @path: path to watch
 105  * @watch: watch to register
 106  * @callback: callback to register
 107  *
 108  * Register a @watch on the given path, using the given xenbus_watch structure
 109  * for storage, and the given @callback function as the callback.  Return 0 on
 110  * success, or -errno on error.  On success, the given @path will be saved as
 111  * @watch->node, and remains the caller's to free.  On error, @watch->node will
 112  * be NULL, the device will switch to %XenbusStateClosing, and the error will
 113  * be saved in the store.
 114  */
 115 int xenbus_watch_path(struct xenbus_device *dev, const char *path,
 116                       struct xenbus_watch *watch,
 117                       void (*callback)(struct xenbus_watch *,
 118                                        const char *, const char *))
 119 {
 120         int err;
 121
 122         watch->node = path;
 123         watch->callback = callback;
 124
 125         err = register_xenbus_watch(watch);
 126
 127         if (err) {
 128                 watch->node = NULL;
 129                 watch->callback = NULL;
 130                 xenbus_dev_fatal(dev, err, "adding watch on %s", path);
 131         }
 132
 133         return err;
 134 }
 135 EXPORT_SYMBOL_GPL(xenbus_watch_path);
 136
 137
 138 /**
 139  * xenbus_watch_pathfmt - register a watch on a sprintf-formatted path
 140  * @dev: xenbus device
 141  * @watch: watch to register
 142  * @callback: callback to register
 143  * @pathfmt: format of path to watch
 144  *
 145  * Register a watch on the given @path, using the given xenbus_watch
 146  * structure for storage, and the given @callback function as the callback.
 147  * Return 0 on success, or -errno on error.  On success, the watched path
 148  * (@path/@path2) will be saved as @watch->node, and becomes the caller's to
 149  * kfree().  On error, watch->node will be NULL, so the caller has nothing to
 150  * free, the device will switch to %XenbusStateClosing, and the error will be
 151  * saved in the store.
 152  */
 153 int xenbus_watch_pathfmt(struct xenbus_device *dev,
 154                          struct xenbus_watch *watch,
 155                          void (*callback)(struct xenbus_watch *,
 156                                           const char *, const char *),
 157                          const char *pathfmt, ...)
 158 {
 159         int err;
 160         va_list ap;
 161         char *path;
 162
 163         va_start(ap, pathfmt);
 164         path = kvasprintf(GFP_NOIO | __GFP_HIGH, pathfmt, ap);
 165         va_end(ap);
 166
 167         if (!path) {
 168                 xenbus_dev_fatal(dev, -ENOMEM, "allocating path for watch");
 169                 return -ENOMEM;
 170         }
 171         err = xenbus_watch_path(dev, path, watch, callback);
 172
 173         if (err)
 174                 kfree(path);
 175         return err;
 176 }
 177 EXPORT_SYMBOL_GPL(xenbus_watch_pathfmt);
 178
 179 static void xenbus_switch_fatal(struct xenbus_device *, int, int,
 180                                 const char *, ...);
 181
 182 static int
 183 __xenbus_switch_state(struct xenbus_device *dev,
 184                       enum xenbus_state state, int depth)
 185 {
 186         /* We check whether the state is currently set to the given value, and
 187            if not, then the state is set.  We don't want to unconditionally
 188            write the given state, because we don't want to fire watches
 189            unnecessarily.  Furthermore, if the node has gone, we don't write
 190            to it, as the device will be tearing down, and we don't want to
 191            resurrect that directory.
 192
 193            Note that, because of this cached value of our state, this
 194            function will not take a caller's Xenstore transaction
 195            (something it was trying to in the past) because dev->state
 196            would not get reset if the transaction was aborted.
 197          */
 198
 199         struct xenbus_transaction xbt;
 200         int current_state;
 201         int err, abort;
 202
 203         if (state == dev->state)
 204                 return 0;
 205
 206 again:
 207         abort = 1;
 208
 209         err = xenbus_transaction_start(&xbt);
 210         if (err) {
 211                 xenbus_switch_fatal(dev, depth, err, "starting transaction");
 212                 return 0;
 213         }
 214
 215         err = xenbus_scanf(xbt, dev->nodename, "state", "%d", &current_state);
 216         if (err != 1)
 217                 goto abort;
 218
 219         err = xenbus_printf(xbt, dev->nodename, "state", "%d", state);
 220         if (err) {
 221                 xenbus_switch_fatal(dev, depth, err, "writing new state");
 222                 goto abort;
 223         }
 224
 225         abort = 0;
 226 abort:
 227         err = xenbus_transaction_end(xbt, abort);
 228         if (err) {
 229                 if (err == -EAGAIN && !abort)
 230                         goto again;
 231                 xenbus_switch_fatal(dev, depth, err, "ending transaction");
 232         } else
 233                 dev->state = state;
 234
 235         return 0;
 236 }
 237
 238 /**
 239  * xenbus_switch_state
 240  * @dev: xenbus device
 241  * @state: new state
 242  *
 243  * Advertise in the store a change of the given driver to the given new_state.
 244  * Return 0 on success, or -errno on error.  On error, the device will switch
 245  * to XenbusStateClosing, and the error will be saved in the store.
 246  */
 247 int xenbus_switch_state(struct xenbus_device *dev, enum xenbus_state state)
 248 {
 249         return __xenbus_switch_state(dev, state, 0);
 250 }
 251
 252 EXPORT_SYMBOL_GPL(xenbus_switch_state);
 253
 254 int xenbus_frontend_closed(struct xenbus_device *dev)
 255 {
 256         xenbus_switch_state(dev, XenbusStateClosed);
 257         complete(&dev->down);
 258         return 0;
 259 }
 260 EXPORT_SYMBOL_GPL(xenbus_frontend_closed);
 261
 262 static void xenbus_va_dev_error(struct xenbus_device *dev, int err,
 263                                 const char *fmt, va_list ap)
 264 {
 265         unsigned int len;
 266         char *printf_buffer;
 267         char *path_buffer;
 268
 269 #define PRINTF_BUFFER_SIZE 4096
 270
 271         printf_buffer = kmalloc(PRINTF_BUFFER_SIZE, GFP_KERNEL);
 272         if (!printf_buffer)
 273                 return;
 274
 275         len = sprintf(printf_buffer, "%i ", -err);
 276         vsnprintf(printf_buffer + len, PRINTF_BUFFER_SIZE - len, fmt, ap);
 277
 278         dev_err(&dev->dev, "%s\n", printf_buffer);
 279
 280         path_buffer = kasprintf(GFP_KERNEL, "error/%s", dev->nodename);
 281         if (!path_buffer ||
 282             xenbus_write(XBT_NIL, path_buffer, "error", printf_buffer))
 283                 dev_err(&dev->dev, "failed to write error node for %s (%s)\n",
 284                         dev->nodename, printf_buffer);
 285
 286         kfree(printf_buffer);
 287         kfree(path_buffer);
 288 }
 289
 290 /**
 291  * xenbus_dev_error
 292  * @dev: xenbus device
 293  * @err: error to report
 294  * @fmt: error message format
 295  *
 296  * Report the given negative errno into the store, along with the given
 297  * formatted message.
 298  */
 299 void xenbus_dev_error(struct xenbus_device *dev, int err, const char *fmt, ...)
 300 {
 301         va_list ap;
 302
 303         va_start(ap, fmt);
 304         xenbus_va_dev_error(dev, err, fmt, ap);
 305         va_end(ap);
 306 }
 307 EXPORT_SYMBOL_GPL(xenbus_dev_error);
 308
 309 /**
 310  * xenbus_dev_fatal
 311  * @dev: xenbus device
 312  * @err: error to report
 313  * @fmt: error message format
 314  *
 315  * Equivalent to xenbus_dev_error(dev, err, fmt, args), followed by
 316  * xenbus_switch_state(dev, XenbusStateClosing) to schedule an orderly
 317  * closedown of this driver and its peer.
 318  */
 319
 320 void xenbus_dev_fatal(struct xenbus_device *dev, int err, const char *fmt, ...)
 321 {
 322         va_list ap;
 323
 324         va_start(ap, fmt);
 325         xenbus_va_dev_error(dev, err, fmt, ap);
 326         va_end(ap);
 327
 328         xenbus_switch_state(dev, XenbusStateClosing);
 329 }
 330 EXPORT_SYMBOL_GPL(xenbus_dev_fatal);
 331
 332 /**
 333  * Equivalent to xenbus_dev_fatal(dev, err, fmt, args), but helps
 334  * avoiding recursion within xenbus_switch_state.
 335  */
 336 static void xenbus_switch_fatal(struct xenbus_device *dev, int depth, int err,
 337                                 const char *fmt, ...)
 338 {
 339         va_list ap;
 340
 341         va_start(ap, fmt);
 342         xenbus_va_dev_error(dev, err, fmt, ap);
 343         va_end(ap);
 344
 345         if (!depth)
 346                 __xenbus_switch_state(dev, XenbusStateClosing, 1);
 347 }
 348
 349 /**
 350  * xenbus_grant_ring
 351  * @dev: xenbus device
 352  * @vaddr: starting virtual address of the ring
 353  * @nr_pages: number of pages to be granted
 354  * @grefs: grant reference array to be filled in
 355  *
 356  * Grant access to the given @vaddr to the peer of the given device.
 357  * Then fill in @grefs with grant references.  Return 0 on success, or
 358  * -errno on error.  On error, the device will switch to
 359  * XenbusStateClosing, and the error will be saved in the store.
 360  */
 361 int xenbus_grant_ring(struct xenbus_device *dev, void *vaddr,
 362                       unsigned int nr_pages, grant_ref_t *grefs)
 363 {
 364         int err;
 365         int i, j;
 366
 367         for (i = 0; i < nr_pages; i++) {
 368                 err = gnttab_grant_foreign_access(dev->otherend_id,
 369                                                   virt_to_gfn(vaddr), 0);
 370                 if (err < 0) {
 371                         xenbus_dev_fatal(dev, err,
 372                                          "granting access to ring page");
 373                         goto fail;
 374                 }
 375                 grefs[i] = err;
 376
 377                 vaddr = vaddr + XEN_PAGE_SIZE;
 378         }
 379
 380         return 0;
 381
 382 fail:
 383         for (j = 0; j < i; j++)
 384                 gnttab_end_foreign_access_ref(grefs[j], 0);
 385         return err;
 386 }
 387 EXPORT_SYMBOL_GPL(xenbus_grant_ring);
 388
 389
 390 /**
 391  * Allocate an event channel for the given xenbus_device, assigning the newly
 392  * created local port to *port.  Return 0 on success, or -errno on error.  On
 393  * error, the device will switch to XenbusStateClosing, and the error will be
 394  * saved in the store.
 395  */
 396 int xenbus_alloc_evtchn(struct xenbus_device *dev, int *port)
 397 {
 398         struct evtchn_alloc_unbound alloc_unbound;
 399         int err;
 400
 401         alloc_unbound.dom = DOMID_SELF;
 402         alloc_unbound.remote_dom = dev->otherend_id;
 403
 404         err = HYPERVISOR_event_channel_op(EVTCHNOP_alloc_unbound,
 405                                           &alloc_unbound);
 406         if (err)
 407                 xenbus_dev_fatal(dev, err, "allocating event channel");
 408         else
 409                 *port = alloc_unbound.port;
 410
 411         return err;
 412 }
 413 EXPORT_SYMBOL_GPL(xenbus_alloc_evtchn);
 414
 415
 416 /**
 417  * Free an existing event channel. Returns 0 on success or -errno on error.
 418  */
 419 int xenbus_free_evtchn(struct xenbus_device *dev, int port)
 420 {
 421         struct evtchn_close close;
 422         int err;
 423
 424         close.port = port;
 425
 426         err = HYPERVISOR_event_channel_op(EVTCHNOP_close, &close);
 427         if (err)
 428                 xenbus_dev_error(dev, err, "freeing event channel %d", port);
 429
 430         return err;
 431 }
 432 EXPORT_SYMBOL_GPL(xenbus_free_evtchn);
 433
 434
 435 /**
 436  * xenbus_map_ring_valloc
 437  * @dev: xenbus device
 438  * @gnt_refs: grant reference array
 439  * @nr_grefs: number of grant references
 440  * @vaddr: pointer to address to be filled out by mapping
 441  *
 442  * Map @nr_grefs pages of memory into this domain from another
 443  * domain's grant table.  xenbus_map_ring_valloc allocates @nr_grefs
 444  * pages of virtual address space, maps the pages to that address, and
 445  * sets *vaddr to that address.  Returns 0 on success, and GNTST_*
 446  * (see xen/include/interface/grant_table.h) or -ENOMEM / -EINVAL on
 447  * error. If an error is returned, device will switch to
 448  * XenbusStateClosing and the error message will be saved in XenStore.
 449  */
 450 int xenbus_map_ring_valloc(struct xenbus_device *dev, grant_ref_t *gnt_refs,
 451                            unsigned int nr_grefs, void **vaddr)
 452 {
 453         return ring_ops->map(dev, gnt_refs, nr_grefs, vaddr);
 454 }
 455 EXPORT_SYMBOL_GPL(xenbus_map_ring_valloc);
 456
 457 /* N.B. sizeof(phys_addr_t) doesn't always equal to sizeof(unsigned
 458  * long), e.g. 32-on-64.  Caller is responsible for preparing the
 459  * right array to feed into this function */
 460 static int __xenbus_map_ring(struct xenbus_device *dev,
 461                              grant_ref_t *gnt_refs,
 462                              unsigned int nr_grefs,
 463                              grant_handle_t *handles,
 464                              phys_addr_t *addrs,
 465                              unsigned int flags,
 466                              bool *leaked)
 467 {
 468         struct gnttab_map_grant_ref map[XENBUS_MAX_RING_GRANTS];
 469         struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_GRANTS];
 470         int i, j;
 471         int err = GNTST_okay;
 472
 473         if (nr_grefs > XENBUS_MAX_RING_GRANTS)
 474                 return -EINVAL;
 475
 476         for (i = 0; i < nr_grefs; i++) {
 477                 memset(&map[i], 0, sizeof(map[i]));
 478                 gnttab_set_map_op(&map[i], addrs[i], flags, gnt_refs[i],
 479                                   dev->otherend_id);
 480                 handles[i] = INVALID_GRANT_HANDLE;
 481         }
 482
 483         gnttab_batch_map(map, i);
 484
 485         for (i = 0; i < nr_grefs; i++) {
 486                 if (map[i].status != GNTST_okay) {
 487                         err = map[i].status;
 488                         xenbus_dev_fatal(dev, map[i].status,
 489                                          "mapping in shared page %d from domain %d",
 490                                          gnt_refs[i], dev->otherend_id);
 491                         goto fail;
 492                 } else
 493                         handles[i] = map[i].handle;
 494         }
 495
 496         return GNTST_okay;
 497
 498  fail:
 499         for (i = j = 0; i < nr_grefs; i++) {
 500                 if (handles[i] != INVALID_GRANT_HANDLE) {
 501                         memset(&unmap[j], 0, sizeof(unmap[j]));
 502                         gnttab_set_unmap_op(&unmap[j], (phys_addr_t)addrs[i],
 503                                             GNTMAP_host_map, handles[i]);
 504                         j++;
 505                 }
 506         }
 507
 508         if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, j))
 509                 BUG();
 510
 511         *leaked = false;
 512         for (i = 0; i < j; i++) {
 513                 if (unmap[i].status != GNTST_okay) {
 514                         *leaked = true;
 515                         break;
 516                 }
 517         }
 518
 519         return err;
 520 }
 521
 522 static int xenbus_map_ring_valloc_pv(struct xenbus_device *dev,
 523                                      grant_ref_t *gnt_refs,
 524                                      unsigned int nr_grefs,
 525                                      void **vaddr)
 526 {
 527         struct xenbus_map_node *node;
 528         struct vm_struct *area;
 529         pte_t *ptes[XENBUS_MAX_RING_GRANTS];
 530         phys_addr_t phys_addrs[XENBUS_MAX_RING_GRANTS];
 531         int err = GNTST_okay;
 532         int i;
 533         bool leaked;
 534
 535         *vaddr = NULL;
 536
 537         if (nr_grefs > XENBUS_MAX_RING_GRANTS)
 538                 return -EINVAL;
 539
 540         node = kzalloc(sizeof(*node), GFP_KERNEL);
 541         if (!node)
 542                 return -ENOMEM;
 543
 544         area = alloc_vm_area(XEN_PAGE_SIZE * nr_grefs, ptes);
 545         if (!area) {
 546                 kfree(node);
 547                 return -ENOMEM;
 548         }
 549
 550         for (i = 0; i < nr_grefs; i++)
 551                 phys_addrs[i] = arbitrary_virt_to_machine(ptes[i]).maddr;
 552
 553         err = __xenbus_map_ring(dev, gnt_refs, nr_grefs, node->handles,
 554                                 phys_addrs,
 555                                 GNTMAP_host_map | GNTMAP_contains_pte,
 556                                 &leaked);
 557         if (err)
 558                 goto failed;
 559
 560         node->nr_handles = nr_grefs;
 561         node->pv.area = area;
 562
 563         spin_lock(&xenbus_valloc_lock);
 564         list_add(&node->next, &xenbus_valloc_pages);
 565         spin_unlock(&xenbus_valloc_lock);
 566
 567         *vaddr = area->addr;
 568         return 0;
 569
 570 failed:
 571         if (!leaked)
 572                 free_vm_area(area);
 573         else
 574                 pr_alert("leaking VM area %p size %u page(s)", area, nr_grefs);
 575
 576         kfree(node);
 577         return err;
 578 }
 579
 580 struct map_ring_valloc_hvm
 581 {
 582         unsigned int idx;
 583
 584         /* Why do we need two arrays? See comment of __xenbus_map_ring */
 585         phys_addr_t phys_addrs[XENBUS_MAX_RING_GRANTS];
 586         unsigned long addrs[XENBUS_MAX_RING_GRANTS];
 587 };
 588
 589 static void xenbus_map_ring_setup_grant_hvm(unsigned long gfn,
 590                                             unsigned int goffset,
 591                                             unsigned int len,
 592                                             void *data)
 593 {
 594         struct map_ring_valloc_hvm *info = data;
 595         unsigned long vaddr = (unsigned long)gfn_to_virt(gfn);
 596
 597         info->phys_addrs[info->idx] = vaddr;
 598         info->addrs[info->idx] = vaddr;
 599
 600         info->idx++;
 601 }
 602
 603 static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev,
 604                                       grant_ref_t *gnt_ref,
 605                                       unsigned int nr_grefs,
 606                                       void **vaddr)
 607 {
 608         struct xenbus_map_node *node;
 609         int err;
 610         void *addr;
 611         bool leaked = false;
 612         struct map_ring_valloc_hvm info = {
 613                 .idx = 0,
 614         };
 615         unsigned int nr_pages = XENBUS_PAGES(nr_grefs);
 616
 617         if (nr_grefs > XENBUS_MAX_RING_GRANTS)
 618                 return -EINVAL;
 619
 620         *vaddr = NULL;
 621
 622         node = kzalloc(sizeof(*node), GFP_KERNEL);
 623         if (!node)
 624                 return -ENOMEM;
 625
 626         err = alloc_xenballooned_pages(nr_pages, node->hvm.pages);
 627         if (err)
 628                 goto out_err;
 629
 630         gnttab_foreach_grant(node->hvm.pages, nr_grefs,
 631                              xenbus_map_ring_setup_grant_hvm,
 632                              &info);
 633
 634         err = __xenbus_map_ring(dev, gnt_ref, nr_grefs, node->handles,
 635                                 info.phys_addrs, GNTMAP_host_map, &leaked);
 636         node->nr_handles = nr_grefs;
 637
 638         if (err)
 639                 goto out_free_ballooned_pages;
 640
 641         addr = vmap(node->hvm.pages, nr_pages, VM_MAP | VM_IOREMAP,
 642                     PAGE_KERNEL);
 643         if (!addr) {
 644                 err = -ENOMEM;
 645                 goto out_xenbus_unmap_ring;
 646         }
 647
 648         node->hvm.addr = addr;
 649
 650         spin_lock(&xenbus_valloc_lock);
 651         list_add(&node->next, &xenbus_valloc_pages);
 652         spin_unlock(&xenbus_valloc_lock);
 653
 654         *vaddr = addr;
 655         return 0;
 656
 657  out_xenbus_unmap_ring:
 658         if (!leaked)
 659                 xenbus_unmap_ring(dev, node->handles, nr_grefs, info.addrs);
 660         else
 661                 pr_alert("leaking %p size %u page(s)",
 662                          addr, nr_pages);
 663  out_free_ballooned_pages:
 664         if (!leaked)
 665                 free_xenballooned_pages(nr_pages, node->hvm.pages);
 666  out_err:
 667         kfree(node);
 668         return err;
 669 }
 670
 671
 672 /**
 673  * xenbus_map_ring
 674  * @dev: xenbus device
 675  * @gnt_refs: grant reference array
 676  * @nr_grefs: number of grant reference
 677  * @handles: pointer to grant handle to be filled
 678  * @vaddrs: addresses to be mapped to
 679  * @leaked: fail to clean up a failed map, caller should not free vaddr
 680  *
 681  * Map pages of memory into this domain from another domain's grant table.
 682  * xenbus_map_ring does not allocate the virtual address space (you must do
 683  * this yourself!). It only maps in the pages to the specified address.
 684  * Returns 0 on success, and GNTST_* (see xen/include/interface/grant_table.h)
 685  * or -ENOMEM / -EINVAL on error. If an error is returned, device will switch to
 686  * XenbusStateClosing and the first error message will be saved in XenStore.
 687  * Further more if we fail to map the ring, caller should check @leaked.
 688  * If @leaked is not zero it means xenbus_map_ring fails to clean up, caller
 689  * should not free the address space of @vaddr.
 690  */
 691 int xenbus_map_ring(struct xenbus_device *dev, grant_ref_t *gnt_refs,
 692                     unsigned int nr_grefs, grant_handle_t *handles,
 693                     unsigned long *vaddrs, bool *leaked)
 694 {
 695         phys_addr_t phys_addrs[XENBUS_MAX_RING_GRANTS];
 696         int i;
 697
 698         if (nr_grefs > XENBUS_MAX_RING_GRANTS)
 699                 return -EINVAL;
 700
 701         for (i = 0; i < nr_grefs; i++)
 702                 phys_addrs[i] = (unsigned long)vaddrs[i];
 703
 704         return __xenbus_map_ring(dev, gnt_refs, nr_grefs, handles,
 705                                  phys_addrs, GNTMAP_host_map, leaked);
 706 }
 707 EXPORT_SYMBOL_GPL(xenbus_map_ring);
 708
 709
 710 /**
 711  * xenbus_unmap_ring_vfree
 712  * @dev: xenbus device
 713  * @vaddr: addr to unmap
 714  *
 715  * Based on Rusty Russell's skeleton driver's unmap_page.
 716  * Unmap a page of memory in this domain that was imported from another domain.
 717  * Use xenbus_unmap_ring_vfree if you mapped in your memory with
 718  * xenbus_map_ring_valloc (it will free the virtual address space).
 719  * Returns 0 on success and returns GNTST_* on error
 720  * (see xen/include/interface/grant_table.h).
 721  */
 722 int xenbus_unmap_ring_vfree(struct xenbus_device *dev, void *vaddr)
 723 {
 724         return ring_ops->unmap(dev, vaddr);
 725 }
 726 EXPORT_SYMBOL_GPL(xenbus_unmap_ring_vfree);
 727
 728 static int xenbus_unmap_ring_vfree_pv(struct xenbus_device *dev, void *vaddr)
 729 {
 730         struct xenbus_map_node *node;
 731         struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_GRANTS];
 732         unsigned int level;
 733         int i;
 734         bool leaked = false;
 735         int err;
 736
 737         spin_lock(&xenbus_valloc_lock);
 738         list_for_each_entry(node, &xenbus_valloc_pages, next) {
 739                 if (node->pv.area->addr == vaddr) {
 740                         list_del(&node->next);
 741                         goto found;
 742                 }
 743         }
 744         node = NULL;
 745  found:
 746         spin_unlock(&xenbus_valloc_lock);
 747
 748         if (!node) {
 749                 xenbus_dev_error(dev, -ENOENT,
 750                                  "can't find mapped virtual address %p", vaddr);
 751                 return GNTST_bad_virt_addr;
 752         }
 753
 754         for (i = 0; i < node->nr_handles; i++) {
 755                 unsigned long addr;
 756
 757                 memset(&unmap[i], 0, sizeof(unmap[i]));
 758                 addr = (unsigned long)vaddr + (XEN_PAGE_SIZE * i);
 759                 unmap[i].host_addr = arbitrary_virt_to_machine(
 760                         lookup_address(addr, &level)).maddr;
 761                 unmap[i].dev_bus_addr = 0;
 762                 unmap[i].handle = node->handles[i];
 763         }
 764
 765         if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, i))
 766                 BUG();
 767
 768         err = GNTST_okay;
 769         leaked = false;
 770         for (i = 0; i < node->nr_handles; i++) {
 771                 if (unmap[i].status != GNTST_okay) {
 772                         leaked = true;
 773                         xenbus_dev_error(dev, unmap[i].status,
 774                                          "unmapping page at handle %d error %d",
 775                                          node->handles[i], unmap[i].status);
 776                         err = unmap[i].status;
 777                         break;
 778                 }
 779         }
 780
 781         if (!leaked)
 782                 free_vm_area(node->pv.area);
 783         else
 784                 pr_alert("leaking VM area %p size %u page(s)",
 785                          node->pv.area, node->nr_handles);
 786
 787         kfree(node);
 788         return err;
 789 }
 790
 791 struct unmap_ring_vfree_hvm
 792 {
 793         unsigned int idx;
 794         unsigned long addrs[XENBUS_MAX_RING_GRANTS];
 795 };
 796
 797 static void xenbus_unmap_ring_setup_grant_hvm(unsigned long gfn,
 798                                               unsigned int goffset,
 799                                               unsigned int len,
 800                                               void *data)
 801 {
 802         struct unmap_ring_vfree_hvm *info = data;
 803
 804         info->addrs[info->idx] = (unsigned long)gfn_to_virt(gfn);
 805
 806         info->idx++;
 807 }
 808
 809 static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr)
 810 {
 811         int rv;
 812         struct xenbus_map_node *node;
 813         void *addr;
 814         struct unmap_ring_vfree_hvm info = {
 815                 .idx = 0,
 816         };
 817         unsigned int nr_pages;
 818
 819         spin_lock(&xenbus_valloc_lock);
 820         list_for_each_entry(node, &xenbus_valloc_pages, next) {
 821                 addr = node->hvm.addr;
 822                 if (addr == vaddr) {
 823                         list_del(&node->next);
 824                         goto found;
 825                 }
 826         }
 827         node = addr = NULL;
 828  found:
 829         spin_unlock(&xenbus_valloc_lock);
 830
 831         if (!node) {
 832                 xenbus_dev_error(dev, -ENOENT,
 833                                  "can't find mapped virtual address %p", vaddr);
 834                 return GNTST_bad_virt_addr;
 835         }
 836
 837         nr_pages = XENBUS_PAGES(node->nr_handles);
 838
 839         gnttab_foreach_grant(node->hvm.pages, node->nr_handles,
 840                              xenbus_unmap_ring_setup_grant_hvm,
 841                              &info);
 842
 843         rv = xenbus_unmap_ring(dev, node->handles, node->nr_handles,
 844                                info.addrs);
 845         if (!rv) {
 846                 vunmap(vaddr);
 847                 free_xenballooned_pages(nr_pages, node->hvm.pages);
 848         }
 849         else
 850                 WARN(1, "Leaking %p, size %u page(s)\n", vaddr, nr_pages);
 851
 852         kfree(node);
 853         return rv;
 854 }
 855
 856 /**
 857  * xenbus_unmap_ring
 858  * @dev: xenbus device
 859  * @handles: grant handle array
 860  * @nr_handles: number of handles in the array
 861  * @vaddrs: addresses to unmap
 862  *
 863  * Unmap memory in this domain that was imported from another domain.
 864  * Returns 0 on success and returns GNTST_* on error
 865  * (see xen/include/interface/grant_table.h).
 866  */
 867 int xenbus_unmap_ring(struct xenbus_device *dev,
 868                       grant_handle_t *handles, unsigned int nr_handles,
 869                       unsigned long *vaddrs)
 870 {
 871         struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_GRANTS];
 872         int i;
 873         int err;
 874
 875         if (nr_handles > XENBUS_MAX_RING_GRANTS)
 876                 return -EINVAL;
 877
 878         for (i = 0; i < nr_handles; i++)
 879                 gnttab_set_unmap_op(&unmap[i], vaddrs[i],
 880                                     GNTMAP_host_map, handles[i]);
 881
 882         if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, i))
 883                 BUG();
 884
 885         err = GNTST_okay;
 886         for (i = 0; i < nr_handles; i++) {
 887                 if (unmap[i].status != GNTST_okay) {
 888                         xenbus_dev_error(dev, unmap[i].status,
 889                                          "unmapping page at handle %d error %d",
 890                                          handles[i], unmap[i].status);
 891                         err = unmap[i].status;
 892                         break;
 893                 }
 894         }
 895
 896         return err;
 897 }
 898 EXPORT_SYMBOL_GPL(xenbus_unmap_ring);
 899
 900
 901 /**
 902  * xenbus_read_driver_state
 903  * @path: path for driver
 904  *
 905  * Return the state of the driver rooted at the given store path, or
 906  * XenbusStateUnknown if no state can be read.
 907  */
 908 enum xenbus_state xenbus_read_driver_state(const char *path)
 909 {
 910         enum xenbus_state result;
 911         int err = xenbus_gather(XBT_NIL, path, "state", "%d", &result, NULL);
 912         if (err)
 913                 result = XenbusStateUnknown;
 914
 915         return result;
 916 }
 917 EXPORT_SYMBOL_GPL(xenbus_read_driver_state);
 918
 919 static const struct xenbus_ring_ops ring_ops_pv = {
 920         .map = xenbus_map_ring_valloc_pv,
 921         .unmap = xenbus_unmap_ring_vfree_pv,
 922 };
 923
 924 static const struct xenbus_ring_ops ring_ops_hvm = {
 925         .map = xenbus_map_ring_valloc_hvm,
 926         .unmap = xenbus_unmap_ring_vfree_hvm,
 927 };
 928
 929 void __init xenbus_ring_ops_init(void)
 930 {
 931         if (!xen_feature(XENFEAT_auto_translated_physmap))
 932                 ring_ops = &ring_ops_pv;
 933         else
 934                 ring_ops = &ring_ops_hvm;
 935 }