lib/netns_linux.c

   1 /*
   2  * NS functions.
   3  * Copyright (C) 2014 6WIND S.A.
   4  *
   5  * This file is part of GNU Zebra.
   6  *
   7  * GNU Zebra is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published
   9  * by the Free Software Foundation; either version 2, or (at your
  10  * option) any later version.
  11  *
  12  * GNU Zebra is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along
  18  * with this program; see the file COPYING; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include <zebra.h>
  23
  24 #ifdef HAVE_NETNS
  25 #undef _GNU_SOURCE
  26 #define _GNU_SOURCE
  27
  28 #include <sched.h>
  29 #endif
  30
  31 /* for basename */
  32 #include <libgen.h>
  33
  34 #include "if.h"
  35 #include "ns.h"
  36 #include "log.h"
  37 #include "memory.h"
  38
  39 #include "command.h"
  40 #include "vty.h"
  41 #include "vrf.h"
  42
  43 DEFINE_MTYPE_STATIC(LIB, NS, "NetNS Context")
  44 DEFINE_MTYPE_STATIC(LIB, NS_NAME, "NetNS Name")
  45
  46 /* default NS ID value used when VRF backend is not NETNS */
  47 #define NS_DEFAULT_INTERNAL 0
  48
  49 static inline int ns_compare(const struct ns *ns, const struct ns *ns2);
  50 static struct ns *ns_lookup_name_internal(const char *name);
  51
  52 RB_GENERATE(ns_head, ns, entry, ns_compare)
  53
  54 struct ns_head ns_tree = RB_INITIALIZER(&ns_tree);
  55
  56 static struct ns *default_ns;
  57 static int ns_current_ns_fd;
  58 static int ns_default_ns_fd;
  59
  60 static int ns_debug;
  61
  62 struct ns_map_nsid {
  63         RB_ENTRY(ns_map_nsid) id_entry;
  64         ns_id_t ns_id_external;
  65         ns_id_t ns_id;
  66 };
  67
  68 static inline int ns_map_compare(const struct ns_map_nsid *a,
  69                                    const struct ns_map_nsid *b)
  70 {
  71         return (a->ns_id - b->ns_id);
  72 }
  73
  74 RB_HEAD(ns_map_nsid_head, ns_map_nsid);
  75 RB_PROTOTYPE(ns_map_nsid_head, ns_map_nsid, id_entry, ns_map_compare);
  76 RB_GENERATE(ns_map_nsid_head, ns_map_nsid, id_entry, ns_map_compare);
  77 struct ns_map_nsid_head ns_map_nsid_list = RB_INITIALIZER(&ns_map_nsid_list);
  78
  79 static ns_id_t ns_id_external_numbering;
  80
  81
  82 #ifndef CLONE_NEWNET
  83 #define CLONE_NEWNET 0x40000000
  84 /* New network namespace (lo, device, names sockets, etc) */
  85 #endif
  86
  87 #ifndef HAVE_SETNS
  88 static inline int setns(int fd, int nstype)
  89 {
  90 #ifdef __NR_setns
  91         return syscall(__NR_setns, fd, nstype);
  92 #else
  93         errno = EINVAL;
  94         return -1;
  95 #endif
  96 }
  97 #endif /* !HAVE_SETNS */
  98
  99 #ifdef HAVE_NETNS
 100 static int have_netns_enabled = -1;
 101 #endif /* HAVE_NETNS */
 102
 103 /* default NS ID value used when VRF backend is not NETNS */
 104 #define NS_DEFAULT_INTERNAL 0
 105
 106 static int have_netns(void)
 107 {
 108 #ifdef HAVE_NETNS
 109         if (have_netns_enabled < 0) {
 110                 int fd = open(NS_DEFAULT_NAME, O_RDONLY);
 111
 112                 if (fd < 0)
 113                         have_netns_enabled = 0;
 114                 else {
 115                         have_netns_enabled = 1;
 116                         close(fd);
 117                 }
 118         }
 119         return have_netns_enabled;
 120 #else
 121         return 0;
 122 #endif
 123 }
 124
 125 /* Holding NS hooks  */
 126 struct ns_master {
 127         int (*ns_new_hook)(struct ns *ns);
 128         int (*ns_delete_hook)(struct ns *ns);
 129         int (*ns_enable_hook)(struct ns *ns);
 130         int (*ns_disable_hook)(struct ns *ns);
 131 } ns_master = {
 132         0,
 133 };
 134
 135 static int ns_is_enabled(struct ns *ns);
 136
 137 static inline int ns_compare(const struct ns *a, const struct ns *b)
 138 {
 139         return (a->ns_id - b->ns_id);
 140 }
 141
 142 /* Look up a NS by identifier. */
 143 static struct ns *ns_lookup_internal(ns_id_t ns_id)
 144 {
 145         struct ns ns;
 146
 147         ns.ns_id = ns_id;
 148         return RB_FIND(ns_head, &ns_tree, &ns);
 149 }
 150
 151 /* Look up a NS by name */
 152 static struct ns *ns_lookup_name_internal(const char *name)
 153 {
 154         struct ns *ns = NULL;
 155
 156         RB_FOREACH (ns, ns_head, &ns_tree) {
 157                 if (ns->name != NULL) {
 158                         if (strcmp(name, ns->name) == 0)
 159                                 return ns;
 160                 }
 161         }
 162         return NULL;
 163 }
 164
 165 static struct ns *ns_get_created_internal(struct ns *ns, char *name,
 166                                           ns_id_t ns_id)
 167 {
 168         int created = 0;
 169         /*
 170          * Initialize interfaces.
 171          */
 172         if (!ns && !name && ns_id != NS_UNKNOWN)
 173                 ns = ns_lookup_internal(ns_id);
 174         if (!ns && name)
 175                 ns = ns_lookup_name_internal(name);
 176         if (!ns) {
 177                 ns = XCALLOC(MTYPE_NS, sizeof(struct ns));
 178                 ns->ns_id = ns_id;
 179                 if (name)
 180                         ns->name = XSTRDUP(MTYPE_NS_NAME, name);
 181                 ns->fd = -1;
 182                 RB_INSERT(ns_head, &ns_tree, ns);
 183                 created = 1;
 184         }
 185         if (ns_id != ns->ns_id) {
 186                 RB_REMOVE(ns_head, &ns_tree, ns);
 187                 ns->ns_id = ns_id;
 188                 RB_INSERT(ns_head, &ns_tree, ns);
 189         }
 190         if (!created)
 191                 return ns;
 192         if (ns_debug) {
 193                 if (ns->ns_id != NS_UNKNOWN)
 194                         zlog_info("NS %u is created.", ns->ns_id);
 195                 else
 196                         zlog_info("NS %s is created.", ns->name);
 197         }
 198         if (ns_master.ns_new_hook)
 199                 (*ns_master.ns_new_hook)(ns);
 200         return ns;
 201 }
 202
 203 /*
 204  * Enable a NS - that is, let the NS be ready to use.
 205  * The NS_ENABLE_HOOK callback will be called to inform
 206  * that they can allocate resources in this NS.
 207  *
 208  * RETURN: 1 - enabled successfully; otherwise, 0.
 209  */
 210 static int ns_enable_internal(struct ns *ns, void (*func)(ns_id_t, void *))
 211 {
 212         if (!ns_is_enabled(ns)) {
 213                 if (have_netns()) {
 214                         ns->fd = open(ns->name, O_RDONLY);
 215                 } else {
 216                         ns->fd = -2;
 217                         /* Remember ns_enable_hook has been called */
 218                         errno = -ENOTSUP;
 219                 }
 220
 221                 if (!ns_is_enabled(ns)) {
 222                         zlog_err("Can not enable NS %u: %s!", ns->ns_id,
 223                                  safe_strerror(errno));
 224                         return 0;
 225                 }
 226
 227                 /* Non default NS. leave */
 228                 if (ns->ns_id == NS_UNKNOWN) {
 229                         zlog_err("Can not enable NS %s %u: Invalid NSID",
 230                                  ns->name, ns->ns_id);
 231                         return 0;
 232                 }
 233                 if (func)
 234                         func(ns->ns_id, (void *)ns->vrf_ctxt);
 235                 if (ns_debug) {
 236                         if (have_netns())
 237                                 zlog_info("NS %u is associated with NETNS %s.",
 238                                           ns->ns_id, ns->name);
 239                         zlog_info("NS %u is enabled.", ns->ns_id);
 240                 }
 241                 /* zebra first receives NS enable event,
 242                  * then VRF enable event
 243                  */
 244                 if (ns_master.ns_enable_hook)
 245                         (*ns_master.ns_enable_hook)(ns);
 246         }
 247
 248         return 1;
 249 }
 250
 251 /*
 252  * Check whether the NS is enabled - that is, whether the NS
 253  * is ready to allocate resources. Currently there's only one
 254  * type of resource: socket.
 255  */
 256 static int ns_is_enabled(struct ns *ns)
 257 {
 258         if (have_netns())
 259                 return ns && ns->fd >= 0;
 260         else
 261                 return ns && ns->fd == -2 && ns->ns_id == NS_DEFAULT;
 262 }
 263
 264 /*
 265  * Disable a NS - that is, let the NS be unusable.
 266  * The NS_DELETE_HOOK callback will be called to inform
 267  * that they must release the resources in the NS.
 268  */
 269 static void ns_disable_internal(struct ns *ns)
 270 {
 271         if (ns_is_enabled(ns)) {
 272                 if (ns_debug)
 273                         zlog_info("NS %u is to be disabled.", ns->ns_id);
 274
 275                 if (ns_master.ns_disable_hook)
 276                         (*ns_master.ns_disable_hook)(ns);
 277
 278                 if (have_netns())
 279                         close(ns->fd);
 280
 281                 ns->fd = -1;
 282         }
 283 }
 284
 285 /* VRF list existance check by name. */
 286 static struct ns_map_nsid *ns_map_nsid_lookup_by_nsid(ns_id_t ns_id)
 287 {
 288         struct ns_map_nsid ns_map;
 289
 290         ns_map.ns_id = ns_id;
 291         return RB_FIND(ns_map_nsid_head, &ns_map_nsid_list, &ns_map);
 292 }
 293
 294 ns_id_t ns_map_nsid_with_external(ns_id_t ns_id, bool map)
 295 {
 296         struct ns_map_nsid *ns_map;
 297         vrf_id_t ns_id_external;
 298
 299         ns_map = ns_map_nsid_lookup_by_nsid(ns_id);
 300         if (ns_map && !map) {
 301                 ns_id_external = ns_map->ns_id_external;
 302                 RB_REMOVE(ns_map_nsid_head, &ns_map_nsid_list, ns_map);
 303                 return ns_id_external;
 304         }
 305         if (ns_map)
 306                 return ns_map->ns_id_external;
 307         ns_map = XCALLOC(MTYPE_NS, sizeof(struct ns_map_nsid));
 308         /* increase vrf_id
 309          * default vrf is the first one : 0
 310          */
 311         ns_map->ns_id_external = ns_id_external_numbering++;
 312         ns_map->ns_id = ns_id;
 313         RB_INSERT(ns_map_nsid_head, &ns_map_nsid_list, ns_map);
 314         return ns_map->ns_id_external;
 315 }
 316
 317 struct ns *ns_get_created(struct ns *ns, char *name, ns_id_t ns_id)
 318 {
 319         return ns_get_created_internal(ns, name, ns_id);
 320 }
 321
 322 int ns_have_netns(void)
 323 {
 324         return have_netns();
 325 }
 326
 327 /* Delete a NS. This is called in ns_terminate(). */
 328 void ns_delete(struct ns *ns)
 329 {
 330         if (ns_debug)
 331                 zlog_info("NS %u is to be deleted.", ns->ns_id);
 332
 333         ns_disable(ns);
 334
 335         if (ns_master.ns_delete_hook)
 336                 (*ns_master.ns_delete_hook)(ns);
 337
 338         /*
 339          * I'm not entirely sure if the vrf->iflist
 340          * needs to be moved into here or not.
 341          */
 342         // if_terminate (&ns->iflist);
 343
 344         RB_REMOVE(ns_head, &ns_tree, ns);
 345         if (ns->name)
 346                 XFREE(MTYPE_NS_NAME, ns->name);
 347
 348         XFREE(MTYPE_NS, ns);
 349 }
 350
 351 /* Look up the data pointer of the specified VRF. */
 352 void *ns_info_lookup(ns_id_t ns_id)
 353 {
 354         struct ns *ns = ns_lookup_internal(ns_id);
 355
 356         return ns ? ns->info : NULL;
 357 }
 358
 359 /* Look up a NS by name */
 360 struct ns *ns_lookup_name(const char *name)
 361 {
 362         return ns_lookup_name_internal(name);
 363 }
 364
 365 int ns_enable(struct ns *ns, void (*func)(ns_id_t, void *))
 366 {
 367         return ns_enable_internal(ns, func);
 368 }
 369
 370 void ns_disable(struct ns *ns)
 371 {
 372         return ns_disable_internal(ns);
 373 }
 374
 375 struct ns *ns_lookup(ns_id_t ns_id)
 376 {
 377         return ns_lookup_internal(ns_id);
 378 }
 379
 380 void ns_walk_func(int (*func)(struct ns *))
 381 {
 382         struct ns *ns = NULL;
 383
 384         RB_FOREACH (ns, ns_head, &ns_tree)
 385                 func(ns);
 386 }
 387
 388 const char *ns_get_name(struct ns *ns)
 389 {
 390         if (!ns)
 391                 return NULL;
 392         return ns->name;
 393 }
 394
 395 /* Add a NS hook. Please add hooks before calling ns_init(). */
 396 void ns_add_hook(int type, int (*func)(struct ns *))
 397 {
 398         switch (type) {
 399         case NS_NEW_HOOK:
 400                 ns_master.ns_new_hook = func;
 401                 break;
 402         case NS_DELETE_HOOK:
 403                 ns_master.ns_delete_hook = func;
 404                 break;
 405         case NS_ENABLE_HOOK:
 406                 ns_master.ns_enable_hook = func;
 407                 break;
 408         case NS_DISABLE_HOOK:
 409                 ns_master.ns_disable_hook = func;
 410                 break;
 411         default:
 412                 break;
 413         }
 414 }
 415
 416 /*
 417  * NS realization with NETNS
 418  */
 419
 420 char *ns_netns_pathname(struct vty *vty, const char *name)
 421 {
 422         static char pathname[PATH_MAX];
 423         char *result;
 424         char *check_base;
 425
 426         if (name[0] == '/') /* absolute pathname */
 427                 result = realpath(name, pathname);
 428         else {
 429                 /* relevant pathname */
 430                 char tmp_name[PATH_MAX];
 431
 432                 snprintf(tmp_name, PATH_MAX, "%s/%s", NS_RUN_DIR, name);
 433                 result = realpath(tmp_name, pathname);
 434         }
 435
 436         if (!result) {
 437                 if (vty)
 438                         vty_out(vty, "Invalid pathname: %s\n",
 439                                 safe_strerror(errno));
 440                 else
 441                         zlog_warn("Invalid pathname: %s", safe_strerror(errno));
 442                 return NULL;
 443         }
 444         check_base = basename(pathname);
 445         if (check_base != NULL && strlen(check_base) + 1 > NS_NAMSIZ) {
 446                 if (vty)
 447                         vty_out(vty, "NS name (%s) invalid: too long (>%d)\n",
 448                                 check_base, NS_NAMSIZ - 1);
 449                 else
 450                         zlog_warn("NS name (%s) invalid: too long (>%d)",
 451                                   check_base, NS_NAMSIZ - 1);
 452                 return NULL;
 453         }
 454         return pathname;
 455 }
 456
 457 void ns_init(void)
 458 {
 459         static int ns_initialised;
 460
 461         ns_debug = 0;
 462         /* silently return as initialisation done */
 463         if (ns_initialised == 1)
 464                 return;
 465         errno = 0;
 466 #ifdef HAVE_NETNS
 467         if (have_netns_enabled < 0) {
 468                 ns_default_ns_fd = open(NS_DEFAULT_NAME, O_RDONLY);
 469                 if (ns_default_ns_fd == -1)
 470                         zlog_err("NS initialization failure %d(%s)",
 471                                  errno, safe_strerror(errno));
 472         } else {
 473                 ns_default_ns_fd = -1;
 474                 default_ns = NULL;
 475         }
 476 #else
 477         ns_default_ns_fd = -1;
 478         default_ns = NULL;
 479 #endif /* HAVE_NETNS */
 480         ns_current_ns_fd = -1;
 481         ns_initialised = 1;
 482 }
 483
 484 /* Initialize NS module. */
 485 void ns_init_management(ns_id_t default_ns_id, ns_id_t internal_ns)
 486 {
 487         int fd;
 488
 489         ns_init();
 490         default_ns = ns_get_created_internal(NULL, NULL, default_ns_id);
 491         if (!default_ns) {
 492                 zlog_err("%s: failed to create the default NS!", __func__);
 493                 exit(1);
 494         }
 495         if (have_netns()) {
 496                 fd = open(NS_DEFAULT_NAME, O_RDONLY);
 497                 default_ns->fd = fd;
 498         }
 499         default_ns->internal_ns_id = internal_ns;
 500
 501         /* Set the default NS name. */
 502         default_ns->name = XSTRDUP(MTYPE_NS_NAME, NS_DEFAULT_NAME);
 503         if (ns_debug)
 504                 zlog_info("%s: default NSID is %u", __func__,
 505                           default_ns->ns_id);
 506
 507         /* Enable the default NS. */
 508         if (!ns_enable(default_ns, NULL)) {
 509                 zlog_err("%s: failed to enable the default NS!", __func__);
 510                 exit(1);
 511         }
 512 }
 513
 514 /* Terminate NS module. */
 515 void ns_terminate(void)
 516 {
 517         struct ns *ns;
 518
 519         while (!RB_EMPTY(ns_head, &ns_tree)) {
 520                 ns = RB_ROOT(ns_head, &ns_tree);
 521
 522                 ns_delete(ns);
 523         }
 524 }
 525
 526 int ns_switch_to_netns(const char *name)
 527 {
 528         int ret;
 529         int fd;
 530
 531         if (name == NULL)
 532                 return -1;
 533         if (ns_default_ns_fd == -1)
 534                 return -1;
 535         fd = open(name, O_RDONLY);
 536         if (fd == -1) {
 537                 errno = EINVAL;
 538                 return -1;
 539         }
 540         ret = setns(fd, CLONE_NEWNET);
 541         ns_current_ns_fd = fd;
 542         close(fd);
 543         return ret;
 544 }
 545
 546 /* returns 1 if switch() was not called before
 547  * return status of setns() otherwise
 548  */
 549 int ns_switchback_to_initial(void)
 550 {
 551         if (ns_current_ns_fd != -1 && ns_default_ns_fd != -1) {
 552                 int ret;
 553
 554                 ret = setns(ns_default_ns_fd, CLONE_NEWNET);
 555                 ns_current_ns_fd = -1;
 556                 return ret;
 557         }
 558         /* silently ignore if setns() is not called */
 559         return 1;
 560 }
 561
 562 /* Create a socket for the NS. */
 563 int ns_socket(int domain, int type, int protocol, ns_id_t ns_id)
 564 {
 565         struct ns *ns = ns_lookup(ns_id);
 566         int ret;
 567
 568         if (!ns || !ns_is_enabled(ns)) {
 569                 errno = EINVAL;
 570                 return -1;
 571         }
 572         if (have_netns()) {
 573                 ret = (ns_id != NS_DEFAULT) ? setns(ns->fd, CLONE_NEWNET) : 0;
 574                 if (ret >= 0) {
 575                         ret = socket(domain, type, protocol);
 576                         if (ns_id != NS_DEFAULT) {
 577                                 setns(ns_lookup(NS_DEFAULT)->fd, CLONE_NEWNET);
 578                                 ns_current_ns_fd = ns_id;
 579                         }
 580                 }
 581         } else
 582                 ret = socket(domain, type, protocol);
 583
 584         return ret;
 585 }
 586
 587 ns_id_t ns_get_default_id(void)
 588 {
 589         if (default_ns)
 590                 return default_ns->ns_id;
 591         return NS_DEFAULT_INTERNAL;
 592 }