palacios/src/palacios/vmm_vnet_core.c

   1 /*
   2  * This file is part of the Palacios Virtual Machine Monitor developed
   3  * by the V3VEE Project with funding from the United States National
   4  * Science Foundation and the Department of Energy.
   5  *
   6  * The V3VEE Project is a joint project between Northwestern University
   7  * and the University of New Mexico.  You can find out more at
   8  * http://www.v3vee.org
   9  *
  10  * Copyright (c) 2010, Lei Xia <lxia@northwestern.edu>
  11  * Copyright (c) 2009, Yuan Tang <ytang@northwestern.edu>
  12  * Copyright (c) 2009, The V3VEE Project <http://www.v3vee.org>
  13  * All rights reserved.
  14  *
  15  * Author: Lei Xia <lxia@northwestern.edu>
  16  *         Yuan Tang <ytang@northwestern.edu>
  17  *
  18  * This is free software.  You are permitted to use,
  19  * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
  20  */
  21
  22 #include <palacios/vmm_vnet.h>
  23 #include <palacios/vm_guest_mem.h>
  24 #include <palacios/vmm_lock.h>
  25 #include <palacios/vmm_queue.h>
  26 #include <palacios/vmm_sprintf.h>
  27 #include <palacios/vmm_ethernet.h>
  28
  29 #ifndef CONFIG_DEBUG_VNET
  30 #undef PrintDebug
  31 #define PrintDebug(fmt, args...)
  32 #endif
  33
  34 struct eth_hdr {
  35     uint8_t dst_mac[ETH_ALEN];
  36     uint8_t src_mac[ETH_ALEN];
  37     uint16_t type; /* indicates layer 3 protocol type */
  38 } __attribute__((packed));
  39
  40
  41 struct vnet_dev {
  42     int dev_id;
  43     uint8_t mac_addr[ETH_ALEN];
  44     struct v3_vm_info * vm;
  45     struct v3_vnet_dev_ops dev_ops;
  46     void * private_data;
  47
  48     int active;
  49     uint8_t mode;  //vmm_drivern or guest_drivern
  50
  51     struct list_head node;
  52 } __attribute__((packed));
  53
  54
  55 struct vnet_brg_dev {
  56     struct v3_vm_info * vm;
  57     struct v3_vnet_bridge_ops brg_ops;
  58
  59     uint8_t type;
  60     uint8_t mode;
  61     int active;
  62     void * private_data;
  63 } __attribute__((packed));
  64
  65
  66
  67 struct vnet_route_info {
  68     struct v3_vnet_route route_def;
  69
  70     struct vnet_dev * dst_dev;
  71     struct vnet_dev * src_dev;
  72
  73     struct list_head node;
  74     struct list_head match_node; // used for route matching
  75 };
  76
  77
  78 struct route_list {
  79     uint8_t hash_buf[VNET_HASH_SIZE];
  80
  81     uint32_t num_routes;
  82     struct vnet_route_info * routes[0];
  83 } __attribute__((packed));
  84
  85
  86 static struct {
  87     struct list_head routes;
  88     struct list_head devs;
  89
  90     int num_routes;
  91     int num_devs;
  92
  93     struct vnet_brg_dev *bridge;
  94
  95     v3_lock_t lock;
  96
  97     struct hashtable * route_cache;
  98 } vnet_state;
  99
 100
 101
 102 #ifdef CONFIG_DEBUG_VNET
 103 static inline void mac_to_string(uint8_t * mac, char * buf) {
 104     snprintf(buf, 100, "%2x:%2x:%2x:%2x:%2x:%2x",
 105              mac[0], mac[1], mac[2],
 106              mac[3], mac[4], mac[5]);
 107 }
 108
 109 static void print_route(struct v3_vnet_route * route){
 110     char str[50];
 111
 112     mac_to_string(route->src_mac, str);
 113     PrintDebug("Src Mac (%s),  src_qual (%d)\n",
 114                str, route->src_mac_qual);
 115     mac_to_string(route->dst_mac, str);
 116     PrintDebug("Dst Mac (%s),  dst_qual (%d)\n",
 117                str, route->dst_mac_qual);
 118     PrintDebug("Src dev id (%d), src type (%d)",
 119                route->src_id,
 120                route->src_type);
 121     PrintDebug("Dst dev id (%d), dst type (%d)\n",
 122                route->dst_id,
 123                route->dst_type);
 124 }
 125
 126 static void dump_routes(){
 127     struct vnet_route_info *route;
 128
 129     int i = 0;
 130     PrintDebug("\n========Dump routes starts ============\n");
 131     list_for_each_entry(route, &(vnet_state.routes), node) {
 132         PrintDebug("\nroute %d:\n", i++);
 133
 134         print_route(&(route->route_def));
 135         if (route->route_def.dst_type == LINK_INTERFACE) {
 136             PrintDebug("dst_dev (%p), dst_dev_id (%d), dst_dev_ops(%p), dst_dev_data (%p)\n",
 137                 route->dst_dev,
 138                 route->dst_dev->dev_id,
 139                 (void *)&(route->dst_dev->dev_ops),
 140                 route->dst_dev->private_data);
 141         }
 142     }
 143
 144     PrintDebug("\n========Dump routes end ============\n");
 145 }
 146
 147 #endif
 148
 149
 150 /*
 151  * A VNET packet is a packed struct with the hashed fields grouped together.
 152  * This means we can generate the hash from an offset into the pkt struct
 153  */
 154 static inline uint_t hash_fn(addr_t hdr_ptr) {
 155     uint8_t * hdr_buf = (uint8_t *)hdr_ptr;
 156
 157     return v3_hash_buffer(hdr_buf, VNET_HASH_SIZE);
 158 }
 159
 160 static inline int hash_eq(addr_t key1, addr_t key2) {
 161     return (memcmp((uint8_t *)key1, (uint8_t *)key2, VNET_HASH_SIZE) == 0);
 162 }
 163
 164 static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list * routes) {
 165     memcpy(routes->hash_buf, pkt->hash_buf, VNET_HASH_SIZE);
 166
 167     if (v3_htable_insert(vnet_state.route_cache, (addr_t)routes->hash_buf, (addr_t)routes) == 0) {
 168         PrintError("VNET/P Core: Failed to insert new route entry to the cache\n");
 169         return -1;
 170     }
 171
 172     return 0;
 173 }
 174
 175 static int clear_hash_cache() {
 176     v3_free_htable(vnet_state.route_cache, 1, 1);
 177     vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
 178
 179     return 0;
 180 }
 181
 182 static int look_into_cache(const struct v3_vnet_pkt * pkt, struct route_list ** routes) {
 183     *routes = (struct route_list *)v3_htable_search(vnet_state.route_cache, (addr_t)(pkt->hash_buf));
 184
 185     return 0;
 186 }
 187
 188
 189 static struct vnet_dev * dev_by_id(int idx) {
 190     struct vnet_dev * dev = NULL;
 191
 192     list_for_each_entry(dev, &(vnet_state.devs), node) {
 193         int dev_id = dev->dev_id;
 194
 195         if (dev_id == idx)
 196             return dev;
 197     }
 198
 199     return NULL;
 200 }
 201
 202 static struct vnet_dev * dev_by_mac(uint8_t * mac) {
 203     struct vnet_dev * dev = NULL;
 204
 205     list_for_each_entry(dev, &(vnet_state.devs), node) {
 206         if (!compare_ethaddr(dev->mac_addr, mac)){
 207             return dev;
 208         }
 209     }
 210
 211     return NULL;
 212 }
 213
 214
 215 int v3_vnet_find_dev(uint8_t  * mac) {
 216     struct vnet_dev * dev = NULL;
 217
 218     dev = dev_by_mac(mac);
 219
 220     if(dev != NULL) {
 221         return dev->dev_id;
 222     }
 223
 224     return -1;
 225 }
 226
 227
 228 int v3_vnet_add_route(struct v3_vnet_route route) {
 229     struct vnet_route_info * new_route = NULL;
 230     unsigned long flags;
 231
 232     new_route = (struct vnet_route_info *)V3_Malloc(sizeof(struct vnet_route_info));
 233     memset(new_route, 0, sizeof(struct vnet_route_info));
 234
 235 #ifdef CONFIG_DEBUG_VNET
 236     PrintDebug("VNET/P Core: add_route_entry:\n");
 237     print_route(&route);
 238 #endif
 239
 240     memcpy(new_route->route_def.src_mac, route.src_mac, ETH_ALEN);
 241     memcpy(new_route->route_def.dst_mac, route.dst_mac, ETH_ALEN);
 242     new_route->route_def.src_mac_qual = route.src_mac_qual;
 243     new_route->route_def.dst_mac_qual = route.dst_mac_qual;
 244     new_route->route_def.dst_type = route.dst_type;
 245     new_route->route_def.src_type = route.src_type;
 246     new_route->route_def.src_id = route.src_id;
 247     new_route->route_def.dst_id = route.dst_id;
 248
 249     if (new_route->route_def.dst_type == LINK_INTERFACE) {
 250         new_route->dst_dev = dev_by_id(new_route->route_def.dst_id);
 251     }
 252
 253     if (new_route->route_def.src_type == LINK_INTERFACE) {
 254         new_route->src_dev = dev_by_id(new_route->route_def.src_id);
 255     }
 256
 257
 258     flags = v3_lock_irqsave(vnet_state.lock);
 259
 260     list_add(&(new_route->node), &(vnet_state.routes));
 261     clear_hash_cache();
 262
 263     v3_unlock_irqrestore(vnet_state.lock, flags);
 264
 265
 266 #ifdef CONFIG_DEBUG_VNET
 267     dump_routes();
 268 #endif
 269
 270     return 0;
 271 }
 272
 273
 274 /* delete all route entries with specfied src or dst device id */
 275 static void inline del_routes_by_dev(int dev_id){
 276     struct vnet_route_info * route = NULL;
 277     unsigned long flags;
 278
 279     flags = v3_lock_irqsave(vnet_state.lock);
 280
 281     list_for_each_entry(route, &(vnet_state.routes), node) {
 282         if((route->route_def.dst_type == LINK_INTERFACE &&
 283              route->route_def.dst_id == dev_id) ||
 284              (route->route_def.src_type == LINK_INTERFACE &&
 285               route->route_def.src_id == dev_id)){
 286
 287             list_del(&(route->node));
 288             list_del(&(route->match_node));
 289             V3_Free(route);
 290         }
 291     }
 292
 293     v3_unlock_irqrestore(vnet_state.lock, flags);
 294 }
 295
 296 /* At the end allocate a route_list
 297  * This list will be inserted into the cache so we don't need to free it
 298  */
 299 static struct route_list * match_route(const struct v3_vnet_pkt * pkt) {
 300     struct vnet_route_info * route = NULL;
 301     struct route_list * matches = NULL;
 302     int num_matches = 0;
 303     int max_rank = 0;
 304     struct list_head match_list;
 305     struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data);
 306  //   uint8_t src_type = pkt->src_type;
 307  //   uint32_t src_link = pkt->src_id;
 308
 309 #ifdef CONFIG_DEBUG_VNET
 310     {
 311         char dst_str[100];
 312         char src_str[100];
 313
 314         mac_to_string(hdr->src_mac, src_str);
 315         mac_to_string(hdr->dst_mac, dst_str);
 316         PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
 317     }
 318 #endif
 319
 320     INIT_LIST_HEAD(&match_list);
 321
 322 #define UPDATE_MATCHES(rank) do {                               \
 323         if (max_rank < (rank)) {                                \
 324             max_rank = (rank);                                  \
 325             INIT_LIST_HEAD(&match_list);                        \
 326                                                                 \
 327             list_add(&(route->match_node), &match_list);        \
 328             num_matches = 1;                                    \
 329         } else if (max_rank == (rank)) {                        \
 330             list_add(&(route->match_node), &match_list);        \
 331             num_matches++;                                      \
 332         }                                                       \
 333     } while (0)
 334
 335
 336     list_for_each_entry(route, &(vnet_state.routes), node) {
 337         struct v3_vnet_route * route_def = &(route->route_def);
 338
 339 /*
 340         // CHECK SOURCE TYPE HERE
 341         if ( (route_def->src_type != LINK_ANY) &&
 342              ( (route_def->src_type != src_type) ||
 343                ( (route_def->src_id != src_link) &&
 344                  (route_def->src_id != -1)))) {
 345             continue;
 346         }
 347 */
 348
 349         if ((route_def->dst_mac_qual == MAC_ANY) &&
 350             (route_def->src_mac_qual == MAC_ANY)) {
 351             UPDATE_MATCHES(3);
 352         }
 353
 354         if (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) {
 355             if (route_def->src_mac_qual != MAC_NOT) {
 356                 if (route_def->dst_mac_qual == MAC_ANY) {
 357                     UPDATE_MATCHES(6);
 358                 } else if (route_def->dst_mac_qual != MAC_NOT &&
 359                            memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
 360                     UPDATE_MATCHES(8);
 361                 }
 362             }
 363         }
 364
 365         if (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) {
 366             if (route_def->dst_mac_qual != MAC_NOT) {
 367                 if (route_def->src_mac_qual == MAC_ANY) {
 368                     UPDATE_MATCHES(6);
 369                 } else if ((route_def->src_mac_qual != MAC_NOT) &&
 370                            (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
 371                     UPDATE_MATCHES(8);
 372                 }
 373             }
 374         }
 375
 376         if ((route_def->dst_mac_qual == MAC_NOT) &&
 377             (memcmp(route_def->dst_mac, hdr->dst_mac, 6) != 0)) {
 378             if (route_def->src_mac_qual == MAC_ANY) {
 379                 UPDATE_MATCHES(5);
 380             } else if ((route_def->src_mac_qual != MAC_NOT) &&
 381                        (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) {
 382                 UPDATE_MATCHES(7);
 383             }
 384         }
 385
 386         if ((route_def->src_mac_qual == MAC_NOT) &&
 387             (memcmp(route_def->src_mac, hdr->src_mac, 6) != 0)) {
 388             if (route_def->dst_mac_qual == MAC_ANY) {
 389                 UPDATE_MATCHES(5);
 390             } else if ((route_def->dst_mac_qual != MAC_NOT) &&
 391                        (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0)) {
 392                 UPDATE_MATCHES(7);
 393             }
 394         }
 395
 396         // Default route
 397         if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) &&
 398              (route_def->dst_mac_qual == MAC_NONE)) {
 399             UPDATE_MATCHES(4);
 400         }
 401     }
 402
 403     PrintDebug("VNET/P Core: match_route: Matches=%d\n", num_matches);
 404
 405     if (num_matches == 0) {
 406         return NULL;
 407     }
 408
 409     matches = (struct route_list *)V3_Malloc(sizeof(struct route_list) +
 410                                 (sizeof(struct vnet_route_info *) * num_matches));
 411
 412     matches->num_routes = num_matches;
 413
 414     {
 415         int i = 0;
 416         list_for_each_entry(route, &match_list, match_node) {
 417             matches->routes[i++] = route;
 418         }
 419     }
 420
 421     return matches;
 422 }
 423
 424
 425 int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) {
 426     struct route_list * matched_routes = NULL;
 427     unsigned long flags;
 428     int i;
 429
 430 #ifdef CONFIG_DEBUG_VNET
 431    {
 432         int cpu = V3_Get_CPU();
 433        PrintDebug("VNET/P Core: cpu %d: pkt (size %d, src_id:%d, src_type: %d, dst_id: %d, dst_type: %d)\n",
 434                   cpu, pkt->size, pkt->src_id,
 435                   pkt->src_type, pkt->dst_id, pkt->dst_type);
 436        //v3_hexdump(pkt->data, pkt->size, NULL, 0);
 437    }
 438 #endif
 439
 440     flags = v3_lock_irqsave(vnet_state.lock);
 441
 442     look_into_cache(pkt, &matched_routes);
 443     if (matched_routes == NULL) {
 444         PrintDebug("VNET/P Core: send pkt Looking into routing table\n");
 445
 446         matched_routes = match_route(pkt);
 447
 448         if (matched_routes) {
 449             add_route_to_cache(pkt, matched_routes);
 450         } else {
 451             PrintDebug("VNET/P Core: Could not find route for packet... discards packet\n");
 452             v3_unlock_irqrestore(vnet_state.lock, flags);
 453             return 0; /* do we return -1 here?*/
 454         }
 455     }
 456
 457     v3_unlock_irqrestore(vnet_state.lock, flags);
 458
 459     PrintDebug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
 460
 461     for (i = 0; i < matched_routes->num_routes; i++) {
 462         struct vnet_route_info * route = matched_routes->routes[i];
 463
 464         if (route->route_def.dst_type == LINK_EDGE) {
 465             struct vnet_brg_dev *bridge = vnet_state.bridge;
 466             pkt->dst_type = LINK_EDGE;
 467             pkt->dst_id = route->route_def.dst_id;
 468
 469             if (bridge == NULL || (bridge->active == 0)) {
 470                 PrintDebug("VNET/P Core: No active bridge to sent data to\n");
 471                  continue;
 472             }
 473
 474             if(bridge->brg_ops.input(bridge->vm, pkt, bridge->private_data) < 0){
 475                 PrintDebug("VNET/P Core: Packet not sent properly to bridge\n");
 476                 continue;
 477             }
 478         } else if (route->route_def.dst_type == LINK_INTERFACE) {
 479             if (route->dst_dev == NULL || route->dst_dev->active == 0){
 480                 PrintDebug("VNET/P Core: No active device to sent data to\n");
 481                 continue;
 482             }
 483
 484             if(route->dst_dev->dev_ops.input(route->dst_dev->vm, pkt, route->dst_dev->private_data) < 0) {
 485                 PrintDebug("VNET/P Core: Packet not sent properly\n");
 486                 continue;
 487             }
 488         } else {
 489             PrintError("VNET/P Core: Wrong dst type\n");
 490         }
 491     }
 492
 493     return 0;
 494 }
 495
 496 int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac,
 497                     struct v3_vnet_dev_ops *ops,
 498                     void * priv_data){
 499     struct vnet_dev * new_dev = NULL;
 500     unsigned long flags;
 501
 502     new_dev = (struct vnet_dev *)V3_Malloc(sizeof(struct vnet_dev));
 503
 504     if (new_dev == NULL) {
 505         PrintError("Malloc fails\n");
 506         return -1;
 507     }
 508
 509     memcpy(new_dev->mac_addr, mac, 6);
 510     new_dev->dev_ops.input = ops->input;
 511     new_dev->dev_ops.poll = ops->poll;
 512     new_dev->private_data = priv_data;
 513     new_dev->vm = vm;
 514     new_dev->dev_id = 0;
 515     new_dev->active = 1;
 516     new_dev->mode = GUEST_DRIVERN;
 517
 518     flags = v3_lock_irqsave(vnet_state.lock);
 519
 520     if (dev_by_mac(mac) == NULL) {
 521         list_add(&(new_dev->node), &(vnet_state.devs));
 522         new_dev->dev_id = ++vnet_state.num_devs;
 523     }
 524
 525     v3_unlock_irqrestore(vnet_state.lock, flags);
 526
 527     /* if the device was found previosly the id should still be 0 */
 528     if (new_dev->dev_id == 0) {
 529         PrintError("VNET/P Core: Device Already exists\n");
 530         return -1;
 531     }
 532
 533     PrintDebug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
 534
 535     return new_dev->dev_id;
 536 }
 537
 538
 539
 540 int v3_vnet_del_dev(int dev_id){
 541     struct vnet_dev * dev = NULL;
 542     unsigned long flags;
 543
 544     flags = v3_lock_irqsave(vnet_state.lock);
 545
 546     dev = dev_by_id(dev_id);
 547     if (dev != NULL){
 548         list_del(&(dev->node));
 549         del_routes_by_dev(dev_id);
 550     }
 551
 552     v3_unlock_irqrestore(vnet_state.lock, flags);
 553
 554     V3_Free(dev);
 555
 556     PrintDebug("VNET/P Core: Remove Device: dev_id %d\n", dev_id);
 557
 558     return 0;
 559 }
 560
 561
 562 static void free_devices(){
 563     struct vnet_dev * dev = NULL;
 564
 565     list_for_each_entry(dev, &(vnet_state.devs), node) {
 566         list_del(&(dev->node));
 567         V3_Free(dev);
 568     }
 569 }
 570
 571 static void free_routes(){
 572     struct vnet_route_info * route = NULL;
 573
 574     list_for_each_entry(route, &(vnet_state.routes), node) {
 575         list_del(&(route->node));
 576         list_del(&(route->match_node));
 577         V3_Free(route);
 578     }
 579 }
 580
 581 int v3_vnet_add_bridge(struct v3_vm_info * vm,
 582                        struct v3_vnet_bridge_ops * ops,
 583                        uint8_t type,
 584                        void * priv_data) {
 585     unsigned long flags;
 586     int bridge_free = 0;
 587     struct vnet_brg_dev * tmp_bridge = NULL;
 588
 589     flags = v3_lock_irqsave(vnet_state.lock);
 590
 591     if (vnet_state.bridge == NULL) {
 592         bridge_free = 1;
 593         vnet_state.bridge = (void *)1;
 594     }
 595
 596     v3_unlock_irqrestore(vnet_state.lock, flags);
 597
 598     if (bridge_free == 0) {
 599         PrintError("VNET/P Core: Bridge already set\n");
 600         return -1;
 601     }
 602
 603     tmp_bridge = (struct vnet_brg_dev *)V3_Malloc(sizeof(struct vnet_brg_dev));
 604
 605     if (tmp_bridge == NULL) {
 606         PrintError("Malloc Fails\n");
 607         vnet_state.bridge = NULL;
 608         return -1;
 609     }
 610
 611     tmp_bridge->vm = vm;
 612     tmp_bridge->brg_ops.input = ops->input;
 613     tmp_bridge->brg_ops.poll = ops->poll;
 614     tmp_bridge->private_data = priv_data;
 615     tmp_bridge->active = 1;
 616     tmp_bridge->mode = GUEST_DRIVERN;
 617     tmp_bridge->type = type;
 618
 619     /* make this atomic to avoid possible race conditions */
 620     flags = v3_lock_irqsave(vnet_state.lock);
 621     vnet_state.bridge = tmp_bridge;
 622     v3_unlock_irqrestore(vnet_state.lock, flags);
 623
 624     return 0;
 625 }
 626
 627
 628 int v3_init_vnet() {
 629     memset(&vnet_state, 0, sizeof(vnet_state));
 630
 631     INIT_LIST_HEAD(&(vnet_state.routes));
 632     INIT_LIST_HEAD(&(vnet_state.devs));
 633
 634     vnet_state.num_devs = 0;
 635     vnet_state.num_routes = 0;
 636
 637     if (v3_lock_init(&(vnet_state.lock)) == -1){
 638         PrintError("VNET/P Core: Fails to initiate lock\n");
 639     }
 640
 641     vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq);
 642
 643     if (vnet_state.route_cache == NULL) {
 644         PrintError("VNET/P Core: Fails to initiate route cache\n");
 645         return -1;
 646     }
 647
 648     PrintDebug("VNET/P Core is initiated\n");
 649
 650     return 0;
 651 }
 652
 653
 654 void v3_deinit_vnet(){
 655
 656     v3_lock_deinit(&(vnet_state.lock));
 657
 658     free_devices();
 659     free_routes();
 660
 661     v3_free_htable(vnet_state.route_cache, 1, 1);
 662     V3_Free(vnet_state.bridge);
 663 }
 664
 665