1 /* See the DRL-LICENSE file for this file's software license. */
4 * ulogd output target for DRL: GRD and FPS
6 * Ken Yocum <kyocum@cs.ucsd.edu>
8 * Original shell of this code from ulogd_NETFLOW
9 * Like that code, we keep track of per-slice data rates
10 * out of this slice. Thus we are rate limiting particular slices
11 * across multiple boxes, ensuring that their outbound rate does not
12 * exceed some fixed limit.
16 * Enforcer: linux drop percentage.
18 * This file reads packets from the netlink socket. It updates all
19 * the hashmaps which track how much data has arrived per flow.
20 * It starts two threads for this limiter.
21 * One thread handles periodic estimation.
22 * The other thread handles communication with other limiters.
27 * ulogd_DRL: attach to netlink socket, accept packets. replaces ratelimit.cc
28 * util.c: generic hashing functions, flow comparisons, sundry items.
29 * gossip.c: Recv gossip, send gossip.
30 * peer_comm.c: Thread to listen for updates from other limiters.
31 * estimate.c: Thread to calculate the local limits.
34 * Ken Yocum <kyocum@cs.ucsd.edu>
37 * Some code appropriated from ulogd_NETFLOW:
39 * Mark Huang <mlhuang@cs.princeton.edu>
40 * Copyright (C) 2004-2005 The Trustees of Princeton University
42 * Based on admindump.pl by Mic Bowman and Paul Brett
43 * Copyright (c) 2002 Intel Corporation
47 /* Enable GNU glibc extensions */
53 /* va_start() and friends */
59 /* strstr() and friends */
62 /* dirname() and basename() */
65 /* fork() and wait() */
66 #include <sys/types.h>
70 /* errno and assert() */
77 /* time() and friends */
82 #include <sys/socket.h>
83 #include <netinet/in.h>
84 #include <arpa/inet.h>
86 /* ICMP definitions */
87 #include <netinet/ip.h>
88 #include <netinet/ip_icmp.h>
93 /* pthread_create() */
99 /* Signal definitions - so that we can catch SIGHUP and update config. */
102 #include <ulogd/ulogd.h>
103 #include <ulogd/conffile.h>
105 /* Perhaps useful for files within vservers? */
106 #if !defined(STANDALONE) && HAVE_LIBPROPER
107 #include <proper/prop.h>
111 * Jenkins hash support
112 * lives in raterouter.h
116 #include "raterouter.h"
118 #include "calendar.h"
119 #include "ratetypes.h" /* needs util and pthread.h */
123 * /etc/ulogd.conf configuration options
124 * Add the config options for DRL.
127 static config_entry_t drl_configfile = {
129 .key = "drl_configfile",
130 .type = CONFIG_TYPE_STRING,
131 .options = CONFIG_OPT_MANDATORY,
132 .u = { .string = "drl.xml" },
135 /** The administrative bandwidth limit (mbps) for the local node. The node
136 * will not set a limit higher than this, even when distributed capacity is
137 * available. Set to 0 for no limit. */
138 static config_entry_t nodelimit = {
139 .next = &drl_configfile,
141 .type = CONFIG_TYPE_INT,
142 .options = CONFIG_OPT_MANDATORY,
146 /** Determines the verbosity of logging. */
147 static config_entry_t drl_loglevel = {
149 .key = "drl_loglevel",
150 .type = CONFIG_TYPE_INT,
151 .options = CONFIG_OPT_MANDATORY,
152 .u = { .value = LOG_WARN },
155 /** The path of the logfile. */
156 static config_entry_t drl_logfile = {
157 .next = &drl_loglevel,
158 .key = "drl_logfile",
159 .type = CONFIG_TYPE_STRING,
160 .options = CONFIG_OPT_MANDATORY,
161 .u = { .string = "drl_logfile.log" },
164 /** The choice of DRL protocol. */
165 static config_entry_t policy = {
166 .next = &drl_logfile,
168 .type = CONFIG_TYPE_STRING,
169 .options = CONFIG_OPT_MANDATORY,
170 .u = { .string = "GRD" },
173 /** The estimate interval, in milliseconds. */
174 static config_entry_t estintms = {
177 .type = CONFIG_TYPE_INT,
178 .options = CONFIG_OPT_MANDATORY,
179 .u = { .value = 100 },
182 #define config_entries (&estintms)
185 * Debug functionality
192 #define NIPQUAD(addr) \
193 ((unsigned char *)&addr)[0], \
194 ((unsigned char *)&addr)[1], \
195 ((unsigned char *)&addr)[2], \
196 ((unsigned char *)&addr)[3]
198 #define IPQUAD(addr) \
199 ((unsigned char *)&addr)[3], \
200 ((unsigned char *)&addr)[2], \
201 ((unsigned char *)&addr)[1], \
202 ((unsigned char *)&addr)[0]
206 /* Salt for the hash functions */
210 * Hash slice name lookups on context ID.
213 /* Special context IDs */
214 #define UNKNOWN_XID -1
218 CONNECTION_REFUSED_XID = 65536, /* MAX_S_CONTEXT + 1 */
225 pthread_t estimate_thread;
226 pthread_t signal_thread;
227 pthread_t comm_thread;
228 uint32_t local_ip = 0;
230 extern FILE *logfile;
231 extern uint8_t system_loglevel;
235 static inline uint32_t
236 hash_flow(uint8_t protocol, uint32_t src_ip, uint16_t src_port, uint32_t dst_ip, uint16_t dst_port)
238 unsigned char mybytes[FLOWKEYSIZE];
239 mybytes[0] = protocol;
240 *(uint32_t*)(&(mybytes[1])) = src_ip;
241 *(uint32_t*)(&(mybytes[5])) = dst_ip;
242 *(uint32_t*)(&(mybytes[9])) = (src_port << 16) | dst_port;
243 return jhash(mybytes,FLOWKEYSIZE,salt) & (FLOW_HASH_SIZE - 1);
246 uint32_t sampled_hasher(const key_flow *key) {
247 return hash_flow(key->protocol, key->source_ip, key->source_port, key->dest_ip, key->dest_port);
250 uint32_t standard_hasher(const key_flow *key) {
251 return hash_flow(key->protocol, key->source_ip, key->source_port, key->dest_ip, key->dest_port);
259 /* Interesting keys */
283 #define INTR_IDS (sizeof(intr_ids)/sizeof(intr_ids[0]))
284 static struct intr_id intr_ids[] = {
285 [OOB_TIME_SEC] = { "oob.time.sec", 0 },
286 [OOB_MARK] = { "oob.mark", 0 },
287 [IP_SADDR] = { "ip.saddr", 0 },
288 [IP_DADDR] = { "ip.daddr", 0 },
289 [IP_TOTLEN] = { "ip.totlen", 0 },
290 [IP_PROTOCOL] = { "ip.protocol", 0 },
291 [TCP_SPORT] = { "tcp.sport", 0 },
292 [TCP_DPORT] { "tcp.dport", 0 },
293 [TCP_ACK] = { "tcp.ack", 0 },
294 [TCP_FIN] = { "tcp.fin", 0 },
295 [TCP_SYN] = { "tcp.syn", 0 },
296 [TCP_RST] = { "tcp.rst", 0 },
297 [UDP_SPORT] = { "udp.sport", 0 },
298 [UDP_DPORT] = { "udp.dport", 0 },
299 [ICMP_TYPE] = { "icmp.type", 0 },
300 [ICMP_CODE] = { "icmp.code", 0 },
301 [GRE_FLAG_KEY] = { "gre.flag.key", 0 },
302 [GRE_VERSION] = { "gre.version", 0 },
303 [GRE_KEY] = { "gre.key", 0 },
304 [PPTP_CALLID] = { "pptp.callid", 0 },
307 #define GET_VALUE(x) intr_ids[x].res->value
309 #define DATE(t) ((t) / (24*60*60) * (24*60*60))
311 static int _output_drl(ulog_iret_t *res)
314 uint32_t src_ip, dst_ip;
315 uint16_t src_port, dst_port;
322 protocol = GET_VALUE(IP_PROTOCOL).ui8;
323 src_ip = GET_VALUE(IP_SADDR).ui32;
324 dst_ip = GET_VALUE(IP_DADDR).ui32;
325 xid = GET_VALUE(OOB_MARK).ui32;
330 src_port = GET_VALUE(TCP_SPORT).ui16;
331 dst_port = GET_VALUE(TCP_DPORT).ui16;
335 /* netflow had an issue with many udp flows and set
336 * src_port=0 to handle it. We don't.
338 src_port = GET_VALUE(UDP_SPORT).ui16;
341 * traceroutes create a large number of flows in the db
342 * this is a quick hack to catch the most common form
343 * of traceroute (basically we're mapping any UDP packet
344 * in the 33435-33524 range to the "trace" port, 33524 is
345 * 3 packets * nhops (30).
347 dst_port = GET_VALUE(UDP_DPORT).ui16;
348 if (dst_port >= 33435 && dst_port <= 33524)
353 src_port = GET_VALUE(ICMP_TYPE).ui8;
354 dst_port = GET_VALUE(ICMP_CODE).ui8;
357 * We special case some of the ICMP traffic that the kernel
358 * always generates. Since this is attributed to root, it
359 * creates significant "noise" in the output. We want to be
360 * able to quickly see that root is generating traffic.
362 if (xid == ROOT_XID) {
363 if (src_port == ICMP_ECHOREPLY)
364 xid = ICMP_ECHOREPLY_XID;
365 else if (src_port == ICMP_UNREACH)
366 xid = ICMP_UNREACH_XID;
371 if (GET_VALUE(GRE_FLAG_KEY).b) {
372 if (GET_VALUE(GRE_VERSION).ui8 == 1) {
373 /* Get PPTP call ID */
374 src_port = GET_VALUE(PPTP_CALLID).ui16;
376 /* XXX Truncate GRE keys to 16 bits */
377 src_port = (uint16_t) GET_VALUE(GRE_KEY).ui32;
380 /* No key available */
387 /* This is the default key for packets from unsupported protocols */
393 key.protocol = protocol;
394 key.source_ip = src_ip;
395 key.dest_ip = dst_ip;
396 key.source_port = src_port;
397 key.dest_port = dst_port;
398 key.packet_size = GET_VALUE(IP_TOTLEN).ui16;
399 key.packet_time = (time_t) GET_VALUE(OOB_TIME_SEC).ui32;
401 pthread_rwlock_rdlock(&limiter.limiter_lock); /* CLUNK! */
403 leaf = (leaf_t *) map_search(limiter.stable_instance.leaf_map, &xid, sizeof(xid));
405 /* Even if the packet doesn't match any specific xid, it should still
406 * count in the machine-type tables. This catches root (xid == 0) and
407 * unclassified (xid = fff) packets, which don't have map entries. */
409 ident = limiter.stable_instance.last_machine;
411 ident = leaf->parent;
415 pthread_mutex_lock(&ident->table_mutex);
417 /* Update the identity's table. */
418 ident->table_sample_function(ident->table, &key);
420 pthread_mutex_unlock(&ident->table_mutex);
422 ident = ident->parent;
425 pthread_rwlock_unlock(&limiter.limiter_lock); /* CLINK! */
430 /* get all key id's for the keys we are intrested in */
431 static int get_ids(void)
434 struct intr_id *cur_id;
436 for (i = 0; i < INTR_IDS; i++) {
437 cur_id = &intr_ids[i];
438 cur_id->res = keyh_getres(keyh_getid(cur_id->name));
440 ulogd_log(ULOGD_ERROR,
441 "Cannot resolve keyhash id for %s\n",
449 static void free_identity(identity_t *ident) {
451 free_comm(&ident->comm);
454 ident->table_destroy_function(ident->table);
457 pthread_mutex_destroy(&ident->table_mutex);
463 static void free_identity_map(map_handle map) {
464 identity_t *tofree = NULL;
466 map_reset_iterate(map);
467 while ((tofree = (identity_t *) map_next(map))) {
468 free_identity(tofree);
474 static void free_instance(drl_instance_t *instance) {
475 if (instance->leaves)
476 free(instance->leaves);
477 if (instance->leaf_map)
478 free_map(instance->leaf_map, 0);
479 if (instance->ident_map)
480 free_identity_map(instance->ident_map);
481 if (instance->machines)
482 free(instance->machines);
484 free(instance->sets);
489 memset(instance, 0, sizeof(drl_instance_t));
492 static void free_failed_config(parsed_configs configs, drl_instance_t *instance) {
494 if (configs.machines)
495 free_ident_list(configs.machines);
497 free_ident_list(configs.sets);
501 free_instance(instance);
504 static identity_t *new_identity(ident_config *config) {
505 identity_t *ident = malloc(sizeof(identity_t));
506 remote_node_t *comm_nodes = malloc(sizeof(remote_node_t)*config->peer_count);
507 ident_peer *peer = config->peers;
514 if (comm_nodes == NULL) {
519 memset(ident, 0, sizeof(identity_t));
520 memset(comm_nodes, 0, config->peer_count * sizeof(remote_node_t));
522 ident->id = config->id;
523 ident->limit = (uint32_t) (((double) config->limit * 1000000.0) / 8.0);
524 ident->fixed_ewma_weight = config->fixed_ewma_weight;
525 ident->intervals = config->intervals;
526 ident->ewma_weight = pow(ident->fixed_ewma_weight,
527 (limiter.estintms/1000.0) * config->intervals);
528 ident->parent = NULL;
530 pthread_mutex_init(&ident->table_mutex, NULL);
531 switch (config->accounting) {
534 standard_table_create(standard_hasher, &ident->common);
536 /* Ugly function pointer casting. Makes things sufficiently
537 * generic, though. */
538 ident->table_sample_function =
539 (int (*)(void *, const key_flow *)) standard_table_sample;
540 ident->table_cleanup_function =
541 (int (*)(void *)) standard_table_cleanup;
542 ident->table_update_function =
543 (void (*)(void *, struct timeval, double)) standard_table_update_flows;
544 ident->table_destroy_function =
545 (void (*)(void *)) standard_table_destroy;
549 ident->table = sampled_table_create(sampled_hasher,
550 ident->limit * IDENT_CLEAN_INTERVAL,
551 1, 20, &ident->common);
553 ident->table_sample_function =
554 (int (*)(void *, const key_flow *)) sampled_table_sample;
555 ident->table_cleanup_function =
556 (int (*)(void *)) sampled_table_cleanup;
557 ident->table_update_function =
558 (void (*)(void *, struct timeval, double)) sampled_table_update_flows;
559 ident->table_destroy_function =
560 (void (*)(void *)) sampled_table_destroy;
564 ident->table = simple_table_create(&ident->common);
566 ident->table_sample_function =
567 (int (*)(void *, const key_flow *)) simple_table_sample;
568 ident->table_cleanup_function =
569 (int (*)(void *)) simple_table_cleanup;
570 ident->table_update_function =
571 (void (*)(void *, struct timeval, double)) simple_table_update_flows;
572 ident->table_destroy_function =
573 (void (*)(void *)) simple_table_destroy;
577 /* Make sure the table was allocated. */
578 if (ident->table == NULL) {
584 comm_nodes[peer_slot].addr = peer->ip;
585 comm_nodes[peer_slot].port = htons(LIMITER_LISTEN_PORT);
590 if (new_comm(&ident->comm, config, comm_nodes)) {
591 printlog(LOG_CRITICAL, "Failed to create communication structure.\n");
595 ident->comm.remote_nodes = comm_nodes;
600 /* Determines the validity of the parameters of one ident_config.
605 static int validate_config(ident_config *config) {
606 /* Limit must be a positive integer. */
607 if (config->limit < 1) {
611 /* Commfabric must be a valid choice (COMM_MESH or COMM_GOSSIP). */
612 if (config->commfabric != COMM_MESH &&
613 config->commfabric != COMM_GOSSIP) {
617 /* If commfabric is COMM_GOSSIP, this must be a positive integer. */
618 if (config->commfabric == COMM_GOSSIP && config->branch < 1) {
622 /* Accounting must be a valid choice (ACT_STANDARD, ACT_SAMPLEHOLD,
624 if (config->accounting != ACT_STANDARD &&
625 config->accounting != ACT_SAMPLEHOLD &&
626 config->accounting != ACT_SIMPLE) {
630 /* Ewma weight must be greater than or equal to zero. */
631 if (config->fixed_ewma_weight < 0) {
635 /* Note: Parsing stage requires that each ident has at least one peer. */
642 static int validate_configs(parsed_configs configs, drl_instance_t *instance) {
644 ident_config *mlist = configs.machines;
645 ident_config *slist = configs.sets;
646 ident_config *tmp = NULL;
650 /* ID must be non-zero and unique. */
651 /* This is ugly and hackish, but this function will be called rarely.
652 * I'm tired of trying to be clever. */
654 printlog(LOG_CRITICAL, "Negative ident id: %d (%x) ?\n", mlist->id, mlist->id);
659 if (mlist->id == tmp->id) {
660 printlog(LOG_CRITICAL, "Duplicate ident id: %d (%x)\n", mlist->id, mlist->id);
667 if (mlist->id == tmp->id) {
668 printlog(LOG_CRITICAL, "Duplicate ident id: %d (%x)\n", mlist->id, mlist->id);
674 if (validate_config(mlist)) {
675 printlog(LOG_CRITICAL, "Invalid ident parameters for id: %d (%x)\n", mlist->id, mlist->id);
682 instance->sets = malloc(configs.set_count * sizeof(identity_t *));
683 if (instance->sets == NULL) {
687 memset(instance->sets, 0, configs.set_count * sizeof(identity_t *));
688 instance->set_count = configs.set_count;
690 /* For sets, make sure that the hierarchy is valid. */
692 ident_member *members = slist->members;
694 /* ID must be non-zero and unique. */
696 printlog(LOG_CRITICAL, "Negative ident id: %d (%x) ?\n", slist->id, slist->id);
701 if (slist->id == tmp->id) {
702 printlog(LOG_CRITICAL, "Duplicate ident id: %d (%x)\n", slist->id, slist->id);
708 if (validate_config(slist)) {
709 printlog(LOG_CRITICAL, "Invalid ident parameters for id: %d (%x)\n", slist->id, slist->id);
713 /* Allocate an identity_t for this set-type identity. */
714 instance->sets[i] = new_identity(slist);
716 if (instance->sets[i] == NULL) {
720 /* Loop through children and look up each in leaf or ident map
721 * depending on the type of child. Set the child's parent pointer
722 * to the identity we just created above, unless it is already set,
723 * in which case we have an error. */
725 identity_t *child_ident = NULL;
726 leaf_t *child_leaf = NULL;
728 switch (members->type) {
730 child_leaf = map_search(instance->leaf_map, &members->value,
731 sizeof(members->value));
732 if (child_leaf == NULL) {
735 if (child_leaf->parent != NULL) {
736 /* Error - This leaf already has a parent. */
739 child_leaf->parent = instance->sets[i];
742 child_ident = map_search(instance->ident_map, &members->value,
743 sizeof(members->value));
744 if (child_ident == NULL) {
747 if (child_ident->parent != NULL) {
748 /* Error - This identity already has a parent. */
751 child_ident->parent = instance->sets[i];
754 /* Error - shouldn't be possible. */
757 members = members->next;
760 map_insert(instance->ident_map, &instance->sets[i]->id,
761 sizeof(instance->sets[i]->id), instance->sets[i]);
769 static int fill_set_leaf_pointer(drl_instance_t *instance, identity_t *ident) {
771 identity_t *current_ident;
772 leaf_t *current_leaf;
773 leaf_t **leaves = malloc(instance->leaf_count * sizeof(leaf_t *));
774 if (leaves == NULL) {
778 map_reset_iterate(instance->leaf_map);
779 while ((current_leaf = (leaf_t *) map_next(instance->leaf_map))) {
780 current_ident = current_leaf->parent;
781 while (current_ident != NULL && current_ident != instance->last_machine) {
782 if (current_ident == ident) {
783 /* Found the ident we were looking for - add the leaf. */
784 leaves[count] = current_leaf;
788 current_ident = current_ident->parent;
792 ident->leaves = leaves;
793 ident->leaf_count = count;
798 static int init_identities(parsed_configs configs, drl_instance_t *instance) {
800 ident_config *config = configs.machines;
803 instance->cal = malloc(sizeof(struct ident_calendar) * SCHEDLEN);
805 if (instance->cal == NULL) {
809 for (i = 0; i < SCHEDLEN; ++i) {
810 TAILQ_INIT(instance->cal + i);
812 instance->cal_slot = 0;
814 instance->machines = malloc(configs.machine_count * sizeof(drl_instance_t *));
816 if (instance->machines == NULL) {
820 memset(instance->machines, 0, configs.machine_count * sizeof(drl_instance_t *));
821 instance->machine_count = configs.machine_count;
823 /* Allocate and add the machine identities. */
824 for (i = 0; i < configs.machine_count; ++i) {
825 instance->machines[i] = new_identity(config);
827 if (instance->machines[i] == NULL) {
831 /* The first has no parent - it is the root. All others have the
832 * previous ident as their parent. */
834 instance->machines[i]->parent = NULL;
836 instance->machines[i]->parent = instance->machines[i - 1];
839 instance->last_machine = instance->machines[i];
841 /* Add the ident to the guid->ident map. */
842 map_insert(instance->ident_map, &instance->machines[i]->id,
843 sizeof(instance->machines[i]->id), instance->machines[i]);
845 config = config->next;
847 TAILQ_INSERT_TAIL(instance->cal + (instance->cal_slot & SCHEDMASK),
848 instance->machines[i], calendar);
850 /* Setup the array of pointers to leaves. This is easy for machines
851 * because a machine node applies to every leaf. */
852 instance->machines[i]->leaves =
853 malloc(instance->leaf_count * sizeof(leaf_t *));
854 if (instance->machines[i]->leaves == NULL) {
857 instance->machines[i]->leaf_count = instance->leaf_count;
858 for (j = 0; j < instance->leaf_count; ++j) {
859 instance->machines[i]->leaves[j] = &instance->leaves[j];
863 /* Connect the set subtree to the machines. Any set or leaf without a
864 * parent will take the last machine as its parent. */
867 map_reset_iterate(instance->leaf_map);
868 while ((leaf = (leaf_t *) map_next(instance->leaf_map))) {
869 if (leaf->parent == NULL) {
870 leaf->parent = instance->last_machine;
875 for (i = 0; i < instance->set_count; ++i) {
876 if (instance->sets[i]->parent == NULL) {
877 instance->sets[i]->parent = instance->last_machine;
880 TAILQ_INSERT_TAIL(instance->cal + (instance->cal_slot & SCHEDMASK),
881 instance->sets[i], calendar);
883 /* Setup the array of pointers to leaves. This is harder for sets,
884 * but this doesn't need to be super-efficient because it happens
885 * rarely and it isn't on the critical path for reconfig(). */
886 if (fill_set_leaf_pointer(instance, instance->sets[i])) {
895 static void print_instance(drl_instance_t *instance) {
897 identity_t *ident = NULL;
899 if (system_loglevel == LOG_DEBUG) {
900 map_reset_iterate(instance->leaf_map);
901 while ((leaf = (leaf_t *) map_next(instance->leaf_map))) {
902 printf("%x:", leaf->xid);
903 ident = leaf->parent;
905 printf("%d:",ident->id);
906 ident = ident->parent;
908 printf("Leaf's parent pointer is %p\n", leaf->parent);
911 printf("instance->last_machine is %p\n", instance->last_machine);
915 static int assign_htb_hierarchy(drl_instance_t *instance) {
917 int next_node = 0x11;
919 /* Chain machine nodes under 1:10. */
920 for (i = 0; i < instance->machine_count; ++i) {
921 if (instance->machines[i]->parent == NULL) {
923 instance->machines[i]->htb_parent = 0x10;
926 instance->machines[i]->htb_parent =
927 instance->machines[i]->parent->htb_node;
930 instance->machines[i]->htb_node = next_node;
936 /* Add set nodes under machine nodes. Iterate backwards to ensure parent is
938 for (j = (instance->set_count - 1); j >= 0; --j) {
939 if (instance->sets[j]->parent == NULL) {
940 instance->sets[j]->htb_parent = 0x10;
942 instance->sets[j]->htb_parent = instance->sets[j]->parent->htb_node;
944 instance->sets[j]->htb_node = next_node;
952 /* Added this so that I could comment one line and kill off all of the
953 * command execution. */
954 static int execute_cmd(const char *cmd) {
958 static int create_htb_hierarchy(drl_instance_t *instance) {
962 /* Nuke the hierarchy. */
963 sprintf(cmd, "tc qdisc del dev eth0 root handle 1: htb");
965 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
967 /* Re-initialize the basics. */
968 sprintf(cmd, "tc qdisc add dev eth0 root handle 1: htb default 1fff");
969 if (execute_cmd(cmd)) {
972 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
973 sprintf(cmd, "tc class add dev eth0 parent 1: classid 1:1 htb rate 1000mbit ceil 1000mbit");
974 if (execute_cmd(cmd)) {
977 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
979 /* Add back 1:10. (Nodelimit : Megabits/sec -> bits/second)*/
980 if (limiter.nodelimit) {
981 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:1 classid 1:10 htb rate 8bit ceil %lubit",
982 (unsigned long) limiter.nodelimit * 1024 * 1024);
984 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:1 classid 1:10 htb rate 8bit ceil 1000mbit");
987 if (execute_cmd(cmd)) {
990 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
993 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:1 classid 1:20 htb rate 8bit ceil 1000mbit");
995 if (execute_cmd(cmd)) {
998 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1002 for (i = 0; i < instance->machine_count; ++i) {
1003 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:%x classid 1:%x htb rate 8bit ceil %lubit",
1004 instance->machines[i]->htb_parent,
1005 instance->machines[i]->htb_node,
1006 (unsigned long) instance->machines[i]->limit * 1024 * 1024);
1008 if (execute_cmd(cmd)) {
1011 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1015 for (j = (instance->set_count - 1); j >= 0; --j) {
1016 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:%x classid 1:%x htb rate 8bit ceil %lubit",
1017 instance->sets[j]->htb_parent,
1018 instance->sets[j]->htb_node,
1019 (unsigned long) instance->sets[j]->limit * 1024 * 1024);
1021 if (execute_cmd(cmd)) {
1024 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1027 /* Add leaves. FIXME: Set static sliver limit as ceil here! */
1028 for (k = 0; k < instance->leaf_count; ++k) {
1029 if (instance->leaves[k].parent == NULL) {
1030 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:10 classid 1:1%x htb rate 8bit ceil %lubit",
1031 instance->leaves[k].xid,
1032 (unsigned long) 100 * 1024 * 1024);
1034 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:%x classid 1:1%x htb rate 8bit ceil %lubit",
1035 instance->leaves[k].parent->htb_node,
1036 instance->leaves[k].xid,
1037 (unsigned long) 100 * 1024 * 1024);
1040 if (execute_cmd(cmd)) {
1043 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1045 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:20 classid 1:2%x htb rate 8bit ceil 1000mbit",
1046 instance->leaves[k].xid);
1048 if (execute_cmd(cmd)) {
1051 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1054 /* Add 1:1000 and 1:2000 */
1055 if (instance->last_machine == NULL) {
1056 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:10 classid 1:1000 htb rate 8bit ceil 1000mbit");
1058 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:%x classid 1:1000 htb rate 8bit ceil 1000mbit",
1059 instance->last_machine->htb_node);
1062 if (execute_cmd(cmd)) {
1065 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1066 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:20 classid 1:2000 htb rate 8bit ceil 1000mbit");
1068 if (execute_cmd(cmd)) {
1071 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1073 /* Add 1:1fff and 1:2fff */
1074 if (instance->last_machine == NULL) {
1075 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:10 classid 1:1fff htb rate 8bit ceil 1000mbit");
1077 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:%x classid 1:1fff htb rate 8bit ceil 1000mbit",
1078 instance->last_machine->htb_node);
1081 if (execute_cmd(cmd)) {
1084 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1085 sprintf(cmd, "/sbin/tc class add dev eth0 parent 1:20 classid 1:2fff htb rate 8bit ceil 1000mbit");
1087 if (execute_cmd(cmd)) {
1090 printlog(LOG_DEBUG, "HTB_cmd: %s\n", cmd);
1093 /* Only for artificial delay testing. */
1094 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:1000 handle 1000 pfifo");
1097 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1000 handle 1000 netem loss 0 delay 40ms");
1099 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:11f9 handle 11f9 pfifo");
1102 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:11f9 handle 11f9 netem loss 0 delay 40ms");
1104 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:11fa handle 11fa pfifo");
1107 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:11fa handle 11fa netem loss 0 delay 40ms");
1109 /* End delay testing */
1115 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:1000 handle 1000 pfifo");
1118 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1000 handle 1000 sfq perturb 20");
1125 static int setup_tc_grd(drl_instance_t *instance) {
1129 for (i = 0; i < instance->leaf_count; ++i) {
1130 /* Delete the old pfifo qdisc that might have been there before. */
1131 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:1%x handle 1%x pfifo",
1132 instance->leaves[i].xid, instance->leaves[i].xid);
1134 if (execute_cmd(cmd)) {
1135 printlog(LOG_DEBUG, "GRD: pfifo qdisc wasn't there!\n");
1138 /* Add the netem qdisc. */
1140 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1%x handle 1%x netem loss 0 delay 40ms",
1141 instance->leaves[i].xid, instance->leaves[i].xid);
1143 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1%x handle 1%x netem loss 0 delay 0ms",
1144 instance->leaves[i].xid, instance->leaves[i].xid);
1147 if (execute_cmd(cmd)) {
1152 /* Do the same for 1000 and 1fff. */
1153 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:1000 handle 1000 pfifo");
1155 if (execute_cmd(cmd)) {
1156 printlog(LOG_DEBUG, "GRD: pfifo qdisc wasn't there!\n");
1159 /* Add the netem qdisc. */
1161 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1000 handle 1000 netem loss 0 delay 40ms");
1163 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1000 handle 1000 netem loss 0 delay 0ms");
1166 if (execute_cmd(cmd)) {
1170 sprintf(cmd, "/sbin/tc qdisc del dev eth0 parent 1:1fff handle 1fff pfifo");
1172 if (execute_cmd(cmd)) {
1173 printlog(LOG_DEBUG, "GRD: pfifo qdisc wasn't there!\n");
1176 /* Add the netem qdisc. */
1178 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1fff handle 1fff netem loss 0 delay 40ms");
1180 sprintf(cmd, "/sbin/tc qdisc replace dev eth0 parent 1:1fff handle 1fff netem loss 0 delay 0ms");
1183 if (execute_cmd(cmd)) {
1192 * Initialize this limiter with options
1193 * Open UDP socket for peer communication
1195 static int init_drl(void) {
1196 parsed_configs configs;
1197 struct sockaddr_in server_address;
1199 memset(&limiter, 0, sizeof(limiter_t));
1201 /* Setup logging. */
1202 system_loglevel = (uint8_t) drl_loglevel.u.value;
1203 logfile = fopen(drl_logfile.u.string, "w");
1205 if (logfile == NULL) {
1206 printf("Couldn't open logfile - ");
1211 printlog(LOG_CRITICAL, "ulogd_DRL initializing . . .\n");
1213 limiter.nodelimit = (uint32_t) (((double) nodelimit.u.value * 1000000.0) / 8.0);
1215 init_hashing(); /* for all hash maps */
1217 pthread_rwlock_init(&limiter.limiter_lock,NULL);
1219 /* determine our local IP by iterating through interfaces */
1220 if ((limiter.ip = get_local_ip())==0) {
1221 printlog(LOG_CRITICAL,
1222 "ulogd_DRL unable to aquire local IP address, not registering.\n");
1225 limiter.localaddr = inet_addr(limiter.ip);
1226 limiter.port = htons(LIMITER_LISTEN_PORT);
1227 limiter.udp_socket = socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP);
1228 if (limiter.udp_socket < 0) {
1229 printlog(LOG_CRITICAL, "Failed to create UDP socket().\n");
1233 memset(&server_address, 0, sizeof(server_address));
1234 server_address.sin_family = AF_INET;
1235 server_address.sin_addr.s_addr = limiter.localaddr;
1236 server_address.sin_port = limiter.port;
1238 if (bind(limiter.udp_socket, (struct sockaddr *) &server_address, sizeof(server_address)) < 0) {
1239 printlog(LOG_CRITICAL, "Failed to bind UDP socket.\n");
1243 printlog(LOG_WARN, " POLICY: %s\n",policy.u.string);
1244 if (strcasecmp(policy.u.string,"GRD") == 0) {
1245 limiter.policy = POLICY_GRD;
1246 } else if (strcasecmp(policy.u.string,"FPS") == 0) {
1247 limiter.policy = POLICY_FPS;
1249 printlog(LOG_CRITICAL,
1250 "Unknown DRL policy %s, aborting.\n",policy.u.string);
1254 limiter.estintms = estintms.u.value;
1255 if (limiter.estintms > 1000) {
1256 printlog(LOG_CRITICAL,
1257 "DRL: sorry estimate intervals must be less than 1 second.");
1258 printlog(LOG_CRITICAL,
1259 " Simple source mods will allow larger intervals. Using 1 second.\n");
1260 limiter.estintms = 1000;
1262 printlog(LOG_WARN, " Est interval: %dms\n",limiter.estintms);
1264 /* Acquire the big limiter lock for writing. Prevents pretty much
1265 * anything else from happening while the hierarchy is being changed. */
1266 pthread_rwlock_wrlock(&limiter.limiter_lock);
1268 limiter.stable_instance.ident_map = allocate_map();
1269 if (limiter.stable_instance.ident_map == NULL) {
1270 printlog(LOG_CRITICAL, "Failed to allocate memory for identity map.\n");
1274 if (get_eligible_leaves(&limiter.stable_instance)) {
1275 printlog(LOG_CRITICAL, "Failed to read eligigle leaves.\n");
1279 if (parse_drl_config(drl_configfile.u.string, &configs)) {
1280 /* Parse error occured. Return non-zero to notify init_drl(). */
1284 /* Validate identity hierarchy! */
1285 if (validate_configs(configs, &limiter.stable_instance)) {
1286 /* Clean up everything. */
1287 free_failed_config(configs, &limiter.stable_instance);
1291 if (init_identities(configs, &limiter.stable_instance)) {
1292 free_failed_config(configs, &limiter.stable_instance);
1296 /* At this point, we should be done with configs. */
1297 free_ident_list(configs.machines);
1298 free_ident_list(configs.sets);
1300 /* Debugging - FIXME: remove this? */
1301 print_instance(&limiter.stable_instance);
1303 switch (limiter.policy) {
1305 if (assign_htb_hierarchy(&limiter.stable_instance)) {
1306 free_instance(&limiter.stable_instance);
1310 if (create_htb_hierarchy(&limiter.stable_instance)) {
1311 free_instance(&limiter.stable_instance);
1317 if (setup_tc_grd(&limiter.stable_instance)) {
1318 free_instance(&limiter.stable_instance);
1327 pthread_rwlock_unlock(&limiter.limiter_lock);
1329 if (pthread_create(&limiter.udp_recv_thread, NULL, limiter_receive_thread, NULL)) {
1330 printlog(LOG_CRITICAL, "Unable to start UDP receive thread.\n");
1334 printlog(LOG_WARN, "ulogd_DRL init finished.\n");
1339 static void reconfig() {
1340 parsed_configs configs;
1342 printlog(LOG_DEBUG, "--Starting reconfig()--\n");
1345 memset(&configs, 0, sizeof(parsed_configs));
1346 memset(&limiter.new_instance, 0, sizeof(drl_instance_t));
1348 limiter.new_instance.ident_map = allocate_map();
1349 if (limiter.new_instance.ident_map == NULL) {
1350 printlog(LOG_CRITICAL, "Failed to allocate ident_map during reconfig().\n");
1354 if (get_eligible_leaves(&limiter.new_instance)) {
1355 free_failed_config(configs, &limiter.new_instance);
1356 printlog(LOG_CRITICAL, "Failed to read leaves during reconfig().\n");
1360 if (parse_drl_config(drl_configfile.u.string, &configs)) {
1361 free_failed_config(configs, &limiter.new_instance);
1362 printlog(LOG_CRITICAL, "Failed to parse config during reconfig().\n");
1366 if (validate_configs(configs, &limiter.new_instance)) {
1367 free_failed_config(configs, &limiter.new_instance);
1368 printlog(LOG_CRITICAL, "Validation failed during reconfig().\n");
1369 pthread_rwlock_unlock(&limiter.limiter_lock);
1373 if (init_identities(configs, &limiter.new_instance)) {
1374 free_failed_config(configs, &limiter.new_instance);
1375 printlog(LOG_CRITICAL, "Initialization failed during reconfig().\n");
1376 pthread_rwlock_unlock(&limiter.limiter_lock);
1380 free_ident_list(configs.machines);
1381 free_ident_list(configs.sets);
1383 /* Debugging - FIXME: remove this? */
1384 print_instance(&limiter.new_instance);
1387 pthread_rwlock_wrlock(&limiter.limiter_lock);
1389 switch (limiter.policy) {
1391 if (assign_htb_hierarchy(&limiter.new_instance)) {
1392 free_instance(&limiter.new_instance);
1393 printlog(LOG_CRITICAL, "Failed to assign HTB hierarchy during reconfig().\n");
1394 pthread_rwlock_unlock(&limiter.limiter_lock);
1398 if (create_htb_hierarchy(&limiter.new_instance)) {
1399 free_instance(&limiter.new_instance);
1400 printlog(LOG_CRITICAL, "Failed to create HTB hierarchy during reconfig().\n");
1402 /* Re-create old instance. */
1403 if (create_htb_hierarchy(&limiter.stable_instance)) {
1404 /* Error reinstating the old one - big problem. */
1405 printlog(LOG_CRITICAL, "Failed to reinstate HTB hierarchy during reconfig().\n");
1406 printlog(LOG_CRITICAL, "Giving up...\n");
1411 pthread_rwlock_unlock(&limiter.limiter_lock);
1417 if (setup_tc_grd(&limiter.new_instance)) {
1418 free_instance(&limiter.new_instance);
1419 printlog(LOG_CRITICAL, "GRD tc calls failed during reconfig().\n");
1421 /* Try to re-create old instance. */
1422 if (setup_tc_grd(&limiter.stable_instance)) {
1423 printlog(LOG_CRITICAL, "Failed to reinstate old GRD qdiscs during reconfig().\n");
1424 printlog(LOG_CRITICAL, "Giving up...\n");
1432 /* Should be impossible. */
1433 printf("Pigs are flying?\n");
1437 /* Switch over new to stable instance. */
1438 free_instance(&limiter.stable_instance);
1439 memcpy(&limiter.stable_instance, &limiter.new_instance, sizeof(drl_instance_t));
1441 /* Success! - Unlock */
1442 pthread_rwlock_unlock(&limiter.limiter_lock);
1445 static ulog_output_t drl_op = {
1447 .output = &_output_drl,
1448 .signal = NULL, /* This appears to be broken. Using my own handler. */
1453 /* Tests the amount of time it takes to call reconfig(). */
1454 static void time_reconfig(int iterations) {
1455 struct timeval start, end;
1458 gettimeofday(&start, NULL);
1459 for (i = 0; i < iterations; ++i) {
1462 gettimeofday(&end, NULL);
1464 printf("%d reconfigs() took %d seconds and %d microseconds.\n",
1465 iterations, end.tv_sec - start.tv_sec, end.tv_usec - start.tv_usec);
1468 // Seems to take about 85ms / iteration
1471 static void *signal_thread_func(void *args) {
1477 sigaddset(&sigs, SIGHUP);
1478 pthread_sigmask(SIG_BLOCK, &sigs, NULL);
1482 sigaddset(&sigs, SIGHUP);
1484 err = sigwait(&sigs, &sig);
1487 printlog(LOG_CRITICAL, "sigwait() returned an error.\n");
1493 printlog(LOG_WARN, "Caught SIGHUP - re-reading XML file.\n");
1495 //time_reconfig(1000); //instrumentation
1499 /* Should be impossible... */
1506 /* register output plugin with ulogd */
1507 static void _drl_reg_op(void)
1509 ulog_output_t *op = &drl_op;
1510 sigset_t signal_mask;
1512 sigemptyset(&signal_mask);
1513 sigaddset(&signal_mask, SIGHUP);
1514 pthread_sigmask(SIG_BLOCK, &signal_mask, NULL);
1516 if (pthread_create(&signal_thread, NULL, &signal_thread_func, NULL) != 0) {
1517 printlog(LOG_CRITICAL, "Failed to create signal handling thread.\n");
1518 fprintf(stderr, "An error has occured starting ulogd_DRL. Refer to your logfile (%s) for additional information.\n", drl_logfile.u.string);
1524 printlog(LOG_CRITICAL, "Init failed. :(\n");
1525 fprintf(stderr, "An error has occured starting ulogd_DRL. Refer to your logfile (%s) for additional information.\n", drl_logfile.u.string);
1530 register_output(op);
1532 /* start up the thread that will periodically estimate the
1533 * local rate and set the local limits
1536 if (pthread_create(&estimate_thread, NULL, (void*(*)(void*)) &handle_estimation, &limiter)!=0) {
1537 printlog(LOG_CRITICAL, "Couldn't start estimate thread.\n");
1538 fprintf(stderr, "An error has occured starting ulogd_DRL. Refer to your logfile (%s) for additional information.\n", drl_logfile.u.string);
1545 /* have the opts parsed */
1546 config_parse_file("DRL", config_entries);
1549 ulogd_log(ULOGD_ERROR, "can't resolve all keyhash id's\n");
1553 /* Seed the hash function */
1554 salt = getpid() ^ time(NULL);