1 /* 2 * IPVS An implementation of the IP virtual server support for the 3 * LINUX operating system. IPVS is now implemented as a module 4 * over the NetFilter framework. IPVS can be used to build a 5 * high-performance and highly available server based on a 6 * cluster of servers. 7 * 8 * Version 1, is capable of handling both version 0 and 1 messages. 9 * Version 0 is the plain old format. 10 * Note Version 0 receivers will just drop Ver 1 messages. 11 * Version 1 is capable of handle IPv6, Persistence data, 12 * time-outs, and firewall marks. 13 * In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order. 14 * Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0 15 * 16 * Definitions Message: is a complete datagram 17 * Sync_conn: is a part of a Message 18 * Param Data is an option to a Sync_conn. 19 * 20 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> 21 * 22 * ip_vs_sync: sync connection info from master load balancer to backups 23 * through multicast 24 * 25 * Changes: 26 * Alexandre Cassen : Added master & backup support at a time. 27 * Alexandre Cassen : Added SyncID support for incoming sync 28 * messages filtering. 29 * Justin Ossevoort : Fix endian problem on sync message size. 30 * Hans Schillstrom : Added Version 1: i.e. IPv6, 31 * Persistence support, fwmark and time-out. 32 */ 33 34 #define KMSG_COMPONENT "IPVS" 35 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 36 37 #include <linux/module.h> 38 #include <linux/slab.h> 39 #include <linux/inetdevice.h> 40 #include <linux/net.h> 41 #include <linux/completion.h> 42 #include <linux/delay.h> 43 #include <linux/skbuff.h> 44 #include <linux/in.h> 45 #include <linux/igmp.h> /* for ip_mc_join_group */ 46 #include <linux/udp.h> 47 #include <linux/err.h> 48 #include <linux/kthread.h> 49 #include <linux/wait.h> 50 #include <linux/kernel.h> 51 52 #include <asm/unaligned.h> /* Used for ntoh_seq and hton_seq */ 53 54 #include <net/ip.h> 55 #include <net/sock.h> 56 57 #include <net/ip_vs.h> 58 59 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */ 60 #define IP_VS_SYNC_PORT 8848 /* multicast port */ 61 62 #define SYNC_PROTO_VER 1 /* Protocol version in header */ 63 64 static struct lock_class_key __ipvs_sync_key; 65 /* 66 * IPVS sync connection entry 67 * Version 0, i.e. original version. 68 */ 69 struct ip_vs_sync_conn_v0 { 70 __u8 reserved; 71 72 /* Protocol, addresses and port numbers */ 73 __u8 protocol; /* Which protocol (TCP/UDP) */ 74 __be16 cport; 75 __be16 vport; 76 __be16 dport; 77 __be32 caddr; /* client address */ 78 __be32 vaddr; /* virtual address */ 79 __be32 daddr; /* destination address */ 80 81 /* Flags and state transition */ 82 __be16 flags; /* status flags */ 83 __be16 state; /* state info */ 84 85 /* The sequence options start here */ 86 }; 87 88 struct ip_vs_sync_conn_options { 89 struct ip_vs_seq in_seq; /* incoming seq. struct */ 90 struct ip_vs_seq out_seq; /* outgoing seq. struct */ 91 }; 92 93 /* 94 Sync Connection format (sync_conn) 95 96 0 1 2 3 97 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 98 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 99 | Type | Protocol | Ver. | Size | 100 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 101 | Flags | 102 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 103 | State | cport | 104 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 105 | vport | dport | 106 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 107 | fwmark | 108 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 109 | timeout (in sec.) | 110 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 111 | ... | 112 | IP-Addresses (v4 or v6) | 113 | ... | 114 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 115 Optional Parameters. 116 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 117 | Param. Type | Param. Length | Param. data | 118 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | 119 | ... | 120 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 121 | | Param Type | Param. Length | 122 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 123 | Param data | 124 | Last Param data should be padded for 32 bit alignment | 125 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 126 */ 127 128 /* 129 * Type 0, IPv4 sync connection format 130 */ 131 struct ip_vs_sync_v4 { 132 __u8 type; 133 __u8 protocol; /* Which protocol (TCP/UDP) */ 134 __be16 ver_size; /* Version msb 4 bits */ 135 /* Flags and state transition */ 136 __be32 flags; /* status flags */ 137 __be16 state; /* state info */ 138 /* Protocol, addresses and port numbers */ 139 __be16 cport; 140 __be16 vport; 141 __be16 dport; 142 __be32 fwmark; /* Firewall mark from skb */ 143 __be32 timeout; /* cp timeout */ 144 __be32 caddr; /* client address */ 145 __be32 vaddr; /* virtual address */ 146 __be32 daddr; /* destination address */ 147 /* The sequence options start here */ 148 /* PE data padded to 32bit alignment after seq. options */ 149 }; 150 /* 151 * Type 2 messages IPv6 152 */ 153 struct ip_vs_sync_v6 { 154 __u8 type; 155 __u8 protocol; /* Which protocol (TCP/UDP) */ 156 __be16 ver_size; /* Version msb 4 bits */ 157 /* Flags and state transition */ 158 __be32 flags; /* status flags */ 159 __be16 state; /* state info */ 160 /* Protocol, addresses and port numbers */ 161 __be16 cport; 162 __be16 vport; 163 __be16 dport; 164 __be32 fwmark; /* Firewall mark from skb */ 165 __be32 timeout; /* cp timeout */ 166 struct in6_addr caddr; /* client address */ 167 struct in6_addr vaddr; /* virtual address */ 168 struct in6_addr daddr; /* destination address */ 169 /* The sequence options start here */ 170 /* PE data padded to 32bit alignment after seq. options */ 171 }; 172 173 union ip_vs_sync_conn { 174 struct ip_vs_sync_v4 v4; 175 struct ip_vs_sync_v6 v6; 176 }; 177 178 /* Bits in Type field in above */ 179 #define STYPE_INET6 0 180 #define STYPE_F_INET6 (1 << STYPE_INET6) 181 182 #define SVER_SHIFT 12 /* Shift to get version */ 183 #define SVER_MASK 0x0fff /* Mask to strip version */ 184 185 #define IPVS_OPT_SEQ_DATA 1 186 #define IPVS_OPT_PE_DATA 2 187 #define IPVS_OPT_PE_NAME 3 188 #define IPVS_OPT_PARAM 7 189 190 #define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1)) 191 #define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1)) 192 #define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1)) 193 #define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1)) 194 195 struct ip_vs_sync_thread_data { 196 struct net *net; 197 struct socket *sock; 198 char *buf; 199 int id; 200 }; 201 202 /* Version 0 definition of packet sizes */ 203 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0)) 204 #define FULL_CONN_SIZE \ 205 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options)) 206 207 208 /* 209 The master mulitcasts messages (Datagrams) to the backup load balancers 210 in the following format. 211 212 Version 1: 213 Note, first byte should be Zero, so ver 0 receivers will drop the packet. 214 215 0 1 2 3 216 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 217 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 218 | 0 | SyncID | Size | 219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 220 | Count Conns | Version | Reserved, set to Zero | 221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 222 | | 223 | IPVS Sync Connection (1) | 224 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 225 | . | 226 ~ . ~ 227 | . | 228 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 229 | | 230 | IPVS Sync Connection (n) | 231 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 232 233 Version 0 Header 234 0 1 2 3 235 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 236 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 237 | Count Conns | SyncID | Size | 238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 239 | IPVS Sync Connection (1) | 240 */ 241 242 #define SYNC_MESG_HEADER_LEN 4 243 #define MAX_CONNS_PER_SYNCBUFF 255 /* nr_conns in ip_vs_sync_mesg is 8 bit */ 244 245 /* Version 0 header */ 246 struct ip_vs_sync_mesg_v0 { 247 __u8 nr_conns; 248 __u8 syncid; 249 __be16 size; 250 251 /* ip_vs_sync_conn entries start here */ 252 }; 253 254 /* Version 1 header */ 255 struct ip_vs_sync_mesg { 256 __u8 reserved; /* must be zero */ 257 __u8 syncid; 258 __be16 size; 259 __u8 nr_conns; 260 __s8 version; /* SYNC_PROTO_VER */ 261 __u16 spare; 262 /* ip_vs_sync_conn entries start here */ 263 }; 264 265 struct ip_vs_sync_buff { 266 struct list_head list; 267 unsigned long firstuse; 268 269 /* pointers for the message data */ 270 struct ip_vs_sync_mesg *mesg; 271 unsigned char *head; 272 unsigned char *end; 273 }; 274 275 /* 276 * Copy of struct ip_vs_seq 277 * From unaligned network order to aligned host order 278 */ 279 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho) 280 { 281 ho->init_seq = get_unaligned_be32(&no->init_seq); 282 ho->delta = get_unaligned_be32(&no->delta); 283 ho->previous_delta = get_unaligned_be32(&no->previous_delta); 284 } 285 286 /* 287 * Copy of struct ip_vs_seq 288 * From Aligned host order to unaligned network order 289 */ 290 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no) 291 { 292 put_unaligned_be32(ho->init_seq, &no->init_seq); 293 put_unaligned_be32(ho->delta, &no->delta); 294 put_unaligned_be32(ho->previous_delta, &no->previous_delta); 295 } 296 297 static inline struct ip_vs_sync_buff * 298 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms) 299 { 300 struct ip_vs_sync_buff *sb; 301 302 spin_lock_bh(&ipvs->sync_lock); 303 if (list_empty(&ms->sync_queue)) { 304 sb = NULL; 305 __set_current_state(TASK_INTERRUPTIBLE); 306 } else { 307 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff, 308 list); 309 list_del(&sb->list); 310 ms->sync_queue_len--; 311 if (!ms->sync_queue_len) 312 ms->sync_queue_delay = 0; 313 } 314 spin_unlock_bh(&ipvs->sync_lock); 315 316 return sb; 317 } 318 319 /* 320 * Create a new sync buffer for Version 1 proto. 321 */ 322 static inline struct ip_vs_sync_buff * 323 ip_vs_sync_buff_create(struct netns_ipvs *ipvs) 324 { 325 struct ip_vs_sync_buff *sb; 326 327 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) 328 return NULL; 329 330 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC); 331 if (!sb->mesg) { 332 kfree(sb); 333 return NULL; 334 } 335 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */ 336 sb->mesg->version = SYNC_PROTO_VER; 337 sb->mesg->syncid = ipvs->master_syncid; 338 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg)); 339 sb->mesg->nr_conns = 0; 340 sb->mesg->spare = 0; 341 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg); 342 sb->end = (unsigned char *)sb->mesg + ipvs->send_mesg_maxlen; 343 344 sb->firstuse = jiffies; 345 return sb; 346 } 347 348 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) 349 { 350 kfree(sb->mesg); 351 kfree(sb); 352 } 353 354 static inline void sb_queue_tail(struct netns_ipvs *ipvs, 355 struct ipvs_master_sync_state *ms) 356 { 357 struct ip_vs_sync_buff *sb = ms->sync_buff; 358 359 spin_lock(&ipvs->sync_lock); 360 if (ipvs->sync_state & IP_VS_STATE_MASTER && 361 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) { 362 if (!ms->sync_queue_len) 363 schedule_delayed_work(&ms->master_wakeup_work, 364 max(IPVS_SYNC_SEND_DELAY, 1)); 365 ms->sync_queue_len++; 366 list_add_tail(&sb->list, &ms->sync_queue); 367 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE) 368 wake_up_process(ms->master_thread); 369 } else 370 ip_vs_sync_buff_release(sb); 371 spin_unlock(&ipvs->sync_lock); 372 } 373 374 /* 375 * Get the current sync buffer if it has been created for more 376 * than the specified time or the specified time is zero. 377 */ 378 static inline struct ip_vs_sync_buff * 379 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms, 380 unsigned long time) 381 { 382 struct ip_vs_sync_buff *sb; 383 384 spin_lock_bh(&ipvs->sync_buff_lock); 385 sb = ms->sync_buff; 386 if (sb && time_after_eq(jiffies - sb->firstuse, time)) { 387 ms->sync_buff = NULL; 388 __set_current_state(TASK_RUNNING); 389 } else 390 sb = NULL; 391 spin_unlock_bh(&ipvs->sync_buff_lock); 392 return sb; 393 } 394 395 static inline int 396 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp) 397 { 398 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask; 399 } 400 401 /* 402 * Create a new sync buffer for Version 0 proto. 403 */ 404 static inline struct ip_vs_sync_buff * 405 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs) 406 { 407 struct ip_vs_sync_buff *sb; 408 struct ip_vs_sync_mesg_v0 *mesg; 409 410 if (!(sb=kmalloc(sizeof(struct ip_vs_sync_buff), GFP_ATOMIC))) 411 return NULL; 412 413 sb->mesg = kmalloc(ipvs->send_mesg_maxlen, GFP_ATOMIC); 414 if (!sb->mesg) { 415 kfree(sb); 416 return NULL; 417 } 418 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg; 419 mesg->nr_conns = 0; 420 mesg->syncid = ipvs->master_syncid; 421 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0)); 422 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0); 423 sb->end = (unsigned char *)mesg + ipvs->send_mesg_maxlen; 424 sb->firstuse = jiffies; 425 return sb; 426 } 427 428 /* Check if connection is controlled by persistence */ 429 static inline bool in_persistence(struct ip_vs_conn *cp) 430 { 431 for (cp = cp->control; cp; cp = cp->control) { 432 if (cp->flags & IP_VS_CONN_F_TEMPLATE) 433 return true; 434 } 435 return false; 436 } 437 438 /* Check if conn should be synced. 439 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check 440 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry 441 * sync_retries times with period of sync_refresh_period/8 442 * - (2) if both sync_refresh_period and sync_period are 0 send sync only 443 * for state changes or only once when pkts matches sync_threshold 444 * - (3) templates: rate can be reduced only with sync_refresh_period or 445 * with (2) 446 */ 447 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs, 448 struct ip_vs_conn *cp, int pkts) 449 { 450 unsigned long orig = ACCESS_ONCE(cp->sync_endtime); 451 unsigned long now = jiffies; 452 unsigned long n = (now + cp->timeout) & ~3UL; 453 unsigned int sync_refresh_period; 454 int sync_period; 455 int force; 456 457 /* Check if we sync in current state */ 458 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE)) 459 force = 0; 460 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp))) 461 return 0; 462 else if (likely(cp->protocol == IPPROTO_TCP)) { 463 if (!((1 << cp->state) & 464 ((1 << IP_VS_TCP_S_ESTABLISHED) | 465 (1 << IP_VS_TCP_S_FIN_WAIT) | 466 (1 << IP_VS_TCP_S_CLOSE) | 467 (1 << IP_VS_TCP_S_CLOSE_WAIT) | 468 (1 << IP_VS_TCP_S_TIME_WAIT)))) 469 return 0; 470 force = cp->state != cp->old_state; 471 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED) 472 goto set; 473 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) { 474 if (!((1 << cp->state) & 475 ((1 << IP_VS_SCTP_S_ESTABLISHED) | 476 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) | 477 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) | 478 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) | 479 (1 << IP_VS_SCTP_S_CLOSED)))) 480 return 0; 481 force = cp->state != cp->old_state; 482 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED) 483 goto set; 484 } else { 485 /* UDP or another protocol with single state */ 486 force = 0; 487 } 488 489 sync_refresh_period = sysctl_sync_refresh_period(ipvs); 490 if (sync_refresh_period > 0) { 491 long diff = n - orig; 492 long min_diff = max(cp->timeout >> 1, 10UL * HZ); 493 494 /* Avoid sync if difference is below sync_refresh_period 495 * and below the half timeout. 496 */ 497 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) { 498 int retries = orig & 3; 499 500 if (retries >= sysctl_sync_retries(ipvs)) 501 return 0; 502 if (time_before(now, orig - cp->timeout + 503 (sync_refresh_period >> 3))) 504 return 0; 505 n |= retries + 1; 506 } 507 } 508 sync_period = sysctl_sync_period(ipvs); 509 if (sync_period > 0) { 510 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) && 511 pkts % sync_period != sysctl_sync_threshold(ipvs)) 512 return 0; 513 } else if (sync_refresh_period <= 0 && 514 pkts != sysctl_sync_threshold(ipvs)) 515 return 0; 516 517 set: 518 cp->old_state = cp->state; 519 n = cmpxchg(&cp->sync_endtime, orig, n); 520 return n == orig || force; 521 } 522 523 /* 524 * Version 0 , could be switched in by sys_ctl. 525 * Add an ip_vs_conn information into the current sync_buff. 526 */ 527 static void ip_vs_sync_conn_v0(struct net *net, struct ip_vs_conn *cp, 528 int pkts) 529 { 530 struct netns_ipvs *ipvs = net_ipvs(net); 531 struct ip_vs_sync_mesg_v0 *m; 532 struct ip_vs_sync_conn_v0 *s; 533 struct ip_vs_sync_buff *buff; 534 struct ipvs_master_sync_state *ms; 535 int id; 536 int len; 537 538 if (unlikely(cp->af != AF_INET)) 539 return; 540 /* Do not sync ONE PACKET */ 541 if (cp->flags & IP_VS_CONN_F_ONE_PACKET) 542 return; 543 544 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts)) 545 return; 546 547 spin_lock_bh(&ipvs->sync_buff_lock); 548 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) { 549 spin_unlock_bh(&ipvs->sync_buff_lock); 550 return; 551 } 552 553 id = select_master_thread_id(ipvs, cp); 554 ms = &ipvs->ms[id]; 555 buff = ms->sync_buff; 556 if (buff) { 557 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg; 558 /* Send buffer if it is for v1 */ 559 if (!m->nr_conns) { 560 sb_queue_tail(ipvs, ms); 561 ms->sync_buff = NULL; 562 buff = NULL; 563 } 564 } 565 if (!buff) { 566 buff = ip_vs_sync_buff_create_v0(ipvs); 567 if (!buff) { 568 spin_unlock_bh(&ipvs->sync_buff_lock); 569 pr_err("ip_vs_sync_buff_create failed.\n"); 570 return; 571 } 572 ms->sync_buff = buff; 573 } 574 575 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : 576 SIMPLE_CONN_SIZE; 577 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg; 578 s = (struct ip_vs_sync_conn_v0 *) buff->head; 579 580 /* copy members */ 581 s->reserved = 0; 582 s->protocol = cp->protocol; 583 s->cport = cp->cport; 584 s->vport = cp->vport; 585 s->dport = cp->dport; 586 s->caddr = cp->caddr.ip; 587 s->vaddr = cp->vaddr.ip; 588 s->daddr = cp->daddr.ip; 589 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); 590 s->state = htons(cp->state); 591 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { 592 struct ip_vs_sync_conn_options *opt = 593 (struct ip_vs_sync_conn_options *)&s[1]; 594 memcpy(opt, &cp->in_seq, sizeof(*opt)); 595 } 596 597 m->nr_conns++; 598 m->size = htons(ntohs(m->size) + len); 599 buff->head += len; 600 601 /* check if there is a space for next one */ 602 if (buff->head + FULL_CONN_SIZE > buff->end) { 603 sb_queue_tail(ipvs, ms); 604 ms->sync_buff = NULL; 605 } 606 spin_unlock_bh(&ipvs->sync_buff_lock); 607 608 /* synchronize its controller if it has */ 609 cp = cp->control; 610 if (cp) { 611 if (cp->flags & IP_VS_CONN_F_TEMPLATE) 612 pkts = atomic_add_return(1, &cp->in_pkts); 613 else 614 pkts = sysctl_sync_threshold(ipvs); 615 ip_vs_sync_conn(net, cp->control, pkts); 616 } 617 } 618 619 /* 620 * Add an ip_vs_conn information into the current sync_buff. 621 * Called by ip_vs_in. 622 * Sending Version 1 messages 623 */ 624 void ip_vs_sync_conn(struct net *net, struct ip_vs_conn *cp, int pkts) 625 { 626 struct netns_ipvs *ipvs = net_ipvs(net); 627 struct ip_vs_sync_mesg *m; 628 union ip_vs_sync_conn *s; 629 struct ip_vs_sync_buff *buff; 630 struct ipvs_master_sync_state *ms; 631 int id; 632 __u8 *p; 633 unsigned int len, pe_name_len, pad; 634 635 /* Handle old version of the protocol */ 636 if (sysctl_sync_ver(ipvs) == 0) { 637 ip_vs_sync_conn_v0(net, cp, pkts); 638 return; 639 } 640 /* Do not sync ONE PACKET */ 641 if (cp->flags & IP_VS_CONN_F_ONE_PACKET) 642 goto control; 643 sloop: 644 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts)) 645 goto control; 646 647 /* Sanity checks */ 648 pe_name_len = 0; 649 if (cp->pe_data_len) { 650 if (!cp->pe_data || !cp->dest) { 651 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n"); 652 return; 653 } 654 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN); 655 } 656 657 spin_lock_bh(&ipvs->sync_buff_lock); 658 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) { 659 spin_unlock_bh(&ipvs->sync_buff_lock); 660 return; 661 } 662 663 id = select_master_thread_id(ipvs, cp); 664 ms = &ipvs->ms[id]; 665 666 #ifdef CONFIG_IP_VS_IPV6 667 if (cp->af == AF_INET6) 668 len = sizeof(struct ip_vs_sync_v6); 669 else 670 #endif 671 len = sizeof(struct ip_vs_sync_v4); 672 673 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) 674 len += sizeof(struct ip_vs_sync_conn_options) + 2; 675 676 if (cp->pe_data_len) 677 len += cp->pe_data_len + 2; /* + Param hdr field */ 678 if (pe_name_len) 679 len += pe_name_len + 2; 680 681 /* check if there is a space for this one */ 682 pad = 0; 683 buff = ms->sync_buff; 684 if (buff) { 685 m = buff->mesg; 686 pad = (4 - (size_t) buff->head) & 3; 687 /* Send buffer if it is for v0 */ 688 if (buff->head + len + pad > buff->end || m->reserved) { 689 sb_queue_tail(ipvs, ms); 690 ms->sync_buff = NULL; 691 buff = NULL; 692 pad = 0; 693 } 694 } 695 696 if (!buff) { 697 buff = ip_vs_sync_buff_create(ipvs); 698 if (!buff) { 699 spin_unlock_bh(&ipvs->sync_buff_lock); 700 pr_err("ip_vs_sync_buff_create failed.\n"); 701 return; 702 } 703 ms->sync_buff = buff; 704 m = buff->mesg; 705 } 706 707 p = buff->head; 708 buff->head += pad + len; 709 m->size = htons(ntohs(m->size) + pad + len); 710 /* Add ev. padding from prev. sync_conn */ 711 while (pad--) 712 *(p++) = 0; 713 714 s = (union ip_vs_sync_conn *)p; 715 716 /* Set message type & copy members */ 717 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0); 718 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */ 719 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED); 720 s->v4.state = htons(cp->state); 721 s->v4.protocol = cp->protocol; 722 s->v4.cport = cp->cport; 723 s->v4.vport = cp->vport; 724 s->v4.dport = cp->dport; 725 s->v4.fwmark = htonl(cp->fwmark); 726 s->v4.timeout = htonl(cp->timeout / HZ); 727 m->nr_conns++; 728 729 #ifdef CONFIG_IP_VS_IPV6 730 if (cp->af == AF_INET6) { 731 p += sizeof(struct ip_vs_sync_v6); 732 s->v6.caddr = cp->caddr.in6; 733 s->v6.vaddr = cp->vaddr.in6; 734 s->v6.daddr = cp->daddr.in6; 735 } else 736 #endif 737 { 738 p += sizeof(struct ip_vs_sync_v4); /* options ptr */ 739 s->v4.caddr = cp->caddr.ip; 740 s->v4.vaddr = cp->vaddr.ip; 741 s->v4.daddr = cp->daddr.ip; 742 } 743 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { 744 *(p++) = IPVS_OPT_SEQ_DATA; 745 *(p++) = sizeof(struct ip_vs_sync_conn_options); 746 hton_seq((struct ip_vs_seq *)p, &cp->in_seq); 747 p += sizeof(struct ip_vs_seq); 748 hton_seq((struct ip_vs_seq *)p, &cp->out_seq); 749 p += sizeof(struct ip_vs_seq); 750 } 751 /* Handle pe data */ 752 if (cp->pe_data_len && cp->pe_data) { 753 *(p++) = IPVS_OPT_PE_DATA; 754 *(p++) = cp->pe_data_len; 755 memcpy(p, cp->pe_data, cp->pe_data_len); 756 p += cp->pe_data_len; 757 if (pe_name_len) { 758 /* Add PE_NAME */ 759 *(p++) = IPVS_OPT_PE_NAME; 760 *(p++) = pe_name_len; 761 memcpy(p, cp->pe->name, pe_name_len); 762 p += pe_name_len; 763 } 764 } 765 766 spin_unlock_bh(&ipvs->sync_buff_lock); 767 768 control: 769 /* synchronize its controller if it has */ 770 cp = cp->control; 771 if (!cp) 772 return; 773 if (cp->flags & IP_VS_CONN_F_TEMPLATE) 774 pkts = atomic_add_return(1, &cp->in_pkts); 775 else 776 pkts = sysctl_sync_threshold(ipvs); 777 goto sloop; 778 } 779 780 /* 781 * fill_param used by version 1 782 */ 783 static inline int 784 ip_vs_conn_fill_param_sync(struct net *net, int af, union ip_vs_sync_conn *sc, 785 struct ip_vs_conn_param *p, 786 __u8 *pe_data, unsigned int pe_data_len, 787 __u8 *pe_name, unsigned int pe_name_len) 788 { 789 #ifdef CONFIG_IP_VS_IPV6 790 if (af == AF_INET6) 791 ip_vs_conn_fill_param(net, af, sc->v6.protocol, 792 (const union nf_inet_addr *)&sc->v6.caddr, 793 sc->v6.cport, 794 (const union nf_inet_addr *)&sc->v6.vaddr, 795 sc->v6.vport, p); 796 else 797 #endif 798 ip_vs_conn_fill_param(net, af, sc->v4.protocol, 799 (const union nf_inet_addr *)&sc->v4.caddr, 800 sc->v4.cport, 801 (const union nf_inet_addr *)&sc->v4.vaddr, 802 sc->v4.vport, p); 803 /* Handle pe data */ 804 if (pe_data_len) { 805 if (pe_name_len) { 806 char buff[IP_VS_PENAME_MAXLEN+1]; 807 808 memcpy(buff, pe_name, pe_name_len); 809 buff[pe_name_len]=0; 810 p->pe = __ip_vs_pe_getbyname(buff); 811 if (!p->pe) { 812 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n", 813 buff); 814 return 1; 815 } 816 } else { 817 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n"); 818 return 1; 819 } 820 821 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC); 822 if (!p->pe_data) { 823 if (p->pe->module) 824 module_put(p->pe->module); 825 return -ENOMEM; 826 } 827 p->pe_data_len = pe_data_len; 828 } 829 return 0; 830 } 831 832 /* 833 * Connection Add / Update. 834 * Common for version 0 and 1 reception of backup sync_conns. 835 * Param: ... 836 * timeout is in sec. 837 */ 838 static void ip_vs_proc_conn(struct net *net, struct ip_vs_conn_param *param, 839 unsigned int flags, unsigned int state, 840 unsigned int protocol, unsigned int type, 841 const union nf_inet_addr *daddr, __be16 dport, 842 unsigned long timeout, __u32 fwmark, 843 struct ip_vs_sync_conn_options *opt) 844 { 845 struct ip_vs_dest *dest; 846 struct ip_vs_conn *cp; 847 struct netns_ipvs *ipvs = net_ipvs(net); 848 849 if (!(flags & IP_VS_CONN_F_TEMPLATE)) 850 cp = ip_vs_conn_in_get(param); 851 else 852 cp = ip_vs_ct_in_get(param); 853 854 if (cp) { 855 /* Free pe_data */ 856 kfree(param->pe_data); 857 858 dest = cp->dest; 859 spin_lock_bh(&cp->lock); 860 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE && 861 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) { 862 if (flags & IP_VS_CONN_F_INACTIVE) { 863 atomic_dec(&dest->activeconns); 864 atomic_inc(&dest->inactconns); 865 } else { 866 atomic_inc(&dest->activeconns); 867 atomic_dec(&dest->inactconns); 868 } 869 } 870 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK; 871 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK; 872 cp->flags = flags; 873 spin_unlock_bh(&cp->lock); 874 if (!dest) 875 ip_vs_try_bind_dest(cp); 876 } else { 877 /* 878 * Find the appropriate destination for the connection. 879 * If it is not found the connection will remain unbound 880 * but still handled. 881 */ 882 rcu_read_lock(); 883 dest = ip_vs_find_dest(net, type, daddr, dport, param->vaddr, 884 param->vport, protocol, fwmark, flags); 885 886 cp = ip_vs_conn_new(param, daddr, dport, flags, dest, fwmark); 887 rcu_read_unlock(); 888 if (!cp) { 889 if (param->pe_data) 890 kfree(param->pe_data); 891 IP_VS_DBG(2, "BACKUP, add new conn. failed\n"); 892 return; 893 } 894 } 895 896 if (opt) 897 memcpy(&cp->in_seq, opt, sizeof(*opt)); 898 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs)); 899 cp->state = state; 900 cp->old_state = cp->state; 901 /* 902 * For Ver 0 messages style 903 * - Not possible to recover the right timeout for templates 904 * - can not find the right fwmark 905 * virtual service. If needed, we can do it for 906 * non-fwmark persistent services. 907 * Ver 1 messages style. 908 * - No problem. 909 */ 910 if (timeout) { 911 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ) 912 timeout = MAX_SCHEDULE_TIMEOUT / HZ; 913 cp->timeout = timeout*HZ; 914 } else { 915 struct ip_vs_proto_data *pd; 916 917 pd = ip_vs_proto_data_get(net, protocol); 918 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table) 919 cp->timeout = pd->timeout_table[state]; 920 else 921 cp->timeout = (3*60*HZ); 922 } 923 ip_vs_conn_put(cp); 924 } 925 926 /* 927 * Process received multicast message for Version 0 928 */ 929 static void ip_vs_process_message_v0(struct net *net, const char *buffer, 930 const size_t buflen) 931 { 932 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer; 933 struct ip_vs_sync_conn_v0 *s; 934 struct ip_vs_sync_conn_options *opt; 935 struct ip_vs_protocol *pp; 936 struct ip_vs_conn_param param; 937 char *p; 938 int i; 939 940 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0); 941 for (i=0; i<m->nr_conns; i++) { 942 unsigned int flags, state; 943 944 if (p + SIMPLE_CONN_SIZE > buffer+buflen) { 945 IP_VS_ERR_RL("BACKUP v0, bogus conn\n"); 946 return; 947 } 948 s = (struct ip_vs_sync_conn_v0 *) p; 949 flags = ntohs(s->flags) | IP_VS_CONN_F_SYNC; 950 flags &= ~IP_VS_CONN_F_HASHED; 951 if (flags & IP_VS_CONN_F_SEQ_MASK) { 952 opt = (struct ip_vs_sync_conn_options *)&s[1]; 953 p += FULL_CONN_SIZE; 954 if (p > buffer+buflen) { 955 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n"); 956 return; 957 } 958 } else { 959 opt = NULL; 960 p += SIMPLE_CONN_SIZE; 961 } 962 963 state = ntohs(s->state); 964 if (!(flags & IP_VS_CONN_F_TEMPLATE)) { 965 pp = ip_vs_proto_get(s->protocol); 966 if (!pp) { 967 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n", 968 s->protocol); 969 continue; 970 } 971 if (state >= pp->num_states) { 972 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n", 973 pp->name, state); 974 continue; 975 } 976 } else { 977 /* protocol in templates is not used for state/timeout */ 978 if (state > 0) { 979 IP_VS_DBG(2, "BACKUP v0, Invalid template state %u\n", 980 state); 981 state = 0; 982 } 983 } 984 985 ip_vs_conn_fill_param(net, AF_INET, s->protocol, 986 (const union nf_inet_addr *)&s->caddr, 987 s->cport, 988 (const union nf_inet_addr *)&s->vaddr, 989 s->vport, ¶m); 990 991 /* Send timeout as Zero */ 992 ip_vs_proc_conn(net, ¶m, flags, state, s->protocol, AF_INET, 993 (union nf_inet_addr *)&s->daddr, s->dport, 994 0, 0, opt); 995 } 996 } 997 998 /* 999 * Handle options 1000 */ 1001 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen, 1002 __u32 *opt_flags, 1003 struct ip_vs_sync_conn_options *opt) 1004 { 1005 struct ip_vs_sync_conn_options *topt; 1006 1007 topt = (struct ip_vs_sync_conn_options *)p; 1008 1009 if (plen != sizeof(struct ip_vs_sync_conn_options)) { 1010 IP_VS_DBG(2, "BACKUP, bogus conn options length\n"); 1011 return -EINVAL; 1012 } 1013 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) { 1014 IP_VS_DBG(2, "BACKUP, conn options found twice\n"); 1015 return -EINVAL; 1016 } 1017 ntoh_seq(&topt->in_seq, &opt->in_seq); 1018 ntoh_seq(&topt->out_seq, &opt->out_seq); 1019 *opt_flags |= IPVS_OPT_F_SEQ_DATA; 1020 return 0; 1021 } 1022 1023 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len, 1024 __u8 **data, unsigned int maxlen, 1025 __u32 *opt_flags, __u32 flag) 1026 { 1027 if (plen > maxlen) { 1028 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen); 1029 return -EINVAL; 1030 } 1031 if (*opt_flags & flag) { 1032 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag); 1033 return -EINVAL; 1034 } 1035 *data_len = plen; 1036 *data = p; 1037 *opt_flags |= flag; 1038 return 0; 1039 } 1040 /* 1041 * Process a Version 1 sync. connection 1042 */ 1043 static inline int ip_vs_proc_sync_conn(struct net *net, __u8 *p, __u8 *msg_end) 1044 { 1045 struct ip_vs_sync_conn_options opt; 1046 union ip_vs_sync_conn *s; 1047 struct ip_vs_protocol *pp; 1048 struct ip_vs_conn_param param; 1049 __u32 flags; 1050 unsigned int af, state, pe_data_len=0, pe_name_len=0; 1051 __u8 *pe_data=NULL, *pe_name=NULL; 1052 __u32 opt_flags=0; 1053 int retc=0; 1054 1055 s = (union ip_vs_sync_conn *) p; 1056 1057 if (s->v6.type & STYPE_F_INET6) { 1058 #ifdef CONFIG_IP_VS_IPV6 1059 af = AF_INET6; 1060 p += sizeof(struct ip_vs_sync_v6); 1061 #else 1062 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n"); 1063 retc = 10; 1064 goto out; 1065 #endif 1066 } else if (!s->v4.type) { 1067 af = AF_INET; 1068 p += sizeof(struct ip_vs_sync_v4); 1069 } else { 1070 return -10; 1071 } 1072 if (p > msg_end) 1073 return -20; 1074 1075 /* Process optional params check Type & Len. */ 1076 while (p < msg_end) { 1077 int ptype; 1078 int plen; 1079 1080 if (p+2 > msg_end) 1081 return -30; 1082 ptype = *(p++); 1083 plen = *(p++); 1084 1085 if (!plen || ((p + plen) > msg_end)) 1086 return -40; 1087 /* Handle seq option p = param data */ 1088 switch (ptype & ~IPVS_OPT_F_PARAM) { 1089 case IPVS_OPT_SEQ_DATA: 1090 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt)) 1091 return -50; 1092 break; 1093 1094 case IPVS_OPT_PE_DATA: 1095 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data, 1096 IP_VS_PEDATA_MAXLEN, &opt_flags, 1097 IPVS_OPT_F_PE_DATA)) 1098 return -60; 1099 break; 1100 1101 case IPVS_OPT_PE_NAME: 1102 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name, 1103 IP_VS_PENAME_MAXLEN, &opt_flags, 1104 IPVS_OPT_F_PE_NAME)) 1105 return -70; 1106 break; 1107 1108 default: 1109 /* Param data mandatory ? */ 1110 if (!(ptype & IPVS_OPT_F_PARAM)) { 1111 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n", 1112 ptype & ~IPVS_OPT_F_PARAM); 1113 retc = 20; 1114 goto out; 1115 } 1116 } 1117 p += plen; /* Next option */ 1118 } 1119 1120 /* Get flags and Mask off unsupported */ 1121 flags = ntohl(s->v4.flags) & IP_VS_CONN_F_BACKUP_MASK; 1122 flags |= IP_VS_CONN_F_SYNC; 1123 state = ntohs(s->v4.state); 1124 1125 if (!(flags & IP_VS_CONN_F_TEMPLATE)) { 1126 pp = ip_vs_proto_get(s->v4.protocol); 1127 if (!pp) { 1128 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n", 1129 s->v4.protocol); 1130 retc = 30; 1131 goto out; 1132 } 1133 if (state >= pp->num_states) { 1134 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n", 1135 pp->name, state); 1136 retc = 40; 1137 goto out; 1138 } 1139 } else { 1140 /* protocol in templates is not used for state/timeout */ 1141 if (state > 0) { 1142 IP_VS_DBG(3, "BACKUP, Invalid template state %u\n", 1143 state); 1144 state = 0; 1145 } 1146 } 1147 if (ip_vs_conn_fill_param_sync(net, af, s, ¶m, pe_data, 1148 pe_data_len, pe_name, pe_name_len)) { 1149 retc = 50; 1150 goto out; 1151 } 1152 /* If only IPv4, just silent skip IPv6 */ 1153 if (af == AF_INET) 1154 ip_vs_proc_conn(net, ¶m, flags, state, s->v4.protocol, af, 1155 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport, 1156 ntohl(s->v4.timeout), ntohl(s->v4.fwmark), 1157 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL) 1158 ); 1159 #ifdef CONFIG_IP_VS_IPV6 1160 else 1161 ip_vs_proc_conn(net, ¶m, flags, state, s->v6.protocol, af, 1162 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport, 1163 ntohl(s->v6.timeout), ntohl(s->v6.fwmark), 1164 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL) 1165 ); 1166 #endif 1167 return 0; 1168 /* Error exit */ 1169 out: 1170 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc); 1171 return retc; 1172 1173 } 1174 /* 1175 * Process received multicast message and create the corresponding 1176 * ip_vs_conn entries. 1177 * Handles Version 0 & 1 1178 */ 1179 static void ip_vs_process_message(struct net *net, __u8 *buffer, 1180 const size_t buflen) 1181 { 1182 struct netns_ipvs *ipvs = net_ipvs(net); 1183 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer; 1184 __u8 *p, *msg_end; 1185 int i, nr_conns; 1186 1187 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) { 1188 IP_VS_DBG(2, "BACKUP, message header too short\n"); 1189 return; 1190 } 1191 1192 if (buflen != ntohs(m2->size)) { 1193 IP_VS_DBG(2, "BACKUP, bogus message size\n"); 1194 return; 1195 } 1196 /* SyncID sanity check */ 1197 if (ipvs->backup_syncid != 0 && m2->syncid != ipvs->backup_syncid) { 1198 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid); 1199 return; 1200 } 1201 /* Handle version 1 message */ 1202 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0) 1203 && (m2->spare == 0)) { 1204 1205 msg_end = buffer + sizeof(struct ip_vs_sync_mesg); 1206 nr_conns = m2->nr_conns; 1207 1208 for (i=0; i<nr_conns; i++) { 1209 union ip_vs_sync_conn *s; 1210 unsigned int size; 1211 int retc; 1212 1213 p = msg_end; 1214 if (p + sizeof(s->v4) > buffer+buflen) { 1215 IP_VS_ERR_RL("BACKUP, Dropping buffer, to small\n"); 1216 return; 1217 } 1218 s = (union ip_vs_sync_conn *)p; 1219 size = ntohs(s->v4.ver_size) & SVER_MASK; 1220 msg_end = p + size; 1221 /* Basic sanity checks */ 1222 if (msg_end > buffer+buflen) { 1223 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n"); 1224 return; 1225 } 1226 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) { 1227 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n", 1228 ntohs(s->v4.ver_size) >> SVER_SHIFT); 1229 return; 1230 } 1231 /* Process a single sync_conn */ 1232 retc = ip_vs_proc_sync_conn(net, p, msg_end); 1233 if (retc < 0) { 1234 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n", 1235 retc); 1236 return; 1237 } 1238 /* Make sure we have 32 bit alignment */ 1239 msg_end = p + ((size + 3) & ~3); 1240 } 1241 } else { 1242 /* Old type of message */ 1243 ip_vs_process_message_v0(net, buffer, buflen); 1244 return; 1245 } 1246 } 1247 1248 1249 /* 1250 * Setup sndbuf (mode=1) or rcvbuf (mode=0) 1251 */ 1252 static void set_sock_size(struct sock *sk, int mode, int val) 1253 { 1254 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */ 1255 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */ 1256 lock_sock(sk); 1257 if (mode) { 1258 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2, 1259 sysctl_wmem_max); 1260 sk->sk_sndbuf = val * 2; 1261 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1262 } else { 1263 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2, 1264 sysctl_rmem_max); 1265 sk->sk_rcvbuf = val * 2; 1266 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1267 } 1268 release_sock(sk); 1269 } 1270 1271 /* 1272 * Setup loopback of outgoing multicasts on a sending socket 1273 */ 1274 static void set_mcast_loop(struct sock *sk, u_char loop) 1275 { 1276 struct inet_sock *inet = inet_sk(sk); 1277 1278 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ 1279 lock_sock(sk); 1280 inet->mc_loop = loop ? 1 : 0; 1281 release_sock(sk); 1282 } 1283 1284 /* 1285 * Specify TTL for outgoing multicasts on a sending socket 1286 */ 1287 static void set_mcast_ttl(struct sock *sk, u_char ttl) 1288 { 1289 struct inet_sock *inet = inet_sk(sk); 1290 1291 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ 1292 lock_sock(sk); 1293 inet->mc_ttl = ttl; 1294 release_sock(sk); 1295 } 1296 1297 /* 1298 * Specifiy default interface for outgoing multicasts 1299 */ 1300 static int set_mcast_if(struct sock *sk, char *ifname) 1301 { 1302 struct net_device *dev; 1303 struct inet_sock *inet = inet_sk(sk); 1304 struct net *net = sock_net(sk); 1305 1306 dev = __dev_get_by_name(net, ifname); 1307 if (!dev) 1308 return -ENODEV; 1309 1310 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) 1311 return -EINVAL; 1312 1313 lock_sock(sk); 1314 inet->mc_index = dev->ifindex; 1315 /* inet->mc_addr = 0; */ 1316 release_sock(sk); 1317 1318 return 0; 1319 } 1320 1321 1322 /* 1323 * Set the maximum length of sync message according to the 1324 * specified interface's MTU. 1325 */ 1326 static int set_sync_mesg_maxlen(struct net *net, int sync_state) 1327 { 1328 struct netns_ipvs *ipvs = net_ipvs(net); 1329 struct net_device *dev; 1330 int num; 1331 1332 if (sync_state == IP_VS_STATE_MASTER) { 1333 dev = __dev_get_by_name(net, ipvs->master_mcast_ifn); 1334 if (!dev) 1335 return -ENODEV; 1336 1337 num = (dev->mtu - sizeof(struct iphdr) - 1338 sizeof(struct udphdr) - 1339 SYNC_MESG_HEADER_LEN - 20) / SIMPLE_CONN_SIZE; 1340 ipvs->send_mesg_maxlen = SYNC_MESG_HEADER_LEN + 1341 SIMPLE_CONN_SIZE * min(num, MAX_CONNS_PER_SYNCBUFF); 1342 IP_VS_DBG(7, "setting the maximum length of sync sending " 1343 "message %d.\n", ipvs->send_mesg_maxlen); 1344 } else if (sync_state == IP_VS_STATE_BACKUP) { 1345 dev = __dev_get_by_name(net, ipvs->backup_mcast_ifn); 1346 if (!dev) 1347 return -ENODEV; 1348 1349 ipvs->recv_mesg_maxlen = dev->mtu - 1350 sizeof(struct iphdr) - sizeof(struct udphdr); 1351 IP_VS_DBG(7, "setting the maximum length of sync receiving " 1352 "message %d.\n", ipvs->recv_mesg_maxlen); 1353 } 1354 1355 return 0; 1356 } 1357 1358 1359 /* 1360 * Join a multicast group. 1361 * the group is specified by a class D multicast address 224.0.0.0/8 1362 * in the in_addr structure passed in as a parameter. 1363 */ 1364 static int 1365 join_mcast_group(struct sock *sk, struct in_addr *addr, char *ifname) 1366 { 1367 struct net *net = sock_net(sk); 1368 struct ip_mreqn mreq; 1369 struct net_device *dev; 1370 int ret; 1371 1372 memset(&mreq, 0, sizeof(mreq)); 1373 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); 1374 1375 dev = __dev_get_by_name(net, ifname); 1376 if (!dev) 1377 return -ENODEV; 1378 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) 1379 return -EINVAL; 1380 1381 mreq.imr_ifindex = dev->ifindex; 1382 1383 lock_sock(sk); 1384 ret = ip_mc_join_group(sk, &mreq); 1385 release_sock(sk); 1386 1387 return ret; 1388 } 1389 1390 1391 static int bind_mcastif_addr(struct socket *sock, char *ifname) 1392 { 1393 struct net *net = sock_net(sock->sk); 1394 struct net_device *dev; 1395 __be32 addr; 1396 struct sockaddr_in sin; 1397 1398 dev = __dev_get_by_name(net, ifname); 1399 if (!dev) 1400 return -ENODEV; 1401 1402 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); 1403 if (!addr) 1404 pr_err("You probably need to specify IP address on " 1405 "multicast interface.\n"); 1406 1407 IP_VS_DBG(7, "binding socket with (%s) %pI4\n", 1408 ifname, &addr); 1409 1410 /* Now bind the socket with the address of multicast interface */ 1411 sin.sin_family = AF_INET; 1412 sin.sin_addr.s_addr = addr; 1413 sin.sin_port = 0; 1414 1415 return sock->ops->bind(sock, (struct sockaddr*)&sin, sizeof(sin)); 1416 } 1417 1418 /* 1419 * Set up sending multicast socket over UDP 1420 */ 1421 static struct socket *make_send_sock(struct net *net, int id) 1422 { 1423 struct netns_ipvs *ipvs = net_ipvs(net); 1424 /* multicast addr */ 1425 struct sockaddr_in mcast_addr = { 1426 .sin_family = AF_INET, 1427 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id), 1428 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP), 1429 }; 1430 struct socket *sock; 1431 int result; 1432 1433 /* First create a socket move it to right name space later */ 1434 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); 1435 if (result < 0) { 1436 pr_err("Error during creation of socket; terminating\n"); 1437 return ERR_PTR(result); 1438 } 1439 /* 1440 * Kernel sockets that are a part of a namespace, should not 1441 * hold a reference to a namespace in order to allow to stop it. 1442 * After sk_change_net should be released using sk_release_kernel. 1443 */ 1444 sk_change_net(sock->sk, net); 1445 result = set_mcast_if(sock->sk, ipvs->master_mcast_ifn); 1446 if (result < 0) { 1447 pr_err("Error setting outbound mcast interface\n"); 1448 goto error; 1449 } 1450 1451 set_mcast_loop(sock->sk, 0); 1452 set_mcast_ttl(sock->sk, 1); 1453 result = sysctl_sync_sock_size(ipvs); 1454 if (result > 0) 1455 set_sock_size(sock->sk, 1, result); 1456 1457 result = bind_mcastif_addr(sock, ipvs->master_mcast_ifn); 1458 if (result < 0) { 1459 pr_err("Error binding address of the mcast interface\n"); 1460 goto error; 1461 } 1462 1463 result = sock->ops->connect(sock, (struct sockaddr *) &mcast_addr, 1464 sizeof(struct sockaddr), 0); 1465 if (result < 0) { 1466 pr_err("Error connecting to the multicast addr\n"); 1467 goto error; 1468 } 1469 1470 return sock; 1471 1472 error: 1473 sk_release_kernel(sock->sk); 1474 return ERR_PTR(result); 1475 } 1476 1477 1478 /* 1479 * Set up receiving multicast socket over UDP 1480 */ 1481 static struct socket *make_receive_sock(struct net *net, int id) 1482 { 1483 struct netns_ipvs *ipvs = net_ipvs(net); 1484 /* multicast addr */ 1485 struct sockaddr_in mcast_addr = { 1486 .sin_family = AF_INET, 1487 .sin_port = cpu_to_be16(IP_VS_SYNC_PORT + id), 1488 .sin_addr.s_addr = cpu_to_be32(IP_VS_SYNC_GROUP), 1489 }; 1490 struct socket *sock; 1491 int result; 1492 1493 /* First create a socket */ 1494 result = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock); 1495 if (result < 0) { 1496 pr_err("Error during creation of socket; terminating\n"); 1497 return ERR_PTR(result); 1498 } 1499 /* 1500 * Kernel sockets that are a part of a namespace, should not 1501 * hold a reference to a namespace in order to allow to stop it. 1502 * After sk_change_net should be released using sk_release_kernel. 1503 */ 1504 sk_change_net(sock->sk, net); 1505 /* it is equivalent to the REUSEADDR option in user-space */ 1506 sock->sk->sk_reuse = SK_CAN_REUSE; 1507 result = sysctl_sync_sock_size(ipvs); 1508 if (result > 0) 1509 set_sock_size(sock->sk, 0, result); 1510 1511 result = sock->ops->bind(sock, (struct sockaddr *) &mcast_addr, 1512 sizeof(struct sockaddr)); 1513 if (result < 0) { 1514 pr_err("Error binding to the multicast addr\n"); 1515 goto error; 1516 } 1517 1518 /* join the multicast group */ 1519 result = join_mcast_group(sock->sk, 1520 (struct in_addr *) &mcast_addr.sin_addr, 1521 ipvs->backup_mcast_ifn); 1522 if (result < 0) { 1523 pr_err("Error joining to the multicast group\n"); 1524 goto error; 1525 } 1526 1527 return sock; 1528 1529 error: 1530 sk_release_kernel(sock->sk); 1531 return ERR_PTR(result); 1532 } 1533 1534 1535 static int 1536 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) 1537 { 1538 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; 1539 struct kvec iov; 1540 int len; 1541 1542 EnterFunction(7); 1543 iov.iov_base = (void *)buffer; 1544 iov.iov_len = length; 1545 1546 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); 1547 1548 LeaveFunction(7); 1549 return len; 1550 } 1551 1552 static int 1553 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) 1554 { 1555 int msize; 1556 int ret; 1557 1558 msize = ntohs(msg->size); 1559 1560 ret = ip_vs_send_async(sock, (char *)msg, msize); 1561 if (ret >= 0 || ret == -EAGAIN) 1562 return ret; 1563 pr_err("ip_vs_send_async error %d\n", ret); 1564 return 0; 1565 } 1566 1567 static int 1568 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) 1569 { 1570 struct msghdr msg = {NULL,}; 1571 struct kvec iov; 1572 int len; 1573 1574 EnterFunction(7); 1575 1576 /* Receive a packet */ 1577 iov.iov_base = buffer; 1578 iov.iov_len = (size_t)buflen; 1579 1580 len = kernel_recvmsg(sock, &msg, &iov, 1, buflen, MSG_DONTWAIT); 1581 1582 if (len < 0) 1583 return len; 1584 1585 LeaveFunction(7); 1586 return len; 1587 } 1588 1589 /* Wakeup the master thread for sending */ 1590 static void master_wakeup_work_handler(struct work_struct *work) 1591 { 1592 struct ipvs_master_sync_state *ms = 1593 container_of(work, struct ipvs_master_sync_state, 1594 master_wakeup_work.work); 1595 struct netns_ipvs *ipvs = ms->ipvs; 1596 1597 spin_lock_bh(&ipvs->sync_lock); 1598 if (ms->sync_queue_len && 1599 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) { 1600 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE; 1601 wake_up_process(ms->master_thread); 1602 } 1603 spin_unlock_bh(&ipvs->sync_lock); 1604 } 1605 1606 /* Get next buffer to send */ 1607 static inline struct ip_vs_sync_buff * 1608 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms) 1609 { 1610 struct ip_vs_sync_buff *sb; 1611 1612 sb = sb_dequeue(ipvs, ms); 1613 if (sb) 1614 return sb; 1615 /* Do not delay entries in buffer for more than 2 seconds */ 1616 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME); 1617 } 1618 1619 static int sync_thread_master(void *data) 1620 { 1621 struct ip_vs_sync_thread_data *tinfo = data; 1622 struct netns_ipvs *ipvs = net_ipvs(tinfo->net); 1623 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id]; 1624 struct sock *sk = tinfo->sock->sk; 1625 struct ip_vs_sync_buff *sb; 1626 1627 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, " 1628 "syncid = %d, id = %d\n", 1629 ipvs->master_mcast_ifn, ipvs->master_syncid, tinfo->id); 1630 1631 for (;;) { 1632 sb = next_sync_buff(ipvs, ms); 1633 if (unlikely(kthread_should_stop())) 1634 break; 1635 if (!sb) { 1636 schedule_timeout(IPVS_SYNC_CHECK_PERIOD); 1637 continue; 1638 } 1639 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) { 1640 int ret = 0; 1641 1642 __wait_event_interruptible(*sk_sleep(sk), 1643 sock_writeable(sk) || 1644 kthread_should_stop(), 1645 ret); 1646 if (unlikely(kthread_should_stop())) 1647 goto done; 1648 } 1649 ip_vs_sync_buff_release(sb); 1650 } 1651 1652 done: 1653 __set_current_state(TASK_RUNNING); 1654 if (sb) 1655 ip_vs_sync_buff_release(sb); 1656 1657 /* clean up the sync_buff queue */ 1658 while ((sb = sb_dequeue(ipvs, ms))) 1659 ip_vs_sync_buff_release(sb); 1660 __set_current_state(TASK_RUNNING); 1661 1662 /* clean up the current sync_buff */ 1663 sb = get_curr_sync_buff(ipvs, ms, 0); 1664 if (sb) 1665 ip_vs_sync_buff_release(sb); 1666 1667 /* release the sending multicast socket */ 1668 sk_release_kernel(tinfo->sock->sk); 1669 kfree(tinfo); 1670 1671 return 0; 1672 } 1673 1674 1675 static int sync_thread_backup(void *data) 1676 { 1677 struct ip_vs_sync_thread_data *tinfo = data; 1678 struct netns_ipvs *ipvs = net_ipvs(tinfo->net); 1679 int len; 1680 1681 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, " 1682 "syncid = %d, id = %d\n", 1683 ipvs->backup_mcast_ifn, ipvs->backup_syncid, tinfo->id); 1684 1685 while (!kthread_should_stop()) { 1686 wait_event_interruptible(*sk_sleep(tinfo->sock->sk), 1687 !skb_queue_empty(&tinfo->sock->sk->sk_receive_queue) 1688 || kthread_should_stop()); 1689 1690 /* do we have data now? */ 1691 while (!skb_queue_empty(&(tinfo->sock->sk->sk_receive_queue))) { 1692 len = ip_vs_receive(tinfo->sock, tinfo->buf, 1693 ipvs->recv_mesg_maxlen); 1694 if (len <= 0) { 1695 if (len != -EAGAIN) 1696 pr_err("receiving message error\n"); 1697 break; 1698 } 1699 1700 ip_vs_process_message(tinfo->net, tinfo->buf, len); 1701 } 1702 } 1703 1704 /* release the sending multicast socket */ 1705 sk_release_kernel(tinfo->sock->sk); 1706 kfree(tinfo->buf); 1707 kfree(tinfo); 1708 1709 return 0; 1710 } 1711 1712 1713 int start_sync_thread(struct net *net, int state, char *mcast_ifn, __u8 syncid) 1714 { 1715 struct ip_vs_sync_thread_data *tinfo; 1716 struct task_struct **array = NULL, *task; 1717 struct socket *sock; 1718 struct netns_ipvs *ipvs = net_ipvs(net); 1719 char *name; 1720 int (*threadfn)(void *data); 1721 int id, count; 1722 int result = -ENOMEM; 1723 1724 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current)); 1725 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %Zd bytes\n", 1726 sizeof(struct ip_vs_sync_conn_v0)); 1727 1728 if (!ipvs->sync_state) { 1729 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX); 1730 ipvs->threads_mask = count - 1; 1731 } else 1732 count = ipvs->threads_mask + 1; 1733 1734 if (state == IP_VS_STATE_MASTER) { 1735 if (ipvs->ms) 1736 return -EEXIST; 1737 1738 strlcpy(ipvs->master_mcast_ifn, mcast_ifn, 1739 sizeof(ipvs->master_mcast_ifn)); 1740 ipvs->master_syncid = syncid; 1741 name = "ipvs-m:%d:%d"; 1742 threadfn = sync_thread_master; 1743 } else if (state == IP_VS_STATE_BACKUP) { 1744 if (ipvs->backup_threads) 1745 return -EEXIST; 1746 1747 strlcpy(ipvs->backup_mcast_ifn, mcast_ifn, 1748 sizeof(ipvs->backup_mcast_ifn)); 1749 ipvs->backup_syncid = syncid; 1750 name = "ipvs-b:%d:%d"; 1751 threadfn = sync_thread_backup; 1752 } else { 1753 return -EINVAL; 1754 } 1755 1756 if (state == IP_VS_STATE_MASTER) { 1757 struct ipvs_master_sync_state *ms; 1758 1759 ipvs->ms = kzalloc(count * sizeof(ipvs->ms[0]), GFP_KERNEL); 1760 if (!ipvs->ms) 1761 goto out; 1762 ms = ipvs->ms; 1763 for (id = 0; id < count; id++, ms++) { 1764 INIT_LIST_HEAD(&ms->sync_queue); 1765 ms->sync_queue_len = 0; 1766 ms->sync_queue_delay = 0; 1767 INIT_DELAYED_WORK(&ms->master_wakeup_work, 1768 master_wakeup_work_handler); 1769 ms->ipvs = ipvs; 1770 } 1771 } else { 1772 array = kzalloc(count * sizeof(struct task_struct *), 1773 GFP_KERNEL); 1774 if (!array) 1775 goto out; 1776 } 1777 set_sync_mesg_maxlen(net, state); 1778 1779 tinfo = NULL; 1780 for (id = 0; id < count; id++) { 1781 if (state == IP_VS_STATE_MASTER) 1782 sock = make_send_sock(net, id); 1783 else 1784 sock = make_receive_sock(net, id); 1785 if (IS_ERR(sock)) { 1786 result = PTR_ERR(sock); 1787 goto outtinfo; 1788 } 1789 tinfo = kmalloc(sizeof(*tinfo), GFP_KERNEL); 1790 if (!tinfo) 1791 goto outsocket; 1792 tinfo->net = net; 1793 tinfo->sock = sock; 1794 if (state == IP_VS_STATE_BACKUP) { 1795 tinfo->buf = kmalloc(ipvs->recv_mesg_maxlen, 1796 GFP_KERNEL); 1797 if (!tinfo->buf) 1798 goto outtinfo; 1799 } else { 1800 tinfo->buf = NULL; 1801 } 1802 tinfo->id = id; 1803 1804 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id); 1805 if (IS_ERR(task)) { 1806 result = PTR_ERR(task); 1807 goto outtinfo; 1808 } 1809 tinfo = NULL; 1810 if (state == IP_VS_STATE_MASTER) 1811 ipvs->ms[id].master_thread = task; 1812 else 1813 array[id] = task; 1814 } 1815 1816 /* mark as active */ 1817 1818 if (state == IP_VS_STATE_BACKUP) 1819 ipvs->backup_threads = array; 1820 spin_lock_bh(&ipvs->sync_buff_lock); 1821 ipvs->sync_state |= state; 1822 spin_unlock_bh(&ipvs->sync_buff_lock); 1823 1824 /* increase the module use count */ 1825 ip_vs_use_count_inc(); 1826 1827 return 0; 1828 1829 outsocket: 1830 sk_release_kernel(sock->sk); 1831 1832 outtinfo: 1833 if (tinfo) { 1834 sk_release_kernel(tinfo->sock->sk); 1835 kfree(tinfo->buf); 1836 kfree(tinfo); 1837 } 1838 count = id; 1839 while (count-- > 0) { 1840 if (state == IP_VS_STATE_MASTER) 1841 kthread_stop(ipvs->ms[count].master_thread); 1842 else 1843 kthread_stop(array[count]); 1844 } 1845 kfree(array); 1846 1847 out: 1848 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) { 1849 kfree(ipvs->ms); 1850 ipvs->ms = NULL; 1851 } 1852 return result; 1853 } 1854 1855 1856 int stop_sync_thread(struct net *net, int state) 1857 { 1858 struct netns_ipvs *ipvs = net_ipvs(net); 1859 struct task_struct **array; 1860 int id; 1861 int retc = -EINVAL; 1862 1863 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current)); 1864 1865 if (state == IP_VS_STATE_MASTER) { 1866 if (!ipvs->ms) 1867 return -ESRCH; 1868 1869 /* 1870 * The lock synchronizes with sb_queue_tail(), so that we don't 1871 * add sync buffers to the queue, when we are already in 1872 * progress of stopping the master sync daemon. 1873 */ 1874 1875 spin_lock_bh(&ipvs->sync_buff_lock); 1876 spin_lock(&ipvs->sync_lock); 1877 ipvs->sync_state &= ~IP_VS_STATE_MASTER; 1878 spin_unlock(&ipvs->sync_lock); 1879 spin_unlock_bh(&ipvs->sync_buff_lock); 1880 1881 retc = 0; 1882 for (id = ipvs->threads_mask; id >= 0; id--) { 1883 struct ipvs_master_sync_state *ms = &ipvs->ms[id]; 1884 int ret; 1885 1886 pr_info("stopping master sync thread %d ...\n", 1887 task_pid_nr(ms->master_thread)); 1888 cancel_delayed_work_sync(&ms->master_wakeup_work); 1889 ret = kthread_stop(ms->master_thread); 1890 if (retc >= 0) 1891 retc = ret; 1892 } 1893 kfree(ipvs->ms); 1894 ipvs->ms = NULL; 1895 } else if (state == IP_VS_STATE_BACKUP) { 1896 if (!ipvs->backup_threads) 1897 return -ESRCH; 1898 1899 ipvs->sync_state &= ~IP_VS_STATE_BACKUP; 1900 array = ipvs->backup_threads; 1901 retc = 0; 1902 for (id = ipvs->threads_mask; id >= 0; id--) { 1903 int ret; 1904 1905 pr_info("stopping backup sync thread %d ...\n", 1906 task_pid_nr(array[id])); 1907 ret = kthread_stop(array[id]); 1908 if (retc >= 0) 1909 retc = ret; 1910 } 1911 kfree(array); 1912 ipvs->backup_threads = NULL; 1913 } 1914 1915 /* decrease the module use count */ 1916 ip_vs_use_count_dec(); 1917 1918 return retc; 1919 } 1920 1921 /* 1922 * Initialize data struct for each netns 1923 */ 1924 int __net_init ip_vs_sync_net_init(struct net *net) 1925 { 1926 struct netns_ipvs *ipvs = net_ipvs(net); 1927 1928 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key); 1929 spin_lock_init(&ipvs->sync_lock); 1930 spin_lock_init(&ipvs->sync_buff_lock); 1931 return 0; 1932 } 1933 1934 void ip_vs_sync_net_cleanup(struct net *net) 1935 { 1936 int retc; 1937 struct netns_ipvs *ipvs = net_ipvs(net); 1938 1939 mutex_lock(&ipvs->sync_mutex); 1940 retc = stop_sync_thread(net, IP_VS_STATE_MASTER); 1941 if (retc && retc != -ESRCH) 1942 pr_err("Failed to stop Master Daemon\n"); 1943 1944 retc = stop_sync_thread(net, IP_VS_STATE_BACKUP); 1945 if (retc && retc != -ESRCH) 1946 pr_err("Failed to stop Backup Daemon\n"); 1947 mutex_unlock(&ipvs->sync_mutex); 1948 } 1949