1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * IPVS An implementation of the IP virtual server support for the 4 * LINUX operating system. IPVS is now implemented as a module 5 * over the NetFilter framework. IPVS can be used to build a 6 * high-performance and highly available server based on a 7 * cluster of servers. 8 * 9 * Version 1, is capable of handling both version 0 and 1 messages. 10 * Version 0 is the plain old format. 11 * Note Version 0 receivers will just drop Ver 1 messages. 12 * Version 1 is capable of handle IPv6, Persistence data, 13 * time-outs, and firewall marks. 14 * In ver.1 "ip_vs_sync_conn_options" will be sent in netw. order. 15 * Ver. 0 can be turned on by sysctl -w net.ipv4.vs.sync_version=0 16 * 17 * Definitions Message: is a complete datagram 18 * Sync_conn: is a part of a Message 19 * Param Data is an option to a Sync_conn. 20 * 21 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org> 22 * 23 * ip_vs_sync: sync connection info from master load balancer to backups 24 * through multicast 25 * 26 * Changes: 27 * Alexandre Cassen : Added master & backup support at a time. 28 * Alexandre Cassen : Added SyncID support for incoming sync 29 * messages filtering. 30 * Justin Ossevoort : Fix endian problem on sync message size. 31 * Hans Schillstrom : Added Version 1: i.e. IPv6, 32 * Persistence support, fwmark and time-out. 33 */ 34 35 #define pr_fmt(fmt) "IPVS: " 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 #include <linux/sched/signal.h> 52 53 #include <linux/unaligned.h> /* Used for ntoh_seq and hton_seq */ 54 55 #include <net/ip.h> 56 #include <net/sock.h> 57 58 #include <net/ip_vs.h> 59 60 #define IP_VS_SYNC_GROUP 0xe0000051 /* multicast addr - 224.0.0.81 */ 61 #define IP_VS_SYNC_PORT 8848 /* multicast port */ 62 63 #define SYNC_PROTO_VER 1 /* Protocol version in header */ 64 65 static struct lock_class_key __ipvs_sync_key; 66 /* 67 * IPVS sync connection entry 68 * Version 0, i.e. original version. 69 */ 70 struct ip_vs_sync_conn_v0 { 71 __u8 reserved; 72 73 /* Protocol, addresses and port numbers */ 74 __u8 protocol; /* Which protocol (TCP/UDP) */ 75 __be16 cport; 76 __be16 vport; 77 __be16 dport; 78 __be32 caddr; /* client address */ 79 __be32 vaddr; /* virtual address */ 80 __be32 daddr; /* destination address */ 81 82 /* Flags and state transition */ 83 __be16 flags; /* status flags */ 84 __be16 state; /* state info */ 85 86 /* The sequence options start here */ 87 }; 88 89 struct ip_vs_sync_conn_options { 90 struct ip_vs_seq in_seq; /* incoming seq. struct */ 91 struct ip_vs_seq out_seq; /* outgoing seq. struct */ 92 }; 93 94 /* 95 Sync Connection format (sync_conn) 96 97 0 1 2 3 98 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 99 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 100 | Type | Protocol | Ver. | Size | 101 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 102 | Flags | 103 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 104 | State | cport | 105 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 106 | vport | dport | 107 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 108 | fwmark | 109 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 110 | timeout (in sec.) | 111 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 112 | ... | 113 | IP-Addresses (v4 or v6) | 114 | ... | 115 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 116 Optional Parameters. 117 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 118 | Param. Type | Param. Length | Param. data | 119 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | 120 | ... | 121 | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 122 | | Param Type | Param. Length | 123 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 124 | Param data | 125 | Last Param data should be padded for 32 bit alignment | 126 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 127 */ 128 129 /* 130 * Type 0, IPv4 sync connection format 131 */ 132 struct ip_vs_sync_v4 { 133 __u8 type; 134 __u8 protocol; /* Which protocol (TCP/UDP) */ 135 __be16 ver_size; /* Version msb 4 bits */ 136 /* Flags and state transition */ 137 __be32 flags; /* status flags */ 138 __be16 state; /* state info */ 139 /* Protocol, addresses and port numbers */ 140 __be16 cport; 141 __be16 vport; 142 __be16 dport; 143 __be32 fwmark; /* Firewall mark from skb */ 144 __be32 timeout; /* cp timeout */ 145 __be32 caddr; /* client address */ 146 __be32 vaddr; /* virtual address */ 147 __be32 daddr; /* destination address */ 148 /* The sequence options start here */ 149 /* PE data padded to 32bit alignment after seq. options */ 150 }; 151 /* 152 * Type 2 messages IPv6 153 */ 154 struct ip_vs_sync_v6 { 155 __u8 type; 156 __u8 protocol; /* Which protocol (TCP/UDP) */ 157 __be16 ver_size; /* Version msb 4 bits */ 158 /* Flags and state transition */ 159 __be32 flags; /* status flags */ 160 __be16 state; /* state info */ 161 /* Protocol, addresses and port numbers */ 162 __be16 cport; 163 __be16 vport; 164 __be16 dport; 165 __be32 fwmark; /* Firewall mark from skb */ 166 __be32 timeout; /* cp timeout */ 167 struct in6_addr caddr; /* client address */ 168 struct in6_addr vaddr; /* virtual address */ 169 struct in6_addr daddr; /* destination address */ 170 /* The sequence options start here */ 171 /* PE data padded to 32bit alignment after seq. options */ 172 }; 173 174 union ip_vs_sync_conn { 175 struct ip_vs_sync_v4 v4; 176 struct ip_vs_sync_v6 v6; 177 }; 178 179 /* Bits in Type field in above */ 180 #define STYPE_INET6 0 181 #define STYPE_F_INET6 (1 << STYPE_INET6) 182 183 #define SVER_SHIFT 12 /* Shift to get version */ 184 #define SVER_MASK 0x0fff /* Mask to strip version */ 185 186 #define IPVS_OPT_SEQ_DATA 1 187 #define IPVS_OPT_PE_DATA 2 188 #define IPVS_OPT_PE_NAME 3 189 #define IPVS_OPT_PARAM 7 190 191 #define IPVS_OPT_F_SEQ_DATA (1 << (IPVS_OPT_SEQ_DATA-1)) 192 #define IPVS_OPT_F_PE_DATA (1 << (IPVS_OPT_PE_DATA-1)) 193 #define IPVS_OPT_F_PE_NAME (1 << (IPVS_OPT_PE_NAME-1)) 194 #define IPVS_OPT_F_PARAM (1 << (IPVS_OPT_PARAM-1)) 195 196 struct ip_vs_sync_thread_data { 197 struct task_struct *task; 198 struct netns_ipvs *ipvs; 199 struct socket *sock; 200 char *buf; 201 int id; 202 }; 203 204 /* Version 0 definition of packet sizes */ 205 #define SIMPLE_CONN_SIZE (sizeof(struct ip_vs_sync_conn_v0)) 206 #define FULL_CONN_SIZE \ 207 (sizeof(struct ip_vs_sync_conn_v0) + sizeof(struct ip_vs_sync_conn_options)) 208 209 210 /* 211 The master mulitcasts messages (Datagrams) to the backup load balancers 212 in the following format. 213 214 Version 1: 215 Note, first byte should be Zero, so ver 0 receivers will drop the packet. 216 217 0 1 2 3 218 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 1 219 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 220 | 0 | SyncID | Size | 221 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 222 | Count Conns | Version | Reserved, set to Zero | 223 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 224 | | 225 | IPVS Sync Connection (1) | 226 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 227 | . | 228 ~ . ~ 229 | . | 230 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 231 | | 232 | IPVS Sync Connection (n) | 233 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 234 235 Version 0 Header 236 0 1 2 3 237 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 1 238 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 239 | Count Conns | SyncID | Size | 240 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ 241 | IPVS Sync Connection (1) | 242 */ 243 244 /* Version 0 header */ 245 struct ip_vs_sync_mesg_v0 { 246 __u8 nr_conns; 247 __u8 syncid; 248 __be16 size; 249 250 /* ip_vs_sync_conn entries start here */ 251 }; 252 253 /* Version 1 header */ 254 struct ip_vs_sync_mesg { 255 __u8 reserved; /* must be zero */ 256 __u8 syncid; 257 __be16 size; 258 __u8 nr_conns; 259 __s8 version; /* SYNC_PROTO_VER */ 260 __u16 spare; 261 /* ip_vs_sync_conn entries start here */ 262 }; 263 264 union ipvs_sockaddr { 265 struct sockaddr_in in; 266 struct sockaddr_in6 in6; 267 }; 268 269 struct ip_vs_sync_buff { 270 struct list_head list; 271 unsigned long firstuse; 272 273 /* pointers for the message data */ 274 struct ip_vs_sync_mesg *mesg; 275 unsigned char *head; 276 unsigned char *end; 277 }; 278 279 /* 280 * Copy of struct ip_vs_seq 281 * From unaligned network order to aligned host order 282 */ 283 static void ntoh_seq(struct ip_vs_seq *no, struct ip_vs_seq *ho) 284 { 285 memset(ho, 0, sizeof(*ho)); 286 ho->init_seq = get_unaligned_be32(&no->init_seq); 287 ho->delta = get_unaligned_be32(&no->delta); 288 ho->previous_delta = get_unaligned_be32(&no->previous_delta); 289 } 290 291 /* 292 * Copy of struct ip_vs_seq 293 * From Aligned host order to unaligned network order 294 */ 295 static void hton_seq(struct ip_vs_seq *ho, struct ip_vs_seq *no) 296 { 297 put_unaligned_be32(ho->init_seq, &no->init_seq); 298 put_unaligned_be32(ho->delta, &no->delta); 299 put_unaligned_be32(ho->previous_delta, &no->previous_delta); 300 } 301 302 static inline struct ip_vs_sync_buff * 303 sb_dequeue(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms) 304 { 305 struct ip_vs_sync_buff *sb; 306 307 spin_lock_bh(&ipvs->sync_lock); 308 if (list_empty(&ms->sync_queue)) { 309 sb = NULL; 310 __set_current_state(TASK_INTERRUPTIBLE); 311 } else { 312 sb = list_entry(ms->sync_queue.next, struct ip_vs_sync_buff, 313 list); 314 list_del(&sb->list); 315 ms->sync_queue_len--; 316 if (!ms->sync_queue_len) 317 ms->sync_queue_delay = 0; 318 } 319 spin_unlock_bh(&ipvs->sync_lock); 320 321 return sb; 322 } 323 324 /* 325 * Create a new sync buffer for Version 1 proto. 326 */ 327 static inline struct ip_vs_sync_buff * 328 ip_vs_sync_buff_create(struct netns_ipvs *ipvs, unsigned int len) 329 { 330 struct ip_vs_sync_buff *sb; 331 332 if (!(sb=kmalloc_obj(struct ip_vs_sync_buff, GFP_ATOMIC))) 333 return NULL; 334 335 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg), 336 ipvs->mcfg.sync_maxlen); 337 sb->mesg = kmalloc(len, GFP_ATOMIC); 338 if (!sb->mesg) { 339 kfree(sb); 340 return NULL; 341 } 342 sb->mesg->reserved = 0; /* old nr_conns i.e. must be zero now */ 343 sb->mesg->version = SYNC_PROTO_VER; 344 sb->mesg->syncid = ipvs->mcfg.syncid; 345 sb->mesg->size = htons(sizeof(struct ip_vs_sync_mesg)); 346 sb->mesg->nr_conns = 0; 347 sb->mesg->spare = 0; 348 sb->head = (unsigned char *)sb->mesg + sizeof(struct ip_vs_sync_mesg); 349 sb->end = (unsigned char *)sb->mesg + len; 350 351 sb->firstuse = jiffies; 352 return sb; 353 } 354 355 static inline void ip_vs_sync_buff_release(struct ip_vs_sync_buff *sb) 356 { 357 kfree(sb->mesg); 358 kfree(sb); 359 } 360 361 static inline void sb_queue_tail(struct netns_ipvs *ipvs, 362 struct ipvs_master_sync_state *ms) 363 { 364 struct ip_vs_sync_buff *sb = ms->sync_buff; 365 366 spin_lock(&ipvs->sync_lock); 367 if (ipvs->sync_state & IP_VS_STATE_MASTER && 368 ms->sync_queue_len < sysctl_sync_qlen_max(ipvs)) { 369 if (!ms->sync_queue_len) 370 schedule_delayed_work(&ms->master_wakeup_work, 371 max(IPVS_SYNC_SEND_DELAY, 1)); 372 ms->sync_queue_len++; 373 list_add_tail(&sb->list, &ms->sync_queue); 374 if ((++ms->sync_queue_delay) == IPVS_SYNC_WAKEUP_RATE) { 375 int id = (int)(ms - ipvs->ms); 376 377 wake_up_process(ipvs->master_tinfo[id].task); 378 } 379 } else 380 ip_vs_sync_buff_release(sb); 381 spin_unlock(&ipvs->sync_lock); 382 } 383 384 /* 385 * Get the current sync buffer if it has been created for more 386 * than the specified time or the specified time is zero. 387 */ 388 static inline struct ip_vs_sync_buff * 389 get_curr_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms, 390 unsigned long time) 391 { 392 struct ip_vs_sync_buff *sb; 393 394 spin_lock_bh(&ipvs->sync_buff_lock); 395 sb = ms->sync_buff; 396 if (sb && time_after_eq(jiffies - sb->firstuse, time)) { 397 ms->sync_buff = NULL; 398 __set_current_state(TASK_RUNNING); 399 } else 400 sb = NULL; 401 spin_unlock_bh(&ipvs->sync_buff_lock); 402 return sb; 403 } 404 405 static inline int 406 select_master_thread_id(struct netns_ipvs *ipvs, struct ip_vs_conn *cp) 407 { 408 return ((long) cp >> (1 + ilog2(sizeof(*cp)))) & ipvs->threads_mask; 409 } 410 411 /* 412 * Create a new sync buffer for Version 0 proto. 413 */ 414 static inline struct ip_vs_sync_buff * 415 ip_vs_sync_buff_create_v0(struct netns_ipvs *ipvs, unsigned int len) 416 { 417 struct ip_vs_sync_buff *sb; 418 struct ip_vs_sync_mesg_v0 *mesg; 419 420 if (!(sb=kmalloc_obj(struct ip_vs_sync_buff, GFP_ATOMIC))) 421 return NULL; 422 423 len = max_t(unsigned int, len + sizeof(struct ip_vs_sync_mesg_v0), 424 ipvs->mcfg.sync_maxlen); 425 sb->mesg = kmalloc(len, GFP_ATOMIC); 426 if (!sb->mesg) { 427 kfree(sb); 428 return NULL; 429 } 430 mesg = (struct ip_vs_sync_mesg_v0 *)sb->mesg; 431 mesg->nr_conns = 0; 432 mesg->syncid = ipvs->mcfg.syncid; 433 mesg->size = htons(sizeof(struct ip_vs_sync_mesg_v0)); 434 sb->head = (unsigned char *)mesg + sizeof(struct ip_vs_sync_mesg_v0); 435 sb->end = (unsigned char *)mesg + len; 436 sb->firstuse = jiffies; 437 return sb; 438 } 439 440 /* Check if connection is controlled by persistence */ 441 static inline bool in_persistence(struct ip_vs_conn *cp) 442 { 443 for (cp = cp->control; cp; cp = cp->control) { 444 if (cp->flags & IP_VS_CONN_F_TEMPLATE) 445 return true; 446 } 447 return false; 448 } 449 450 /* Check if conn should be synced. 451 * pkts: conn packets, use sysctl_sync_threshold to avoid packet check 452 * - (1) sync_refresh_period: reduce sync rate. Additionally, retry 453 * sync_retries times with period of sync_refresh_period/8 454 * - (2) if both sync_refresh_period and sync_period are 0 send sync only 455 * for state changes or only once when pkts matches sync_threshold 456 * - (3) templates: rate can be reduced only with sync_refresh_period or 457 * with (2) 458 */ 459 static int ip_vs_sync_conn_needed(struct netns_ipvs *ipvs, 460 struct ip_vs_conn *cp, int pkts) 461 { 462 unsigned long orig = READ_ONCE(cp->sync_endtime); 463 unsigned long now = jiffies; 464 unsigned long n = (now + cp->timeout) & ~3UL; 465 unsigned int sync_refresh_period; 466 int sync_period; 467 int force; 468 469 /* Check if we sync in current state */ 470 if (unlikely(cp->flags & IP_VS_CONN_F_TEMPLATE)) 471 force = 0; 472 else if (unlikely(sysctl_sync_persist_mode(ipvs) && in_persistence(cp))) 473 return 0; 474 else if (likely(cp->protocol == IPPROTO_TCP)) { 475 if (!((1 << cp->state) & 476 ((1 << IP_VS_TCP_S_ESTABLISHED) | 477 (1 << IP_VS_TCP_S_FIN_WAIT) | 478 (1 << IP_VS_TCP_S_CLOSE) | 479 (1 << IP_VS_TCP_S_CLOSE_WAIT) | 480 (1 << IP_VS_TCP_S_TIME_WAIT)))) 481 return 0; 482 force = cp->state != cp->old_state; 483 if (force && cp->state != IP_VS_TCP_S_ESTABLISHED) 484 goto set; 485 } else if (unlikely(cp->protocol == IPPROTO_SCTP)) { 486 if (!((1 << cp->state) & 487 ((1 << IP_VS_SCTP_S_ESTABLISHED) | 488 (1 << IP_VS_SCTP_S_SHUTDOWN_SENT) | 489 (1 << IP_VS_SCTP_S_SHUTDOWN_RECEIVED) | 490 (1 << IP_VS_SCTP_S_SHUTDOWN_ACK_SENT) | 491 (1 << IP_VS_SCTP_S_CLOSED)))) 492 return 0; 493 force = cp->state != cp->old_state; 494 if (force && cp->state != IP_VS_SCTP_S_ESTABLISHED) 495 goto set; 496 } else { 497 /* UDP or another protocol with single state */ 498 force = 0; 499 } 500 501 sync_refresh_period = sysctl_sync_refresh_period(ipvs); 502 if (sync_refresh_period > 0) { 503 long diff = n - orig; 504 long min_diff = max(cp->timeout >> 1, 10UL * HZ); 505 506 /* Avoid sync if difference is below sync_refresh_period 507 * and below the half timeout. 508 */ 509 if (abs(diff) < min_t(long, sync_refresh_period, min_diff)) { 510 int retries = orig & 3; 511 512 if (retries >= sysctl_sync_retries(ipvs)) 513 return 0; 514 if (time_before(now, orig - cp->timeout + 515 (sync_refresh_period >> 3))) 516 return 0; 517 n |= retries + 1; 518 } 519 } 520 sync_period = sysctl_sync_period(ipvs); 521 if (sync_period > 0) { 522 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE) && 523 pkts % sync_period != sysctl_sync_threshold(ipvs)) 524 return 0; 525 } else if (!sync_refresh_period && 526 pkts != sysctl_sync_threshold(ipvs)) 527 return 0; 528 529 set: 530 cp->old_state = cp->state; 531 n = cmpxchg(&cp->sync_endtime, orig, n); 532 return n == orig || force; 533 } 534 535 /* 536 * Version 0 , could be switched in by sys_ctl. 537 * Add an ip_vs_conn information into the current sync_buff. 538 */ 539 static void ip_vs_sync_conn_v0(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, 540 int pkts) 541 { 542 struct ip_vs_sync_mesg_v0 *m; 543 struct ip_vs_sync_conn_v0 *s; 544 struct ip_vs_sync_buff *buff; 545 struct ipvs_master_sync_state *ms; 546 int id; 547 unsigned int len; 548 549 if (unlikely(cp->af != AF_INET)) 550 return; 551 /* Do not sync ONE PACKET */ 552 if (cp->flags & IP_VS_CONN_F_ONE_PACKET) 553 return; 554 555 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts)) 556 return; 557 558 spin_lock_bh(&ipvs->sync_buff_lock); 559 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) { 560 spin_unlock_bh(&ipvs->sync_buff_lock); 561 return; 562 } 563 564 id = select_master_thread_id(ipvs, cp); 565 ms = &ipvs->ms[id]; 566 buff = ms->sync_buff; 567 len = (cp->flags & IP_VS_CONN_F_SEQ_MASK) ? FULL_CONN_SIZE : 568 SIMPLE_CONN_SIZE; 569 if (buff) { 570 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg; 571 /* Send buffer if it is for v1 */ 572 if (buff->head + len > buff->end || !m->nr_conns) { 573 sb_queue_tail(ipvs, ms); 574 ms->sync_buff = NULL; 575 buff = NULL; 576 } 577 } 578 if (!buff) { 579 buff = ip_vs_sync_buff_create_v0(ipvs, len); 580 if (!buff) { 581 spin_unlock_bh(&ipvs->sync_buff_lock); 582 pr_err("ip_vs_sync_buff_create failed.\n"); 583 return; 584 } 585 ms->sync_buff = buff; 586 } 587 588 m = (struct ip_vs_sync_mesg_v0 *) buff->mesg; 589 s = (struct ip_vs_sync_conn_v0 *) buff->head; 590 591 /* copy members */ 592 s->reserved = 0; 593 s->protocol = cp->protocol; 594 s->cport = cp->cport; 595 s->vport = cp->vport; 596 s->dport = cp->dport; 597 s->caddr = cp->caddr.ip; 598 s->vaddr = cp->vaddr.ip; 599 s->daddr = cp->daddr.ip; 600 s->flags = htons(cp->flags & ~IP_VS_CONN_F_HASHED); 601 s->state = htons(cp->state); 602 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { 603 struct ip_vs_sync_conn_options *opt = 604 (struct ip_vs_sync_conn_options *)&s[1]; 605 memcpy(opt, &cp->sync_conn_opt, sizeof(*opt)); 606 } 607 608 m->nr_conns++; 609 m->size = htons(ntohs(m->size) + len); 610 buff->head += len; 611 spin_unlock_bh(&ipvs->sync_buff_lock); 612 613 /* synchronize its controller if it has */ 614 cp = cp->control; 615 if (cp) { 616 if (cp->flags & IP_VS_CONN_F_TEMPLATE) 617 pkts = atomic_inc_return(&cp->in_pkts); 618 else 619 pkts = sysctl_sync_threshold(ipvs); 620 ip_vs_sync_conn(ipvs, cp, pkts); 621 } 622 } 623 624 /* 625 * Add an ip_vs_conn information into the current sync_buff. 626 * Called by ip_vs_in. 627 * Sending Version 1 messages 628 */ 629 void ip_vs_sync_conn(struct netns_ipvs *ipvs, struct ip_vs_conn *cp, int pkts) 630 { 631 struct ip_vs_sync_mesg *m; 632 union ip_vs_sync_conn *s; 633 struct ip_vs_sync_buff *buff; 634 struct ipvs_master_sync_state *ms; 635 int id; 636 __u8 *p; 637 unsigned int len, pe_name_len, pad; 638 639 /* Handle old version of the protocol */ 640 if (sysctl_sync_ver(ipvs) == 0) { 641 ip_vs_sync_conn_v0(ipvs, cp, pkts); 642 return; 643 } 644 /* Do not sync ONE PACKET */ 645 if (cp->flags & IP_VS_CONN_F_ONE_PACKET) 646 goto control; 647 sloop: 648 if (!ip_vs_sync_conn_needed(ipvs, cp, pkts)) 649 goto control; 650 651 /* Sanity checks */ 652 pe_name_len = 0; 653 if (cp->pe_data_len) { 654 if (!cp->pe_data || !cp->dest) { 655 IP_VS_ERR_RL("SYNC, connection pe_data invalid\n"); 656 return; 657 } 658 pe_name_len = strnlen(cp->pe->name, IP_VS_PENAME_MAXLEN); 659 } 660 661 spin_lock_bh(&ipvs->sync_buff_lock); 662 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) { 663 spin_unlock_bh(&ipvs->sync_buff_lock); 664 return; 665 } 666 667 id = select_master_thread_id(ipvs, cp); 668 ms = &ipvs->ms[id]; 669 670 #ifdef CONFIG_IP_VS_IPV6 671 if (cp->af == AF_INET6) 672 len = sizeof(struct ip_vs_sync_v6); 673 else 674 #endif 675 len = sizeof(struct ip_vs_sync_v4); 676 677 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) 678 len += sizeof(struct ip_vs_sync_conn_options) + 2; 679 680 if (cp->pe_data_len) 681 len += cp->pe_data_len + 2; /* + Param hdr field */ 682 if (pe_name_len) 683 len += pe_name_len + 2; 684 685 /* check if there is a space for this one */ 686 pad = 0; 687 buff = ms->sync_buff; 688 if (buff) { 689 m = buff->mesg; 690 pad = (4 - (size_t) buff->head) & 3; 691 /* Send buffer if it is for v0 */ 692 if (buff->head + len + pad > buff->end || m->reserved) { 693 sb_queue_tail(ipvs, ms); 694 ms->sync_buff = NULL; 695 buff = NULL; 696 pad = 0; 697 } 698 } 699 700 if (!buff) { 701 buff = ip_vs_sync_buff_create(ipvs, len); 702 if (!buff) { 703 spin_unlock_bh(&ipvs->sync_buff_lock); 704 pr_err("ip_vs_sync_buff_create failed.\n"); 705 return; 706 } 707 ms->sync_buff = buff; 708 m = buff->mesg; 709 } 710 711 p = buff->head; 712 buff->head += pad + len; 713 m->size = htons(ntohs(m->size) + pad + len); 714 /* Add ev. padding from prev. sync_conn */ 715 while (pad--) 716 *(p++) = 0; 717 718 s = (union ip_vs_sync_conn *)p; 719 720 /* Set message type & copy members */ 721 s->v4.type = (cp->af == AF_INET6 ? STYPE_F_INET6 : 0); 722 s->v4.ver_size = htons(len & SVER_MASK); /* Version 0 */ 723 s->v4.flags = htonl(cp->flags & ~IP_VS_CONN_F_HASHED); 724 s->v4.state = htons(cp->state); 725 s->v4.protocol = cp->protocol; 726 s->v4.cport = cp->cport; 727 s->v4.vport = cp->vport; 728 s->v4.dport = cp->dport; 729 s->v4.fwmark = htonl(cp->fwmark); 730 s->v4.timeout = htonl(cp->timeout / HZ); 731 m->nr_conns++; 732 733 #ifdef CONFIG_IP_VS_IPV6 734 if (cp->af == AF_INET6) { 735 p += sizeof(struct ip_vs_sync_v6); 736 s->v6.caddr = cp->caddr.in6; 737 s->v6.vaddr = cp->vaddr.in6; 738 s->v6.daddr = cp->daddr.in6; 739 } else 740 #endif 741 { 742 p += sizeof(struct ip_vs_sync_v4); /* options ptr */ 743 s->v4.caddr = cp->caddr.ip; 744 s->v4.vaddr = cp->vaddr.ip; 745 s->v4.daddr = cp->daddr.ip; 746 } 747 if (cp->flags & IP_VS_CONN_F_SEQ_MASK) { 748 *(p++) = IPVS_OPT_SEQ_DATA; 749 *(p++) = sizeof(struct ip_vs_sync_conn_options); 750 hton_seq(&cp->in_seq, (struct ip_vs_seq *)p); 751 p += sizeof(struct ip_vs_seq); 752 hton_seq(&cp->out_seq, (struct ip_vs_seq *)p); 753 p += sizeof(struct ip_vs_seq); 754 } 755 /* Handle pe data */ 756 if (cp->pe_data_len && cp->pe_data) { 757 *(p++) = IPVS_OPT_PE_DATA; 758 *(p++) = cp->pe_data_len; 759 memcpy(p, cp->pe_data, cp->pe_data_len); 760 p += cp->pe_data_len; 761 if (pe_name_len) { 762 /* Add PE_NAME */ 763 *(p++) = IPVS_OPT_PE_NAME; 764 *(p++) = pe_name_len; 765 memcpy(p, cp->pe->name, pe_name_len); 766 p += pe_name_len; 767 } 768 } 769 770 spin_unlock_bh(&ipvs->sync_buff_lock); 771 772 control: 773 /* synchronize its controller if it has */ 774 cp = cp->control; 775 if (!cp) 776 return; 777 if (cp->flags & IP_VS_CONN_F_TEMPLATE) 778 pkts = atomic_inc_return(&cp->in_pkts); 779 else 780 pkts = sysctl_sync_threshold(ipvs); 781 goto sloop; 782 } 783 784 /* 785 * fill_param used by version 1 786 */ 787 static inline int 788 ip_vs_conn_fill_param_sync(struct netns_ipvs *ipvs, int af, union ip_vs_sync_conn *sc, 789 struct ip_vs_conn_param *p, 790 __u8 *pe_data, unsigned int pe_data_len, 791 __u8 *pe_name, unsigned int pe_name_len) 792 { 793 #ifdef CONFIG_IP_VS_IPV6 794 if (af == AF_INET6) 795 ip_vs_conn_fill_param(ipvs, af, sc->v6.protocol, 796 (const union nf_inet_addr *)&sc->v6.caddr, 797 sc->v6.cport, 798 (const union nf_inet_addr *)&sc->v6.vaddr, 799 sc->v6.vport, p); 800 else 801 #endif 802 ip_vs_conn_fill_param(ipvs, af, sc->v4.protocol, 803 (const union nf_inet_addr *)&sc->v4.caddr, 804 sc->v4.cport, 805 (const union nf_inet_addr *)&sc->v4.vaddr, 806 sc->v4.vport, p); 807 /* Handle pe data */ 808 if (pe_data_len) { 809 if (pe_name_len) { 810 char buff[IP_VS_PENAME_MAXLEN+1]; 811 812 memcpy(buff, pe_name, pe_name_len); 813 buff[pe_name_len]=0; 814 p->pe = __ip_vs_pe_getbyname(buff); 815 if (!p->pe) { 816 IP_VS_DBG(3, "BACKUP, no %s engine found/loaded\n", 817 buff); 818 return 1; 819 } 820 } else { 821 IP_VS_ERR_RL("BACKUP, Invalid PE parameters\n"); 822 return 1; 823 } 824 825 p->pe_data = kmemdup(pe_data, pe_data_len, GFP_ATOMIC); 826 if (!p->pe_data) { 827 module_put(p->pe->module); 828 return -ENOMEM; 829 } 830 p->pe_data_len = pe_data_len; 831 } 832 return 0; 833 } 834 835 /* 836 * Connection Add / Update. 837 * Common for version 0 and 1 reception of backup sync_conns. 838 * Param: ... 839 * timeout is in sec. 840 */ 841 static void ip_vs_proc_conn(struct netns_ipvs *ipvs, struct ip_vs_conn_param *param, 842 unsigned int flags, unsigned int state, 843 unsigned int protocol, unsigned int type, 844 const union nf_inet_addr *daddr, __be16 dport, 845 unsigned long timeout, __u32 fwmark, 846 struct ip_vs_sync_conn_options *opt) 847 { 848 struct ip_vs_dest *dest; 849 struct ip_vs_conn *cp; 850 851 if (!(flags & IP_VS_CONN_F_TEMPLATE)) { 852 cp = ip_vs_conn_in_get(param); 853 if (cp && ((cp->dport != dport) || 854 !ip_vs_addr_equal(cp->daf, &cp->daddr, daddr))) { 855 if (!(flags & IP_VS_CONN_F_INACTIVE)) { 856 ip_vs_conn_expire_now(cp); 857 __ip_vs_conn_put(cp); 858 cp = NULL; 859 } else { 860 /* This is the expiration message for the 861 * connection that was already replaced, so we 862 * just ignore it. 863 */ 864 __ip_vs_conn_put(cp); 865 kfree(param->pe_data); 866 return; 867 } 868 } 869 } else { 870 cp = ip_vs_ct_in_get(param); 871 } 872 873 if (cp) { 874 /* Free pe_data */ 875 kfree(param->pe_data); 876 877 dest = cp->dest; 878 spin_lock_bh(&cp->lock); 879 if ((cp->flags ^ flags) & IP_VS_CONN_F_INACTIVE && 880 !(flags & IP_VS_CONN_F_TEMPLATE) && dest) { 881 if (flags & IP_VS_CONN_F_INACTIVE) 882 atomic_dec(&dest->activeconns); 883 else 884 atomic_inc(&dest->activeconns); 885 } 886 flags &= IP_VS_CONN_F_BACKUP_UPD_MASK; 887 flags |= cp->flags & ~IP_VS_CONN_F_BACKUP_UPD_MASK; 888 cp->flags = flags; 889 spin_unlock_bh(&cp->lock); 890 if (!dest) 891 ip_vs_try_bind_dest(cp); 892 } else { 893 /* 894 * Find the appropriate destination for the connection. 895 * If it is not found the connection will remain unbound 896 * but still handled. 897 */ 898 rcu_read_lock(); 899 /* This function is only invoked by the synchronization 900 * code. We do not currently support heterogeneous pools 901 * with synchronization, so we can make the assumption that 902 * the svc_af is the same as the dest_af 903 */ 904 dest = ip_vs_find_dest(ipvs, type, type, daddr, dport, 905 param->vaddr, param->vport, protocol, 906 fwmark, flags); 907 908 cp = ip_vs_conn_new(param, type, daddr, dport, flags, dest, 909 fwmark); 910 rcu_read_unlock(); 911 if (!cp) { 912 kfree(param->pe_data); 913 IP_VS_DBG(2, "BACKUP, add new conn. failed\n"); 914 return; 915 } 916 if (!(flags & IP_VS_CONN_F_TEMPLATE)) 917 kfree(param->pe_data); 918 } 919 920 if (opt) { 921 cp->in_seq = opt->in_seq; 922 cp->out_seq = opt->out_seq; 923 } 924 atomic_set(&cp->in_pkts, sysctl_sync_threshold(ipvs)); 925 cp->state = state; 926 cp->old_state = cp->state; 927 /* 928 * For Ver 0 messages style 929 * - Not possible to recover the right timeout for templates 930 * - can not find the right fwmark 931 * virtual service. If needed, we can do it for 932 * non-fwmark persistent services. 933 * Ver 1 messages style. 934 * - No problem. 935 */ 936 if (timeout) { 937 if (timeout > MAX_SCHEDULE_TIMEOUT / HZ) 938 timeout = MAX_SCHEDULE_TIMEOUT / HZ; 939 cp->timeout = timeout*HZ; 940 } else { 941 struct ip_vs_proto_data *pd; 942 943 pd = ip_vs_proto_data_get(ipvs, protocol); 944 if (!(flags & IP_VS_CONN_F_TEMPLATE) && pd && pd->timeout_table) 945 cp->timeout = pd->timeout_table[state]; 946 else 947 cp->timeout = (3*60*HZ); 948 } 949 ip_vs_conn_put(cp); 950 } 951 952 /* Check for incompatible flags */ 953 static bool ip_vs_sync_validate_flags(u32 flags) 954 { 955 /* We do not expect NO_CPORT, especially to allow lookups 956 * to hit templates 957 */ 958 if (flags & IP_VS_CONN_F_NO_CPORT) { 959 if (flags & IP_VS_CONN_F_TEMPLATE) 960 return false; 961 } 962 if (flags & IP_VS_CONN_F_ONE_PACKET) 963 return false; 964 return true; 965 } 966 967 /* 968 * Process received multicast message for Version 0 969 */ 970 static void ip_vs_process_message_v0(struct netns_ipvs *ipvs, const char *buffer, 971 const size_t buflen) 972 { 973 struct ip_vs_sync_mesg_v0 *m = (struct ip_vs_sync_mesg_v0 *)buffer; 974 struct ip_vs_sync_conn_v0 *s; 975 struct ip_vs_sync_conn_options *opt; 976 struct ip_vs_protocol *pp; 977 struct ip_vs_conn_param param; 978 char *p; 979 int i; 980 981 p = (char *)buffer + sizeof(struct ip_vs_sync_mesg_v0); 982 for (i=0; i<m->nr_conns; i++) { 983 unsigned int flags, state; 984 985 if (p + SIMPLE_CONN_SIZE > buffer+buflen) { 986 IP_VS_ERR_RL("BACKUP v0, bogus conn\n"); 987 return; 988 } 989 s = (struct ip_vs_sync_conn_v0 *) p; 990 flags = ntohs(s->flags); 991 if (flags & IP_VS_CONN_F_SEQ_MASK) { 992 opt = (struct ip_vs_sync_conn_options *)&s[1]; 993 p += FULL_CONN_SIZE; 994 if (p > buffer+buflen) { 995 IP_VS_ERR_RL("BACKUP v0, Dropping buffer bogus conn options\n"); 996 return; 997 } 998 } else { 999 opt = NULL; 1000 p += SIMPLE_CONN_SIZE; 1001 } 1002 1003 if (!ip_vs_sync_validate_flags(flags)) { 1004 IP_VS_DBG(2, "BACKUP v0, Invalid flags 0x%X\n", flags); 1005 continue; 1006 } 1007 flags &= IP_VS_CONN_F_BACKUP_MASK; 1008 flags |= IP_VS_CONN_F_SYNC; 1009 1010 state = ntohs(s->state); 1011 if (!(flags & IP_VS_CONN_F_TEMPLATE)) { 1012 pp = ip_vs_proto_get(s->protocol); 1013 if (!pp) { 1014 IP_VS_DBG(2, "BACKUP v0, Unsupported protocol %u\n", 1015 s->protocol); 1016 continue; 1017 } 1018 if (state >= pp->num_states) { 1019 IP_VS_DBG(2, "BACKUP v0, Invalid %s state %u\n", 1020 pp->name, state); 1021 continue; 1022 } 1023 } else if (state >= IP_VS_CTPL_S_LAST) { 1024 IP_VS_DBG(7, "BACKUP v0, Invalid tpl state %u\n", 1025 state); 1026 continue; 1027 } 1028 1029 ip_vs_conn_fill_param(ipvs, AF_INET, s->protocol, 1030 (const union nf_inet_addr *)&s->caddr, 1031 s->cport, 1032 (const union nf_inet_addr *)&s->vaddr, 1033 s->vport, ¶m); 1034 1035 /* Send timeout as Zero */ 1036 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->protocol, AF_INET, 1037 (union nf_inet_addr *)&s->daddr, s->dport, 1038 0, 0, opt); 1039 } 1040 } 1041 1042 /* 1043 * Handle options 1044 */ 1045 static inline int ip_vs_proc_seqopt(__u8 *p, unsigned int plen, 1046 __u32 *opt_flags, 1047 struct ip_vs_sync_conn_options *opt) 1048 { 1049 struct ip_vs_sync_conn_options *topt; 1050 1051 topt = (struct ip_vs_sync_conn_options *)p; 1052 1053 if (plen != sizeof(struct ip_vs_sync_conn_options)) { 1054 IP_VS_DBG(2, "BACKUP, bogus conn options length\n"); 1055 return -EINVAL; 1056 } 1057 if (*opt_flags & IPVS_OPT_F_SEQ_DATA) { 1058 IP_VS_DBG(2, "BACKUP, conn options found twice\n"); 1059 return -EINVAL; 1060 } 1061 ntoh_seq(&topt->in_seq, &opt->in_seq); 1062 ntoh_seq(&topt->out_seq, &opt->out_seq); 1063 *opt_flags |= IPVS_OPT_F_SEQ_DATA; 1064 return 0; 1065 } 1066 1067 static int ip_vs_proc_str(__u8 *p, unsigned int plen, unsigned int *data_len, 1068 __u8 **data, unsigned int maxlen, 1069 __u32 *opt_flags, __u32 flag) 1070 { 1071 if (plen > maxlen) { 1072 IP_VS_DBG(2, "BACKUP, bogus par.data len > %d\n", maxlen); 1073 return -EINVAL; 1074 } 1075 if (*opt_flags & flag) { 1076 IP_VS_DBG(2, "BACKUP, Par.data found twice 0x%x\n", flag); 1077 return -EINVAL; 1078 } 1079 *data_len = plen; 1080 *data = p; 1081 *opt_flags |= flag; 1082 return 0; 1083 } 1084 /* 1085 * Process a Version 1 sync. connection 1086 */ 1087 static inline int ip_vs_proc_sync_conn(struct netns_ipvs *ipvs, __u8 *p, __u8 *msg_end) 1088 { 1089 struct ip_vs_sync_conn_options opt; 1090 union ip_vs_sync_conn *s; 1091 struct ip_vs_protocol *pp; 1092 struct ip_vs_conn_param param; 1093 __u32 flags; 1094 unsigned int af, state, pe_data_len=0, pe_name_len=0; 1095 __u8 *pe_data=NULL, *pe_name=NULL; 1096 __u32 opt_flags=0; 1097 int retc=0; 1098 1099 s = (union ip_vs_sync_conn *) p; 1100 1101 if (s->v6.type & STYPE_F_INET6) { 1102 #ifdef CONFIG_IP_VS_IPV6 1103 af = AF_INET6; 1104 p += sizeof(struct ip_vs_sync_v6); 1105 #else 1106 IP_VS_DBG(3,"BACKUP, IPv6 msg received, and IPVS is not compiled for IPv6\n"); 1107 retc = 10; 1108 goto out; 1109 #endif 1110 } else if (!s->v4.type) { 1111 af = AF_INET; 1112 p += sizeof(struct ip_vs_sync_v4); 1113 } else { 1114 return -10; 1115 } 1116 if (p > msg_end) 1117 return -20; 1118 1119 /* Process optional params check Type & Len. */ 1120 while (p < msg_end) { 1121 int ptype; 1122 int plen; 1123 1124 if (p+2 > msg_end) 1125 return -30; 1126 ptype = *(p++); 1127 plen = *(p++); 1128 1129 if (!plen || ((p + plen) > msg_end)) 1130 return -40; 1131 /* Handle seq option p = param data */ 1132 switch (ptype & ~IPVS_OPT_F_PARAM) { 1133 case IPVS_OPT_SEQ_DATA: 1134 if (ip_vs_proc_seqopt(p, plen, &opt_flags, &opt)) 1135 return -50; 1136 break; 1137 1138 case IPVS_OPT_PE_DATA: 1139 if (ip_vs_proc_str(p, plen, &pe_data_len, &pe_data, 1140 IP_VS_PEDATA_MAXLEN, &opt_flags, 1141 IPVS_OPT_F_PE_DATA)) 1142 return -60; 1143 break; 1144 1145 case IPVS_OPT_PE_NAME: 1146 if (ip_vs_proc_str(p, plen,&pe_name_len, &pe_name, 1147 IP_VS_PENAME_MAXLEN, &opt_flags, 1148 IPVS_OPT_F_PE_NAME)) 1149 return -70; 1150 break; 1151 1152 default: 1153 /* Param data mandatory ? */ 1154 if (!(ptype & IPVS_OPT_F_PARAM)) { 1155 IP_VS_DBG(3, "BACKUP, Unknown mandatory param %d found\n", 1156 ptype & ~IPVS_OPT_F_PARAM); 1157 retc = 20; 1158 goto out; 1159 } 1160 } 1161 p += plen; /* Next option */ 1162 } 1163 1164 /* Get flags and Mask off unsupported */ 1165 flags = ntohl(s->v4.flags); 1166 if (!ip_vs_sync_validate_flags(flags)) { 1167 IP_VS_DBG(3, "BACKUP, Invalid flags 0x%X\n", flags); 1168 retc = 25; 1169 goto out; 1170 } 1171 flags &= IP_VS_CONN_F_BACKUP_MASK; 1172 flags |= IP_VS_CONN_F_SYNC; 1173 state = ntohs(s->v4.state); 1174 1175 if (!(flags & IP_VS_CONN_F_TEMPLATE)) { 1176 pp = ip_vs_proto_get(s->v4.protocol); 1177 if (!pp) { 1178 IP_VS_DBG(3,"BACKUP, Unsupported protocol %u\n", 1179 s->v4.protocol); 1180 retc = 30; 1181 goto out; 1182 } 1183 if (state >= pp->num_states) { 1184 IP_VS_DBG(3, "BACKUP, Invalid %s state %u\n", 1185 pp->name, state); 1186 retc = 40; 1187 goto out; 1188 } 1189 } else if (state >= IP_VS_CTPL_S_LAST) { 1190 IP_VS_DBG(7, "BACKUP, Invalid tpl state %u\n", state); 1191 retc = 40; 1192 goto out; 1193 } 1194 if (ip_vs_conn_fill_param_sync(ipvs, af, s, ¶m, pe_data, 1195 pe_data_len, pe_name, pe_name_len)) { 1196 retc = 50; 1197 goto out; 1198 } 1199 /* If only IPv4, just silent skip IPv6 */ 1200 if (af == AF_INET) 1201 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v4.protocol, af, 1202 (union nf_inet_addr *)&s->v4.daddr, s->v4.dport, 1203 ntohl(s->v4.timeout), ntohl(s->v4.fwmark), 1204 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL) 1205 ); 1206 #ifdef CONFIG_IP_VS_IPV6 1207 else 1208 ip_vs_proc_conn(ipvs, ¶m, flags, state, s->v6.protocol, af, 1209 (union nf_inet_addr *)&s->v6.daddr, s->v6.dport, 1210 ntohl(s->v6.timeout), ntohl(s->v6.fwmark), 1211 (opt_flags & IPVS_OPT_F_SEQ_DATA ? &opt : NULL) 1212 ); 1213 #endif 1214 ip_vs_pe_put(param.pe); 1215 return 0; 1216 /* Error exit */ 1217 out: 1218 IP_VS_DBG(2, "BACKUP, Single msg dropped err:%d\n", retc); 1219 return retc; 1220 1221 } 1222 /* 1223 * Process received multicast message and create the corresponding 1224 * ip_vs_conn entries. 1225 * Handles Version 0 & 1 1226 */ 1227 static void ip_vs_process_message(struct netns_ipvs *ipvs, __u8 *buffer, 1228 const size_t buflen) 1229 { 1230 struct ip_vs_sync_mesg *m2 = (struct ip_vs_sync_mesg *)buffer; 1231 __u8 *p, *msg_end; 1232 int i, nr_conns; 1233 1234 if (buflen < sizeof(struct ip_vs_sync_mesg_v0)) { 1235 IP_VS_DBG(2, "BACKUP, message header too short\n"); 1236 return; 1237 } 1238 1239 if (buflen != ntohs(m2->size)) { 1240 IP_VS_DBG(2, "BACKUP, bogus message size\n"); 1241 return; 1242 } 1243 /* SyncID sanity check */ 1244 if (ipvs->bcfg.syncid != 0 && m2->syncid != ipvs->bcfg.syncid) { 1245 IP_VS_DBG(7, "BACKUP, Ignoring syncid = %d\n", m2->syncid); 1246 return; 1247 } 1248 /* Handle version 1 message */ 1249 if ((m2->version == SYNC_PROTO_VER) && (m2->reserved == 0) 1250 && (m2->spare == 0)) { 1251 1252 msg_end = buffer + sizeof(struct ip_vs_sync_mesg); 1253 nr_conns = m2->nr_conns; 1254 1255 for (i=0; i<nr_conns; i++) { 1256 union ip_vs_sync_conn *s; 1257 unsigned int size; 1258 int retc; 1259 1260 p = msg_end; 1261 if (p + sizeof(s->v4) > buffer+buflen) { 1262 IP_VS_ERR_RL("BACKUP, Dropping buffer, too small\n"); 1263 return; 1264 } 1265 s = (union ip_vs_sync_conn *)p; 1266 size = ntohs(s->v4.ver_size) & SVER_MASK; 1267 msg_end = p + size; 1268 /* Basic sanity checks */ 1269 if (msg_end > buffer+buflen) { 1270 IP_VS_ERR_RL("BACKUP, Dropping buffer, msg > buffer\n"); 1271 return; 1272 } 1273 if (ntohs(s->v4.ver_size) >> SVER_SHIFT) { 1274 IP_VS_ERR_RL("BACKUP, Dropping buffer, Unknown version %d\n", 1275 ntohs(s->v4.ver_size) >> SVER_SHIFT); 1276 return; 1277 } 1278 /* Process a single sync_conn */ 1279 retc = ip_vs_proc_sync_conn(ipvs, p, msg_end); 1280 if (retc < 0) { 1281 IP_VS_ERR_RL("BACKUP, Dropping buffer, Err: %d in decoding\n", 1282 retc); 1283 return; 1284 } 1285 /* Make sure we have 32 bit alignment */ 1286 msg_end = p + ((size + 3) & ~3); 1287 } 1288 } else { 1289 /* Old type of message */ 1290 ip_vs_process_message_v0(ipvs, buffer, buflen); 1291 return; 1292 } 1293 } 1294 1295 1296 /* 1297 * Setup sndbuf (mode=1) or rcvbuf (mode=0) 1298 */ 1299 static void set_sock_size(struct sock *sk, int mode, int val) 1300 { 1301 /* setsockopt(sock, SOL_SOCKET, SO_SNDBUF, &val, sizeof(val)); */ 1302 /* setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)); */ 1303 lock_sock(sk); 1304 if (mode) { 1305 val = clamp_t(int, val, (SOCK_MIN_SNDBUF + 1) / 2, 1306 READ_ONCE(sysctl_wmem_max)); 1307 sk->sk_sndbuf = val * 2; 1308 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1309 } else { 1310 val = clamp_t(int, val, (SOCK_MIN_RCVBUF + 1) / 2, 1311 READ_ONCE(sysctl_rmem_max)); 1312 sk->sk_rcvbuf = val * 2; 1313 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 1314 } 1315 release_sock(sk); 1316 } 1317 1318 /* 1319 * Setup loopback of outgoing multicasts on a sending socket 1320 */ 1321 static void set_mcast_loop(struct sock *sk, u_char loop) 1322 { 1323 /* setsockopt(sock, SOL_IP, IP_MULTICAST_LOOP, &loop, sizeof(loop)); */ 1324 inet_assign_bit(MC_LOOP, sk, loop); 1325 #ifdef CONFIG_IP_VS_IPV6 1326 if (READ_ONCE(sk->sk_family) == AF_INET6) { 1327 /* IPV6_MULTICAST_LOOP */ 1328 inet6_assign_bit(MC6_LOOP, sk, loop); 1329 } 1330 #endif 1331 } 1332 1333 /* 1334 * Specify TTL for outgoing multicasts on a sending socket 1335 */ 1336 static void set_mcast_ttl(struct sock *sk, u_char ttl) 1337 { 1338 struct inet_sock *inet = inet_sk(sk); 1339 1340 /* setsockopt(sock, SOL_IP, IP_MULTICAST_TTL, &ttl, sizeof(ttl)); */ 1341 lock_sock(sk); 1342 WRITE_ONCE(inet->mc_ttl, ttl); 1343 #ifdef CONFIG_IP_VS_IPV6 1344 if (sk->sk_family == AF_INET6) { 1345 struct ipv6_pinfo *np = inet6_sk(sk); 1346 1347 /* IPV6_MULTICAST_HOPS */ 1348 WRITE_ONCE(np->mcast_hops, ttl); 1349 } 1350 #endif 1351 release_sock(sk); 1352 } 1353 1354 /* Control fragmentation of messages */ 1355 static void set_mcast_pmtudisc(struct sock *sk, int val) 1356 { 1357 struct inet_sock *inet = inet_sk(sk); 1358 1359 /* setsockopt(sock, SOL_IP, IP_MTU_DISCOVER, &val, sizeof(val)); */ 1360 lock_sock(sk); 1361 WRITE_ONCE(inet->pmtudisc, val); 1362 #ifdef CONFIG_IP_VS_IPV6 1363 if (sk->sk_family == AF_INET6) { 1364 struct ipv6_pinfo *np = inet6_sk(sk); 1365 1366 /* IPV6_MTU_DISCOVER */ 1367 WRITE_ONCE(np->pmtudisc, val); 1368 } 1369 #endif 1370 release_sock(sk); 1371 } 1372 1373 /* 1374 * Specifiy default interface for outgoing multicasts 1375 */ 1376 static int set_mcast_if(struct sock *sk, struct net_device *dev) 1377 { 1378 struct inet_sock *inet = inet_sk(sk); 1379 1380 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) 1381 return -EINVAL; 1382 1383 lock_sock(sk); 1384 inet->mc_index = dev->ifindex; 1385 /* inet->mc_addr = 0; */ 1386 #ifdef CONFIG_IP_VS_IPV6 1387 if (sk->sk_family == AF_INET6) { 1388 struct ipv6_pinfo *np = inet6_sk(sk); 1389 1390 /* IPV6_MULTICAST_IF */ 1391 WRITE_ONCE(np->mcast_oif, dev->ifindex); 1392 } 1393 #endif 1394 release_sock(sk); 1395 1396 return 0; 1397 } 1398 1399 1400 /* 1401 * Join a multicast group. 1402 * the group is specified by a class D multicast address 224.0.0.0/8 1403 * in the in_addr structure passed in as a parameter. 1404 */ 1405 static int 1406 join_mcast_group(struct sock *sk, struct in_addr *addr, struct net_device *dev) 1407 { 1408 struct ip_mreqn mreq; 1409 int ret; 1410 1411 memset(&mreq, 0, sizeof(mreq)); 1412 memcpy(&mreq.imr_multiaddr, addr, sizeof(struct in_addr)); 1413 1414 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) 1415 return -EINVAL; 1416 1417 mreq.imr_ifindex = dev->ifindex; 1418 1419 lock_sock(sk); 1420 ret = ip_mc_join_group(sk, &mreq); 1421 release_sock(sk); 1422 1423 return ret; 1424 } 1425 1426 #ifdef CONFIG_IP_VS_IPV6 1427 static int join_mcast_group6(struct sock *sk, struct in6_addr *addr, 1428 struct net_device *dev) 1429 { 1430 int ret; 1431 1432 if (sk->sk_bound_dev_if && dev->ifindex != sk->sk_bound_dev_if) 1433 return -EINVAL; 1434 1435 lock_sock(sk); 1436 ret = ipv6_sock_mc_join(sk, dev->ifindex, addr); 1437 release_sock(sk); 1438 1439 return ret; 1440 } 1441 #endif 1442 1443 static int bind_mcastif_addr(struct socket *sock, struct net_device *dev) 1444 { 1445 __be32 addr; 1446 struct sockaddr_in sin; 1447 1448 addr = inet_select_addr(dev, 0, RT_SCOPE_UNIVERSE); 1449 if (!addr) 1450 pr_err("You probably need to specify IP address on " 1451 "multicast interface.\n"); 1452 1453 IP_VS_DBG(7, "binding socket with (%s) %pI4\n", 1454 dev->name, &addr); 1455 1456 /* Now bind the socket with the address of multicast interface */ 1457 sin.sin_family = AF_INET; 1458 sin.sin_addr.s_addr = addr; 1459 sin.sin_port = 0; 1460 1461 return kernel_bind(sock, (struct sockaddr_unsized *)&sin, sizeof(sin)); 1462 } 1463 1464 static void get_mcast_sockaddr(union ipvs_sockaddr *sa, int *salen, 1465 struct ipvs_sync_daemon_cfg *c, int id) 1466 { 1467 if (AF_INET6 == c->mcast_af) { 1468 sa->in6 = (struct sockaddr_in6) { 1469 .sin6_family = AF_INET6, 1470 .sin6_port = htons(c->mcast_port + id), 1471 }; 1472 sa->in6.sin6_addr = c->mcast_group.in6; 1473 *salen = sizeof(sa->in6); 1474 } else { 1475 sa->in = (struct sockaddr_in) { 1476 .sin_family = AF_INET, 1477 .sin_port = htons(c->mcast_port + id), 1478 }; 1479 sa->in.sin_addr = c->mcast_group.in; 1480 *salen = sizeof(sa->in); 1481 } 1482 } 1483 1484 /* 1485 * Set up sending multicast socket over UDP 1486 */ 1487 static int make_send_sock(struct netns_ipvs *ipvs, int id, 1488 struct net_device *dev, struct socket **sock_ret) 1489 { 1490 /* multicast addr */ 1491 union ipvs_sockaddr mcast_addr; 1492 struct socket *sock; 1493 int result, salen; 1494 1495 /* First create a socket */ 1496 result = sock_create_kern(ipvs->net, ipvs->mcfg.mcast_af, SOCK_DGRAM, 1497 IPPROTO_UDP, &sock); 1498 if (result < 0) { 1499 pr_err("Error during creation of socket; terminating\n"); 1500 goto error; 1501 } 1502 *sock_ret = sock; 1503 result = set_mcast_if(sock->sk, dev); 1504 if (result < 0) { 1505 pr_err("Error setting outbound mcast interface\n"); 1506 goto error; 1507 } 1508 1509 set_mcast_loop(sock->sk, 0); 1510 set_mcast_ttl(sock->sk, ipvs->mcfg.mcast_ttl); 1511 /* Allow fragmentation if MTU changes */ 1512 set_mcast_pmtudisc(sock->sk, IP_PMTUDISC_DONT); 1513 result = sysctl_sync_sock_size(ipvs); 1514 if (result > 0) 1515 set_sock_size(sock->sk, 1, result); 1516 1517 if (AF_INET == ipvs->mcfg.mcast_af) 1518 result = bind_mcastif_addr(sock, dev); 1519 else 1520 result = 0; 1521 if (result < 0) { 1522 pr_err("Error binding address of the mcast interface\n"); 1523 goto error; 1524 } 1525 1526 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->mcfg, id); 1527 result = kernel_connect(sock, (struct sockaddr_unsized *)&mcast_addr, 1528 salen, 0); 1529 if (result < 0) { 1530 pr_err("Error connecting to the multicast addr\n"); 1531 goto error; 1532 } 1533 1534 return 0; 1535 1536 error: 1537 return result; 1538 } 1539 1540 1541 /* 1542 * Set up receiving multicast socket over UDP 1543 */ 1544 static int make_receive_sock(struct netns_ipvs *ipvs, int id, 1545 struct net_device *dev, struct socket **sock_ret) 1546 { 1547 /* multicast addr */ 1548 union ipvs_sockaddr mcast_addr; 1549 struct socket *sock; 1550 int result, salen; 1551 1552 /* First create a socket */ 1553 result = sock_create_kern(ipvs->net, ipvs->bcfg.mcast_af, SOCK_DGRAM, 1554 IPPROTO_UDP, &sock); 1555 if (result < 0) { 1556 pr_err("Error during creation of socket; terminating\n"); 1557 goto error; 1558 } 1559 *sock_ret = sock; 1560 /* it is equivalent to the REUSEADDR option in user-space */ 1561 sock->sk->sk_reuse = SK_CAN_REUSE; 1562 result = sysctl_sync_sock_size(ipvs); 1563 if (result > 0) 1564 set_sock_size(sock->sk, 0, result); 1565 1566 get_mcast_sockaddr(&mcast_addr, &salen, &ipvs->bcfg, id); 1567 sock->sk->sk_bound_dev_if = dev->ifindex; 1568 result = kernel_bind(sock, (struct sockaddr_unsized *)&mcast_addr, salen); 1569 if (result < 0) { 1570 pr_err("Error binding to the multicast addr\n"); 1571 goto error; 1572 } 1573 1574 /* join the multicast group */ 1575 #ifdef CONFIG_IP_VS_IPV6 1576 if (ipvs->bcfg.mcast_af == AF_INET6) 1577 result = join_mcast_group6(sock->sk, &mcast_addr.in6.sin6_addr, 1578 dev); 1579 else 1580 #endif 1581 result = join_mcast_group(sock->sk, &mcast_addr.in.sin_addr, 1582 dev); 1583 if (result < 0) { 1584 pr_err("Error joining to the multicast group\n"); 1585 goto error; 1586 } 1587 1588 return 0; 1589 1590 error: 1591 return result; 1592 } 1593 1594 1595 static int 1596 ip_vs_send_async(struct socket *sock, const char *buffer, const size_t length) 1597 { 1598 struct msghdr msg = {.msg_flags = MSG_DONTWAIT|MSG_NOSIGNAL}; 1599 struct kvec iov; 1600 int len; 1601 1602 iov.iov_base = (void *)buffer; 1603 iov.iov_len = length; 1604 1605 len = kernel_sendmsg(sock, &msg, &iov, 1, (size_t)(length)); 1606 1607 return len; 1608 } 1609 1610 static int 1611 ip_vs_send_sync_msg(struct socket *sock, struct ip_vs_sync_mesg *msg) 1612 { 1613 int msize; 1614 int ret; 1615 1616 msize = ntohs(msg->size); 1617 1618 ret = ip_vs_send_async(sock, (char *)msg, msize); 1619 if (ret >= 0 || ret == -EAGAIN) 1620 return ret; 1621 pr_err("ip_vs_send_async error %d\n", ret); 1622 return 0; 1623 } 1624 1625 static int 1626 ip_vs_receive(struct socket *sock, char *buffer, const size_t buflen) 1627 { 1628 struct msghdr msg = {NULL,}; 1629 struct kvec iov = {buffer, buflen}; 1630 int len; 1631 1632 /* Receive a packet */ 1633 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &iov, 1, buflen); 1634 len = sock_recvmsg(sock, &msg, MSG_DONTWAIT); 1635 if (len < 0) 1636 return len; 1637 1638 return len; 1639 } 1640 1641 /* Wakeup the master thread for sending */ 1642 static void master_wakeup_work_handler(struct work_struct *work) 1643 { 1644 struct ipvs_master_sync_state *ms = 1645 container_of(work, struct ipvs_master_sync_state, 1646 master_wakeup_work.work); 1647 struct netns_ipvs *ipvs = ms->ipvs; 1648 1649 spin_lock_bh(&ipvs->sync_lock); 1650 if (ms->sync_queue_len && 1651 ms->sync_queue_delay < IPVS_SYNC_WAKEUP_RATE) { 1652 int id = (int)(ms - ipvs->ms); 1653 1654 ms->sync_queue_delay = IPVS_SYNC_WAKEUP_RATE; 1655 wake_up_process(ipvs->master_tinfo[id].task); 1656 } 1657 spin_unlock_bh(&ipvs->sync_lock); 1658 } 1659 1660 /* Get next buffer to send */ 1661 static inline struct ip_vs_sync_buff * 1662 next_sync_buff(struct netns_ipvs *ipvs, struct ipvs_master_sync_state *ms) 1663 { 1664 struct ip_vs_sync_buff *sb; 1665 1666 sb = sb_dequeue(ipvs, ms); 1667 if (sb) 1668 return sb; 1669 /* Do not delay entries in buffer for more than 2 seconds */ 1670 return get_curr_sync_buff(ipvs, ms, IPVS_SYNC_FLUSH_TIME); 1671 } 1672 1673 static int sync_thread_master(void *data) 1674 { 1675 struct ip_vs_sync_thread_data *tinfo = data; 1676 struct netns_ipvs *ipvs = tinfo->ipvs; 1677 struct ipvs_master_sync_state *ms = &ipvs->ms[tinfo->id]; 1678 struct sock *sk = tinfo->sock->sk; 1679 struct ip_vs_sync_buff *sb; 1680 1681 pr_info("sync thread started: state = MASTER, mcast_ifn = %s, " 1682 "syncid = %d, id = %d\n", 1683 ipvs->mcfg.mcast_ifn, ipvs->mcfg.syncid, tinfo->id); 1684 1685 for (;;) { 1686 sb = next_sync_buff(ipvs, ms); 1687 if (unlikely(kthread_should_stop())) 1688 break; 1689 if (!sb) { 1690 schedule_timeout(IPVS_SYNC_CHECK_PERIOD); 1691 continue; 1692 } 1693 while (ip_vs_send_sync_msg(tinfo->sock, sb->mesg) < 0) { 1694 /* (Ab)use interruptible sleep to avoid increasing 1695 * the load avg. 1696 */ 1697 __wait_event_interruptible(*sk_sleep(sk), 1698 sock_writeable(sk) || 1699 kthread_should_stop()); 1700 if (unlikely(kthread_should_stop())) 1701 goto done; 1702 } 1703 ip_vs_sync_buff_release(sb); 1704 } 1705 1706 done: 1707 __set_current_state(TASK_RUNNING); 1708 if (sb) 1709 ip_vs_sync_buff_release(sb); 1710 1711 /* clean up the sync_buff queue */ 1712 while ((sb = sb_dequeue(ipvs, ms))) 1713 ip_vs_sync_buff_release(sb); 1714 __set_current_state(TASK_RUNNING); 1715 1716 /* clean up the current sync_buff */ 1717 sb = get_curr_sync_buff(ipvs, ms, 0); 1718 if (sb) 1719 ip_vs_sync_buff_release(sb); 1720 1721 return 0; 1722 } 1723 1724 1725 static int sync_thread_backup(void *data) 1726 { 1727 struct ip_vs_sync_thread_data *tinfo = data; 1728 struct netns_ipvs *ipvs = tinfo->ipvs; 1729 struct sock *sk = tinfo->sock->sk; 1730 struct udp_sock *up = udp_sk(sk); 1731 int len; 1732 1733 pr_info("sync thread started: state = BACKUP, mcast_ifn = %s, " 1734 "syncid = %d, id = %d\n", 1735 ipvs->bcfg.mcast_ifn, ipvs->bcfg.syncid, tinfo->id); 1736 1737 while (!kthread_should_stop()) { 1738 wait_event_interruptible(*sk_sleep(sk), 1739 !skb_queue_empty_lockless(&sk->sk_receive_queue) || 1740 !skb_queue_empty_lockless(&up->reader_queue) || 1741 kthread_should_stop()); 1742 1743 /* do we have data now? */ 1744 while (!skb_queue_empty_lockless(&sk->sk_receive_queue) || 1745 !skb_queue_empty_lockless(&up->reader_queue)) { 1746 len = ip_vs_receive(tinfo->sock, tinfo->buf, 1747 ipvs->bcfg.sync_maxlen); 1748 if (len <= 0) { 1749 if (len != -EAGAIN) 1750 pr_err("receiving message error\n"); 1751 break; 1752 } 1753 1754 ip_vs_process_message(ipvs, tinfo->buf, len); 1755 } 1756 } 1757 1758 return 0; 1759 } 1760 1761 1762 int start_sync_thread(struct netns_ipvs *ipvs, struct ipvs_sync_daemon_cfg *c, 1763 int state) 1764 { 1765 struct ip_vs_sync_thread_data *ti = NULL, *tinfo; 1766 struct task_struct *task; 1767 struct net_device *dev; 1768 char *name; 1769 int (*threadfn)(void *data); 1770 int id = 0, count, hlen; 1771 int result = -ENOMEM; 1772 u16 mtu, min_mtu; 1773 1774 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current)); 1775 IP_VS_DBG(7, "Each ip_vs_sync_conn entry needs %zd bytes\n", 1776 sizeof(struct ip_vs_sync_conn_v0)); 1777 1778 /* increase the module use count */ 1779 if (!ip_vs_use_count_inc()) 1780 return -ENOPROTOOPT; 1781 1782 /* Backup server can be started without services just to sync conns, 1783 * make sure conn_tab is created even if ipvs->enable is 0. 1784 */ 1785 if (state == IP_VS_STATE_BACKUP) { 1786 mutex_lock(&ipvs->service_mutex); 1787 if (!rcu_dereference_protected(ipvs->conn_tab, 1)) { 1788 int lfactor = sysctl_conn_lfactor(ipvs); 1789 int new_size = ip_vs_conn_desired_size(ipvs, NULL, 1790 lfactor); 1791 struct ip_vs_rht *tc_new; 1792 1793 tc_new = ip_vs_conn_tab_alloc(ipvs, new_size, lfactor); 1794 if (!tc_new) { 1795 mutex_unlock(&ipvs->service_mutex); 1796 result = -ENOMEM; 1797 goto out_module; 1798 } 1799 rcu_assign_pointer(ipvs->conn_tab, tc_new); 1800 } 1801 mutex_unlock(&ipvs->service_mutex); 1802 } 1803 1804 /* Do not hold one mutex and then to block on another */ 1805 for (;;) { 1806 rtnl_lock(); 1807 if (mutex_trylock(&ipvs->sync_mutex)) 1808 break; 1809 rtnl_unlock(); 1810 mutex_lock(&ipvs->sync_mutex); 1811 if (rtnl_trylock()) 1812 break; 1813 mutex_unlock(&ipvs->sync_mutex); 1814 } 1815 1816 if (!ipvs->sync_state) { 1817 count = clamp(sysctl_sync_ports(ipvs), 1, IPVS_SYNC_PORTS_MAX); 1818 ipvs->threads_mask = count - 1; 1819 } else 1820 count = ipvs->threads_mask + 1; 1821 1822 if (c->mcast_af == AF_UNSPEC) { 1823 c->mcast_af = AF_INET; 1824 c->mcast_group.ip = cpu_to_be32(IP_VS_SYNC_GROUP); 1825 } 1826 if (!c->mcast_port) 1827 c->mcast_port = IP_VS_SYNC_PORT; 1828 if (!c->mcast_ttl) 1829 c->mcast_ttl = 1; 1830 1831 dev = __dev_get_by_name(ipvs->net, c->mcast_ifn); 1832 if (!dev) { 1833 pr_err("Unknown mcast interface: %s\n", c->mcast_ifn); 1834 result = -ENODEV; 1835 goto out_early; 1836 } 1837 hlen = (AF_INET6 == c->mcast_af) ? 1838 sizeof(struct ipv6hdr) + sizeof(struct udphdr) : 1839 sizeof(struct iphdr) + sizeof(struct udphdr); 1840 mtu = (state == IP_VS_STATE_BACKUP) ? 1841 clamp(dev->mtu, 1500U, 65535U) : 1500U; 1842 min_mtu = (state == IP_VS_STATE_BACKUP) ? 1024 : 1; 1843 1844 if (c->sync_maxlen) 1845 c->sync_maxlen = clamp_t(unsigned int, 1846 c->sync_maxlen, min_mtu, 1847 65535 - hlen); 1848 else 1849 c->sync_maxlen = mtu - hlen; 1850 1851 if (state == IP_VS_STATE_MASTER) { 1852 result = -EEXIST; 1853 if (ipvs->ms) 1854 goto out_early; 1855 1856 ipvs->mcfg = *c; 1857 name = "ipvs-m:%d:%d"; 1858 threadfn = sync_thread_master; 1859 } else if (state == IP_VS_STATE_BACKUP) { 1860 result = -EEXIST; 1861 if (ipvs->backup_tinfo) 1862 goto out_early; 1863 1864 ipvs->bcfg = *c; 1865 name = "ipvs-b:%d:%d"; 1866 threadfn = sync_thread_backup; 1867 } else { 1868 result = -EINVAL; 1869 goto out_early; 1870 } 1871 1872 if (state == IP_VS_STATE_MASTER) { 1873 struct ipvs_master_sync_state *ms; 1874 1875 result = -ENOMEM; 1876 ipvs->ms = kzalloc_objs(ipvs->ms[0], count); 1877 if (!ipvs->ms) 1878 goto out; 1879 ms = ipvs->ms; 1880 for (id = 0; id < count; id++, ms++) { 1881 INIT_LIST_HEAD(&ms->sync_queue); 1882 ms->sync_queue_len = 0; 1883 ms->sync_queue_delay = 0; 1884 INIT_DELAYED_WORK(&ms->master_wakeup_work, 1885 master_wakeup_work_handler); 1886 ms->ipvs = ipvs; 1887 } 1888 } 1889 result = -ENOMEM; 1890 ti = kzalloc_objs(struct ip_vs_sync_thread_data, count); 1891 if (!ti) 1892 goto out; 1893 1894 for (id = 0; id < count; id++) { 1895 tinfo = &ti[id]; 1896 tinfo->ipvs = ipvs; 1897 if (state == IP_VS_STATE_BACKUP) { 1898 result = -ENOMEM; 1899 tinfo->buf = kmalloc(ipvs->bcfg.sync_maxlen, 1900 GFP_KERNEL); 1901 if (!tinfo->buf) 1902 goto out; 1903 } 1904 tinfo->id = id; 1905 if (state == IP_VS_STATE_MASTER) 1906 result = make_send_sock(ipvs, id, dev, &tinfo->sock); 1907 else 1908 result = make_receive_sock(ipvs, id, dev, &tinfo->sock); 1909 if (result < 0) 1910 goto out; 1911 1912 task = kthread_run(threadfn, tinfo, name, ipvs->gen, id); 1913 if (IS_ERR(task)) { 1914 result = PTR_ERR(task); 1915 goto out; 1916 } 1917 tinfo->task = task; 1918 } 1919 1920 /* mark as active */ 1921 1922 if (state == IP_VS_STATE_MASTER) 1923 ipvs->master_tinfo = ti; 1924 else 1925 ipvs->backup_tinfo = ti; 1926 spin_lock_bh(&ipvs->sync_buff_lock); 1927 ipvs->sync_state |= state; 1928 spin_unlock_bh(&ipvs->sync_buff_lock); 1929 1930 mutex_unlock(&ipvs->sync_mutex); 1931 rtnl_unlock(); 1932 1933 return 0; 1934 1935 out: 1936 /* We do not need RTNL lock anymore, release it here so that 1937 * sock_release below can use rtnl_lock to leave the mcast group. 1938 */ 1939 rtnl_unlock(); 1940 id = min(id, count - 1); 1941 if (ti) { 1942 for (tinfo = ti + id; tinfo >= ti; tinfo--) { 1943 if (tinfo->task) 1944 kthread_stop(tinfo->task); 1945 } 1946 } 1947 if (!(ipvs->sync_state & IP_VS_STATE_MASTER)) { 1948 kfree(ipvs->ms); 1949 ipvs->ms = NULL; 1950 } 1951 mutex_unlock(&ipvs->sync_mutex); 1952 1953 /* No more mutexes, release socks */ 1954 if (ti) { 1955 for (tinfo = ti + id; tinfo >= ti; tinfo--) { 1956 if (tinfo->sock) 1957 sock_release(tinfo->sock); 1958 kfree(tinfo->buf); 1959 } 1960 kfree(ti); 1961 } 1962 1963 /* decrease the module use count */ 1964 ip_vs_use_count_dec(); 1965 return result; 1966 1967 out_early: 1968 mutex_unlock(&ipvs->sync_mutex); 1969 rtnl_unlock(); 1970 1971 out_module: 1972 /* decrease the module use count */ 1973 ip_vs_use_count_dec(); 1974 return result; 1975 } 1976 1977 1978 int stop_sync_thread(struct netns_ipvs *ipvs, int state) 1979 { 1980 struct ip_vs_sync_thread_data *ti, *tinfo; 1981 int id; 1982 int retc = -EINVAL; 1983 1984 IP_VS_DBG(7, "%s(): pid %d\n", __func__, task_pid_nr(current)); 1985 1986 mutex_lock(&ipvs->sync_mutex); 1987 if (state == IP_VS_STATE_MASTER) { 1988 retc = -ESRCH; 1989 if (!ipvs->ms) 1990 goto err; 1991 ti = ipvs->master_tinfo; 1992 1993 /* 1994 * The lock synchronizes with sb_queue_tail(), so that we don't 1995 * add sync buffers to the queue, when we are already in 1996 * progress of stopping the master sync daemon. 1997 */ 1998 1999 spin_lock_bh(&ipvs->sync_buff_lock); 2000 spin_lock(&ipvs->sync_lock); 2001 ipvs->sync_state &= ~IP_VS_STATE_MASTER; 2002 spin_unlock(&ipvs->sync_lock); 2003 spin_unlock_bh(&ipvs->sync_buff_lock); 2004 2005 retc = 0; 2006 for (id = ipvs->threads_mask; id >= 0; id--) { 2007 struct ipvs_master_sync_state *ms = &ipvs->ms[id]; 2008 int ret; 2009 2010 tinfo = &ti[id]; 2011 pr_info("stopping master sync thread %d ...\n", 2012 task_pid_nr(tinfo->task)); 2013 cancel_delayed_work_sync(&ms->master_wakeup_work); 2014 ret = kthread_stop(tinfo->task); 2015 if (retc >= 0) 2016 retc = ret; 2017 } 2018 kfree(ipvs->ms); 2019 ipvs->ms = NULL; 2020 ipvs->master_tinfo = NULL; 2021 } else if (state == IP_VS_STATE_BACKUP) { 2022 retc = -ESRCH; 2023 if (!ipvs->backup_tinfo) 2024 goto err; 2025 ti = ipvs->backup_tinfo; 2026 2027 ipvs->sync_state &= ~IP_VS_STATE_BACKUP; 2028 retc = 0; 2029 for (id = ipvs->threads_mask; id >= 0; id--) { 2030 int ret; 2031 2032 tinfo = &ti[id]; 2033 pr_info("stopping backup sync thread %d ...\n", 2034 task_pid_nr(tinfo->task)); 2035 ret = kthread_stop(tinfo->task); 2036 if (retc >= 0) 2037 retc = ret; 2038 } 2039 ipvs->backup_tinfo = NULL; 2040 } else { 2041 goto err; 2042 } 2043 id = ipvs->threads_mask; 2044 mutex_unlock(&ipvs->sync_mutex); 2045 2046 /* No more mutexes, release socks */ 2047 for (tinfo = ti + id; tinfo >= ti; tinfo--) { 2048 if (tinfo->sock) 2049 sock_release(tinfo->sock); 2050 kfree(tinfo->buf); 2051 } 2052 kfree(ti); 2053 2054 /* decrease the module use count */ 2055 ip_vs_use_count_dec(); 2056 return retc; 2057 2058 err: 2059 mutex_unlock(&ipvs->sync_mutex); 2060 return retc; 2061 } 2062 2063 /* 2064 * Initialize data struct for each netns 2065 */ 2066 int __net_init ip_vs_sync_net_init(struct netns_ipvs *ipvs) 2067 { 2068 __mutex_init(&ipvs->sync_mutex, "ipvs->sync_mutex", &__ipvs_sync_key); 2069 spin_lock_init(&ipvs->sync_lock); 2070 spin_lock_init(&ipvs->sync_buff_lock); 2071 return 0; 2072 } 2073 2074 void ip_vs_sync_net_cleanup(struct netns_ipvs *ipvs) 2075 { 2076 int retc; 2077 2078 retc = stop_sync_thread(ipvs, IP_VS_STATE_MASTER); 2079 if (retc && retc != -ESRCH) 2080 pr_err("Failed to stop Master Daemon\n"); 2081 2082 retc = stop_sync_thread(ipvs, IP_VS_STATE_BACKUP); 2083 if (retc && retc != -ESRCH) 2084 pr_err("Failed to stop Backup Daemon\n"); 2085 } 2086