1// SPDX-License-Identifier: GPL-2.0-or-later
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 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
10 * Peter Kese <peter.kese@ijs.si>
11 * Julian Anastasov <ja@ssi.bg>
12 *
13 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
14 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
15 * and others. Many code here is taken from IP MASQ code of kernel 2.2.
16 *
17 * Changes:
18 */
19
20#define KMSG_COMPONENT "IPVS"
21#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
22
23#include <linux/interrupt.h>
24#include <linux/in.h>
25#include <linux/inet.h>
26#include <linux/net.h>
27#include <linux/kernel.h>
28#include <linux/module.h>
29#include <linux/proc_fs.h> /* for proc_net_* */
30#include <linux/slab.h>
31#include <linux/seq_file.h>
32#include <linux/jhash.h>
33#include <linux/random.h>
34#include <linux/rcupdate_wait.h>
35
36#include <net/net_namespace.h>
37#include <net/ip_vs.h>
38
39
40#ifndef CONFIG_IP_VS_TAB_BITS
41#define CONFIG_IP_VS_TAB_BITS 12
42#endif
43
44/*
45 * Connection hash size. Default is what was selected at compile time.
46*/
47static int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
48module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
49MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
50
51/* size and mask values */
52int ip_vs_conn_tab_size __read_mostly;
53static int ip_vs_conn_tab_mask __read_mostly;
54
55/*
56 * Connection hash table: for input and output packets lookups of IPVS
57 */
58static struct hlist_head *ip_vs_conn_tab __read_mostly;
59
60/* SLAB cache for IPVS connections */
61static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
62
63/* counter for no client port connections */
64static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
65
66/* random value for IPVS connection hash */
67static unsigned int ip_vs_conn_rnd __read_mostly;
68
69/*
70 * Fine locking granularity for big connection hash table
71 */
72#define CT_LOCKARRAY_BITS 5
73#define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
74#define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
75
76/* We need an addrstrlen that works with or without v6 */
77#ifdef CONFIG_IP_VS_IPV6
78#define IP_VS_ADDRSTRLEN INET6_ADDRSTRLEN
79#else
80#define IP_VS_ADDRSTRLEN (8+1)
81#endif
82
83struct ip_vs_aligned_lock
84{
85 spinlock_t l;
86} __attribute__((__aligned__(SMP_CACHE_BYTES)));
87
88/* lock array for conn table */
89static struct ip_vs_aligned_lock
90__ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
91
92static inline void ct_write_lock_bh(unsigned int key)
93{
94 spin_lock_bh(lock: &__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
95}
96
97static inline void ct_write_unlock_bh(unsigned int key)
98{
99 spin_unlock_bh(lock: &__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
100}
101
102static void ip_vs_conn_expire(struct timer_list *t);
103
104/*
105 * Returns hash value for IPVS connection entry
106 */
107static unsigned int ip_vs_conn_hashkey(struct netns_ipvs *ipvs, int af, unsigned int proto,
108 const union nf_inet_addr *addr,
109 __be16 port)
110{
111#ifdef CONFIG_IP_VS_IPV6
112 if (af == AF_INET6)
113 return (jhash_3words(a: jhash(key: addr, length: 16, initval: ip_vs_conn_rnd),
114 b: (__force u32)port, c: proto, initval: ip_vs_conn_rnd) ^
115 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
116#endif
117 return (jhash_3words(a: (__force u32)addr->ip, b: (__force u32)port, c: proto,
118 initval: ip_vs_conn_rnd) ^
119 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
120}
121
122static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
123 bool inverse)
124{
125 const union nf_inet_addr *addr;
126 __be16 port;
127
128 if (p->pe_data && p->pe->hashkey_raw)
129 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
130 ip_vs_conn_tab_mask;
131
132 if (likely(!inverse)) {
133 addr = p->caddr;
134 port = p->cport;
135 } else {
136 addr = p->vaddr;
137 port = p->vport;
138 }
139
140 return ip_vs_conn_hashkey(ipvs: p->ipvs, af: p->af, proto: p->protocol, addr, port);
141}
142
143static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
144{
145 struct ip_vs_conn_param p;
146
147 ip_vs_conn_fill_param(ipvs: cp->ipvs, af: cp->af, protocol: cp->protocol,
148 caddr: &cp->caddr, cport: cp->cport, NULL, vport: 0, p: &p);
149
150 if (cp->pe) {
151 p.pe = cp->pe;
152 p.pe_data = cp->pe_data;
153 p.pe_data_len = cp->pe_data_len;
154 }
155
156 return ip_vs_conn_hashkey_param(p: &p, inverse: false);
157}
158
159/*
160 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
161 * returns bool success.
162 */
163static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
164{
165 unsigned int hash;
166 int ret;
167
168 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
169 return 0;
170
171 /* Hash by protocol, client address and port */
172 hash = ip_vs_conn_hashkey_conn(cp);
173
174 ct_write_lock_bh(key: hash);
175 spin_lock(lock: &cp->lock);
176
177 if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
178 cp->flags |= IP_VS_CONN_F_HASHED;
179 refcount_inc(r: &cp->refcnt);
180 hlist_add_head_rcu(n: &cp->c_list, h: &ip_vs_conn_tab[hash]);
181 ret = 1;
182 } else {
183 pr_err("%s(): request for already hashed, called from %pS\n",
184 __func__, __builtin_return_address(0));
185 ret = 0;
186 }
187
188 spin_unlock(lock: &cp->lock);
189 ct_write_unlock_bh(key: hash);
190
191 return ret;
192}
193
194
195/*
196 * UNhashes ip_vs_conn from ip_vs_conn_tab.
197 * returns bool success. Caller should hold conn reference.
198 */
199static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
200{
201 unsigned int hash;
202 int ret;
203
204 /* unhash it and decrease its reference counter */
205 hash = ip_vs_conn_hashkey_conn(cp);
206
207 ct_write_lock_bh(key: hash);
208 spin_lock(lock: &cp->lock);
209
210 if (cp->flags & IP_VS_CONN_F_HASHED) {
211 hlist_del_rcu(n: &cp->c_list);
212 cp->flags &= ~IP_VS_CONN_F_HASHED;
213 refcount_dec(r: &cp->refcnt);
214 ret = 1;
215 } else
216 ret = 0;
217
218 spin_unlock(lock: &cp->lock);
219 ct_write_unlock_bh(key: hash);
220
221 return ret;
222}
223
224/* Try to unlink ip_vs_conn from ip_vs_conn_tab.
225 * returns bool success.
226 */
227static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
228{
229 unsigned int hash;
230 bool ret = false;
231
232 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
233 return refcount_dec_if_one(r: &cp->refcnt);
234
235 hash = ip_vs_conn_hashkey_conn(cp);
236
237 ct_write_lock_bh(key: hash);
238 spin_lock(lock: &cp->lock);
239
240 if (cp->flags & IP_VS_CONN_F_HASHED) {
241 /* Decrease refcnt and unlink conn only if we are last user */
242 if (refcount_dec_if_one(r: &cp->refcnt)) {
243 hlist_del_rcu(n: &cp->c_list);
244 cp->flags &= ~IP_VS_CONN_F_HASHED;
245 ret = true;
246 }
247 }
248
249 spin_unlock(lock: &cp->lock);
250 ct_write_unlock_bh(key: hash);
251
252 return ret;
253}
254
255
256/*
257 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
258 * Called for pkts coming from OUTside-to-INside.
259 * p->caddr, p->cport: pkt source address (foreign host)
260 * p->vaddr, p->vport: pkt dest address (load balancer)
261 */
262static inline struct ip_vs_conn *
263__ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
264{
265 unsigned int hash;
266 struct ip_vs_conn *cp;
267
268 hash = ip_vs_conn_hashkey_param(p, inverse: false);
269
270 rcu_read_lock();
271
272 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
273 if (p->cport == cp->cport && p->vport == cp->vport &&
274 cp->af == p->af &&
275 ip_vs_addr_equal(af: p->af, a: p->caddr, b: &cp->caddr) &&
276 ip_vs_addr_equal(af: p->af, a: p->vaddr, b: &cp->vaddr) &&
277 ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
278 p->protocol == cp->protocol &&
279 cp->ipvs == p->ipvs) {
280 if (!__ip_vs_conn_get(cp))
281 continue;
282 /* HIT */
283 rcu_read_unlock();
284 return cp;
285 }
286 }
287
288 rcu_read_unlock();
289
290 return NULL;
291}
292
293struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
294{
295 struct ip_vs_conn *cp;
296
297 cp = __ip_vs_conn_in_get(p);
298 if (!cp && atomic_read(v: &ip_vs_conn_no_cport_cnt)) {
299 struct ip_vs_conn_param cport_zero_p = *p;
300 cport_zero_p.cport = 0;
301 cp = __ip_vs_conn_in_get(p: &cport_zero_p);
302 }
303
304 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
305 ip_vs_proto_name(p->protocol),
306 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
307 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
308 cp ? "hit" : "not hit");
309
310 return cp;
311}
312
313static int
314ip_vs_conn_fill_param_proto(struct netns_ipvs *ipvs,
315 int af, const struct sk_buff *skb,
316 const struct ip_vs_iphdr *iph,
317 struct ip_vs_conn_param *p)
318{
319 __be16 _ports[2], *pptr;
320
321 pptr = frag_safe_skb_hp(skb, offset: iph->len, len: sizeof(_ports), buffer: _ports);
322 if (pptr == NULL)
323 return 1;
324
325 if (likely(!ip_vs_iph_inverse(iph)))
326 ip_vs_conn_fill_param(ipvs, af, protocol: iph->protocol, caddr: &iph->saddr,
327 cport: pptr[0], vaddr: &iph->daddr, vport: pptr[1], p);
328 else
329 ip_vs_conn_fill_param(ipvs, af, protocol: iph->protocol, caddr: &iph->daddr,
330 cport: pptr[1], vaddr: &iph->saddr, vport: pptr[0], p);
331 return 0;
332}
333
334struct ip_vs_conn *
335ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af,
336 const struct sk_buff *skb,
337 const struct ip_vs_iphdr *iph)
338{
339 struct ip_vs_conn_param p;
340
341 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, p: &p))
342 return NULL;
343
344 return ip_vs_conn_in_get(p: &p);
345}
346EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
347
348/* Get reference to connection template */
349struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
350{
351 unsigned int hash;
352 struct ip_vs_conn *cp;
353
354 hash = ip_vs_conn_hashkey_param(p, inverse: false);
355
356 rcu_read_lock();
357
358 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
359 if (unlikely(p->pe_data && p->pe->ct_match)) {
360 if (cp->ipvs != p->ipvs)
361 continue;
362 if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
363 if (__ip_vs_conn_get(cp))
364 goto out;
365 }
366 continue;
367 }
368
369 if (cp->af == p->af &&
370 ip_vs_addr_equal(af: p->af, a: p->caddr, b: &cp->caddr) &&
371 /* protocol should only be IPPROTO_IP if
372 * p->vaddr is a fwmark */
373 ip_vs_addr_equal(af: p->protocol == IPPROTO_IP ? AF_UNSPEC :
374 p->af, a: p->vaddr, b: &cp->vaddr) &&
375 p->vport == cp->vport && p->cport == cp->cport &&
376 cp->flags & IP_VS_CONN_F_TEMPLATE &&
377 p->protocol == cp->protocol &&
378 cp->ipvs == p->ipvs) {
379 if (__ip_vs_conn_get(cp))
380 goto out;
381 }
382 }
383 cp = NULL;
384
385 out:
386 rcu_read_unlock();
387
388 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
389 ip_vs_proto_name(p->protocol),
390 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
391 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
392 cp ? "hit" : "not hit");
393
394 return cp;
395}
396
397/* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
398 * Called for pkts coming from inside-to-OUTside.
399 * p->caddr, p->cport: pkt source address (inside host)
400 * p->vaddr, p->vport: pkt dest address (foreign host) */
401struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
402{
403 unsigned int hash;
404 struct ip_vs_conn *cp, *ret=NULL;
405 const union nf_inet_addr *saddr;
406 __be16 sport;
407
408 /*
409 * Check for "full" addressed entries
410 */
411 hash = ip_vs_conn_hashkey_param(p, inverse: true);
412
413 rcu_read_lock();
414
415 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
416 if (p->vport != cp->cport)
417 continue;
418
419 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
420 sport = cp->vport;
421 saddr = &cp->vaddr;
422 } else {
423 sport = cp->dport;
424 saddr = &cp->daddr;
425 }
426
427 if (p->cport == sport && cp->af == p->af &&
428 ip_vs_addr_equal(af: p->af, a: p->vaddr, b: &cp->caddr) &&
429 ip_vs_addr_equal(af: p->af, a: p->caddr, b: saddr) &&
430 p->protocol == cp->protocol &&
431 cp->ipvs == p->ipvs) {
432 if (!__ip_vs_conn_get(cp))
433 continue;
434 /* HIT */
435 ret = cp;
436 break;
437 }
438 }
439
440 rcu_read_unlock();
441
442 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
443 ip_vs_proto_name(p->protocol),
444 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
445 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
446 ret ? "hit" : "not hit");
447
448 return ret;
449}
450
451struct ip_vs_conn *
452ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
453 const struct sk_buff *skb,
454 const struct ip_vs_iphdr *iph)
455{
456 struct ip_vs_conn_param p;
457
458 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, p: &p))
459 return NULL;
460
461 return ip_vs_conn_out_get(p: &p);
462}
463EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
464
465/*
466 * Put back the conn and restart its timer with its timeout
467 */
468static void __ip_vs_conn_put_timer(struct ip_vs_conn *cp)
469{
470 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
471 0 : cp->timeout;
472 mod_timer(timer: &cp->timer, expires: jiffies+t);
473
474 __ip_vs_conn_put(cp);
475}
476
477void ip_vs_conn_put(struct ip_vs_conn *cp)
478{
479 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) &&
480 (refcount_read(r: &cp->refcnt) == 1) &&
481 !timer_pending(timer: &cp->timer))
482 /* expire connection immediately */
483 ip_vs_conn_expire(t: &cp->timer);
484 else
485 __ip_vs_conn_put_timer(cp);
486}
487
488/*
489 * Fill a no_client_port connection with a client port number
490 */
491void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
492{
493 if (ip_vs_conn_unhash(cp)) {
494 spin_lock_bh(lock: &cp->lock);
495 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
496 atomic_dec(v: &ip_vs_conn_no_cport_cnt);
497 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
498 cp->cport = cport;
499 }
500 spin_unlock_bh(lock: &cp->lock);
501
502 /* hash on new dport */
503 ip_vs_conn_hash(cp);
504 }
505}
506
507
508/*
509 * Bind a connection entry with the corresponding packet_xmit.
510 * Called by ip_vs_conn_new.
511 */
512static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
513{
514 switch (IP_VS_FWD_METHOD(cp)) {
515 case IP_VS_CONN_F_MASQ:
516 cp->packet_xmit = ip_vs_nat_xmit;
517 break;
518
519 case IP_VS_CONN_F_TUNNEL:
520#ifdef CONFIG_IP_VS_IPV6
521 if (cp->daf == AF_INET6)
522 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
523 else
524#endif
525 cp->packet_xmit = ip_vs_tunnel_xmit;
526 break;
527
528 case IP_VS_CONN_F_DROUTE:
529 cp->packet_xmit = ip_vs_dr_xmit;
530 break;
531
532 case IP_VS_CONN_F_LOCALNODE:
533 cp->packet_xmit = ip_vs_null_xmit;
534 break;
535
536 case IP_VS_CONN_F_BYPASS:
537 cp->packet_xmit = ip_vs_bypass_xmit;
538 break;
539 }
540}
541
542#ifdef CONFIG_IP_VS_IPV6
543static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
544{
545 switch (IP_VS_FWD_METHOD(cp)) {
546 case IP_VS_CONN_F_MASQ:
547 cp->packet_xmit = ip_vs_nat_xmit_v6;
548 break;
549
550 case IP_VS_CONN_F_TUNNEL:
551 if (cp->daf == AF_INET6)
552 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
553 else
554 cp->packet_xmit = ip_vs_tunnel_xmit;
555 break;
556
557 case IP_VS_CONN_F_DROUTE:
558 cp->packet_xmit = ip_vs_dr_xmit_v6;
559 break;
560
561 case IP_VS_CONN_F_LOCALNODE:
562 cp->packet_xmit = ip_vs_null_xmit;
563 break;
564
565 case IP_VS_CONN_F_BYPASS:
566 cp->packet_xmit = ip_vs_bypass_xmit_v6;
567 break;
568 }
569}
570#endif
571
572
573static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
574{
575 return atomic_read(v: &dest->activeconns)
576 + atomic_read(v: &dest->inactconns);
577}
578
579/*
580 * Bind a connection entry with a virtual service destination
581 * Called just after a new connection entry is created.
582 */
583static inline void
584ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
585{
586 unsigned int conn_flags;
587 __u32 flags;
588
589 /* if dest is NULL, then return directly */
590 if (!dest)
591 return;
592
593 /* Increase the refcnt counter of the dest */
594 ip_vs_dest_hold(dest);
595
596 conn_flags = atomic_read(v: &dest->conn_flags);
597 if (cp->protocol != IPPROTO_UDP)
598 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
599 flags = cp->flags;
600 /* Bind with the destination and its corresponding transmitter */
601 if (flags & IP_VS_CONN_F_SYNC) {
602 /* if the connection is not template and is created
603 * by sync, preserve the activity flag.
604 */
605 if (!(flags & IP_VS_CONN_F_TEMPLATE))
606 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
607 /* connections inherit forwarding method from dest */
608 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
609 }
610 flags |= conn_flags;
611 cp->flags = flags;
612 cp->dest = dest;
613
614 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
615 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
616 "dest->refcnt:%d\n",
617 ip_vs_proto_name(cp->protocol),
618 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
619 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
620 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
621 ip_vs_fwd_tag(cp), cp->state,
622 cp->flags, refcount_read(&cp->refcnt),
623 refcount_read(&dest->refcnt));
624
625 /* Update the connection counters */
626 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
627 /* It is a normal connection, so modify the counters
628 * according to the flags, later the protocol can
629 * update them on state change
630 */
631 if (!(flags & IP_VS_CONN_F_INACTIVE))
632 atomic_inc(v: &dest->activeconns);
633 else
634 atomic_inc(v: &dest->inactconns);
635 } else {
636 /* It is a persistent connection/template, so increase
637 the persistent connection counter */
638 atomic_inc(v: &dest->persistconns);
639 }
640
641 if (dest->u_threshold != 0 &&
642 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
643 dest->flags |= IP_VS_DEST_F_OVERLOAD;
644}
645
646
647/*
648 * Check if there is a destination for the connection, if so
649 * bind the connection to the destination.
650 */
651void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
652{
653 struct ip_vs_dest *dest;
654
655 rcu_read_lock();
656
657 /* This function is only invoked by the synchronization code. We do
658 * not currently support heterogeneous pools with synchronization,
659 * so we can make the assumption that the svc_af is the same as the
660 * dest_af
661 */
662 dest = ip_vs_find_dest(ipvs: cp->ipvs, svc_af: cp->af, dest_af: cp->af, daddr: &cp->daddr,
663 dport: cp->dport, vaddr: &cp->vaddr, vport: cp->vport,
664 protocol: cp->protocol, fwmark: cp->fwmark, flags: cp->flags);
665 if (dest) {
666 struct ip_vs_proto_data *pd;
667
668 spin_lock_bh(lock: &cp->lock);
669 if (cp->dest) {
670 spin_unlock_bh(lock: &cp->lock);
671 rcu_read_unlock();
672 return;
673 }
674
675 /* Applications work depending on the forwarding method
676 * but better to reassign them always when binding dest */
677 if (cp->app)
678 ip_vs_unbind_app(cp);
679
680 ip_vs_bind_dest(cp, dest);
681 spin_unlock_bh(lock: &cp->lock);
682
683 /* Update its packet transmitter */
684 cp->packet_xmit = NULL;
685#ifdef CONFIG_IP_VS_IPV6
686 if (cp->af == AF_INET6)
687 ip_vs_bind_xmit_v6(cp);
688 else
689#endif
690 ip_vs_bind_xmit(cp);
691
692 pd = ip_vs_proto_data_get(ipvs: cp->ipvs, proto: cp->protocol);
693 if (pd && atomic_read(v: &pd->appcnt))
694 ip_vs_bind_app(cp, pp: pd->pp);
695 }
696 rcu_read_unlock();
697}
698
699
700/*
701 * Unbind a connection entry with its VS destination
702 * Called by the ip_vs_conn_expire function.
703 */
704static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
705{
706 struct ip_vs_dest *dest = cp->dest;
707
708 if (!dest)
709 return;
710
711 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
712 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
713 "dest->refcnt:%d\n",
714 ip_vs_proto_name(cp->protocol),
715 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
716 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
717 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
718 ip_vs_fwd_tag(cp), cp->state,
719 cp->flags, refcount_read(&cp->refcnt),
720 refcount_read(&dest->refcnt));
721
722 /* Update the connection counters */
723 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
724 /* It is a normal connection, so decrease the inactconns
725 or activeconns counter */
726 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
727 atomic_dec(v: &dest->inactconns);
728 } else {
729 atomic_dec(v: &dest->activeconns);
730 }
731 } else {
732 /* It is a persistent connection/template, so decrease
733 the persistent connection counter */
734 atomic_dec(v: &dest->persistconns);
735 }
736
737 if (dest->l_threshold != 0) {
738 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
739 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
740 } else if (dest->u_threshold != 0) {
741 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
742 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
743 } else {
744 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
745 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
746 }
747
748 ip_vs_dest_put(dest);
749}
750
751static int expire_quiescent_template(struct netns_ipvs *ipvs,
752 struct ip_vs_dest *dest)
753{
754#ifdef CONFIG_SYSCTL
755 return ipvs->sysctl_expire_quiescent_template &&
756 (atomic_read(v: &dest->weight) == 0);
757#else
758 return 0;
759#endif
760}
761
762/*
763 * Checking if the destination of a connection template is available.
764 * If available, return 1, otherwise invalidate this connection
765 * template and return 0.
766 */
767int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest)
768{
769 struct ip_vs_dest *dest = ct->dest;
770 struct netns_ipvs *ipvs = ct->ipvs;
771
772 /*
773 * Checking the dest server status.
774 */
775 if ((dest == NULL) ||
776 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
777 expire_quiescent_template(ipvs, dest) ||
778 (cdest && (dest != cdest))) {
779 IP_VS_DBG_BUF(9, "check_template: dest not available for "
780 "protocol %s s:%s:%d v:%s:%d "
781 "-> d:%s:%d\n",
782 ip_vs_proto_name(ct->protocol),
783 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
784 ntohs(ct->cport),
785 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
786 ntohs(ct->vport),
787 IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
788 ntohs(ct->dport));
789
790 /*
791 * Invalidate the connection template
792 */
793 if (ct->vport != htons(0xffff)) {
794 if (ip_vs_conn_unhash(cp: ct)) {
795 ct->dport = htons(0xffff);
796 ct->vport = htons(0xffff);
797 ct->cport = 0;
798 ip_vs_conn_hash(cp: ct);
799 }
800 }
801
802 /*
803 * Simply decrease the refcnt of the template,
804 * don't restart its timer.
805 */
806 __ip_vs_conn_put(cp: ct);
807 return 0;
808 }
809 return 1;
810}
811
812static void ip_vs_conn_rcu_free(struct rcu_head *head)
813{
814 struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
815 rcu_head);
816
817 ip_vs_pe_put(cp->pe);
818 kfree(objp: cp->pe_data);
819 kmem_cache_free(s: ip_vs_conn_cachep, objp: cp);
820}
821
822/* Try to delete connection while not holding reference */
823static void ip_vs_conn_del(struct ip_vs_conn *cp)
824{
825 if (del_timer(timer: &cp->timer)) {
826 /* Drop cp->control chain too */
827 if (cp->control)
828 cp->timeout = 0;
829 ip_vs_conn_expire(t: &cp->timer);
830 }
831}
832
833/* Try to delete connection while holding reference */
834static void ip_vs_conn_del_put(struct ip_vs_conn *cp)
835{
836 if (del_timer(timer: &cp->timer)) {
837 /* Drop cp->control chain too */
838 if (cp->control)
839 cp->timeout = 0;
840 __ip_vs_conn_put(cp);
841 ip_vs_conn_expire(t: &cp->timer);
842 } else {
843 __ip_vs_conn_put(cp);
844 }
845}
846
847static void ip_vs_conn_expire(struct timer_list *t)
848{
849 struct ip_vs_conn *cp = from_timer(cp, t, timer);
850 struct netns_ipvs *ipvs = cp->ipvs;
851
852 /*
853 * do I control anybody?
854 */
855 if (atomic_read(v: &cp->n_control))
856 goto expire_later;
857
858 /* Unlink conn if not referenced anymore */
859 if (likely(ip_vs_conn_unlink(cp))) {
860 struct ip_vs_conn *ct = cp->control;
861
862 /* delete the timer if it is activated by other users */
863 del_timer(timer: &cp->timer);
864
865 /* does anybody control me? */
866 if (ct) {
867 bool has_ref = !cp->timeout && __ip_vs_conn_get(cp: ct);
868
869 ip_vs_control_del(cp);
870 /* Drop CTL or non-assured TPL if not used anymore */
871 if (has_ref && !atomic_read(v: &ct->n_control) &&
872 (!(ct->flags & IP_VS_CONN_F_TEMPLATE) ||
873 !(ct->state & IP_VS_CTPL_S_ASSURED))) {
874 IP_VS_DBG(4, "drop controlling connection\n");
875 ip_vs_conn_del_put(cp: ct);
876 } else if (has_ref) {
877 __ip_vs_conn_put(cp: ct);
878 }
879 }
880
881 if ((cp->flags & IP_VS_CONN_F_NFCT) &&
882 !(cp->flags & IP_VS_CONN_F_ONE_PACKET)) {
883 /* Do not access conntracks during subsys cleanup
884 * because nf_conntrack_find_get can not be used after
885 * conntrack cleanup for the net.
886 */
887 smp_rmb();
888 if (ipvs->enable)
889 ip_vs_conn_drop_conntrack(cp);
890 }
891
892 if (unlikely(cp->app != NULL))
893 ip_vs_unbind_app(cp);
894 ip_vs_unbind_dest(cp);
895 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
896 atomic_dec(v: &ip_vs_conn_no_cport_cnt);
897 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
898 ip_vs_conn_rcu_free(head: &cp->rcu_head);
899 else
900 call_rcu(head: &cp->rcu_head, func: ip_vs_conn_rcu_free);
901 atomic_dec(v: &ipvs->conn_count);
902 return;
903 }
904
905 expire_later:
906 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
907 refcount_read(&cp->refcnt),
908 atomic_read(&cp->n_control));
909
910 refcount_inc(r: &cp->refcnt);
911 cp->timeout = 60*HZ;
912
913 if (ipvs->sync_state & IP_VS_STATE_MASTER)
914 ip_vs_sync_conn(ipvs, cp, pkts: sysctl_sync_threshold(ipvs));
915
916 __ip_vs_conn_put_timer(cp);
917}
918
919/* Modify timer, so that it expires as soon as possible.
920 * Can be called without reference only if under RCU lock.
921 * We can have such chain of conns linked with ->control: DATA->CTL->TPL
922 * - DATA (eg. FTP) and TPL (persistence) can be present depending on setup
923 * - cp->timeout=0 indicates all conns from chain should be dropped but
924 * TPL is not dropped if in assured state
925 */
926void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
927{
928 /* Using mod_timer_pending will ensure the timer is not
929 * modified after the final del_timer in ip_vs_conn_expire.
930 */
931 if (timer_pending(timer: &cp->timer) &&
932 time_after(cp->timer.expires, jiffies))
933 mod_timer_pending(timer: &cp->timer, expires: jiffies);
934}
935
936
937/*
938 * Create a new connection entry and hash it into the ip_vs_conn_tab
939 */
940struct ip_vs_conn *
941ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
942 const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
943 struct ip_vs_dest *dest, __u32 fwmark)
944{
945 struct ip_vs_conn *cp;
946 struct netns_ipvs *ipvs = p->ipvs;
947 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs: p->ipvs,
948 proto: p->protocol);
949
950 cp = kmem_cache_alloc(cachep: ip_vs_conn_cachep, GFP_ATOMIC);
951 if (cp == NULL) {
952 IP_VS_ERR_RL("%s(): no memory\n", __func__);
953 return NULL;
954 }
955
956 INIT_HLIST_NODE(h: &cp->c_list);
957 timer_setup(&cp->timer, ip_vs_conn_expire, 0);
958 cp->ipvs = ipvs;
959 cp->af = p->af;
960 cp->daf = dest_af;
961 cp->protocol = p->protocol;
962 ip_vs_addr_set(af: p->af, dst: &cp->caddr, src: p->caddr);
963 cp->cport = p->cport;
964 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
965 ip_vs_addr_set(af: p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
966 dst: &cp->vaddr, src: p->vaddr);
967 cp->vport = p->vport;
968 ip_vs_addr_set(af: cp->daf, dst: &cp->daddr, src: daddr);
969 cp->dport = dport;
970 cp->flags = flags;
971 cp->fwmark = fwmark;
972 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
973 ip_vs_pe_get(p->pe);
974 cp->pe = p->pe;
975 cp->pe_data = p->pe_data;
976 cp->pe_data_len = p->pe_data_len;
977 } else {
978 cp->pe = NULL;
979 cp->pe_data = NULL;
980 cp->pe_data_len = 0;
981 }
982 spin_lock_init(&cp->lock);
983
984 /*
985 * Set the entry is referenced by the current thread before hashing
986 * it in the table, so that other thread run ip_vs_random_dropentry
987 * but cannot drop this entry.
988 */
989 refcount_set(r: &cp->refcnt, n: 1);
990
991 cp->control = NULL;
992 atomic_set(v: &cp->n_control, i: 0);
993 atomic_set(v: &cp->in_pkts, i: 0);
994
995 cp->packet_xmit = NULL;
996 cp->app = NULL;
997 cp->app_data = NULL;
998 /* reset struct ip_vs_seq */
999 cp->in_seq.delta = 0;
1000 cp->out_seq.delta = 0;
1001
1002 atomic_inc(v: &ipvs->conn_count);
1003 if (flags & IP_VS_CONN_F_NO_CPORT)
1004 atomic_inc(v: &ip_vs_conn_no_cport_cnt);
1005
1006 /* Bind the connection with a destination server */
1007 cp->dest = NULL;
1008 ip_vs_bind_dest(cp, dest);
1009
1010 /* Set its state and timeout */
1011 cp->state = 0;
1012 cp->old_state = 0;
1013 cp->timeout = 3*HZ;
1014 cp->sync_endtime = jiffies & ~3UL;
1015
1016 /* Bind its packet transmitter */
1017#ifdef CONFIG_IP_VS_IPV6
1018 if (p->af == AF_INET6)
1019 ip_vs_bind_xmit_v6(cp);
1020 else
1021#endif
1022 ip_vs_bind_xmit(cp);
1023
1024 if (unlikely(pd && atomic_read(&pd->appcnt)))
1025 ip_vs_bind_app(cp, pp: pd->pp);
1026
1027 /*
1028 * Allow conntrack to be preserved. By default, conntrack
1029 * is created and destroyed for every packet.
1030 * Sometimes keeping conntrack can be useful for
1031 * IP_VS_CONN_F_ONE_PACKET too.
1032 */
1033
1034 if (ip_vs_conntrack_enabled(ipvs))
1035 cp->flags |= IP_VS_CONN_F_NFCT;
1036
1037 /* Hash it in the ip_vs_conn_tab finally */
1038 ip_vs_conn_hash(cp);
1039
1040 return cp;
1041}
1042
1043/*
1044 * /proc/net/ip_vs_conn entries
1045 */
1046#ifdef CONFIG_PROC_FS
1047struct ip_vs_iter_state {
1048 struct seq_net_private p;
1049 struct hlist_head *l;
1050};
1051
1052static void *ip_vs_conn_array(struct seq_file *seq, loff_t pos)
1053{
1054 int idx;
1055 struct ip_vs_conn *cp;
1056 struct ip_vs_iter_state *iter = seq->private;
1057
1058 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1059 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1060 /* __ip_vs_conn_get() is not needed by
1061 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1062 */
1063 if (pos-- == 0) {
1064 iter->l = &ip_vs_conn_tab[idx];
1065 return cp;
1066 }
1067 }
1068 cond_resched_rcu();
1069 }
1070
1071 return NULL;
1072}
1073
1074static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1075 __acquires(RCU)
1076{
1077 struct ip_vs_iter_state *iter = seq->private;
1078
1079 iter->l = NULL;
1080 rcu_read_lock();
1081 return *pos ? ip_vs_conn_array(seq, pos: *pos - 1) :SEQ_START_TOKEN;
1082}
1083
1084static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1085{
1086 struct ip_vs_conn *cp = v;
1087 struct ip_vs_iter_state *iter = seq->private;
1088 struct hlist_node *e;
1089 struct hlist_head *l = iter->l;
1090 int idx;
1091
1092 ++*pos;
1093 if (v == SEQ_START_TOKEN)
1094 return ip_vs_conn_array(seq, pos: 0);
1095
1096 /* more on same hash chain? */
1097 e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1098 if (e)
1099 return hlist_entry(e, struct ip_vs_conn, c_list);
1100
1101 idx = l - ip_vs_conn_tab;
1102 while (++idx < ip_vs_conn_tab_size) {
1103 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1104 iter->l = &ip_vs_conn_tab[idx];
1105 return cp;
1106 }
1107 cond_resched_rcu();
1108 }
1109 iter->l = NULL;
1110 return NULL;
1111}
1112
1113static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1114 __releases(RCU)
1115{
1116 rcu_read_unlock();
1117}
1118
1119static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1120{
1121
1122 if (v == SEQ_START_TOKEN)
1123 seq_puts(m: seq,
1124 s: "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1125 else {
1126 const struct ip_vs_conn *cp = v;
1127 struct net *net = seq_file_net(seq);
1128 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1129 size_t len = 0;
1130 char dbuf[IP_VS_ADDRSTRLEN];
1131
1132 if (!net_eq(net1: cp->ipvs->net, net2: net))
1133 return 0;
1134 if (cp->pe_data) {
1135 pe_data[0] = ' ';
1136 len = strlen(cp->pe->name);
1137 memcpy(pe_data + 1, cp->pe->name, len);
1138 pe_data[len + 1] = ' ';
1139 len += 2;
1140 len += cp->pe->show_pe_data(cp, pe_data + len);
1141 }
1142 pe_data[len] = '\0';
1143
1144#ifdef CONFIG_IP_VS_IPV6
1145 if (cp->daf == AF_INET6)
1146 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%pI6", &cp->daddr.in6);
1147 else
1148#endif
1149 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%08X",
1150 ntohl(cp->daddr.ip));
1151
1152#ifdef CONFIG_IP_VS_IPV6
1153 if (cp->af == AF_INET6)
1154 seq_printf(m: seq, fmt: "%-3s %pI6 %04X %pI6 %04X "
1155 "%s %04X %-11s %7u%s\n",
1156 ip_vs_proto_name(proto: cp->protocol),
1157 &cp->caddr.in6, ntohs(cp->cport),
1158 &cp->vaddr.in6, ntohs(cp->vport),
1159 dbuf, ntohs(cp->dport),
1160 ip_vs_state_name(cp),
1161 jiffies_delta_to_msecs(delta: cp->timer.expires -
1162 jiffies) / 1000,
1163 pe_data);
1164 else
1165#endif
1166 seq_printf(m: seq,
1167 fmt: "%-3s %08X %04X %08X %04X"
1168 " %s %04X %-11s %7u%s\n",
1169 ip_vs_proto_name(proto: cp->protocol),
1170 ntohl(cp->caddr.ip), ntohs(cp->cport),
1171 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1172 dbuf, ntohs(cp->dport),
1173 ip_vs_state_name(cp),
1174 jiffies_delta_to_msecs(delta: cp->timer.expires -
1175 jiffies) / 1000,
1176 pe_data);
1177 }
1178 return 0;
1179}
1180
1181static const struct seq_operations ip_vs_conn_seq_ops = {
1182 .start = ip_vs_conn_seq_start,
1183 .next = ip_vs_conn_seq_next,
1184 .stop = ip_vs_conn_seq_stop,
1185 .show = ip_vs_conn_seq_show,
1186};
1187
1188static const char *ip_vs_origin_name(unsigned int flags)
1189{
1190 if (flags & IP_VS_CONN_F_SYNC)
1191 return "SYNC";
1192 else
1193 return "LOCAL";
1194}
1195
1196static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1197{
1198 char dbuf[IP_VS_ADDRSTRLEN];
1199
1200 if (v == SEQ_START_TOKEN)
1201 seq_puts(m: seq,
1202 s: "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1203 else {
1204 const struct ip_vs_conn *cp = v;
1205 struct net *net = seq_file_net(seq);
1206
1207 if (!net_eq(net1: cp->ipvs->net, net2: net))
1208 return 0;
1209
1210#ifdef CONFIG_IP_VS_IPV6
1211 if (cp->daf == AF_INET6)
1212 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%pI6", &cp->daddr.in6);
1213 else
1214#endif
1215 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%08X",
1216 ntohl(cp->daddr.ip));
1217
1218#ifdef CONFIG_IP_VS_IPV6
1219 if (cp->af == AF_INET6)
1220 seq_printf(m: seq, fmt: "%-3s %pI6 %04X %pI6 %04X "
1221 "%s %04X %-11s %-6s %7u\n",
1222 ip_vs_proto_name(proto: cp->protocol),
1223 &cp->caddr.in6, ntohs(cp->cport),
1224 &cp->vaddr.in6, ntohs(cp->vport),
1225 dbuf, ntohs(cp->dport),
1226 ip_vs_state_name(cp),
1227 ip_vs_origin_name(flags: cp->flags),
1228 jiffies_delta_to_msecs(delta: cp->timer.expires -
1229 jiffies) / 1000);
1230 else
1231#endif
1232 seq_printf(m: seq,
1233 fmt: "%-3s %08X %04X %08X %04X "
1234 "%s %04X %-11s %-6s %7u\n",
1235 ip_vs_proto_name(proto: cp->protocol),
1236 ntohl(cp->caddr.ip), ntohs(cp->cport),
1237 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1238 dbuf, ntohs(cp->dport),
1239 ip_vs_state_name(cp),
1240 ip_vs_origin_name(flags: cp->flags),
1241 jiffies_delta_to_msecs(delta: cp->timer.expires -
1242 jiffies) / 1000);
1243 }
1244 return 0;
1245}
1246
1247static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1248 .start = ip_vs_conn_seq_start,
1249 .next = ip_vs_conn_seq_next,
1250 .stop = ip_vs_conn_seq_stop,
1251 .show = ip_vs_conn_sync_seq_show,
1252};
1253#endif
1254
1255
1256/* Randomly drop connection entries before running out of memory
1257 * Can be used for DATA and CTL conns. For TPL conns there are exceptions:
1258 * - traffic for services in OPS mode increases ct->in_pkts, so it is supported
1259 * - traffic for services not in OPS mode does not increase ct->in_pkts in
1260 * all cases, so it is not supported
1261 */
1262static inline int todrop_entry(struct ip_vs_conn *cp)
1263{
1264 /*
1265 * The drop rate array needs tuning for real environments.
1266 * Called from timer bh only => no locking
1267 */
1268 static const signed char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1269 static signed char todrop_counter[9] = {0};
1270 int i;
1271
1272 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1273 This will leave enough time for normal connection to get
1274 through. */
1275 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1276 return 0;
1277
1278 /* Don't drop the entry if its number of incoming packets is not
1279 located in [0, 8] */
1280 i = atomic_read(v: &cp->in_pkts);
1281 if (i > 8 || i < 0) return 0;
1282
1283 if (!todrop_rate[i]) return 0;
1284 if (--todrop_counter[i] > 0) return 0;
1285
1286 todrop_counter[i] = todrop_rate[i];
1287 return 1;
1288}
1289
1290static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn *cp)
1291{
1292 struct ip_vs_service *svc;
1293
1294 if (!cp->dest)
1295 return false;
1296 svc = rcu_dereference(cp->dest->svc);
1297 return svc && (svc->flags & IP_VS_SVC_F_ONEPACKET);
1298}
1299
1300/* Called from keventd and must protect itself from softirqs */
1301void ip_vs_random_dropentry(struct netns_ipvs *ipvs)
1302{
1303 int idx;
1304 struct ip_vs_conn *cp;
1305
1306 rcu_read_lock();
1307 /*
1308 * Randomly scan 1/32 of the whole table every second
1309 */
1310 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1311 unsigned int hash = get_random_u32() & ip_vs_conn_tab_mask;
1312
1313 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1314 if (cp->ipvs != ipvs)
1315 continue;
1316 if (atomic_read(v: &cp->n_control))
1317 continue;
1318 if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
1319 /* connection template of OPS */
1320 if (ip_vs_conn_ops_mode(cp))
1321 goto try_drop;
1322 if (!(cp->state & IP_VS_CTPL_S_ASSURED))
1323 goto drop;
1324 continue;
1325 }
1326 if (cp->protocol == IPPROTO_TCP) {
1327 switch(cp->state) {
1328 case IP_VS_TCP_S_SYN_RECV:
1329 case IP_VS_TCP_S_SYNACK:
1330 break;
1331
1332 case IP_VS_TCP_S_ESTABLISHED:
1333 if (todrop_entry(cp))
1334 break;
1335 continue;
1336
1337 default:
1338 continue;
1339 }
1340 } else if (cp->protocol == IPPROTO_SCTP) {
1341 switch (cp->state) {
1342 case IP_VS_SCTP_S_INIT1:
1343 case IP_VS_SCTP_S_INIT:
1344 break;
1345 case IP_VS_SCTP_S_ESTABLISHED:
1346 if (todrop_entry(cp))
1347 break;
1348 continue;
1349 default:
1350 continue;
1351 }
1352 } else {
1353try_drop:
1354 if (!todrop_entry(cp))
1355 continue;
1356 }
1357
1358drop:
1359 IP_VS_DBG(4, "drop connection\n");
1360 ip_vs_conn_del(cp);
1361 }
1362 cond_resched_rcu();
1363 }
1364 rcu_read_unlock();
1365}
1366
1367
1368/*
1369 * Flush all the connection entries in the ip_vs_conn_tab
1370 */
1371static void ip_vs_conn_flush(struct netns_ipvs *ipvs)
1372{
1373 int idx;
1374 struct ip_vs_conn *cp, *cp_c;
1375
1376flush_again:
1377 rcu_read_lock();
1378 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1379
1380 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1381 if (cp->ipvs != ipvs)
1382 continue;
1383 if (atomic_read(v: &cp->n_control))
1384 continue;
1385 cp_c = cp->control;
1386 IP_VS_DBG(4, "del connection\n");
1387 ip_vs_conn_del(cp);
1388 if (cp_c && !atomic_read(v: &cp_c->n_control)) {
1389 IP_VS_DBG(4, "del controlling connection\n");
1390 ip_vs_conn_del(cp: cp_c);
1391 }
1392 }
1393 cond_resched_rcu();
1394 }
1395 rcu_read_unlock();
1396
1397 /* the counter may be not NULL, because maybe some conn entries
1398 are run by slow timer handler or unhashed but still referred */
1399 if (atomic_read(v: &ipvs->conn_count) != 0) {
1400 schedule();
1401 goto flush_again;
1402 }
1403}
1404
1405#ifdef CONFIG_SYSCTL
1406void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
1407{
1408 int idx;
1409 struct ip_vs_conn *cp, *cp_c;
1410 struct ip_vs_dest *dest;
1411
1412 rcu_read_lock();
1413 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1414 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1415 if (cp->ipvs != ipvs)
1416 continue;
1417
1418 dest = cp->dest;
1419 if (!dest || (dest->flags & IP_VS_DEST_F_AVAILABLE))
1420 continue;
1421
1422 if (atomic_read(v: &cp->n_control))
1423 continue;
1424
1425 cp_c = cp->control;
1426 IP_VS_DBG(4, "del connection\n");
1427 ip_vs_conn_del(cp);
1428 if (cp_c && !atomic_read(v: &cp_c->n_control)) {
1429 IP_VS_DBG(4, "del controlling connection\n");
1430 ip_vs_conn_del(cp: cp_c);
1431 }
1432 }
1433 cond_resched_rcu();
1434
1435 /* netns clean up started, abort delayed work */
1436 if (!ipvs->enable)
1437 break;
1438 }
1439 rcu_read_unlock();
1440}
1441#endif
1442
1443/*
1444 * per netns init and exit
1445 */
1446int __net_init ip_vs_conn_net_init(struct netns_ipvs *ipvs)
1447{
1448 atomic_set(v: &ipvs->conn_count, i: 0);
1449
1450#ifdef CONFIG_PROC_FS
1451 if (!proc_create_net("ip_vs_conn", 0, ipvs->net->proc_net,
1452 &ip_vs_conn_seq_ops,
1453 sizeof(struct ip_vs_iter_state)))
1454 goto err_conn;
1455
1456 if (!proc_create_net("ip_vs_conn_sync", 0, ipvs->net->proc_net,
1457 &ip_vs_conn_sync_seq_ops,
1458 sizeof(struct ip_vs_iter_state)))
1459 goto err_conn_sync;
1460#endif
1461
1462 return 0;
1463
1464#ifdef CONFIG_PROC_FS
1465err_conn_sync:
1466 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1467err_conn:
1468 return -ENOMEM;
1469#endif
1470}
1471
1472void __net_exit ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs)
1473{
1474 /* flush all the connection entries first */
1475 ip_vs_conn_flush(ipvs);
1476#ifdef CONFIG_PROC_FS
1477 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1478 remove_proc_entry("ip_vs_conn_sync", ipvs->net->proc_net);
1479#endif
1480}
1481
1482int __init ip_vs_conn_init(void)
1483{
1484 size_t tab_array_size;
1485 int max_avail;
1486#if BITS_PER_LONG > 32
1487 int max = 27;
1488#else
1489 int max = 20;
1490#endif
1491 int min = 8;
1492 int idx;
1493
1494 max_avail = order_base_2(totalram_pages()) + PAGE_SHIFT;
1495 max_avail -= 2; /* ~4 in hash row */
1496 max_avail -= 1; /* IPVS up to 1/2 of mem */
1497 max_avail -= order_base_2(sizeof(struct ip_vs_conn));
1498 max = clamp(max, min, max_avail);
1499 ip_vs_conn_tab_bits = clamp_val(ip_vs_conn_tab_bits, min, max);
1500 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1501 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1502
1503 /*
1504 * Allocate the connection hash table and initialize its list heads
1505 */
1506 tab_array_size = array_size(ip_vs_conn_tab_size,
1507 sizeof(*ip_vs_conn_tab));
1508 ip_vs_conn_tab = kvmalloc_array(n: ip_vs_conn_tab_size,
1509 size: sizeof(*ip_vs_conn_tab), GFP_KERNEL);
1510 if (!ip_vs_conn_tab)
1511 return -ENOMEM;
1512
1513 /* Allocate ip_vs_conn slab cache */
1514 ip_vs_conn_cachep = KMEM_CACHE(ip_vs_conn, SLAB_HWCACHE_ALIGN);
1515 if (!ip_vs_conn_cachep) {
1516 kvfree(addr: ip_vs_conn_tab);
1517 return -ENOMEM;
1518 }
1519
1520 pr_info("Connection hash table configured (size=%d, memory=%zdKbytes)\n",
1521 ip_vs_conn_tab_size, tab_array_size / 1024);
1522 IP_VS_DBG(0, "Each connection entry needs %zd bytes at least\n",
1523 sizeof(struct ip_vs_conn));
1524
1525 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1526 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1527
1528 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1529 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1530 }
1531
1532 /* calculate the random value for connection hash */
1533 get_random_bytes(buf: &ip_vs_conn_rnd, len: sizeof(ip_vs_conn_rnd));
1534
1535 return 0;
1536}
1537
1538void ip_vs_conn_cleanup(void)
1539{
1540 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1541 rcu_barrier();
1542 /* Release the empty cache */
1543 kmem_cache_destroy(s: ip_vs_conn_cachep);
1544 kvfree(addr: ip_vs_conn_tab);
1545}
1546

source code of linux/net/netfilter/ipvs/ip_vs_conn.c