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 pr_fmt(fmt) "IPVS: " fmt
21
22#include <linux/interrupt.h>
23#include <linux/in.h>
24#include <linux/inet.h>
25#include <linux/net.h>
26#include <linux/kernel.h>
27#include <linux/module.h>
28#include <linux/proc_fs.h> /* for proc_net_* */
29#include <linux/slab.h>
30#include <linux/seq_file.h>
31#include <linux/jhash.h>
32#include <linux/random.h>
33#include <linux/rcupdate_wait.h>
34
35#include <net/net_namespace.h>
36#include <net/ip_vs.h>
37
38
39#ifndef CONFIG_IP_VS_TAB_BITS
40#define CONFIG_IP_VS_TAB_BITS 12
41#endif
42
43/*
44 * Connection hash size. Default is what was selected at compile time.
45*/
46static int ip_vs_conn_tab_bits = CONFIG_IP_VS_TAB_BITS;
47module_param_named(conn_tab_bits, ip_vs_conn_tab_bits, int, 0444);
48MODULE_PARM_DESC(conn_tab_bits, "Set connections' hash size");
49
50/* size and mask values */
51int ip_vs_conn_tab_size __read_mostly;
52static int ip_vs_conn_tab_mask __read_mostly;
53
54/*
55 * Connection hash table: for input and output packets lookups of IPVS
56 */
57static struct hlist_head *ip_vs_conn_tab __read_mostly;
58
59/* SLAB cache for IPVS connections */
60static struct kmem_cache *ip_vs_conn_cachep __read_mostly;
61
62/* counter for no client port connections */
63static atomic_t ip_vs_conn_no_cport_cnt = ATOMIC_INIT(0);
64
65/* random value for IPVS connection hash */
66static unsigned int ip_vs_conn_rnd __read_mostly;
67
68/*
69 * Fine locking granularity for big connection hash table
70 */
71#define CT_LOCKARRAY_BITS 5
72#define CT_LOCKARRAY_SIZE (1<<CT_LOCKARRAY_BITS)
73#define CT_LOCKARRAY_MASK (CT_LOCKARRAY_SIZE-1)
74
75/* We need an addrstrlen that works with or without v6 */
76#ifdef CONFIG_IP_VS_IPV6
77#define IP_VS_ADDRSTRLEN INET6_ADDRSTRLEN
78#else
79#define IP_VS_ADDRSTRLEN (8+1)
80#endif
81
82struct ip_vs_aligned_lock
83{
84 spinlock_t l;
85} __attribute__((__aligned__(SMP_CACHE_BYTES)));
86
87/* lock array for conn table */
88static struct ip_vs_aligned_lock
89__ip_vs_conntbl_lock_array[CT_LOCKARRAY_SIZE] __cacheline_aligned;
90
91static inline void ct_write_lock_bh(unsigned int key)
92{
93 spin_lock_bh(lock: &__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
94}
95
96static inline void ct_write_unlock_bh(unsigned int key)
97{
98 spin_unlock_bh(lock: &__ip_vs_conntbl_lock_array[key&CT_LOCKARRAY_MASK].l);
99}
100
101static void ip_vs_conn_expire(struct timer_list *t);
102
103/*
104 * Returns hash value for IPVS connection entry
105 */
106static unsigned int ip_vs_conn_hashkey(struct netns_ipvs *ipvs, int af, unsigned int proto,
107 const union nf_inet_addr *addr,
108 __be16 port)
109{
110#ifdef CONFIG_IP_VS_IPV6
111 if (af == AF_INET6)
112 return (jhash_3words(a: jhash(key: addr, length: 16, initval: ip_vs_conn_rnd),
113 b: (__force u32)port, c: proto, initval: ip_vs_conn_rnd) ^
114 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
115#endif
116 return (jhash_3words(a: (__force u32)addr->ip, b: (__force u32)port, c: proto,
117 initval: ip_vs_conn_rnd) ^
118 ((size_t)ipvs>>8)) & ip_vs_conn_tab_mask;
119}
120
121static unsigned int ip_vs_conn_hashkey_param(const struct ip_vs_conn_param *p,
122 bool inverse)
123{
124 const union nf_inet_addr *addr;
125 __be16 port;
126
127 if (p->pe_data && p->pe->hashkey_raw)
128 return p->pe->hashkey_raw(p, ip_vs_conn_rnd, inverse) &
129 ip_vs_conn_tab_mask;
130
131 if (likely(!inverse)) {
132 addr = p->caddr;
133 port = p->cport;
134 } else {
135 addr = p->vaddr;
136 port = p->vport;
137 }
138
139 return ip_vs_conn_hashkey(ipvs: p->ipvs, af: p->af, proto: p->protocol, addr, port);
140}
141
142static unsigned int ip_vs_conn_hashkey_conn(const struct ip_vs_conn *cp)
143{
144 struct ip_vs_conn_param p;
145
146 ip_vs_conn_fill_param(ipvs: cp->ipvs, af: cp->af, protocol: cp->protocol,
147 caddr: &cp->caddr, cport: cp->cport, NULL, vport: 0, p: &p);
148
149 if (cp->pe) {
150 p.pe = cp->pe;
151 p.pe_data = cp->pe_data;
152 p.pe_data_len = cp->pe_data_len;
153 }
154
155 return ip_vs_conn_hashkey_param(p: &p, inverse: false);
156}
157
158/*
159 * Hashes ip_vs_conn in ip_vs_conn_tab by netns,proto,addr,port.
160 * returns bool success.
161 */
162static inline int ip_vs_conn_hash(struct ip_vs_conn *cp)
163{
164 unsigned int hash;
165 int ret;
166
167 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
168 return 0;
169
170 /* Hash by protocol, client address and port */
171 hash = ip_vs_conn_hashkey_conn(cp);
172
173 ct_write_lock_bh(key: hash);
174 spin_lock(lock: &cp->lock);
175
176 if (!(cp->flags & IP_VS_CONN_F_HASHED)) {
177 cp->flags |= IP_VS_CONN_F_HASHED;
178 refcount_inc(r: &cp->refcnt);
179 hlist_add_head_rcu(n: &cp->c_list, h: &ip_vs_conn_tab[hash]);
180 ret = 1;
181 } else {
182 pr_err("%s(): request for already hashed, called from %pS\n",
183 __func__, __builtin_return_address(0));
184 ret = 0;
185 }
186
187 spin_unlock(lock: &cp->lock);
188 ct_write_unlock_bh(key: hash);
189
190 return ret;
191}
192
193
194/*
195 * UNhashes ip_vs_conn from ip_vs_conn_tab.
196 * returns bool success. Caller should hold conn reference.
197 */
198static inline int ip_vs_conn_unhash(struct ip_vs_conn *cp)
199{
200 unsigned int hash;
201 int ret;
202
203 /* unhash it and decrease its reference counter */
204 hash = ip_vs_conn_hashkey_conn(cp);
205
206 ct_write_lock_bh(key: hash);
207 spin_lock(lock: &cp->lock);
208
209 if (cp->flags & IP_VS_CONN_F_HASHED) {
210 hlist_del_rcu(n: &cp->c_list);
211 cp->flags &= ~IP_VS_CONN_F_HASHED;
212 refcount_dec(r: &cp->refcnt);
213 ret = 1;
214 } else
215 ret = 0;
216
217 spin_unlock(lock: &cp->lock);
218 ct_write_unlock_bh(key: hash);
219
220 return ret;
221}
222
223/* Try to unlink ip_vs_conn from ip_vs_conn_tab.
224 * returns bool success.
225 */
226static inline bool ip_vs_conn_unlink(struct ip_vs_conn *cp)
227{
228 unsigned int hash;
229 bool ret = false;
230
231 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
232 return refcount_dec_if_one(r: &cp->refcnt);
233
234 hash = ip_vs_conn_hashkey_conn(cp);
235
236 ct_write_lock_bh(key: hash);
237 spin_lock(lock: &cp->lock);
238
239 if (cp->flags & IP_VS_CONN_F_HASHED) {
240 /* Decrease refcnt and unlink conn only if we are last user */
241 if (refcount_dec_if_one(r: &cp->refcnt)) {
242 hlist_del_rcu(n: &cp->c_list);
243 cp->flags &= ~IP_VS_CONN_F_HASHED;
244 ret = true;
245 }
246 }
247
248 spin_unlock(lock: &cp->lock);
249 ct_write_unlock_bh(key: hash);
250
251 return ret;
252}
253
254
255/*
256 * Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
257 * Called for pkts coming from OUTside-to-INside.
258 * p->caddr, p->cport: pkt source address (foreign host)
259 * p->vaddr, p->vport: pkt dest address (load balancer)
260 */
261static inline struct ip_vs_conn *
262__ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
263{
264 unsigned int hash;
265 struct ip_vs_conn *cp;
266
267 hash = ip_vs_conn_hashkey_param(p, inverse: false);
268
269 rcu_read_lock();
270
271 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
272 if (p->cport == cp->cport && p->vport == cp->vport &&
273 cp->af == p->af &&
274 ip_vs_addr_equal(af: p->af, a: p->caddr, b: &cp->caddr) &&
275 ip_vs_addr_equal(af: p->af, a: p->vaddr, b: &cp->vaddr) &&
276 ((!p->cport) ^ (!(cp->flags & IP_VS_CONN_F_NO_CPORT))) &&
277 p->protocol == cp->protocol &&
278 cp->ipvs == p->ipvs) {
279 if (!__ip_vs_conn_get(cp))
280 continue;
281 /* HIT */
282 rcu_read_unlock();
283 return cp;
284 }
285 }
286
287 rcu_read_unlock();
288
289 return NULL;
290}
291
292struct ip_vs_conn *ip_vs_conn_in_get(const struct ip_vs_conn_param *p)
293{
294 struct ip_vs_conn *cp;
295
296 cp = __ip_vs_conn_in_get(p);
297 if (!cp && atomic_read(v: &ip_vs_conn_no_cport_cnt)) {
298 struct ip_vs_conn_param cport_zero_p = *p;
299 cport_zero_p.cport = 0;
300 cp = __ip_vs_conn_in_get(p: &cport_zero_p);
301 }
302
303 IP_VS_DBG_BUF(9, "lookup/in %s %s:%d->%s:%d %s\n",
304 ip_vs_proto_name(p->protocol),
305 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
306 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
307 cp ? "hit" : "not hit");
308
309 return cp;
310}
311
312static int
313ip_vs_conn_fill_param_proto(struct netns_ipvs *ipvs,
314 int af, const struct sk_buff *skb,
315 const struct ip_vs_iphdr *iph,
316 struct ip_vs_conn_param *p)
317{
318 __be16 _ports[2], *pptr;
319
320 pptr = frag_safe_skb_hp(skb, offset: iph->len, len: sizeof(_ports), buffer: _ports);
321 if (pptr == NULL)
322 return 1;
323
324 if (likely(!ip_vs_iph_inverse(iph)))
325 ip_vs_conn_fill_param(ipvs, af, protocol: iph->protocol, caddr: &iph->saddr,
326 cport: pptr[0], vaddr: &iph->daddr, vport: pptr[1], p);
327 else
328 ip_vs_conn_fill_param(ipvs, af, protocol: iph->protocol, caddr: &iph->daddr,
329 cport: pptr[1], vaddr: &iph->saddr, vport: pptr[0], p);
330 return 0;
331}
332
333struct ip_vs_conn *
334ip_vs_conn_in_get_proto(struct netns_ipvs *ipvs, int af,
335 const struct sk_buff *skb,
336 const struct ip_vs_iphdr *iph)
337{
338 struct ip_vs_conn_param p;
339
340 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, p: &p))
341 return NULL;
342
343 return ip_vs_conn_in_get(p: &p);
344}
345EXPORT_SYMBOL_GPL(ip_vs_conn_in_get_proto);
346
347/* Get reference to connection template */
348struct ip_vs_conn *ip_vs_ct_in_get(const struct ip_vs_conn_param *p)
349{
350 unsigned int hash;
351 struct ip_vs_conn *cp;
352
353 hash = ip_vs_conn_hashkey_param(p, inverse: false);
354
355 rcu_read_lock();
356
357 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
358 if (unlikely(p->pe_data && p->pe->ct_match)) {
359 if (cp->ipvs != p->ipvs)
360 continue;
361 if (p->pe == cp->pe && p->pe->ct_match(p, cp)) {
362 if (__ip_vs_conn_get(cp))
363 goto out;
364 }
365 continue;
366 }
367
368 if (cp->af == p->af &&
369 ip_vs_addr_equal(af: p->af, a: p->caddr, b: &cp->caddr) &&
370 /* protocol should only be IPPROTO_IP if
371 * p->vaddr is a fwmark */
372 ip_vs_addr_equal(af: p->protocol == IPPROTO_IP ? AF_UNSPEC :
373 p->af, a: p->vaddr, b: &cp->vaddr) &&
374 p->vport == cp->vport && p->cport == cp->cport &&
375 cp->flags & IP_VS_CONN_F_TEMPLATE &&
376 p->protocol == cp->protocol &&
377 cp->ipvs == p->ipvs) {
378 if (__ip_vs_conn_get(cp))
379 goto out;
380 }
381 }
382 cp = NULL;
383
384 out:
385 rcu_read_unlock();
386
387 IP_VS_DBG_BUF(9, "template lookup/in %s %s:%d->%s:%d %s\n",
388 ip_vs_proto_name(p->protocol),
389 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
390 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
391 cp ? "hit" : "not hit");
392
393 return cp;
394}
395
396/* Gets ip_vs_conn associated with supplied parameters in the ip_vs_conn_tab.
397 * Called for pkts coming from inside-to-OUTside.
398 * p->caddr, p->cport: pkt source address (inside host)
399 * p->vaddr, p->vport: pkt dest address (foreign host) */
400struct ip_vs_conn *ip_vs_conn_out_get(const struct ip_vs_conn_param *p)
401{
402 unsigned int hash;
403 struct ip_vs_conn *cp, *ret=NULL;
404 const union nf_inet_addr *saddr;
405 __be16 sport;
406
407 /*
408 * Check for "full" addressed entries
409 */
410 hash = ip_vs_conn_hashkey_param(p, inverse: true);
411
412 rcu_read_lock();
413
414 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
415 if (p->vport != cp->cport)
416 continue;
417
418 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
419 sport = cp->vport;
420 saddr = &cp->vaddr;
421 } else {
422 sport = cp->dport;
423 saddr = &cp->daddr;
424 }
425
426 if (p->cport == sport && cp->af == p->af &&
427 ip_vs_addr_equal(af: p->af, a: p->vaddr, b: &cp->caddr) &&
428 ip_vs_addr_equal(af: p->af, a: p->caddr, b: saddr) &&
429 p->protocol == cp->protocol &&
430 cp->ipvs == p->ipvs) {
431 if (!__ip_vs_conn_get(cp))
432 continue;
433 /* HIT */
434 ret = cp;
435 break;
436 }
437 }
438
439 rcu_read_unlock();
440
441 IP_VS_DBG_BUF(9, "lookup/out %s %s:%d->%s:%d %s\n",
442 ip_vs_proto_name(p->protocol),
443 IP_VS_DBG_ADDR(p->af, p->caddr), ntohs(p->cport),
444 IP_VS_DBG_ADDR(p->af, p->vaddr), ntohs(p->vport),
445 ret ? "hit" : "not hit");
446
447 return ret;
448}
449
450struct ip_vs_conn *
451ip_vs_conn_out_get_proto(struct netns_ipvs *ipvs, int af,
452 const struct sk_buff *skb,
453 const struct ip_vs_iphdr *iph)
454{
455 struct ip_vs_conn_param p;
456
457 if (ip_vs_conn_fill_param_proto(ipvs, af, skb, iph, p: &p))
458 return NULL;
459
460 return ip_vs_conn_out_get(p: &p);
461}
462EXPORT_SYMBOL_GPL(ip_vs_conn_out_get_proto);
463
464/*
465 * Put back the conn and restart its timer with its timeout
466 */
467static void __ip_vs_conn_put_timer(struct ip_vs_conn *cp)
468{
469 unsigned long t = (cp->flags & IP_VS_CONN_F_ONE_PACKET) ?
470 0 : cp->timeout;
471 mod_timer(timer: &cp->timer, expires: jiffies+t);
472
473 __ip_vs_conn_put(cp);
474}
475
476void ip_vs_conn_put(struct ip_vs_conn *cp)
477{
478 if ((cp->flags & IP_VS_CONN_F_ONE_PACKET) &&
479 (refcount_read(r: &cp->refcnt) == 1) &&
480 !timer_pending(timer: &cp->timer))
481 /* expire connection immediately */
482 ip_vs_conn_expire(t: &cp->timer);
483 else
484 __ip_vs_conn_put_timer(cp);
485}
486
487/*
488 * Fill a no_client_port connection with a client port number
489 */
490void ip_vs_conn_fill_cport(struct ip_vs_conn *cp, __be16 cport)
491{
492 if (ip_vs_conn_unhash(cp)) {
493 spin_lock_bh(lock: &cp->lock);
494 if (cp->flags & IP_VS_CONN_F_NO_CPORT) {
495 atomic_dec(v: &ip_vs_conn_no_cport_cnt);
496 cp->flags &= ~IP_VS_CONN_F_NO_CPORT;
497 cp->cport = cport;
498 }
499 spin_unlock_bh(lock: &cp->lock);
500
501 /* hash on new dport */
502 ip_vs_conn_hash(cp);
503 }
504}
505
506
507/*
508 * Bind a connection entry with the corresponding packet_xmit.
509 * Called by ip_vs_conn_new.
510 */
511static inline void ip_vs_bind_xmit(struct ip_vs_conn *cp)
512{
513 switch (IP_VS_FWD_METHOD(cp)) {
514 case IP_VS_CONN_F_MASQ:
515 cp->packet_xmit = ip_vs_nat_xmit;
516 break;
517
518 case IP_VS_CONN_F_TUNNEL:
519#ifdef CONFIG_IP_VS_IPV6
520 if (cp->daf == AF_INET6)
521 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
522 else
523#endif
524 cp->packet_xmit = ip_vs_tunnel_xmit;
525 break;
526
527 case IP_VS_CONN_F_DROUTE:
528 cp->packet_xmit = ip_vs_dr_xmit;
529 break;
530
531 case IP_VS_CONN_F_LOCALNODE:
532 cp->packet_xmit = ip_vs_null_xmit;
533 break;
534
535 case IP_VS_CONN_F_BYPASS:
536 cp->packet_xmit = ip_vs_bypass_xmit;
537 break;
538 }
539}
540
541#ifdef CONFIG_IP_VS_IPV6
542static inline void ip_vs_bind_xmit_v6(struct ip_vs_conn *cp)
543{
544 switch (IP_VS_FWD_METHOD(cp)) {
545 case IP_VS_CONN_F_MASQ:
546 cp->packet_xmit = ip_vs_nat_xmit_v6;
547 break;
548
549 case IP_VS_CONN_F_TUNNEL:
550 if (cp->daf == AF_INET6)
551 cp->packet_xmit = ip_vs_tunnel_xmit_v6;
552 else
553 cp->packet_xmit = ip_vs_tunnel_xmit;
554 break;
555
556 case IP_VS_CONN_F_DROUTE:
557 cp->packet_xmit = ip_vs_dr_xmit_v6;
558 break;
559
560 case IP_VS_CONN_F_LOCALNODE:
561 cp->packet_xmit = ip_vs_null_xmit;
562 break;
563
564 case IP_VS_CONN_F_BYPASS:
565 cp->packet_xmit = ip_vs_bypass_xmit_v6;
566 break;
567 }
568}
569#endif
570
571
572static inline int ip_vs_dest_totalconns(struct ip_vs_dest *dest)
573{
574 return atomic_read(v: &dest->activeconns)
575 + atomic_read(v: &dest->inactconns);
576}
577
578/*
579 * Bind a connection entry with a virtual service destination
580 * Called just after a new connection entry is created.
581 */
582static inline void
583ip_vs_bind_dest(struct ip_vs_conn *cp, struct ip_vs_dest *dest)
584{
585 unsigned int conn_flags;
586 __u32 flags;
587
588 /* if dest is NULL, then return directly */
589 if (!dest)
590 return;
591
592 /* Increase the refcnt counter of the dest */
593 ip_vs_dest_hold(dest);
594
595 conn_flags = atomic_read(v: &dest->conn_flags);
596 if (cp->protocol != IPPROTO_UDP)
597 conn_flags &= ~IP_VS_CONN_F_ONE_PACKET;
598 flags = cp->flags;
599 /* Bind with the destination and its corresponding transmitter */
600 if (flags & IP_VS_CONN_F_SYNC) {
601 /* if the connection is not template and is created
602 * by sync, preserve the activity flag.
603 */
604 if (!(flags & IP_VS_CONN_F_TEMPLATE))
605 conn_flags &= ~IP_VS_CONN_F_INACTIVE;
606 /* connections inherit forwarding method from dest */
607 flags &= ~(IP_VS_CONN_F_FWD_MASK | IP_VS_CONN_F_NOOUTPUT);
608 }
609 flags |= conn_flags;
610 cp->flags = flags;
611 cp->dest = dest;
612
613 IP_VS_DBG_BUF(7, "Bind-dest %s c:%s:%d v:%s:%d "
614 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
615 "dest->refcnt:%d\n",
616 ip_vs_proto_name(cp->protocol),
617 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
618 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
619 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
620 ip_vs_fwd_tag(cp), cp->state,
621 cp->flags, refcount_read(&cp->refcnt),
622 refcount_read(&dest->refcnt));
623
624 /* Update the connection counters */
625 if (!(flags & IP_VS_CONN_F_TEMPLATE)) {
626 /* It is a normal connection, so modify the counters
627 * according to the flags, later the protocol can
628 * update them on state change
629 */
630 if (!(flags & IP_VS_CONN_F_INACTIVE))
631 atomic_inc(v: &dest->activeconns);
632 else
633 atomic_inc(v: &dest->inactconns);
634 } else {
635 /* It is a persistent connection/template, so increase
636 the persistent connection counter */
637 atomic_inc(v: &dest->persistconns);
638 }
639
640 if (dest->u_threshold != 0 &&
641 ip_vs_dest_totalconns(dest) >= dest->u_threshold)
642 dest->flags |= IP_VS_DEST_F_OVERLOAD;
643}
644
645
646/*
647 * Check if there is a destination for the connection, if so
648 * bind the connection to the destination.
649 */
650void ip_vs_try_bind_dest(struct ip_vs_conn *cp)
651{
652 struct ip_vs_dest *dest;
653
654 rcu_read_lock();
655
656 /* This function is only invoked by the synchronization code. We do
657 * not currently support heterogeneous pools with synchronization,
658 * so we can make the assumption that the svc_af is the same as the
659 * dest_af
660 */
661 dest = ip_vs_find_dest(ipvs: cp->ipvs, svc_af: cp->af, dest_af: cp->af, daddr: &cp->daddr,
662 dport: cp->dport, vaddr: &cp->vaddr, vport: cp->vport,
663 protocol: cp->protocol, fwmark: cp->fwmark, flags: cp->flags);
664 if (dest) {
665 struct ip_vs_proto_data *pd;
666
667 spin_lock_bh(lock: &cp->lock);
668 if (cp->dest) {
669 spin_unlock_bh(lock: &cp->lock);
670 rcu_read_unlock();
671 return;
672 }
673
674 /* Applications work depending on the forwarding method
675 * but better to reassign them always when binding dest */
676 if (cp->app)
677 ip_vs_unbind_app(cp);
678
679 ip_vs_bind_dest(cp, dest);
680 spin_unlock_bh(lock: &cp->lock);
681
682 /* Update its packet transmitter */
683 cp->packet_xmit = NULL;
684#ifdef CONFIG_IP_VS_IPV6
685 if (cp->af == AF_INET6)
686 ip_vs_bind_xmit_v6(cp);
687 else
688#endif
689 ip_vs_bind_xmit(cp);
690
691 pd = ip_vs_proto_data_get(ipvs: cp->ipvs, proto: cp->protocol);
692 if (pd && atomic_read(v: &pd->appcnt))
693 ip_vs_bind_app(cp, pp: pd->pp);
694 }
695 rcu_read_unlock();
696}
697
698
699/*
700 * Unbind a connection entry with its VS destination
701 * Called by the ip_vs_conn_expire function.
702 */
703static inline void ip_vs_unbind_dest(struct ip_vs_conn *cp)
704{
705 struct ip_vs_dest *dest = cp->dest;
706
707 if (!dest)
708 return;
709
710 IP_VS_DBG_BUF(7, "Unbind-dest %s c:%s:%d v:%s:%d "
711 "d:%s:%d fwd:%c s:%u conn->flags:%X conn->refcnt:%d "
712 "dest->refcnt:%d\n",
713 ip_vs_proto_name(cp->protocol),
714 IP_VS_DBG_ADDR(cp->af, &cp->caddr), ntohs(cp->cport),
715 IP_VS_DBG_ADDR(cp->af, &cp->vaddr), ntohs(cp->vport),
716 IP_VS_DBG_ADDR(cp->daf, &cp->daddr), ntohs(cp->dport),
717 ip_vs_fwd_tag(cp), cp->state,
718 cp->flags, refcount_read(&cp->refcnt),
719 refcount_read(&dest->refcnt));
720
721 /* Update the connection counters */
722 if (!(cp->flags & IP_VS_CONN_F_TEMPLATE)) {
723 /* It is a normal connection, so decrease the inactconns
724 or activeconns counter */
725 if (cp->flags & IP_VS_CONN_F_INACTIVE) {
726 atomic_dec(v: &dest->inactconns);
727 } else {
728 atomic_dec(v: &dest->activeconns);
729 }
730 } else {
731 /* It is a persistent connection/template, so decrease
732 the persistent connection counter */
733 atomic_dec(v: &dest->persistconns);
734 }
735
736 if (dest->l_threshold != 0) {
737 if (ip_vs_dest_totalconns(dest) < dest->l_threshold)
738 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
739 } else if (dest->u_threshold != 0) {
740 if (ip_vs_dest_totalconns(dest) * 4 < dest->u_threshold * 3)
741 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
742 } else {
743 if (dest->flags & IP_VS_DEST_F_OVERLOAD)
744 dest->flags &= ~IP_VS_DEST_F_OVERLOAD;
745 }
746
747 ip_vs_dest_put(dest);
748}
749
750static int expire_quiescent_template(struct netns_ipvs *ipvs,
751 struct ip_vs_dest *dest)
752{
753#ifdef CONFIG_SYSCTL
754 return ipvs->sysctl_expire_quiescent_template &&
755 (atomic_read(v: &dest->weight) == 0);
756#else
757 return 0;
758#endif
759}
760
761/*
762 * Checking if the destination of a connection template is available.
763 * If available, return 1, otherwise invalidate this connection
764 * template and return 0.
765 */
766int ip_vs_check_template(struct ip_vs_conn *ct, struct ip_vs_dest *cdest)
767{
768 struct ip_vs_dest *dest = ct->dest;
769 struct netns_ipvs *ipvs = ct->ipvs;
770
771 /*
772 * Checking the dest server status.
773 */
774 if ((dest == NULL) ||
775 !(dest->flags & IP_VS_DEST_F_AVAILABLE) ||
776 expire_quiescent_template(ipvs, dest) ||
777 (cdest && (dest != cdest))) {
778 IP_VS_DBG_BUF(9, "check_template: dest not available for "
779 "protocol %s s:%s:%d v:%s:%d "
780 "-> d:%s:%d\n",
781 ip_vs_proto_name(ct->protocol),
782 IP_VS_DBG_ADDR(ct->af, &ct->caddr),
783 ntohs(ct->cport),
784 IP_VS_DBG_ADDR(ct->af, &ct->vaddr),
785 ntohs(ct->vport),
786 IP_VS_DBG_ADDR(ct->daf, &ct->daddr),
787 ntohs(ct->dport));
788
789 /*
790 * Invalidate the connection template
791 */
792 if (ct->vport != htons(0xffff)) {
793 if (ip_vs_conn_unhash(cp: ct)) {
794 ct->dport = htons(0xffff);
795 ct->vport = htons(0xffff);
796 ct->cport = 0;
797 ip_vs_conn_hash(cp: ct);
798 }
799 }
800
801 /*
802 * Simply decrease the refcnt of the template,
803 * don't restart its timer.
804 */
805 __ip_vs_conn_put(cp: ct);
806 return 0;
807 }
808 return 1;
809}
810
811static void ip_vs_conn_rcu_free(struct rcu_head *head)
812{
813 struct ip_vs_conn *cp = container_of(head, struct ip_vs_conn,
814 rcu_head);
815
816 ip_vs_pe_put(cp->pe);
817 kfree(objp: cp->pe_data);
818 kmem_cache_free(s: ip_vs_conn_cachep, objp: cp);
819}
820
821/* Try to delete connection while not holding reference */
822static void ip_vs_conn_del(struct ip_vs_conn *cp)
823{
824 if (timer_delete(timer: &cp->timer)) {
825 /* Drop cp->control chain too */
826 if (cp->control)
827 cp->timeout = 0;
828 ip_vs_conn_expire(t: &cp->timer);
829 }
830}
831
832/* Try to delete connection while holding reference */
833static void ip_vs_conn_del_put(struct ip_vs_conn *cp)
834{
835 if (timer_delete(timer: &cp->timer)) {
836 /* Drop cp->control chain too */
837 if (cp->control)
838 cp->timeout = 0;
839 __ip_vs_conn_put(cp);
840 ip_vs_conn_expire(t: &cp->timer);
841 } else {
842 __ip_vs_conn_put(cp);
843 }
844}
845
846static void ip_vs_conn_expire(struct timer_list *t)
847{
848 struct ip_vs_conn *cp = timer_container_of(cp, t, timer);
849 struct netns_ipvs *ipvs = cp->ipvs;
850
851 /*
852 * do I control anybody?
853 */
854 if (atomic_read(v: &cp->n_control))
855 goto expire_later;
856
857 /* Unlink conn if not referenced anymore */
858 if (likely(ip_vs_conn_unlink(cp))) {
859 struct ip_vs_conn *ct = cp->control;
860
861 /* delete the timer if it is activated by other users */
862 timer_delete(timer: &cp->timer);
863
864 /* does anybody control me? */
865 if (ct) {
866 bool has_ref = !cp->timeout && __ip_vs_conn_get(cp: ct);
867
868 ip_vs_control_del(cp);
869 /* Drop CTL or non-assured TPL if not used anymore */
870 if (has_ref && !atomic_read(v: &ct->n_control) &&
871 (!(ct->flags & IP_VS_CONN_F_TEMPLATE) ||
872 !(ct->state & IP_VS_CTPL_S_ASSURED))) {
873 IP_VS_DBG(4, "drop controlling connection\n");
874 ip_vs_conn_del_put(cp: ct);
875 } else if (has_ref) {
876 __ip_vs_conn_put(cp: ct);
877 }
878 }
879
880 if ((cp->flags & IP_VS_CONN_F_NFCT) &&
881 !(cp->flags & IP_VS_CONN_F_ONE_PACKET)) {
882 /* Do not access conntracks during subsys cleanup
883 * because nf_conntrack_find_get can not be used after
884 * conntrack cleanup for the net.
885 */
886 smp_rmb();
887 if (READ_ONCE(ipvs->enable))
888 ip_vs_conn_drop_conntrack(cp);
889 }
890
891 if (unlikely(cp->app != NULL))
892 ip_vs_unbind_app(cp);
893 ip_vs_unbind_dest(cp);
894 if (cp->flags & IP_VS_CONN_F_NO_CPORT)
895 atomic_dec(v: &ip_vs_conn_no_cport_cnt);
896 if (cp->flags & IP_VS_CONN_F_ONE_PACKET)
897 ip_vs_conn_rcu_free(head: &cp->rcu_head);
898 else
899 call_rcu(head: &cp->rcu_head, func: ip_vs_conn_rcu_free);
900 atomic_dec(v: &ipvs->conn_count);
901 return;
902 }
903
904 expire_later:
905 IP_VS_DBG(7, "delayed: conn->refcnt=%d conn->n_control=%d\n",
906 refcount_read(&cp->refcnt),
907 atomic_read(&cp->n_control));
908
909 refcount_inc(r: &cp->refcnt);
910 cp->timeout = 60*HZ;
911
912 if (ipvs->sync_state & IP_VS_STATE_MASTER)
913 ip_vs_sync_conn(ipvs, cp, pkts: sysctl_sync_threshold(ipvs));
914
915 __ip_vs_conn_put_timer(cp);
916}
917
918/* Modify timer, so that it expires as soon as possible.
919 * Can be called without reference only if under RCU lock.
920 * We can have such chain of conns linked with ->control: DATA->CTL->TPL
921 * - DATA (eg. FTP) and TPL (persistence) can be present depending on setup
922 * - cp->timeout=0 indicates all conns from chain should be dropped but
923 * TPL is not dropped if in assured state
924 */
925void ip_vs_conn_expire_now(struct ip_vs_conn *cp)
926{
927 /* Using mod_timer_pending will ensure the timer is not
928 * modified after the final timer_delete in ip_vs_conn_expire.
929 */
930 if (timer_pending(timer: &cp->timer) &&
931 time_after(cp->timer.expires, jiffies))
932 mod_timer_pending(timer: &cp->timer, expires: jiffies);
933}
934
935
936/*
937 * Create a new connection entry and hash it into the ip_vs_conn_tab
938 */
939struct ip_vs_conn *
940ip_vs_conn_new(const struct ip_vs_conn_param *p, int dest_af,
941 const union nf_inet_addr *daddr, __be16 dport, unsigned int flags,
942 struct ip_vs_dest *dest, __u32 fwmark)
943{
944 struct ip_vs_conn *cp;
945 struct netns_ipvs *ipvs = p->ipvs;
946 struct ip_vs_proto_data *pd = ip_vs_proto_data_get(ipvs: p->ipvs,
947 proto: p->protocol);
948
949 cp = kmem_cache_alloc(ip_vs_conn_cachep, GFP_ATOMIC);
950 if (cp == NULL) {
951 IP_VS_ERR_RL("%s(): no memory\n", __func__);
952 return NULL;
953 }
954
955 INIT_HLIST_NODE(h: &cp->c_list);
956 timer_setup(&cp->timer, ip_vs_conn_expire, 0);
957 cp->ipvs = ipvs;
958 cp->af = p->af;
959 cp->daf = dest_af;
960 cp->protocol = p->protocol;
961 ip_vs_addr_set(af: p->af, dst: &cp->caddr, src: p->caddr);
962 cp->cport = p->cport;
963 /* proto should only be IPPROTO_IP if p->vaddr is a fwmark */
964 ip_vs_addr_set(af: p->protocol == IPPROTO_IP ? AF_UNSPEC : p->af,
965 dst: &cp->vaddr, src: p->vaddr);
966 cp->vport = p->vport;
967 ip_vs_addr_set(af: cp->daf, dst: &cp->daddr, src: daddr);
968 cp->dport = dport;
969 cp->flags = flags;
970 cp->fwmark = fwmark;
971 if (flags & IP_VS_CONN_F_TEMPLATE && p->pe) {
972 ip_vs_pe_get(p->pe);
973 cp->pe = p->pe;
974 cp->pe_data = p->pe_data;
975 cp->pe_data_len = p->pe_data_len;
976 } else {
977 cp->pe = NULL;
978 cp->pe_data = NULL;
979 cp->pe_data_len = 0;
980 }
981 spin_lock_init(&cp->lock);
982
983 /*
984 * Set the entry is referenced by the current thread before hashing
985 * it in the table, so that other thread run ip_vs_random_dropentry
986 * but cannot drop this entry.
987 */
988 refcount_set(r: &cp->refcnt, n: 1);
989
990 cp->control = NULL;
991 atomic_set(v: &cp->n_control, i: 0);
992 atomic_set(v: &cp->in_pkts, i: 0);
993
994 cp->packet_xmit = NULL;
995 cp->app = NULL;
996 cp->app_data = NULL;
997 /* reset struct ip_vs_seq */
998 cp->in_seq.delta = 0;
999 cp->out_seq.delta = 0;
1000
1001 atomic_inc(v: &ipvs->conn_count);
1002 if (flags & IP_VS_CONN_F_NO_CPORT)
1003 atomic_inc(v: &ip_vs_conn_no_cport_cnt);
1004
1005 /* Bind the connection with a destination server */
1006 cp->dest = NULL;
1007 ip_vs_bind_dest(cp, dest);
1008
1009 /* Set its state and timeout */
1010 cp->state = 0;
1011 cp->old_state = 0;
1012 cp->timeout = 3*HZ;
1013 cp->sync_endtime = jiffies & ~3UL;
1014
1015 /* Bind its packet transmitter */
1016#ifdef CONFIG_IP_VS_IPV6
1017 if (p->af == AF_INET6)
1018 ip_vs_bind_xmit_v6(cp);
1019 else
1020#endif
1021 ip_vs_bind_xmit(cp);
1022
1023 if (unlikely(pd && atomic_read(&pd->appcnt)))
1024 ip_vs_bind_app(cp, pp: pd->pp);
1025
1026 /*
1027 * Allow conntrack to be preserved. By default, conntrack
1028 * is created and destroyed for every packet.
1029 * Sometimes keeping conntrack can be useful for
1030 * IP_VS_CONN_F_ONE_PACKET too.
1031 */
1032
1033 if (ip_vs_conntrack_enabled(ipvs))
1034 cp->flags |= IP_VS_CONN_F_NFCT;
1035
1036 /* Hash it in the ip_vs_conn_tab finally */
1037 ip_vs_conn_hash(cp);
1038
1039 return cp;
1040}
1041
1042/*
1043 * /proc/net/ip_vs_conn entries
1044 */
1045#ifdef CONFIG_PROC_FS
1046struct ip_vs_iter_state {
1047 struct seq_net_private p;
1048 unsigned int bucket;
1049 unsigned int skip_elems;
1050};
1051
1052static void *ip_vs_conn_array(struct ip_vs_iter_state *iter)
1053{
1054 int idx;
1055 struct ip_vs_conn *cp;
1056
1057 for (idx = iter->bucket; idx < ip_vs_conn_tab_size; idx++) {
1058 unsigned int skip = 0;
1059
1060 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1061 /* __ip_vs_conn_get() is not needed by
1062 * ip_vs_conn_seq_show and ip_vs_conn_sync_seq_show
1063 */
1064 if (skip >= iter->skip_elems) {
1065 iter->bucket = idx;
1066 return cp;
1067 }
1068
1069 ++skip;
1070 }
1071
1072 iter->skip_elems = 0;
1073 cond_resched_rcu();
1074 }
1075
1076 iter->bucket = idx;
1077 return NULL;
1078}
1079
1080static void *ip_vs_conn_seq_start(struct seq_file *seq, loff_t *pos)
1081 __acquires(RCU)
1082{
1083 struct ip_vs_iter_state *iter = seq->private;
1084
1085 rcu_read_lock();
1086 if (*pos == 0) {
1087 iter->skip_elems = 0;
1088 iter->bucket = 0;
1089 return SEQ_START_TOKEN;
1090 }
1091
1092 return ip_vs_conn_array(iter);
1093}
1094
1095static void *ip_vs_conn_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1096{
1097 struct ip_vs_conn *cp = v;
1098 struct ip_vs_iter_state *iter = seq->private;
1099 struct hlist_node *e;
1100
1101 ++*pos;
1102 if (v == SEQ_START_TOKEN)
1103 return ip_vs_conn_array(iter);
1104
1105 /* more on same hash chain? */
1106 e = rcu_dereference(hlist_next_rcu(&cp->c_list));
1107 if (e) {
1108 iter->skip_elems++;
1109 return hlist_entry(e, struct ip_vs_conn, c_list);
1110 }
1111
1112 iter->skip_elems = 0;
1113 iter->bucket++;
1114
1115 return ip_vs_conn_array(iter);
1116}
1117
1118static void ip_vs_conn_seq_stop(struct seq_file *seq, void *v)
1119 __releases(RCU)
1120{
1121 rcu_read_unlock();
1122}
1123
1124static int ip_vs_conn_seq_show(struct seq_file *seq, void *v)
1125{
1126
1127 if (v == SEQ_START_TOKEN)
1128 seq_puts(m: seq,
1129 s: "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Expires PEName PEData\n");
1130 else {
1131 const struct ip_vs_conn *cp = v;
1132 struct net *net = seq_file_net(seq);
1133 char pe_data[IP_VS_PENAME_MAXLEN + IP_VS_PEDATA_MAXLEN + 3];
1134 size_t len = 0;
1135 char dbuf[IP_VS_ADDRSTRLEN];
1136
1137 if (!net_eq(net1: cp->ipvs->net, net2: net))
1138 return 0;
1139 if (cp->pe_data) {
1140 pe_data[0] = ' ';
1141 len = strlen(cp->pe->name);
1142 memcpy(pe_data + 1, cp->pe->name, len);
1143 pe_data[len + 1] = ' ';
1144 len += 2;
1145 len += cp->pe->show_pe_data(cp, pe_data + len);
1146 }
1147 pe_data[len] = '\0';
1148
1149#ifdef CONFIG_IP_VS_IPV6
1150 if (cp->daf == AF_INET6)
1151 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%pI6", &cp->daddr.in6);
1152 else
1153#endif
1154 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%08X",
1155 ntohl(cp->daddr.ip));
1156
1157#ifdef CONFIG_IP_VS_IPV6
1158 if (cp->af == AF_INET6)
1159 seq_printf(m: seq, fmt: "%-3s %pI6 %04X %pI6 %04X "
1160 "%s %04X %-11s %7u%s\n",
1161 ip_vs_proto_name(proto: cp->protocol),
1162 &cp->caddr.in6, ntohs(cp->cport),
1163 &cp->vaddr.in6, ntohs(cp->vport),
1164 dbuf, ntohs(cp->dport),
1165 ip_vs_state_name(cp),
1166 jiffies_delta_to_msecs(delta: cp->timer.expires -
1167 jiffies) / 1000,
1168 pe_data);
1169 else
1170#endif
1171 seq_printf(m: seq,
1172 fmt: "%-3s %08X %04X %08X %04X"
1173 " %s %04X %-11s %7u%s\n",
1174 ip_vs_proto_name(proto: cp->protocol),
1175 ntohl(cp->caddr.ip), ntohs(cp->cport),
1176 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1177 dbuf, ntohs(cp->dport),
1178 ip_vs_state_name(cp),
1179 jiffies_delta_to_msecs(delta: cp->timer.expires -
1180 jiffies) / 1000,
1181 pe_data);
1182 }
1183 return 0;
1184}
1185
1186static const struct seq_operations ip_vs_conn_seq_ops = {
1187 .start = ip_vs_conn_seq_start,
1188 .next = ip_vs_conn_seq_next,
1189 .stop = ip_vs_conn_seq_stop,
1190 .show = ip_vs_conn_seq_show,
1191};
1192
1193static const char *ip_vs_origin_name(unsigned int flags)
1194{
1195 if (flags & IP_VS_CONN_F_SYNC)
1196 return "SYNC";
1197 else
1198 return "LOCAL";
1199}
1200
1201static int ip_vs_conn_sync_seq_show(struct seq_file *seq, void *v)
1202{
1203 char dbuf[IP_VS_ADDRSTRLEN];
1204
1205 if (v == SEQ_START_TOKEN)
1206 seq_puts(m: seq,
1207 s: "Pro FromIP FPrt ToIP TPrt DestIP DPrt State Origin Expires\n");
1208 else {
1209 const struct ip_vs_conn *cp = v;
1210 struct net *net = seq_file_net(seq);
1211
1212 if (!net_eq(net1: cp->ipvs->net, net2: net))
1213 return 0;
1214
1215#ifdef CONFIG_IP_VS_IPV6
1216 if (cp->daf == AF_INET6)
1217 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%pI6", &cp->daddr.in6);
1218 else
1219#endif
1220 snprintf(buf: dbuf, size: sizeof(dbuf), fmt: "%08X",
1221 ntohl(cp->daddr.ip));
1222
1223#ifdef CONFIG_IP_VS_IPV6
1224 if (cp->af == AF_INET6)
1225 seq_printf(m: seq, fmt: "%-3s %pI6 %04X %pI6 %04X "
1226 "%s %04X %-11s %-6s %7u\n",
1227 ip_vs_proto_name(proto: cp->protocol),
1228 &cp->caddr.in6, ntohs(cp->cport),
1229 &cp->vaddr.in6, ntohs(cp->vport),
1230 dbuf, ntohs(cp->dport),
1231 ip_vs_state_name(cp),
1232 ip_vs_origin_name(flags: cp->flags),
1233 jiffies_delta_to_msecs(delta: cp->timer.expires -
1234 jiffies) / 1000);
1235 else
1236#endif
1237 seq_printf(m: seq,
1238 fmt: "%-3s %08X %04X %08X %04X "
1239 "%s %04X %-11s %-6s %7u\n",
1240 ip_vs_proto_name(proto: cp->protocol),
1241 ntohl(cp->caddr.ip), ntohs(cp->cport),
1242 ntohl(cp->vaddr.ip), ntohs(cp->vport),
1243 dbuf, ntohs(cp->dport),
1244 ip_vs_state_name(cp),
1245 ip_vs_origin_name(flags: cp->flags),
1246 jiffies_delta_to_msecs(delta: cp->timer.expires -
1247 jiffies) / 1000);
1248 }
1249 return 0;
1250}
1251
1252static const struct seq_operations ip_vs_conn_sync_seq_ops = {
1253 .start = ip_vs_conn_seq_start,
1254 .next = ip_vs_conn_seq_next,
1255 .stop = ip_vs_conn_seq_stop,
1256 .show = ip_vs_conn_sync_seq_show,
1257};
1258#endif
1259
1260
1261/* Randomly drop connection entries before running out of memory
1262 * Can be used for DATA and CTL conns. For TPL conns there are exceptions:
1263 * - traffic for services in OPS mode increases ct->in_pkts, so it is supported
1264 * - traffic for services not in OPS mode does not increase ct->in_pkts in
1265 * all cases, so it is not supported
1266 */
1267static inline int todrop_entry(struct ip_vs_conn *cp)
1268{
1269 /*
1270 * The drop rate array needs tuning for real environments.
1271 * Called from timer bh only => no locking
1272 */
1273 static const signed char todrop_rate[9] = {0, 1, 2, 3, 4, 5, 6, 7, 8};
1274 static signed char todrop_counter[9] = {0};
1275 int i;
1276
1277 /* if the conn entry hasn't lasted for 60 seconds, don't drop it.
1278 This will leave enough time for normal connection to get
1279 through. */
1280 if (time_before(cp->timeout + jiffies, cp->timer.expires + 60*HZ))
1281 return 0;
1282
1283 /* Don't drop the entry if its number of incoming packets is not
1284 located in [0, 8] */
1285 i = atomic_read(v: &cp->in_pkts);
1286 if (i > 8 || i < 0) return 0;
1287
1288 if (!todrop_rate[i]) return 0;
1289 if (--todrop_counter[i] > 0) return 0;
1290
1291 todrop_counter[i] = todrop_rate[i];
1292 return 1;
1293}
1294
1295static inline bool ip_vs_conn_ops_mode(struct ip_vs_conn *cp)
1296{
1297 struct ip_vs_service *svc;
1298
1299 if (!cp->dest)
1300 return false;
1301 svc = rcu_dereference(cp->dest->svc);
1302 return svc && (svc->flags & IP_VS_SVC_F_ONEPACKET);
1303}
1304
1305/* Called from keventd and must protect itself from softirqs */
1306void ip_vs_random_dropentry(struct netns_ipvs *ipvs)
1307{
1308 int idx;
1309 struct ip_vs_conn *cp;
1310
1311 rcu_read_lock();
1312 /*
1313 * Randomly scan 1/32 of the whole table every second
1314 */
1315 for (idx = 0; idx < (ip_vs_conn_tab_size>>5); idx++) {
1316 unsigned int hash = get_random_u32() & ip_vs_conn_tab_mask;
1317
1318 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[hash], c_list) {
1319 if (cp->ipvs != ipvs)
1320 continue;
1321 if (atomic_read(v: &cp->n_control))
1322 continue;
1323 if (cp->flags & IP_VS_CONN_F_TEMPLATE) {
1324 /* connection template of OPS */
1325 if (ip_vs_conn_ops_mode(cp))
1326 goto try_drop;
1327 if (!(cp->state & IP_VS_CTPL_S_ASSURED))
1328 goto drop;
1329 continue;
1330 }
1331 if (cp->protocol == IPPROTO_TCP) {
1332 switch(cp->state) {
1333 case IP_VS_TCP_S_SYN_RECV:
1334 case IP_VS_TCP_S_SYNACK:
1335 break;
1336
1337 case IP_VS_TCP_S_ESTABLISHED:
1338 if (todrop_entry(cp))
1339 break;
1340 continue;
1341
1342 default:
1343 continue;
1344 }
1345 } else if (cp->protocol == IPPROTO_SCTP) {
1346 switch (cp->state) {
1347 case IP_VS_SCTP_S_INIT1:
1348 case IP_VS_SCTP_S_INIT:
1349 break;
1350 case IP_VS_SCTP_S_ESTABLISHED:
1351 if (todrop_entry(cp))
1352 break;
1353 continue;
1354 default:
1355 continue;
1356 }
1357 } else {
1358try_drop:
1359 if (!todrop_entry(cp))
1360 continue;
1361 }
1362
1363drop:
1364 IP_VS_DBG(4, "drop connection\n");
1365 ip_vs_conn_del(cp);
1366 }
1367 cond_resched_rcu();
1368 }
1369 rcu_read_unlock();
1370}
1371
1372
1373/*
1374 * Flush all the connection entries in the ip_vs_conn_tab
1375 */
1376static void ip_vs_conn_flush(struct netns_ipvs *ipvs)
1377{
1378 int idx;
1379 struct ip_vs_conn *cp, *cp_c;
1380
1381flush_again:
1382 rcu_read_lock();
1383 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1384
1385 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1386 if (cp->ipvs != ipvs)
1387 continue;
1388 if (atomic_read(v: &cp->n_control))
1389 continue;
1390 cp_c = cp->control;
1391 IP_VS_DBG(4, "del connection\n");
1392 ip_vs_conn_del(cp);
1393 if (cp_c && !atomic_read(v: &cp_c->n_control)) {
1394 IP_VS_DBG(4, "del controlling connection\n");
1395 ip_vs_conn_del(cp: cp_c);
1396 }
1397 }
1398 cond_resched_rcu();
1399 }
1400 rcu_read_unlock();
1401
1402 /* the counter may be not NULL, because maybe some conn entries
1403 are run by slow timer handler or unhashed but still referred */
1404 if (atomic_read(v: &ipvs->conn_count) != 0) {
1405 schedule();
1406 goto flush_again;
1407 }
1408}
1409
1410#ifdef CONFIG_SYSCTL
1411void ip_vs_expire_nodest_conn_flush(struct netns_ipvs *ipvs)
1412{
1413 int idx;
1414 struct ip_vs_conn *cp, *cp_c;
1415 struct ip_vs_dest *dest;
1416
1417 rcu_read_lock();
1418 for (idx = 0; idx < ip_vs_conn_tab_size; idx++) {
1419 hlist_for_each_entry_rcu(cp, &ip_vs_conn_tab[idx], c_list) {
1420 if (cp->ipvs != ipvs)
1421 continue;
1422
1423 dest = cp->dest;
1424 if (!dest || (dest->flags & IP_VS_DEST_F_AVAILABLE))
1425 continue;
1426
1427 if (atomic_read(v: &cp->n_control))
1428 continue;
1429
1430 cp_c = cp->control;
1431 IP_VS_DBG(4, "del connection\n");
1432 ip_vs_conn_del(cp);
1433 if (cp_c && !atomic_read(v: &cp_c->n_control)) {
1434 IP_VS_DBG(4, "del controlling connection\n");
1435 ip_vs_conn_del(cp: cp_c);
1436 }
1437 }
1438 cond_resched_rcu();
1439
1440 /* netns clean up started, abort delayed work */
1441 if (!READ_ONCE(ipvs->enable))
1442 break;
1443 }
1444 rcu_read_unlock();
1445}
1446#endif
1447
1448/*
1449 * per netns init and exit
1450 */
1451int __net_init ip_vs_conn_net_init(struct netns_ipvs *ipvs)
1452{
1453 atomic_set(v: &ipvs->conn_count, i: 0);
1454
1455#ifdef CONFIG_PROC_FS
1456 if (!proc_create_net("ip_vs_conn", 0, ipvs->net->proc_net,
1457 &ip_vs_conn_seq_ops,
1458 sizeof(struct ip_vs_iter_state)))
1459 goto err_conn;
1460
1461 if (!proc_create_net("ip_vs_conn_sync", 0, ipvs->net->proc_net,
1462 &ip_vs_conn_sync_seq_ops,
1463 sizeof(struct ip_vs_iter_state)))
1464 goto err_conn_sync;
1465#endif
1466
1467 return 0;
1468
1469#ifdef CONFIG_PROC_FS
1470err_conn_sync:
1471 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1472err_conn:
1473 return -ENOMEM;
1474#endif
1475}
1476
1477void __net_exit ip_vs_conn_net_cleanup(struct netns_ipvs *ipvs)
1478{
1479 /* flush all the connection entries first */
1480 ip_vs_conn_flush(ipvs);
1481#ifdef CONFIG_PROC_FS
1482 remove_proc_entry("ip_vs_conn", ipvs->net->proc_net);
1483 remove_proc_entry("ip_vs_conn_sync", ipvs->net->proc_net);
1484#endif
1485}
1486
1487int __init ip_vs_conn_init(void)
1488{
1489 size_t tab_array_size;
1490 int max_avail;
1491#if BITS_PER_LONG > 32
1492 int max = 27;
1493#else
1494 int max = 20;
1495#endif
1496 int min = 8;
1497 int idx;
1498
1499 max_avail = order_base_2(totalram_pages()) + PAGE_SHIFT;
1500 max_avail -= 2; /* ~4 in hash row */
1501 max_avail -= 1; /* IPVS up to 1/2 of mem */
1502 max_avail -= order_base_2(sizeof(struct ip_vs_conn));
1503 max = clamp(max_avail, min, max);
1504 ip_vs_conn_tab_bits = clamp(ip_vs_conn_tab_bits, min, max);
1505 ip_vs_conn_tab_size = 1 << ip_vs_conn_tab_bits;
1506 ip_vs_conn_tab_mask = ip_vs_conn_tab_size - 1;
1507
1508 /*
1509 * Allocate the connection hash table and initialize its list heads
1510 */
1511 tab_array_size = array_size(ip_vs_conn_tab_size,
1512 sizeof(*ip_vs_conn_tab));
1513 ip_vs_conn_tab = kvmalloc_array(ip_vs_conn_tab_size,
1514 sizeof(*ip_vs_conn_tab), GFP_KERNEL);
1515 if (!ip_vs_conn_tab)
1516 return -ENOMEM;
1517
1518 /* Allocate ip_vs_conn slab cache */
1519 ip_vs_conn_cachep = KMEM_CACHE(ip_vs_conn, SLAB_HWCACHE_ALIGN);
1520 if (!ip_vs_conn_cachep) {
1521 kvfree(addr: ip_vs_conn_tab);
1522 return -ENOMEM;
1523 }
1524
1525 pr_info("Connection hash table configured (size=%d, memory=%zdKbytes)\n",
1526 ip_vs_conn_tab_size, tab_array_size / 1024);
1527 IP_VS_DBG(0, "Each connection entry needs %zd bytes at least\n",
1528 sizeof(struct ip_vs_conn));
1529
1530 for (idx = 0; idx < ip_vs_conn_tab_size; idx++)
1531 INIT_HLIST_HEAD(&ip_vs_conn_tab[idx]);
1532
1533 for (idx = 0; idx < CT_LOCKARRAY_SIZE; idx++) {
1534 spin_lock_init(&__ip_vs_conntbl_lock_array[idx].l);
1535 }
1536
1537 /* calculate the random value for connection hash */
1538 get_random_bytes(buf: &ip_vs_conn_rnd, len: sizeof(ip_vs_conn_rnd));
1539
1540 return 0;
1541}
1542
1543void ip_vs_conn_cleanup(void)
1544{
1545 /* Wait all ip_vs_conn_rcu_free() callbacks to complete */
1546 rcu_barrier();
1547 /* Release the empty cache */
1548 kmem_cache_destroy(s: ip_vs_conn_cachep);
1549 kvfree(addr: ip_vs_conn_tab);
1550}
1551

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