]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
cb7f6a7b JV |
2 | /* |
3 | * IPVS: Destination Hashing scheduling module | |
4 | * | |
5 | * Authors: Wensong Zhang <wensong@gnuchina.org> | |
6 | * | |
7 | * Inspired by the consistent hashing scheduler patch from | |
8 | * Thomas Proell <proellt@gmx.de> | |
9 | * | |
cb7f6a7b | 10 | * Changes: |
cb7f6a7b JV |
11 | */ |
12 | ||
13 | /* | |
14 | * The dh algorithm is to select server by the hash key of destination IP | |
15 | * address. The pseudo code is as follows: | |
16 | * | |
17 | * n <- servernode[dest_ip]; | |
18 | * if (n is dead) OR | |
19 | * (n is overloaded) OR (n.weight <= 0) then | |
20 | * return NULL; | |
21 | * | |
22 | * return n; | |
23 | * | |
24 | * Notes that servernode is a 256-bucket hash table that maps the hash | |
25 | * index derived from packet destination IP address to the current server | |
26 | * array. If the dh scheduler is used in cache cluster, it is good to | |
27 | * combine it with cache_bypass feature. When the statically assigned | |
28 | * server is dead or overloaded, the load balancer can bypass the cache | |
29 | * server and send requests to the original server directly. | |
30 | * | |
31 | */ | |
32 | ||
9aada7ac HE |
33 | #define KMSG_COMPONENT "IPVS" |
34 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt | |
35 | ||
cb7f6a7b | 36 | #include <linux/ip.h> |
5a0e3ad6 | 37 | #include <linux/slab.h> |
cb7f6a7b JV |
38 | #include <linux/module.h> |
39 | #include <linux/kernel.h> | |
40 | #include <linux/skbuff.h> | |
9a17740e | 41 | #include <linux/hash.h> |
cb7f6a7b JV |
42 | |
43 | #include <net/ip_vs.h> | |
44 | ||
45 | ||
46 | /* | |
47 | * IPVS DH bucket | |
48 | */ | |
49 | struct ip_vs_dh_bucket { | |
8f3d0023 | 50 | struct ip_vs_dest __rcu *dest; /* real server (cache) */ |
cb7f6a7b JV |
51 | }; |
52 | ||
53 | /* | |
54 | * for IPVS DH entry hash table | |
55 | */ | |
56 | #ifndef CONFIG_IP_VS_DH_TAB_BITS | |
57 | #define CONFIG_IP_VS_DH_TAB_BITS 8 | |
58 | #endif | |
59 | #define IP_VS_DH_TAB_BITS CONFIG_IP_VS_DH_TAB_BITS | |
60 | #define IP_VS_DH_TAB_SIZE (1 << IP_VS_DH_TAB_BITS) | |
61 | #define IP_VS_DH_TAB_MASK (IP_VS_DH_TAB_SIZE - 1) | |
62 | ||
8f3d0023 JA |
63 | struct ip_vs_dh_state { |
64 | struct ip_vs_dh_bucket buckets[IP_VS_DH_TAB_SIZE]; | |
65 | struct rcu_head rcu_head; | |
66 | }; | |
cb7f6a7b JV |
67 | |
68 | /* | |
69 | * Returns hash value for IPVS DH entry | |
70 | */ | |
95c96174 | 71 | static inline unsigned int ip_vs_dh_hashkey(int af, const union nf_inet_addr *addr) |
cb7f6a7b | 72 | { |
20971a0a JV |
73 | __be32 addr_fold = addr->ip; |
74 | ||
75 | #ifdef CONFIG_IP_VS_IPV6 | |
76 | if (af == AF_INET6) | |
77 | addr_fold = addr->ip6[0]^addr->ip6[1]^ | |
78 | addr->ip6[2]^addr->ip6[3]; | |
79 | #endif | |
9a17740e | 80 | return hash_32(ntohl(addr_fold), IP_VS_DH_TAB_BITS); |
cb7f6a7b JV |
81 | } |
82 | ||
83 | ||
84 | /* | |
85 | * Get ip_vs_dest associated with supplied parameters. | |
86 | */ | |
87 | static inline struct ip_vs_dest * | |
8f3d0023 | 88 | ip_vs_dh_get(int af, struct ip_vs_dh_state *s, const union nf_inet_addr *addr) |
cb7f6a7b | 89 | { |
8f3d0023 | 90 | return rcu_dereference(s->buckets[ip_vs_dh_hashkey(af, addr)].dest); |
cb7f6a7b JV |
91 | } |
92 | ||
93 | ||
94 | /* | |
95 | * Assign all the hash buckets of the specified table with the service. | |
96 | */ | |
97 | static int | |
8f3d0023 | 98 | ip_vs_dh_reassign(struct ip_vs_dh_state *s, struct ip_vs_service *svc) |
cb7f6a7b JV |
99 | { |
100 | int i; | |
101 | struct ip_vs_dh_bucket *b; | |
102 | struct list_head *p; | |
103 | struct ip_vs_dest *dest; | |
8f3d0023 | 104 | bool empty; |
cb7f6a7b | 105 | |
8f3d0023 | 106 | b = &s->buckets[0]; |
cb7f6a7b | 107 | p = &svc->destinations; |
8f3d0023 | 108 | empty = list_empty(p); |
cb7f6a7b | 109 | for (i=0; i<IP_VS_DH_TAB_SIZE; i++) { |
8f3d0023 JA |
110 | dest = rcu_dereference_protected(b->dest, 1); |
111 | if (dest) | |
112 | ip_vs_dest_put(dest); | |
113 | if (empty) | |
114 | RCU_INIT_POINTER(b->dest, NULL); | |
115 | else { | |
cb7f6a7b JV |
116 | if (p == &svc->destinations) |
117 | p = p->next; | |
118 | ||
119 | dest = list_entry(p, struct ip_vs_dest, n_list); | |
8f3d0023 JA |
120 | ip_vs_dest_hold(dest); |
121 | RCU_INIT_POINTER(b->dest, dest); | |
cb7f6a7b JV |
122 | |
123 | p = p->next; | |
124 | } | |
125 | b++; | |
126 | } | |
127 | return 0; | |
128 | } | |
129 | ||
130 | ||
131 | /* | |
132 | * Flush all the hash buckets of the specified table. | |
133 | */ | |
8f3d0023 | 134 | static void ip_vs_dh_flush(struct ip_vs_dh_state *s) |
cb7f6a7b JV |
135 | { |
136 | int i; | |
137 | struct ip_vs_dh_bucket *b; | |
8f3d0023 | 138 | struct ip_vs_dest *dest; |
cb7f6a7b | 139 | |
8f3d0023 | 140 | b = &s->buckets[0]; |
cb7f6a7b | 141 | for (i=0; i<IP_VS_DH_TAB_SIZE; i++) { |
8f3d0023 JA |
142 | dest = rcu_dereference_protected(b->dest, 1); |
143 | if (dest) { | |
144 | ip_vs_dest_put(dest); | |
145 | RCU_INIT_POINTER(b->dest, NULL); | |
cb7f6a7b JV |
146 | } |
147 | b++; | |
148 | } | |
149 | } | |
150 | ||
151 | ||
152 | static int ip_vs_dh_init_svc(struct ip_vs_service *svc) | |
153 | { | |
8f3d0023 | 154 | struct ip_vs_dh_state *s; |
cb7f6a7b JV |
155 | |
156 | /* allocate the DH table for this service */ | |
8f3d0023 JA |
157 | s = kzalloc(sizeof(struct ip_vs_dh_state), GFP_KERNEL); |
158 | if (s == NULL) | |
cb7f6a7b | 159 | return -ENOMEM; |
0a9ee813 | 160 | |
8f3d0023 | 161 | svc->sched_data = s; |
5b5e0928 | 162 | IP_VS_DBG(6, "DH hash table (memory=%zdbytes) allocated for " |
cb7f6a7b JV |
163 | "current service\n", |
164 | sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE); | |
165 | ||
8f3d0023 JA |
166 | /* assign the hash buckets with current dests */ |
167 | ip_vs_dh_reassign(s, svc); | |
cb7f6a7b JV |
168 | |
169 | return 0; | |
170 | } | |
171 | ||
172 | ||
ed3ffc4e | 173 | static void ip_vs_dh_done_svc(struct ip_vs_service *svc) |
cb7f6a7b | 174 | { |
8f3d0023 | 175 | struct ip_vs_dh_state *s = svc->sched_data; |
cb7f6a7b JV |
176 | |
177 | /* got to clean up hash buckets here */ | |
8f3d0023 | 178 | ip_vs_dh_flush(s); |
cb7f6a7b JV |
179 | |
180 | /* release the table itself */ | |
8f3d0023 | 181 | kfree_rcu(s, rcu_head); |
5b5e0928 | 182 | IP_VS_DBG(6, "DH hash table (memory=%zdbytes) released\n", |
cb7f6a7b | 183 | sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE); |
cb7f6a7b JV |
184 | } |
185 | ||
186 | ||
8f3d0023 JA |
187 | static int ip_vs_dh_dest_changed(struct ip_vs_service *svc, |
188 | struct ip_vs_dest *dest) | |
cb7f6a7b | 189 | { |
8f3d0023 | 190 | struct ip_vs_dh_state *s = svc->sched_data; |
cb7f6a7b JV |
191 | |
192 | /* assign the hash buckets with the updated service */ | |
8f3d0023 | 193 | ip_vs_dh_reassign(s, svc); |
cb7f6a7b JV |
194 | |
195 | return 0; | |
196 | } | |
197 | ||
198 | ||
199 | /* | |
200 | * If the dest flags is set with IP_VS_DEST_F_OVERLOAD, | |
201 | * consider that the server is overloaded here. | |
202 | */ | |
203 | static inline int is_overloaded(struct ip_vs_dest *dest) | |
204 | { | |
205 | return dest->flags & IP_VS_DEST_F_OVERLOAD; | |
206 | } | |
207 | ||
208 | ||
209 | /* | |
210 | * Destination hashing scheduling | |
211 | */ | |
212 | static struct ip_vs_dest * | |
bba54de5 JA |
213 | ip_vs_dh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb, |
214 | struct ip_vs_iphdr *iph) | |
cb7f6a7b JV |
215 | { |
216 | struct ip_vs_dest *dest; | |
8f3d0023 | 217 | struct ip_vs_dh_state *s; |
cb7f6a7b | 218 | |
1e3e238e | 219 | IP_VS_DBG(6, "%s(): Scheduling...\n", __func__); |
cb7f6a7b | 220 | |
8f3d0023 | 221 | s = (struct ip_vs_dh_state *) svc->sched_data; |
bba54de5 | 222 | dest = ip_vs_dh_get(svc->af, s, &iph->daddr); |
cb7f6a7b JV |
223 | if (!dest |
224 | || !(dest->flags & IP_VS_DEST_F_AVAILABLE) | |
225 | || atomic_read(&dest->weight) <= 0 | |
226 | || is_overloaded(dest)) { | |
8f3d0023 | 227 | ip_vs_scheduler_err(svc, "no destination available"); |
cb7f6a7b JV |
228 | return NULL; |
229 | } | |
230 | ||
20971a0a | 231 | IP_VS_DBG_BUF(6, "DH: destination IP address %s --> server %s:%d\n", |
bba54de5 | 232 | IP_VS_DBG_ADDR(svc->af, &iph->daddr), |
4d316f3f | 233 | IP_VS_DBG_ADDR(dest->af, &dest->addr), |
20971a0a | 234 | ntohs(dest->port)); |
cb7f6a7b JV |
235 | |
236 | return dest; | |
237 | } | |
238 | ||
239 | ||
240 | /* | |
241 | * IPVS DH Scheduler structure | |
242 | */ | |
243 | static struct ip_vs_scheduler ip_vs_dh_scheduler = | |
244 | { | |
245 | .name = "dh", | |
246 | .refcnt = ATOMIC_INIT(0), | |
247 | .module = THIS_MODULE, | |
248 | .n_list = LIST_HEAD_INIT(ip_vs_dh_scheduler.n_list), | |
cb7f6a7b JV |
249 | .init_service = ip_vs_dh_init_svc, |
250 | .done_service = ip_vs_dh_done_svc, | |
8f3d0023 JA |
251 | .add_dest = ip_vs_dh_dest_changed, |
252 | .del_dest = ip_vs_dh_dest_changed, | |
cb7f6a7b JV |
253 | .schedule = ip_vs_dh_schedule, |
254 | }; | |
255 | ||
256 | ||
257 | static int __init ip_vs_dh_init(void) | |
258 | { | |
259 | return register_ip_vs_scheduler(&ip_vs_dh_scheduler); | |
260 | } | |
261 | ||
262 | ||
263 | static void __exit ip_vs_dh_cleanup(void) | |
264 | { | |
265 | unregister_ip_vs_scheduler(&ip_vs_dh_scheduler); | |
ceec4c38 | 266 | synchronize_rcu(); |
cb7f6a7b JV |
267 | } |
268 | ||
269 | ||
270 | module_init(ip_vs_dh_init); | |
271 | module_exit(ip_vs_dh_cleanup); | |
272 | MODULE_LICENSE("GPL"); |