blob: 03e9154f7e687efef63c91878e33427672bc4036 [file] [log] [blame]
Glauber Costad1a4c0b2011-12-11 21:47:04 +00001#include <net/tcp.h>
2#include <net/tcp_memcontrol.h>
3#include <net/sock.h>
Glauber Costa3dc43e32011-12-11 21:47:05 +00004#include <net/ip.h>
5#include <linux/nsproxy.h>
Glauber Costad1a4c0b2011-12-11 21:47:04 +00006#include <linux/memcontrol.h>
7#include <linux/module.h>
8
Glauber Costad1a4c0b2011-12-11 21:47:04 +00009static void memcg_tcp_enter_memory_pressure(struct sock *sk)
10{
Dan Carpenterc48e0742011-12-15 01:05:10 +000011 if (sk->sk_cgrp->memory_pressure)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070012 sk->sk_cgrp->memory_pressure = 1;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000013}
14EXPORT_SYMBOL(memcg_tcp_enter_memory_pressure);
15
Glauber Costa1d62e432012-04-09 19:36:33 -030016int tcp_init_cgroup(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
Glauber Costad1a4c0b2011-12-11 21:47:04 +000017{
18 /*
19 * The root cgroup does not use res_counters, but rather,
20 * rely on the data already collected by the network
21 * subsystem
22 */
23 struct res_counter *res_parent = NULL;
24 struct cg_proto *cg_proto, *parent_cg;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000025 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
26
27 cg_proto = tcp_prot.proto_cgroup(memcg);
28 if (!cg_proto)
Tejun Heo6bc10342012-04-01 12:09:55 -070029 return 0;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000030
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070031 cg_proto->sysctl_mem[0] = sysctl_tcp_mem[0];
32 cg_proto->sysctl_mem[1] = sysctl_tcp_mem[1];
33 cg_proto->sysctl_mem[2] = sysctl_tcp_mem[2];
34 cg_proto->memory_pressure = 0;
35 cg_proto->memcg = memcg;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000036
37 parent_cg = tcp_prot.proto_cgroup(parent);
38 if (parent_cg)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070039 res_parent = &parent_cg->memory_allocated;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000040
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070041 res_counter_init(&cg_proto->memory_allocated, res_parent);
42 percpu_counter_init(&cg_proto->sockets_allocated, 0);
Glauber Costad1a4c0b2011-12-11 21:47:04 +000043
Tejun Heo6bc10342012-04-01 12:09:55 -070044 return 0;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000045}
46EXPORT_SYMBOL(tcp_init_cgroup);
47
Glauber Costa1d62e432012-04-09 19:36:33 -030048void tcp_destroy_cgroup(struct mem_cgroup *memcg)
Glauber Costad1a4c0b2011-12-11 21:47:04 +000049{
Glauber Costad1a4c0b2011-12-11 21:47:04 +000050 struct cg_proto *cg_proto;
Glauber Costad1a4c0b2011-12-11 21:47:04 +000051
52 cg_proto = tcp_prot.proto_cgroup(memcg);
53 if (!cg_proto)
54 return;
55
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070056 percpu_counter_destroy(&cg_proto->sockets_allocated);
Glauber Costad1a4c0b2011-12-11 21:47:04 +000057}
58EXPORT_SYMBOL(tcp_destroy_cgroup);
Glauber Costa3aaabe22011-12-11 21:47:06 +000059
60static int tcp_update_limit(struct mem_cgroup *memcg, u64 val)
61{
Glauber Costa3aaabe22011-12-11 21:47:06 +000062 struct cg_proto *cg_proto;
63 u64 old_lim;
64 int i;
65 int ret;
66
67 cg_proto = tcp_prot.proto_cgroup(memcg);
68 if (!cg_proto)
69 return -EINVAL;
70
Sha Zhengju6de5a8b2013-09-12 15:13:47 -070071 if (val > RES_COUNTER_MAX)
72 val = RES_COUNTER_MAX;
Glauber Costa3aaabe22011-12-11 21:47:06 +000073
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070074 old_lim = res_counter_read_u64(&cg_proto->memory_allocated, RES_LIMIT);
75 ret = res_counter_set_limit(&cg_proto->memory_allocated, val);
Glauber Costa3aaabe22011-12-11 21:47:06 +000076 if (ret)
77 return ret;
78
79 for (i = 0; i < 3; i++)
Eric W. Biederman2e685ca2013-10-19 16:26:19 -070080 cg_proto->sysctl_mem[i] = min_t(long, val >> PAGE_SHIFT,
81 sysctl_tcp_mem[i]);
Glauber Costa3aaabe22011-12-11 21:47:06 +000082
Sha Zhengju6de5a8b2013-09-12 15:13:47 -070083 if (val == RES_COUNTER_MAX)
Glauber Costa3f134612012-05-29 15:07:11 -070084 clear_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
Sha Zhengju6de5a8b2013-09-12 15:13:47 -070085 else if (val != RES_COUNTER_MAX) {
Glauber Costa3f134612012-05-29 15:07:11 -070086 /*
87 * The active bit needs to be written after the static_key
88 * update. This is what guarantees that the socket activation
89 * function is the last one to run. See sock_update_memcg() for
90 * details, and note that we don't mark any socket as belonging
91 * to this memcg until that flag is up.
92 *
93 * We need to do this, because static_keys will span multiple
94 * sites, but we can't control their order. If we mark a socket
95 * as accounted, but the accounting functions are not patched in
96 * yet, we'll lose accounting.
97 *
98 * We never race with the readers in sock_update_memcg(),
99 * because when this value change, the code to process it is not
100 * patched in yet.
101 *
102 * The activated bit is used to guarantee that no two writers
103 * will do the update in the same memcg. Without that, we can't
104 * properly shutdown the static key.
105 */
106 if (!test_and_set_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags))
107 static_key_slow_inc(&memcg_socket_limit_enabled);
108 set_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
109 }
Glauber Costa3aaabe22011-12-11 21:47:06 +0000110
111 return 0;
112}
113
Tejun Heo182446d2013-08-08 20:11:24 -0400114static int tcp_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft,
Glauber Costa3aaabe22011-12-11 21:47:06 +0000115 const char *buffer)
116{
Tejun Heo182446d2013-08-08 20:11:24 -0400117 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa3aaabe22011-12-11 21:47:06 +0000118 unsigned long long val;
119 int ret = 0;
120
121 switch (cft->private) {
122 case RES_LIMIT:
123 /* see memcontrol.c */
124 ret = res_counter_memparse_write_strategy(buffer, &val);
125 if (ret)
126 break;
127 ret = tcp_update_limit(memcg, val);
128 break;
129 default:
130 ret = -EINVAL;
131 break;
132 }
133 return ret;
134}
135
136static u64 tcp_read_stat(struct mem_cgroup *memcg, int type, u64 default_val)
137{
Glauber Costa3aaabe22011-12-11 21:47:06 +0000138 struct cg_proto *cg_proto;
139
140 cg_proto = tcp_prot.proto_cgroup(memcg);
141 if (!cg_proto)
142 return default_val;
143
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700144 return res_counter_read_u64(&cg_proto->memory_allocated, type);
Glauber Costa3aaabe22011-12-11 21:47:06 +0000145}
146
Glauber Costa5a6dd342011-12-11 21:47:07 +0000147static u64 tcp_read_usage(struct mem_cgroup *memcg)
148{
Glauber Costa5a6dd342011-12-11 21:47:07 +0000149 struct cg_proto *cg_proto;
150
151 cg_proto = tcp_prot.proto_cgroup(memcg);
152 if (!cg_proto)
153 return atomic_long_read(&tcp_memory_allocated) << PAGE_SHIFT;
154
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700155 return res_counter_read_u64(&cg_proto->memory_allocated, RES_USAGE);
Glauber Costa5a6dd342011-12-11 21:47:07 +0000156}
157
Tejun Heo182446d2013-08-08 20:11:24 -0400158static u64 tcp_cgroup_read(struct cgroup_subsys_state *css, struct cftype *cft)
Glauber Costa3aaabe22011-12-11 21:47:06 +0000159{
Tejun Heo182446d2013-08-08 20:11:24 -0400160 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
Glauber Costa3aaabe22011-12-11 21:47:06 +0000161 u64 val;
162
163 switch (cft->private) {
164 case RES_LIMIT:
Sha Zhengju6de5a8b2013-09-12 15:13:47 -0700165 val = tcp_read_stat(memcg, RES_LIMIT, RES_COUNTER_MAX);
Glauber Costa3aaabe22011-12-11 21:47:06 +0000166 break;
Glauber Costa5a6dd342011-12-11 21:47:07 +0000167 case RES_USAGE:
168 val = tcp_read_usage(memcg);
169 break;
Glauber Costaffea59e2011-12-11 21:47:08 +0000170 case RES_FAILCNT:
Glauber Costa0850f0f2011-12-11 21:47:09 +0000171 case RES_MAX_USAGE:
172 val = tcp_read_stat(memcg, cft->private, 0);
Glauber Costaffea59e2011-12-11 21:47:08 +0000173 break;
Glauber Costa3aaabe22011-12-11 21:47:06 +0000174 default:
175 BUG();
176 }
177 return val;
178}
179
Tejun Heo182446d2013-08-08 20:11:24 -0400180static int tcp_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event)
Glauber Costaffea59e2011-12-11 21:47:08 +0000181{
182 struct mem_cgroup *memcg;
Glauber Costaffea59e2011-12-11 21:47:08 +0000183 struct cg_proto *cg_proto;
184
Tejun Heo182446d2013-08-08 20:11:24 -0400185 memcg = mem_cgroup_from_css(css);
Glauber Costaffea59e2011-12-11 21:47:08 +0000186 cg_proto = tcp_prot.proto_cgroup(memcg);
187 if (!cg_proto)
188 return 0;
Glauber Costaffea59e2011-12-11 21:47:08 +0000189
190 switch (event) {
Glauber Costa0850f0f2011-12-11 21:47:09 +0000191 case RES_MAX_USAGE:
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700192 res_counter_reset_max(&cg_proto->memory_allocated);
Glauber Costa0850f0f2011-12-11 21:47:09 +0000193 break;
Glauber Costaffea59e2011-12-11 21:47:08 +0000194 case RES_FAILCNT:
Eric W. Biederman2e685ca2013-10-19 16:26:19 -0700195 res_counter_reset_failcnt(&cg_proto->memory_allocated);
Glauber Costaffea59e2011-12-11 21:47:08 +0000196 break;
197 }
198
199 return 0;
200}
201
Tejun Heo676f7c82012-04-01 12:09:55 -0700202static struct cftype tcp_files[] = {
203 {
204 .name = "kmem.tcp.limit_in_bytes",
205 .write_string = tcp_cgroup_write,
206 .read_u64 = tcp_cgroup_read,
207 .private = RES_LIMIT,
208 },
209 {
210 .name = "kmem.tcp.usage_in_bytes",
211 .read_u64 = tcp_cgroup_read,
212 .private = RES_USAGE,
213 },
214 {
215 .name = "kmem.tcp.failcnt",
216 .private = RES_FAILCNT,
217 .trigger = tcp_cgroup_reset,
218 .read_u64 = tcp_cgroup_read,
219 },
220 {
221 .name = "kmem.tcp.max_usage_in_bytes",
222 .private = RES_MAX_USAGE,
223 .trigger = tcp_cgroup_reset,
224 .read_u64 = tcp_cgroup_read,
225 },
Tejun Heo6bc10342012-04-01 12:09:55 -0700226 { } /* terminate */
Tejun Heo676f7c82012-04-01 12:09:55 -0700227};
Tejun Heo6bc10342012-04-01 12:09:55 -0700228
229static int __init tcp_memcontrol_init(void)
230{
231 WARN_ON(cgroup_add_cftypes(&mem_cgroup_subsys, tcp_files));
232 return 0;
233}
234__initcall(tcp_memcontrol_init);