2 #include <net/tcp_memcontrol.h>
5 #include <linux/nsproxy.h>
6 #include <linux/memcontrol.h>
7 #include <linux/module.h>
9 int tcp_init_cgroup(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
12 * The root cgroup does not use page_counters, but rather,
13 * rely on the data already collected by the network
16 struct mem_cgroup *parent = parent_mem_cgroup(memcg);
17 struct page_counter *counter_parent = NULL;
18 struct cg_proto *cg_proto, *parent_cg;
20 cg_proto = tcp_prot.proto_cgroup(memcg);
24 cg_proto->sysctl_mem[0] = sysctl_tcp_mem[0];
25 cg_proto->sysctl_mem[1] = sysctl_tcp_mem[1];
26 cg_proto->sysctl_mem[2] = sysctl_tcp_mem[2];
27 cg_proto->memory_pressure = 0;
28 cg_proto->memcg = memcg;
30 parent_cg = tcp_prot.proto_cgroup(parent);
32 counter_parent = &parent_cg->memory_allocated;
34 page_counter_init(&cg_proto->memory_allocated, counter_parent);
35 percpu_counter_init(&cg_proto->sockets_allocated, 0, GFP_KERNEL);
39 EXPORT_SYMBOL(tcp_init_cgroup);
41 void tcp_destroy_cgroup(struct mem_cgroup *memcg)
43 struct cg_proto *cg_proto;
45 cg_proto = tcp_prot.proto_cgroup(memcg);
49 percpu_counter_destroy(&cg_proto->sockets_allocated);
51 EXPORT_SYMBOL(tcp_destroy_cgroup);
53 static int tcp_update_limit(struct mem_cgroup *memcg, unsigned long nr_pages)
55 struct cg_proto *cg_proto;
59 cg_proto = tcp_prot.proto_cgroup(memcg);
63 ret = page_counter_limit(&cg_proto->memory_allocated, nr_pages);
67 for (i = 0; i < 3; i++)
68 cg_proto->sysctl_mem[i] = min_t(long, nr_pages,
71 if (nr_pages == PAGE_COUNTER_MAX)
72 clear_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
75 * The active bit needs to be written after the static_key
76 * update. This is what guarantees that the socket activation
77 * function is the last one to run. See sock_update_memcg() for
78 * details, and note that we don't mark any socket as belonging
79 * to this memcg until that flag is up.
81 * We need to do this, because static_keys will span multiple
82 * sites, but we can't control their order. If we mark a socket
83 * as accounted, but the accounting functions are not patched in
84 * yet, we'll lose accounting.
86 * We never race with the readers in sock_update_memcg(),
87 * because when this value change, the code to process it is not
90 * The activated bit is used to guarantee that no two writers
91 * will do the update in the same memcg. Without that, we can't
92 * properly shutdown the static key.
94 if (!test_and_set_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags))
95 static_key_slow_inc(&memcg_socket_limit_enabled);
96 set_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
109 static DEFINE_MUTEX(tcp_limit_mutex);
111 static ssize_t tcp_cgroup_write(struct kernfs_open_file *of,
112 char *buf, size_t nbytes, loff_t off)
114 struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of));
115 unsigned long nr_pages;
120 switch (of_cft(of)->private) {
122 /* see memcontrol.c */
123 ret = page_counter_memparse(buf, "-1", &nr_pages);
126 mutex_lock(&tcp_limit_mutex);
127 ret = tcp_update_limit(memcg, nr_pages);
128 mutex_unlock(&tcp_limit_mutex);
134 return ret ?: nbytes;
137 static u64 tcp_cgroup_read(struct cgroup_subsys_state *css, struct cftype *cft)
139 struct mem_cgroup *memcg = mem_cgroup_from_css(css);
140 struct cg_proto *cg_proto = tcp_prot.proto_cgroup(memcg);
143 switch (cft->private) {
146 return PAGE_COUNTER_MAX;
147 val = cg_proto->memory_allocated.limit;
152 val = atomic_long_read(&tcp_memory_allocated);
154 val = page_counter_read(&cg_proto->memory_allocated);
160 val = cg_proto->memory_allocated.failcnt;
165 val = cg_proto->memory_allocated.watermark;
174 static ssize_t tcp_cgroup_reset(struct kernfs_open_file *of,
175 char *buf, size_t nbytes, loff_t off)
177 struct mem_cgroup *memcg;
178 struct cg_proto *cg_proto;
180 memcg = mem_cgroup_from_css(of_css(of));
181 cg_proto = tcp_prot.proto_cgroup(memcg);
185 switch (of_cft(of)->private) {
187 page_counter_reset_watermark(&cg_proto->memory_allocated);
190 cg_proto->memory_allocated.failcnt = 0;
197 static struct cftype tcp_files[] = {
199 .name = "kmem.tcp.limit_in_bytes",
200 .write = tcp_cgroup_write,
201 .read_u64 = tcp_cgroup_read,
202 .private = RES_LIMIT,
205 .name = "kmem.tcp.usage_in_bytes",
206 .read_u64 = tcp_cgroup_read,
207 .private = RES_USAGE,
210 .name = "kmem.tcp.failcnt",
211 .private = RES_FAILCNT,
212 .write = tcp_cgroup_reset,
213 .read_u64 = tcp_cgroup_read,
216 .name = "kmem.tcp.max_usage_in_bytes",
217 .private = RES_MAX_USAGE,
218 .write = tcp_cgroup_reset,
219 .read_u64 = tcp_cgroup_read,
224 static int __init tcp_memcontrol_init(void)
226 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, tcp_files));
229 __initcall(tcp_memcontrol_init);