request_sock.c revision 7a1c8e5ab120a5f352e78bbc1fa5bb64e6f23639 
1/*
2 * NET		Generic infrastructure for Network protocols.
3 *
4 * Authors:	Arnaldo Carvalho de Melo <acme@conectiva.com.br>
5 *
6 * 		From code originally in include/net/tcp.h
7 *
8 *		This program is free software; you can redistribute it and/or
9 *		modify it under the terms of the GNU General Public License
10 *		as published by the Free Software Foundation; either version
11 *		2 of the License, or (at your option) any later version.
12 */
13
14#include <linux/module.h>
15#include <linux/random.h>
16#include <linux/slab.h>
17#include <linux/string.h>
18#include <linux/vmalloc.h>
19
20#include <net/request_sock.h>
21
22/*
23 * Maximum number of SYN_RECV sockets in queue per LISTEN socket.
24 * One SYN_RECV socket costs about 80bytes on a 32bit machine.
25 * It would be better to replace it with a global counter for all sockets
26 * but then some measure against one socket starving all other sockets
27 * would be needed.
28 *
29 * It was 128 by default. Experiments with real servers show, that
30 * it is absolutely not enough even at 100conn/sec. 256 cures most
31 * of problems. This value is adjusted to 128 for very small machines
32 * (<=32Mb of memory) and to 1024 on normal or better ones (>=256Mb).
33 * Note : Dont forget somaxconn that may limit backlog too.
34 */
35int sysctl_max_syn_backlog = 256;
36
37int reqsk_queue_alloc(struct request_sock_queue *queue,
38		      unsigned int nr_table_entries)
39{
40	size_t lopt_size = sizeof(struct listen_sock);
41	struct listen_sock *lopt;
42
43	nr_table_entries = min_t(u32, nr_table_entries, sysctl_max_syn_backlog);
44	nr_table_entries = max_t(u32, nr_table_entries, 8);
45	nr_table_entries = roundup_pow_of_two(nr_table_entries + 1);
46	lopt_size += nr_table_entries * sizeof(struct request_sock *);
47	if (lopt_size > PAGE_SIZE)
48		lopt = vzalloc(lopt_size);
49	else
50		lopt = kzalloc(lopt_size, GFP_KERNEL);
51	if (lopt == NULL)
52		return -ENOMEM;
53
54	for (lopt->max_qlen_log = 3;
55	     (1 << lopt->max_qlen_log) < nr_table_entries;
56	     lopt->max_qlen_log++);
57
58	get_random_bytes(&lopt->hash_rnd, sizeof(lopt->hash_rnd));
59	rwlock_init(&queue->syn_wait_lock);
60	queue->rskq_accept_head = NULL;
61	lopt->nr_table_entries = nr_table_entries;
62
63	write_lock_bh(&queue->syn_wait_lock);
64	queue->listen_opt = lopt;
65	write_unlock_bh(&queue->syn_wait_lock);
66
67	return 0;
68}
69
70void __reqsk_queue_destroy(struct request_sock_queue *queue)
71{
72	struct listen_sock *lopt;
73	size_t lopt_size;
74
75	/*
76	 * this is an error recovery path only
77	 * no locking needed and the lopt is not NULL
78	 */
79
80	lopt = queue->listen_opt;
81	lopt_size = sizeof(struct listen_sock) +
82		lopt->nr_table_entries * sizeof(struct request_sock *);
83
84	if (lopt_size > PAGE_SIZE)
85		vfree(lopt);
86	else
87		kfree(lopt);
88}
89
90static inline struct listen_sock *reqsk_queue_yank_listen_sk(
91		struct request_sock_queue *queue)
92{
93	struct listen_sock *lopt;
94
95	write_lock_bh(&queue->syn_wait_lock);
96	lopt = queue->listen_opt;
97	queue->listen_opt = NULL;
98	write_unlock_bh(&queue->syn_wait_lock);
99
100	return lopt;
101}
102
103void reqsk_queue_destroy(struct request_sock_queue *queue)
104{
105	/* make all the listen_opt local to us */
106	struct listen_sock *lopt = reqsk_queue_yank_listen_sk(queue);
107	size_t lopt_size = sizeof(struct listen_sock) +
108		lopt->nr_table_entries * sizeof(struct request_sock *);
109
110	if (lopt->qlen != 0) {
111		unsigned int i;
112
113		for (i = 0; i < lopt->nr_table_entries; i++) {
114			struct request_sock *req;
115
116			while ((req = lopt->syn_table[i]) != NULL) {
117				lopt->syn_table[i] = req->dl_next;
118				lopt->qlen--;
119				reqsk_free(req);
120			}
121		}
122	}
123
124	WARN_ON(lopt->qlen != 0);
125	if (lopt_size > PAGE_SIZE)
126		vfree(lopt);
127	else
128		kfree(lopt);
129}
130
131