diff options
Diffstat (limited to 'include/net/tcp.h')
-rw-r--r-- | include/net/tcp.h | 120 |
1 files changed, 2 insertions, 118 deletions
diff --git a/include/net/tcp.h b/include/net/tcp.h index 6c9f6f7cab5c..ff5d30ac2b06 100644 --- a/include/net/tcp.h +++ b/include/net/tcp.h | |||
@@ -30,6 +30,7 @@ | |||
30 | #include <linux/slab.h> | 30 | #include <linux/slab.h> |
31 | #include <linux/cache.h> | 31 | #include <linux/cache.h> |
32 | #include <linux/percpu.h> | 32 | #include <linux/percpu.h> |
33 | #include <net/inet_hashtables.h> | ||
33 | #include <net/checksum.h> | 34 | #include <net/checksum.h> |
34 | #include <net/request_sock.h> | 35 | #include <net/request_sock.h> |
35 | #include <net/sock.h> | 36 | #include <net/sock.h> |
@@ -40,101 +41,6 @@ | |||
40 | #endif | 41 | #endif |
41 | #include <linux/seq_file.h> | 42 | #include <linux/seq_file.h> |
42 | 43 | ||
43 | /* This is for all connections with a full identity, no wildcards. | ||
44 | * New scheme, half the table is for TIME_WAIT, the other half is | ||
45 | * for the rest. I'll experiment with dynamic table growth later. | ||
46 | */ | ||
47 | struct inet_ehash_bucket { | ||
48 | rwlock_t lock; | ||
49 | struct hlist_head chain; | ||
50 | } __attribute__((__aligned__(8))); | ||
51 | |||
52 | /* This is for listening sockets, thus all sockets which possess wildcards. */ | ||
53 | #define INET_LHTABLE_SIZE 32 /* Yes, really, this is all you need. */ | ||
54 | |||
55 | /* There are a few simple rules, which allow for local port reuse by | ||
56 | * an application. In essence: | ||
57 | * | ||
58 | * 1) Sockets bound to different interfaces may share a local port. | ||
59 | * Failing that, goto test 2. | ||
60 | * 2) If all sockets have sk->sk_reuse set, and none of them are in | ||
61 | * TCP_LISTEN state, the port may be shared. | ||
62 | * Failing that, goto test 3. | ||
63 | * 3) If all sockets are bound to a specific inet_sk(sk)->rcv_saddr local | ||
64 | * address, and none of them are the same, the port may be | ||
65 | * shared. | ||
66 | * Failing this, the port cannot be shared. | ||
67 | * | ||
68 | * The interesting point, is test #2. This is what an FTP server does | ||
69 | * all day. To optimize this case we use a specific flag bit defined | ||
70 | * below. As we add sockets to a bind bucket list, we perform a | ||
71 | * check of: (newsk->sk_reuse && (newsk->sk_state != TCP_LISTEN)) | ||
72 | * As long as all sockets added to a bind bucket pass this test, | ||
73 | * the flag bit will be set. | ||
74 | * The resulting situation is that tcp_v[46]_verify_bind() can just check | ||
75 | * for this flag bit, if it is set and the socket trying to bind has | ||
76 | * sk->sk_reuse set, we don't even have to walk the owners list at all, | ||
77 | * we return that it is ok to bind this socket to the requested local port. | ||
78 | * | ||
79 | * Sounds like a lot of work, but it is worth it. In a more naive | ||
80 | * implementation (ie. current FreeBSD etc.) the entire list of ports | ||
81 | * must be walked for each data port opened by an ftp server. Needless | ||
82 | * to say, this does not scale at all. With a couple thousand FTP | ||
83 | * users logged onto your box, isn't it nice to know that new data | ||
84 | * ports are created in O(1) time? I thought so. ;-) -DaveM | ||
85 | */ | ||
86 | struct inet_bind_bucket { | ||
87 | unsigned short port; | ||
88 | signed short fastreuse; | ||
89 | struct hlist_node node; | ||
90 | struct hlist_head owners; | ||
91 | }; | ||
92 | |||
93 | #define inet_bind_bucket_for_each(tb, node, head) \ | ||
94 | hlist_for_each_entry(tb, node, head, node) | ||
95 | |||
96 | struct inet_bind_hashbucket { | ||
97 | spinlock_t lock; | ||
98 | struct hlist_head chain; | ||
99 | }; | ||
100 | |||
101 | struct inet_hashinfo { | ||
102 | /* This is for sockets with full identity only. Sockets here will | ||
103 | * always be without wildcards and will have the following invariant: | ||
104 | * | ||
105 | * TCP_ESTABLISHED <= sk->sk_state < TCP_CLOSE | ||
106 | * | ||
107 | * First half of the table is for sockets not in TIME_WAIT, second half | ||
108 | * is for TIME_WAIT sockets only. | ||
109 | */ | ||
110 | struct inet_ehash_bucket *ehash; | ||
111 | |||
112 | /* Ok, let's try this, I give up, we do need a local binding | ||
113 | * TCP hash as well as the others for fast bind/connect. | ||
114 | */ | ||
115 | struct inet_bind_hashbucket *bhash; | ||
116 | |||
117 | int bhash_size; | ||
118 | int ehash_size; | ||
119 | |||
120 | /* All sockets in TCP_LISTEN state will be in here. This is the only | ||
121 | * table where wildcard'd TCP sockets can exist. Hash function here | ||
122 | * is just local port number. | ||
123 | */ | ||
124 | struct hlist_head listening_hash[INET_LHTABLE_SIZE]; | ||
125 | |||
126 | /* All the above members are written once at bootup and | ||
127 | * never written again _or_ are predominantly read-access. | ||
128 | * | ||
129 | * Now align to a new cache line as all the following members | ||
130 | * are often dirty. | ||
131 | */ | ||
132 | rwlock_t lhash_lock ____cacheline_aligned; | ||
133 | atomic_t lhash_users; | ||
134 | wait_queue_head_t lhash_wait; | ||
135 | spinlock_t portalloc_lock; | ||
136 | }; | ||
137 | |||
138 | extern struct inet_hashinfo tcp_hashinfo; | 44 | extern struct inet_hashinfo tcp_hashinfo; |
139 | #define tcp_ehash (tcp_hashinfo.ehash) | 45 | #define tcp_ehash (tcp_hashinfo.ehash) |
140 | #define tcp_bhash (tcp_hashinfo.bhash) | 46 | #define tcp_bhash (tcp_hashinfo.bhash) |
@@ -147,19 +53,8 @@ extern struct inet_hashinfo tcp_hashinfo; | |||
147 | #define tcp_portalloc_lock (tcp_hashinfo.portalloc_lock) | 53 | #define tcp_portalloc_lock (tcp_hashinfo.portalloc_lock) |
148 | 54 | ||
149 | extern kmem_cache_t *tcp_bucket_cachep; | 55 | extern kmem_cache_t *tcp_bucket_cachep; |
150 | extern struct inet_bind_bucket * | ||
151 | inet_bind_bucket_create(kmem_cache_t *cachep, | ||
152 | struct inet_bind_hashbucket *head, | ||
153 | const unsigned short snum); | ||
154 | extern void inet_bind_bucket_destroy(kmem_cache_t *cachep, | ||
155 | struct inet_bind_bucket *tb); | ||
156 | extern int tcp_port_rover; | ||
157 | 56 | ||
158 | /* These are AF independent. */ | 57 | extern int tcp_port_rover; |
159 | static inline int inet_bhashfn(const __u16 lport, const int bhash_size) | ||
160 | { | ||
161 | return lport & (bhash_size - 1); | ||
162 | } | ||
163 | 58 | ||
164 | extern void tcp_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, | 59 | extern void tcp_bind_hash(struct sock *sk, struct inet_bind_bucket *tb, |
165 | unsigned short snum); | 60 | unsigned short snum); |
@@ -359,17 +254,6 @@ extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw); | |||
359 | ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \ | 254 | ipv6_addr_equal(&inet6_sk(__sk)->rcv_saddr, (__daddr)) && \ |
360 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) | 255 | (!((__sk)->sk_bound_dev_if) || ((__sk)->sk_bound_dev_if == (__dif)))) |
361 | 256 | ||
362 | /* These can have wildcards, don't try too hard. */ | ||
363 | static inline int inet_lhashfn(const unsigned short num) | ||
364 | { | ||
365 | return num & (INET_LHTABLE_SIZE - 1); | ||
366 | } | ||
367 | |||
368 | static inline int inet_sk_listen_hashfn(const struct sock *sk) | ||
369 | { | ||
370 | return inet_lhashfn(inet_sk(sk)->num); | ||
371 | } | ||
372 | |||
373 | #define MAX_TCP_HEADER (128 + MAX_HEADER) | 257 | #define MAX_TCP_HEADER (128 + MAX_HEADER) |
374 | 258 | ||
375 | /* | 259 | /* |