This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / include / net / request_sock.h
diff --git a/include/net/request_sock.h b/include/net/request_sock.h
new file mode 100644 (file)
index 0000000..c5d7f92
--- /dev/null
@@ -0,0 +1,261 @@
+/*
+ * NET         Generic infrastructure for Network protocols.
+ *
+ *             Definitions for request_sock 
+ *
+ * Authors:    Arnaldo Carvalho de Melo <acme@conectiva.com.br>
+ *
+ *             From code originally in include/net/tcp.h
+ *
+ *             This program is free software; you can redistribute it and/or
+ *             modify it under the terms of the GNU General Public License
+ *             as published by the Free Software Foundation; either version
+ *             2 of the License, or (at your option) any later version.
+ */
+#ifndef _REQUEST_SOCK_H
+#define _REQUEST_SOCK_H
+
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/types.h>
+
+#include <net/sock.h>
+
+struct request_sock;
+struct sk_buff;
+struct dst_entry;
+struct proto;
+
+struct request_sock_ops {
+       int             family;
+       kmem_cache_t    *slab;
+       int             obj_size;
+       int             (*rtx_syn_ack)(struct sock *sk,
+                                      struct request_sock *req,
+                                      struct dst_entry *dst);
+       void            (*send_ack)(struct sk_buff *skb,
+                                   struct request_sock *req);
+       void            (*send_reset)(struct sk_buff *skb);
+       void            (*destructor)(struct request_sock *req);
+};
+
+/* struct request_sock - mini sock to represent a connection request
+ */
+struct request_sock {
+       struct request_sock             *dl_next; /* Must be first member! */
+       u16                             mss;
+       u8                              retrans;
+       u8                              __pad;
+       /* The following two fields can be easily recomputed I think -AK */
+       u32                             window_clamp; /* window clamp at creation time */
+       u32                             rcv_wnd;          /* rcv_wnd offered first time */
+       u32                             ts_recent;
+       unsigned long                   expires;
+       struct request_sock_ops         *rsk_ops;
+       struct sock                     *sk;
+};
+
+static inline struct request_sock *reqsk_alloc(struct request_sock_ops *ops)
+{
+       struct request_sock *req = kmem_cache_alloc(ops->slab, SLAB_ATOMIC);
+
+       if (req != NULL)
+               req->rsk_ops = ops;
+
+       return req;
+}
+
+static inline void __reqsk_free(struct request_sock *req)
+{
+       kmem_cache_free(req->rsk_ops->slab, req);
+}
+
+static inline void reqsk_free(struct request_sock *req)
+{
+       req->rsk_ops->destructor(req);
+       __reqsk_free(req);
+}
+
+extern int sysctl_max_syn_backlog;
+
+/** struct listen_sock - listen state
+ *
+ * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
+ */
+struct listen_sock {
+       u8                      max_qlen_log;
+       /* 3 bytes hole, try to use */
+       int                     qlen;
+       int                     qlen_young;
+       int                     clock_hand;
+       u32                     hash_rnd;
+       u32                     nr_table_entries;
+       struct request_sock     *syn_table[0];
+};
+
+/** struct request_sock_queue - queue of request_socks
+ *
+ * @rskq_accept_head - FIFO head of established children
+ * @rskq_accept_tail - FIFO tail of established children
+ * @rskq_defer_accept - User waits for some data after accept()
+ * @syn_wait_lock - serializer
+ *
+ * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
+ * lock sock while browsing the listening hash (otherwise it's deadlock prone).
+ *
+ * This lock is acquired in read mode only from listening_get_next() seq_file
+ * op and it's acquired in write mode _only_ from code that is actively
+ * changing rskq_accept_head. All readers that are holding the master sock lock
+ * don't need to grab this lock in read mode too as rskq_accept_head. writes
+ * are always protected from the main sock lock.
+ */
+struct request_sock_queue {
+       struct request_sock     *rskq_accept_head;
+       struct request_sock     *rskq_accept_tail;
+       rwlock_t                syn_wait_lock;
+       u8                      rskq_defer_accept;
+       /* 3 bytes hole, try to pack */
+       struct listen_sock      *listen_opt;
+};
+
+extern int reqsk_queue_alloc(struct request_sock_queue *queue,
+                            const int nr_table_entries);
+
+static inline struct listen_sock *reqsk_queue_yank_listen_sk(struct request_sock_queue *queue)
+{
+       struct listen_sock *lopt;
+
+       write_lock_bh(&queue->syn_wait_lock);
+       lopt = queue->listen_opt;
+       queue->listen_opt = NULL;
+       write_unlock_bh(&queue->syn_wait_lock);
+
+       return lopt;
+}
+
+static inline void __reqsk_queue_destroy(struct request_sock_queue *queue)
+{
+       kfree(reqsk_queue_yank_listen_sk(queue));
+}
+
+extern void reqsk_queue_destroy(struct request_sock_queue *queue);
+
+static inline struct request_sock *
+       reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
+{
+       struct request_sock *req = queue->rskq_accept_head;
+
+       queue->rskq_accept_head = NULL;
+       return req;
+}
+
+static inline int reqsk_queue_empty(struct request_sock_queue *queue)
+{
+       return queue->rskq_accept_head == NULL;
+}
+
+static inline void reqsk_queue_unlink(struct request_sock_queue *queue,
+                                     struct request_sock *req,
+                                     struct request_sock **prev_req)
+{
+       write_lock(&queue->syn_wait_lock);
+       *prev_req = req->dl_next;
+       write_unlock(&queue->syn_wait_lock);
+}
+
+static inline void reqsk_queue_add(struct request_sock_queue *queue,
+                                  struct request_sock *req,
+                                  struct sock *parent,
+                                  struct sock *child)
+{
+       req->sk = child;
+       sk_acceptq_added(parent);
+
+       if (queue->rskq_accept_head == NULL)
+               queue->rskq_accept_head = req;
+       else
+               queue->rskq_accept_tail->dl_next = req;
+
+       queue->rskq_accept_tail = req;
+       req->dl_next = NULL;
+}
+
+static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
+{
+       struct request_sock *req = queue->rskq_accept_head;
+
+       BUG_TRAP(req != NULL);
+
+       queue->rskq_accept_head = req->dl_next;
+       if (queue->rskq_accept_head == NULL)
+               queue->rskq_accept_tail = NULL;
+
+       return req;
+}
+
+static inline struct sock *reqsk_queue_get_child(struct request_sock_queue *queue,
+                                                struct sock *parent)
+{
+       struct request_sock *req = reqsk_queue_remove(queue);
+       struct sock *child = req->sk;
+
+       BUG_TRAP(child != NULL);
+
+       sk_acceptq_removed(parent);
+       __reqsk_free(req);
+       return child;
+}
+
+static inline int reqsk_queue_removed(struct request_sock_queue *queue,
+                                     struct request_sock *req)
+{
+       struct listen_sock *lopt = queue->listen_opt;
+
+       if (req->retrans == 0)
+               --lopt->qlen_young;
+
+       return --lopt->qlen;
+}
+
+static inline int reqsk_queue_added(struct request_sock_queue *queue)
+{
+       struct listen_sock *lopt = queue->listen_opt;
+       const int prev_qlen = lopt->qlen;
+
+       lopt->qlen_young++;
+       lopt->qlen++;
+       return prev_qlen;
+}
+
+static inline int reqsk_queue_len(const struct request_sock_queue *queue)
+{
+       return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0;
+}
+
+static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
+{
+       return queue->listen_opt->qlen_young;
+}
+
+static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
+{
+       return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log;
+}
+
+static inline void reqsk_queue_hash_req(struct request_sock_queue *queue,
+                                       u32 hash, struct request_sock *req,
+                                       unsigned long timeout)
+{
+       struct listen_sock *lopt = queue->listen_opt;
+
+       req->expires = jiffies + timeout;
+       req->retrans = 0;
+       req->sk = NULL;
+       req->dl_next = lopt->syn_table[hash];
+
+       write_lock(&queue->syn_wait_lock);
+       lopt->syn_table[hash] = req;
+       write_unlock(&queue->syn_wait_lock);
+}
+
+#endif /* _REQUEST_SOCK_H */