[NETFILTER] nfnetlink_log: add sequence numbers for log events
authorHarald Welte <laforge@netfilter.org>
Tue, 21 Mar 2006 01:15:11 +0000 (17:15 -0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 21 Mar 2006 01:15:11 +0000 (17:15 -0800)
By using a sequence number for every logged netfilter event, we can
determine from userspace whether logging information was lots somewhere
downstream.

The user has a choice of either having per-instance local sequence
counters, or using a global sequence counter, or both.

Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/linux/netfilter/nfnetlink_log.h
net/netfilter/nfnetlink_log.c

index b04b038..a7497c7 100644 (file)
@@ -47,6 +47,8 @@ enum nfulnl_attr_type {
        NFULA_PAYLOAD,                  /* opaque data payload */
        NFULA_PREFIX,                   /* string prefix */
        NFULA_UID,                      /* user id of socket */
+       NFULA_SEQ,                      /* instance-local sequence number */
+       NFULA_SEQ_GLOBAL,               /* global sequence number */
 
        __NFULA_MAX
 };
@@ -77,6 +79,7 @@ enum nfulnl_attr_config {
        NFULA_CFG_NLBUFSIZ,             /* u_int32_t buffer size */
        NFULA_CFG_TIMEOUT,              /* u_int32_t in 1/100 s */
        NFULA_CFG_QTHRESH,              /* u_int32_t */
+       NFULA_CFG_FLAGS,                /* u_int16_t */
        __NFULA_CFG_MAX
 };
 #define NFULA_CFG_MAX (__NFULA_CFG_MAX -1)
@@ -85,4 +88,7 @@ enum nfulnl_attr_config {
 #define NFULNL_COPY_META       0x01
 #define NFULNL_COPY_PACKET     0x02
 
+#define NFULNL_CFG_F_SEQ       0x0001
+#define NFULNL_CFG_F_SEQ_GLOBAL        0x0002
+
 #endif /* _NFNETLINK_LOG_H */
index 3b3c781..54cbbaa 100644 (file)
  * it under the terms of the GNU General Public License version 2 as
  * published by the Free Software Foundation.
  *
+ * 2006-01-26 Harald Welte <laforge@netfilter.org>
+ *     - Add optional local and global sequence number to detect lost
+ *       events from userspace
+ *
  */
 #include <linux/module.h>
 #include <linux/skbuff.h>
@@ -68,11 +72,14 @@ struct nfulnl_instance {
        unsigned int nlbufsiz;          /* netlink buffer allocation size */
        unsigned int qthreshold;        /* threshold of the queue */
        u_int32_t copy_range;
+       u_int32_t seq;                  /* instance-local sequential counter */
        u_int16_t group_num;            /* number of this queue */
+       u_int16_t flags;
        u_int8_t copy_mode;     
 };
 
 static DEFINE_RWLOCK(instances_lock);
+static atomic_t global_seq;
 
 #define INSTANCE_BUCKETS       16
 static struct hlist_head instance_table[INSTANCE_BUCKETS];
@@ -310,6 +317,16 @@ nfulnl_set_qthresh(struct nfulnl_instance *inst, u_int32_t qthresh)
        return 0;
 }
 
+static int
+nfulnl_set_flags(struct nfulnl_instance *inst, u_int16_t flags)
+{
+       spin_lock_bh(&inst->lock);
+       inst->flags = ntohs(flags);
+       spin_unlock_bh(&inst->lock);
+
+       return 0;
+}
+
 static struct sk_buff *nfulnl_alloc_skb(unsigned int inst_size, 
                                        unsigned int pkt_size)
 {
@@ -377,6 +394,8 @@ static void nfulnl_timer(unsigned long data)
        spin_unlock_bh(&inst->lock);
 }
 
+/* This is an inline function, we don't really care about a long
+ * list of arguments */
 static inline int 
 __build_packet_message(struct nfulnl_instance *inst,
                        const struct sk_buff *skb, 
@@ -515,6 +534,17 @@ __build_packet_message(struct nfulnl_instance *inst,
                        read_unlock_bh(&skb->sk->sk_callback_lock);
        }
 
+       /* local sequence number */
+       if (inst->flags & NFULNL_CFG_F_SEQ) {
+               tmp_uint = htonl(inst->seq++);
+               NFA_PUT(inst->skb, NFULA_SEQ, sizeof(tmp_uint), &tmp_uint);
+       }
+       /* global sequence number */
+       if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL) {
+               tmp_uint = atomic_inc_return(&global_seq);
+               NFA_PUT(inst->skb, NFULA_SEQ_GLOBAL, sizeof(tmp_uint), &tmp_uint);
+       }
+
        if (data_len) {
                struct nfattr *nfa;
                int size = NFA_LENGTH(data_len);
@@ -607,6 +637,11 @@ nfulnl_log_packet(unsigned int pf,
 
        spin_lock_bh(&inst->lock);
 
+       if (inst->flags & NFULNL_CFG_F_SEQ)
+               size += NFA_SPACE(sizeof(u_int32_t));
+       if (inst->flags & NFULNL_CFG_F_SEQ_GLOBAL)
+               size += NFA_SPACE(sizeof(u_int32_t));
+
        qthreshold = inst->qthreshold;
        /* per-rule qthreshold overrides per-instance */
        if (qthreshold > li->u.ulog.qthreshold)
@@ -736,10 +771,14 @@ static const int nfula_min[NFULA_MAX] = {
        [NFULA_TIMESTAMP-1]     = sizeof(struct nfulnl_msg_packet_timestamp),
        [NFULA_IFINDEX_INDEV-1] = sizeof(u_int32_t),
        [NFULA_IFINDEX_OUTDEV-1]= sizeof(u_int32_t),
+       [NFULA_IFINDEX_PHYSINDEV-1]     = sizeof(u_int32_t),
+       [NFULA_IFINDEX_PHYSOUTDEV-1]    = sizeof(u_int32_t),
        [NFULA_HWADDR-1]        = sizeof(struct nfulnl_msg_packet_hw),
        [NFULA_PAYLOAD-1]       = 0,
        [NFULA_PREFIX-1]        = 0,
        [NFULA_UID-1]           = sizeof(u_int32_t),
+       [NFULA_SEQ-1]           = sizeof(u_int32_t),
+       [NFULA_SEQ_GLOBAL-1]    = sizeof(u_int32_t),
 };
 
 static const int nfula_cfg_min[NFULA_CFG_MAX] = {
@@ -748,6 +787,7 @@ static const int nfula_cfg_min[NFULA_CFG_MAX] = {
        [NFULA_CFG_TIMEOUT-1]   = sizeof(u_int32_t),
        [NFULA_CFG_QTHRESH-1]   = sizeof(u_int32_t),
        [NFULA_CFG_NLBUFSIZ-1]  = sizeof(u_int32_t),
+       [NFULA_CFG_FLAGS-1]     = sizeof(u_int16_t),
 };
 
 static int
@@ -859,6 +899,12 @@ nfulnl_recv_config(struct sock *ctnl, struct sk_buff *skb,
                nfulnl_set_qthresh(inst, ntohl(qthresh));
        }
 
+       if (nfula[NFULA_CFG_FLAGS-1]) {
+               u_int16_t flags =
+                       *(u_int16_t *)NFA_DATA(nfula[NFULA_CFG_FLAGS-1]);
+               nfulnl_set_flags(inst, ntohl(flags));
+       }
+
 out_put:
        instance_put(inst);
        return ret;