net-snmp API分为两种，一种叫传统API(Traditional API)，一种叫单个API（Single API）。早期的neet-snmp没有考虑到多线程的问题，所有的会话共享同一个资源，这些就是传统API，后来支持多线程的就叫做单个API。Single API是线程安全的。
先熟悉术语，尤其要注意对Session的官方理解：
APDU Application Protocol Data Unit
Session Concept embodying the management of transacting SNMP APDUs.
下面先列出Traditional API和Single API之间主要的区别：
The functions in the following table are functionally equivalent,
with the exception of these behaviors:
- The Traditional API manages many sessions
- The Traditional API passes a struct snmp_session pointer,
and touches the Sessions list
- The Single API manages only one session
- The Single API passes an opaque pointer, and does not use Sessions list
再看看API之间的区别:
Traditional Single Comment
=========== ============== =======
snmp_sess_init snmp_sess_init Call before either open
snmp_open snmp_sess_open Single not on Sessions lists
snmp_sess_session Exposes snmp_session pointer
snmp_send snmp_sess_send Send one APDU
snmp_async_send snmp_sess_async_send Send one APDU with callback
snmp_select_info snmp_sess_select_info Which session(s) have input
snmp_read snmp_sess_read Read APDUs
snmp_timeout snmp_sess_timeout Check for timeout
snmp_close snmp_sess_close Single not on Sessions list
snmp_synch_response snmp_sess_synch_response Send/receive one APDU
snmp_error snmp_sess_error Get library,system errno
从上面可知snmp_session是贯穿始终的，有必要了解一下snmp_session,先看一下定义：
/** @struct snmp_session * The snmp session structure. */
struct snmp_session {
/* * Protocol-version independent fields */
/** snmp version */
long version;
/** Number of retries before timeout. */
int retries;
/** Number of uS until first timeout, then exponential backoff */
long timeout;
u_long flags;
struct snmp_session *subsession;
struct snmp_session *next;
/** name or address of default peer (may include transport specifier and/or port number) */
char *peername;
/** UDP port number of peer. (NO LONGER USED - USE peername INSTEAD) */
u_short remote_port;
/** My Domain name or dotted IP address, 0 for default */
char *localname;
/** My UDP port number, 0 for default, picked randomly */
u_short local_port;
/** * Authentication function or NULL if null authentication is used s */
u_char *(*authenticator) (u_char *, size_t *, u_char *, size_t);
/** Function to interpret incoming data */
netsnmp_callback callback;
/** * Pointer to data that the callback function may consider important */
void *callback_magic;
/** copy of system errno */
int s_errno;
/** copy of library errno */
int s_snmp_errno;
/** Session id - AgentX only */
long sessid;
/* * SNMPv1 & SNMPv2c fields */
/** community for outgoing requests. */
u_char *community;
/** Length of community name. */
size_t community_len;
/** Largest message to try to receive. */
size_t rcvMsgMaxSize;
/** Largest message to try to send. */
size_t sndMsgMaxSize;
/* * SNMPv3 fields */
/** are we the authoritative engine? */
u_char isAuthoritative;
/** authoritative snmpEngineID */
u_char *contextEngineID;
/** Length of contextEngineID */
size_t contextEngineIDLen;
/** initial engineBoots for remote engine */
u_int engineBoots;
/** initial engineTime for remote engine */
u_int engineTime;
/** authoritative contextName */
char *contextName;
/** Length of contextName */
size_t contextNameLen;
/** authoritative snmpEngineID */
u_char *securityEngineID;
/** Length of contextEngineID */
size_t securityEngineIDLen;
/** on behalf of this principal */
char *securityName;
/** Length of securityName. */
size_t securityNameLen;
/** auth protocol oid */
oid *securityAuthProto;
/** Length of auth protocol oid */
size_t securityAuthProtoLen;
/** Ku for auth protocol XXX */
u_char securityAuthKey[USM_AUTH_KU_LEN];
/** Length of Ku for auth protocol */
size_t securityAuthKeyLen;
/** Kul for auth protocol */
u_char *securityAuthLocalKey;
/** Length of Kul for auth protocol XXX */
size_t securityAuthLocalKeyLen;
/** priv protocol oid */
oid *securityPrivProto;
/** Length of priv protocol oid */
size_t securityPrivProtoLen;
/** Ku for privacy protocol XXX */
u_char securityPrivKey[USM_PRIV_KU_LEN];
/** Length of Ku for priv protocol */
size_t securityPrivKeyLen;
/** Kul for priv protocol */
u_char *securityPrivLocalKey;
/** Length of Kul for priv protocol XXX */
size_t securityPrivLocalKeyLen;
/** snmp security model, v1, v2c, usm */
int securityModel;
/** noAuthNoPriv, authNoPriv, authPriv */
int securityLevel;
/** target param name */
char *paramName;
/** * security module specific */
void *securityInfo;
/** * transport specific configuration */
struct netsnmp_container_s *transport_configuration;
/** * use as you want data * used by 'SNMP_FLAGS_RESP_CALLBACK' handling in the agent
* XXX: or should we add a new field into this structure? */
void *myvoid;
};
这里面记录了我们发一个snmp APDUs所需要的重要信息，比如本地机和远程机的端口，名字，community string, timeout值，errorno,retries等等，参考一下Readme.thread，可以动态的修改session里面的参数。
After the struct snmp_session is setup, the thread must call snmp_sess_open() to create an SNMP session. If at any time the thread must change the Session configuration, snmp_sess_session() returns the pointer to the internal configuration
structure (a struct snmp_session, copied from snmp_sess_open).
The thread can adjust parameters such as the session timeout or the community string with this returned struct snmp_session pointer. Changes to the remote or local port values have no effect on an opened Session.
1）分析采用的示例代码源自net-snmp官方教程中一片异步APP代码，详细可以点击这里
2）只列出了若干个API，更多的可以查看源码
3）这里分析的net-snmp源码版本为5.6.1
正文
if (!(hs->sess = snmp_open(&sess))) {
snmp_perror("snmp_open");
continue;
}
上面是snmp_open使用的演示代码，下面看看snmp_open里具体做了什么事情
netsnmp_session *
snmp_open(netsnmp_session *session)
{
struct session_list *slp;
slp = (struct session_list *) snmp_sess_open(session); //调用singleAPI创建
if (!slp) {
return NULL;
}
snmp_res_lock(MT_LIBRARY_ID, MT_LIB_SESSION); //这个函数是唬人的，根本没锁
slp->next = Sessions;//在snmp_api.c开头定义全局变量struct session_list *Sessions = NULL; /* MT_LIB_SESSION */
Sessions = slp; //添加到共享的Sessions链上
snmp_res_unlock(MT_LIBRARY_ID, MT_LIB_SESSION);//同样是唬人的s
return (slp->session);
}
snmp_open是传统API，这里可以看出所有的会话共享全局的Sessions链表。
snmp_res_lock为什么说是唬人的呢？我们明明在mt_suppotr.h和m_support.c里有看到支持跨平台的代码啊？注意看这两个文件里的宏编译之类NETSNMP_REENTRANT,可以在net-snmp-config.h里看到如下的注释：
/* add in recent resource lock functions (not complete) */
/* #undef NETSNMP_REENTRANT */
原来是还没有完全写完，OK，期待后续版本不用我们来自己写资源锁吧。
snmp_send
介绍：下面这些函数使用活动的会话发送PDUs
* snmp_send - traditional API, no callback
* snmp_async_send - traditional API, with callback(异步方式)
* snmp_sess_send - single session API, no callback
* snmp_sess_async_send - single session API, with callback(异步方式)
调用snmp_build来创建连续的包（即pdu），必要时采用会话的默认项设置某些pdu数据。
如果这个PDU有额外的响应，那就需要排列这个会话的外出请求并且存储这些回调向量。
通过会话向指定目标发送pdu。
如果成功，返回这个pdu请求的id，并且这个pdu被释放。如果失败，返回0，调用者必须调用snmmp_free_pdu释放资源。
snmp_send调用snmp_asyn_send，后者又调用snmp_sess_asyn_send，callback和cb_data参数都为NULL。
int snmp_async_send(netsnmp_session * session,
netsnmp_pdu *pdu, snmp_callback callback, void *cb_data)
{
void *sessp = snmp_sess_pointer(session);
return snmp_sess_async_send(sessp, pdu, callback, cb_data);
}
snmp_sess_pointer函数在全局变量Sessions里查找当前这个session，如果存在返回这个会话指针，否则返回NULL，snmp_error同时设置为SNMPERR_BAD_SESSION。
snmp_select_info
介绍：
输入：如果输入的timeout没有被定义，block设为1；如果输入的timeout被定义了，block设为0。
输出：如果输出的timeout没有被定义，block被视为1；如果输出的timeout被定义了，block被设为0。
上面的输入输出指定是参数timeout和block。
该函数的返回值为可以操作的socket数量，并且这些socket已经被选到了fdset里，供后续的select操作。
示例代码如下
int fds = 0, block = 1;
fd_set fdset;
struct timeval timeout;
FD_ZERO(&fdset);
snmp_select_info(&fds, &fdset, &timeout, &block);
fds = select(fds, &fdset, NULL, NULL, block ? NULL : &timeout);
if (fds < 0) {
perror("select failed");
exit(1);
}
if (fds) snmp_read(&fdset);
else snmp_timeout();
因为输入timeout没有定义，block为1，那么输出后timeout值为0，block值被设为0 s
这里需要注意的是是里面调用了netsnmp_large_fd_set这个结构，它的介绍如源码注释所说
/*
* Structure for holding a set of file descriptors, similar to fd_set.
*
* This structure however can hold so-called large file descriptors
* (>= FD_SETSIZE or 1024) on Unix systems or more than FD_SETSIZE (64)
* sockets on Windows systems.
*
* It is safe to allocate this structure on the stack.
*
* This structure must be initialized by calling netsnmp_large_fd_set_init()
* and must be cleaned up via netsnmp_large_fd_set_cleanup(). If this last
* function is not called this may result in a memory leak.
* The members of this structure are:
* lfs_setsize: maximum set size.
* lsf_setptr: points to lfs_set if lfs_setsize <= FD_SETSIZE, and otherwise
* to dynamically allocated memory.
* lfs_set: file descriptor / socket set data if lfs_setsize <= FD_SETSIZE.
*/
typedef struct netsnmp_large_fd_set_s {
unsigned lfs_setsize;
fd_set *lfs_setptr;
fd_set lfs_set;
} netsnmp_large_fd_set;
snmp_read
介绍：校验看看fd里面的集合是否属于snmp。每个socket的fd集合都有一个从它读取的包，同时snmp_parse被调用来接收包。pud的结果被传递给那个会话的回调例程。如果回调例程返回成功，这个pdu和它的请求就被删除掉。
snmp_timeout
介绍：当snmp_select_info里设定的超时期满的时候，这个函数应当被调用，但是它是幂等（idempotent）的，所以snmp_timeout能够被检验（大概一个cpu时间）。snmp_timeout检验查看每一个拥有对外请求的会话是否已经超时。如果它发现一个（或多个），并且那个pdu拥有多余可用的尝试次数，这个pud就构造一个新报文并且重新发送。如果没有多余的可用次数，这个会话的回调函数就会被用来通知用户超时了。