SYNOPSIS

#include <sys/types.h> #include <sys/socket.h> #include <netdb.h> #include <asr.h>
int
asr_run(struct asr_query *aq, struct asr_result *ar);
int
asr_run_sync(struct asr_query *aq, struct asr_result *ar);
void
asr_abort(struct asr_query *aq);
struct asr_query *
res_send_async(const unsigned char *pkt, int pktlen, void *asr);
struct asr_query *
res_query_async(const char *name, int class, int type, void *asr);
struct asr_query *
res_search_async(const char *name, int class, int type, void *asr);
struct asr_query *
getrrsetbyname_async(const char *hostname, unsigned int rdclass, unsigned int rdtype, unsigned int flags, void *asr);
struct asr_query *
gethostbyname_async(const char *name, void *asr);
struct asr_query *
gethostbyname2_async(const char *name, int af, void *asr);
struct asr_query *
gethostbyaddr_async(const void *addr, socklen_t len, int af, void *asr);
struct asr_query *
getnetbyname_async(const char *name, void *asr);
struct asr_query *
getnetbyaddr_async(in_addr_t net, int type, void *asr);
struct asr_query *
getaddrinfo_async(const char *hostname, const char *servname, const struct addrinfo *hints, void *asr);
struct asr_query *
getnameinfo_async(const struct sockaddr *sa, socklen_t salen, char *host, size_t hostlen, char *serv, size_t servlen, int flags, void *asr);

DESCRIPTION

The asr functions provide a simple interface for asynchronous address resolution and nameserver querying. They should be used in place of the classical resolver functions of libc when blocking is not desirable.

The principle of operation is as follows: All async requests are made against an asr context which basically defines a list of sources to query and a strategy to do so. The user creates a query through one of the dedicated functions, and gets a handle representing the internal query. A query is a state-machine that can be run to try to fulfill a particular request. This is done by calling in a generic API that performs the state transitions until it needs to give the control back to the user, either because a result is available, or because the next transition implies a blocking call (a file descriptor needs to be read from or written to). The user is responsible for dealing with the situation: either get the result, or wait until the fd conditions are met, and then call back into the resolving machinery when it is ready to proceed.

The asr_run() function drives the resolving process. It runs the asynchronous query represented by the handle until a result is available, or until it cannot continue without blocking. The results are returned to the user through the parameter, which must be a valid pointer to user allocated memory. is defined as:

struct asr_result {
        /* Fields set if the query is not done yet (asr_run returns 0) */
        int      ar_cond;       /* ASR_WANT_READ or ASR_WANT_WRITE */
        int      ar_fd;         /* the fd waiting for io condition */
        int      ar_timeout;    /* time to wait for in milliseconds */
        /* Error fields.  Depends on the query type. */
        int      ar_errno;
        int      ar_h_errno;
        int      ar_gai_errno;
        int      ar_rrset_errno;
        /* Result for res_*_async() calls */
        int      ar_count;      /* number of answers in the dns reply */
        int      ar_rcode;      /* response code in the dns reply */
        void    *ar_data;       /* raw reply packet (must be freed) */
        int      ar_datalen;    /* reply packet length */
        struct sockaddr_storage ar_ns; /* nameserver that responded */
        /* Result for other calls. Must be freed properly. */
        struct addrinfo  *ar_addrinfo;
        struct rrsetinfo *ar_rrsetinfo;
        struct hostent   *ar_hostent;
        struct netent    *ar_netent;
};

The function returns one of the following values:

0

The query cannot be processed further until a specific condition on a file descriptor becomes true. The following members of the structure are filled:

one of ASR_WANT_READ or ASR_WANT_WRITE,

the file descriptor waiting for an IO operation,

the amount of time to wait for in milliseconds.

The caller is expected to call asr_run() again once the condition holds or the timeout expires.

1

The query is completed. The members relevant to the actual async query type are set accordingly, including error conditions. In any case, the query is cleared and its handle is invalidated.

Note that although the query itself may fail (the error being properly reported in the structure), the asr_run() function itself cannot fail and it always preserves errno. The asr_run_sync() function is a wrapper around asr_run() that handles the read/write conditions, thus falling back to a blocking interface. It only returns 1. It also preserves errno. The asr_abort() function clears a running query. It can be called when the query is waiting on a file descriptor. Note that a completed query is already cleared when asr_run() returns, so asr_abort() must not be called in this case. The remaining functions are used to initiate different kinds of query on the resolver context. The specific operational details for each of them are described below. All functions return a handle to an internal query, or NULL if they could not allocate the necessary resources to initiate the query. All other errors (especially invalid parameters) are reported when calling asr_run(). They usually have the same interface as an existing resolver function, with an additional asr argument, which specifies the context to use for this request. For now, the argument must always be NULL, which will use the default context for the current thread. The res_send_async(), res_query_async() and res_search_async() functions are asynchronous versions of the standard libc resolver routines. Their interface is very similar, except that the response buffer is always allocated internally. The return value is found upon completion in the member of the response structure. In addition, the structure contains the address of the DNS server that sent the response, contains the code returned by the server in the DNS response packet, and contains the number of answers in the packet. If a response is received it is placed in a newly allocated buffer and returned as member. This buffer must be freed by the caller. On error, the and members are set accordingly. The getrrsetbyname_async() function is an asynchronous version of getrrsetbyname(3). Upon completion, the return code is found in and the address to the newly allocated result set is set in As for the blocking function, it must be freed by calling freerrset(3). The gethostbyname_async(), gethostbyname2_async() and gethostbyaddr_async() functions provide an asynchronous version of the network host entry functions. Upon completion, ar_h_errno is set and the resulting hostent address, if found, is set in the ar_hostent field. Note that unlike their blocking counterparts, these functions always return a pointer to newly allocated memory, which must be released by the caller using free(3). Similarly, the getnetbyname_async() and getnetbyaddr_async() functions provide an asynchronous version of the network entry functions. Upon completion, ar_h_errno is set and the resulting netent address, if found, is set in the ar_netent field. The memory there is also allocated for the request, and it must be freed by free(3). The getaddrinfo_async() function is an asynchronous version of the getaddrinfo(3) call. It provides a chain of addrinfo structures with all valid combinations of socket address for the given , and Those three parameters have the same meaning as for the blocking counterpart. Upon completion the return code is set in The member may also be set. On success, the member points to a newly allocated list of addrinfo. This list must be freed with freeaddrinfo(3).

WORKING WITH THREADS

This implementation of the asynchronous resolver interface is thread-safe and lock-free internally, but the following restriction applies: Two different threads must not create queries on the same context or run queries originating from the same context at the same time. If they want to do that, all calls must be protected by a mutex around that context.

It is generally not a problem since the main point of the asynchronous resolver is to multiplex queries within a single thread of control, so sharing a resolver among threads is not useful.

CAVEATS

This DNS resolver implementation doesn't support the EDNS0 protocol extension yet.