auto import from //branches/cupcake/...@132276

4 files changed, 1227 insertions, 301 deletions

diff --git a/libc/netbsd/gethnamaddr.c b/libc/netbsd/gethnamaddr.c
index e6f919e..1c219b2 100644
--- a/libc/netbsd/gethnamaddr.c
+++ b/libc/netbsd/gethnamaddr.c
@@ -626,37 +626,12 @@ gethostbyname_internal(const char *name, int af, res_state res)
 				break;
 		}
 
-#ifdef ANDROID_CHANGES
-        cache = __get_res_cache();
-        if (cache != NULL) {
-            hp = _resolv_cache_lookup( cache, name, af );
-            if (hp == _RESOLV_HOSTENT_NONE) {
-                h_errno = HOST_NOT_FOUND;
-                return NULL;
-            }
-            if (hp != NULL) {
-                h_errno = NETDB_SUCCESS;
-                return hp;
-            }
-        }
-#endif
-
 	hp = NULL;
 	h_errno = NETDB_INTERNAL;
 	if (nsdispatch(&hp, dtab, NSDB_HOSTS, "gethostbyname",
 	    default_dns_files, name, strlen(name), af) != NS_SUCCESS) {
-#ifdef ANDROID_CHANGES
-                /* cache negative DNS entry */
-                if (h_errno == HOST_NOT_FOUND && cache != NULL)
-                    _resolv_cache_add( cache, name, af, _RESOLV_HOSTENT_NONE );
-#endif
 		return NULL;
         }
-#ifdef ANDROID_CHANGES
-    if (cache != NULL) {
-        _resolv_cache_add( cache, name, af, hp );
-    }
-#endif
 	h_errno = NETDB_SUCCESS;
 	return hp;
 }
diff --git a/libc/netbsd/resolv/res_cache.c b/libc/netbsd/resolv/res_cache.c
index 210dfdd..2c912de 100644
--- a/libc/netbsd/resolv/res_cache.c
+++ b/libc/netbsd/resolv/res_cache.c
@@ -28,30 +28,104 @@
 
 #include "resolv_cache.h"
 #include <stdlib.h>
+#include <string.h>
 #include <time.h>
 #include "pthread.h"
 
-/* this code implements a small DNS resolver cache for all gethostbyname* functions
+/* This code implements a small and *simple* DNS resolver cache.
  *
- * the cache is shared between all threads of the current process, but the result of
- * a succesful lookup is always copied to a thread-local variable to honor the persistence
- * rules of the gethostbyname*() APIs.
+ * It is only used to cache DNS answers for a maximum of CONFIG_SECONDS seconds
+ * in order to reduce DNS traffic. It is not supposed to be a full DNS cache,
+ * since we plan to implement that in the future in a dedicated process running
+ * on the system.
+ *
+ * Note that its design is kept simple very intentionally, i.e.:
+ *
+ *  - it takes raw DNS query packet data as input, and returns raw DNS
+ *    answer packet data as output
+ *
+ *    (this means that two similar queries that encode the DNS name
+ *     differently will be treated distinctly).
+ *
+ *  - the TTLs of answer RRs are ignored. our DNS resolver library does not use
+ *    them anyway, but it means that records with a TTL smaller than
+ *    CONFIG_SECONDS will be kept in the cache anyway.
+ *
+ *    this is bad, but we absolutely want to avoid parsing the answer packets
+ *    (and should be solved by the later full DNS cache process).
+ *
+ *  - the implementation is just a (query-data) => (answer-data) hash table
+ *    with a trivial least-recently-used expiration policy.
+ *
+ * Doing this keeps the code simple and avoids to deal with a lot of things
+ * that a full DNS cache is expected to do.
+ *
+ * The API is also very simple:
+ *
+ *   - the client calls _resolv_cache_get() to obtain a handle to the cache.
+ *     this will initialize the cache on first usage. the result can be NULL
+ *     if the cache is disabled.
+ *
+ *   - the client calls _resolv_cache_lookup() before performing a query
+ *
+ *     if the function returns RESOLV_CACHE_FOUND, a copy of the answer data
+ *     has been copied into the client-provided answer buffer.
+ *
+ *     if the function returns RESOLV_CACHE_NOTFOUND, the client should perform
+ *     a request normally, *then* call _resolv_cache_add() to add the received
+ *     answer to the cache.
+ *
+ *     if the function returns RESOLV_CACHE_UNSUPPORTED, the client should
+ *     perform a request normally, and *not* call _resolv_cache_add()
+ *
+ *     note that RESOLV_CACHE_UNSUPPORTED is also returned if the answer buffer
+ *     is too short to accomodate the cached result.
+ *
+ *  - when network settings change, the cache must be flushed since the list
+ *    of DNS servers probably changed. this is done by calling
+ *    _resolv_cache_reset()
+ *
+ *    the parameter to this function must be an ever-increasing generation
+ *    number corresponding to the current network settings state.
+ *
+ *    This is done because several threads could detect the same network
+ *    settings change (but at different times) and will all end up calling the
+ *    same function. Comparing with the last used generation number ensures
+ *    that the cache is only flushed once per network change.
  */
 
-/* the name of an environment variable that will be checked the first time this code is called
- * if its value is "0", then the resolver cache is disabled.
+/* the name of an environment variable that will be checked the first time
+ * this code is called if its value is "0", then the resolver cache is
+ * disabled.
  */
 #define  CONFIG_ENV  "BIONIC_DNSCACHE"
 
-/* entries older than CONFIG_SECONDS seconds are always discarded. otherwise we could keep
- * stale addresses in the cache and be oblivious to DNS server changes
+/* entries older than CONFIG_SECONDS seconds are always discarded.
  */
 #define  CONFIG_SECONDS    (60*10)    /* 10 minutes */
 
-/* maximum number of entries kept in the cache. be frugal, this is the C library
- * this MUST BE A POWER OF 2
+/* maximum number of entries kept in the cache. This value has been
+ * determined by browsing through various sites and counting the number
+ * of corresponding requests. Keep in mind that our framework is currently
+ * performing two requests per name lookup (one for IPv4, the other for IPv6)
+ *
+ *    www.google.com      4
+ *    www.ysearch.com     6
+ *    www.amazon.com      8
+ *    www.nytimes.com     22
+ *    www.espn.com        28
+ *    www.msn.com         28
+ *    www.lemonde.fr      35
+ *
+ * (determined in 2009-2-17 from Paris, France, results may vary depending
+ *  on location)
+ *
+ * most high-level websites use lots of media/ad servers with different names
+ * but these are generally reused when browsing through the site.
+ *
+ * As such, a valud of 64 should be relatively conformtable at the moment.
  */
-#define  CONFIG_MAX_ENTRIES   128
+#define  CONFIG_MAX_ENTRIES    64
 
 /****************************************************************************/
 /****************************************************************************/
@@ -61,42 +135,204 @@
 /****************************************************************************/
 /****************************************************************************/
 
-#define  DEBUG  0
+/* set to 1 to debug cache operations */
+#define  DEBUG       0
+
+/* set to 1 to debug query data */
+#define  DEBUG_DATA  0
 
 #if DEBUG
-#include <fcntl.h>
-#include <errno.h>
-#include <stdarg.h>
+#  include <logd.h>
+#  define  XLOG(...)   \
+    __libc_android_log_print(ANDROID_LOG_DEBUG,"libc",__VA_ARGS__)
+
 #include <stdio.h>
-static void
-xlog( const char*  fmt, ... )
+#include <stdarg.h>
+
+/** BOUNDED BUFFER FORMATTING
+ **/
+
+/* technical note:
+ *
+ *   the following debugging routines are used to append data to a bounded
+ *   buffer they take two parameters that are:
+ *
+ *   - p : a pointer to the current cursor position in the buffer
+ *         this value is initially set to the buffer's address.
+ *
+ *   - end : the address of the buffer's limit, i.e. of the first byte
+ *           after the buffer. this address should never be touched.
+ *
+ *           IMPORTANT: it is assumed that end > buffer_address, i.e.
+ *                      that the buffer is at least one byte.
+ *
+ *   the _bprint_() functions return the new value of 'p' after the data
+ *   has been appended, and also ensure the following:
+ *
+ *   - the returned value will never be strictly greater than 'end'
+ *
+ *   - a return value equal to 'end' means that truncation occured
+ *     (in which case, end[-1] will be set to 0)
+ *
+ *   - after returning from a _bprint_() function, the content of the buffer
+ *     is always 0-terminated, even in the event of truncation.
+ *
+ *  these conventions allow you to call _bprint_ functions multiple times and
+ *  only check for truncation at the end of the sequence, as in:
+ *
+ *     char  buff[1000], *p = buff, *end = p + sizeof(buff);
+ *
+ *     p = _bprint_c(p, end, '"');
+ *     p = _bprint_s(p, end, my_string);
+ *     p = _bprint_c(p, end, '"');
+ *
+ *     if (p >= end) {
+ *        // buffer was too small
+ *     }
+ *
+ *     printf( "%s", buff );
+ */
+
+/* add a char to a bounded buffer */
+static char*
+_bprint_c( char*  p, char*  end, int  c )
+{
+    if (p < end) {
+        if (p+1 == end)
+            *p++ = 0;
+        else {
+            *p++ = (char) c;
+            *p   = 0;
+        }
+    }
+    return p;
+}
+
+/* add a sequence of bytes to a bounded buffer */
+static char*
+_bprint_b( char*  p, char*  end, const char*  buf, int  len )
+{
+    int  avail = end - p;
+
+    if (avail <= 0 || len <= 0)
+        return p;
+
+    if (avail > len)
+        avail = len;
+
+    memcpy( p, buf, avail );
+    p += avail;
+
+    if (p < end)
+        p[0] = 0;
+    else
+        end[-1] = 0;
+
+    return p;
+}
+
+/* add a string to a bounded buffer */
+static char*
+_bprint_s( char*  p, char*  end, const char*  str )
 {
-    static int  fd = -2;
-    int         ret;
+    return _bprint_b(p, end, str, strlen(str));
+}
+
+/* add a formatted string to a bounded buffer */
+static char*
+_bprint( char*  p, char*  end, const char*  format, ... )
+{
+    int      avail, n;
+    va_list  args;
+
+    avail = end - p;
+
+    if (avail <= 0)
+        return p;
+
+    va_start(args, format);
+    n = snprintf( p, avail, format, args);
+    va_end(args);
+
+    /* certain C libraries return -1 in case of truncation */
+    if (n < 0 || n > avail)
+        n = avail;
 
-    if (fd == -2) {
-        do {
-            fd = open( "/data/dns.log", O_CREAT | O_APPEND | O_WRONLY, 0666 );
-        } while (fd < 0 && errno == EINTR);
+    p += n;
+    /* certain C libraries do not zero-terminate in case of truncation */
+    if (p == end)
+        p[-1] = 0;
+
+    return p;
+}
+
+/* add a hex value to a bounded buffer, up to 8 digits */
+static char*
+_bprint_hex( char*  p, char*  end, unsigned  value, int  numDigits )
+{
+    char   text[sizeof(unsigned)*2];
+    int    nn = 0;
+
+    while (numDigits-- > 0) {
+        text[nn++] = "0123456789abcdef"[(value >> (numDigits*4)) & 15];
     }
+    return _bprint_b(p, end, text, nn);
+}
+
+/* add the hexadecimal dump of some memory area to a bounded buffer */
+static char*
+_bprint_hexdump( char*  p, char*  end, const uint8_t*  data, int  datalen )
+{
+    int   lineSize = 16;
+
+    while (datalen > 0) {
+        int  avail = datalen;
+        int  nn;
+
+        if (avail > lineSize)
+            avail = lineSize;
+
+        for (nn = 0; nn < avail; nn++) {
+            if (nn > 0)
+                p = _bprint_c(p, end, ' ');
+            p = _bprint_hex(p, end, data[nn], 2);
+        }
+        for ( ; nn < lineSize; nn++ ) {
+            p = _bprint_s(p, end, "   ");
+        }
+        p = _bprint_s(p, end, "  ");
 
-    if (fd >= 0) {
-        char  temp[128];
-        va_list  args;
-        va_start(args, fmt);
-        vsnprintf( temp, sizeof(temp), fmt, args);
-        va_end(args);
+        for (nn = 0; nn < avail; nn++) {
+            int  c = data[nn];
 
-        do {
-            ret = write( fd, temp, strlen(temp) );
-        } while (ret == -1 && errno == EINTR);
+            if (c < 32 || c > 127)
+                c = '.';
+
+            p = _bprint_c(p, end, c);
+        }
+        p = _bprint_c(p, end, '\n');
+
+        data    += avail;
+        datalen -= avail;
     }
+    return p;
 }
-#define  XLOG(...)   xlog(__VA_ARGS__)
-#else
-#define  XLOG(...)   ((void)0)
-#endif
 
+/* dump the content of a query of packet to the log */
+static void
+XLOG_BYTES( const void*  base, int  len )
+{
+    char  buff[1024];
+    char*  p = buff, *end = p + sizeof(buff);
+
+    p = _bprint_hexdump(p, end, base, len);
+    XLOG("%s",buff);
+}
+
+#else /* !DEBUG */
+#  define  XLOG(...)        ((void)0)
+#  define  XLOG_BYTES(a,b)  ((void)0)
+#endif
 
 static time_t
 _time_now( void )
@@ -107,96 +343,625 @@ _time_now( void )
     return tv.tv_sec;
 }
 
-/****************************************************************************/
-/****************************************************************************/
-/*****                                                                  *****/
-/*****                                                                  *****/
-/*****                                                                  *****/
-/****************************************************************************/
-/****************************************************************************/
+/* reminder: the general format of a DNS packet is the following:
+ *
+ *    HEADER  (12 bytes)
+ *    QUESTION  (variable)
+ *    ANSWER (variable)
+ *    AUTHORITY (variable)
+ *    ADDITIONNAL (variable)
+ *
+ * the HEADER is made of:
+ *
+ *   ID     : 16 : 16-bit unique query identification field
+ *
+ *   QR     :  1 : set to 0 for queries, and 1 for responses
+ *   Opcode :  4 : set to 0 for queries
+ *   AA     :  1 : set to 0 for queries
+ *   TC     :  1 : truncation flag, will be set to 0 in queries
+ *   RD     :  1 : recursion desired
+ *
+ *   RA     :  1 : recursion available (0 in queries)
+ *   Z      :  3 : three reserved zero bits
+ *   RCODE  :  4 : response code (always 0=NOERROR in queries)
+ *
+ *   QDCount: 16 : question count
+ *   ANCount: 16 : Answer count (0 in queries)
+ *   NSCount: 16: Authority Record count (0 in queries)
+ *   ARCount: 16: Additionnal Record count (0 in queries)
+ *
+ * the QUESTION is made of QDCount Question Record (QRs)
+ * the ANSWER is made of ANCount RRs
+ * the AUTHORITY is made of NSCount RRs
+ * the ADDITIONNAL is made of ARCount RRs
+ *
+ * Each Question Record (QR) is made of:
+ *
+ *   QNAME   : variable : Query DNS NAME
+ *   TYPE    : 16       : type of query (A=1, PTR=12, MX=15, AAAA=28, ALL=255)
+ *   CLASS   : 16       : class of query (IN=1)
+ *
+ * Each Resource Record (RR) is made of:
+ *
+ *   NAME    : variable : DNS NAME
+ *   TYPE    : 16       : type of query (A=1, PTR=12, MX=15, AAAA=28, ALL=255)
+ *   CLASS   : 16       : class of query (IN=1)
+ *   TTL     : 32       : seconds to cache this RR (0=none)
+ *   RDLENGTH: 16       : size of RDDATA in bytes
+ *   RDDATA  : variable : RR data (depends on TYPE)
+ *
+ * Each QNAME contains a domain name encoded as a sequence of 'labels'
+ * terminated by a zero. Each label has the following format:
+ *
+ *    LEN  : 8     : lenght of label (MUST be < 64)
+ *    NAME : 8*LEN : label length (must exclude dots)
+ *
+ * A value of 0 in the encoding is interpreted as the 'root' domain and
+ * terminates the encoding. So 'www.android.com' will be encoded as:
+ *
+ *   <3>www<7>android<3>com<0>
+ *
+ * Where <n> represents the byte with value 'n'
+ *
+ * Each NAME reflects the QNAME of the question, but has a slightly more
+ * complex encoding in order to provide message compression. This is achieved
+ * by using a 2-byte pointer, with format:
+ *
+ *    TYPE   : 2  : 0b11 to indicate a pointer, 0b01 and 0b10 are reserved
+ *    OFFSET : 14 : offset to another part of the DNS packet
+ *
+ * The offset is relative to the start of the DNS packet and must point
+ * A pointer terminates the encoding.
+ *
+ * The NAME can be encoded in one of the following formats:
+ *
+ *   - a sequence of simple labels terminated by 0 (like QNAMEs)
+ *   - a single pointer
+ *   - a sequence of simple labels terminated by a pointer
+ *
+ * A pointer shall always point to either a pointer of a sequence of
+ * labels (which can themselves be terminated by either a 0 or a pointer)
+ *
+ * The expanded length of a given domain name should not exceed 255 bytes.
+ *
+ * NOTE: we don't parse the answer packets, so don't need to deal with NAME
+ *       records, only QNAMEs.
+ */
+
+#define  DNS_HEADER_SIZE  12
+
+#define  DNS_TYPE_A   "\00\01"   /* big-endian decimal 1 */
+#define  DNS_TYPE_PTR "\00\014"  /* big-endian decimal 12 */
+#define  DNS_TYPE_MX  "\00\017"  /* big-endian decimal 15 */
+#define  DNS_TYPE_AAAA "\00\034" /* big-endian decimal 28 */
+#define  DNS_TYPE_ALL "\00\0377" /* big-endian decimal 255 */
+
+#define  DNS_CLASS_IN "\00\01"   /* big-endian decimal 1 */
+
+typedef struct {
+    const uint8_t*  base;
+    const uint8_t*  end;
+    const uint8_t*  cursor;
+} DnsPacket;
+
+static void
+_dnsPacket_init( DnsPacket*  packet, const uint8_t*  buff, int  bufflen )
+{
+    packet->base   = buff;
+    packet->end    = buff + bufflen;
+    packet->cursor = buff;
+}
+
+static void
+_dnsPacket_rewind( DnsPacket*  packet )
+{
+    packet->cursor = packet->base;
+}
+
+static void
+_dnsPacket_skip( DnsPacket*  packet, int  count )
+{
+    const uint8_t*  p = packet->cursor + count;
+
+    if (p > packet->end)
+        p = packet->end;
+
+    packet->cursor = p;
+}
 
-/* used to define the content of _RESOLV_HOSTENT_NONE
+static int
+_dnsPacket_readInt16( DnsPacket*  packet )
+{
+    const uint8_t*  p = packet->cursor;
+
+    if (p+2 > packet->end)
+        return -1;
+
+    packet->cursor = p+2;
+    return (p[0]<< 8) | p[1];
+}
+
+/** QUERY CHECKING
+ **/
+
+/* check bytes in a dns packet. returns 1 on success, 0 on failure.
+ * the cursor is only advanced in the case of success
  */
-const struct hostent  _resolv_hostent_none_cst = {
-    NULL,
-    NULL,
-    AF_INET,
-    4,
-    NULL
-};
+static int
+_dnsPacket_checkBytes( DnsPacket*  packet, int  numBytes, const void*  bytes )
+{
+    const uint8_t*  p = packet->cursor;
+
+    if (p + numBytes > packet->end)
+        return 0;
+
+    if (memcmp(p, bytes, numBytes) != 0)
+        return 0;
+
+    packet->cursor = p + numBytes;
+    return 1;
+}
 
-struct hostent*
-_resolv_hostent_copy( struct hostent*  hp )
+/* parse and skip a given QNAME stored in a query packet,
+ * from the current cursor position. returns 1 on success,
+ * or 0 for malformed data.
+ */
+static int
+_dnsPacket_checkQName( DnsPacket*  packet )
 {
-    struct hostent*  dst;
-    int              nn, len;
-    char*            p;
-    int              num_aliases = 0, num_addresses = 0;
+    const uint8_t*  p   = packet->cursor;
+    const uint8_t*  end = packet->end;
 
-    if (hp == NULL)
-        return NULL;
+    for (;;) {
+        int  c;
 
-    if (hp == _RESOLV_HOSTENT_NONE)
-        return hp;
+        if (p >= end)
+            break;
 
-    len  = sizeof(*hp);
-    len += strlen(hp->h_name) + 1;
+        c = *p++;
 
-    if (hp->h_aliases != NULL) {
-        for (nn = 0; hp->h_aliases[nn] != NULL; nn++)
-            len += sizeof(char*) + strlen(hp->h_aliases[nn]) + 1;
-        num_aliases = nn;
+        if (c == 0) {
+            packet->cursor = p;
+            return 1;
+        }
+
+        /* we don't expect label compression in QNAMEs */
+        if (c >= 64)
+            break;
+
+        p += c;
+        /* we rely on the bound check at the start
+         * of the loop here */
     }
-    len += sizeof(char*);
+    /* malformed data */
+    XLOG("malformed QNAME");
+    return 0;
+}
 
-    for (nn = 0; hp->h_addr_list[nn] != NULL; nn++) {
-        len += sizeof(char*) + hp->h_length;
+/* parse and skip a given QR stored in a packet.
+ * returns 1 on success, and 0 on failure
+ */
+static int
+_dnsPacket_checkQR( DnsPacket*  packet )
+{
+    int  len;
+
+    if (!_dnsPacket_checkQName(packet))
+        return 0;
+
+    /* TYPE must be one of the things we support */
+    if (!_dnsPacket_checkBytes(packet, 2, DNS_TYPE_A) &&
+        !_dnsPacket_checkBytes(packet, 2, DNS_TYPE_PTR) &&
+        !_dnsPacket_checkBytes(packet, 2, DNS_TYPE_MX) &&
+        !_dnsPacket_checkBytes(packet, 2, DNS_TYPE_AAAA) &&
+        !_dnsPacket_checkBytes(packet, 2, DNS_TYPE_ALL))
+    {
+        XLOG("unsupported TYPE");
+        return 0;
+    }
+    /* CLASS must be IN */
+    if (!_dnsPacket_checkBytes(packet, 2, DNS_CLASS_IN)) {
+        XLOG("unsupported CLASS");
+        return 0;
     }
-    num_addresses = nn;
-    len += sizeof(char*);
 
-    dst = malloc( len );
-    if (dst == NULL)
-        return NULL;
+    return 1;
+}
 
-    dst->h_aliases   = (char**)(dst + 1);
-    dst->h_addr_list = dst->h_aliases + num_aliases + 1;
-    dst->h_length    = hp->h_length;
-    dst->h_addrtype  = hp->h_addrtype;
+/* check the header of a DNS Query packet, return 1 if it is one
+ * type of query we can cache, or 0 otherwise
+ */
+static int
+_dnsPacket_checkQuery( DnsPacket*  packet )
+{
+    const uint8_t*  p = packet->base;
+    int             qdCount, anCount, dnCount, arCount;
+
+    if (p + DNS_HEADER_SIZE > packet->end) {
+        XLOG("query packet too small");
+        return 0;
+    }
 
-    p = (char*)(dst->h_addr_list + num_addresses + 1);
+    /* QR must be set to 0, opcode must be 0 and AA must be 0 */
+    /* RA, Z, and RCODE must be 0 */
+    if ((p[2] & 0xFC) != 0 || p[3] != 0) {
+        XLOG("query packet flags unsupported");
+        return 0;
+    }
 
-    /* write the addresses first, to help with alignment issues */
-    for (nn = 0; nn < num_addresses; nn++) {
-        dst->h_addr_list[nn] = p;
-        len = hp->h_length;
-        memcpy( p, hp->h_addr_list[nn], len );
-        p += len;
+    /* Note that we ignore the TC and RD bits here for the
+     * following reasons:
+     *
+     * - there is no point for a query packet sent to a server
+     *   to have the TC bit set, but the implementation might
+     *   set the bit in the query buffer for its own needs
+     *   between a _resolv_cache_lookup and a
+     *   _resolv_cache_add. We should not freak out if this
+     *   is the case.
+     *
+     * - we consider that the result from a RD=0 or a RD=1
+     *   query might be different, hence that the RD bit
+     *   should be used to differentiate cached result.
+     *
+     *   this implies that RD is checked when hashing or
+     *   comparing query packets, but not TC
+     */
+
+    /* ANCOUNT, DNCOUNT and ARCOUNT must be 0 */
+    qdCount = (p[4] << 8) | p[5];
+    anCount = (p[6] << 8) | p[7];
+    dnCount = (p[8] << 8) | p[9];
+    arCount = (p[10]<< 8) | p[11];
+
+    if (anCount != 0 || dnCount != 0 || arCount != 0) {
+        XLOG("query packet contains non-query records");
+        return 0;
     }
-    dst->h_addr_list[nn] = NULL;
 
-    for (nn = 0; nn < num_aliases; nn++) {
-        dst->h_aliases[nn] = p;
-        len = strlen(hp->h_aliases[nn]) + 1;
-        memcpy( p, hp->h_aliases[nn], len );
-        p += len;
+    if (qdCount == 0) {
+        XLOG("query packet doesn't contain query record");
+        return 0;
     }
-    dst->h_aliases[nn] = NULL;
 
-    dst->h_name      = p;
-    len = strlen(hp->h_name) + 1;
-    memcpy(p, hp->h_name, len);
-    p += len;
+    /* Check QDCOUNT QRs */
+    packet->cursor = p + DNS_HEADER_SIZE;
+
+    for (;qdCount > 0; qdCount--)
+        if (!_dnsPacket_checkQR(packet))
+            return 0;
 
-    return dst;
+    return 1;
 }
 
+/** QUERY DEBUGGING
+ **/
+#if DEBUG
+static char*
+_dnsPacket_bprintQName(DnsPacket*  packet, char*  bp, char*  bend)
+{
+    const uint8_t*  p   = packet->cursor;
+    const uint8_t*  end = packet->end;
+    int             first = 1;
 
-void
-_resolv_hostent_free( struct hostent*  hp )
+    for (;;) {
+        int  c;
+
+        if (p >= end)
+            break;
+
+        c = *p++;
+
+        if (c == 0) {
+            packet->cursor = p;
+            return bp;
+        }
+
+        /* we don't expect label compression in QNAMEs */
+        if (c >= 64)
+            break;
+
+        if (first)
+            first = 0;
+        else
+            bp = _bprint_c(bp, bend, '.');
+
+        bp = _bprint_b(bp, bend, (const char*)p, c);
+
+        p += c;
+        /* we rely on the bound check at the start
+         * of the loop here */
+    }
+    /* malformed data */
+    bp = _bprint_s(bp, bend, "<MALFORMED>");
+    return bp;
+}
+
+static char*
+_dnsPacket_bprintQR(DnsPacket*  packet, char*  p, char*  end)
 {
-    if (hp && hp != _RESOLV_HOSTENT_NONE)
-        free(hp);
+#define  QQ(x)   { DNS_TYPE_##x, #x }
+    static const struct { 
+        const char*  typeBytes;
+        const char*  typeString; 
+    } qTypes[] =
+    {
+        QQ(A), QQ(PTR), QQ(MX), QQ(AAAA), QQ(ALL),
+        { NULL, NULL }
+    };
+    int          nn;
+    const char*  typeString = NULL;
+
+    /* dump QNAME */
+    p = _dnsPacket_bprintQName(packet, p, end);
+
+    /* dump TYPE */
+    p = _bprint_s(p, end, " (");
+
+    for (nn = 0; qTypes[nn].typeBytes != NULL; nn++) {
+        if (_dnsPacket_checkBytes(packet, 2, qTypes[nn].typeBytes)) {
+            typeString = qTypes[nn].typeString;
+            break;
+        }
+    }
+
+    if (typeString != NULL)
+        p = _bprint_s(p, end, typeString);
+    else {
+        int  typeCode = _dnsPacket_readInt16(packet);
+        p = _bprint(p, end, "UNKNOWN-%d", typeCode);
+    }
+
+    p = _bprint_c(p, end, ')');
+
+    /* skip CLASS */
+    _dnsPacket_skip(packet, 2);
+    return p;
+}
+
+/* this function assumes the packet has already been checked */
+static char*
+_dnsPacket_bprintQuery( DnsPacket*  packet, char*  p, char*  end )
+{
+    int   qdCount;
+
+    if (packet->base[2] & 0x1) {
+        p = _bprint_s(p, end, "RECURSIVE ");
+    }
+
+    _dnsPacket_skip(packet, 4);
+    qdCount = _dnsPacket_readInt16(packet);
+    _dnsPacket_skip(packet, 6);
+
+    for ( ; qdCount > 0; qdCount-- ) {
+        p = _dnsPacket_bprintQR(packet, p, end);
+    }
+    return p;
+}
+#endif
+
+
+/** QUERY HASHING SUPPORT
+ **
+ ** THE FOLLOWING CODE ASSUMES THAT THE INPUT PACKET HAS ALREADY
+ ** BEEN SUCCESFULLY CHECKED.
+ **/
+
+/* use 32-bit FNV hash function */
+#define  FNV_MULT   16777619U
+#define  FNV_BASIS  2166136261U
+
+static unsigned
+_dnsPacket_hashBytes( DnsPacket*  packet, int  numBytes, unsigned  hash )
+{
+    const uint8_t*  p   = packet->cursor;
+    const uint8_t*  end = packet->end;
+
+    while (numBytes > 0 && p < end) {
+        hash = hash*FNV_MULT ^ *p++;
+    }
+    packet->cursor = p;
+    return hash;
+}
+
+
+static unsigned
+_dnsPacket_hashQName( DnsPacket*  packet, unsigned  hash )
+{
+    const uint8_t*  p   = packet->cursor;
+    const uint8_t*  end = packet->end;
+
+    for (;;) {
+        int  c;
+
+        if (p >= end) {  /* should not happen */
+            XLOG("%s: INTERNAL_ERROR: read-overflow !!\n", __FUNCTION__);
+            break;
+        }
+
+        c = *p++;
+
+        if (c == 0)
+            break;
+
+        if (c >= 64) {
+            XLOG("%s: INTERNAL_ERROR: malformed domain !!\n", __FUNCTION__);
+            break;
+        }
+        if (p + c >= end) {
+            XLOG("%s: INTERNAL_ERROR: simple label read-overflow !!\n",
+                    __FUNCTION__);
+            break;
+        }
+        while (c > 0) {
+            hash = hash*FNV_MULT ^ *p++;
+            c   -= 1;
+        }
+    }
+    packet->cursor = p;
+    return hash;
+}
+
+static unsigned
+_dnsPacket_hashQR( DnsPacket*  packet, unsigned  hash )
+{
+    int   len;
+
+    hash = _dnsPacket_hashQName(packet, hash);
+    hash = _dnsPacket_hashBytes(packet, 4, hash); /* TYPE and CLASS */
+    return hash;
+}
+
+static unsigned
+_dnsPacket_hashQuery( DnsPacket*  packet )
+{
+    unsigned  hash = FNV_BASIS;
+    int       count;
+    _dnsPacket_rewind(packet);
+
+    /* we ignore the TC bit for reasons explained in
+     * _dnsPacket_checkQuery().
+     *
+     * however we hash the RD bit to differentiate
+     * between answers for recursive and non-recursive
+     * queries.
+     */
+    hash = hash*FNV_MULT ^ (packet->base[2] & 1);
+
+    /* assume: other flags are 0 */
+    _dnsPacket_skip(packet, 4);
+
+    /* read QDCOUNT */
+    count = _dnsPacket_readInt16(packet);
+
+    /* assume: ANcount, NScount, ARcount are 0 */
+    _dnsPacket_skip(packet, 6);
+
+    /* hash QDCOUNT QRs */
+    for ( ; count > 0; count-- )
+        hash = _dnsPacket_hashQR(packet, hash);
+
+    return hash;
+}
+
+
+/** QUERY COMPARISON
+ **
+ ** THE FOLLOWING CODE ASSUMES THAT THE INPUT PACKETS HAVE ALREADY
+ ** BEEN SUCCESFULLY CHECKED.
+ **/
+
+static int
+_dnsPacket_isEqualDomainName( DnsPacket*  pack1, DnsPacket*  pack2 )
+{
+    const uint8_t*  p1   = pack1->cursor;
+    const uint8_t*  end1 = pack1->end;
+    const uint8_t*  p2   = pack2->cursor;
+    const uint8_t*  end2 = pack2->end;
+
+    for (;;) {
+        int  c1, c2;
+
+        if (p1 >= end1 || p2 >= end2) {
+            XLOG("%s: INTERNAL_ERROR: read-overflow !!\n", __FUNCTION__);
+            break;
+        }
+        c1 = *p1++;
+        c2 = *p2++;
+        if (c1 != c2)
+            break;
+
+        if (c1 == 0) {
+            pack1->cursor = p1;
+            pack2->cursor = p2;
+            return 1;
+        }
+        if (c1 >= 64) {
+            XLOG("%s: INTERNAL_ERROR: malformed domain !!\n", __FUNCTION__);
+            break;
+        }
+        if ((p1+c1 > end1) || (p2+c1 > end2)) {
+            XLOG("%s: INTERNAL_ERROR: simple label read-overflow !!\n",
+                    __FUNCTION__);
+            break;
+        }
+        if (memcmp(p1, p2, c1) != 0)
+            break;
+        p1 += c1;
+        p2 += c1;
+        /* we rely on the bound checks at the start of the loop */
+    }
+    /* not the same, or one is malformed */
+    XLOG("different DN");
+    return 0;
+}
+
+static int
+_dnsPacket_isEqualBytes( DnsPacket*  pack1, DnsPacket*  pack2, int  numBytes )
+{
+    const uint8_t*  p1 = pack1->cursor;
+    const uint8_t*  p2 = pack2->cursor;
+
+    if ( p1 + numBytes > pack1->end || p2 + numBytes > pack2->end )
+        return 0;
+
+    if ( memcmp(p1, p2, numBytes) != 0 )
+        return 0;
+
+    pack1->cursor += numBytes;
+    pack2->cursor += numBytes;
+    return 1;
+}
+
+static int
+_dnsPacket_isEqualQR( DnsPacket*  pack1, DnsPacket*  pack2 )
+{
+    /* compare domain name encoding + TYPE + CLASS */
+    if ( !_dnsPacket_isEqualDomainName(pack1, pack2) ||
+         !_dnsPacket_isEqualBytes(pack1, pack2, 2+2) )
+        return 0;
+
+    return 1;
+}
+
+static int
+_dnsPacket_isEqualQuery( DnsPacket*  pack1, DnsPacket*  pack2 )
+{
+    int  count1, count2;
+
+    /* compare the headers, ignore most fields */
+    _dnsPacket_rewind(pack1);
+    _dnsPacket_rewind(pack2);
+
+    /* compare RD, ignore TC, see comment in _dnsPacket_checkQuery */
+    if ((pack1->base[2] & 1) != (pack2->base[2] & 1)) {
+        XLOG("different RD");
+        return 0;
+    }
+
+    /* assume: other flags are all 0 */
+    _dnsPacket_skip(pack1, 4);
+    _dnsPacket_skip(pack2, 4);
+
+    /* compare QDCOUNT */
+    count1 = _dnsPacket_readInt16(pack1);
+    count2 = _dnsPacket_readInt16(pack2);
+    if (count1 != count2 || count1 < 0) {
+        XLOG("different QDCOUNT");
+        return 0;
+    }
+
+    /* assume: ANcount, NScount and ARcount are all 0 */
+    _dnsPacket_skip(pack1, 6);
+    _dnsPacket_skip(pack2, 6);
+
+    /* compare the QDCOUNT QRs */
+    for ( ; count1 > 0; count1-- ) {
+        if (!_dnsPacket_isEqualQR(pack1, pack2)) {
+            XLOG("different QR");
+            return 0;
+        }
+    }
+    return 1;
 }
 
 /****************************************************************************/
@@ -207,15 +972,23 @@ _resolv_hostent_free( struct hostent*  hp )
 /****************************************************************************/
 /****************************************************************************/
 
+/* cache entry. for simplicity, 'hash' and 'hlink' are inlined in this
+ * structure though they are conceptually part of the hash table.
+ *
+ * similarly, mru_next and mru_prev are part of the global MRU list
+ */
 typedef struct Entry {
-    unsigned int     hash;
-    const char*      name;
-    short            af;
-    short            index;
+    unsigned int     hash;   /* hash value */
+    struct Entry*    hlink;  /* next in collision chain */
     struct Entry*    mru_prev;
     struct Entry*    mru_next;
-    time_t           when;
-    struct hostent*  hp;
+
+    const uint8_t*   query;
+    int              querylen;
+    const uint8_t*   answer;
+    int              answerlen;
+    time_t           when;   /* time_t when entry was added to table */
+    int              id;     /* for debugging purpose */
 } Entry;
 
 
@@ -224,30 +997,10 @@ entry_free( Entry*  e )
 {
     /* everything is allocated in a single memory block */
     if (e) {
-        _resolv_hostent_free(e->hp);
         free(e);
     }
 }
 
-static void
-entry_init_key( Entry*  e, const char*  name, int  af )
-{
-    unsigned     h = 0;
-    const char*  p = name;
-
-    /* compute hash */
-    p = name;
-    while (*p) {
-        h = h*33 + *p++;
-    }
-    h += af*17;
-
-    e->hash = h;
-    e->name = name;
-    e->af   = (short) af;
-}
-
-
 static __inline__ void
 entry_mru_remove( Entry*  e )
 {
@@ -267,46 +1020,75 @@ entry_mru_add( Entry*  e, Entry*  list )
     first->mru_prev = e;
 }
 
+/* compute the hash of a given entry, this is a hash of most
+ * data in the query (key) */
+static unsigned
+entry_hash( const Entry*  e )
+{
+    DnsPacket  pack[1];
+
+    _dnsPacket_init(pack, e->query, e->querylen);
+    return _dnsPacket_hashQuery(pack);
+}
 
+/* initialize an Entry as a search key, this also checks the input query packet
+ * returns 1 on success, or 0 in case of unsupported/malformed data */
+static int
+entry_init_key( Entry*  e, const void*  query, int  querylen )
+{
+    DnsPacket  pack[1];
+
+    memset(e, 0, sizeof(*e));
+
+    e->query    = query;
+    e->querylen = querylen;
+    e->hash     = entry_hash(e);
+
+    _dnsPacket_init(pack, query, querylen);
+
+    return _dnsPacket_checkQuery(pack);
+}
+
+/* allocate a new entry as a cache node */
 static Entry*
-entry_alloc( const char*  name, int  af, int  index, struct hostent*  hp )
+entry_alloc( const Entry*  init, const void*  answer, int  answerlen )
 {
     Entry*  e;
-    int     num_aliases   = 0;
-    int     num_addresses = 0;
-    char**  aliases;
-    char**  addresses;
+    int     size;
 
-    /* compute the length of the memory block that will contain everything */
-    int   len = sizeof(*e) + strlen(name)+1;
-
-    e = malloc(len);
+    size = sizeof(*e) + init->querylen + answerlen;
+    e    = calloc(size, 1);
     if (e == NULL)
         return e;
 
-    entry_init_key(e, name, af);
+    e->hash     = init->hash;
+    e->query    = (const uint8_t*)(e+1);
+    e->querylen = init->querylen;
 
-    e->mru_next = e->mru_prev = e;
-    e->index    = (short) index;
-    e->when     = _time_now();
-    e->hp       = _resolv_hostent_copy(hp);
+    memcpy( (char*)e->query, init->query, e->querylen );
 
-    if (e->hp == NULL) {
-        free(e);
-        return NULL;
-    }
+    e->answer    = e->query + e->querylen;
+    e->answerlen = answerlen;
+
+    memcpy( (char*)e->answer, answer, e->answerlen );
+
+    e->when  = _time_now();
 
-    e->name = (char*)(e+1);
-    len     = strlen(name)+1;
-    memcpy( (char*)e->name, name, len );
     return e;
 }
 
-
-static __inline__ int
+static int
 entry_equals( const Entry*  e1, const Entry*  e2 )
 {
-    return e1->hash == e2->hash && e1->af == e2->af && !strcmp( e1->name, e2->name );
+    DnsPacket  pack1[1], pack2[1];
+
+    if (e1->querylen != e2->querylen) {
+        return 0;
+    }
+    _dnsPacket_init(pack1, e1->query, e1->querylen);
+    _dnsPacket_init(pack2, e2->query, e2->querylen);
+
+    return _dnsPacket_isEqualQuery(pack1, pack2);
 }
 
 /****************************************************************************/
@@ -317,30 +1099,49 @@ entry_equals( const Entry*  e1, const Entry*  e2 )
 /****************************************************************************/
 /****************************************************************************/
 
+/* We use a simple hash table with external collision lists
+ * for simplicity, the hash-table fields 'hash' and 'hlink' are
+ * inlined in the Entry structure.
+ */
 #define  MAX_HASH_ENTRIES   (2*CONFIG_MAX_ENTRIES)
 
 typedef struct resolv_cache {
-    int               num_entries;
-    Entry             mru_list;
-    pthread_mutex_t   lock;
-    int               disabled;
-    Entry*            entries[ MAX_HASH_ENTRIES ];      /* hash-table of pointers to entries */
+    int              num_entries;
+    Entry            mru_list;
+    pthread_mutex_t  lock;
+    unsigned         generation;
+    int              last_id;
+    Entry*           entries[ MAX_HASH_ENTRIES ];
 } Cache;
 
 
-void
-_resolv_cache_destroy( struct resolv_cache*  cache )
+#define  HTABLE_VALID(x)  ((x) != NULL && (x) != HTABLE_DELETED)
+
+static void
+_cache_flush_locked( Cache*  cache )
 {
-    if (cache != NULL) {
-        int  nn;
-        for (nn = 0; nn < MAX_HASH_ENTRIES; nn++) {
-            entry_free(cache->entries[nn]);
+    int     nn;
+    time_t  now = _time_now();
+
+    for (nn = 0; nn < MAX_HASH_ENTRIES; nn++) 
+    {
+        Entry**  pnode = &cache->entries[nn];
+
+        while (*pnode != NULL) {
+            Entry*  node = *pnode;
+            *pnode = node->hlink;
+            entry_free(node);
         }
-        pthread_mutex_destroy(&cache->lock);
-        free(cache);
     }
-}
 
+    cache->mru_list.mru_next = cache->mru_list.mru_prev = &cache->mru_list;
+    cache->num_entries       = 0;
+    cache->last_id           = 0;
+
+    XLOG("*************************\n"
+         "*** DNS CACHE FLUSHED ***\n"
+         "*************************");
+}
 
 struct resolv_cache*
 _resolv_cache_create( void )
@@ -349,116 +1150,203 @@ _resolv_cache_create( void )
 
     cache = calloc(sizeof(*cache), 1);
     if (cache) {
-        const char*  env = getenv(CONFIG_ENV);
-
-        if (env && atoi(env) == 0)
-            cache->disabled = 1;
-
+        cache->generation = ~0U;
         pthread_mutex_init( &cache->lock, NULL );
         cache->mru_list.mru_prev = cache->mru_list.mru_next = &cache->mru_list;
-        XLOG("%s: cache=%p %s\n", __FUNCTION__, cache, cache->disabled ? "disabled" : "enabled" );
+        XLOG("%s: cache created\n", __FUNCTION__);
     }
     return cache;
 }
 
 
-static int
-_resolv_cache_find_index( Cache*       cache,
-                          const char*  name,
-                          int          af )
+#if DEBUG
+static void
+_dump_query( const uint8_t*  query, int  querylen )
 {
-    Entry  key;
-    int    nn, step, tries;
+    char       temp[256], *p=temp, *end=p+sizeof(temp);
+    DnsPacket  pack[1];
 
-    entry_init_key( &key, name, af );
+    _dnsPacket_init(pack, query, querylen);
+    p = _dnsPacket_bprintQuery(pack, p, end);
+    XLOG("QUERY: %s", temp);
+}
 
-    tries = MAX_HASH_ENTRIES;
-    nn    = key.hash % MAX_HASH_ENTRIES;
-    step  = 5;
+static void
+_cache_dump_mru( Cache*  cache )
+{
+    char    temp[512], *p=temp, *end=p+sizeof(temp);
+    Entry*  e;
 
-    while (tries > 0) {
-        Entry*  key2 = cache->entries[nn];
+    p = _bprint(temp, end, "MRU LIST (%2d): ", cache->num_entries);
+    for (e = cache->mru_list.mru_next; e != &cache->mru_list; e = e->mru_next)
+        p = _bprint(p, end, " %d", e->id);
 
-        if (key2 == NULL) {
-            return -(nn + 1);
-        }
+    XLOG("%s", temp);
+}
+#endif
 
-        if (entry_equals( &key, key2 ) ) {
-            return nn;
-        }
+#if DEBUG
+#  define  XLOG_QUERY(q,len)   _dump_query((q), (len))
+#else
+#  define  XLOG_QUERY(q,len)   ((void)0)
+#endif
+
+/* This function tries to find a key within the hash table
+ * In case of success, it will return a *pointer* to the hashed key.
+ * In case of failure, it will return a *pointer* to NULL
+ *
+ * So, the caller must check '*result' to check for success/failure.
+ *
+ * The main idea is that the result can later be used directly in
+ * calls to _resolv_cache_add or _resolv_cache_remove as the 'lookup'
+ * parameter. This makes the code simpler and avoids re-searching
+ * for the key position in the htable.
+ *
+ * The result of a lookup_p is only valid until you alter the hash
+ * table.
+ */
+static Entry**
+_cache_lookup_p( Cache*   cache,
+                 Entry*   key )
+{
+    int      index = key->hash % MAX_HASH_ENTRIES;
+    Entry**  pnode = &cache->entries[ key->hash % MAX_HASH_ENTRIES ];
+
+    while (*pnode != NULL) {
+        Entry*  node = *pnode;
+
+        if (node == NULL)
+            break;
 
-        nn     = (nn + step) % MAX_HASH_ENTRIES;
-        tries -= 1;
+        if (node->hash == key->hash && entry_equals(node, key))
+            break;
+
+        pnode = &node->hlink;
     }
-    return -(MAX_HASH_ENTRIES+1);
+    return pnode; 
 }
 
+/* Add a new entry to the hash table. 'lookup' must be the
+ * result of an immediate previous failed _lookup_p() call
+ * (i.e. with *lookup == NULL), and 'e' is the pointer to the
+ * newly created entry
+ */
+static void
+_cache_add_p( Cache*   cache,
+              Entry**  lookup,
+              Entry*   e )
+{
+    *lookup = e;
+    e->id = ++cache->last_id;
+    entry_mru_add(e, &cache->mru_list);
+    cache->num_entries += 1;
+
+    XLOG("%s: entry %d added (count=%d)", __FUNCTION__,
+         e->id, cache->num_entries);
+}
 
+/* Remove an existing entry from the hash table,
+ * 'lookup' must be the result of an immediate previous
+ * and succesful _lookup_p() call.
+ */
 static void
-_resolv_cache_remove( struct resolv_cache*  cache,
-                      Entry*                e )
+_cache_remove_p( Cache*   cache,
+                 Entry**  lookup )
 {
-    XLOG("%s: name='%s' af=%d\n", __FUNCTION__, e->name, e->af);
-    cache->entries[ e->index ] = NULL;  /* remove from hash table */
-    entry_mru_remove( e );
-    entry_free( e );
+    Entry*  e  = *lookup;
+
+    XLOG("%s: entry %d removed (count=%d)", __FUNCTION__,
+         e->id, cache->num_entries-1);
+
+    entry_mru_remove(e);
+    *lookup = e->hlink;
+    entry_free(e);
     cache->num_entries -= 1;
 }
 
-
-struct hostent*
-_resolv_cache_lookup( struct resolv_cache*  cache,
-                      const char*           name,
-                      int                   af )
+/* Remove the oldest entry from the hash table.
+ */
+static void
+_cache_remove_oldest( Cache*  cache )
 {
-    int               index;
-    struct hostent*   result = NULL;
+    Entry*   oldest = cache->mru_list.mru_prev;
+    Entry**  lookup = _cache_lookup_p(cache, oldest);
 
-    if (cache->disabled)
-        return NULL;
+    if (*lookup == NULL) { /* should not happen */
+        XLOG("%s: OLDEST NOT IN HTABLE ?", __FUNCTION__);
+        return;
+    }
+    _cache_remove_p(cache, lookup);
+}
 
+
+ResolvCacheStatus
+_resolv_cache_lookup( struct resolv_cache*  cache,
+                      const void*           query,
+                      int                   querylen,
+                      void*                 answer,
+                      int                   answersize,
+                      int                  *answerlen )
+{
+    DnsPacket  pack[1];
+    Entry      key[1];
+    int        index;
+    Entry**    lookup;
+    Entry*     e;
+    time_t     now;
+
+    ResolvCacheStatus  result = RESOLV_CACHE_NOTFOUND;
+
+    XLOG("%s: lookup", __FUNCTION__);
+    XLOG_QUERY(query, querylen);
+
+    /* we don't cache malformed queries */
+    if (!entry_init_key(key, query, querylen)) {
+        XLOG("%s: unsupported query", __FUNCTION__);
+        return RESOLV_CACHE_UNSUPPORTED;
+    }
+    /* lookup cache */
     pthread_mutex_lock( &cache->lock );
 
-    XLOG( "%s: cache=%p name='%s' af=%d ", __FUNCTION__, cache, name, af );
-    index = _resolv_cache_find_index( cache, name, af );
-    if (index >= 0) {
-        Entry*  e   = cache->entries[index];
-        time_t  now = _time_now();
-        struct hostent**  pht;
+    /* see the description of _lookup_p to understand this.
+     * the function always return a non-NULL pointer.
+     */
+    lookup = _cache_lookup_p(cache, key);
+    e      = *lookup;
 
-        /* ignore stale entries, they will be discarded in _resolv_cache_add */
-        if ( (unsigned)(now - e->when) >= CONFIG_SECONDS ) {
-            XLOG( " OLD\n" );
-            goto Exit;
-        }
+    if (e == NULL) {
+        XLOG( "NOT IN CACHE");
+        goto Exit;
+    }
 
-        /* bump up this entry to the top of the MRU list */
-        if (e != cache->mru_list.mru_next) {
-            entry_mru_remove( e );
-            entry_mru_add( e, &cache->mru_list );
-        }
+    now = _time_now();
 
-        /* now copy the result into a thread-local variable */
-        pht = __get_res_cache_hostent_p();
-        if (pht == NULL) {
-            XLOG( " NOTLS\n" );  /* shouldn't happen */
-            goto Exit;
-        }
+    /* remove stale entries here */
+    if ( (unsigned)(now - e->when) >= CONFIG_SECONDS ) {
+        XLOG( " NOT IN CACHE (STALE ENTRY %p DISCARDED)", *lookup );
+        _cache_remove_p(cache, lookup);
+        goto Exit;
+    }
 
-        if (pht[0]) {
-            _resolv_hostent_free( pht[0] );  /* clear previous value */
-            pht[0] = NULL;
-        }
-        result = _resolv_hostent_copy( e->hp );
-        if (result == NULL) {
-            XLOG( " NOMEM\n" );  /* bummer */
-            goto Exit;
-        }
-        XLOG( " OK\n" );
-        pht[0] = result;
+    *answerlen = e->answerlen;
+    if (e->answerlen > answersize) {
+        /* NOTE: we return UNSUPPORTED if the answer buffer is too short */
+        result = RESOLV_CACHE_UNSUPPORTED;
+        XLOG(" ANSWER TOO LONG");
         goto Exit;
     }
-    XLOG( " KO\n" );
+
+    memcpy( answer, e->answer, e->answerlen );
+
+    /* bump up this entry to the top of the MRU list */
+    if (e != cache->mru_list.mru_next) {
+        entry_mru_remove( e );
+        entry_mru_add( e, &cache->mru_list );
+    }
+
+    XLOG( "FOUND IN CACHE entry=%p", e );
+    result = RESOLV_CACHE_FOUND;
+
 Exit:
     pthread_mutex_unlock( &cache->lock );
     return result;
@@ -467,42 +1355,59 @@ Exit:
 
 void
 _resolv_cache_add( struct resolv_cache*  cache,
-                   const char*           name,
-                   int                   af,
-                   struct hostent*       hp )
+                   const void*           query,
+                   int                   querylen,
+                   const void*           answer,
+                   int                   answerlen )
 {
+    Entry    key[1];
     Entry*   e;
-    int      index;
+    Entry**  lookup;
 
-    if (cache->disabled)
+    /* don't assume that the query has already been cached
+     */
+    if (!entry_init_key( key, query, querylen )) {
+        XLOG( "%s: passed invalid query ?", __FUNCTION__);
         return;
+    }
 
     pthread_mutex_lock( &cache->lock );
 
-    XLOG( "%s: cache=%p name='%s' af=%d\n", __FUNCTION__, cache, name, af);
+    XLOG( "%s: query:", __FUNCTION__ );
+    XLOG_QUERY(query,querylen);
+#if DEBUG_DATA
+    XLOG( "answer:");
+    XLOG_BYTES(answer,answerlen);
+#endif
+
+    lookup = _cache_lookup_p(cache, key);
+    e      = *lookup;
 
-    /* get rid of the oldest entry if needed */
-    if (cache->num_entries >= CONFIG_MAX_ENTRIES) {
-        Entry*  oldest = cache->mru_list.mru_prev;
-        _resolv_cache_remove( cache, oldest );
+    if (e != NULL) { /* should not happen */
+        XLOG("%s: ALREADY IN CACHE (%p) ? IGNORING ADD",
+             __FUNCTION__, e);
+        goto Exit;
     }
 
-    index = _resolv_cache_find_index( cache, name, af );
-    if (index >= 0) {
-        /* discard stale entry */
-        _resolv_cache_remove( cache, cache->entries[index] );
-    } else {
-        index = -(index+1);
-        if (index >= MAX_HASH_ENTRIES)
-            goto Exit;  /* should not happen */
+    if (cache->num_entries >= CONFIG_MAX_ENTRIES) {
+        _cache_remove_oldest(cache);
+        /* need to lookup again */
+        lookup = _cache_lookup_p(cache, key);
+        e      = *lookup;
+        if (e != NULL) {
+            XLOG("%s: ALREADY IN CACHE (%p) ? IGNORING ADD",
+                __FUNCTION__, e);
+            goto Exit;
+        }
     }
 
-    e = entry_alloc( name, af, index, hp );
+    e = entry_alloc( key, answer, answerlen );
     if (e != NULL) {
-        entry_mru_add( e, &cache->mru_list );
-        cache->entries[index] = e;
-        cache->num_entries   += 1;
+        _cache_add_p(cache, lookup, e);
     }
+#if DEBUG
+    _cache_dump_mru(cache);
+#endif
 Exit:
     pthread_mutex_unlock( &cache->lock );
 }
@@ -521,6 +1426,13 @@ static pthread_once_t        _res_cache_once;
 static void
 _res_cache_init( void )
 {
+    const char*  env = getenv(CONFIG_ENV);
+
+    if (env && atoi(env) == 0) {
+        /* the cache is disabled */
+        return;
+    }
+
     _res_cache = _resolv_cache_create();
 }
 
@@ -531,3 +1443,19 @@ __get_res_cache( void )
     pthread_once( &_res_cache_once, _res_cache_init );
     return _res_cache;
 }
+
+void
+_resolv_cache_reset( unsigned  generation )
+{
+    XLOG("%s: generation=%d", __FUNCTION__, generation);
+
+    if (_res_cache == NULL)
+        return;
+
+    pthread_mutex_lock( &_res_cache->lock );
+    if (_res_cache->generation != generation) {
+        _cache_flush_locked(_res_cache);
+        _res_cache->generation = generation;
+    }
+    pthread_mutex_unlock( &_res_cache->lock );
+}
diff --git a/libc/netbsd/resolv/res_send.c b/libc/netbsd/resolv/res_send.c
index 24b740a..3aca760 100644
--- a/libc/netbsd/resolv/res_send.c
+++ b/libc/netbsd/resolv/res_send.c
@@ -81,6 +81,9 @@ __RCSID("$NetBSD: res_send.c,v 1.9 2006/01/24 17:41:25 christos Exp $");
 #endif
 #endif /* LIBC_SCCS and not lint */
 
+/* set to 1 to use our small/simple/limited DNS cache */
+#define  USE_RESOLV_CACHE  1
+
 /*
  * Send query to name server and wait for reply.
  */
@@ -111,6 +114,10 @@ __RCSID("$NetBSD: res_send.c,v 1.9 2006/01/24 17:41:25 christos Exp $");
 
 #include <isc/eventlib.h>
 
+#if USE_RESOLV_CACHE
+#  include <resolv_cache.h>
+#endif
+
 #ifndef DE_CONST
 #define DE_CONST(c,v)   v = ((c) ? \
     strchr((const void *)(c), *(const char *)(const void *)(c)) : NULL)
@@ -344,12 +351,17 @@ res_queriesmatch(const u_char *buf1, const u_char *eom1,
 	return (1);
 }
 
+
 int
 res_nsend(res_state statp,
 	  const u_char *buf, int buflen, u_char *ans, int anssiz)
 {
 	int gotsomewhere, terrno, try, v_circuit, resplen, ns, n;
 	char abuf[NI_MAXHOST];
+#if USE_RESOLV_CACHE
+        struct resolv_cache*  cache;
+        ResolvCacheStatus     cache_status = RESOLV_CACHE_UNSUPPORTED;
+#endif
 
 	if (statp->nscount == 0) {
 		errno = ESRCH;
@@ -365,6 +377,20 @@ res_nsend(res_state statp,
 	gotsomewhere = 0;
 	terrno = ETIMEDOUT;
 
+#if USE_RESOLV_CACHE
+        cache = __get_res_cache();
+        if (cache != NULL) {
+            int  anslen = 0;
+            cache_status = _resolv_cache_lookup(
+                                cache, buf, buflen,
+                                ans, anssiz, &anslen);
+
+            if (cache_status == RESOLV_CACHE_FOUND) {
+                return anslen;
+            }
+        }
+#endif
+
 	/*
 	 * If the ns_addr_list in the resolver context has changed, then
 	 * invalidate our cached copy and the associated timing data.
@@ -534,6 +560,12 @@ res_nsend(res_state statp,
 			(stdout, "%s", ""),
 			ans, (resplen > anssiz) ? anssiz : resplen);
 
+#if USE_RESOLV_CACHE
+                if (cache_status == RESOLV_CACHE_NOTFOUND) {
+                    _resolv_cache_add(cache, buf, buflen,
+                                      ans, resplen);
+                }
+#endif
 		/*
 		 * If we have temporarily opened a virtual circuit,
 		 * or if we haven't been asked to keep a socket open,
diff --git a/libc/netbsd/resolv/res_state.c b/libc/netbsd/resolv/res_state.c
index 8f2851a..3a2301d 100644
--- a/libc/netbsd/resolv/res_state.c
+++ b/libc/netbsd/resolv/res_state.c
@@ -46,7 +46,6 @@ typedef struct {
     struct __res_state     _nres[1];
     unsigned               _serial;
     struct prop_info*      _pi;
-    struct hostent*        _hostent;
     struct res_static      _rstatic[1];
 } _res_thread;
 
@@ -66,9 +65,9 @@ _res_thread_alloc(void)
         if ( res_ninit( rt->_nres ) < 0 ) {
             free(rt);
             rt = NULL;
+        } else {
+            memset(rt->_rstatic, 0, sizeof rt->_rstatic);
         }
-        rt->_hostent = NULL;
-        memset(rt->_rstatic, 0, sizeof rt->_rstatic);
     }
     return rt;
 }
@@ -93,7 +92,6 @@ _res_thread_free( void*  _rt )
     _res_thread*  rt = _rt;
 
     _res_static_done(rt->_rstatic);
-    _resolv_hostent_free(rt->_hostent);
     res_ndestroy(rt->_nres);
     free(rt);
 }
@@ -132,6 +130,7 @@ _res_thread_get(void)
         rt = NULL;
         pthread_setspecific( _res_key, rt );
     }
+    _resolv_cache_reset(rt->_serial);
     return rt;
 }
 
@@ -177,14 +176,6 @@ __res_put_state(res_state res)
     res=res;
 }
 
-struct hostent**
-__get_res_cache_hostent_p(void)
-{
-    _res_thread*  rt = _res_thread_get();
-
-    return rt ? &rt->_hostent : NULL;
-}
-
 res_static
 __res_get_static(void)
 {