{182782396} Look up peer's short hostname

This patch includes a slight refactoring around hostname lookup,
also fixes incorrect IP address when peer drops.

Signed-off-by: Rivers Zhang <hzhang320@bloomberg.net>

Author: riverszhang89

bdb/bdb_net.c

@@ -41,6 +41,7 @@
 #include "printformats.h"
 #include "crc32c.h"
 #include <timer_util.h>
+#include <hostname_support.h>
 
 #undef UDP_DEBUG
 #undef UDP_TRACE
@@ -200,33 +201,23 @@ void comdb2_early_ack(DB_ENV *dbenv, DB_LSN permlsn, uint32_t commit_gen, uint32
 char *print_addr(struct sockaddr_in *addr, char *buf)
 {
     buf[0] = '\0';
+    short port = ntohs(addr->sin_port);
     if (addr == NULL) {
         return buf;
     }
     if(addr->sin_addr.s_addr == htonl(INADDR_ANY)) {
-        sprintf(buf, "[0.0.0.0 0.0.0.0:%d ]", ntohs(addr->sin_port));
+        sprintf(buf, "[0.0.0.0 0.0.0.0:%d ]", port);
         return buf;
     }
     char ip[32] = {0};
-    char name[256] = {0};
-    char service[256] = {0};
+    const char *name;
     char errbuf[256] = {0};
-    socklen_t len;
 
-    len = sizeof(*addr);
-    int rc = getnameinfo((struct sockaddr *)addr, len, name, sizeof(name),
-                         service, sizeof(service), 0);
-    if (rc) {
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wunused-result"
-        strerror_r(errno, errbuf, sizeof(errbuf));
-#pragma GCC diagnostic pop
-        sprintf(buf, "%s:getnameinfo errbuf=%s", __func__, errbuf);
+    if ((name = get_cached_hostname_by_addr(addr)) == NULL)
         return buf;
-    }
 
     if (inet_ntop(addr->sin_family, &addr->sin_addr.s_addr, ip, sizeof(ip))) {
-        sprintf(buf, "[%s %s:%s] ", name, ip, service);
+        sprintf(buf, "[%s %s:%hd] ", name, ip, port);
     } else {
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wunused-result"

db/comdb2.h

@@ -3021,8 +3021,8 @@ void process_nodestats(void);
 void nodestats_report(FILE *fh, const char *prefix, int disp_rates);
 void nodestats_node_report(FILE *fh, const char *prefix, int disp_rates,
                            char *host);
-struct rawnodestats *get_raw_node_stats(const char *task, const char *stack,
-                                        char *host, int fd, int is_ssl);
+struct rawnodestats *get_raw_node_stats(const char *task, const char *stack, char *host, struct sockaddr_in *addr,
+                                        int is_ssl);
 int release_node_stats(const char *task, const char *stack, char *host);
 struct summary_nodestats *get_nodestats_summary(unsigned *nodes_cnt,
                                                 int disp_rates);

db/config.c

@@ -458,7 +458,8 @@ static char *legacy_options[] = {
     "recovery_ckp 0",
     "sc_current_version 3",
     "disable_sql_table_replacement 1",
-    "endianize_locklist 0"
+    "endianize_locklist 0",
+    "get_peer_fqdn 0"
 };
 // clang-format on
 

db/db_tunables.c

@@ -624,6 +624,7 @@ int gbl_nudge_replication_when_idle = 100;
 
 extern int gbl_new_connection_grace_ms;
 extern int gbl_accept_headroom;
+extern int gbl_get_peer_fqdn;
 
 int parse_int64(const char *value, int64_t *num);
 

db/db_tunables.h

@@ -2607,5 +2607,6 @@ REGISTER_TUNABLE("new_connection_grace_ms", "Time (in ms) before new connection
                  TUNABLE_INTEGER, &gbl_new_connection_grace_ms, INTERNAL, NULL, NULL, NULL, NULL);
 REGISTER_TUNABLE("accept_headroom", "", TUNABLE_INTEGER, &gbl_accept_headroom, INTERNAL, NULL, NULL, NULL, NULL);
 REGISTER_TUNABLE("simpleauth", NULL, TUNABLE_BOOLEAN, &gbl_uses_simpleauth, NOARG | READEARLY, NULL, NULL, NULL, NULL);
-
+REGISTER_TUNABLE("get_peer_fqdn", "When set, use peer's FQDN", TUNABLE_BOOLEAN, &gbl_get_peer_fqdn, INTERNAL, NULL,
+                 NULL, NULL, NULL);
 #endif /* _DB_TUNABLES_H */

db/reqlog.c

@@ -2508,7 +2508,7 @@ void release_clientstats_lock()
     Pthread_rwlock_unlock(&clientstats_lk);
 }
 
-static void init_clientstats(nodestats_t *entry, int task_len, char *host, int fd)
+static void init_clientstats(nodestats_t *entry, int task_len, char *host, struct sockaddr_in *addr)
 {
     entry->task = entry->mem;
     entry->stack = entry->mem + task_len;
@@ -2521,21 +2521,10 @@ static void init_clientstats(nodestats_t *entry, int task_len, char *host, int f
     Pthread_mutex_init(&entry->mtx, 0);
     Pthread_mutex_init(&entry->rawtotals.lk, NULL);
 
-    if (fd < 0) {
+    if (addr == NULL) {
         bzero(&(entry->addr), sizeof(struct in_addr));
     } else {
-        struct sockaddr_in peeraddr;
-        socklen_t len = sizeof(peeraddr);
-        bzero(&peeraddr, sizeof(peeraddr));
-        if (getpeername(fd, (struct sockaddr *)&peeraddr, &len) < 0) {
-            if (errno != ENOTCONN)
-                logmsg(LOGMSG_ERROR, "%s: getpeername failed fd %d: %d %s\n", __func__, fd, errno,
-                        strerror(errno));
-            bzero(&(entry->addr), sizeof(struct in_addr));
-        } else {
-            memcpy(&(entry->addr), &peeraddr.sin_addr,
-                    sizeof(struct in_addr));
-        }
+        memcpy(&(entry->addr), &addr->sin_addr, sizeof(struct in_addr));
     }
 }
 
@@ -2549,9 +2538,8 @@ static void update_clientstats_cache(nodestats_t *entry) {
     Pthread_mutex_unlock(&clientstats_cache_mtx);
 }
 
-static nodestats_t *add_clientstats(unsigned checksum,
-                                    const char *task_and_stack, int task_len,
-                                    int stack_len, char *host, int node, int fd)
+static nodestats_t *add_clientstats(unsigned checksum, const char *task_and_stack, int task_len, int stack_len,
+                                    char *host, int node, struct sockaddr_in *addr)
 {
     nodestats_t *entry = calloc(1, offsetof(nodestats_t, mem) + task_len + stack_len);
     if (entry == NULL) {
@@ -2606,15 +2594,15 @@ static nodestats_t *add_clientstats(unsigned checksum,
         goto done;
     }
 
-    init_clientstats(entry, task_len, host, fd);
+    init_clientstats(entry, task_len, host, addr);
     hash_add(clientstats, entry);
 
 done:
     release_clientstats_lock();
     return entry;
 }
 
-static nodestats_t *find_clientstats(unsigned checksum, int node, int fd)
+static nodestats_t *find_clientstats(unsigned checksum, int node, struct sockaddr_in *addr)
 {
     nodestats_t key;
     key.checksum = checksum;
@@ -2630,19 +2618,8 @@ static nodestats_t *find_clientstats(unsigned checksum, int node, int fd)
     entry->ref++;
     update_clientstats_cache(entry);
 
-    if (*(unsigned *)&(entry->addr) == 0 && fd > 0) {
-        struct sockaddr_in peeraddr;
-        socklen_t len = sizeof(peeraddr);
-        bzero(&peeraddr, sizeof(peeraddr));
-        if (getpeername(fd, (struct sockaddr *)&peeraddr, &len) < 0) {
-            if (errno != ENOTCONN)
-                logmsg(LOGMSG_ERROR, "%s: getpeername failed fd %d: %d %s\n", __func__, fd, errno,
-                        strerror(errno));
-            bzero(&(entry->addr), sizeof(struct in_addr));
-        } else {
-            memcpy(&(entry->addr), &peeraddr.sin_addr,
-                    sizeof(struct in_addr));
-        }
+    if (*(unsigned *)&(entry->addr) == 0 && addr != NULL) {
+        memcpy(&(entry->addr), &addr->sin_addr, sizeof(struct in_addr));
     }
     Pthread_mutex_unlock(&entry->mtx);
 
@@ -2692,8 +2669,8 @@ nodestats_t *get_next_clientstats_entry(void **curr, unsigned int *iter)
     return entry;
 }
 
-struct rawnodestats *get_raw_node_stats(const char *task, const char *stack,
-                                        char *host, int fd, int is_ssl)
+struct rawnodestats *get_raw_node_stats(const char *task, const char *stack, char *host, struct sockaddr_in *addr,
+                                        int is_ssl)
 {
     unsigned checksum;
     int namelen, node;
@@ -2718,10 +2695,10 @@ struct rawnodestats *get_raw_node_stats(const char *task, const char *stack,
     memcpy(tmp, task, task_len);
     memcpy(tmp + task_len, stack, stack_len);
     checksum = crc32c((const uint8_t *)tmp, namelen);
-    
-    nodestats_t *nodestats = find_clientstats(checksum, node, fd);
+
+    nodestats_t *nodestats = find_clientstats(checksum, node, addr);
     if (!nodestats) {
-        nodestats = add_clientstats(checksum, tmp, task_len, stack_len, host, node, fd);
+        nodestats = add_clientstats(checksum, tmp, task_len, stack_len, host, node, addr);
     }
 
     if (nodestats) {

db/sltdbt.c

@@ -427,9 +427,12 @@ int handle_ireq(struct ireq *iq)
     reqlog_pushprefixf(iq->reqlogger, "%s:REQ %s ", getorigin(iq),
                        req2a(iq->opcode));
 
-    if (!iq->sorese) /* don't count osql */
-        iq->rawnodestats =
-            get_raw_node_stats(NULL, NULL, iq->frommach, cdb2buf_fileno(iq->sb), 0 /* tag does not support ssl */);
+    if (!iq->sorese) { /* don't count osql */
+        struct sockaddr_in addr;
+        socklen_t len = sizeof(addr);
+        getpeername(cdb2buf_fileno(iq->sb), (struct sockaddr *)&addr, &len);
+        iq->rawnodestats = get_raw_node_stats(NULL, NULL, iq->frommach, &addr, 0 /* tag does not support ssl */);
+    }
     if (iq->rawnodestats && iq->opcode >= 0 && iq->opcode < MAXTYPCNT)
         iq->rawnodestats->opcode_counts[iq->opcode]++;
     if (gbl_print_deadlock_cycles && IQ_HAS_SNAPINFO(iq))

db/sql.h

@@ -843,6 +843,7 @@ struct sqlclntstate {
     int authgen;
 
     char *origin;
+    struct sockaddr_in addr;
 
     TAILQ_HEAD(, session_tbl) session_tbls;
 

net/net.c

@@ -73,6 +73,7 @@
 #include "thread_util.h"
 #include <timer_util.h>
 #include <comdb2_atomic.h>
+#include <hostname_support.h>
 
 #ifdef UDP_DEBUG
 static int curr_udp_cnt = 0;
@@ -2524,23 +2525,19 @@ static int get_subnet_incomming_syn(host_node_type *host_node_ptr)
         return 0;
     }
 
-    char host[NI_MAXHOST], service[NI_MAXSERV];
     /* get our host name for local _into_ address of connection */
-    int s = getnameinfo((struct sockaddr *)&lcl_addr_inet, lcl_len, host,
-                        NI_MAXHOST, service, NI_MAXSERV, 0);
-
-    if (s != 0) {
-        logmsg(LOGMSG_WARN, "Incoming connection into unknown (%s:%u): %s\n",
-               inet_ntoa(lcl_addr_inet.sin_addr),
-               (unsigned)ntohs(lcl_addr_inet.sin_port), gai_strerror(s));
+    const char *host = get_cached_hostname_by_addr(&lcl_addr_inet);
+    if (host == NULL) {
+        logmsg(LOGMSG_WARN, "Incoming connection into unknown (%s:%u)\n", inet_ntoa(lcl_addr_inet.sin_addr),
+               (unsigned)ntohs(lcl_addr_inet.sin_port));
         return 0;
     }
 
     /* extract the suffix of subnet ex. '_n3' in name node1_n3 */
     int myh_len = strlen(host_node_ptr->netinfo_ptr->myhostname);
     if (strncmp(host_node_ptr->netinfo_ptr->myhostname, host, myh_len) == 0) {
         assert(myh_len <= sizeof(host));
-        char *subnet = &host[myh_len];
+        const char *subnet = &host[myh_len];
         if (subnet[0])
             strncpy0(host_node_ptr->subnet, subnet, HOSTNAME_LEN);
     }

plugins/newsql/newsql.c

@@ -2456,8 +2456,7 @@ newsql_loop_result newsql_loop(struct sqlclntstate *clnt, CDB2SQLQUERY *sql_quer
     }
     if (clnt->rawnodestats == NULL) {
         clnt->rawnodestats =
-            get_raw_node_stats(clnt->argv0, clnt->stack, clnt->origin,
-                               clnt->plugin.get_fileno(clnt), clnt->plugin.has_ssl(clnt));
+            get_raw_node_stats(clnt->argv0, clnt->stack, clnt->origin, &clnt->addr, clnt->plugin.has_ssl(clnt));
     }
     if (process_set_commands(clnt, sql_query)) {
         return NEWSQL_ERROR;