relayd: add fd-cap option to limit the number of opened FDs

[lttng-tools.git] / src / bin / lttng-relayd / main.c

diff --git a/src/bin/lttng-relayd/main.c b/src/bin/lttng-relayd/main.c
index 749feb6f6f319a95c86d62a68af0b5f693a7c4d3..f61f1f13e69ce72eedfc53d3a1941951ae50041c 100644 (file)
--- a/src/bin/lttng-relayd/main.c
+++ b/src/bin/lttng-relayd/main.c
@@ -34,6 +34,7 @@
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/wait.h>
+#include <sys/resource.h>
 #include <inttypes.h>
 #include <urcu/futex.h>
 #include <urcu/uatomic.h>
@@ -41,6 +42,7 @@
 #include <unistd.h>
 #include <fcntl.h>
 #include <strings.h>
+#include <ctype.h>
 
 #include <lttng/lttng.h>
 #include <common/common.h>
@@ -160,6 +162,9 @@ static uint64_t last_relay_stream_id;
  */
 static struct relay_conn_queue relay_conn_queue;
 
+/* Cap of file desriptors to be in simultaneous use by the relay daemon. */
+static unsigned int lttng_opt_fd_cap;
+
 /* Global relay stream hash table. */
 struct lttng_ht *relay_streams_ht;
 
@@ -181,6 +186,7 @@ static struct option long_options[] = {
        { "daemonize", 0, 0, 'd', },
        { "background", 0, 0, 'b', },
        { "group", 1, 0, 'g', },
+       { "fd-cap", 1, 0, '\0', },
        { "help", 0, 0, 'h', },
        { "output", 1, 0, 'o', },
        { "verbose", 0, 0, 'v', },
@@ -224,9 +230,34 @@ static int set_option(int opt, const char *arg, const char *optname)
 
        switch (opt) {
        case 0:
-               fprintf(stderr, "option %s", optname);
-               if (arg) {
-                       fprintf(stderr, " with arg %s\n", arg);
+               if (!strcmp(optname, "fd-cap")) {
+                       unsigned long v;
+
+                       errno = 0;
+                       v = strtoul(arg, NULL, 0);
+                       if (errno != 0 || !isdigit(arg[0])) {
+                               ERR("Wrong value in --fd-cap parameter: %s",
+                                               arg);
+                               ret = -1;
+                               goto end;
+                       }
+                       if (v < DEFAULT_RELAYD_MINIMAL_FD_CAP) {
+                               ERR("File descriptor cap must be set to at least %d",
+                                               DEFAULT_RELAYD_MINIMAL_FD_CAP);
+                       }
+                       if (v >= UINT_MAX) {
+                               ERR("File descriptor cap overflow in --fd-cap parameter: %s",
+                                               arg);
+                               ret = -1;
+                               goto end;
+                       }
+                       lttng_opt_fd_cap = (unsigned int) v;
+                       DBG3("File descriptor cap set to %u", lttng_opt_fd_cap);
+               } else {
+                       fprintf(stderr, "unknown option %s", optname);
+                       if (arg) {
+                               fprintf(stderr, " with arg %s\n", arg);
+                       }
                }
                break;
        case 'C':
@@ -563,6 +594,18 @@ static int set_options(int argc, char **argv)
                        goto exit;
                }
        }
+       if (lttng_opt_fd_cap == 0) {
+               int ret;
+               struct rlimit rlimit;
+
+               ret = getrlimit(RLIMIT_NOFILE, &rlimit);
+               if (ret) {
+                       PERROR("Failed to get file descriptor limit");
+                       retval = -1;
+               }
+
+               lttng_opt_fd_cap = rlimit.rlim_cur;
+       }
 
        if (opt_group_output_by == RELAYD_GROUP_OUTPUT_BY_UNKNOWN) {
                opt_group_output_by = RELAYD_GROUP_OUTPUT_BY_HOST;
@@ -2372,7 +2415,6 @@ static int relay_rotate_session_streams(
                } else {
                        chunk_id_str = chunk_id_buf;
                }
-               session->has_rotated = true;
        }
 
        DBG("Rotate %" PRIu32 " streams of session \"%s\" to chunk \"%s\"",
@@ -2700,6 +2742,20 @@ static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
        }
 
        pthread_mutex_lock(&session->lock);
+       if (close_command.is_set &&
+                       close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_DELETE) {
+               /*
+                * Clear command. It is a protocol error to ask for a
+                * clear on a relay which does not allow it. Querying
+                * the configuration allows figuring out whether
+                * clearing is allowed before doing the clear.
+                */
+               if (!opt_allow_clear) {
+                       ret = -1;
+                       reply_code = LTTNG_ERR_INVALID_PROTOCOL;
+                       goto end_unlock_session;
+               }
+       }
        if (session->pending_closure_trace_chunk &&
                        session->pending_closure_trace_chunk != chunk) {
                ERR("Trace chunk close command for session \"%s\" does not target the trace chunk pending closure",
@@ -2803,6 +2859,10 @@ static int relay_close_trace_chunk(const struct lttcomm_relayd_hdr *recv_hdr,
                        goto end_unlock_session;
                }
        }
+       if (close_command.is_set &&
+                       close_command.value == LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED) {
+               session->has_rotated = true;
+       }
        DBG("Reply chunk path on close: %s", closed_trace_chunk_path);
        path_length = strlen(closed_trace_chunk_path) + 1;
        if (path_length > UINT32_MAX) {