Use ust_lock during sock_info operations for atomicity against fork
[lttng-ust.git] / liblttng-ust / lttng-ust-comm.c
index 08e6299ce3a5bd7793e4ddf04c556e9d04e2e7f8..0cf5ecd9687536be5fc7794c4bfeb28ed135a11f 100644 (file)
@@ -35,6 +35,7 @@
 #include <time.h>
 #include <assert.h>
 #include <signal.h>
+#include <limits.h>
 #include <urcu/uatomic.h>
 #include <urcu/futex.h>
 #include <urcu/compiler.h>
 #include <lttng/ust-ctl.h>
 #include <urcu/tls-compat.h>
 #include <ust-comm.h>
+#include <ust-fd.h>
 #include <usterr-signal-safe.h>
 #include <helper.h>
 #include "tracepoint-internal.h"
 #include "lttng-tracer-core.h"
 #include "compat.h"
-#include "../libringbuffer/tlsfixup.h"
+#include "../libringbuffer/rb-init.h"
 #include "lttng-ust-statedump.h"
 #include "clock.h"
 #include "../libringbuffer/getcpu.h"
@@ -106,6 +108,14 @@ static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
 /* Should the ust comm thread quit ? */
 static int lttng_ust_comm_should_quit;
 
+/*
+ * This variable can be tested by applications to check whether
+ * lttng-ust is loaded. They simply have to define their own
+ * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
+ * library constructor.
+ */
+int lttng_ust_loaded __attribute__((weak));
+
 /*
  * Return 0 on success, -1 if should quit.
  * The lock is taken in both cases.
@@ -356,11 +366,11 @@ const char *get_lttng_home_dir(void)
 {
        const char *val;
 
-       val = (const char *) lttng_secure_getenv("LTTNG_HOME");
+       val = (const char *) lttng_getenv("LTTNG_HOME");
        if (val != NULL) {
                return val;
        }
-       return (const char *) lttng_secure_getenv("HOME");
+       return (const char *) lttng_getenv("HOME");
 }
 
 /*
@@ -388,6 +398,17 @@ void lttng_fixup_urcu_bp_tls(void)
        rcu_read_unlock();
 }
 
+void lttng_ust_fixup_tls(void)
+{
+       lttng_fixup_urcu_bp_tls();
+       lttng_fixup_ringbuffer_tls();
+       lttng_fixup_vtid_tls();
+       lttng_fixup_nest_count_tls();
+       lttng_fixup_procname_tls();
+       lttng_fixup_ust_mutex_nest_tls();
+       lttng_ust_fixup_fd_tracker_tls();
+}
+
 int lttng_get_notify_socket(void *owner)
 {
        struct sock_info *info = owner;
@@ -441,7 +462,7 @@ int setup_local_apps(void)
 }
 
 /*
- * Get notify_sock timeout, in ms.
+ * Get socket timeout, in ms.
  * -1: wait forever. 0: don't wait. >0: timeout, in ms.
  */
 static
@@ -450,20 +471,31 @@ long get_timeout(void)
        long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
 
        if (!got_timeout_env) {
-               str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
+               str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
                got_timeout_env = 1;
        }
        if (str_timeout)
                constructor_delay_ms = strtol(str_timeout, NULL, 10);
+       /* All negative values are considered as "-1". */
+       if (constructor_delay_ms < -1)
+               constructor_delay_ms = -1;
        return constructor_delay_ms;
 }
 
+/* Timeout for notify socket send and recv. */
 static
 long get_notify_sock_timeout(void)
 {
        return get_timeout();
 }
 
+/* Timeout for connecting to cmd and notify sockets. */
+static
+long get_connect_sock_timeout(void)
+{
+       return get_timeout();
+}
+
 /*
  * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
  */
@@ -502,6 +534,19 @@ int get_constructor_timeout(struct timespec *constructor_timeout)
        return 1;
 }
 
+static
+void get_allow_blocking(void)
+{
+       const char *str_allow_blocking =
+               lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
+
+       if (str_allow_blocking) {
+               DBG("%s environment variable is set",
+                       "LTTNG_UST_ALLOW_BLOCKING");
+               lttng_ust_ringbuffer_set_allow_blocking();
+       }
+}
+
 static
 int register_to_sessiond(int socket, enum ustctl_socket_type type)
 {
@@ -588,6 +633,7 @@ int handle_message(struct sock_info *sock_info,
        const struct lttng_ust_objd_ops *ops;
        struct ustcomm_ust_reply lur;
        union ust_args args;
+       char ctxstr[LTTNG_UST_SYM_NAME_LEN];    /* App context string. */
        ssize_t len;
 
        memset(&lur, 0, sizeof(lur));
@@ -807,6 +853,64 @@ int handle_message(struct sock_info *sock_info,
                        ret = -ENOSYS;
                break;
        }
+       case LTTNG_UST_CONTEXT:
+               switch (lum->u.context.ctx) {
+               case LTTNG_UST_CONTEXT_APP_CONTEXT:
+               {
+                       char *p;
+                       size_t ctxlen, recvlen;
+
+                       ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
+                                       + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
+                       if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
+                               ERR("Application context string length size is too large: %zu bytes",
+                                       ctxlen);
+                               ret = -EINVAL;
+                               goto error;
+                       }
+                       strcpy(ctxstr, "$app.");
+                       p = &ctxstr[strlen("$app.")];
+                       recvlen = ctxlen - strlen("$app.");
+                       len = ustcomm_recv_unix_sock(sock, p, recvlen);
+                       switch (len) {
+                       case 0: /* orderly shutdown */
+                               ret = 0;
+                               goto error;
+                       default:
+                               if (len == recvlen) {
+                                       DBG("app context data received");
+                                       break;
+                               } else if (len < 0) {
+                                       DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
+                                       if (len == -ECONNRESET) {
+                                               ERR("%s remote end closed connection", sock_info->name);
+                                               ret = len;
+                                               goto error;
+                                       }
+                                       ret = len;
+                                       goto error;
+                               } else {
+                                       DBG("incorrect app context data message size: %zd", len);
+                                       ret = -EINVAL;
+                                       goto error;
+                               }
+                       }
+                       /* Put : between provider and ctxname. */
+                       p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
+                       args.app_context.ctxname = ctxstr;
+                       break;
+               }
+               default:
+                       break;
+               }
+               if (ops->cmd) {
+                       ret = ops->cmd(lum->handle, lum->cmd,
+                                       (unsigned long) &lum->u,
+                                       &args, sock_info);
+               } else {
+                       ret = -ENOSYS;
+               }
+               break;
        default:
                if (ops->cmd)
                        ret = ops->cmd(lum->handle, lum->cmd,
@@ -866,6 +970,21 @@ int handle_message(struct sock_info *sock_info,
                }
        }
        DBG("Return value: %d", lur.ret_val);
+
+       ust_unlock();
+
+       /*
+        * Performed delayed statedump operations outside of the UST
+        * lock. We need to take the dynamic loader lock before we take
+        * the UST lock internally within handle_pending_statedump().
+         */
+       handle_pending_statedump(sock_info);
+
+       if (ust_lock()) {
+               ret = -LTTNG_UST_ERR_EXITING;
+               goto error;
+       }
+
        ret = send_reply(sock, &lur);
        if (ret < 0) {
                DBG("error sending reply");
@@ -896,13 +1015,6 @@ int handle_message(struct sock_info *sock_info,
 error:
        ust_unlock();
 
-       /*
-        * Performed delayed statedump operations outside of the UST
-        * lock. We need to take the dynamic loader lock before we take
-        * the UST lock internally within handle_pending_statedump().
-         */
-       handle_pending_statedump(sock_info);
-
        return ret;
 }
 
@@ -948,7 +1060,12 @@ void cleanup_sock_info(struct sock_info *sock_info, int exiting)
                long page_size;
 
                page_size = sysconf(_SC_PAGE_SIZE);
-               if (page_size > 0) {
+               if (page_size <= 0) {
+                       if (!page_size) {
+                               errno = EINVAL;
+                       }
+                       PERROR("Error in sysconf(_SC_PAGE_SIZE)");
+               } else {
                        ret = munmap(sock_info->wait_shm_mmap, page_size);
                        if (ret) {
                                ERR("Error unmapping wait shm");
@@ -1130,21 +1247,47 @@ char *get_map_shm(struct sock_info *sock_info)
        char *wait_shm_mmap;
 
        page_size = sysconf(_SC_PAGE_SIZE);
-       if (page_size < 0) {
+       if (page_size <= 0) {
+               if (!page_size) {
+                       errno = EINVAL;
+               }
+               PERROR("Error in sysconf(_SC_PAGE_SIZE)");
                goto error;
        }
 
+       lttng_ust_lock_fd_tracker();
        wait_shm_fd = get_wait_shm(sock_info, page_size);
        if (wait_shm_fd < 0) {
+               lttng_ust_unlock_fd_tracker();
+               goto error;
+       }
+
+       ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
+       if (ret < 0) {
+               ret = close(wait_shm_fd);
+               if (!ret) {
+                       PERROR("Error closing fd");
+               }
+               lttng_ust_unlock_fd_tracker();
                goto error;
        }
+
+       wait_shm_fd = ret;
+       lttng_ust_unlock_fd_tracker();
+
        wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
                  MAP_SHARED, wait_shm_fd, 0);
+
        /* close shm fd immediately after taking the mmap reference */
+       lttng_ust_lock_fd_tracker();
        ret = close(wait_shm_fd);
-       if (ret) {
+       if (!ret) {
+               lttng_ust_delete_fd_from_tracker(wait_shm_fd);
+       } else {
                PERROR("Error closing fd");
        }
+       lttng_ust_unlock_fd_tracker();
+
        if (wait_shm_mmap == MAP_FAILED) {
                DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
                goto error;
@@ -1221,9 +1364,19 @@ static
 void *ust_listener_thread(void *arg)
 {
        struct sock_info *sock_info = arg;
-       int sock, ret, prev_connect_failed = 0, has_waited = 0;
+       int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
        long timeout;
 
+       lttng_ust_fixup_tls();
+       /*
+        * If available, add '-ust' to the end of this thread's
+        * process name
+        */
+       ret = lttng_ust_setustprocname();
+       if (ret) {
+               ERR("Unable to set UST process name");
+       }
+
        /* Restart trying to connect to the session daemon */
 restart:
        if (prev_connect_failed) {
@@ -1237,12 +1390,18 @@ restart:
                         * deals with a killed or broken session daemon.
                         */
                        sleep(5);
+               } else {
+                       has_waited = 1;
                }
-               has_waited = 1;
                prev_connect_failed = 0;
        }
 
+       if (ust_lock()) {
+               goto quit;
+       }
+
        if (sock_info->socket != -1) {
+               /* FD tracker is updated by ustcomm_close_unix_sock() */
                ret = ustcomm_close_unix_sock(sock_info->socket);
                if (ret) {
                        ERR("Error closing %s ust cmd socket",
@@ -1251,6 +1410,7 @@ restart:
                sock_info->socket = -1;
        }
        if (sock_info->notify_socket != -1) {
+               /* FD tracker is updated by ustcomm_close_unix_sock() */
                ret = ustcomm_close_unix_sock(sock_info->notify_socket);
                if (ret) {
                        ERR("Error closing %s ust notify socket",
@@ -1259,6 +1419,7 @@ restart:
                sock_info->notify_socket = -1;
        }
 
+
        /*
         * Register. We need to perform both connect and sending
         * registration message before doing the next connect otherwise
@@ -1267,15 +1428,14 @@ restart:
         * first connect registration message.
         */
        /* Connect cmd socket */
-       ret = ustcomm_connect_unix_sock(sock_info->sock_path);
+       lttng_ust_lock_fd_tracker();
+       ret = ustcomm_connect_unix_sock(sock_info->sock_path,
+               get_connect_sock_timeout());
        if (ret < 0) {
+               lttng_ust_unlock_fd_tracker();
                DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
                prev_connect_failed = 1;
 
-               if (ust_lock()) {
-                       goto quit;
-               }
-
                /*
                 * If we cannot find the sessiond daemon, don't delay
                 * constructor execution.
@@ -1285,8 +1445,28 @@ restart:
                ust_unlock();
                goto restart;
        }
+       fd = ret;
+       ret = lttng_ust_add_fd_to_tracker(fd);
+       if (ret < 0) {
+               ret = close(fd);
+               if (ret) {
+                       PERROR("close on sock_info->socket");
+               }
+               ret = -1;
+               lttng_ust_unlock_fd_tracker();
+               ust_unlock();
+               goto quit;
+       }
+
        sock_info->socket = ret;
+       lttng_ust_unlock_fd_tracker();
 
+       ust_unlock();
+       /*
+        * Unlock/relock ust lock because connect is blocking (with
+        * timeout). Don't delay constructors on the ust lock for too
+        * long.
+        */
        if (ust_lock()) {
                goto quit;
        }
@@ -1321,17 +1501,24 @@ restart:
        }
 
        ust_unlock();
+       /*
+        * Unlock/relock ust lock because connect is blocking (with
+        * timeout). Don't delay constructors on the ust lock for too
+        * long.
+        */
+       if (ust_lock()) {
+               goto quit;
+       }
 
        /* Connect notify socket */
-       ret = ustcomm_connect_unix_sock(sock_info->sock_path);
+       lttng_ust_lock_fd_tracker();
+       ret = ustcomm_connect_unix_sock(sock_info->sock_path,
+               get_connect_sock_timeout());
        if (ret < 0) {
+               lttng_ust_unlock_fd_tracker();
                DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
                prev_connect_failed = 1;
 
-               if (ust_lock()) {
-                       goto quit;
-               }
-
                /*
                 * If we cannot find the sessiond daemon, don't delay
                 * constructor execution.
@@ -1341,7 +1528,32 @@ restart:
                ust_unlock();
                goto restart;
        }
+
+       fd = ret;
+       ret = lttng_ust_add_fd_to_tracker(fd);
+       if (ret < 0) {
+               ret = close(fd);
+               if (ret) {
+                       PERROR("close on sock_info->notify_socket");
+               }
+               ret = -1;
+               lttng_ust_unlock_fd_tracker();
+               ust_unlock();
+               goto quit;
+       }
+
        sock_info->notify_socket = ret;
+       lttng_ust_unlock_fd_tracker();
+
+       ust_unlock();
+       /*
+        * Unlock/relock ust lock because connect is blocking (with
+        * timeout). Don't delay constructors on the ust lock for too
+        * long.
+        */
+       if (ust_lock()) {
+               goto quit;
+       }
 
        timeout = get_notify_sock_timeout();
        if (timeout >= 0) {
@@ -1365,10 +1577,6 @@ restart:
                WARN("Unsupported timeout value %ld", timeout);
        }
 
-       if (ust_lock()) {
-               goto quit;
-       }
-
        ret = register_to_sessiond(sock_info->notify_socket,
                        USTCTL_SOCKET_NOTIFY);
        if (ret < 0) {
@@ -1486,12 +1694,9 @@ void __attribute__((constructor)) lttng_ust_init(void)
         * to be the dynamic linker mutex) and ust_lock, taken within
         * the ust lock.
         */
-       lttng_fixup_urcu_bp_tls();
-       lttng_fixup_ringbuffer_tls();
-       lttng_fixup_vtid_tls();
-       lttng_fixup_nest_count_tls();
-       lttng_fixup_procname_tls();
-       lttng_fixup_ust_mutex_nest_tls();
+       lttng_ust_fixup_tls();
+
+       lttng_ust_loaded = 1;
 
        /*
         * We want precise control over the order in which we construct
@@ -1500,7 +1705,9 @@ void __attribute__((constructor)) lttng_ust_init(void)
         * sessiond before the init functions are completed).
         */
        init_usterr();
+       lttng_ust_getenv_init();        /* Needs init_usterr() to be completed. */
        init_tracepoint();
+       lttng_ust_init_fd_tracker();
        lttng_ust_clock_init();
        lttng_ust_getcpu_init();
        lttng_ust_statedump_init();
@@ -1510,7 +1717,6 @@ void __attribute__((constructor)) lttng_ust_init(void)
        lttng_ring_buffer_client_discard_init();
        lttng_ring_buffer_client_discard_rt_init();
        lttng_perf_counter_init();
-       lttng_context_init();
        /*
         * Invoke ust malloc wrapper init before starting other threads.
         */
@@ -1518,8 +1724,12 @@ void __attribute__((constructor)) lttng_ust_init(void)
 
        timeout_mode = get_constructor_timeout(&constructor_timeout);
 
+       get_allow_blocking();
+
        ret = sem_init(&constructor_wait, 0, 0);
-       assert(!ret);
+       if (ret) {
+               PERROR("sem_init");
+       }
 
        ret = setup_local_apps();
        if (ret) {
@@ -1584,17 +1794,34 @@ void __attribute__((constructor)) lttng_ust_init(void)
                        ret = sem_timedwait(&constructor_wait,
                                        &constructor_timeout);
                } while (ret < 0 && errno == EINTR);
-               if (ret < 0 && errno == ETIMEDOUT) {
-                       ERR("Timed out waiting for lttng-sessiond");
-               } else {
-                       assert(!ret);
+               if (ret < 0) {
+                       switch (errno) {
+                       case ETIMEDOUT:
+                               ERR("Timed out waiting for lttng-sessiond");
+                               break;
+                       case EINVAL:
+                               PERROR("sem_timedwait");
+                               break;
+                       default:
+                               ERR("Unexpected error \"%s\" returned by sem_timedwait",
+                                       strerror(errno));
+                       }
                }
                break;
        case -1:/* wait forever */
                do {
                        ret = sem_wait(&constructor_wait);
                } while (ret < 0 && errno == EINTR);
-               assert(!ret);
+               if (ret < 0) {
+                       switch (errno) {
+                       case EINVAL:
+                               PERROR("sem_wait");
+                               break;
+                       default:
+                               ERR("Unexpected error \"%s\" returned by sem_wait",
+                                       strerror(errno));
+                       }
+               }
                break;
        case 0: /* no timeout */
                break;
@@ -1606,6 +1833,7 @@ void lttng_ust_cleanup(int exiting)
 {
        cleanup_sock_info(&global_apps, exiting);
        cleanup_sock_info(&local_apps, exiting);
+       local_apps.allowed = 0;
        /*
         * The teardown in this function all affect data structures
         * accessed under the UST lock by the listener thread. This
@@ -1615,7 +1843,6 @@ void lttng_ust_cleanup(int exiting)
         */
        lttng_ust_abi_exit();
        lttng_ust_events_exit();
-       lttng_context_exit();
        lttng_perf_counter_exit();
        lttng_ring_buffer_client_discard_rt_exit();
        lttng_ring_buffer_client_discard_exit();
@@ -1702,6 +1929,9 @@ void ust_before_fork(sigset_t *save_sigset)
        sigset_t all_sigs;
        int ret;
 
+       /* Fixup lttng-ust TLS. */
+       lttng_ust_fixup_tls();
+
        if (URCU_TLS(lttng_ust_nest_count))
                return;
        /* Disable signals */
@@ -1756,11 +1986,11 @@ void ust_after_fork_child(sigset_t *restore_sigset)
 {
        if (URCU_TLS(lttng_ust_nest_count))
                return;
+       lttng_context_vtid_reset();
        DBG("process %d", getpid());
        /* Release urcu mutexes */
        rcu_bp_after_fork_child();
        lttng_ust_cleanup(0);
-       lttng_context_vtid_reset();
        /* Release mutexes and reenable signals */
        ust_after_fork_common(restore_sigset);
        lttng_ust_init();
This page took 0.02977 seconds and 4 git commands to generate.