Fix: reset vtid cache before releasing urcu locks
[lttng-ust.git] / liblttng-ust / lttng-ust-comm.c
index a73808e1128f7b63fbb62422e5f7bccc5115a9b4..a84316a9358caeda6be0138a439c3828dd979b8e 100644 (file)
@@ -20,6 +20,7 @@
  */
 
 #define _LGPL_SOURCE
+#define _GNU_SOURCE
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <sys/mman.h>
 #include "lttng-tracer-core.h"
 #include "compat.h"
 #include "../libringbuffer/tlsfixup.h"
-#include "lttng-ust-baddr.h"
+#include "lttng-ust-statedump.h"
+#include "clock.h"
+#include "../libringbuffer/getcpu.h"
+#include "getenv.h"
 
 /*
  * Has lttng ust comm constructor been called ?
@@ -68,9 +72,14 @@ static int initialized;
  *
  * ust_exit_mutex must never nest in ust_mutex.
  *
+ * ust_fork_mutex must never nest in ust_mutex.
+ *
  * ust_mutex_nest is a per-thread nesting counter, allowing the perf
  * counter lazy initialization called by events within the statedump,
  * which traces while the ust_mutex is held.
+ *
+ * ust_lock nests within the dynamic loader lock (within glibc) because
+ * it is taken within the library constructor.
  */
 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
 
@@ -97,6 +106,14 @@ 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.
@@ -105,8 +122,15 @@ static int lttng_ust_comm_should_quit;
 int ust_lock(void)
 {
        sigset_t sig_all_blocked, orig_mask;
-       int ret;
+       int ret, oldstate;
 
+       ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
+       if (ret) {
+               ERR("pthread_setcancelstate: %s", strerror(ret));
+       }
+       if (oldstate != PTHREAD_CANCEL_ENABLE) {
+               ERR("pthread_setcancelstate: unexpected oldstate");
+       }
        sigfillset(&sig_all_blocked);
        ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
        if (ret) {
@@ -134,8 +158,15 @@ int ust_lock(void)
 void ust_lock_nocheck(void)
 {
        sigset_t sig_all_blocked, orig_mask;
-       int ret;
+       int ret, oldstate;
 
+       ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
+       if (ret) {
+               ERR("pthread_setcancelstate: %s", strerror(ret));
+       }
+       if (oldstate != PTHREAD_CANCEL_ENABLE) {
+               ERR("pthread_setcancelstate: unexpected oldstate");
+       }
        sigfillset(&sig_all_blocked);
        ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
        if (ret) {
@@ -155,7 +186,7 @@ void ust_lock_nocheck(void)
 void ust_unlock(void)
 {
        sigset_t sig_all_blocked, orig_mask;
-       int ret;
+       int ret, oldstate;
 
        sigfillset(&sig_all_blocked);
        ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
@@ -168,6 +199,13 @@ void ust_unlock(void)
        if (ret) {
                ERR("pthread_sigmask: %s", strerror(ret));
        }
+       ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
+       if (ret) {
+               ERR("pthread_setcancelstate: %s", strerror(ret));
+       }
+       if (oldstate != PTHREAD_CANCEL_DISABLE) {
+               ERR("pthread_setcancelstate: unexpected oldstate");
+       }
 }
 
 /*
@@ -299,6 +337,24 @@ extern void lttng_ring_buffer_client_discard_exit(void);
 extern void lttng_ring_buffer_client_discard_rt_exit(void);
 extern void lttng_ring_buffer_metadata_client_exit(void);
 
+ssize_t lttng_ust_read(int fd, void *buf, size_t len)
+{
+       ssize_t ret;
+       size_t copied = 0, to_copy = len;
+
+       do {
+               ret = read(fd, buf + copied, to_copy);
+               if (ret > 0) {
+                       copied += ret;
+                       to_copy -= ret;
+               }
+       } while ((ret > 0 && to_copy > 0)
+               || (ret < 0 && errno == EINTR));
+       if (ret > 0) {
+               ret = copied;
+       }
+       return ret;
+}
 /*
  * Returns the HOME directory path. Caller MUST NOT free(3) the returned
  * pointer.
@@ -308,11 +364,11 @@ const char *get_lttng_home_dir(void)
 {
        const char *val;
 
-       val = (const char *) getenv("LTTNG_HOME");
+       val = (const char *) lttng_secure_getenv("LTTNG_HOME");
        if (val != NULL) {
                return val;
        }
-       return (const char *) getenv("HOME");
+       return (const char *) lttng_secure_getenv("HOME");
 }
 
 /*
@@ -330,6 +386,16 @@ void lttng_fixup_ust_mutex_nest_tls(void)
        asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
 }
 
+/*
+ * Fixup urcu bp TLS.
+ */
+static
+void lttng_fixup_urcu_bp_tls(void)
+{
+       rcu_read_lock();
+       rcu_read_unlock();
+}
+
 int lttng_get_notify_socket(void *owner)
 {
        struct sock_info *info = owner;
@@ -384,7 +450,7 @@ int setup_local_apps(void)
 
 /*
  * Get notify_sock timeout, in ms.
- * -1: don't wait. 0: wait forever. >0: timeout, in ms.
+ * -1: wait forever. 0: don't wait. >0: timeout, in ms.
  */
 static
 long get_timeout(void)
@@ -397,6 +463,9 @@ long get_timeout(void)
        }
        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;
 }
 
@@ -407,7 +476,7 @@ long get_notify_sock_timeout(void)
 }
 
 /*
- * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
+ * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
  */
 static
 int get_constructor_timeout(struct timespec *constructor_timeout)
@@ -430,7 +499,8 @@ int get_constructor_timeout(struct timespec *constructor_timeout)
         */
        ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
        if (ret) {
-               return -1;
+               /* Don't wait. */
+               return 0;
        }
        constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
        constructor_timeout->tv_nsec +=
@@ -439,6 +509,7 @@ int get_constructor_timeout(struct timespec *constructor_timeout)
                constructor_timeout->tv_sec++;
                constructor_timeout->tv_nsec -= 1000000000UL;
        }
+       /* Timeout wait (constructor_delay_ms). */
        return 1;
 }
 
@@ -513,8 +584,8 @@ void handle_pending_statedump(struct sock_info *sock_info)
        int ctor_passed = sock_info->constructor_sem_posted;
 
        if (ctor_passed && sock_info->statedump_pending) {
-               pthread_mutex_lock(&ust_fork_mutex);
                sock_info->statedump_pending = 0;
+               pthread_mutex_lock(&ust_fork_mutex);
                lttng_handle_pending_statedump(sock_info);
                pthread_mutex_unlock(&ust_fork_mutex);
        }
@@ -528,19 +599,20 @@ 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));
 
        if (ust_lock()) {
                ret = -LTTNG_UST_ERR_EXITING;
-               goto end;
+               goto error;
        }
 
        ops = objd_ops(lum->handle);
        if (!ops) {
                ret = -ENOENT;
-               goto end;
+               goto error;
        }
 
        switch (lum->cmd) {
@@ -601,12 +673,12 @@ int handle_message(struct sock_info *sock_info,
                                }
                                ret = len;
                                free(bytecode);
-                               goto end;
+                               goto error;
                        } else {
                                DBG("incorrect filter data message size: %zd", len);
                                ret = -EINVAL;
                                free(bytecode);
-                               goto end;
+                               goto error;
                        }
                }
                bytecode->bc.len = lum->u.filter.data_size;
@@ -666,12 +738,12 @@ int handle_message(struct sock_info *sock_info,
                                }
                                ret = len;
                                free(node);
-                               goto end;
+                               goto error;
                        } else {
                                DBG("Incorrect exclusion data message size: %zd", len);
                                ret = -EINVAL;
                                free(node);
-                               goto end;
+                               goto error;
                        }
                }
                if (ops->cmd) {
@@ -712,11 +784,11 @@ int handle_message(struct sock_info *sock_info,
                                        goto error;
                                }
                                ret = len;
-                               goto end;
+                               goto error;
                        } else {
                                DBG("incorrect channel data message size: %zd", len);
                                ret = -EINVAL;
-                               goto end;
+                               goto error;
                        }
                }
                args.channel.chan_data = chan_data;
@@ -737,7 +809,7 @@ int handle_message(struct sock_info *sock_info,
                        &args.stream.shm_fd,
                        &args.stream.wakeup_fd);
                if (ret) {
-                       goto end;
+                       goto error;
                }
                if (ops->cmd)
                        ret = ops->cmd(lum->handle, lum->cmd,
@@ -747,6 +819,64 @@ 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,
@@ -757,7 +887,6 @@ int handle_message(struct sock_info *sock_info,
                break;
        }
 
-end:
        lur.handle = lum->handle;
        lur.cmd = lum->cmd;
        lur.ret_val = ret;
@@ -889,7 +1018,12 @@ 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");
@@ -1071,7 +1205,11 @@ 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;
        }
 
@@ -1099,8 +1237,6 @@ error:
 static
 void wait_for_sessiond(struct sock_info *sock_info)
 {
-       int ret;
-
        if (ust_lock()) {
                goto quit;
        }
@@ -1116,23 +1252,32 @@ void wait_for_sessiond(struct sock_info *sock_info)
 
        DBG("Waiting for %s apps sessiond", sock_info->name);
        /* Wait for futex wakeup */
-       if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
-               ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
-                       FUTEX_WAIT, 0, NULL, NULL, 0);
-               if (ret < 0) {
-                       if (errno == EFAULT) {
-                               wait_poll_fallback = 1;
-                               DBG(
+       if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
+               goto end_wait;
+
+       while (futex_async((int32_t *) sock_info->wait_shm_mmap,
+                       FUTEX_WAIT, 0, NULL, NULL, 0)) {
+               switch (errno) {
+               case EWOULDBLOCK:
+                       /* Value already changed. */
+                       goto end_wait;
+               case EINTR:
+                       /* Retry if interrupted by signal. */
+                       break;  /* Get out of switch. */
+               case EFAULT:
+                       wait_poll_fallback = 1;
+                       DBG(
 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
 "Please upgrade your kernel "
 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
 "mainline). LTTng-UST will use polling mode fallback.");
-                               if (ust_debug())
-                                       PERROR("futex");
-                       }
+                       if (ust_debug())
+                               PERROR("futex");
+                       goto end_wait;
                }
        }
+end_wait:
        return;
 
 quit:
@@ -1158,6 +1303,15 @@ void *ust_listener_thread(void *arg)
        int sock, ret, prev_connect_failed = 0, has_waited = 0;
        long timeout;
 
+       /*
+        * 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) {
@@ -1171,8 +1325,9 @@ restart:
                         * deals with a killed or broken session daemon.
                         */
                        sleep(5);
+               } else {
+                       has_waited = 1;
                }
-               has_waited = 1;
                prev_connect_failed = 0;
        }
 
@@ -1351,7 +1506,13 @@ restart:
                        print_cmd(lum.cmd, lum.handle);
                        ret = handle_message(sock_info, sock, &lum);
                        if (ret) {
-                               ERR("Error handling message for %s socket", sock_info->name);
+                               ERR("Error handling message for %s socket",
+                                       sock_info->name);
+                               /*
+                                * Close socket if protocol error is
+                                * detected.
+                                */
+                               goto end;
                        }
                        continue;
                default:
@@ -1414,12 +1575,15 @@ 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_loaded = 1;
+
        /*
         * We want precise control over the order in which we construct
         * our sub-libraries vs starting to receive commands from
@@ -1428,14 +1592,15 @@ void __attribute__((constructor)) lttng_ust_init(void)
         */
        init_usterr();
        init_tracepoint();
-       lttng_ust_baddr_statedump_init();
+       lttng_ust_clock_init();
+       lttng_ust_getcpu_init();
+       lttng_ust_statedump_init();
        lttng_ring_buffer_metadata_client_init();
        lttng_ring_buffer_client_overwrite_init();
        lttng_ring_buffer_client_overwrite_rt_init();
        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.
         */
@@ -1444,7 +1609,9 @@ void __attribute__((constructor)) lttng_ust_init(void)
        timeout_mode = get_constructor_timeout(&constructor_timeout);
 
        ret = sem_init(&constructor_wait, 0, 0);
-       assert(!ret);
+       if (ret) {
+               PERROR("sem_init");
+       }
 
        ret = setup_local_apps();
        if (ret) {
@@ -1509,17 +1676,34 @@ 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;
@@ -1530,9 +1714,8 @@ static
 void lttng_ust_cleanup(int exiting)
 {
        cleanup_sock_info(&global_apps, exiting);
-       if (local_apps.allowed) {
-               cleanup_sock_info(&local_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
@@ -1542,14 +1725,13 @@ 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();
        lttng_ring_buffer_client_overwrite_rt_exit();
        lttng_ring_buffer_client_overwrite_exit();
        lttng_ring_buffer_metadata_client_exit();
-       lttng_ust_baddr_statedump_destroy();
+       lttng_ust_statedump_destroy();
        exit_tracepoint();
        if (!exiting) {
                /* Reinitialize values for fork */
@@ -1683,11 +1865,11 @@ 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 0.030458 seconds and 4 git commands to generate.