X-Git-Url: http://git.liburcu.org/?a=blobdiff_plain;f=liblttng-ust%2Flttng-ust-comm.c;h=47ba36e5c9f1d9f1871c267469049708a7dd1e1e;hb=793d29c9367a58c2a368bc1ccb11b01c2d2bdac2;hp=5df33f5f789e529a86249899b0ff4a2655757c7c;hpb=d58d145447ffcc9497801769e1fe754b30dca5b8;p=lttng-ust.git diff --git a/liblttng-ust/lttng-ust-comm.c b/liblttng-ust/lttng-ust-comm.c index 5df33f5f..47ba36e5 100644 --- a/liblttng-ust/lttng-ust-comm.c +++ b/liblttng-ust/lttng-ust-comm.c @@ -20,12 +20,14 @@ */ #define _LGPL_SOURCE +#define _GNU_SOURCE #include #include #include #include #include #include +#include #include #include #include @@ -34,6 +36,7 @@ #include #include #include +#include #include #include #include @@ -45,13 +48,20 @@ #include #include #include +#include #include #include #include "tracepoint-internal.h" #include "lttng-tracer-core.h" #include "compat.h" -#include "../libringbuffer/tlsfixup.h" -#include "lttng-ust-baddr.h" +#include "../libringbuffer/rb-init.h" +#include "lttng-ust-statedump.h" +#include "clock.h" +#include "../libringbuffer/getcpu.h" +#include "getenv.h" + +/* Concatenate lttng ust shared library name with its major version number. */ +#define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR) /* * Has lttng ust comm constructor been called ? @@ -68,9 +78,16 @@ 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. + * + * The ust fd tracker lock nests within the ust_mutex. */ static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER; @@ -86,9 +103,25 @@ static DEFINE_URCU_TLS(int, ust_mutex_nest); */ static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER; +/* + * ust_fork_mutex protects base address statedump tracing against forks. It + * prevents the dynamic loader lock to be taken (by base address statedump + * tracing) while a fork is happening, thus preventing deadlock issues with + * the dynamic loader lock. + */ +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. @@ -97,8 +130,15 @@ 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) { @@ -126,8 +166,15 @@ 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) { @@ -147,7 +194,7 @@ 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); @@ -160,6 +207,13 @@ 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"); + } } /* @@ -176,7 +230,11 @@ static sem_t constructor_wait; /* * Doing this for both the global and local sessiond. */ -static int sem_count = { 2 }; +enum { + sem_count_initial_value = 4, +}; + +static int sem_count = sem_count_initial_value; /* * Counting nesting within lttng-ust. Used to ensure that calling fork() @@ -191,7 +249,7 @@ struct sock_info { const char *name; pthread_t ust_listener; /* listener thread */ int root_handle; - int constructor_sem_posted; + int registration_done; int allowed; int global; int thread_active; @@ -204,6 +262,7 @@ struct sock_info { char *wait_shm_mmap; /* Keep track of lazy state dump not performed yet. */ int statedump_pending; + int initial_statedump_done; }; /* Socket from app (connect) to session daemon (listen) for communication */ @@ -212,7 +271,8 @@ struct sock_info global_apps = { .global = 1, .root_handle = -1, - .allowed = 1, + .registration_done = 0, + .allowed = 0, .thread_active = 0, .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME, @@ -222,6 +282,7 @@ struct sock_info global_apps = { .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME, .statedump_pending = 0, + .initial_statedump_done = 0, }; /* TODO: allow global_apps_sock_path override */ @@ -230,6 +291,7 @@ struct sock_info local_apps = { .name = "local", .global = 0, .root_handle = -1, + .registration_done = 0, .allowed = 0, /* Check setuid bit first */ .thread_active = 0, @@ -237,6 +299,7 @@ struct sock_info local_apps = { .notify_socket = -1, .statedump_pending = 0, + .initial_statedump_done = 0, }; static int wait_poll_fallback; @@ -291,6 +354,26 @@ 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); +static char *get_map_shm(struct sock_info *sock_info); + +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. @@ -300,11 +383,11 @@ const char *get_lttng_home_dir(void) { const char *val; - val = (const char *) getenv("LTTNG_HOME"); + val = (const char *) lttng_getenv("LTTNG_HOME"); if (val != NULL) { return val; } - return (const char *) getenv("HOME"); + return (const char *) lttng_getenv("HOME"); } /* @@ -322,6 +405,27 @@ 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(); +} + +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; @@ -343,25 +447,48 @@ void print_cmd(int cmd, int handle) lttng_ust_obj_get_name(handle), handle); } +static +int setup_global_apps(void) +{ + int ret = 0; + assert(!global_apps.wait_shm_mmap); + + global_apps.wait_shm_mmap = get_map_shm(&global_apps); + if (!global_apps.wait_shm_mmap) { + WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing."); + global_apps.allowed = 0; + ret = -EIO; + goto error; + } + + global_apps.allowed = 1; +error: + return ret; +} static int setup_local_apps(void) { + int ret = 0; const char *home_dir; uid_t uid; + assert(!local_apps.wait_shm_mmap); + uid = getuid(); /* * Disallow per-user tracing for setuid binaries. */ if (uid != geteuid()) { assert(local_apps.allowed == 0); - return 0; + ret = 0; + goto end; } home_dir = get_lttng_home_dir(); if (!home_dir) { WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); assert(local_apps.allowed == 0); - return -ENOENT; + ret = -ENOENT; + goto end; } local_apps.allowed = 1; snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s", @@ -371,12 +498,21 @@ int setup_local_apps(void) snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u", LTTNG_UST_WAIT_FILENAME, uid); - return 0; + + local_apps.wait_shm_mmap = get_map_shm(&local_apps); + if (!local_apps.wait_shm_mmap) { + WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing."); + local_apps.allowed = 0; + ret = -EIO; + goto end; + } +end: + return ret; } /* - * Get notify_sock timeout, in ms. - * -1: don't wait. 0: wait forever. >0: timeout, in ms. + * Get socket timeout, in ms. + * -1: wait forever. 0: don't wait. >0: timeout, in ms. */ static long get_timeout(void) @@ -384,22 +520,33 @@ 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: 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) @@ -422,7 +569,8 @@ 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 += @@ -431,9 +579,23 @@ int get_constructor_timeout(struct timespec *constructor_timeout) constructor_timeout->tv_sec++; constructor_timeout->tv_nsec -= 1000000000UL; } + /* Timeout wait (constructor_delay_ms). */ 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) { @@ -470,43 +632,85 @@ int send_reply(int sock, struct ustcomm_ust_reply *lur) } static -int handle_register_done(struct sock_info *sock_info) +void decrement_sem_count(unsigned int count) { int ret; - if (sock_info->constructor_sem_posted) - return 0; - sock_info->constructor_sem_posted = 1; + assert(uatomic_read(&sem_count) >= count); + if (uatomic_read(&sem_count) <= 0) { - return 0; + return; } - ret = uatomic_add_return(&sem_count, -1); + + ret = uatomic_add_return(&sem_count, -count); if (ret == 0) { ret = sem_post(&constructor_wait); assert(!ret); } +} + +static +int handle_register_done(struct sock_info *sock_info) +{ + if (sock_info->registration_done) + return 0; + sock_info->registration_done = 1; + + decrement_sem_count(1); + if (!sock_info->statedump_pending) { + sock_info->initial_statedump_done = 1; + decrement_sem_count(1); + } + + return 0; +} + +static +int handle_register_failed(struct sock_info *sock_info) +{ + if (sock_info->registration_done) + return 0; + sock_info->registration_done = 1; + sock_info->initial_statedump_done = 1; + + decrement_sem_count(2); + return 0; } /* * Only execute pending statedump after the constructor semaphore has - * been posted by each listener thread. This means statedump will only - * be performed after the "registration done" command is received from - * each session daemon the application is connected to. + * been posted by the current listener thread. This means statedump will + * only be performed after the "registration done" command is received + * from this thread's session daemon. * * This ensures we don't run into deadlock issues with the dynamic * loader mutex, which is held while the constructor is called and * waiting on the constructor semaphore. All operations requiring this * dynamic loader lock need to be postponed using this mechanism. + * + * In a scenario with two session daemons connected to the application, + * it is possible that the first listener thread which receives the + * registration done command issues its statedump while the dynamic + * loader lock is still held by the application constructor waiting on + * the semaphore. It will however be allowed to proceed when the + * second session daemon sends the registration done command to the + * second listener thread. This situation therefore does not produce + * a deadlock. */ static void handle_pending_statedump(struct sock_info *sock_info) { - int ctor_passed = sock_info->constructor_sem_posted; - - if (ctor_passed && sock_info->statedump_pending) { + if (sock_info->registration_done && sock_info->statedump_pending) { sock_info->statedump_pending = 0; + pthread_mutex_lock(&ust_fork_mutex); lttng_handle_pending_statedump(sock_info); + pthread_mutex_unlock(&ust_fork_mutex); + + if (!sock_info->initial_statedump_done) { + sock_info->initial_statedump_done = 1; + decrement_sem_count(1); + } } } @@ -518,19 +722,20 @@ 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) { @@ -591,12 +796,12 @@ 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; @@ -656,12 +861,12 @@ 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) { @@ -702,11 +907,11 @@ 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; @@ -723,12 +928,13 @@ int handle_message(struct sock_info *sock_info, { /* Receive shm_fd, wakeup_fd */ ret = ustcomm_recv_stream_from_sessiond(sock, - &lum->u.stream.len, + NULL, &args.stream.shm_fd, &args.stream.wakeup_fd); if (ret) { - goto end; + goto error; } + if (ops->cmd) ret = ops->cmd(lum->handle, lum->cmd, (unsigned long) &lum->u, @@ -737,6 +943,64 @@ 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, @@ -747,7 +1011,6 @@ int handle_message(struct sock_info *sock_info, break; } -end: lur.handle = lum->handle; lur.cmd = lum->cmd; lur.ret_val = ret; @@ -797,6 +1060,21 @@ end: } } 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"); @@ -827,13 +1105,6 @@ end: 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; } @@ -849,7 +1120,8 @@ void cleanup_sock_info(struct sock_info *sock_info, int exiting) } sock_info->root_handle = -1; } - sock_info->constructor_sem_posted = 0; + sock_info->registration_done = 0; + sock_info->initial_statedump_done = 0; /* * wait_shm_mmap, socket and notify socket are used by listener @@ -879,7 +1151,12 @@ 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"); @@ -1061,21 +1338,47 @@ 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; @@ -1089,40 +1392,45 @@ error: static void wait_for_sessiond(struct sock_info *sock_info) { - int ret; - + /* Use ust_lock to check if we should quit. */ if (ust_lock()) { goto quit; } if (wait_poll_fallback) { goto error; } - if (!sock_info->wait_shm_mmap) { - sock_info->wait_shm_mmap = get_map_shm(sock_info); - if (!sock_info->wait_shm_mmap) - goto error; - } ust_unlock(); + assert(sock_info->wait_shm_mmap); + 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: @@ -1145,9 +1453,19 @@ 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) { @@ -1161,12 +1479,18 @@ 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", @@ -1175,6 +1499,7 @@ 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", @@ -1183,6 +1508,7 @@ restart: sock_info->notify_socket = -1; } + /* * Register. We need to perform both connect and sending * registration message before doing the next connect otherwise @@ -1191,26 +1517,45 @@ 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. */ - ret = handle_register_done(sock_info); + ret = handle_register_failed(sock_info); assert(!ret); 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; } @@ -1238,34 +1583,66 @@ restart: * If we cannot register to the sessiond daemon, don't * delay constructor execution. */ - ret = handle_register_done(sock_info); + ret = handle_register_failed(sock_info); assert(!ret); ust_unlock(); goto 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. */ - ret = handle_register_done(sock_info); + ret = handle_register_failed(sock_info); assert(!ret); 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) { @@ -1289,10 +1666,6 @@ 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) { @@ -1303,7 +1676,7 @@ restart: * If we cannot register to the sessiond daemon, don't * delay constructor execution. */ - ret = handle_register_done(sock_info); + ret = handle_register_failed(sock_info); assert(!ret); ust_unlock(); goto restart; @@ -1333,7 +1706,7 @@ restart: * If we cannot register to the sessiond daemon, don't * delay constructor execution. */ - ret = handle_register_done(sock_info); + ret = handle_register_failed(sock_info); assert(!ret); ust_unlock(); goto end; @@ -1341,7 +1714,13 @@ 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: @@ -1395,6 +1774,7 @@ void __attribute__((constructor)) lttng_ust_init(void) pthread_attr_t thread_attr; int timeout_mode; int ret; + void *handle; if (uatomic_xchg(&initialized, 1) == 1) return; @@ -1404,11 +1784,29 @@ void __attribute__((constructor)) lttng_ust_init(void) * to be the dynamic linker mutex) and ust_lock, taken within * the ust lock. */ - 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 need to ensure that the liblttng-ust library is not unloaded to avoid + * the unloading of code used by the ust_listener_threads as we can not + * reliably know when they exited. To do that, manually load + * liblttng-ust.so to increment the dynamic loader's internal refcount for + * this library so it never becomes zero, thus never gets unloaded from the + * address space of the process. Since we are already running in the + * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will + * simply increment the refcount and no additionnal work is needed by the + * dynamic loader as the shared library is already loaded in the address + * space. As a safe guard, we use the RTLD_NODELETE flag to prevent + * unloading of the UST library if its refcount becomes zero (which should + * never happen). Do the return value check but discard the handle at the + * end of the function as it's not needed. + */ + handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE); + if (!handle) { + ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME); + } /* * We want precise control over the order in which we construct @@ -1417,15 +1815,18 @@ 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_baddr_statedump_init(); + lttng_ust_init_fd_tracker(); + 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. */ @@ -1433,11 +1834,22 @@ 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_global_apps(); + if (ret) { + assert(global_apps.allowed == 0); + DBG("global apps setup returned %d", ret); + } ret = setup_local_apps(); if (ret) { + assert(local_apps.allowed == 0); DBG("local apps setup returned %d", ret); } @@ -1461,14 +1873,18 @@ void __attribute__((constructor)) lttng_ust_init(void) ERR("pthread_attr_setdetachstate: %s", strerror(ret)); } - pthread_mutex_lock(&ust_exit_mutex); - ret = pthread_create(&global_apps.ust_listener, &thread_attr, - ust_listener_thread, &global_apps); - if (ret) { - ERR("pthread_create global: %s", strerror(ret)); + if (global_apps.allowed) { + pthread_mutex_lock(&ust_exit_mutex); + ret = pthread_create(&global_apps.ust_listener, &thread_attr, + ust_listener_thread, &global_apps); + if (ret) { + ERR("pthread_create global: %s", strerror(ret)); + } + global_apps.thread_active = 1; + pthread_mutex_unlock(&ust_exit_mutex); + } else { + handle_register_done(&global_apps); } - global_apps.thread_active = 1; - pthread_mutex_unlock(&ust_exit_mutex); if (local_apps.allowed) { pthread_mutex_lock(&ust_exit_mutex); @@ -1499,17 +1915,34 @@ 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; @@ -1520,9 +1953,9 @@ 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; + global_apps.allowed = 0; /* * The teardown in this function all affect data structures * accessed under the UST lock by the listener thread. This @@ -1532,18 +1965,17 @@ 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 */ - sem_count = 2; + sem_count = sem_count_initial_value; lttng_ust_comm_should_quit = 0; initialized = 0; } @@ -1619,6 +2051,9 @@ 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 */ @@ -1627,8 +2062,12 @@ void ust_before_fork(sigset_t *save_sigset) if (ret == -1) { PERROR("sigprocmask"); } + + pthread_mutex_lock(&ust_fork_mutex); + ust_lock_nocheck(); - rcu_bp_before_fork(); + urcu_bp_before_fork(); + lttng_ust_lock_fd_tracker(); } static void ust_after_fork_common(sigset_t *restore_sigset) @@ -1636,7 +2075,11 @@ static void ust_after_fork_common(sigset_t *restore_sigset) int ret; DBG("process %d", getpid()); + lttng_ust_unlock_fd_tracker(); ust_unlock(); + + pthread_mutex_unlock(&ust_fork_mutex); + /* Restore signals */ ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); if (ret == -1) { @@ -1649,7 +2092,7 @@ void ust_after_fork_parent(sigset_t *restore_sigset) if (URCU_TLS(lttng_ust_nest_count)) return; DBG("process %d", getpid()); - rcu_bp_after_fork_parent(); + urcu_bp_after_fork_parent(); /* Release mutexes and reenable signals */ ust_after_fork_common(restore_sigset); } @@ -1667,11 +2110,13 @@ void ust_after_fork_child(sigset_t *restore_sigset) { if (URCU_TLS(lttng_ust_nest_count)) return; + lttng_context_vpid_reset(); + lttng_context_vtid_reset(); + lttng_context_procname_reset(); DBG("process %d", getpid()); /* Release urcu mutexes */ - rcu_bp_after_fork_child(); + urcu_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();