X-Git-Url: http://git.lttng.org/?p=lttng-ust.git;a=blobdiff_plain;f=liblttng-ust%2Flttng-ust-comm.c;h=3bddf76744216c4d684e940290f0089e67c13c43;hp=5c4e4e76609b9d637ace3622505c27304f19d618;hb=bf5ff35ed0a3a7f60e92d654a5b97e73b94da852;hpb=1cb5dd98436f121126596a31368b50f938b1fc16 diff --git a/liblttng-ust/lttng-ust-comm.c b/liblttng-ust/lttng-ust-comm.c index 5c4e4e76..3bddf767 100644 --- a/liblttng-ust/lttng-ust-comm.c +++ b/liblttng-ust/lttng-ust-comm.c @@ -22,7 +22,6 @@ #define _LGPL_SOURCE #include #include -#include #include #include #include @@ -46,6 +45,8 @@ #include #include "tracepoint-internal.h" #include "ltt-tracer-core.h" +#include "compat.h" +#include "../libringbuffer/tlsfixup.h" /* * Has lttng ust comm constructor been called ? @@ -77,6 +78,12 @@ static sem_t constructor_wait; */ static int sem_count = { 2 }; +/* + * Counting nesting within lttng-ust. Used to ensure that calling fork() + * from liblttng-ust does not execute the pre/post fork handlers. + */ +static int __thread lttng_ust_nest_count; + /* * Info about socket and associated listener thread. */ @@ -129,6 +136,15 @@ extern void ltt_ring_buffer_client_overwrite_exit(void); extern void ltt_ring_buffer_client_discard_exit(void); extern void ltt_ring_buffer_metadata_client_exit(void); +/* + * Force a read (imply TLS fixup for dlopen) of TLS variables. + */ +static +void lttng_fixup_nest_count_tls(void) +{ + asm volatile ("" : : "m" (lttng_ust_nest_count)); +} + static int setup_local_apps(void) { @@ -140,14 +156,16 @@ int setup_local_apps(void) * Disallow per-user tracing for setuid binaries. */ if (uid != geteuid()) { - local_apps.allowed = 0; + assert(local_apps.allowed == 0); return 0; - } else { - local_apps.allowed = 1; } home_dir = (const char *) getenv("HOME"); - if (!home_dir) + if (!home_dir) { + WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); + assert(local_apps.allowed == 0); return -ENOENT; + } + local_apps.allowed = 1; snprintf(local_apps.sock_path, PATH_MAX, DEFAULT_HOME_APPS_UNIX_SOCK, home_dir); snprintf(local_apps.wait_shm_path, PATH_MAX, @@ -159,7 +177,6 @@ static int register_app_to_sessiond(int socket) { ssize_t ret; - int prctl_ret; struct { uint32_t major; uint32_t minor; @@ -178,11 +195,7 @@ int register_app_to_sessiond(int socket) reg_msg.uid = getuid(); reg_msg.gid = getgid(); reg_msg.bits_per_long = CAA_BITS_PER_LONG; - prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0); - if (prctl_ret) { - ERR("Error executing prctl"); - return -errno; - } + lttng_ust_getprocname(reg_msg.name); ret = ustcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg)); if (ret >= 0 && ret != sizeof(reg_msg)) @@ -239,6 +252,8 @@ int handle_message(struct sock_info *sock_info, const struct lttng_ust_objd_ops *ops; struct ustcomm_ust_reply lur; int shm_fd, wait_fd; + char *shm_path, *wait_pipe_path; + union ust_args args; ust_lock(); @@ -271,7 +286,8 @@ int handle_message(struct sock_info *sock_info, default: if (ops->cmd) ret = ops->cmd(lum->handle, lum->cmd, - (unsigned long) &lum->u); + (unsigned long) &lum->u, + &args); else ret = -ENOSYS; break; @@ -287,28 +303,34 @@ end: //lur.ret_code = USTCOMM_SESSION_FAIL; lur.ret_code = ret; } - switch (lum->cmd) { - case LTTNG_UST_STREAM: - /* - * Special-case reply to send stream info. - * Use lum.u output. - */ - lur.u.stream.memory_map_size = lum->u.stream.memory_map_size; - shm_fd = lum->u.stream.shm_fd; - wait_fd = lum->u.stream.wait_fd; - break; - case LTTNG_UST_METADATA: - case LTTNG_UST_CHANNEL: - lur.u.channel.memory_map_size = lum->u.channel.memory_map_size; - shm_fd = lum->u.channel.shm_fd; - wait_fd = lum->u.channel.wait_fd; - break; - case LTTNG_UST_TRACER_VERSION: - lur.u.version = lum->u.version; - break; - case LTTNG_UST_TRACEPOINT_LIST_GET: - memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); - break; + if (ret >= 0) { + switch (lum->cmd) { + case LTTNG_UST_STREAM: + /* + * Special-case reply to send stream info. + * Use lum.u output. + */ + lur.u.stream.memory_map_size = *args.stream.memory_map_size; + shm_fd = *args.stream.shm_fd; + shm_path = args.stream.shm_path; + wait_fd = *args.stream.wait_fd; + wait_pipe_path = args.stream.wait_pipe_path; + break; + case LTTNG_UST_METADATA: + case LTTNG_UST_CHANNEL: + lur.u.channel.memory_map_size = *args.channel.memory_map_size; + shm_fd = *args.channel.shm_fd; + shm_path = args.channel.shm_path; + wait_fd = *args.channel.wait_fd; + wait_pipe_path = args.channel.wait_pipe_path; + break; + case LTTNG_UST_TRACER_VERSION: + lur.u.version = lum->u.version; + break; + case LTTNG_UST_TRACEPOINT_LIST_GET: + memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); + break; + } } ret = send_reply(sock, &lur); if (ret < 0) { @@ -320,34 +342,84 @@ end: || lum->cmd == LTTNG_UST_CHANNEL || lum->cmd == LTTNG_UST_METADATA) && lur.ret_code == USTCOMM_OK) { - /* we also need to send the file descriptors. */ - ret = ustcomm_send_fds_unix_sock(sock, - &shm_fd, &shm_fd, - 1, sizeof(int)); + int sendret = 0; + + /* send the shm path */ + ret = ustcomm_send_string(sock, shm_path, strlen(shm_path)); if (ret < 0) { - perror("send shm_fd"); - goto error; + perror("send shm_path"); + sendret = ret; } - ret = ustcomm_send_fds_unix_sock(sock, - &wait_fd, &wait_fd, - 1, sizeof(int)); + /* + * The sessiond expects 2 file descriptors, even upon + * error. + */ + ret = ustcomm_send_string(sock, wait_pipe_path, strlen(wait_pipe_path)); if (ret < 0) { - perror("send wait_fd"); + perror("send wait_pipe_path"); + goto error; + } + if (sendret) { + ret = sendret; goto error; } } + /* + * We still have the memory map reference, and the fds have been + * sent to the sessiond. We can therefore close those fds. Note + * that we keep the write side of the wait_fd open, but close + * the read side. + */ + if (lur.ret_code == USTCOMM_OK) { + switch (lum->cmd) { + case LTTNG_UST_STREAM: + if (shm_fd >= 0) { + ret = close(shm_fd); + if (ret) { + PERROR("Error closing stream shm_fd"); + } + *args.stream.shm_fd = -1; + } + if (wait_fd >= 0) { + ret = close(wait_fd); + if (ret) { + PERROR("Error closing stream wait_fd"); + } + *args.stream.wait_fd = -1; + } + break; + case LTTNG_UST_METADATA: + case LTTNG_UST_CHANNEL: + if (shm_fd >= 0) { + ret = close(shm_fd); + if (ret) { + PERROR("Error closing channel shm_fd"); + } + *args.channel.shm_fd = -1; + } + if (wait_fd >= 0) { + ret = close(wait_fd); + if (ret) { + PERROR("Error closing channel wait_fd"); + } + *args.channel.wait_fd = -1; + } + break; + } + } + error: ust_unlock(); return ret; } static -void cleanup_sock_info(struct sock_info *sock_info) +void cleanup_sock_info(struct sock_info *sock_info, int exiting) { int ret; if (sock_info->socket != -1) { - ret = close(sock_info->socket); + ret = ustcomm_close_unix_sock(sock_info->socket); if (ret) { ERR("Error closing apps socket"); } @@ -361,7 +433,13 @@ void cleanup_sock_info(struct sock_info *sock_info) sock_info->root_handle = -1; } sock_info->constructor_sem_posted = 0; - if (sock_info->wait_shm_mmap) { + /* + * wait_shm_mmap is used by listener threads outside of the + * ust lock, so we cannot tear it down ourselves, because we + * cannot join on these threads. Leave this task to the OS + * process exit. + */ + if (!exiting && sock_info->wait_shm_mmap) { ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); if (ret) { ERR("Error unmapping wait shm"); @@ -404,7 +482,9 @@ int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) * If the open failed because the file did not exist, try * creating it ourself. */ + lttng_ust_nest_count++; pid = fork(); + lttng_ust_nest_count--; if (pid > 0) { int status; @@ -413,7 +493,7 @@ int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) * shared memory map will have been created. */ pid = wait(&status); - if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { + if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { wait_shm_fd = -1; goto end; } @@ -453,9 +533,9 @@ int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) ret = ftruncate(wait_shm_fd, mmap_size); if (ret) { PERROR("ftruncate"); - exit(EXIT_FAILURE); + _exit(EXIT_FAILURE); } - exit(EXIT_SUCCESS); + _exit(EXIT_SUCCESS); } /* * For local shm, we need to have rw access to accept @@ -467,13 +547,13 @@ int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) */ if (!sock_info->global && errno != EACCES) { ERR("Error opening shm %s", sock_info->wait_shm_path); - exit(EXIT_FAILURE); + _exit(EXIT_FAILURE); } /* * The shm exists, but we cannot open it RW. Report * success. */ - exit(EXIT_SUCCESS); + _exit(EXIT_SUCCESS); } else { return -1; } @@ -567,7 +647,8 @@ void wait_for_sessiond(struct sock_info *sock_info) "Please upgrade your kernel " "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " "mainline). LTTng-UST will use polling mode fallback."); - PERROR("futex"); + if (ust_debug()) + PERROR("futex"); } } } @@ -620,7 +701,7 @@ restart: } if (sock_info->socket != -1) { - ret = close(sock_info->socket); + ret = ustcomm_close_unix_sock(sock_info->socket); if (ret) { ERR("Error closing %s apps socket", sock_info->name); } @@ -771,12 +852,24 @@ int get_timeout(struct timespec *constructor_timeout) void __attribute__((constructor)) lttng_ust_init(void) { struct timespec constructor_timeout; + sigset_t sig_all_blocked, orig_parent_mask; + pthread_attr_t thread_attr; int timeout_mode; int ret; if (uatomic_xchg(&initialized, 1) == 1) return; + /* + * Fixup interdependency between TLS fixup mutex (which happens + * to be the dynamic linker mutex) and ust_lock, taken within + * the ust lock. + */ + lttng_fixup_event_tls(); + lttng_fixup_ringbuffer_tls(); + lttng_fixup_vtid_tls(); + lttng_fixup_nest_count_tls(); + /* * We want precise control over the order in which we construct * our sub-libraries vs starting to receive commands from @@ -796,17 +889,53 @@ void __attribute__((constructor)) lttng_ust_init(void) ret = setup_local_apps(); if (ret) { - ERR("Error setting up to local apps"); + DBG("local apps setup returned %d", ret); + } + + /* A new thread created by pthread_create inherits the signal mask + * from the parent. To avoid any signal being received by the + * listener thread, we block all signals temporarily in the parent, + * while we create the listener thread. + */ + sigfillset(&sig_all_blocked); + ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask); + if (ret) { + ERR("pthread_sigmask: %s", strerror(ret)); + } + + ret = pthread_attr_init(&thread_attr); + if (ret) { + ERR("pthread_attr_init: %s", strerror(ret)); + } + ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED); + if (ret) { + ERR("pthread_attr_setdetachstate: %s", strerror(ret)); } - ret = pthread_create(&local_apps.ust_listener, NULL, - ust_listener_thread, &local_apps); + ret = pthread_create(&global_apps.ust_listener, &thread_attr, + ust_listener_thread, &global_apps); + if (ret) { + ERR("pthread_create global: %s", strerror(ret)); + } if (local_apps.allowed) { - ret = pthread_create(&global_apps.ust_listener, NULL, - ust_listener_thread, &global_apps); + ret = pthread_create(&local_apps.ust_listener, &thread_attr, + ust_listener_thread, &local_apps); + if (ret) { + ERR("pthread_create local: %s", strerror(ret)); + } } else { handle_register_done(&local_apps); } + ret = pthread_attr_destroy(&thread_attr); + if (ret) { + ERR("pthread_attr_destroy: %s", strerror(ret)); + } + + /* Restore original signal mask in parent */ + ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL); + if (ret) { + ERR("pthread_sigmask: %s", strerror(ret)); + } switch (timeout_mode) { case 1: /* timeout wait */ @@ -834,10 +963,17 @@ void __attribute__((constructor)) lttng_ust_init(void) static void lttng_ust_cleanup(int exiting) { - cleanup_sock_info(&global_apps); + cleanup_sock_info(&global_apps, exiting); if (local_apps.allowed) { - cleanup_sock_info(&local_apps); + cleanup_sock_info(&local_apps, exiting); } + /* + * The teardown in this function all affect data structures + * accessed under the UST lock by the listener thread. This + * lock, along with the lttng_ust_comm_should_quit flag, ensure + * that none of these threads are accessing this data at this + * point. + */ lttng_ust_abi_exit(); lttng_ust_events_exit(); ltt_ring_buffer_client_discard_exit(); @@ -871,16 +1007,27 @@ void __attribute__((destructor)) lttng_ust_exit(void) lttng_ust_comm_should_quit = 1; ust_unlock(); + /* cancel threads */ ret = pthread_cancel(global_apps.ust_listener); if (ret) { - ERR("Error cancelling global ust listener thread"); + ERR("Error cancelling global ust listener thread: %s", + strerror(ret)); } if (local_apps.allowed) { ret = pthread_cancel(local_apps.ust_listener); if (ret) { - ERR("Error cancelling local ust listener thread"); + ERR("Error cancelling local ust listener thread: %s", + strerror(ret)); } } + /* + * Do NOT join threads: use of sys_futex makes it impossible to + * join the threads without using async-cancel, but async-cancel + * is delivered by a signal, which could hit the target thread + * anywhere in its code path, including while the ust_lock() is + * held, causing a deadlock for the other thread. Let the OS + * cleanup the threads if there are stalled in a syscall. + */ lttng_ust_cleanup(1); } @@ -902,6 +1049,8 @@ void ust_before_fork(sigset_t *save_sigset) sigset_t all_sigs; int ret; + if (lttng_ust_nest_count) + return; /* Disable signals */ sigfillset(&all_sigs); ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); @@ -927,6 +1076,8 @@ static void ust_after_fork_common(sigset_t *restore_sigset) void ust_after_fork_parent(sigset_t *restore_sigset) { + if (lttng_ust_nest_count) + return; DBG("process %d", getpid()); rcu_bp_after_fork_parent(); /* Release mutexes and reenable signals */ @@ -944,6 +1095,8 @@ void ust_after_fork_parent(sigset_t *restore_sigset) */ void ust_after_fork_child(sigset_t *restore_sigset) { + if (lttng_ust_nest_count) + return; DBG("process %d", getpid()); /* Release urcu mutexes */ rcu_bp_after_fork_child();