X-Git-Url: http://git.efficios.com/?p=lttng-tools.git;a=blobdiff_plain;f=ltt-sessiond%2Fmain.c;h=d2240d9dd89a1f9cec011e233955b9c33a441b7e;hp=64493521ba77e6edf5df99d097683b930205a0ae;hb=e848fc768a950acc1b823fcb139057a4fac1442a;hpb=773168b7255d45fd28f018055292e93b7a76482d diff --git a/ltt-sessiond/main.c b/ltt-sessiond/main.c index 64493521b..d2240d9dd 100644 --- a/ltt-sessiond/main.c +++ b/ltt-sessiond/main.c @@ -1,6 +1,6 @@ /* * Copyright (C) 2011 - David Goulet - * Copyright (C) 2011 - Mathieu Desnoyers + * Mathieu Desnoyers * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License @@ -22,36 +22,37 @@ #include #include #include -#include #include #include #include #include #include #include -#include +#include #include -#include +#include #include #include #include -#include -#include +#include +#include #include -#include /* URCU list library (-lurcu) */ -#include -#include +#include +#include +#include +#include +#include "compat/poll.h" #include "context.h" -#include "ltt-sessiond.h" -#include "lttngerr.h" +#include "futex.h" #include "kernel-ctl.h" -#include "ust-ctl.h" -#include "session.h" +#include "ltt-sessiond.h" +#include "shm.h" #include "traceable-app.h" -#include "lttng-kconsumerd.h" +#include "ust-ctl.h" #include "utils.h" +#include "ust-ctl.h" /* Const values */ const char default_home_dir[] = DEFAULT_HOME_DIR; @@ -71,12 +72,13 @@ static int opt_daemon; static int is_root; /* Set to 1 if the daemon is running as root */ static pid_t ppid; /* Parent PID for --sig-parent option */ static pid_t kconsumerd_pid; -static struct pollfd *kernel_pollfd; +static int dispatch_thread_exit; static char apps_unix_sock_path[PATH_MAX]; /* Global application Unix socket path */ static char client_unix_sock_path[PATH_MAX]; /* Global client Unix socket path */ static char kconsumerd_err_unix_sock_path[PATH_MAX]; /* kconsumerd error Unix socket path */ static char kconsumerd_cmd_unix_sock_path[PATH_MAX]; /* kconsumerd command Unix socket path */ +static char wait_shm_path[PATH_MAX]; /* global wait shm path for UST */ /* Sockets and FDs */ static int client_sock; @@ -92,16 +94,33 @@ static int kernel_poll_pipe[2]; */ static int thread_quit_pipe[2]; +/* + * This pipe is used to inform the thread managing application communication + * that a command is queued and ready to be processed. + */ +static int apps_cmd_pipe[2]; + /* Pthread, Mutexes and Semaphores */ static pthread_t kconsumerd_thread; static pthread_t apps_thread; +static pthread_t reg_apps_thread; static pthread_t client_thread; static pthread_t kernel_thread; +static pthread_t dispatch_thread; static sem_t kconsumerd_sem; -static pthread_mutex_t kconsumerd_pid_mutex; /* Mutex to control kconsumerd pid assignation */ -static int modprobe_remove_kernel_modules(void); +/* Mutex to control kconsumerd pid assignation */ +static pthread_mutex_t kconsumerd_pid_mutex; + +/* + * UST registration command queue. This queue is tied with a futex and uses a N + * wakers / 1 waiter implemented and detailed in futex.c/.h + * + * The thread_manage_apps and thread_dispatch_ust_registration interact with + * this queue and the wait/wake scheme. + */ +static struct ust_cmd_queue ust_cmd_queue; /* * Pointer initialized before thread creation. @@ -115,6 +134,88 @@ static int modprobe_remove_kernel_modules(void); */ static struct ltt_session_list *session_list_ptr; +/* + * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. + */ +static int create_thread_poll_set(struct lttng_poll_event *events, + unsigned int size) +{ + int ret; + + if (events == NULL || size == 0) { + ret = -1; + goto error; + } + + ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); + if (ret < 0) { + goto error; + } + + /* Add quit pipe */ + ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN); + if (ret < 0) { + goto error; + } + + return 0; + +error: + return ret; +} + +/* + * Check if the thread quit pipe was triggered. + * + * Return 1 if it was triggered else 0; + */ +static int check_thread_quit_pipe(int fd, uint32_t events) +{ + if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { + return 1; + } + + return 0; +} + +/* + * Remove modules in reverse load order. + */ +static int modprobe_remove_kernel_modules(void) +{ + int ret = 0, i; + char modprobe[256]; + + for (i = ARRAY_SIZE(kernel_modules_list) - 1; i >= 0; i--) { + ret = snprintf(modprobe, sizeof(modprobe), + "/sbin/modprobe --remove --quiet %s", + kernel_modules_list[i].name); + if (ret < 0) { + perror("snprintf modprobe --remove"); + goto error; + } + modprobe[sizeof(modprobe) - 1] = '\0'; + ret = system(modprobe); + if (ret == -1) { + ERR("Unable to launch modprobe --remove for module %s", + kernel_modules_list[i].name); + } else if (kernel_modules_list[i].required + && WEXITSTATUS(ret) != 0) { + ERR("Unable to remove module %s", + kernel_modules_list[i].name); + } else { + DBG("Modprobe removal successful %s", + kernel_modules_list[i].name); + } + } + +error: + return ret; +} + +/* + * Return group ID of the tracing group or -1 if not found. + */ static gid_t allowed_group(void) { struct group *grp; @@ -132,7 +233,7 @@ static gid_t allowed_group(void) } /* - * Init quit pipe. + * Init thread quit pipe. * * Return -1 on error or 0 if all pipes are created. */ @@ -158,18 +259,38 @@ static void teardown_kernel_session(struct ltt_session *session) { if (session->kernel_session != NULL) { DBG("Tearing down kernel session"); - trace_destroy_kernel_session(session->kernel_session); + + /* + * If a custom kernel consumer was registered, close the socket before + * tearing down the complete kernel session structure + */ + if (session->kernel_session->consumer_fd != kconsumerd_cmd_sock) { + lttcomm_close_unix_sock(session->kernel_session->consumer_fd); + } + + trace_kernel_destroy_session(session->kernel_session); /* Extra precaution */ session->kernel_session = NULL; } } +/* + * Stop all threads by closing the thread quit pipe. + */ static void stop_threads(void) { + int ret; + /* Stopping all threads */ DBG("Terminating all threads"); - close(thread_quit_pipe[0]); - close(thread_quit_pipe[1]); + ret = write(thread_quit_pipe[1], "!", 1); + if (ret < 0) { + ERR("write error on thread quit pipe"); + } + + /* Dispatch thread */ + dispatch_thread_exit = 1; + futex_nto1_wake(&ust_cmd_queue.futex); } /* @@ -184,21 +305,23 @@ static void cleanup(void) DBG("Cleaning up"); /* */ - MSG("\n%c[%d;%dm*** assert failed *** ==> %c[%dm%c[%d;%dm" + MSG("%c[%d;%dm*** assert failed *** ==> %c[%dm%c[%d;%dm" "Matthew, BEET driven development works!%c[%dm", 27, 1, 31, 27, 0, 27, 1, 33, 27, 0); /* */ - DBG("Removing %s directory", LTTNG_RUNDIR); - ret = asprintf(&cmd, "rm -rf " LTTNG_RUNDIR); - if (ret < 0) { - ERR("asprintf failed. Something is really wrong!"); - } + if (is_root) { + DBG("Removing %s directory", LTTNG_RUNDIR); + ret = asprintf(&cmd, "rm -rf " LTTNG_RUNDIR); + if (ret < 0) { + ERR("asprintf failed. Something is really wrong!"); + } - /* Remove lttng run directory */ - ret = system(cmd); - if (ret < 0) { - ERR("Unable to clean " LTTNG_RUNDIR); + /* Remove lttng run directory */ + ret = system(cmd); + if (ret < 0) { + ERR("Unable to clean " LTTNG_RUNDIR); + } } DBG("Cleaning up all session"); @@ -214,13 +337,21 @@ static void cleanup(void) } } + DBG("Closing all UST sockets"); + clean_traceable_apps_list(); + pthread_mutex_destroy(&kconsumerd_pid_mutex); DBG("Closing kernel fd"); close(kernel_tracer_fd); - DBG("Unloading kernel modules"); - modprobe_remove_kernel_modules(); + if (is_root) { + DBG("Unloading kernel modules"); + modprobe_remove_kernel_modules(); + } + + close(thread_quit_pipe[0]); + close(thread_quit_pipe[1]); } /* @@ -288,7 +419,9 @@ static int send_kconsumerd_channel_fds(int sock, struct ltt_kernel_channel *chan lkm.fd = stream->fd; lkm.state = stream->state; lkm.max_sb_size = channel->channel->attr.subbuf_size; + lkm.output = channel->channel->attr.output; strncpy(lkm.path_name, stream->pathname, PATH_MAX); + lkm.path_name[PATH_MAX - 1] = '\0'; DBG("Sending fd %d to kconsumerd", lkm.fd); @@ -311,7 +444,7 @@ error: /* * Send all stream fds of the kernel session to the consumer. */ -static int send_kconsumerd_fds(int sock, struct ltt_kernel_session *session) +static int send_kconsumerd_fds(struct ltt_kernel_session *session) { int ret; struct ltt_kernel_channel *chan; @@ -324,7 +457,7 @@ static int send_kconsumerd_fds(int sock, struct ltt_kernel_session *session) DBG("Sending kconsumerd header for metadata"); - ret = lttcomm_send_unix_sock(sock, &lkh, sizeof(struct lttcomm_kconsumerd_header)); + ret = lttcomm_send_unix_sock(session->consumer_fd, &lkh, sizeof(struct lttcomm_kconsumerd_header)); if (ret < 0) { perror("send kconsumerd header"); goto error; @@ -332,14 +465,21 @@ static int send_kconsumerd_fds(int sock, struct ltt_kernel_session *session) DBG("Sending metadata stream fd"); + /* Extra protection. It's NOT suppose to be set to 0 at this point */ + if (session->consumer_fd == 0) { + session->consumer_fd = kconsumerd_cmd_sock; + } + if (session->metadata_stream_fd != 0) { /* Send metadata stream fd first */ lkm.fd = session->metadata_stream_fd; lkm.state = ACTIVE_FD; lkm.max_sb_size = session->metadata->conf->attr.subbuf_size; + lkm.output = DEFAULT_KERNEL_CHANNEL_OUTPUT; strncpy(lkm.path_name, session->metadata->pathname, PATH_MAX); + lkm.path_name[PATH_MAX - 1] = '\0'; - ret = lttcomm_send_fds_unix_sock(sock, &lkm, &lkm.fd, 1, sizeof(lkm)); + ret = lttcomm_send_fds_unix_sock(session->consumer_fd, &lkm, &lkm.fd, 1, sizeof(lkm)); if (ret < 0) { perror("send kconsumerd fd"); goto error; @@ -347,7 +487,7 @@ static int send_kconsumerd_fds(int sock, struct ltt_kernel_session *session) } cds_list_for_each_entry(chan, &session->channel_list.head, list) { - ret = send_kconsumerd_channel_fds(sock, chan); + ret = send_kconsumerd_channel_fds(session->consumer_fd, chan); if (ret < 0) { goto error; } @@ -361,64 +501,29 @@ error: return ret; } -#ifdef DISABLED /* - * Return a socket connected to the libust communication socket of the - * application identified by the pid. - * - * If the pid is not found in the traceable list, return -1 to indicate error. + * Notify UST applications using the shm mmap futex. */ -static int ust_connect_app(pid_t pid) +static int notify_ust_apps(int active) { - int sock; - struct ltt_traceable_app *lta; - - DBG("Connect to application pid %d", pid); - - lta = find_app_by_pid(pid); - if (lta == NULL) { - /* App not found */ - DBG("Application pid %d not found", pid); - return -1; - } - - sock = ustctl_connect_pid(lta->pid); - if (sock < 0) { - ERR("Fail connecting to the PID %d", pid); - } + char *wait_shm_mmap; - return sock; -} -#endif /* DISABLED */ + DBG("Notifying applications of session daemon state: %d", active); -/* - * Notify apps by writing 42 to a named pipe using name. Every applications - * waiting for a ltt-sessiond will be notified and re-register automatically to - * the session daemon. - * - * Return open or write error value. - */ -static int notify_apps(const char *name) -{ - int fd; - int ret = -1; - - DBG("Notify the global application pipe"); - - /* Try opening the global pipe */ - fd = open(name, O_WRONLY); - if (fd < 0) { + /* See shm.c for this call implying mmap, shm and futex calls */ + wait_shm_mmap = shm_ust_get_mmap(wait_shm_path, is_root); + if (wait_shm_mmap == NULL) { goto error; } - /* Notify by writing on the pipe */ - ret = write(fd, "42", 2); - if (ret < 0) { - perror("write"); - } + /* Wake waiting process */ + futex_wait_update((int32_t *) wait_shm_mmap, active); + + /* Apps notified successfully */ + return 0; error: - return ret; + return -1; } /* @@ -455,23 +560,17 @@ error: } /* - * Update the kernel pollfd set of all channel fd available over all tracing + * Update the kernel poll set of all channel fd available over all tracing * session. Add the wakeup pipe at the end of the set. */ -static int update_kernel_pollfd(void) +static int update_kernel_poll(struct lttng_poll_event *events) { - int i = 0; - /* - * The wakup pipe and the quit pipe are needed so the number of fds starts - * at 2 for those pipes. - */ - unsigned int nb_fd = 2; + int ret; struct ltt_session *session; struct ltt_kernel_channel *channel; - DBG("Updating kernel_pollfd"); + DBG("Updating kernel poll set"); - /* Get the number of channel of all kernel session */ lock_session_list(); cds_list_for_each_entry(session, &session_list_ptr->head, list) { lock_session(session); @@ -479,46 +578,21 @@ static int update_kernel_pollfd(void) unlock_session(session); continue; } - nb_fd += session->kernel_session->channel_count; - unlock_session(session); - } - - DBG("Resizing kernel_pollfd to size %d", nb_fd); - - kernel_pollfd = realloc(kernel_pollfd, nb_fd * sizeof(struct pollfd)); - if (kernel_pollfd == NULL) { - perror("malloc kernel_pollfd"); - goto error; - } - cds_list_for_each_entry(session, &session_list_ptr->head, list) { - lock_session(session); - if (session->kernel_session == NULL) { - unlock_session(session); - continue; - } - if (i >= nb_fd) { - ERR("To much channel for kernel_pollfd size"); - unlock_session(session); - break; - } cds_list_for_each_entry(channel, &session->kernel_session->channel_list.head, list) { - kernel_pollfd[i].fd = channel->fd; - kernel_pollfd[i].events = POLLIN | POLLRDNORM; - i++; + /* Add channel fd to the kernel poll set */ + ret = lttng_poll_add(events, channel->fd, LPOLLIN | LPOLLRDNORM); + if (ret < 0) { + unlock_session(session); + goto error; + } + DBG("Channel fd %d added to kernel set", channel->fd); } unlock_session(session); } unlock_session_list(); - /* Adding wake up pipe */ - kernel_pollfd[nb_fd - 2].fd = kernel_poll_pipe[0]; - kernel_pollfd[nb_fd - 2].events = POLLIN; - - /* Adding the quit pipe */ - kernel_pollfd[nb_fd - 1].fd = thread_quit_pipe[0]; - - return nb_fd; + return 0; error: unlock_session_list(); @@ -546,34 +620,44 @@ static int update_kernel_stream(int fd) unlock_session(session); continue; } - cds_list_for_each_entry(channel, &session->kernel_session->channel_list.head, list) { + + /* This is not suppose to be 0 but this is an extra security check */ + if (session->kernel_session->consumer_fd == 0) { + session->kernel_session->consumer_fd = kconsumerd_cmd_sock; + } + + cds_list_for_each_entry(channel, + &session->kernel_session->channel_list.head, list) { if (channel->fd == fd) { DBG("Channel found, updating kernel streams"); ret = kernel_open_channel_stream(channel); if (ret < 0) { - goto end; + goto error; } + /* - * Have we already sent fds to the consumer? If yes, it means that - * tracing is started so it is safe to send our updated stream fds. + * Have we already sent fds to the consumer? If yes, it means + * that tracing is started so it is safe to send our updated + * stream fds. */ if (session->kernel_session->kconsumer_fds_sent == 1) { - ret = send_kconsumerd_channel_fds(kconsumerd_cmd_sock, channel); + ret = send_kconsumerd_channel_fds( + session->kernel_session->consumer_fd, channel); if (ret < 0) { - goto end; + goto error; } } - goto end; + goto error; } } unlock_session(session); } + unlock_session_list(); + return ret; -end: +error: + unlock_session(session); unlock_session_list(); - if (session) { - unlock_session(session); - } return ret; } @@ -585,79 +669,93 @@ end: */ static void *thread_manage_kernel(void *data) { - int ret, i, nb_fd = 0; + int ret, i, pollfd, update_poll_flag = 1; + uint32_t revents, nb_fd; char tmp; - int update_poll_flag = 1; + struct lttng_poll_event events; DBG("Thread manage kernel started"); + ret = create_thread_poll_set(&events, 2); + if (ret < 0) { + goto error; + } + + ret = lttng_poll_add(&events, kernel_poll_pipe[0], LPOLLIN); + if (ret < 0) { + goto error; + } + while (1) { if (update_poll_flag == 1) { - nb_fd = update_kernel_pollfd(); - if (nb_fd < 0) { + ret = update_kernel_poll(&events); + if (ret < 0) { goto error; } update_poll_flag = 0; } - DBG("Polling on %d fds", nb_fd); + nb_fd = LTTNG_POLL_GETNB(&events); + + DBG("Thread kernel polling on %d fds", nb_fd); + + /* Zeroed the poll events */ + lttng_poll_reset(&events); /* Poll infinite value of time */ - ret = poll(kernel_pollfd, nb_fd, -1); + ret = lttng_poll_wait(&events, -1); if (ret < 0) { - perror("poll kernel thread"); goto error; } else if (ret == 0) { /* Should not happen since timeout is infinite */ + ERR("Return value of poll is 0 with an infinite timeout.\n" + "This should not have happened! Continuing..."); continue; } - /* Thread quit pipe has been closed. Killing thread. */ - if (kernel_pollfd[nb_fd - 1].revents == POLLNVAL) { - goto error; - } - - DBG("Kernel poll event triggered"); + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); - /* - * Check if the wake up pipe was triggered. If so, the kernel_pollfd - * must be updated. - */ - switch (kernel_pollfd[nb_fd - 2].revents) { - case POLLIN: - ret = read(kernel_poll_pipe[0], &tmp, 1); - update_poll_flag = 1; - continue; - case POLLERR: - goto error; - default: - break; - } + /* Thread quit pipe has been closed. Killing thread. */ + ret = check_thread_quit_pipe(pollfd, revents); + if (ret) { + goto error; + } - for (i = 0; i < nb_fd; i++) { - switch (kernel_pollfd[i].revents) { - /* - * New CPU detected by the kernel. Adding kernel stream to kernel - * session and updating the kernel consumer - */ - case POLLIN | POLLRDNORM: - ret = update_kernel_stream(kernel_pollfd[i].fd); - if (ret < 0) { - continue; + /* Check for data on kernel pipe */ + if (pollfd == kernel_poll_pipe[0] && (revents & LPOLLIN)) { + ret = read(kernel_poll_pipe[0], &tmp, 1); + update_poll_flag = 1; + continue; + } else { + /* + * New CPU detected by the kernel. Adding kernel stream to + * kernel session and updating the kernel consumer + */ + if (revents & LPOLLIN) { + ret = update_kernel_stream(pollfd); + if (ret < 0) { + continue; + } + break; + /* + * TODO: We might want to handle the LPOLLERR | LPOLLHUP + * and unregister kernel stream at this point. + */ } - break; } } } error: DBG("Kernel thread dying"); - if (kernel_pollfd) { - free(kernel_pollfd); - } - close(kernel_poll_pipe[0]); close(kernel_poll_pipe[1]); + + lttng_poll_clean(&events); + return NULL; } @@ -666,9 +764,10 @@ error: */ static void *thread_manage_kconsumerd(void *data) { - int sock = 0, ret; + int sock = 0, i, ret, pollfd; + uint32_t revents, nb_fd; enum lttcomm_return_code code; - struct pollfd pollfd[2]; + struct lttng_poll_event events; DBG("[thread] Manage kconsumerd started"); @@ -677,26 +776,46 @@ static void *thread_manage_kconsumerd(void *data) goto error; } - /* First fd is always the quit pipe */ - pollfd[0].fd = thread_quit_pipe[0]; + /* + * Pass 2 as size here for the thread quit pipe and kconsumerd_err_sock. + * Nothing more will be added to this poll set. + */ + ret = create_thread_poll_set(&events, 2); + if (ret < 0) { + goto error; + } + + ret = lttng_poll_add(&events, kconsumerd_err_sock, LPOLLIN | LPOLLRDHUP); + if (ret < 0) { + goto error; + } - /* Apps socket */ - pollfd[1].fd = kconsumerd_err_sock; - pollfd[1].events = POLLIN; + nb_fd = LTTNG_POLL_GETNB(&events); /* Inifinite blocking call, waiting for transmission */ - ret = poll(pollfd, 2, -1); + ret = lttng_poll_wait(&events, -1); if (ret < 0) { - perror("poll kconsumerd thread"); goto error; } - /* Thread quit pipe has been closed. Killing thread. */ - if (pollfd[0].revents == POLLNVAL) { - goto error; - } else if (pollfd[1].revents == POLLERR) { - ERR("Kconsumerd err socket poll error"); - goto error; + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Thread quit pipe has been closed. Killing thread. */ + ret = check_thread_quit_pipe(pollfd, revents); + if (ret) { + goto error; + } + + /* Event on the registration socket */ + if (pollfd == kconsumerd_err_sock) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Kconsumerd err socket poll error"); + goto error; + } + } } sock = lttcomm_accept_unix_sock(kconsumerd_err_sock); @@ -726,6 +845,46 @@ static void *thread_manage_kconsumerd(void *data) goto error; } + /* Remove the kconsumerd error socket since we have established a connexion */ + ret = lttng_poll_del(&events, kconsumerd_err_sock); + if (ret < 0) { + goto error; + } + + ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP); + if (ret < 0) { + goto error; + } + + /* Update number of fd */ + nb_fd = LTTNG_POLL_GETNB(&events); + + /* Inifinite blocking call, waiting for transmission */ + ret = lttng_poll_wait(&events, -1); + if (ret < 0) { + goto error; + } + + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Thread quit pipe has been closed. Killing thread. */ + ret = check_thread_quit_pipe(pollfd, revents); + if (ret) { + goto error; + } + + /* Event on the kconsumerd socket */ + if (pollfd == sock) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Kconsumerd err socket second poll error"); + goto error; + } + } + } + /* Wait for any kconsumerd error */ ret = lttcomm_recv_unix_sock(sock, &code, sizeof(enum lttcomm_return_code)); if (ret <= 0) { @@ -737,117 +896,338 @@ static void *thread_manage_kconsumerd(void *data) error: DBG("Kconsumerd thread dying"); - if (kconsumerd_err_sock) { - close(kconsumerd_err_sock); - } - if (kconsumerd_cmd_sock) { - close(kconsumerd_cmd_sock); - } - if (sock) { - close(sock); - } + close(kconsumerd_err_sock); + close(kconsumerd_cmd_sock); + close(sock); unlink(kconsumerd_err_unix_sock_path); unlink(kconsumerd_cmd_unix_sock_path); - kconsumerd_pid = 0; + + lttng_poll_clean(&events); + return NULL; } /* - * This thread manage the application socket communication + * This thread manage application communication. */ static void *thread_manage_apps(void *data) { - int sock = 0, ret; - struct pollfd pollfd[2]; + int i, ret, pollfd; + uint32_t revents, nb_fd; + struct ust_command ust_cmd; + struct lttng_poll_event events; - /* TODO: Something more elegant is needed but fine for now */ - /* FIXME: change all types to either uint8_t, uint32_t, uint64_t - * for 32-bit vs 64-bit compat processes. */ - /* replicate in ust with version number */ - struct { - int reg; /* 1:register, 0:unregister */ - pid_t pid; - uid_t uid; - } reg_msg; + DBG("[thread] Manage application started"); - DBG("[thread] Manage apps started"); + ret = create_thread_poll_set(&events, 2); + if (ret < 0) { + goto error; + } + + ret = lttng_poll_add(&events, apps_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); + if (ret < 0) { + goto error; + } + + while (1) { + /* Zeroed the events structure */ + lttng_poll_reset(&events); + + nb_fd = LTTNG_POLL_GETNB(&events); + + DBG("Apps thread polling on %d fds", nb_fd); + + /* Inifinite blocking call, waiting for transmission */ + ret = lttng_poll_wait(&events, -1); + if (ret < 0) { + goto error; + } + + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Thread quit pipe has been closed. Killing thread. */ + ret = check_thread_quit_pipe(pollfd, revents); + if (ret) { + goto error; + } + + /* Inspect the apps cmd pipe */ + if (pollfd == apps_cmd_pipe[0]) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Apps command pipe error"); + goto error; + } else if (revents & LPOLLIN) { + /* Empty pipe */ + ret = read(apps_cmd_pipe[0], &ust_cmd, sizeof(ust_cmd)); + if (ret < 0 || ret < sizeof(ust_cmd)) { + perror("read apps cmd pipe"); + goto error; + } + + /* Register applicaton to the session daemon */ + ret = register_traceable_app(&ust_cmd.reg_msg, ust_cmd.sock); + if (ret < 0) { + /* Only critical ENOMEM error can be returned here */ + goto error; + } + + ret = ustctl_register_done(ust_cmd.sock); + if (ret < 0) { + /* + * If the registration is not possible, we simply + * unregister the apps and continue + */ + unregister_traceable_app(ust_cmd.sock); + } else { + /* + * We just need here to monitor the close of the UST + * socket and poll set monitor those by default. + */ + ret = lttng_poll_add(&events, ust_cmd.sock, 0); + if (ret < 0) { + goto error; + } + + DBG("Apps with sock %d added to poll set", ust_cmd.sock); + } + break; + } + } else { + /* + * At this point, we know that a registered application made the + * event at poll_wait. + */ + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + /* Removing from the poll set */ + ret = lttng_poll_del(&events, pollfd); + if (ret < 0) { + goto error; + } + + /* Socket closed */ + unregister_traceable_app(pollfd); + break; + } + } + } + } + +error: + DBG("Application communication apps dying"); + close(apps_cmd_pipe[0]); + close(apps_cmd_pipe[1]); + + lttng_poll_clean(&events); + + return NULL; +} + +/* + * Dispatch request from the registration threads to the application + * communication thread. + */ +static void *thread_dispatch_ust_registration(void *data) +{ + int ret; + struct cds_wfq_node *node; + struct ust_command *ust_cmd = NULL; + + DBG("[thread] Dispatch UST command started"); + + while (!dispatch_thread_exit) { + /* Atomically prepare the queue futex */ + futex_nto1_prepare(&ust_cmd_queue.futex); + + do { + /* Dequeue command for registration */ + node = cds_wfq_dequeue_blocking(&ust_cmd_queue.queue); + if (node == NULL) { + DBG("Waked up but nothing in the UST command queue"); + /* Continue thread execution */ + break; + } + + ust_cmd = caa_container_of(node, struct ust_command, node); + + DBG("Dispatching UST registration pid:%d ppid:%d uid:%d" + " gid:%d sock:%d name:%s (version %d.%d)", + ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid, + ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid, + ust_cmd->sock, ust_cmd->reg_msg.name, + ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor); + /* + * Inform apps thread of the new application registration. This + * call is blocking so we can be assured that the data will be read + * at some point in time or wait to the end of the world :) + */ + ret = write(apps_cmd_pipe[1], ust_cmd, + sizeof(struct ust_command)); + if (ret < 0) { + perror("write apps cmd pipe"); + if (errno == EBADF) { + /* + * We can't inform the application thread to process + * registration. We will exit or else application + * registration will not occur and tracing will never + * start. + */ + goto error; + } + } + free(ust_cmd); + } while (node != NULL); + + /* Futex wait on queue. Blocking call on futex() */ + futex_nto1_wait(&ust_cmd_queue.futex); + } + +error: + DBG("Dispatch thread dying"); + return NULL; +} + +/* + * This thread manage application registration. + */ +static void *thread_registration_apps(void *data) +{ + int sock = 0, i, ret, pollfd; + uint32_t revents, nb_fd; + struct lttng_poll_event events; + /* + * Get allocated in this thread, enqueued to a global queue, dequeued and + * freed in the manage apps thread. + */ + struct ust_command *ust_cmd = NULL; + + DBG("[thread] Manage application registration started"); ret = lttcomm_listen_unix_sock(apps_sock); if (ret < 0) { goto error; } - /* First fd is always the quit pipe */ - pollfd[0].fd = thread_quit_pipe[0]; + /* + * Pass 2 as size here for the thread quit pipe and apps socket. Nothing + * more will be added to this poll set. + */ + ret = create_thread_poll_set(&events, 2); + if (ret < 0) { + goto error; + } - /* Apps socket */ - pollfd[1].fd = apps_sock; - pollfd[1].events = POLLIN; + /* Add the application registration socket */ + ret = lttng_poll_add(&events, apps_sock, LPOLLIN | LPOLLRDHUP); + if (ret < 0) { + goto error; + } /* Notify all applications to register */ - notify_apps(default_global_apps_pipe); + ret = notify_ust_apps(1); + if (ret < 0) { + ERR("Failed to notify applications or create the wait shared memory.\n" + "Execution continues but there might be problem for already running\n" + "applications that wishes to register."); + } while (1) { DBG("Accepting application registration"); + nb_fd = LTTNG_POLL_GETNB(&events); + /* Inifinite blocking call, waiting for transmission */ - ret = poll(pollfd, 2, -1); + ret = lttng_poll_wait(&events, -1); if (ret < 0) { - perror("poll apps thread"); goto error; } - /* Thread quit pipe has been closed. Killing thread. */ - if (pollfd[0].revents == POLLNVAL) { - goto error; - } else if (pollfd[1].revents == POLLERR) { - ERR("Apps socket poll error"); - goto error; - } + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); - sock = lttcomm_accept_unix_sock(apps_sock); - if (sock < 0) { - goto error; - } + /* Thread quit pipe has been closed. Killing thread. */ + ret = check_thread_quit_pipe(pollfd, revents); + if (ret) { + goto error; + } - /* - * Basic recv here to handle the very simple data - * that the libust send to register (reg_msg). - */ - ret = recv(sock, ®_msg, sizeof(reg_msg), 0); - if (ret < 0) { - perror("recv"); - continue; - } + /* Event on the registration socket */ + if (pollfd == apps_sock) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Register apps socket poll error"); + goto error; + } else if (revents & LPOLLIN) { + sock = lttcomm_accept_unix_sock(apps_sock); + if (sock < 0) { + goto error; + } - /* Add application to the global traceable list */ - if (reg_msg.reg == 1) { - /* Registering */ - ret = register_traceable_app(reg_msg.pid, reg_msg.uid); - if (ret < 0) { - /* register_traceable_app only return an error with - * ENOMEM. At this point, we better stop everything. - */ - goto error; + /* Create UST registration command for enqueuing */ + ust_cmd = malloc(sizeof(struct ust_command)); + if (ust_cmd == NULL) { + perror("ust command malloc"); + goto error; + } + + /* + * Using message-based transmissions to ensure we don't + * have to deal with partially received messages. + */ + ret = lttcomm_recv_unix_sock(sock, &ust_cmd->reg_msg, + sizeof(struct ust_register_msg)); + if (ret < 0 || ret < sizeof(struct ust_register_msg)) { + if (ret < 0) { + perror("lttcomm_recv_unix_sock register apps"); + } else { + ERR("Wrong size received on apps register"); + } + free(ust_cmd); + close(sock); + continue; + } + + ust_cmd->sock = sock; + + DBG("UST registration received with pid:%d ppid:%d uid:%d" + " gid:%d sock:%d name:%s (version %d.%d)", + ust_cmd->reg_msg.pid, ust_cmd->reg_msg.ppid, + ust_cmd->reg_msg.uid, ust_cmd->reg_msg.gid, + ust_cmd->sock, ust_cmd->reg_msg.name, + ust_cmd->reg_msg.major, ust_cmd->reg_msg.minor); + /* + * Lock free enqueue the registration request. The red pill + * has been taken! This apps will be part of the *system* :) + */ + cds_wfq_enqueue(&ust_cmd_queue.queue, &ust_cmd->node); + + /* + * Wake the registration queue futex. Implicit memory + * barrier with the exchange in cds_wfq_enqueue. + */ + futex_nto1_wake(&ust_cmd_queue.futex); + } } - } else { - /* Unregistering */ - unregister_traceable_app(reg_msg.pid); } } error: - DBG("Apps thread dying"); - if (apps_sock) { - close(apps_sock); - } - if (sock) { - close(sock); - } + DBG("UST Registration thread dying"); + + /* Notify that the registration thread is gone */ + notify_ust_apps(0); + close(apps_sock); + close(sock); unlink(apps_unix_sock_path); + + lttng_poll_clean(&events); + return NULL; } @@ -883,6 +1263,9 @@ error: return ret; } +/* + * Join kernel consumer thread + */ static int join_kconsumerd_thread(void) { void *status; @@ -1005,7 +1388,7 @@ static int modprobe_kernel_modules(void) ERR("Unable to launch modprobe for module %s", kernel_modules_list[i].name); } else if (kernel_modules_list[i].required - && WEXITSTATUS(ret) != 0) { + && WEXITSTATUS(ret) != 0) { ERR("Unable to load module %s", kernel_modules_list[i].name); } else { @@ -1018,42 +1401,6 @@ error: return ret; } -/* - * modprobe_remove_kernel_modules - * Remove modules in reverse load order. - */ -static int modprobe_remove_kernel_modules(void) -{ - int ret = 0, i; - char modprobe[256]; - - for (i = ARRAY_SIZE(kernel_modules_list) - 1; i >= 0; i--) { - ret = snprintf(modprobe, sizeof(modprobe), - "/sbin/modprobe --remove --quiet %s", - kernel_modules_list[i].name); - if (ret < 0) { - perror("snprintf modprobe --remove"); - goto error; - } - modprobe[sizeof(modprobe) - 1] = '\0'; - ret = system(modprobe); - if (ret == -1) { - ERR("Unable to launch modprobe --remove for module %s", - kernel_modules_list[i].name); - } else if (kernel_modules_list[i].required - && WEXITSTATUS(ret) != 0) { - ERR("Unable to remove module %s", - kernel_modules_list[i].name); - } else { - DBG("Modprobe removal successful %s", - kernel_modules_list[i].name); - } - } - -error: - return ret; -} - /* * mount_debugfs */ @@ -1169,7 +1516,16 @@ static int start_kernel_trace(struct ltt_kernel_session *session) int ret = 0; if (session->kconsumer_fds_sent == 0) { - ret = send_kconsumerd_fds(kconsumerd_cmd_sock, session); + /* + * Assign default kernel consumer if no consumer assigned to the kernel + * session. At this point, it's NOT suppose to be 0 but this is an extra + * security check. + */ + if (session->consumer_fd == 0) { + session->consumer_fd = kconsumerd_cmd_sock; + } + + ret = send_kconsumerd_fds(session); if (ret < 0) { ERR("Send kconsumerd fds failed"); ret = LTTCOMM_KERN_CONSUMER_FAIL; @@ -1184,9 +1540,9 @@ error: } /* - * Notify kernel thread to update it's pollfd. + * Notify kernel thread to update it's poll set. */ -static int notify_kernel_pollfd(void) +static int notify_kernel_channels_update(void) { int ret; @@ -1202,7 +1558,8 @@ static int notify_kernel_pollfd(void) /* * Allocate a channel structure and fill it. */ -static struct lttng_channel *init_default_channel(char *name) +static struct lttng_channel *init_default_channel(enum lttng_domain_type domain_type, + char *name) { struct lttng_channel *chan; @@ -1214,18 +1571,66 @@ static struct lttng_channel *init_default_channel(char *name) if (snprintf(chan->name, NAME_MAX, "%s", name) < 0) { perror("snprintf channel name"); - return NULL; + goto error; } chan->attr.overwrite = DEFAULT_CHANNEL_OVERWRITE; - chan->attr.subbuf_size = DEFAULT_CHANNEL_SUBBUF_SIZE; - chan->attr.num_subbuf = DEFAULT_CHANNEL_SUBBUF_NUM; chan->attr.switch_timer_interval = DEFAULT_CHANNEL_SWITCH_TIMER; chan->attr.read_timer_interval = DEFAULT_CHANNEL_READ_TIMER; - chan->attr.output = DEFAULT_KERNEL_CHANNEL_OUTPUT; -error: + switch (domain_type) { + case LTTNG_DOMAIN_KERNEL: + chan->attr.subbuf_size = DEFAULT_KERNEL_CHANNEL_SUBBUF_SIZE; + chan->attr.num_subbuf = DEFAULT_KERNEL_CHANNEL_SUBBUF_NUM; + chan->attr.output = DEFAULT_KERNEL_CHANNEL_OUTPUT; + break; + /* TODO: add UST */ + default: + goto error; /* Not implemented */ + } + return chan; + +error: + free(chan); + return NULL; +} + +/* + * Create an UST session and add it to the session ust list. + */ +static int create_ust_session(pid_t pid, struct ltt_session *session) +{ + int ret = -1; + struct ltt_ust_session *lus; + + DBG("Creating UST session"); + + lus = trace_ust_create_session(session->path, pid); + if (lus == NULL) { + goto error; + } + + ret = mkdir_recursive(lus->path, S_IRWXU | S_IRWXG, + geteuid(), allowed_group()); + if (ret < 0) { + if (ret != -EEXIST) { + ERR("Trace directory creation error"); + goto error; + } + } + + /* Create session on the UST tracer */ + ret = ustctl_create_session(lus); + if (ret < 0) { + goto error; + } + + return 0; + +error: + free(lus); + return ret; } /* @@ -1243,8 +1648,13 @@ static int create_kernel_session(struct ltt_session *session) goto error; } - ret = mkdir_recursive(session->path, S_IRWXU | S_IRWXG, - geteuid(), allowed_group()); + /* Set kernel consumer socket fd */ + if (kconsumerd_cmd_sock) { + session->kernel_session->consumer_fd = kconsumerd_cmd_sock; + } + + ret = mkdir_recursive(session->kernel_session->trace_path, + S_IRWXU | S_IRWXG, geteuid(), allowed_group()); if (ret < 0) { if (ret != -EEXIST) { ERR("Trace directory creation error"); @@ -1275,7 +1685,9 @@ static void list_lttng_sessions(struct lttng_session *sessions) */ cds_list_for_each_entry(session, &session_list_ptr->head, list) { strncpy(sessions[i].path, session->path, PATH_MAX); + sessions[i].path[PATH_MAX - 1] = '\0'; strncpy(sessions[i].name, session->name, NAME_MAX); + sessions[i].name[NAME_MAX - 1] = '\0'; i++; } } @@ -1321,6 +1733,7 @@ static void list_lttng_events(struct ltt_kernel_channel *kchan, /* Kernel channels */ cds_list_for_each_entry(event, &kchan->events_list.head , list) { strncpy(events[i].name, event->event->name, LTTNG_SYMBOL_NAME_LEN); + events[i].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0'; events[i].enabled = event->enabled; switch (event->event->instrumentation) { case LTTNG_KERNEL_TRACEPOINT: @@ -1355,7 +1768,9 @@ static int process_client_msg(struct command_ctx *cmd_ctx) DBG("Processing client command %d", cmd_ctx->lsm->cmd_type); - /* Listing commands don't need a session */ + /* + * Commands that DO NOT need a session. + */ switch (cmd_ctx->lsm->cmd_type) { case LTTNG_CREATE_SESSION: case LTTNG_LIST_SESSIONS: @@ -1393,13 +1808,12 @@ static int process_client_msg(struct command_ctx *cmd_ctx) goto error; } } - /* Need a session for kernel command */ switch (cmd_ctx->lsm->cmd_type) { + case LTTNG_CALIBRATE: case LTTNG_CREATE_SESSION: case LTTNG_LIST_SESSIONS: case LTTNG_LIST_TRACEPOINTS: - case LTTNG_CALIBRATE: break; default: if (cmd_ctx->session->kernel_session == NULL) { @@ -1408,9 +1822,9 @@ static int process_client_msg(struct command_ctx *cmd_ctx) ret = LTTCOMM_KERN_SESS_FAIL; goto error; } - /* Start the kernel consumer daemon */ - if (kconsumerd_pid == 0) { + if (kconsumerd_pid == 0 && + cmd_ctx->lsm->cmd_type != LTTNG_REGISTER_CONSUMER) { ret = start_kconsumerd(); if (ret < 0) { goto error; @@ -1419,6 +1833,8 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } } break; + case LTTNG_DOMAIN_UST_PID: + break; default: break; } @@ -1444,6 +1860,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) strncpy(kctx.u.perf_counter.name, cmd_ctx->lsm->u.context.ctx.u.perf_counter.name, LTTNG_SYMBOL_NAME_LEN); + kctx.u.perf_counter.name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0'; /* Add kernel context to kernel tracer. See context.c */ ret = add_kernel_context(cmd_ctx->session->kernel_session, &kctx, @@ -1474,7 +1891,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: - kchan = get_kernel_channel_by_name(cmd_ctx->lsm->u.disable.channel_name, + kchan = trace_kernel_get_channel_by_name(cmd_ctx->lsm->u.disable.channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { ret = LTTCOMM_KERN_CHAN_NOT_FOUND; @@ -1512,14 +1929,14 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: - kchan = get_kernel_channel_by_name(cmd_ctx->lsm->u.disable.channel_name, + kchan = trace_kernel_get_channel_by_name(cmd_ctx->lsm->u.disable.channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { ret = LTTCOMM_KERN_CHAN_NOT_FOUND; goto error; } - kevent = get_kernel_event_by_name(cmd_ctx->lsm->u.disable.name, kchan); + kevent = trace_kernel_get_event_by_name(cmd_ctx->lsm->u.disable.name, kchan); if (kevent != NULL) { DBG("Disabling kernel event %s for channel %s.", kevent->event->name, kchan->channel->name); @@ -1555,7 +1972,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: DBG("Disabling all enabled kernel events"); - kchan = get_kernel_channel_by_name(cmd_ctx->lsm->u.disable.channel_name, + kchan = trace_kernel_get_channel_by_name(cmd_ctx->lsm->u.disable.channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { ret = LTTCOMM_KERN_CHAN_NOT_FOUND; @@ -1596,21 +2013,24 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: - kchan = get_kernel_channel_by_name(cmd_ctx->lsm->u.enable.channel_name, + kchan = trace_kernel_get_channel_by_name( + cmd_ctx->lsm->u.enable.channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { /* Channel not found, creating it */ - DBG("Creating kernel channel"); + DBG("Creating kernel channel %s", + cmd_ctx->lsm->u.enable.channel_name); ret = kernel_create_channel(cmd_ctx->session->kernel_session, - &cmd_ctx->lsm->u.channel.chan, cmd_ctx->session->path); + &cmd_ctx->lsm->u.channel.chan, + cmd_ctx->session->kernel_session->trace_path); if (ret < 0) { ret = LTTCOMM_KERN_CHAN_FAIL; goto error; } /* Notify kernel thread that there is a new channel */ - ret = notify_kernel_pollfd(); + ret = notify_kernel_channels_update(); if (ret < 0) { ret = LTTCOMM_FATAL; goto error; @@ -1626,9 +2046,11 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } kernel_wait_quiescent(kernel_tracer_fd); + break; + case LTTNG_DOMAIN_UST_PID: + break; default: - /* TODO: Userspace tracing */ ret = LTTCOMM_NOT_IMPLEMENTED; goto error; } @@ -1653,28 +2075,40 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: - do { - kchan = get_kernel_channel_by_name(channel_name, + kchan = trace_kernel_get_channel_by_name(channel_name, + cmd_ctx->session->kernel_session); + if (kchan == NULL) { + DBG("Channel not found. Creating channel %s", channel_name); + + chan = init_default_channel(cmd_ctx->lsm->domain.type, channel_name); + if (chan == NULL) { + ret = LTTCOMM_FATAL; + goto error; + } + + ret = kernel_create_channel(cmd_ctx->session->kernel_session, + chan, cmd_ctx->session->kernel_session->trace_path); + if (ret < 0) { + ret = LTTCOMM_KERN_CHAN_FAIL; + goto error; + } + kchan = trace_kernel_get_channel_by_name(channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { - DBG("Channel not found. Creating channel %s", channel_name); - - chan = init_default_channel(channel_name); - if (chan == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } + ERR("Channel %s not found after creation. Internal error, giving up.", + channel_name); + ret = LTTCOMM_FATAL; + goto error; + } - ret = kernel_create_channel(cmd_ctx->session->kernel_session, - chan, cmd_ctx->session->path); - if (ret < 0) { - ret = LTTCOMM_KERN_CHAN_FAIL; - goto error; - } + ret = notify_kernel_channels_update(); + if (ret < 0) { + ret = LTTCOMM_FATAL; + goto error; } - } while (kchan == NULL); + } - kevent = get_kernel_event_by_name(cmd_ctx->lsm->u.enable.event.name, kchan); + kevent = trace_kernel_get_event_by_name(cmd_ctx->lsm->u.enable.event.name, kchan); if (kevent == NULL) { DBG("Creating kernel event %s for channel %s.", cmd_ctx->lsm->u.enable.event.name, channel_name); @@ -1725,26 +2159,38 @@ static int process_client_msg(struct command_ctx *cmd_ctx) switch (cmd_ctx->lsm->domain.type) { case LTTNG_DOMAIN_KERNEL: - do { - kchan = get_kernel_channel_by_name(channel_name, + kchan = trace_kernel_get_channel_by_name(channel_name, + cmd_ctx->session->kernel_session); + if (kchan == NULL) { + DBG("Channel not found. Creating channel %s", channel_name); + + chan = init_default_channel(cmd_ctx->lsm->domain.type, channel_name); + if (chan == NULL) { + ret = LTTCOMM_FATAL; + goto error; + } + + ret = kernel_create_channel(cmd_ctx->session->kernel_session, + chan, cmd_ctx->session->kernel_session->trace_path); + if (ret < 0) { + ret = LTTCOMM_KERN_CHAN_FAIL; + goto error; + } + kchan = trace_kernel_get_channel_by_name(channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { - DBG("Channel not found. Creating channel %s", channel_name); - - chan = init_default_channel(channel_name); - if (chan == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } + ERR("Channel %s not found after creation. Internal error, giving up.", + channel_name); + ret = LTTCOMM_FATAL; + goto error; + } - ret = kernel_create_channel(cmd_ctx->session->kernel_session, - chan, cmd_ctx->session->path); - if (ret < 0) { - ret = LTTCOMM_KERN_CHAN_FAIL; - goto error; - } + ret = notify_kernel_channels_update(); + if (ret < 0) { + ret = LTTCOMM_FATAL; + goto error; } - } while (kchan == NULL); + } /* For each event in the kernel session */ cds_list_for_each_entry(kevent, &kchan->events_list.head, list) { @@ -1763,7 +2209,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } for (i = 0; i < size; i++) { - kevent = get_kernel_event_by_name(event_list[i].name, kchan); + kevent = trace_kernel_get_event_by_name(event_list[i].name, kchan); if (kevent == NULL) { /* Default event type for enable all */ event_list[i].type = LTTNG_EVENT_TRACEPOINT; @@ -1843,7 +2289,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) if (cmd_ctx->session->kernel_session->metadata == NULL) { DBG("Open kernel metadata"); ret = kernel_open_metadata(cmd_ctx->session->kernel_session, - cmd_ctx->session->path); + cmd_ctx->session->kernel_session->trace_path); if (ret < 0) { ret = LTTCOMM_KERN_META_FAIL; goto error; @@ -1877,16 +2323,16 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } } - DBG("Start kernel tracing"); - ret = kernel_start_session(cmd_ctx->session->kernel_session); + ret = start_kernel_trace(cmd_ctx->session->kernel_session); if (ret < 0) { - ERR("Kernel start session failed"); ret = LTTCOMM_KERN_START_FAIL; goto error; } - ret = start_kernel_trace(cmd_ctx->session->kernel_session); + DBG("Start kernel tracing"); + ret = kernel_start_session(cmd_ctx->session->kernel_session); if (ret < 0) { + ERR("Kernel start session failed"); ret = LTTCOMM_KERN_START_FAIL; goto error; } @@ -1980,10 +2426,10 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } /* - * Must notify the kernel thread here to update it's pollfd in order to - * remove the channel(s)' fd just destroyed. + * Must notify the kernel thread here to update it's poll setin order + * to remove the channel(s)' fd just destroyed. */ - ret = notify_kernel_pollfd(); + ret = notify_kernel_channels_update(); if (ret < 0) { ret = LTTCOMM_FATAL; goto error; @@ -2000,7 +2446,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) nb_dom++; } - nb_dom += cmd_ctx->session->ust_trace_count; + nb_dom += cmd_ctx->session->ust_session_list.count; ret = setup_lttng_msg(cmd_ctx, sizeof(struct lttng_domain) * nb_dom); if (ret < 0) { @@ -2047,7 +2493,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) struct ltt_kernel_channel *kchan = NULL; if (cmd_ctx->session->kernel_session != NULL) { - kchan = get_kernel_channel_by_name(cmd_ctx->lsm->u.list.channel_name, + kchan = trace_kernel_get_channel_by_name(cmd_ctx->lsm->u.list.channel_name, cmd_ctx->session->kernel_session); if (kchan == NULL) { ret = LTTCOMM_KERN_CHAN_NOT_FOUND; @@ -2095,7 +2541,6 @@ static int process_client_msg(struct command_ctx *cmd_ctx) ret = LTTCOMM_OK; break; } - case LTTNG_CALIBRATE: { /* Setup lttng message with no payload */ @@ -2125,6 +2570,43 @@ static int process_client_msg(struct command_ctx *cmd_ctx) ret = LTTCOMM_OK; break; } + case LTTNG_REGISTER_CONSUMER: + { + int sock; + + /* Setup lttng message with no payload */ + ret = setup_lttng_msg(cmd_ctx, 0); + if (ret < 0) { + goto setup_error; + } + + switch (cmd_ctx->lsm->domain.type) { + case LTTNG_DOMAIN_KERNEL: + { + /* Can't register a consumer if there is already one */ + if (cmd_ctx->session->kernel_session->consumer_fd != 0) { + ret = LTTCOMM_CONNECT_FAIL; + goto error; + } + + sock = lttcomm_connect_unix_sock(cmd_ctx->lsm->u.reg.path); + if (sock < 0) { + ret = LTTCOMM_CONNECT_FAIL; + goto error; + } + + cmd_ctx->session->kernel_session->consumer_fd = sock; + break; + } + default: + /* TODO: Userspace tracing */ + ret = LTTCOMM_NOT_IMPLEMENTED; + goto error; + } + + ret = LTTCOMM_OK; + break; + } default: /* Undefined command */ @@ -2169,9 +2651,10 @@ setup_error: */ static void *thread_manage_clients(void *data) { - int sock = 0, ret; + int sock = 0, ret, i, pollfd; + uint32_t revents, nb_fd; struct command_ctx *cmd_ctx = NULL; - struct pollfd pollfd[2]; + struct lttng_poll_event events; DBG("[thread] Manage client started"); @@ -2180,15 +2663,23 @@ static void *thread_manage_clients(void *data) goto error; } - /* First fd is always the quit pipe */ - pollfd[0].fd = thread_quit_pipe[0]; + /* + * Pass 2 as size here for the thread quit pipe and client_sock. Nothing + * more will be added to this poll set. + */ + ret = create_thread_poll_set(&events, 2); + if (ret < 0) { + goto error; + } - /* Apps socket */ - pollfd[1].fd = client_sock; - pollfd[1].events = POLLIN; + /* Add the application registration socket */ + ret = lttng_poll_add(&events, client_sock, LPOLLIN | LPOLLPRI); + if (ret < 0) { + goto error; + } - /* Notify parent pid that we are ready - * to accept command for client side. + /* + * Notify parent pid that we are ready to accept command for client side. */ if (opt_sig_parent) { kill(ppid, SIGCHLD); @@ -2197,21 +2688,36 @@ static void *thread_manage_clients(void *data) while (1) { DBG("Accepting client command ..."); + nb_fd = LTTNG_POLL_GETNB(&events); + /* Inifinite blocking call, waiting for transmission */ - ret = poll(pollfd, 2, -1); + ret = lttng_poll_wait(&events, -1); if (ret < 0) { - perror("poll client thread"); goto error; } - /* Thread quit pipe has been closed. Killing thread. */ - if (pollfd[0].revents == POLLNVAL) { - goto error; - } else if (pollfd[1].revents == POLLERR) { - ERR("Client socket poll error"); - goto error; + for (i = 0; i < nb_fd; i++) { + /* Fetch once the poll data */ + revents = LTTNG_POLL_GETEV(&events, i); + pollfd = LTTNG_POLL_GETFD(&events, i); + + /* Thread quit pipe has been closed. Killing thread. */ + ret = check_thread_quit_pipe(pollfd, revents); + if (ret) { + goto error; + } + + /* Event on the registration socket */ + if (pollfd == client_sock) { + if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { + ERR("Client socket poll error"); + goto error; + } + } } + DBG("Wait for client response"); + sock = lttcomm_accept_unix_sock(client_sock); if (sock < 0) { goto error; @@ -2219,9 +2725,18 @@ static void *thread_manage_clients(void *data) /* Allocate context command to process the client request */ cmd_ctx = malloc(sizeof(struct command_ctx)); + if (cmd_ctx == NULL) { + perror("malloc cmd_ctx"); + goto error; + } /* Allocate data buffer for reception */ cmd_ctx->lsm = malloc(sizeof(struct lttcomm_session_msg)); + if (cmd_ctx->lsm == NULL) { + perror("malloc cmd_ctx->lsm"); + goto error; + } + cmd_ctx->llm = NULL; cmd_ctx->session = NULL; @@ -2231,12 +2746,17 @@ static void *thread_manage_clients(void *data) * the client. */ DBG("Receiving data from client ..."); - ret = lttcomm_recv_unix_sock(sock, cmd_ctx->lsm, sizeof(struct lttcomm_session_msg)); + ret = lttcomm_recv_unix_sock(sock, cmd_ctx->lsm, + sizeof(struct lttcomm_session_msg)); if (ret <= 0) { + DBG("Nothing recv() from client... continuing"); + close(sock); + free(cmd_ctx); continue; } - // TODO: Validate cmd_ctx including sanity check for security purpose. + // TODO: Validate cmd_ctx including sanity check for + // security purpose. /* * This function dispatch the work to the kernel or userspace tracer @@ -2246,16 +2766,19 @@ static void *thread_manage_clients(void *data) */ ret = process_client_msg(cmd_ctx); if (ret < 0) { - /* TODO: Inform client somehow of the fatal error. At this point, - * ret < 0 means that a malloc failed (ENOMEM). */ - /* Error detected but still accept command */ + /* + * TODO: Inform client somehow of the fatal error. At + * this point, ret < 0 means that a malloc failed + * (ENOMEM). Error detected but still accept command. + */ clean_command_ctx(&cmd_ctx); continue; } DBG("Sending response (size: %d, retcode: %d)", cmd_ctx->lttng_msg_size, cmd_ctx->llm->ret_code); - ret = send_unix_sock(sock, cmd_ctx->llm, cmd_ctx->lttng_msg_size); + ret = send_unix_sock(sock, cmd_ctx->llm, + cmd_ctx->lttng_msg_size); if (ret < 0) { ERR("Failed to send data back to client"); } @@ -2268,15 +2791,11 @@ static void *thread_manage_clients(void *data) error: DBG("Client thread dying"); - if (client_sock) { - close(client_sock); - } - if (sock) { - close(sock); - } - unlink(client_unix_sock_path); + close(client_sock); + close(sock); + lttng_poll_clean(&events); clean_command_ctx(&cmd_ctx); return NULL; } @@ -2442,13 +2961,15 @@ end: static int check_existing_daemon(void) { if (access(client_unix_sock_path, F_OK) < 0 && - access(apps_unix_sock_path, F_OK) < 0) + access(apps_unix_sock_path, F_OK) < 0) { return 0; + } /* Is there anybody out there ? */ - if (lttng_session_daemon_alive()) + if (lttng_session_daemon_alive()) { return -EEXIST; - else + } else { return 0; + } } /* @@ -2509,6 +3030,14 @@ static int create_kernel_poll_pipe(void) return pipe2(kernel_poll_pipe, O_CLOEXEC); } +/* + * Create the application command pipe to wake thread_manage_apps. + */ +static int create_apps_cmd_pipe(void) +{ + return pipe2(apps_cmd_pipe, O_CLOEXEC); +} + /* * Create the lttng run directory needed for all global sockets and pipe. */ @@ -2704,6 +3233,12 @@ int main(int argc, char **argv) snprintf(client_unix_sock_path, PATH_MAX, DEFAULT_GLOBAL_CLIENT_UNIX_SOCK); } + + /* Set global SHM for ust */ + if (strlen(wait_shm_path) == 0) { + snprintf(wait_shm_path, PATH_MAX, + DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH); + } } else { home_path = get_home_dir(); if (home_path == NULL) { @@ -2723,6 +3258,12 @@ int main(int argc, char **argv) snprintf(client_unix_sock_path, PATH_MAX, DEFAULT_HOME_CLIENT_UNIX_SOCK, home_path); } + + /* Set global SHM for ust */ + if (strlen(wait_shm_path) == 0) { + snprintf(wait_shm_path, PATH_MAX, + DEFAULT_HOME_APPS_WAIT_SHM_PATH, geteuid()); + } } DBG("Client socket path %s", client_unix_sock_path); @@ -2785,30 +3326,58 @@ int main(int argc, char **argv) goto exit; } + /* Setup the thread apps communication pipe. */ + if ((ret = create_apps_cmd_pipe()) < 0) { + goto exit; + } + + /* Init UST command queue. */ + cds_wfq_init(&ust_cmd_queue.queue); + /* * Get session list pointer. This pointer MUST NOT be free(). * This list is statically declared in session.c */ session_list_ptr = get_session_list(); + /* Set up max poll set size */ + lttng_poll_set_max_size(); + /* Create thread to manage the client socket */ - ret = pthread_create(&client_thread, NULL, thread_manage_clients, (void *) NULL); + ret = pthread_create(&client_thread, NULL, + thread_manage_clients, (void *) NULL); if (ret != 0) { - perror("pthread_create"); + perror("pthread_create clients"); goto exit_client; } + /* Create thread to dispatch registration */ + ret = pthread_create(&dispatch_thread, NULL, + thread_dispatch_ust_registration, (void *) NULL); + if (ret != 0) { + perror("pthread_create dispatch"); + goto exit_dispatch; + } + + /* Create thread to manage application registration. */ + ret = pthread_create(®_apps_thread, NULL, + thread_registration_apps, (void *) NULL); + if (ret != 0) { + perror("pthread_create registration"); + goto exit_reg_apps; + } + /* Create thread to manage application socket */ ret = pthread_create(&apps_thread, NULL, thread_manage_apps, (void *) NULL); if (ret != 0) { - perror("pthread_create"); + perror("pthread_create apps"); goto exit_apps; } /* Create kernel thread to manage kernel event */ ret = pthread_create(&kernel_thread, NULL, thread_manage_kernel, (void *) NULL); if (ret != 0) { - perror("pthread_create"); + perror("pthread_create kernel"); goto exit_kernel; } @@ -2826,6 +3395,20 @@ exit_kernel: } exit_apps: + ret = pthread_join(reg_apps_thread, &status); + if (ret != 0) { + perror("pthread_join"); + goto error; /* join error, exit without cleanup */ + } + +exit_reg_apps: + ret = pthread_join(dispatch_thread, &status); + if (ret != 0) { + perror("pthread_join"); + goto error; /* join error, exit without cleanup */ + } + +exit_dispatch: ret = pthread_join(client_thread, &status); if (ret != 0) { perror("pthread_join");