X-Git-Url: https://git.lttng.org/?p=lttng-tools.git;a=blobdiff_plain;f=ltt-sessiond%2Fmain.c;h=91f79b3c5cdfc363de10dd080550cc251eb6a330;hp=5636700d8fad1cd3127c9c6571b03a648782261d;hb=90014c57e763d8c3f10148800bfb8461a66fb987;hpb=d0254c7c0bb43c62abafdf866b16de0da1ad4d5d diff --git a/ltt-sessiond/main.c b/ltt-sessiond/main.c index 5636700d8..91f79b3c5 100644 --- a/ltt-sessiond/main.c +++ b/ltt-sessiond/main.c @@ -1,5 +1,6 @@ /* * Copyright (C) 2011 - David Goulet + * Mathieu Desnoyers * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License @@ -28,29 +29,30 @@ #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 "context.h" -#include "liblttsessiondcomm.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; @@ -60,6 +62,7 @@ const char default_global_apps_pipe[] = DEFAULT_GLOBAL_APPS_PIPE; /* Variables */ int opt_verbose; /* Not static for lttngerr.h */ +int opt_verbose_kconsumerd; /* Not static for lttngerr.h */ int opt_quiet; /* Not static for lttngerr.h */ const char *progname; @@ -70,11 +73,13 @@ 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; @@ -90,16 +95,31 @@ 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); +/* + * 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. @@ -113,6 +133,60 @@ static int modprobe_remove_kernel_modules(void); */ static struct ltt_session_list *session_list_ptr; +/* + * 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; + + if (opt_tracing_group) { + grp = getgrnam(opt_tracing_group); + } else { + grp = getgrnam(default_tracing_group); + } + if (!grp) { + return -1; + } else { + return grp->gr_gid; + } +} + /* * Init quit pipe. * @@ -140,16 +214,39 @@ 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) +{ + /* Stopping all threads */ + DBG("Terminating all threads"); + close(thread_quit_pipe[0]); + close(thread_quit_pipe[1]); + /* Dispatch thread */ + dispatch_thread_exit = 1; + futex_nto1_wake(&ust_cmd_queue.futex); +} + /* * Cleanup the daemon */ -static void cleanup() +static void cleanup(void) { int ret; char *cmd; @@ -158,16 +255,11 @@ static void cleanup() 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); /* */ - /* Stopping all threads */ - DBG("Terminating all threads"); - close(thread_quit_pipe[0]); - close(thread_quit_pipe[1]); - DBG("Removing %s directory", LTTNG_RUNDIR); ret = asprintf(&cmd, "rm -rf " LTTNG_RUNDIR); if (ret < 0) { @@ -193,13 +285,18 @@ static void cleanup() } } + 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(); + } } /* @@ -267,7 +364,9 @@ 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); @@ -290,7 +389,7 @@ 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; @@ -303,7 +402,7 @@ 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; @@ -311,14 +410,21 @@ 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; @@ -326,7 +432,7 @@ 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; } @@ -340,64 +446,29 @@ 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; + char *wait_shm_mmap; - DBG("Connect to application pid %d", pid); + DBG("Notifying applications of session daemon state: %d", active); - 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); - } - - return sock; -} -#endif /* DISABLED */ - -/* - * 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; } /* @@ -496,6 +567,8 @@ static int update_kernel_pollfd(void) /* Adding the quit pipe */ kernel_pollfd[nb_fd - 1].fd = thread_quit_pipe[0]; + kernel_pollfd[nb_fd - 1].events = + POLLHUP | POLLNVAL | POLLERR | POLLIN | POLLRDHUP | POLLPRI; return nb_fd; @@ -525,6 +598,12 @@ static int update_kernel_stream(int fd) unlock_session(session); continue; } + + /* 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"); @@ -532,12 +611,14 @@ static int update_kernel_stream(int fd) if (ret < 0) { goto end; } + /* * 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; } @@ -588,6 +669,8 @@ static void *thread_manage_kernel(void *data) 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; } @@ -705,6 +788,25 @@ static void *thread_manage_kconsumerd(void *data) goto error; } + /* Kconsumerd err socket */ + pollfd[1].fd = sock; + pollfd[1].events = POLLIN; + + /* Inifinite blocking call, waiting for transmission */ + ret = poll(pollfd, 2, -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 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) { @@ -734,24 +836,275 @@ error: } /* - * This thread manage the application socket communication + * Reallocate the apps command pollfd structure of nb_fd size. + * + * The first two fds must be there at all time. + */ +static int update_apps_cmd_pollfd(unsigned int nb_fd, unsigned int old_nb_fd, + struct pollfd **pollfd) +{ + int i, count; + struct pollfd *old_pollfd = NULL; + + /* Can't accept pollfd less than 2 */ + if (nb_fd < 2) { + goto end; + } + + if (*pollfd) { + /* Save pointer */ + old_pollfd = *pollfd; + } + + *pollfd = malloc(nb_fd * sizeof(struct pollfd)); + if (*pollfd == NULL) { + perror("malloc manage apps pollfd"); + goto error; + } + + /* First fd is always the quit pipe */ + (*pollfd)[0].fd = thread_quit_pipe[0]; + (*pollfd)[0].events = + POLLHUP | POLLNVAL | POLLERR | POLLIN | POLLRDHUP | POLLPRI; + + /* Apps command pipe */ + (*pollfd)[1].fd = apps_cmd_pipe[0]; + (*pollfd)[1].events = POLLIN; + + /* Start count after the two pipes below */ + count = 2; + for (i = 2; i < old_nb_fd; i++) { + /* Add to new pollfd */ + if (old_pollfd[i].fd != -1) { + (*pollfd)[count].fd = old_pollfd[i].fd; + (*pollfd)[count].events = POLLHUP | POLLNVAL | POLLERR; + count++; + } + + if (count > nb_fd) { + ERR("Updating poll fd wrong size"); + goto error; + } + } + + if (nb_fd < 2) { + /* + * There should *always* be at least two fds in the pollfd. This safety + * check make sure the poll() will actually try on those two pipes at + * best which are the thread_quit_pipe and apps_cmd_pipe. + */ + nb_fd = 2; + MSG("nb_fd < 2 --> Not good! Continuing..."); + } + + /* Destroy old pollfd */ + free(old_pollfd); + + DBG("Apps cmd pollfd realloc of size %d", nb_fd); + +end: + return 0; + +error: + /* Destroy old pollfd */ + free(old_pollfd); + return -1; +} + +/* + * This thread manage application communication. */ static void *thread_manage_apps(void *data) +{ + int i, ret, current_nb_fd; + unsigned int nb_fd = 2; + int update_poll_flag = 1; + struct pollfd *pollfd = NULL; + struct ust_command ust_cmd; + + DBG("[thread] Manage application started"); + + ust_cmd.sock = -1; + current_nb_fd = nb_fd; + + while (1) { + /* See if we have a valid socket to add to pollfd */ + if (ust_cmd.sock != -1) { + nb_fd++; + update_poll_flag = 1; + } + + /* The pollfd struct must be updated */ + if (update_poll_flag) { + ret = update_apps_cmd_pollfd(nb_fd, current_nb_fd, &pollfd); + if (ret < 0) { + /* malloc failed so we quit */ + goto error; + } + + if (ust_cmd.sock != -1) { + /* Update pollfd with the new UST socket */ + DBG("Adding sock %d to apps cmd pollfd", ust_cmd.sock); + pollfd[nb_fd - 1].fd = ust_cmd.sock; + pollfd[nb_fd - 1].events = POLLHUP | POLLNVAL | POLLERR; + ust_cmd.sock = -1; + } + } + + DBG("Apps thread polling on %d fds", nb_fd); + + /* Inifinite blocking call, waiting for transmission */ + ret = poll(pollfd, nb_fd, -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 { + /* apps_cmd_pipe pipe events */ + switch (pollfd[1].revents) { + case POLLERR: + ERR("Apps command pipe poll error"); + goto error; + case POLLIN: + /* 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); + } + break; + } + } + + current_nb_fd = nb_fd; + for (i = 2; i < current_nb_fd; i++) { + /* Apps socket is closed/hungup */ + switch (pollfd[i].revents) { + case POLLERR: + case POLLHUP: + case POLLNVAL: + /* Pipe closed */ + unregister_traceable_app(pollfd[i].fd); + /* Indicate to remove this fd from the pollfd */ + pollfd[i].fd = -1; + nb_fd--; + break; + } + } + + if (nb_fd != current_nb_fd) { + update_poll_flag = 1; + } + } + +error: + DBG("Application communication apps dying"); + close(apps_cmd_pipe[0]); + close(apps_cmd_pipe[1]); + + free(pollfd); + + 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, ret; struct pollfd pollfd[2]; + /* + * Get allocated in this thread, enqueued to a global queue, dequeued and + * freed in the manage apps thread. + */ + struct ust_command *ust_cmd = NULL; - /* 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 apps started"); + DBG("[thread] Manage application registration started"); ret = lttcomm_listen_unix_sock(apps_sock); if (ret < 0) { @@ -760,13 +1113,20 @@ static void *thread_manage_apps(void *data) /* First fd is always the quit pipe */ pollfd[0].fd = thread_quit_pipe[0]; + pollfd[0].events = + POLLHUP | POLLNVAL | POLLERR | POLLIN | POLLRDHUP | POLLPRI; /* Apps socket */ pollfd[1].fd = apps_sock; pollfd[1].events = POLLIN; /* 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"); @@ -774,59 +1134,86 @@ static void *thread_manage_apps(void *data) /* Inifinite blocking call, waiting for transmission */ ret = poll(pollfd, 2, -1); if (ret < 0) { - perror("poll apps thread"); + perror("poll register 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; } - sock = lttcomm_accept_unix_sock(apps_sock); - if (sock < 0) { + switch (pollfd[1].revents) { + case POLLNVAL: + case POLLHUP: + case POLLRDHUP: + case POLLERR: + ERR("Register apps socket poll error"); 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; - } + case POLLIN: + 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. - */ + /* Create UST registration command for enqueuing */ + ust_cmd = malloc(sizeof(struct ust_command)); + if (ust_cmd == NULL) { + perror("ust command malloc"); goto error; } - } else { - /* Unregistering */ - unregister_traceable_app(reg_msg.pid); + + /* + * 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); + break; } } 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); + return NULL; } @@ -862,6 +1249,26 @@ error: return ret; } +/* + * Join kernel consumer thread + */ +static int join_kconsumerd_thread(void) +{ + void *status; + int ret; + + if (kconsumerd_pid != 0) { + ret = kill(kconsumerd_pid, SIGTERM); + if (ret) { + ERR("Error killing kconsumerd"); + return ret; + } + return pthread_join(kconsumerd_thread, &status); + } else { + return 0; + } +} + /* * Fork and exec a kernel consumer daemon (kconsumerd). * @@ -880,7 +1287,7 @@ static pid_t spawn_kconsumerd(void) /* * Exec kconsumerd. */ - if (opt_verbose > 1) { + if (opt_verbose > 1 || opt_verbose_kconsumerd) { verbosity = "--verbose"; } else { verbosity = "--quiet"; @@ -926,88 +1333,52 @@ static int start_kconsumerd(void) /* Setting up the global kconsumerd_pid */ kconsumerd_pid = ret; - pthread_mutex_unlock(&kconsumerd_pid_mutex); - - DBG("Kconsumerd pid %d", ret); - - DBG("Spawning kconsumerd thread"); - ret = spawn_kconsumerd_thread(); - if (ret < 0) { - ERR("Fatal error spawning kconsumerd thread"); - goto error; - } - -end: - return 0; - -error: - return ret; -} - -/* - * modprobe_kernel_modules - */ -static int modprobe_kernel_modules(void) -{ - int ret = 0, i; - char modprobe[256]; - - for (i = 0; i < ARRAY_SIZE(kernel_modules_list); i++) { - ret = snprintf(modprobe, sizeof(modprobe), - "/sbin/modprobe %s%s", - kernel_modules_list[i].required ? "" : "--quiet ", - kernel_modules_list[i].name); - if (ret < 0) { - perror("snprintf modprobe"); - goto error; - } - modprobe[sizeof(modprobe) - 1] = '\0'; - ret = system(modprobe); - if (ret == -1) { - ERR("Unable to launch modprobe for module %s", - kernel_modules_list[i].name); - } else if (kernel_modules_list[i].required - && WEXITSTATUS(ret) != 0) { - ERR("Unable to load module %s", - kernel_modules_list[i].name); - } else { - DBG("Modprobe successfully %s", - kernel_modules_list[i].name); - } + pthread_mutex_unlock(&kconsumerd_pid_mutex); + + DBG("Kconsumerd pid %d", ret); + + DBG("Spawning kconsumerd thread"); + ret = spawn_kconsumerd_thread(); + if (ret < 0) { + ERR("Fatal error spawning kconsumerd thread"); + goto error; } +end: + return 0; + error: return ret; } /* - * modprobe_remove_kernel_modules - * Remove modules in reverse load order. + * modprobe_kernel_modules */ -static int modprobe_remove_kernel_modules(void) +static int modprobe_kernel_modules(void) { int ret = 0, i; char modprobe[256]; - for (i = ARRAY_SIZE(kernel_modules_list) - 1; i >= 0; i--) { + for (i = 0; i < ARRAY_SIZE(kernel_modules_list); i++) { ret = snprintf(modprobe, sizeof(modprobe), - "/sbin/modprobe --remove --quiet %s", + "/sbin/modprobe %s%s", + kernel_modules_list[i].required ? "" : "--quiet ", kernel_modules_list[i].name); if (ret < 0) { - perror("snprintf modprobe --remove"); + perror("snprintf modprobe"); goto error; } modprobe[sizeof(modprobe) - 1] = '\0'; ret = system(modprobe); if (ret == -1) { - ERR("Unable to launch modprobe --remove for module %s", + ERR("Unable to launch modprobe for module %s", kernel_modules_list[i].name); } else if (kernel_modules_list[i].required - && WEXITSTATUS(ret) != 0) { - ERR("Unable to remove module %s", + && WEXITSTATUS(ret) != 0) { + ERR("Unable to load module %s", kernel_modules_list[i].name); } else { - DBG("Modprobe removal successful %s", + DBG("Modprobe successfully %s", kernel_modules_list[i].name); } } @@ -1024,7 +1395,7 @@ static int mount_debugfs(char *path) int ret; char *type = "debugfs"; - ret = mkdir_recursive(path, S_IRWXU | S_IRWXG); + ret = mkdir_recursive(path, S_IRWXU | S_IRWXG, geteuid(), getegid()); if (ret < 0) { goto error; } @@ -1131,7 +1502,16 @@ 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; @@ -1164,7 +1544,8 @@ 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; @@ -1176,18 +1557,66 @@ 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; } /* @@ -1205,9 +1634,15 @@ static int create_kernel_session(struct ltt_session *session) goto error; } - ret = mkdir_recursive(session->path, S_IRWXU | S_IRWXG ); + /* 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) { + if (ret != -EEXIST) { ERR("Trace directory creation error"); goto error; } @@ -1236,7 +1671,9 @@ 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++; } } @@ -1282,6 +1719,7 @@ 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: @@ -1316,7 +1754,9 @@ 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: @@ -1354,13 +1794,12 @@ 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) { @@ -1369,9 +1808,9 @@ 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; @@ -1380,6 +1819,8 @@ static int process_client_msg(struct command_ctx *cmd_ctx) } } break; + case LTTNG_DOMAIN_UST_PID: + break; default: break; } @@ -1405,6 +1846,7 @@ 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, @@ -1435,7 +1877,7 @@ 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; @@ -1473,14 +1915,14 @@ 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); @@ -1516,7 +1958,7 @@ 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; @@ -1557,14 +1999,15 @@ 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"); 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; @@ -1587,9 +2030,11 @@ 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; } @@ -1614,28 +2059,34 @@ 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, - cmd_ctx->session->kernel_session); - if (kchan == NULL) { - DBG("Channel not found. Creating channel %s", 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(channel_name); - if (chan == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } + 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->path); - if (ret < 0) { - ret = LTTCOMM_KERN_CHAN_FAIL; - 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; } - } while (kchan == NULL); + kchan = trace_kernel_get_channel_by_name(channel_name, + cmd_ctx->session->kernel_session); + if (kchan == NULL) { + ERR("Channel %s not found after creation. Internal error, giving up.", + channel_name); + ret = LTTCOMM_FATAL; + goto error; + } + } - 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); @@ -1686,26 +2137,32 @@ 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, - cmd_ctx->session->kernel_session); - if (kchan == NULL) { - DBG("Channel not found. Creating channel %s", 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(channel_name); - if (chan == NULL) { - ret = LTTCOMM_FATAL; - goto error; - } + 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->path); - if (ret < 0) { - ret = LTTCOMM_KERN_CHAN_FAIL; - 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) { + ERR("Channel %s not found after creation. Internal error, giving up.", + channel_name); + 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) { @@ -1724,7 +2181,7 @@ 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; @@ -1804,7 +2261,7 @@ 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; @@ -1838,16 +2295,16 @@ 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; } @@ -1961,7 +2418,7 @@ 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) { @@ -2008,7 +2465,7 @@ 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; @@ -2023,7 +2480,7 @@ static int process_client_msg(struct command_ctx *cmd_ctx) goto setup_error; } - DBG("Listing events (%ld events)", nb_event); + DBG("Listing events (%zu events)", nb_event); list_lttng_events(kchan, (struct lttng_event *)(cmd_ctx->llm->payload)); @@ -2056,7 +2513,6 @@ static int process_client_msg(struct command_ctx *cmd_ctx) ret = LTTCOMM_OK; break; } - case LTTNG_CALIBRATE: { /* Setup lttng message with no payload */ @@ -2086,6 +2542,43 @@ 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 */ @@ -2143,6 +2636,8 @@ static void *thread_manage_clients(void *data) /* First fd is always the quit pipe */ pollfd[0].fd = thread_quit_pipe[0]; + pollfd[0].events = + POLLHUP | POLLNVAL | POLLERR | POLLIN | POLLRDHUP | POLLPRI; /* Apps socket */ pollfd[1].fd = client_sock; @@ -2260,6 +2755,7 @@ static void usage(void) fprintf(stderr, " -S, --sig-parent Send SIGCHLD to parent pid to notify readiness.\n"); fprintf(stderr, " -q, --quiet No output at all.\n"); fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); + fprintf(stderr, " --verbose-kconsumerd Verbose mode for kconsumerd. Activate DBG() macro.\n"); } /* @@ -2281,12 +2777,13 @@ static int parse_args(int argc, char **argv) { "version", 0, 0, 'V' }, { "quiet", 0, 0, 'q' }, { "verbose", 0, 0, 'v' }, + { "verbose-kconsumerd", 0, 0, 'Z' }, { NULL, 0, 0, 0 } }; while (1) { int option_index = 0; - c = getopt_long(argc, argv, "dhqvVS" "a:c:g:s:E:C:", long_options, &option_index); + c = getopt_long(argc, argv, "dhqvVS" "a:c:g:s:E:C:Z", long_options, &option_index); if (c == -1) { break; } @@ -2332,6 +2829,9 @@ static int parse_args(int argc, char **argv) /* Verbose level can increase using multiple -v */ opt_verbose += 1; break; + case 'Z': + opt_verbose_kconsumerd += 1; + break; default: /* Unknown option or other error. * Error is printed by getopt, just return */ @@ -2347,7 +2847,7 @@ static int parse_args(int argc, char **argv) * apps_sock - The communication socket for all UST apps. * client_sock - The communication of the cli tool (lttng). */ -static int init_daemon_socket() +static int init_daemon_socket(void) { int ret = 0; mode_t old_umask; @@ -2392,18 +2892,21 @@ end: } /* - * Check if the global socket is available. If yes, error is returned. + * Check if the global socket is available, and if a daemon is answering + * at the other side. If yes, error is returned. */ -static int check_existing_daemon() +static int check_existing_daemon(void) { - int ret; - - ret = access(client_unix_sock_path, F_OK); - if (ret == 0) { - ret = access(apps_unix_sock_path, F_OK); + if (access(client_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()) { + return -EEXIST; + } else { + return 0; } - - return ret; } /* @@ -2415,14 +2918,10 @@ static int check_existing_daemon() static int set_permissions(void) { int ret; - struct group *grp; + gid_t gid; - /* Decide which group name to use */ - (opt_tracing_group != NULL) ? - (grp = getgrnam(opt_tracing_group)) : - (grp = getgrnam(default_tracing_group)); - - if (grp == NULL) { + gid = allowed_group(); + if (gid < 0) { if (is_root) { WARN("No tracing group detected"); ret = 0; @@ -2434,21 +2933,21 @@ static int set_permissions(void) } /* Set lttng run dir */ - ret = chown(LTTNG_RUNDIR, 0, grp->gr_gid); + ret = chown(LTTNG_RUNDIR, 0, gid); if (ret < 0) { ERR("Unable to set group on " LTTNG_RUNDIR); perror("chown"); } /* lttng client socket path */ - ret = chown(client_unix_sock_path, 0, grp->gr_gid); + ret = chown(client_unix_sock_path, 0, gid); if (ret < 0) { ERR("Unable to set group on %s", client_unix_sock_path); perror("chown"); } /* kconsumerd error socket path */ - ret = chown(kconsumerd_err_unix_sock_path, 0, grp->gr_gid); + ret = chown(kconsumerd_err_unix_sock_path, 0, gid); if (ret < 0) { ERR("Unable to set group on %s", kconsumerd_err_unix_sock_path); perror("chown"); @@ -2468,6 +2967,14 @@ 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. */ @@ -2536,26 +3043,27 @@ error: /* * Signal handler for the daemon + * + * Simply stop all worker threads, leaving main() return gracefully + * after joining all threads and calling cleanup(). */ static void sighandler(int sig) { switch (sig) { - case SIGPIPE: - DBG("SIGPIPE catched"); - return; - case SIGINT: - DBG("SIGINT catched"); - cleanup(); - break; - case SIGTERM: - DBG("SIGTERM catched"); - cleanup(); - break; - default: - break; + case SIGPIPE: + DBG("SIGPIPE catched"); + return; + case SIGINT: + DBG("SIGINT catched"); + stop_threads(); + break; + case SIGTERM: + DBG("SIGTERM catched"); + stop_threads(); + break; + default: + break; } - - exit(EXIT_SUCCESS); } /* @@ -2625,14 +3133,14 @@ int main(int argc, char **argv) const char *home_path; /* Create thread quit pipe */ - if (init_thread_quit_pipe() < 0) { - goto exit; + if ((ret = init_thread_quit_pipe()) < 0) { + goto error; } /* Parse arguments */ progname = argv[0]; if ((ret = parse_args(argc, argv) < 0)) { - goto exit; + goto error; } /* Daemonize */ @@ -2640,7 +3148,7 @@ int main(int argc, char **argv) ret = daemon(0, 0); if (ret < 0) { perror("daemon"); - goto exit; + goto error; } } @@ -2650,7 +3158,7 @@ int main(int argc, char **argv) if (is_root) { ret = create_lttng_rundir(); if (ret < 0) { - goto exit; + goto error; } if (strlen(apps_unix_sock_path) == 0) { @@ -2662,12 +3170,19 @@ 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) { /* TODO: Add --socket PATH option */ ERR("Can't get HOME directory for sockets creation."); - goto exit; + ret = -EPERM; + goto error; } if (strlen(apps_unix_sock_path) == 0) { @@ -2680,23 +3195,27 @@ 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); DBG("Application socket path %s", apps_unix_sock_path); /* - * See if daemon already exist. If any of the two socket needed by the - * daemon are present, this test fails. However, if the daemon is killed - * with a SIGKILL, those unix socket must be unlinked by hand. + * See if daemon already exist. */ - if ((ret = check_existing_daemon()) == 0) { + if ((ret = check_existing_daemon()) < 0) { ERR("Already running daemon.\n"); /* - * We do not goto error because we must not cleanup() because a daemon - * is already running. + * We do not goto exit because we must not cleanup() + * because a daemon is already running. */ - goto exit; + goto error; } /* After this point, we can safely call cleanup() so goto error is used */ @@ -2710,7 +3229,7 @@ int main(int argc, char **argv) if (is_root) { ret = set_kconsumerd_sockets(); if (ret < 0) { - goto error; + goto exit; } /* Setup kernel tracer */ @@ -2720,18 +3239,18 @@ int main(int argc, char **argv) set_ulimit(); } - if (set_signal_handler() < 0) { - goto error; + if ((ret = set_signal_handler()) < 0) { + goto exit; } /* Setup the needed unix socket */ - if (init_daemon_socket() < 0) { - goto error; + if ((ret = init_daemon_socket()) < 0) { + goto exit; } /* Set credentials to socket */ - if (is_root && (set_permissions() < 0)) { - goto error; + if (is_root && ((ret = set_permissions()) < 0)) { + goto exit; } /* Get parent pid if -S, --sig-parent is specified. */ @@ -2740,51 +3259,110 @@ int main(int argc, char **argv) } /* Setup the kernel pipe for waking up the kernel thread */ - if (create_kernel_poll_pipe() < 0) { - goto error; + if ((ret = create_kernel_poll_pipe()) < 0) { + 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(); - while (1) { - /* Create thread to manage the client socket */ - ret = pthread_create(&client_thread, NULL, thread_manage_clients, (void *) NULL); - if (ret != 0) { - perror("pthread_create"); - goto error; - } + /* Create thread to manage the client socket */ + ret = pthread_create(&client_thread, NULL, + thread_manage_clients, (void *) NULL); + if (ret != 0) { + perror("pthread_create clients"); + goto exit_client; + } - /* Create thread to manage application socket */ - ret = pthread_create(&apps_thread, NULL, thread_manage_apps, (void *) NULL); - if (ret != 0) { - perror("pthread_create"); - goto error; - } + /* 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 kernel thread to manage kernel event */ - ret = pthread_create(&kernel_thread, NULL, thread_manage_kernel, (void *) NULL); - if (ret != 0) { - perror("pthread_create"); - goto error; - } + /* 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; + } - ret = pthread_join(client_thread, &status); - if (ret != 0) { - perror("pthread_join"); - goto error; - } + /* Create thread to manage application socket */ + ret = pthread_create(&apps_thread, NULL, thread_manage_apps, (void *) NULL); + if (ret != 0) { + perror("pthread_create apps"); + goto exit_apps; } - cleanup(); - exit(EXIT_SUCCESS); + /* Create kernel thread to manage kernel event */ + ret = pthread_create(&kernel_thread, NULL, thread_manage_kernel, (void *) NULL); + if (ret != 0) { + perror("pthread_create kernel"); + goto exit_kernel; + } -error: - cleanup(); + ret = pthread_join(kernel_thread, &status); + if (ret != 0) { + perror("pthread_join"); + goto error; /* join error, exit without cleanup */ + } + +exit_kernel: + ret = pthread_join(apps_thread, &status); + if (ret != 0) { + perror("pthread_join"); + goto error; /* join error, exit without cleanup */ + } +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"); + goto error; /* join error, exit without cleanup */ + } + + ret = join_kconsumerd_thread(); + if (ret != 0) { + perror("join_kconsumerd"); + goto error; /* join error, exit without cleanup */ + } + +exit_client: exit: + /* + * cleanup() is called when no other thread is running. + */ + cleanup(); + if (!ret) + exit(EXIT_SUCCESS); +error: exit(EXIT_FAILURE); }