X-Git-Url: http://git.lttng.org/?a=blobdiff_plain;f=ustd%2Fustd.c;h=4db550222407ba7d0773aa8b15e1b33cda777de2;hb=2a79ceebcba229c91c93272705444c61c458bb06;hp=6f4acf8f5e0722b6dd6e9bef8d42a80c49a369b8;hpb=409e2abe93d854d2c1e0e5284c5cd32acb2ff6b0;p=ust.git

diff --git a/ustd/ustd.c b/ustd/ustd.c
index 6f4acf8..4db5502 100644
--- a/ustd/ustd.c
+++ b/ustd/ustd.c
@@ -33,9 +33,8 @@
 #include <getopt.h>
 
 #include "ustd.h"
-#include "localerr.h"
+#include "usterr.h"
 #include "ustcomm.h"
-#include "share.h"
 
 /* return value: 0 = subbuffer is finished, it won't produce data anymore
  *               1 = got subbuffer successfully
@@ -49,9 +48,12 @@
 #define PUT_SUBBUF_OK 1
 #define PUT_SUBBUF_DIED 0
 #define PUT_SUBBUF_PUSHED 2
+#define PUT_SUBBUF_DONE 3
 
 char *sock_path=NULL;
 char *trace_path=NULL;
+int daemon_mode = 0;
+char *pidfile = NULL;
 
 /* Number of active buffers and the mutex to protect it. */
 int active_buffers = 0;
@@ -69,7 +71,7 @@ int get_subbuffer(struct buffer_info *buf)
 
 	asprintf(&send_msg, "get_subbuffer %s", buf->name);
 	result = ustcomm_send_request(&buf->conn, send_msg, &received_msg);
-	if(result == -1 && errno == EPIPE || result == 0) {
+	if((result == -1 && errno == EPIPE) || result == 0) {
 		DBG("app died while being traced");
 		retval = GET_SUBBUF_DIED;
 		goto end;
@@ -97,6 +99,11 @@ int get_subbuffer(struct buffer_info *buf)
 		retval = GET_SUBBUF_DONE;
 		goto end_rep;
 	}
+	else if(!strcmp(received_msg, "NOTFOUND")) {
+		WARN("For buffer %s, the trace was not found. This likely means it was destroyed by the user.", buf->name);
+		retval = GET_SUBBUF_DONE;
+		goto end_rep;
+	}
 	else {
 		DBG("error getting subbuffer %s", buf->name);
 		retval = -1;
@@ -129,11 +136,18 @@ int put_subbuffer(struct buffer_info *buf)
 		retval = PUT_SUBBUF_DIED;
 		goto end;
 	}
-	if(result < 0) {
+	else if(result < 0) {
 		ERR("put_subbuffer: send_message failed");
 		retval = -1;
 		goto end;
 	}
+	else if(result == 0) {
+		/* Program seems finished. However this might not be
+		 * the last subbuffer that has to be collected.
+		 */
+		retval = PUT_SUBBUF_DIED;
+		goto end;
+	}
 
 	result = sscanf(received_msg, "%as", &rep_code);
 	if(result != 1) {
@@ -146,6 +160,14 @@ int put_subbuffer(struct buffer_info *buf)
 		DBG("subbuffer put %s", buf->name);
 		retval = PUT_SUBBUF_OK;
 	}
+	else if(!strcmp(received_msg, "NOTFOUND")) {
+		WARN("For buffer %s, the trace was not found. This likely means it was destroyed by the user.", buf->name);
+		/* However, maybe this was not the last subbuffer. So
+		 * we return the program died.
+		 */
+		retval = PUT_SUBBUF_DIED;
+		goto end_rep;
+	}
 	else {
 		DBG("put_subbuffer: received error, we were pushed");
 		retval = PUT_SUBBUF_PUSHED;
@@ -172,65 +194,6 @@ void decrement_active_buffers(void *arg)
 	pthread_mutex_unlock(&active_buffers_mutex);
 }
 
-void *consumer_thread(void *arg)
-{
-	struct buffer_info *buf = (struct buffer_info *) arg;
-	int result;
-
-	pthread_cleanup_push(decrement_active_buffers, NULL);
-
-	for(;;) {
-		/* get the subbuffer */
-		result = get_subbuffer(buf);
-		if(result == -1) {
-			ERR("error getting subbuffer");
-			continue;
-		}
-		else if(result == GET_SUBBUF_DONE) {
-			/* this is done */
-			break;
-		}
-		else if(result == GET_SUBBUF_DIED) {
-			finish_consuming_dead_subbuffer(buf);
-			break;
-		}
-
-		/* write data to file */
-		result = patient_write(buf->file_fd, buf->mem + (buf->consumed_old & (buf->n_subbufs * buf->subbuf_size-1)), buf->subbuf_size);
-		if(result == -1) {
-			PERROR("write");
-			/* FIXME: maybe drop this trace */
-		}
-
-		/* put the subbuffer */
-		result = put_subbuffer(buf);
-		if(result == -1) {
-			ERR("unknown error putting subbuffer (channel=%s)", buf->name);
-			break;
-		}
-		else if(result == PUT_SUBBUF_PUSHED) {
-			ERR("Buffer overflow (channel=%s), reader pushed. This channel will not be usable passed this point.", buf->name);
-			break;
-		}
-		else if(result == PUT_SUBBUF_DIED) {
-			WARN("application died while putting subbuffer");
-			/* FIXME: probably need to skip the first subbuffer in finish_consuming_dead_subbuffer */
-			finish_consuming_dead_subbuffer(buf);
-			break;
-		}
-		else if(result == PUT_SUBBUF_OK) {
-		}
-	}
-
-	DBG("thread for buffer %s is stopping", buf->name);
-
-	/* FIXME: destroy, unalloc... */
-
-	pthread_cleanup_pop(1);
-
-	return NULL;
-}
-
 int create_dir_if_needed(char *dir)
 {
 	int result;
@@ -263,7 +226,7 @@ int is_directory(const char *dir)
 	return 1;
 }
 
-int add_buffer(pid_t pid, char *bufname)
+struct buffer_info *connect_buffer(pid_t pid, const char *bufname)
 {
 	struct buffer_info *buf;
 	char *send_msg;
@@ -271,13 +234,12 @@ int add_buffer(pid_t pid, char *bufname)
 	int result;
 	char *tmp;
 	int fd;
-	pthread_t thr;
 	struct shmid_ds shmds;
 
 	buf = (struct buffer_info *) malloc(sizeof(struct buffer_info));
 	if(buf == NULL) {
 		ERR("add_buffer: insufficient memory");
-		return -1;
+		return NULL;
 	}
 
 	buf->name = bufname;
@@ -287,7 +249,7 @@ int add_buffer(pid_t pid, char *bufname)
 	result = ustcomm_connect_app(buf->pid, &buf->conn);
 	if(result) {
 		WARN("unable to connect to process, it probably died before we were able to connect");
-		return -1;
+		return NULL;
 	}
 
 	/* get pidunique */
@@ -296,13 +258,16 @@ int add_buffer(pid_t pid, char *bufname)
 	free(send_msg);
 	if(result == -1) {
 		ERR("problem in ustcomm_send_request(get_pidunique)");
-		return -1;
+		return NULL;
+	}
+	if(result == 0) {
+		goto error;
 	}
 
 	result = sscanf(received_msg, "%lld", &buf->pidunique);
 	if(result != 1) {
 		ERR("unable to parse response to get_pidunique");
-		return -1;
+		return NULL;
 	}
 	free(received_msg);
 	DBG("got pidunique %lld", buf->pidunique);
@@ -313,13 +278,16 @@ int add_buffer(pid_t pid, char *bufname)
 	free(send_msg);
 	if(result == -1) {
 		ERR("problem in ustcomm_send_request(get_shmid)");
-		return -1;
+		return NULL;
+	}
+	if(result == 0) {
+		goto error;
 	}
 
 	result = sscanf(received_msg, "%d %d", &buf->shmid, &buf->bufstruct_shmid);
 	if(result != 2) {
-		ERR("unable to parse response to get_shmid");
-		return -1;
+		ERR("unable to parse response to get_shmid (\"%s\")", received_msg);
+		return NULL;
 	}
 	free(received_msg);
 	DBG("got shmids %d %d", buf->shmid, buf->bufstruct_shmid);
@@ -330,26 +298,36 @@ int add_buffer(pid_t pid, char *bufname)
 	free(send_msg);
 	if(result == -1) {
 		ERR("problem in ustcomm_send_request(g_n_subbufs)");
-		return -1;
+		return NULL;
+	}
+	if(result == 0) {
+		goto error;
 	}
 
 	result = sscanf(received_msg, "%d", &buf->n_subbufs);
 	if(result != 1) {
 		ERR("unable to parse response to get_n_subbufs");
-		return -1;
+		return NULL;
 	}
 	free(received_msg);
 	DBG("got n_subbufs %d", buf->n_subbufs);
 
 	/* get subbuf size */
 	asprintf(&send_msg, "get_subbuf_size %s", buf->name);
-	ustcomm_send_request(&buf->conn, send_msg, &received_msg);
+	result = ustcomm_send_request(&buf->conn, send_msg, &received_msg);
 	free(send_msg);
+	if(result == -1) {
+		ERR("problem in ustcomm_send_request(get_subbuf_size)");
+		return NULL;
+	}
+	if(result == 0) {
+		goto error;
+	}
 
 	result = sscanf(received_msg, "%d", &buf->subbuf_size);
 	if(result != 1) {
 		ERR("unable to parse response to get_subbuf_size");
-		return -1;
+		return NULL;
 	}
 	free(received_msg);
 	DBG("got subbuf_size %d", buf->subbuf_size);
@@ -358,14 +336,14 @@ int add_buffer(pid_t pid, char *bufname)
 	buf->mem = shmat(buf->shmid, NULL, 0);
 	if(buf->mem == (void *) 0) {
 		PERROR("shmat");
-		return -1;
+		return NULL;
 	}
 	DBG("successfully attached buffer memory");
 
 	buf->bufstruct_mem = shmat(buf->bufstruct_shmid, NULL, 0);
 	if(buf->bufstruct_mem == (void *) 0) {
 		PERROR("shmat");
-		return -1;
+		return NULL;
 	}
 	DBG("successfully attached buffer bufstruct memory");
 
@@ -373,7 +351,7 @@ int add_buffer(pid_t pid, char *bufname)
 	result = shmctl(buf->shmid, IPC_STAT, &shmds);
 	if(result == -1) {
 		PERROR("shmctl");
-		return -1;
+		return NULL;
 	}
 	buf->memlen = shmds.shm_segsz;
 
@@ -386,7 +364,7 @@ int add_buffer(pid_t pid, char *bufname)
 		result = create_dir_if_needed(USTD_DEFAULT_TRACE_PATH);
 		if(result == -1) {
 			ERR("could not create directory %s", USTD_DEFAULT_TRACE_PATH);
-			return -1;
+			return NULL;
 		}
 
 		trace_path = USTD_DEFAULT_TRACE_PATH;
@@ -397,16 +375,16 @@ int add_buffer(pid_t pid, char *bufname)
 	if(result == -1) {
 		ERR("could not create directory %s", tmp);
 		free(tmp);
-		return -1;
+		return NULL;
 	}
 	free(tmp);
 
-	asprintf(&tmp, "%s/%u_%lld/%s_0", trace_path, buf->pid, buf->pidunique, buf->name);
+	asprintf(&tmp, "%s/%u_%lld/%s", trace_path, buf->pid, buf->pidunique, buf->name);
 	result = fd = open(tmp, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL, 00600);
 	if(result == -1) {
 		PERROR("open");
 		ERR("failed opening trace file %s", tmp);
-		return -1;
+		return NULL;
 	}
 	buf->file_fd = fd;
 	free(tmp);
@@ -415,7 +393,173 @@ int add_buffer(pid_t pid, char *bufname)
 	active_buffers++;
 	pthread_mutex_unlock(&active_buffers_mutex);
 
-	pthread_create(&thr, NULL, consumer_thread, buf);
+	return buf;
+
+error:
+	free(buf);
+	return NULL;
+}
+
+int unwrite_last_subbuffer(struct buffer_info *buf)
+{
+	int result;
+
+	result = ftruncate(buf->file_fd, buf->previous_offset);
+	if(result == -1) {
+		PERROR("ftruncate");
+		return -1;
+	}
+
+	result = lseek(buf->file_fd, buf->previous_offset, SEEK_SET);
+	if(result == (int)(off_t)-1) {
+		PERROR("lseek");
+		return -1;
+	}
+
+	return 0;
+}
+
+int write_current_subbuffer(struct buffer_info *buf)
+{
+	int result;
+
+	void *subbuf_mem = buf->mem + (buf->consumed_old & (buf->n_subbufs * buf->subbuf_size-1));
+
+	size_t cur_sb_size = subbuffer_data_size(subbuf_mem);
+
+	off_t cur_offset = lseek(buf->file_fd, 0, SEEK_CUR);
+	if(cur_offset == (off_t)-1) {
+		PERROR("lseek");
+		return -1;
+	}
+
+	buf->previous_offset = cur_offset;
+	DBG("previous_offset: %ld", cur_offset);
+
+	result = patient_write(buf->file_fd, subbuf_mem, cur_sb_size);
+	if(result == -1) {
+		PERROR("write");
+		/* FIXME: maybe drop this trace */
+		return -1;
+	}
+
+	return 0;
+}
+
+int consumer_loop(struct buffer_info *buf)
+{
+	int result;
+
+	pthread_cleanup_push(decrement_active_buffers, NULL);
+
+	for(;;) {
+		/* get the subbuffer */
+		result = get_subbuffer(buf);
+		if(result == -1) {
+			ERR("error getting subbuffer");
+			continue;
+		}
+		else if(result == GET_SUBBUF_DONE) {
+			/* this is done */
+			break;
+		}
+		else if(result == GET_SUBBUF_DIED) {
+			finish_consuming_dead_subbuffer(buf);
+			break;
+		}
+
+		/* write data to file */
+		write_current_subbuffer(buf);
+		/* FIXME: handle return value? */
+
+		/* put the subbuffer */
+		result = put_subbuffer(buf);
+		if(result == -1) {
+			ERR("unknown error putting subbuffer (channel=%s)", buf->name);
+			break;
+		}
+		else if(result == PUT_SUBBUF_PUSHED) {
+			ERR("Buffer overflow (channel=%s), reader pushed. This channel will not be usable passed this point.", buf->name);
+			break;
+		}
+		else if(result == PUT_SUBBUF_DIED) {
+			WARN("application died while putting subbuffer");
+			/* Skip the first subbuffer. We are not sure it is trustable
+			 * because the put_subbuffer() did not complete.
+			 */
+			unwrite_last_subbuffer(buf);
+			finish_consuming_dead_subbuffer(buf);
+			break;
+		}
+		else if(result == PUT_SUBBUF_DONE) {
+			/* Done with this subbuffer */
+			/* FIXME: add a case where this branch is used? Upon
+			 * normal trace termination, at put_subbuf time, a
+			 * special last-subbuffer code could be returned by
+			 * the listener.
+			 */
+			break;
+		}
+		else if(result == PUT_SUBBUF_OK) {
+		}
+	}
+
+	DBG("thread for buffer %s is stopping", buf->name);
+
+	/* FIXME: destroy, unalloc... */
+
+	pthread_cleanup_pop(1);
+
+	return 0;
+}
+
+void free_buffer(struct buffer_info *buf)
+{
+}
+
+struct consumer_thread_args {
+	pid_t pid;
+	const char *bufname;
+};
+
+void *consumer_thread(void *arg)
+{
+	struct buffer_info *buf = (struct buffer_info *) arg;
+	struct consumer_thread_args *args = (struct consumer_thread_args *) arg;
+
+	DBG("GOT ARGS: pid %d bufname %s", args->pid, args->bufname);
+
+	buf = connect_buffer(args->pid, args->bufname);
+	if(buf == NULL) {
+		ERR("failed to connect to buffer");
+		goto end;
+	}
+
+	consumer_loop(buf);
+
+	free_buffer(buf);
+
+	end:
+	/* bufname is free'd in free_buffer() */
+	free(args);
+	return NULL;
+}
+
+int start_consuming_buffer(pid_t pid, const char *bufname)
+{
+	pthread_t thr;
+	struct consumer_thread_args *args;
+
+	DBG("beginning of start_consuming_buffer: args: pid %d bufname %s", pid, bufname);
+
+	args = (struct consumer_thread_args *) malloc(sizeof(struct consumer_thread_args));
+
+	args->pid = pid;
+	args->bufname = strdup(bufname);
+	DBG("beginning2 of start_consuming_buffer: args: pid %d bufname %s", args->pid, args->bufname);
+
+	pthread_create(&thr, NULL, consumer_thread, args);
+	DBG("end of start_consuming_buffer: args: pid %d bufname %s", args->pid, args->bufname);
 
 	return 0;
 }
@@ -425,7 +569,9 @@ void usage(void)
 	fprintf(stderr, "Usage:\nustd OPTIONS\n\nOptions:\n"
 			"\t-h\t\tDisplay this usage.\n"
 			"\t-o DIR\t\tSpecify the directory where to output the traces.\n"
-			"\t-s PATH\t\tSpecify the path to use for the daemon socket.\n");
+			"\t-s PATH\t\tSpecify the path to use for the daemon socket.\n"
+			"\t-d\t\tStart as a daemon.\n"
+			"\t--pidfile FILE\tWrite the PID in this file (when using -d).\n");
 }
 
 int parse_args(int argc, char **argv)
@@ -435,12 +581,13 @@ int parse_args(int argc, char **argv)
 	while (1) {
 		int option_index = 0;
 		static struct option long_options[] = {
+			{"pidfile", 1, 0, 'p'},
 			{"help", 0, 0, 'h'},
 			{"version", 0, 0, 'V'},
 			{0, 0, 0, 0}
 		};
 
-		c = getopt_long(argc, argv, "hs:o:", long_options, &option_index);
+		c = getopt_long(argc, argv, "hs:o:d", long_options, &option_index);
 		if (c == -1)
 			break;
 
@@ -461,6 +608,12 @@ int parse_args(int argc, char **argv)
 				return -1;
 			}
 			break;
+		case 'd':
+			daemon_mode = 1;
+			break;
+		case 'p':
+			pidfile = strdup(optarg);
+			break;
 		case 'h':
 			usage();
 			exit(0);
@@ -483,7 +636,25 @@ void sigterm_handler(int sig)
 	terminate_req = 1;
 }
 
-int main(int argc, char **argv)
+static int write_pidfile(const char *file_name, pid_t pid)
+{
+	FILE *pidfp;
+
+	pidfp = fopen(file_name, "w");
+	if(!pidfp) {
+		PERROR("fopen (%s)", pidfile);
+		WARN("killing child process");
+		return -1;
+	}
+
+	fprintf(pidfp, "%d\n", pid);
+
+	fclose(pidfp);
+
+	return 0;
+}
+
+int start_ustd(int fd)
 {
 	struct ustcomm_ustd ustd;
 	int result;
@@ -503,10 +674,10 @@ int main(int argc, char **argv)
 		PERROR("sigaction");
 		return 1;
 	}
-
-	result = parse_args(argc, argv);
+	result = sigaction(SIGINT, &sa, NULL);
 	if(result == -1) {
-		exit(1);
+		PERROR("sigaction");
+		return 1;
 	}
 
 	result = ustcomm_init_ustd(&ustd, sock_path);
@@ -532,6 +703,33 @@ int main(int argc, char **argv)
 		return 1;
 	}
 
+	/* Write pidfile */
+	if(pidfile) {
+		result = write_pidfile(pidfile, getpid());
+		if(result == -1) {
+			ERR("failed to write pidfile");
+			return 1;
+		}
+	}
+
+	/* Notify parent that we are successfully started. */
+	if(fd != -1) {
+		/* write any one character */
+		result = write(fd, "!", 1);
+		if(result == -1) {
+			PERROR("write");
+			return -1;
+		}
+		if(result != 1) {
+			ERR("Problem sending confirmation of daemon start to parent");
+			return -1;
+		}
+		result = close(fd);
+		if(result == -1) {
+			PERROR("close");
+		}
+	}
+
 	/* app loop */
 	for(;;) {
 		char *recvbuf;
@@ -540,7 +738,7 @@ int main(int argc, char **argv)
 		result = ustcomm_ustd_recv_message(&ustd, &recvbuf, NULL, 100);
 		if(result == -1) {
 			ERR("error in ustcomm_ustd_recv_message");
-			continue;
+			goto loop_end;
 		}
 		if(result > 0) {
 			if(!strncmp(recvbuf, "collect", 7)) {
@@ -550,19 +748,28 @@ int main(int argc, char **argv)
 
 				result = sscanf(recvbuf, "%*s %d %50as", &pid, &bufname);
 				if(result != 2) {
-					fprintf(stderr, "parsing error: %s\n", recvbuf);
+					ERR("parsing error: %s", recvbuf);
+					goto free_bufname;
 				}
 
-				result = add_buffer(pid, bufname);
+				result = start_consuming_buffer(pid, bufname);
 				if(result < 0) {
 					ERR("error in add_buffer");
-					continue;
+					goto free_bufname;
 				}
+
+				free_bufname:
+				free(bufname);
+			}
+			else {
+				WARN("unknown command: %s", recvbuf);
 			}
 
 			free(recvbuf);
 		}
 
+		loop_end:
+
 		if(terminate_req) {
 			pthread_mutex_lock(&active_buffers_mutex);
 			if(active_buffers == 0) {
@@ -573,5 +780,69 @@ int main(int argc, char **argv)
 		}
 	}
 
+	ustcomm_fini_ustd(&ustd);
+
 	return 0;
 }
+
+int start_ustd_daemon()
+{
+	int result;
+	int fd[2];
+	pid_t child_pid;
+
+	result = pipe(fd);
+
+	result = child_pid = fork();
+	if(result == -1) {
+		PERROR("fork");
+		return -1;
+	}
+	else if(result == 0) {
+		return start_ustd(fd[1]);
+	}
+	else {
+		char buf;
+
+		result = read(fd[0], &buf, 1);
+		if(result == -1) {
+			PERROR("read");
+			return -1;
+		}
+		if(result != 1) {
+			ERR("did not receive valid confirmation that the daemon is started");
+			return -1;
+		}
+
+		result = close(fd[0]);
+		if(result == -1) {
+			PERROR("close");
+		}
+
+		DBG("The daemon is now successfully started");
+	}
+
+	/* Wait for confirmation that the server is ready. */
+
+
+	return 0;
+}
+
+int main(int argc, char **argv)
+{
+	int result;
+
+	result = parse_args(argc, argv);
+	if(result == -1) {
+		exit(1);
+	}
+
+	if(daemon_mode) {
+		result = start_ustd_daemon();
+	}
+	else {
+		result = start_ustd(-1);
+	}
+
+	return result;
+}