From ececeffd85db1b27c07cdf91a921fd203006daf6 Mon Sep 17 00:00:00 2001
From: Rusty Russell <rusty@rustcorp.com.au>
Date: Wed, 24 Feb 2010 10:53:05 +1030
Subject: tdb: add -k option to tdbtorture

To test the case of death of a process during transaction commit, add
a -k (kill random) option to tdbtorture.  The easiest way to do this
is to make every worker a child (unless there's only one child), which
is why this patch is bigger than you might expect.

Using -k without -t (always transactions) you expect corruption, though
it doesn't happen every time.  With -t, we currently get corruption but
the next patch fixes that.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
---
 lib/tdb/tools/tdbtorture.c | 199 ++++++++++++++++++++++++++++++++-------------
 1 file changed, 142 insertions(+), 57 deletions(-)

(limited to 'lib/tdb')

diff --git a/lib/tdb/tools/tdbtorture.c b/lib/tdb/tools/tdbtorture.c
index b0221a2503..5ee6991825 100644
--- a/lib/tdb/tools/tdbtorture.c
+++ b/lib/tdb/tools/tdbtorture.c
@@ -30,6 +30,10 @@ static struct tdb_context *db;
 static int in_transaction;
 static int error_count;
 static int always_transaction = 0;
+static int hash_size = 2;
+static int loopnum;
+static int count_pipe;
+static struct tdb_logging_context log_ctx;
 
 #ifdef PRINTF_ATTRIBUTE
 static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const char *format, ...) PRINTF_ATTRIBUTE(3,4);
@@ -48,8 +52,9 @@ static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const c
 	va_end(ap);
 	fflush(stdout);
 #if 0
-	{
+	if (level != TDB_DEBUG_TRACE) {
 		char *ptr;
+		signal(SIGUSR1, SIG_IGN);
 		asprintf(&ptr,"xterm -e gdb /proc/%d/exe %d", getpid(), getpid());
 		system(ptr);
 		free(ptr);
@@ -211,24 +216,74 @@ static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
 
 static void usage(void)
 {
-	printf("Usage: tdbtorture [-t] [-n NUM_PROCS] [-l NUM_LOOPS] [-s SEED] [-H HASH_SIZE]\n");
+	printf("Usage: tdbtorture [-t] [-k] [-n NUM_PROCS] [-l NUM_LOOPS] [-s SEED] [-H HASH_SIZE]\n");
 	exit(0);
 }
 
- int main(int argc, char * const *argv)
+static void send_count_and_suicide(int sig)
+{
+	/* This ensures our successor can continue where we left off. */
+	write(count_pipe, &loopnum, sizeof(loopnum));
+	/* This gives a unique signature. */
+	kill(getpid(), SIGUSR2);
+}
+
+static int run_child(int i, int seed, unsigned num_loops, unsigned start)
+{
+	db = tdb_open_ex("torture.tdb", hash_size, TDB_DEFAULT,
+			 O_RDWR | O_CREAT, 0600, &log_ctx, NULL);
+	if (!db) {
+		fatal("db open failed");
+	}
+
+	srand(seed + i);
+	srandom(seed + i);
+
+	/* Set global, then we're ready to handle being killed. */
+	loopnum = start;
+	signal(SIGUSR1, send_count_and_suicide);
+
+	for (;loopnum<num_loops && error_count == 0;loopnum++) {
+		addrec_db();
+	}
+
+	if (error_count == 0) {
+		tdb_traverse_read(db, NULL, NULL);
+		if (always_transaction) {
+			while (in_transaction) {
+				tdb_transaction_cancel(db);
+				in_transaction--;
+			}
+			if (tdb_transaction_start(db) != 0)
+				fatal("tdb_transaction_start failed");
+		}
+		tdb_traverse(db, traverse_fn, NULL);
+		tdb_traverse(db, traverse_fn, NULL);
+		if (always_transaction) {
+			if (tdb_transaction_commit(db) != 0)
+				fatal("tdb_transaction_commit failed");
+		}
+	}
+
+	tdb_close(db);
+
+	return (error_count < 100 ? error_count : 100);
+}
+
+int main(int argc, char * const *argv)
 {
 	int i, seed = -1;
-	int num_procs = 3;
 	int num_loops = 5000;
-	int hash_size = 2;
-	int c;
+	int num_procs = 3;
+	int c, pfds[2];
 	extern char *optarg;
 	pid_t *pids;
+	int kill_random = 0;
+	int *done;
 
-	struct tdb_logging_context log_ctx;
 	log_ctx.log_fn = tdb_log;
 
-	while ((c = getopt(argc, argv, "n:l:s:H:th")) != -1) {
+	while ((c = getopt(argc, argv, "n:l:s:H:thk")) != -1) {
 		switch (c) {
 		case 'n':
 			num_procs = strtol(optarg, NULL, 0);
@@ -245,6 +300,9 @@ static void usage(void)
 		case 't':
 			always_transaction = 1;
 			break;
+		case 'k':
+			kill_random = 1;
+			break;
 		default:
 			usage();
 		}
@@ -252,93 +310,120 @@ static void usage(void)
 
 	unlink("torture.tdb");
 
-	pids = (pid_t *)calloc(sizeof(pid_t), num_procs);
-	pids[0] = getpid();
-
-	for (i=0;i<num_procs-1;i++) {
-		if ((pids[i+1]=fork()) == 0) break;
-	}
-
-	db = tdb_open_ex("torture.tdb", hash_size, TDB_CLEAR_IF_FIRST, 
-			 O_RDWR | O_CREAT, 0600, &log_ctx, NULL);
-	if (!db) {
-		fatal("db open failed");
-	}
-
 	if (seed == -1) {
 		seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
 	}
 
-	if (i == 0) {
-		printf("testing with %d processes, %d loops, %d hash_size, seed=%d%s\n",
-		       num_procs, num_loops, hash_size, seed, always_transaction ? " (all within transactions)" : "");
+	if (num_procs == 1 && !kill_random) {
+		/* Don't fork for this case, makes debugging easier. */
+		error_count = run_child(0, seed, num_loops, 0);
+		goto done;
 	}
 
-	srand(seed + i);
-	srandom(seed + i);
+	pids = (pid_t *)calloc(sizeof(pid_t), num_procs);
+	done = (int *)calloc(sizeof(int), num_procs);
 
-	for (i=0;i<num_loops && error_count == 0;i++) {
-		addrec_db();
+	if (pipe(pfds) != 0) {
+		perror("Creating pipe");
+		exit(1);
 	}
-
-	if (error_count == 0) {
-		tdb_traverse_read(db, NULL, NULL);
-		if (always_transaction) {
-			while (in_transaction) {
-				tdb_transaction_cancel(db);
-				in_transaction--;
+	count_pipe = pfds[1];
+
+	for (i=0;i<num_procs;i++) {
+		if ((pids[i]=fork()) == 0) {
+			close(pfds[0]);
+			if (i == 0) {
+				printf("testing with %d processes, %d loops, %d hash_size, seed=%d%s\n",
+				       num_procs, num_loops, hash_size, seed, always_transaction ? " (all within transactions)" : "");
 			}
-			if (tdb_transaction_start(db) != 0)
-				fatal("tdb_transaction_start failed");
-		}
-		tdb_traverse(db, traverse_fn, NULL);
-		tdb_traverse(db, traverse_fn, NULL);
-		if (always_transaction) {
-			if (tdb_transaction_commit(db) != 0)
-				fatal("tdb_transaction_commit failed");
+			exit(run_child(i, seed, num_loops, 0));
 		}
 	}
 
-	tdb_close(db);
-
-	if (getpid() != pids[0]) {
-		return error_count;
-	}
-
-	for (i=1;i<num_procs;i++) {
+	while (num_procs) {
 		int status, j;
 		pid_t pid;
+
 		if (error_count != 0) {
 			/* try and stop the test on any failure */
-			for (j=1;j<num_procs;j++) {
+			for (j=0;j<num_procs;j++) {
 				if (pids[j] != 0) {
 					kill(pids[j], SIGTERM);
 				}
 			}
 		}
-		pid = waitpid(-1, &status, 0);
+
+		pid = waitpid(-1, &status, kill_random ? WNOHANG : 0);
+		if (pid == 0) {
+			struct timespec ts;
+
+			/* Sleep for 1/10 second. */
+			ts.tv_sec = 0;
+			ts.tv_nsec = 100000000;
+			nanosleep(&ts, NULL);
+
+			/* Kill someone. */
+			kill(pids[random() % num_procs], SIGUSR1);
+			continue;
+		}
+
 		if (pid == -1) {
 			perror("failed to wait for child\n");
 			exit(1);
 		}
-		for (j=1;j<num_procs;j++) {
+
+		for (j=0;j<num_procs;j++) {
 			if (pids[j] == pid) break;
 		}
 		if (j == num_procs) {
 			printf("unknown child %d exited!?\n", (int)pid);
 			exit(1);
 		}
-		if (WEXITSTATUS(status) != 0) {
-			printf("child %d exited with status %d\n",
-			       (int)pid, WEXITSTATUS(status));
+		if (WIFSIGNALED(status)) {
+			if (WTERMSIG(status) == SIGUSR2
+			    || WTERMSIG(status) == SIGUSR1) {
+				/* SIGUSR2 means they wrote to pipe. */
+				if (WTERMSIG(status) == SIGUSR2) {
+					read(pfds[0], &done[j],
+					     sizeof(done[j]));
+				}
+				pids[j] = fork();
+				if (pids[j] == 0)
+					exit(run_child(j, seed, num_loops,
+						       done[j]));
+				printf("Restarting child %i for %u-%u\n",
+				       j, done[j], num_loops);
+				continue;
+			}
+			printf("child %d exited with signal %d\n",
+			       (int)pid, WTERMSIG(status));
 			error_count++;
+		} else {
+			if (WEXITSTATUS(status) != 0) {
+				printf("child %d exited with status %d\n",
+				       (int)pid, WEXITSTATUS(status));
+				error_count++;
+			}
 		}
-		pids[j] = 0;
+		memmove(&pids[j], &pids[j+1],
+			(num_procs - j - 1)*sizeof(pids[0]));
+		num_procs--;
 	}
 
 	free(pids);
 
+done:
 	if (error_count == 0) {
+		db = tdb_open_ex("torture.tdb", hash_size, TDB_DEFAULT,
+				 O_RDWR, 0, &log_ctx, NULL);
+		if (!db) {
+			fatal("db open failed");
+		}
+		if (tdb_check(db, NULL, NULL) == -1) {
+			printf("db check failed");
+			exit(1);
+		}
+		tdb_close(db);
 		printf("OK\n");
 	}
 
-- 
cgit