diff --git a/lustre/include/obd_support.h b/lustre/include/obd_support.h
index 62207d137ee59829fa5faecbbea8f0047d6634ae..84c0790545aadd33efddf231a06d7a1bb89277e4 100644
--- a/lustre/include/obd_support.h
+++ b/lustre/include/obd_support.h
@@ -229,6 +229,7 @@ extern unsigned int obd_alloc_fail_rate;
#define OBD_FAIL_LDLM_PAUSE_CANCEL 0x312
#define OBD_FAIL_LDLM_CLOSE_THREAD 0x313
#define OBD_FAIL_LDLM_CANCEL_BL_CB_RACE 0x314
+#define OBD_FAIL_LDLM_CP_CB_WAIT 0x315
#define OBD_FAIL_OSC 0x400
#define OBD_FAIL_OSC_BRW_READ_BULK 0x401
diff --git a/lustre/ldlm/ldlm_flock.c b/lustre/ldlm/ldlm_flock.c
index f5eb0b94221d69e20fbfc1b578c5535a7bf80ef0..b158d1d448997766f3ac280940793580f75e1d2d 100644
--- a/lustre/ldlm/ldlm_flock.c
+++ b/lustre/ldlm/ldlm_flock.c
@@ -558,6 +558,7 @@ ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data)
RETURN(rc);
granted:
+ OBD_FAIL_TIMEOUT(OBD_FAIL_LDLM_CP_CB_WAIT, 10);
LDLM_DEBUG(lock, "client-side enqueue granted");
ns = lock->l_resource->lr_namespace;
lock_res_and_lock(lock);
diff --git a/lustre/tests/Makefile.am b/lustre/tests/Makefile.am
index 93712fa549a09d771016bc19de17fe70261de44c..65b0f4a6ad851e68bc56636d44f7a210b0f15009 100644
--- a/lustre/tests/Makefile.am
+++ b/lustre/tests/Makefile.am
@@ -59,6 +59,9 @@ multiop_LDADD=$(LIBLUSTREAPI)
ll_dirstripe_verify_SOURCES= ll_dirstripe_verify.c
ll_dirstripe_verify_LDADD= -L$(top_builddir)/lustre/utils -llustreapi
+flocks_test_SOURCES=flocks_test.c
+flocks_test_LDADD=-lpthread
+
if MPITESTS
#LAM_LD_FLAGS=-L/opt/lam/lib -lmpi -llam -lpthread
LAM_LD_FLAGS=-lmpich -lpthread
diff --git a/lustre/tests/flocks_test.c b/lustre/tests/flocks_test.c
index 633e1b1b9a4c1cbcc42c9f8cf800305f47f0ce6e..97890d85f57976935eba4bad1722c14bca77616c 100644
--- a/lustre/tests/flocks_test.c
+++ b/lustre/tests/flocks_test.c
@@ -40,38 +40,118 @@
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
-
+#include <pthread.h>
#include <sys/file.h>
+#include <stdarg.h>
-void usage(void)
+#define MAX_PATH_LENGTH 4096
+/**
+ * helper functions
+ */
+int t_fcntl(int fd, int cmd, ...)
+{
+ va_list ap;
+ long arg;
+ struct flock *lock;
+ int rc = -1;
+
+ va_start(ap, cmd);
+ switch (cmd) {
+ case F_GETFL:
+ va_end(ap);
+ rc = fcntl(fd, cmd);
+ if (rc == -1) {
+ fprintf(stderr, "fcntl GETFL failed: %s\n",
+ strerror(errno));
+ return(1);
+ }
+ break;
+ case F_SETFL:
+ arg = va_arg(ap, long);
+ va_end(ap);
+ rc = fcntl(fd, cmd, arg);
+ if (rc == -1) {
+ fprintf(stderr, "fcntl SETFL %ld failed: %s\n",
+ arg, strerror(errno));
+ return(1);
+ }
+ break;
+ case F_GETLK:
+ case F_SETLK:
+ case F_SETLKW:
+ lock = va_arg(ap, struct flock *);
+ va_end(ap);
+ rc = fcntl(fd, cmd, lock);
+ if (rc == -1) {
+ fprintf(stderr, "fcntl cmd %d failed: %s\n",
+ cmd, strerror(errno));
+ return(1);
+ }
+ break;
+ case F_DUPFD:
+ arg = va_arg(ap, long);
+ va_end(ap);
+ rc = fcntl(fd, cmd, arg);
+ if (rc == -1) {
+ fprintf(stderr, "fcntl F_DUPFD %d failed: %s\n",
+ (int)arg, strerror(errno));
+ return(1);
+ }
+ break;
+ default:
+ va_end(ap);
+ fprintf(stderr, "fcntl cmd %d not supported\n", cmd);
+ return(1);
+ }
+ return rc;
+}
+
+int t_unlink(const char *path)
{
- fprintf(stderr, "usage: ./flocks_test on|off -c|-f|-l /path/to/file\n");
- exit(EXIT_FAILURE);
+ int rc;
+
+ rc = unlink(path);
+ if (rc)
+ fprintf(stderr, "unlink(%s) error: %s\n", path, strerror(errno));
+ return rc;
}
-int main(int argc, char *argv[])
+/** =================================================================
+ * test number 1
+ *
+ * normal flock test
+ */
+void t1_usage(void)
+{
+ fprintf(stderr, "usage: ./flocks_test 1 on|off -c|-f|-l /path/to/file\n");
+}
+
+int t1(int argc, char *argv[])
{
int fd;
int mount_with_flock = 0;
int error = 0;
- if (argc != 4)
- usage();
-
- if (!strncmp(argv[1], "on", 3)) {
+ if (argc != 5) {
+ t1_usage();
+ return EXIT_FAILURE;
+ }
+
+ if (!strncmp(argv[2], "on", 3)) {
mount_with_flock = 1;
- } else if (!strncmp(argv[1], "off", 4)) {
+ } else if (!strncmp(argv[2], "off", 4)) {
mount_with_flock = 0;
} else {
- usage();
+ t1_usage();
+ return EXIT_FAILURE;
}
- if ((fd = open(argv[3], O_RDWR)) < 0) {
- fprintf(stderr, "Couldn't open file: %s\n", argv[2]);
- exit(EXIT_FAILURE);
+ if ((fd = open(argv[4], O_RDWR)) < 0) {
+ fprintf(stderr, "Couldn't open file: %s\n", argv[3]);
+ return EXIT_FAILURE;
}
- if (!strncmp(argv[2], "-c", 3)) {
+ if (!strncmp(argv[3], "-c", 3)) {
struct flock fl;
fl.l_type = F_RDLCK;
@@ -80,12 +160,13 @@ int main(int argc, char *argv[])
fl.l_len = 1;
error = fcntl(fd, F_SETLK, &fl);
- } else if (!strncmp(argv[2], "-l", 3)) {
+ } else if (!strncmp(argv[3], "-l", 3)) {
error = lockf(fd, F_LOCK, 1);
- } else if (!strncmp(argv[2], "-f", 3)) {
+ } else if (!strncmp(argv[3], "-f", 3)) {
error = flock(fd, LOCK_EX);
} else {
- usage();
+ t1_usage();
+ return EXIT_FAILURE;
}
if (mount_with_flock)
@@ -93,3 +174,128 @@ int main(int argc, char *argv[])
else
return((error == 0) ? EXIT_FAILURE : EXIT_SUCCESS);
}
+
+/** ===============================================================
+ * test number 2
+ *
+ * 2 threads flock ops interweave
+ */
+typedef struct {
+ struct flock* lock;
+ int fd;
+} th_data;
+
+void* t2_thread1(void *arg)
+{
+ struct flock *lock = ((th_data *)arg)->lock;
+ int fd = ((th_data *)arg)->fd;
+
+ printf("thread 1: set write lock (blocking)\n");
+ lock->l_type = F_WRLCK;
+ t_fcntl(fd, F_SETLKW, lock);
+ printf("thread 1: set write lock done\n");
+ t_fcntl(fd, F_GETLK, lock);
+ printf("thread 1: unlock\n");
+ lock->l_type = F_UNLCK;
+ t_fcntl(fd, F_SETLK, lock);
+ printf("thread 1: unlock done\n");
+ return 0;
+}
+
+void* t2_thread2(void *arg)
+{
+ struct flock *lock = ((th_data *)arg)->lock;
+ int fd = ((th_data *)arg)->fd;
+
+ sleep(2);
+ printf("thread 2: unlock\n");
+ lock->l_type = F_UNLCK;
+ t_fcntl(fd, F_SETLK, lock);
+ printf("thread 2: unlock done\n");
+ printf("thread 2: set write lock (non-blocking)\n");
+ lock->l_type = F_WRLCK;
+ t_fcntl(fd, F_SETLK, lock);
+ printf("thread 2: set write lock done\n");
+ t_fcntl(fd, F_GETLK, lock);
+ return 0;
+}
+
+int t2(int argc, char* argv[])
+{
+ struct flock lock = {
+ .l_type = F_RDLCK,
+ .l_whence = SEEK_SET,
+ };
+ char file[MAX_PATH_LENGTH] = "";
+ int fd, rc;
+ pthread_t th1, th2;
+ th_data ta;
+
+ snprintf(file, MAX_PATH_LENGTH, "%s/test_t2_file", argv[2]);
+
+ fd = open(file, O_RDWR|O_CREAT, (mode_t)0666);
+ if (fd < 0) {
+ fprintf(stderr, "error open file: %s\n", file);
+ return EXIT_FAILURE;
+ }
+
+ t_fcntl(fd, F_SETFL, O_APPEND);
+ if (!(rc = t_fcntl(fd, F_GETFL)) & O_APPEND) {
+ fprintf(stderr, "error get flag: ret %x\n", rc);
+ return EXIT_FAILURE;
+ }
+
+ ta.lock = &lock;
+ ta.fd = fd;
+ rc = pthread_create(&th1, NULL, t2_thread1, &ta);
+ if (rc) {
+ fprintf(stderr, "error create thread 1\n");
+ rc = EXIT_FAILURE;
+ goto out;
+ }
+ rc = pthread_create(&th2, NULL, t2_thread2, &ta);
+ if (rc) {
+ fprintf(stderr, "error create thread 2\n");
+ rc = EXIT_FAILURE;
+ goto out;
+ }
+ (void)pthread_join(th1, NULL);
+ (void)pthread_join(th2, NULL);
+out:
+ t_unlink(file);
+ close(fd);
+ return rc;
+}
+
+/** ==============================================================
+ * program entry
+ */
+void usage(void)
+{
+ fprintf(stderr, "usage: ./flocks_test test# [corresponding arguments]\n");
+}
+
+int main(int argc, char* argv[])
+{
+ int test_no;
+ int rc = EXIT_SUCCESS;
+
+ if (argc < 1) {
+ usage();
+ exit(EXIT_FAILURE);
+ }
+ test_no = atoi(argv[1]);
+
+ switch(test_no) {
+ case 1:
+ rc = t1(argc, argv);
+ break;
+ case 2:
+ rc = t2(argc, argv);
+ break;
+ default:
+ fprintf(stderr, "unknow test number %s\n", argv[1]);
+ break;
+ }
+ return rc;
+}
diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 40d4b6f3e5bc39156e92f48face716f4177bcd5f..a4c96567c4c5bc0ca801613f4a52979ea683a484 100644
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -4093,9 +4093,9 @@ test_105a() {
touch $DIR/$tfile
if [ -n "`mount | grep \"$DIR.*flock\" | grep -v noflock`" ];
then
- flocks_test on -f $DIR/$tfile || error "fail flock on"
+ flocks_test 1 on -f $DIR/$tfile || error "fail flock on"
else
- flocks_test off -f $DIR/$tfile || error "fail flock off"
+ flocks_test 1 off -f $DIR/$tfile || error "fail flock off"
fi
}
run_test 105a "flock when mounted without -o flock test ========"
@@ -4104,9 +4104,9 @@ test_105b() {
touch $DIR/$tfile
if [ -n "`mount | grep \"$DIR.*flock\" | grep -v noflock`" ];
then
- flocks_test on -c $DIR/$tfile || error "fail flock on"
+ flocks_test 1 on -c $DIR/$tfile || error "fail flock on"
else
- flocks_test off -c $DIR/$tfile || error "fail flock off"
+ flocks_test 1 off -c $DIR/$tfile || error "fail flock off"
fi
}
run_test 105b "fcntl when mounted without -o flock test ========"
@@ -4115,13 +4115,23 @@ test_105c() {
touch $DIR/$tfile
if [ -n "`mount | grep \"$DIR.*flock\" | grep -v noflock`" ];
then
- flocks_test on -l $DIR/$tfile || error "fail flock on"
+ flocks_test 1 on -l $DIR/$tfile || error "fail flock on"
else
- flocks_test off -l $DIR/$tfile || error "fail flock off"
+ flocks_test 1 off -l $DIR/$tfile || error "fail flock off"
fi
}
run_test 105c "lockf when mounted without -o flock test ========"
+test_105d() { # bug 15924
+ mkdir -p $DIR/$tdir
+ [ -z "`mount | grep \"$DIR.*flock\" | grep -v noflock`" ] && \
+ skip "mount w/o flock enabled" && return
+ #define OBD_FAIL_LDLM_CP_CB_WAIT 0x315
+ $LCTL set_param fail_loc=0x80000315
+ flocks_test 2 $DIR/$tdir
+}
+run_test 105d "flock race (should not freeze) ========"
+
test_106() { #bug 10921
mkdir -p $DIR/$tdir
$DIR/$tdir && error "exec $DIR/$tdir succeeded"