LU-1205 tests: sanityn test_18 sometimes takes long time to run

This is a live-lock problem where two processes are writing to the
same mmaped file via two nodes. To write a mmap region, both processes
will do:

  acquire cl_lock -> read page -> release cl_lock-> install page.

During the above steps, the page can be truncated after the lock is
released and then immediately cancelled by the other process, so
kernel has to do page fault again and never complete.

Lustre can't handle this case well so this test case is disabled.

Signed-off-by: Jinshan Xiong <jinshan.xiong@whamcloud.com>
Signed-off-by: Bob Glossman <bogl@whamcloud.com>
Change-Id: I0cbb00a1ca68715a0b97ce369a18c53fa8de19cb
Reviewed-on: http://review.whamcloud.com/2723
Tested-by: Hudson
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/tests/mmap_sanity.c b/lustre/tests/mmap_sanity.c
index 853c484..7ab03ca 100644 (file)
--- a/lustre/tests/mmap_sanity.c
+++ b/lustre/tests/mmap_sanity.c
@@ -55,9 +55,11 @@ char mmap_sanity[256];
 
 static void usage(void)
 {
-        printf("Usage: mmap_sanity -d dir [-m dir2]\n");
-        printf("       dir      lustre mount point\n");
-        printf("       dir2     another mount point\n");
+        printf("Usage: mmap_sanity -d dir [-m dir2] [-e <test cases>]\n");
+        printf("       -d dir        lustre mount point\n");
+        printf("       -m dir2       another mount point\n");
+        printf("       -e testcases  skipped test cases, -e 1 -e 2 to exclude"
+                                    " test cases 1 and 2.\n");
         exit(127);
 }
 
@@ -715,6 +717,7 @@ struct test_case {
         char    *desc;                  /* test description */
         int     (*test_fn)(char *mnt);  /* test function */
         int     node_cnt;               /* node count */
+        int     skipped;                /* skipped by caller */
 };
 
 struct test_case tests[] = {
@@ -733,36 +736,47 @@ struct test_case tests[] = {
 
 int main(int argc, char **argv)
 {
-        struct test_case *test;
-        int c, rc = 0;
-
-        while ((c = getopt(argc, argv, "d:m:")) != -1) {
-                switch (c) {
-                case 'd':
-                        dir = optarg;
-                        break;
-                case 'm':
-                        dir2 = optarg;
-                        break;
-                default:
-                        usage();
-                        break;
-                }
-        }
+       struct test_case *test;
+       int nr_cases = sizeof(tests)/sizeof(*test);
+       int c, rc = 0;
+
+       while ((c = getopt(argc, argv, "d:m:e:")) != -1) {
+               switch (c) {
+               case 'd':
+                       dir = optarg;
+                       break;
+               case 'm':
+                       dir2 = optarg;
+                       break;
+               case 'e': {
+                       char *endptr = NULL;
+                       rc = strtol(optarg, &endptr, 10);
+                       if (endptr != NULL && *endptr != '\0')
+                               usage();
+                       if (rc > 0 && rc < nr_cases)
+                               tests[rc - 1].skipped = 1;
+                       break;
+               }
+               default:
+                       usage();
+                       break;
+               }
+       }
 
-        if (dir == NULL)
-                usage();
+       if (dir == NULL)
+               usage();
 
         if (mmap_initialize(argv[0]) != 0) {
                 fprintf(stderr, "mmap_initialize failed!\n");
                 return -EINVAL;
         }
 
+        rc = 0;
         for (test = tests; test->tc; test++) {
-                double duration;
-                char *rs;
+                double duration = 0.0;
+                char *rs = "SKIPPED";
 
-                if (test->node_cnt == 1 || dir2 != NULL) {
+                if (!test->skipped && (test->node_cnt == 1 || dir2 != NULL)) {
                         struct timeval start, end;
 
                         gettimeofday(&start, NULL);
@@ -772,10 +786,8 @@ int main(int argc, char **argv)
                         duration = (double)(end.tv_sec - start.tv_sec) +
                                 (double)(end.tv_usec - start.tv_usec) / 1000000;
                         rs = rc ? "FAIL" : "PASS";
-                } else {
-                        duration = 0.0;
-                        rs = "SKIP";
                 }
+
                 fprintf(stderr, "%s (%s, %.5gs)\n", test->desc, rs, duration);
                 if (rc)
                         break;

diff --git a/lustre/tests/sanityn.sh b/lustre/tests/sanityn.sh
index a229c61..0d0ca1a 100644 (file)
--- a/lustre/tests/sanityn.sh
+++ b/lustre/tests/sanityn.sh
@@ -3,8 +3,8 @@
 set -e
 
 ONLY=${ONLY:-"$*"}
-# bug number for skipped test: 3192 15528/3811 16929 9977 15528/11549 18080
-ALWAYS_EXCEPT="                14b  19         22    28   29          35    $SANITYN_EXCEPT"
+# bug number for skipped test: 3192 LU-1205 15528/3811 16929 9977 15528/11549 18080
+ALWAYS_EXCEPT="                14b  18c     19         22    28   29          35    $SANITYN_EXCEPT"
 # UPDATE THE COMMENT ABOVE WITH BUG NUMBERS WHEN CHANGING ALWAYS_EXCEPT!
 
 # bug number for skipped test:                                                    12652 12652
@@ -350,7 +350,17 @@ test_17() { # bug 3513, 3667
 run_test 17 "resource creation/LVB creation race ==============="
 
 test_18() {
-       $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2
+        # turn e.g. ALWAYS_EXCEPT="18c" into "-e 3"
+        local idx
+        local excepts=
+        for idx in {a..z}; do
+                local ptr=EXCEPT_ALWAYS_18$idx
+                [ x${!ptr} = xtrue ] || continue
+
+                excepts="$excepts -e $(($(printf %d \'$idx)-96))"
+        done
+
+       $LUSTRE/tests/mmap_sanity -d $MOUNT1 -m $MOUNT2 $excepts
        sync; sleep 1; sync
 }
 run_test 18 "mmap sanity check ================================="