Demonstrate a bug we have when dealing with real os-level share modes

Another one of those where you stare at logfiles for hours, and when you found
it, it's absolutely obvious what is happening...

diff --git a/source3/torture/torture.c b/source3/torture/torture.c
index 2c68ee609a4cfaa450292f74c6ca730362453629..162ba2bc82323d128bbaef904187dcfe96e83059 100644 (file)
--- a/source3/torture/torture.c
+++ b/source3/torture/torture.c
@@ -2083,6 +2083,80 @@ fail:
        return correct;
 }
 
+/*
+ * This demonstrates a problem with our use of GPFS share modes: A file
+ * descriptor sitting in the pending close queue holding a GPFS share mode
+ * blocks opening a file another time. Happens with Word 2007 temp files.
+ * With "posix locking = yes" and "gpfs:sharemodes = yes" enabled, the third
+ * open is denied with NT_STATUS_SHARING_VIOLATION.
+ */
+
+static bool run_locktest8(int dummy)
+{
+       struct cli_state *cli1;
+       const char *fname = "\\lockt8.lck";
+       uint16_t fnum1, fnum2;
+       char buf[200];
+       bool correct = False;
+       NTSTATUS status;
+
+       if (!torture_open_connection(&cli1, 0)) {
+               return False;
+       }
+
+       cli_sockopt(cli1, sockops);
+
+       printf("starting locktest8\n");
+
+       cli_unlink(cli1, fname, aSYSTEM | aHIDDEN);
+
+       status = cli_open(cli1, fname, O_RDWR|O_CREAT|O_EXCL, DENY_WRITE,
+                         &fnum1);
+       if (!NT_STATUS_IS_OK(status)) {
+               d_fprintf(stderr, "cli_open returned %s\n", cli_errstr(cli1));
+               return false;
+       }
+
+       memset(buf, 0, sizeof(buf));
+
+       status = cli_open(cli1, fname, O_RDONLY, DENY_NONE, &fnum2);
+       if (!NT_STATUS_IS_OK(status)) {
+               d_fprintf(stderr, "cli_open second time returned %s\n",
+                         cli_errstr(cli1));
+               goto fail;
+       }
+
+       if (!cli_lock(cli1, fnum2, 1, 1, 0, READ_LOCK)) {
+               printf("Unable to apply read lock on range 1:1, error was "
+                      "%s\n", cli_errstr(cli1));
+               goto fail;
+       }
+
+       status = cli_close(cli1, fnum1);
+       if (!NT_STATUS_IS_OK(status)) {
+               d_fprintf(stderr, "cli_close(fnum1) %s\n", cli_errstr(cli1));
+               goto fail;
+       }
+
+       status = cli_open(cli1, fname, O_RDWR, DENY_NONE, &fnum1);
+       if (!NT_STATUS_IS_OK(status)) {
+               d_fprintf(stderr, "cli_open third time returned %s\n",
+                          cli_errstr(cli1));
+                goto fail;
+        }
+
+       correct = true;
+
+fail:
+       cli_close(cli1, fnum1);
+       cli_close(cli1, fnum2);
+       cli_unlink(cli1, fname, aSYSTEM | aHIDDEN);
+       torture_close_connection(cli1);
+
+       printf("finished locktest8\n");
+       return correct;
+}
+
 /*
 test whether fnums and tids open on one VC are available on another (a major
 security hole)
@@ -6009,6 +6083,7 @@ static struct {
        {"LOCK5",  run_locktest5,  0},
        {"LOCK6",  run_locktest6,  0},
        {"LOCK7",  run_locktest7,  0},
+       {"LOCK8",  run_locktest8,  0},
        {"UNLINK", run_unlinktest, 0},
        {"BROWSE", run_browsetest, 0},
        {"ATTR",   run_attrtest,   0},