On Sun, Aug 1, 2010 at 9:36 PM, Mitani <[email protected]> wrote: >> -----Original Message----- >> From: Garrett Cooper [mailto:[email protected]] >> Sent: Saturday, July 17, 2010 7:51 AM >> To: Mitani >> Cc: [email protected] >> Subject: Re: [LTP] ”Ｃｏｎｔａｉｎｅｒｓ” test was "TFAIL" but "PASS" >> is in log results >> >> On Fri, Jul 16, 2010 at 3:50 PM, Garrett Cooper <[email protected]> >> wrote: >> > On Thu, Jul 15, 2010 at 9:19 PM, Mitani <[email protected]> wrote: >> >> Hi, >> >> >> >> With RHEL5.4/5.5 systems, I noticed that "Containers" test failed >> but >> >> "Containers" is PASS in log results: >> >> >> >> ---<log results file>------------- >> >> ... >> >> hangup01 PASS 0 >> >> Containers PASS 0 >> >> BindMounts PASS 0 >> >> ... >> >> ---------------- >> >> >> >> ---<detail in redirected file>------------- >> >> ... >> >> check_netns_enabled 1 TFAIL : unshare syscall smoke test >> failed: >> >> errno=EINVAL(22): Invalid argument >> >> Network namespaces not enabled in kernel. Not running netns tests. >> >> ... >> >> ---------------- >> >> >> >> The source which returned with "Invalid argument" is >> >> "check_netns_enabled.c": >> >> >> >> ---<check_netns_enabled.c>------------- >> >> main() >> >> { >> >> /* Checking if the kernel supports unshare with netns >> capabilities. >> >> */ >> >> if (syscall(__NR_unshare, CLONE_NEWNET | CLONE_NEWNS) < 0) >> { >> >> tst_resm (TFAIL | TERRNO, "unshare syscall smoke >> test >> >> failed"); >> >> } >> >> tst_exit(); >> >> } >> >> ---------------- >> >> >> >> In RHEL5.4/5.5 systems, unshare() does not support the argument >> >> "CLONE_NEWNET". >> >> But I feel strange that log results is "PASS" in spite of "TFAIL" >> exit. >> >> >> >> "container_test.sh" which calls "check_netns_enabled.c" is as >> follows: >> >> ---<container_test.sh>------------- >> >> ... >> >> check_netns_enabled >> >> if [ $? -eq 0 ]; then >> >> echo "Running netns tests." >> >> runnetnstest.sh >> >> else >> >> echo "Network namespaces not enabled in kernel. Not >> running netns >> >> tests." >> >> fi >> >> ... >> >> ---------------- >> >> >> >> According to above source, "check_netns_enabled.c" is only the >> judgement >> >> whether "runnetnstest.sh" test can be called or not. >> >> So, I think that "TFAIL" is not suitable for the test impossibility >> >> message. >> >> "TCONF" is more suitable for the message than "TFAIL", I think. >> >> But, if "TCONF" is used instead of "TFAIL", "tst_exit" returns with >> 0. >> >> And the test shell "runnetnstest.sh" is called against intention. >> >> >> >> >> >> I thought up two ways of solutions. >> >> (1) Insert "exit(1)" just after "tst_resm()" of >> "check_netns_enabled.c" >> >> (2) Replace "TFAIL" with "TWARN", instead of "TCONF". >> >> >> >> But above solutions are both irregular measures, I think. >> >> >> >> If there are other good measures, please give me advice. >> >> >> >> >> >> In the following, I nominate a temporary patch for an example: >> >> >> >> Signed-off-by: Tomonori Mitani <[email protected]> >> >> >> >> ===(1)========= >> >> --- >> a/testcases/kernel/containers/netns/check_netns_enabled.c 2010- >> 07-12 >> >> 02:01:45.000000000 +0900 >> >> +++ >> b/testcases/kernel/containers/netns/check_netns_enabled.c 2010- >> 07-16 >> >> 11:58:14.000000000 +0900 >> >> @@ -43,7 +43,8 @@ >> >> { >> >> /* Checking if the kernel supports unshare with netns >> capabilities. >> >> */ >> >> if (syscall(__NR_unshare, CLONE_NEWNET | CLONE_NEWNS) < >> 0) { >> >> - tst_resm (TFAIL | TERRNO, "unshare syscall smoke >> test >> >> failed"); >> >> + tst_resm (TCONF | TERRNO, "unshare syscall smoke >> test >> >> failed"); >> >> + exit (1); >> >> } >> >> tst_exit(); >> >> } >> >> ============ >> >> >> >> ===(2)========= >> >> --- >> a/testcases/kernel/containers/netns/check_netns_enabled.c 2010- >> 07-12 >> >> 02:01:45.000000000 +0900 >> >> +++ >> b/testcases/kernel/containers/netns/check_netns_enabled.c 2010- >> 07-16 >> >> 11:56:47.000000000 +0900 >> >> @@ -43,7 +43,7 @@ >> >> { >> >> /* Checking if the kernel supports unshare with netns >> capabilities. >> >> */ >> >> if (syscall(__NR_unshare, CLONE_NEWNET | CLONE_NEWNS) < >> 0) { >> >> - tst_resm (TFAIL | TERRNO, "unshare syscall smoke >> test >> >> failed"); >> >> + tst_resm (TWARN | TERRNO, "unshare syscall smoke >> test >> >> failed"); >> >> } >> >> tst_exit(); >> >> } >> >> ============ >> > >> > Uh, that would introduce a more serious problem because the netns >> > tests are keyed on whether or not that script passes or fails. >> >> s/script/app/ >> >> > If you can provide a more stable means of detecting whether or not >> > CLONE_* is available with netns, and integrate that into the tests, >> > that would be a better idea. >> >> Thanks, >> -Garrett > > > Sorry for my late answer. > > I considered how to judge detecting whether or not CLONE_* is available. > However, I hit only about errno judgment. > > Signed-off-by: Tomonori Mitani <[email protected]> > > ============ > --- a/testcases/kernel/containers/netns/check_netns_enabled.c 2010-07-30 > 02:23:32.000000000 +0900 > +++ b/testcases/kernel/containers/netns/check_netns_enabled.c 2010-08-02 > 10:38:31.000000000 +0900 > @@ -42,8 +42,12 @@ > main() > { > /* Checking if the kernel supports unshare with netns capabilities. */ > - if (syscall(__NR_unshare, CLONE_NEWNET | CLONE_NEWNS) < 0) { > - tst_resm (TFAIL | TERRNO, "unshare syscall smoke test > failed"); > + if (syscall(__NR_unshare, CLONE_NEWNET | CLONE_NEWNS) < 0) { > + if (errno == EINVAL) { > + tst_resm (TWARN | TERRNO, "unshare syscall not > support CLONE_NEWNET or CLONE_NEWNS"); > + } else { > + tst_resm (TFAIL | TERRNO, "unshare syscall smoke test > failed"); > + } > } > tst_exit(); > }
There are only 3 errors allowed for unshare(2), according to the manpage: EPERM flags specified CLONE_NEWNS but the calling process was not privileged (did not have the CAP_SYS_ADMIN capability). ENOMEM Cannot allocate sufficient memory to copy parts of caller's context that need to be unshared. EINVAL An invalid bit was specified in flags. Given that TWARN will show up as an error, I suppose this is ok.. but it seems silly to single out only that case. Committed anyhow :P. -Garrett ------------------------------------------------------------------------------ The Palm PDK Hot Apps Program offers developers who use the Plug-In Development Kit to bring their C/C++ apps to Palm for a share of $1 Million in cash or HP Products. Visit us here for more details: http://p.sf.net/sfu/dev2dev-palm _______________________________________________ Ltp-list mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/ltp-list
