| Message ID | 20190307233154.12873-1-saravanak@google.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v2] setpriority01: Skip only PRIO_USER when unable to add test user | expand |
Hi Saravana, > We don't need to skip all the tests just because we are unable to add > a test user. Not having a test user only affects PRIO_USER test case. > So just skip that one and continue running the rest of the tests when > useradd is not available. > If useradd is present and it still fails, then consider it an error in > the test preparation phase. > This also allows this test case to be built and run on Android. > Signed-off-by: Saravana Kannan <saravanak@google.com> Reviewed-by: Petr Vorel <pvorel@suse.cz> ... > v2: > - Skip only PRIO_USER test when useradd isn't present > - Fail entire test with TBROK for all other cases where user can't be added > I still think v1 is the right approach. But if you want TCONF for only > for cases like Android where useradd itself isn't present, then v2 would > be the right patch. Agree with it, but waiting for Cyril approval to merge it. Kind regards, Petr
Hi! > > v2: > > - Skip only PRIO_USER test when useradd isn't present > > - Fail entire test with TBROK for all other cases where user can't be added > > > I still think v1 is the right approach. But if you want TCONF for only > > for cases like Android where useradd itself isn't present, then v2 would > > be the right patch. > Agree with it, but waiting for Cyril approval to merge it. As far as I can tell this patch still removes the read-only rootfs check introduced in: commit 8cab24427ec4584389cd0462eb87bd9d43261861 Author: the_hoang0709@yahoo.com <the_hoang0709@yahoo.com> Date: Sun Apr 9 15:38:05 2017 +0700 setpriority01: Fix failure on read-only machine Return 32 if /etc/passwd is read-only. Fixes: #137 See also: #88
Hi Cyril, > > > I still think v1 is the right approach. But if you want TCONF for only > > > for cases like Android where useradd itself isn't present, then v2 would > > > be the right patch. > > Agree with it, but waiting for Cyril approval to merge it. > As far as I can tell this patch still removes the read-only rootfs check > introduced in: Good catch, eaccess check should stay in. I'll put it back again and push, ok? Kind regards, Petr > commit 8cab24427ec4584389cd0462eb87bd9d43261861 > Author: the_hoang0709@yahoo.com <the_hoang0709@yahoo.com> > Date: Sun Apr 9 15:38:05 2017 +0700 > setpriority01: Fix failure on read-only machine > Return 32 if /etc/passwd is read-only. > Fixes: #137 > See also: #88
Hi! > > > > I still think v1 is the right approach. But if you want TCONF for only > > > > for cases like Android where useradd itself isn't present, then v2 would > > > > be the right patch. > > > Agree with it, but waiting for Cyril approval to merge it. > > > As far as I can tell this patch still removes the read-only rootfs check > > introduced in: > Good catch, eaccess check should stay in. > I'll put it back again and push, ok? Or we can check the return value 1 from useradd and map it to TCONF as suggested by Saravana.
Hi, > > > > > I still think v1 is the right approach. But if you want TCONF for only > > > > > for cases like Android where useradd itself isn't present, then v2 would > > > > > be the right patch. > > > > Agree with it, but waiting for Cyril approval to merge it. > > > As far as I can tell this patch still removes the read-only rootfs check > > > introduced in: > > Good catch, eaccess check should stay in. > > I'll put it back again and push, ok? > Or we can check the return value 1 from useradd and map it to TCONF as > suggested by Saravana. Probably. According to useradd sources, exit 1 (E_PW_UPDATE) also happen when problems to lock /etc/passwd, which could theoretically on different circumstance than read-only /, but I guess we can ignore it. Or we could use statvfs to detect read-only / ("can't update password file" does can be for different purposes), but I wouldn't bother and user return value 1. Kind regards, Petr [1] https://github.com/shadow-maint/shadow/blob/master/src/useradd.c#L1626
The problem with using eaccess is that it doesn't build on Android as eaccess isn't implemented. I vote for mapping useradd exit code 1 to TCONF. Sure there are some very unlikely cases where useradd might fail with 1 when the file system is writable. Excluding that corner case if it means being able to run on a lot of Android devices seems like a good compromise. -Saravana On Tue, Mar 19, 2019 at 4:28 AM Petr Vorel <pvorel@suse.cz> wrote: > Hi, > > > > > > > I still think v1 is the right approach. But if you want TCONF > for only > > > > > > for cases like Android where useradd itself isn't present, then > v2 would > > > > > > be the right patch. > > > > > Agree with it, but waiting for Cyril approval to merge it. > > > > > As far as I can tell this patch still removes the read-only rootfs > check > > > > introduced in: > > > Good catch, eaccess check should stay in. > > > I'll put it back again and push, ok? > > > Or we can check the return value 1 from useradd and map it to TCONF as > > suggested by Saravana. > Probably. According to useradd sources, exit 1 (E_PW_UPDATE) also happen > when > problems to lock /etc/passwd, which could theoretically on different > circumstance than read-only /, but I guess we can ignore it. > > Or we could use statvfs to detect read-only / ("can't update password > file" does > can be for different purposes), but I wouldn't bother and user return > value 1. > > Kind regards, > Petr > > [1] https://github.com/shadow-maint/shadow/blob/master/src/useradd.c#L1626 > > -- > You received this message because you are subscribed to the Google Groups > "kernel-team" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to kernel-team+unsubscribe@android.com. > > <div dir="ltr">The problem with using eaccess is that it doesn't build on Android as eaccess isn't implemented.<div><br></div><div>I vote for mapping useradd exit code 1 to TCONF. Sure there are some very unlikely cases where useradd might fail with 1 when the file system is writable. Excluding that corner case if it means being able to run on a lot of Android devices seems like a good compromise.</div><div><br></div><div>-Saravana</div><div><br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Tue, Mar 19, 2019 at 4:28 AM Petr Vorel <<a href="mailto:pvorel@suse.cz">pvorel@suse.cz</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi,<br> <br> > > > > > I still think v1 is the right approach. But if you want TCONF for only<br> > > > > > for cases like Android where useradd itself isn't present, then v2 would<br> > > > > > be the right patch.<br> > > > > Agree with it, but waiting for Cyril approval to merge it.<br> <br> > > > As far as I can tell this patch still removes the read-only rootfs check<br> > > > introduced in:<br> > > Good catch, eaccess check should stay in.<br> > > I'll put it back again and push, ok?<br> <br> > Or we can check the return value 1 from useradd and map it to TCONF as<br> > suggested by Saravana.<br> Probably. According to useradd sources, exit 1 (E_PW_UPDATE) also happen when<br> problems to lock /etc/passwd, which could theoretically on different<br> circumstance than read-only /, but I guess we can ignore it.<br> <br> Or we could use statvfs to detect read-only / ("can't update password file" does<br> can be for different purposes), but I wouldn't bother and user return value 1.<br> <br> Kind regards,<br> Petr<br> <br> [1] <a href="https://github.com/shadow-maint/shadow/blob/master/src/useradd.c#L1626" rel="noreferrer" target="_blank">https://github.com/shadow-maint/shadow/blob/master/src/useradd.c#L1626</a><br> <br> -- <br> You received this message because you are subscribed to the Google Groups "kernel-team" group.<br> To unsubscribe from this group and stop receiving emails from it, send an email to <a href="mailto:kernel-team%2Bunsubscribe@android.com" target="_blank">kernel-team+unsubscribe@android.com</a>.<br> <br> </blockquote></div>
diff --git a/testcases/kernel/syscalls/setpriority/Makefile b/testcases/kernel/syscalls/setpriority/Makefile index 5d00984ea..7a1a87a28 100644 --- a/testcases/kernel/syscalls/setpriority/Makefile +++ b/testcases/kernel/syscalls/setpriority/Makefile @@ -19,9 +19,4 @@ top_srcdir ?= ../../../.. include $(top_srcdir)/include/mk/testcases.mk - -ifeq ($(ANDROID), 1) -FILTER_OUT_MAKE_TARGETS += setpriority01 -endif - include $(top_srcdir)/include/mk/generic_leaf_target.mk diff --git a/testcases/kernel/syscalls/setpriority/setpriority01.c b/testcases/kernel/syscalls/setpriority/setpriority01.c index 38b77b77f..92b1b4ef4 100644 --- a/testcases/kernel/syscalls/setpriority/setpriority01.c +++ b/testcases/kernel/syscalls/setpriority/setpriority01.c @@ -92,9 +92,16 @@ static void verify_setpriority(unsigned int n) { struct tcase *tc = &tcases[n]; + if (tc->which == PRIO_USER && !user_added) { + tst_res(TCONF, "setpriority(%s(%d), %d, -20..19) skipped - Can't add user", + str_which(tc->which), tc->which, *tc->who); + return; + } + pid = SAFE_FORK(); if (pid == 0) { - SAFE_SETUID(uid); + if (user_added) + SAFE_SETUID(uid); SAFE_SETPGID(0, 0); TST_CHECKPOINT_WAKE_AND_WAIT(0); @@ -115,15 +122,19 @@ static void setup(void) { const char *const cmd_useradd[] = {"useradd", username, NULL}; struct passwd *ltpuser; + int rc; - if (eaccess("/etc/passwd", W_OK)) - tst_brk(TCONF, "/etc/passwd is not accessible"); - - tst_run_cmd(cmd_useradd, NULL, NULL, 0); - user_added = 1; - - ltpuser = SAFE_GETPWNAM(username); - uid = ltpuser->pw_uid; + switch ((rc = tst_run_cmd(cmd_useradd, NULL, NULL, 1))) { + case 0: + user_added = 1; + ltpuser = SAFE_GETPWNAM(username); + uid = ltpuser->pw_uid; + return; + case 255: + return; + default: + tst_brk(TBROK, "Useradd failed (%d)", rc); + } } static void cleanup(void)
We don't need to skip all the tests just because we are unable to add a test user. Not having a test user only affects PRIO_USER test case. So just skip that one and continue running the rest of the tests when useradd is not available. If useradd is present and it still fails, then consider it an error in the test preparation phase. This also allows this test case to be built and run on Android. Signed-off-by: Saravana Kannan <saravanak@google.com> --- v1: - Skip only PRIO_USER test when unable to add user for any reason. v2: - Skip only PRIO_USER test when useradd isn't present - Fail entire test with TBROK for all other cases where user can't be added I still think v1 is the right approach. But if you want TCONF for only for cases like Android where useradd itself isn't present, then v2 would be the right patch. .../kernel/syscalls/setpriority/Makefile | 5 ---- .../syscalls/setpriority/setpriority01.c | 29 +++++++++++++------ 2 files changed, 20 insertions(+), 14 deletions(-)