| Message ID | 20250401015557.113872-1-lufei@uniontech.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] unshare_test: set nr_open using soft limit | expand |
On 3/31/25 19:55, lufei wrote: > Set maximum file descriptor number limit by rlimit.rlim_max than > nr_open(hard limit). Hard limit may cause dup2 fail. > > Signed-off-by: lufei <lufei@uniontech.com> Petr, Al, Okay to take this patch? thanks, -- Shuah
> On 3/31/25 19:55, lufei wrote: > > Set maximum file descriptor number limit by rlimit.rlim_max than > > nr_open(hard limit). Hard limit may cause dup2 fail. > > Signed-off-by: lufei <lufei@uniontech.com> > Petr, Al, > Okay to take this patch? LGTM, hopefully I haven't overlook anything. Reviewed-by: Petr Vorel <pvorel@suse.cz> FYI Cyril Hrubis suggested a different approach (for LTP tests which is based on this Al's test): https://lore.kernel.org/ltp/Z-u7yYJpTBG8Hi6A@yuki.lan/ https://lore.kernel.org/ltp/Z-vwYZxLms8juTjX@yuki.lan/ filedescriptor that is >= 64 and set the nr_open to 64. That would help to avoid using /proc/sys/fs/nr_open. Kind regards, Petr > thanks, > -- Shuah
On 4/8/25 23:56, Petr Vorel wrote: >> On 3/31/25 19:55, lufei wrote: >>> Set maximum file descriptor number limit by rlimit.rlim_max than >>> nr_open(hard limit). Hard limit may cause dup2 fail. > >>> Signed-off-by: lufei <lufei@uniontech.com> > >> Petr, Al, > >> Okay to take this patch? > > LGTM, hopefully I haven't overlook anything. > > Reviewed-by: Petr Vorel <pvorel@suse.cz> > > FYI Cyril Hrubis suggested a different approach (for LTP tests which is based on > this Al's test): > > https://lore.kernel.org/ltp/Z-u7yYJpTBG8Hi6A@yuki.lan/ > https://lore.kernel.org/ltp/Z-vwYZxLms8juTjX@yuki.lan/ > > filedescriptor that is >= 64 and set the nr_open to 64. > > That would help to avoid using /proc/sys/fs/nr_open. > Thank you for this suggestion. lufei, Can you look into this as a solution? Also the short log should include the subsystem: e.: selftests:core: unshare_test: ----- thanks, -- Shuah
On Wed, Apr 09, 2025 at 03:30:32PM -0600, Shuah Khan wrote: > On 4/8/25 23:56, Petr Vorel wrote: > > > On 3/31/25 19:55, lufei wrote: > > > > Set maximum file descriptor number limit by rlimit.rlim_max than > > > > nr_open(hard limit). Hard limit may cause dup2 fail. > > > > > > Signed-off-by: lufei <lufei@uniontech.com> > > > > > Petr, Al, > > > > > Okay to take this patch? > > > > LGTM, hopefully I haven't overlook anything. > > > > Reviewed-by: Petr Vorel <pvorel@suse.cz> > > > > FYI Cyril Hrubis suggested a different approach (for LTP tests which is based on > > this Al's test): > > > > https://lore.kernel.org/ltp/Z-u7yYJpTBG8Hi6A@yuki.lan/ > > https://lore.kernel.org/ltp/Z-vwYZxLms8juTjX@yuki.lan/ > > > > filedescriptor that is >= 64 and set the nr_open to 64. > > > > That would help to avoid using /proc/sys/fs/nr_open. > > > > Thank you for this suggestion. > > lufei, Can you look into this as a solution? Also the short log > should include the subsystem: e.: > > selftests:core: unshare_test: ----- > > thanks, > -- Shuah > Hi Shuah, Thanks very much for reply. I'll submit a v3 patch with dup a filedescriptor >= sizeof(long)*8 and set nr_open to sizeof(long)*8 to trigger EMFILE. Hi Petr, Shuah, I don't have a i686 environment, only on x86_64 has been tested. Is this all right? I'll re-write short log as asked in v3 patch.
diff --git a/tools/testing/selftests/core/unshare_test.c b/tools/testing/selftests/core/unshare_test.c index 7fec9dfb1b0e..66b651168887 100644 --- a/tools/testing/selftests/core/unshare_test.c +++ b/tools/testing/selftests/core/unshare_test.c @@ -30,6 +30,7 @@ TEST(unshare_EMFILE) static char buf[512], buf2[512]; struct rlimit rlimit; int nr_open; + int rlimit_max; fd = open("/proc/sys/fs/nr_open", O_RDWR); ASSERT_GE(fd, 0); @@ -42,22 +43,24 @@ TEST(unshare_EMFILE) ASSERT_EQ(0, getrlimit(RLIMIT_NOFILE, &rlimit)); - /* bump fs.nr_open */ - n2 = sprintf(buf2, "%d\n", nr_open + 1024); + rlimit_max = rlimit.rlim_max; + + /* bump rlimit.rlim_max */ + n2 = sprintf(buf2, "%d\n", rlimit_max + 1024); lseek(fd, 0, SEEK_SET); write(fd, buf2, n2); /* bump ulimit -n */ - rlimit.rlim_cur = nr_open + 1024; - rlimit.rlim_max = nr_open + 1024; + rlimit.rlim_cur = rlimit_max + 1024; + rlimit.rlim_max = rlimit_max + 1024; EXPECT_EQ(0, setrlimit(RLIMIT_NOFILE, &rlimit)) { lseek(fd, 0, SEEK_SET); write(fd, buf, n); exit(EXIT_FAILURE); } - /* get a descriptor past the old fs.nr_open */ - EXPECT_GE(dup2(2, nr_open + 64), 0) { + /* get a descriptor past the old rlimit.rlim_max */ + EXPECT_GE(dup2(2, rlimit_max + 64), 0) { lseek(fd, 0, SEEK_SET); write(fd, buf, n); exit(EXIT_FAILURE); @@ -74,15 +77,20 @@ TEST(unshare_EMFILE) if (pid == 0) { int err; - /* restore fs.nr_open */ + n2 = sprintf(buf2, "%d\n", rlimit_max); lseek(fd, 0, SEEK_SET); - write(fd, buf, n); + write(fd, buf2, n2); + /* ... and now unshare(CLONE_FILES) must fail with EMFILE */ err = unshare(CLONE_FILES); EXPECT_EQ(err, -1) exit(EXIT_FAILURE); EXPECT_EQ(errno, EMFILE) exit(EXIT_FAILURE); + + /* restore fs.nr_open */ + lseek(fd, 0, SEEK_SET); + write(fd, buf, n); exit(EXIT_SUCCESS); }
Set maximum file descriptor number limit by rlimit.rlim_max than nr_open(hard limit). Hard limit may cause dup2 fail. Signed-off-by: lufei <lufei@uniontech.com> --- tools/testing/selftests/core/unshare_test.c | 24 ++++++++++++++------- 1 file changed, 16 insertions(+), 8 deletions(-)