From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Stancek Date: Wed, 11 Oct 2017 04:42:47 -0400 (EDT) Subject: [LTP] [PATCH v2 04/15] lib/tst_test: Add .all_filesystems flag In-Reply-To: <20171010123205.20147-4-chrubis@suse.cz> References: <20171010123205.20147-1-chrubis@suse.cz> <20171010123205.20147-4-chrubis@suse.cz> Message-ID: <1573092724.28358902.1507711367946.JavaMail.zimbra@redhat.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ltp@lists.linux.it ----- Original Message ----- > This commit adds a support for re-runing a test on all available > filesystems simply by turning on .all_filesystems flag in the tst_test > structure. > > The .all_filesystems flag implies .needs_device, the .format_filesystem and > .mount_filesystem works as usuall but the device is formatted and mounted for > each filesystem before we run the test function. The .setup and .cleanup > functions are executed for each filesystem. The currently tested filesystem > type is stored in the tst_device->fs_type. > > This allows us to easily run a test that tests filesystem specific > syscall for all supported filesystems without a need to hardcode all fs > types into a runtest file and with a minimal changes to the test code. > > Signed-off-by: Cyril Hrubis Hi, Overall series looks good to me. My tests on RHEL 5.6/6.0/7.4 and 4.14-rc kernels went all without issues. Couple notes below. 1) When this is merged, we should also add some notes to docs. For example: timeout applies for one filesystem, not all combined. 2) > +static int run_tcases_per_fs(void) > +{ > + int ret = 0; > + unsigned int i; > + const char *const *filesystems = tst_get_supported_fs_types(); > + > + if (!filesystems[0]) > + tst_brk(TCONF, "There are no supported filesystems"); > + > + for (i = 0; filesystems[i]; i++) { > + tdev.fs_type = filesystems[i]; > + > + prepare_device(); > + > + ret = fork_testrun(); > + > + if (mntpoint_mounted) { > + tst_umount(tst_test->mntpoint); > + mntpoint_mounted = 0; > + } > + > + if (ret == TCONF) { > + update_results(ret); > + continue; > + } > + > + if (ret == 0) > + continue; > + > + do_exit(ret); > + } I'm on the fence here, if we should stop immediately after we get a TFAIL or TBROK. For example: Is a TFAIL on ext2 reason to stop testing xfs? 3) fs_fill test is still quite verbose # ./fs_fill > log2 2>&1 # du -h log2 688K log2 # grep Got log2 fs_fill.c:96: PASS: Got 5713 ENOSPC fs_fill.c:96: PASS: Got 9 ENOSPC fs_fill.c:96: PASS: Got 1709 ENOSPC I'm running this on single CPU KVM guest, which is (presumably due to caching and host SSD) able to trigger ENOSPC many times in those 2 seconds. I expect baremetal systems with many CPUs producing lot of output as well, because we start tst_fill_fs() in NR_CPU+2 threads. Regards, Jan