From: Cyril Hrubis Date: Tue, 22 Jun 2021 16:14:13 +0200 Subject: [LTP] [PATCH v4 3/3] API: Remove TST_ERR usage from rtnetlink/netdevice In-Reply-To: <3dfda1bd-3829-7188-6add-e1755e91df4f@suse.cz> References: <20210615074045.23974-1-rpalethorpe@suse.com> <20210615074045.23974-3-rpalethorpe@suse.com> <87czsk51fc.fsf@suse.de> <1e7ecce7-2e46-ea46-3c5b-357d53a238c3@suse.cz> <3dfda1bd-3829-7188-6add-e1755e91df4f@suse.cz> Message-ID: List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ltp@lists.linux.it Hi! > >>> The test author is guaranteed that the library will not set TST_ERR > >>> except via the TEST macro and similar. > >> > >> Hi, who decided to guarantee this and where is the guarantee documented? > > Guaranteeing that TST_RET and TST_ERR will not be modified makes sense > for SAFE_SYSCALL()s because they will be used extensively throughout > test code. But the case is not so clear for primarily setup()/cleanup() > functions like the af_alg, rtnetlink and netdevice libraries. And note > that the rtnetlink library allows you to check ACKs manually without > calling the two functions which will modify TST_ERR. > > So again, where is the extent of this guarantee documented? I haven't > found any mention of it in the doc/ dir and Richie didn't add any docs > changes in his patchset either. Documenting this is necessary for both > test writers and library maintainers. I guess that it should be added to doc/library-api-writing-guidelines.txt but that is orthogonal to the attempt to fix the the library code itself. > >> Changing the return value to always return errno will be a major PITA > >> because 95% of error handling happens after some safe_syscall() which > >> clobbers errno by calling tst_brk() after failure. And if you only want > >> to return error codes from rtnetlink ACK messages, then there's the > >> problem what to return when a safe_syscall() fails during cleanup(). > > > > For the current code it would be as easy as: > > That code will only result in RTNL_SEND_VALIDATE() always returning 0 > regardless of success or failure, except when tst_brk() terminates the > whole program. Ah right, we have to return non-zero if the tst_rtnl_send() or tst_rtnl_rect() failed. I would be nice to propagete the errno properly, which we do not do at the moment, but that looks like a lot of effort so we may as well hardcode something non-zero there as well. -- Cyril Hrubis chrubis@suse.cz