Message ID | 20220608130527.26e50cba@gandalf.local.home (mailing list archive) |
---|---|
State | Superseded |
Headers | show |
Series | trace-cmd record: Verify that splice works before using it | expand |
On Wed, 2022-06-08 at 13:05 -0400, Steven Rostedt wrote: > From: "Steven Rostedt (Google)" <rostedt@goodmis.org> > > Add a test to make sure that splice works on the source directory before > using it, and if not automatically switch over to read/write method. > > Suggested-by: Johannes Berg <johannes@sipsolutions.net> > Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=213659 > Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> > --- > Johannes, can you test this? > I want to make sure that it fixes the issue. > Umm, yeah, long time ago. Sorry, I've been meaning to do that, but it isn't easy, so I kept pushing it off. My use case for this was UML with hostfs, which I've since fixed (kernel upstream commit 1568cb0e6d97 ("hostfs: support splice_write"); I can revert that easily of course, but then I still have to build my own trace-cmd with that patch... But that needs libtracefs 1.6 which even Fedora doesn't ship. Given that you've merged the patch already, I think I'll just give up on testing it, sorry about that. johannes
On Thu, 19 Jan 2023 22:03:27 +0100 Johannes Berg <johannes@sipsolutions.net> wrote: > On Wed, 2022-06-08 at 13:05 -0400, Steven Rostedt wrote: > > From: "Steven Rostedt (Google)" <rostedt@goodmis.org> > > > > Add a test to make sure that splice works on the source directory before > > using it, and if not automatically switch over to read/write method. > > > > Suggested-by: Johannes Berg <johannes@sipsolutions.net> > > Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=213659 > > Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org> > > --- > > Johannes, can you test this? > > I want to make sure that it fixes the issue. > > > > Umm, yeah, long time ago. Sorry, I've been meaning to do that, but it > isn't easy, so I kept pushing it off. > > My use case for this was UML with hostfs, which I've since fixed (kernel > upstream commit 1568cb0e6d97 ("hostfs: support splice_write"); I can > revert that easily of course, but then I still have to build my own > trace-cmd with that patch... But that needs libtracefs 1.6 which even > Fedora doesn't ship. They should ;-) > > Given that you've merged the patch already, I think I'll just give up on > testing it, sorry about that. No worries. I'll just close the bug. Thanks, for responding. -- Steve
diff --git a/lib/trace-cmd/trace-recorder.c b/lib/trace-cmd/trace-recorder.c index c83337899098..c7ef13c851d2 100644 --- a/lib/trace-cmd/trace-recorder.c +++ b/lib/trace-cmd/trace-recorder.c @@ -220,6 +220,41 @@ tracecmd_create_buffer_recorder_fd2(int fd, int fd2, int cpu, unsigned flags, return NULL; } +static void verify_splice(const char *file, unsigned *flags) +{ + int brass[2]; + int ret; + int fd; + + fd = open(file, O_WRONLY | O_CREAT | O_TRUNC | O_LARGEFILE, 0644); + if (fd < 0) + return; /* Will fail by the caller too */ + + if (pipe(brass) < 0) + goto fail_pipe; + + ret = splice(brass[0], NULL, fd, NULL, 0, SPLICE_F_NONBLOCK); + if (ret < 0) + goto fail_splice; + + out_pipe: + close(brass[0]); + close(brass[1]); + out: + close(fd); + return; + + fail_pipe: + tracecmd_warning("Failed opening pipe, trying read/write"); + *flags |= TRACECMD_RECORD_NOSPLICE; + goto out; + + fail_splice: + tracecmd_warning("Failed splice to file, trying read/write"); + *flags |= TRACECMD_RECORD_NOSPLICE; + goto out_pipe; +} + struct tracecmd_recorder * tracecmd_create_buffer_recorder_fd(int fd, int cpu, unsigned flags, const char *buffer) { @@ -233,6 +268,9 @@ __tracecmd_create_buffer_recorder(const char *file, int cpu, unsigned flags, struct tracecmd_recorder *recorder; int fd; + if (!(flags & TRACECMD_RECORD_NOSPLICE)) + verify_splice(file, &flags); + fd = open(file, O_WRONLY | O_CREAT | O_TRUNC | O_LARGEFILE, 0644); if (fd < 0) return NULL;