Re: [PATCH 04/24] perf daemon: Add server socket support

[Jiri Olsa <jolsa@redhat.com>]

On Wed, Feb 03, 2021 at 06:04:23PM -0300, Arnaldo Carvalho de Melo wrote:
> Em Sun, Jan 31, 2021 at 12:48:36AM +0100, Jiri Olsa escreveu:
> > Add support to create server socket that listens for client
> > commands and process them.
> > 
> > This patch adds only the core support, all commands using
> > this functionality are coming in following patches.
> > 
> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > ---
> >  tools/perf/builtin-daemon.c | 101 +++++++++++++++++++++++++++++++++++-
> >  1 file changed, 100 insertions(+), 1 deletion(-)
> > 
> > diff --git a/tools/perf/builtin-daemon.c b/tools/perf/builtin-daemon.c
> > index 8d0ac44ec808..756d60616d7d 100644
> > --- a/tools/perf/builtin-daemon.c
> > +++ b/tools/perf/builtin-daemon.c
> > @@ -1,5 +1,6 @@
> >  // SPDX-License-Identifier: GPL-2.0
> >  #include <subcmd/parse-options.h>
> > +#include <api/fd/array.h>
> >  #include <linux/limits.h>
> >  #include <string.h>
> >  #include <signal.h>
> > @@ -7,6 +8,10 @@
> >  #include <stdio.h>
> >  #include <unistd.h>
> >  #include <errno.h>
> > +#include <sys/types.h>
> > +#include <sys/socket.h>
> > +#include <sys/un.h>
> > +#include <poll.h>
> >  #include "builtin.h"
> >  #include "perf.h"
> >  #include "debug.h"
> > @@ -37,6 +42,78 @@ static void sig_handler(int sig __maybe_unused)
> >  	done = true;
> >  }
> >  
> > +static int setup_server_socket(struct daemon *daemon)
> > +{
> > +	struct sockaddr_un addr;
> > +	char path[100];
> > +	int fd;
> > +
> > +	fd = socket(AF_UNIX, SOCK_STREAM, 0);
> 
> Minor, combine decl with use, since line isn't long and its one after
> the other, i.e.:
> 
> 	int fd = socket(AF_UNIX, SOCK_STREAM, 0);

hum, sure, but I'm missing the point.. I think it's less readable

> 
> > +	if (fd < 0) {
> > +		fprintf(stderr, "socket: %s\n", strerror(errno));
> > +		return -1;
> > +	}
> > +
> > +	fcntl(fd, F_SETFD, FD_CLOEXEC);
> 
> Don't we have to check its return?

yep, will add

> 
> > +
> > +	scnprintf(path, PATH_MAX, "%s/control", daemon->base);
> 
> Humm the safe thing here is to use:
> 
> 	scnprintf(path, sizeof(path), "%s/control", daemon->base);
> 
> Using it like that would avoid the bug in your code, as path has only
> 100 bytes, not PATH_MAX bytes ;-)

right, will change

> 
> > +
> > +	memset(&addr, 0, sizeof(addr));
> > +	addr.sun_family = AF_UNIX;
> > +
> > +	strncpy(addr.sun_path, path, sizeof(addr.sun_path) - 1);
> 
> strncpy may end up not adding the final \0 see the NOTES in its man
> page. Consider using strlcpy instead. See:
> 
>   bef0b8970f27da5c ("perf probe: Fix unchecked usage of strncpy()")

hum, it's memset-ed to 0 for that an there's -1 in the size,
so I'd think there's zero at the end, but we can use strlcpy
to make it more obvious

SNIP

> > +		fprintf(stderr, "accept: %s\n", strerror(errno));
> > +		return -1;
> > +	}
> > +
> > +	if (sizeof(cmd) != read(fd, &cmd, sizeof(cmd))) {
> > +		fprintf(stderr, "read: %s\n", strerror(errno));
> 
> close fd
> 
> > +		return -1;
> > +	}
> > +
> > +	out = fdopen(fd, "w");
> > +	if (!out) {
> > +		perror("failed: fdopen");
> 
> close fd
> 
> I.e. goto out_close;
> 
> > +		return -1;
> > +	}
> > +
> > +	switch (cmd.cmd) {
> > +	default:
> > +		break;
> > +	}
> > +
> > +	fclose(out);
> 
> out_close:
> 
> > +	close(fd);
> > +	return ret;

ugh, I overlooked this one

thanks
jirka