I recently came accross some anecdotal internet advice _(in the comments section of an entirely unrelated reddit post)_ that `startx` should be rewritten. A link to the source was provided by the commentator, showing this message: "[Site administrators are STRONGLY urged to write nicer versions](https://gitlab.freedesktop.org/xorg/app/xinit/-/blob/master/startx.cpp#L11) " As well as an alternative solution : ``` #!/bin/sh trap "$HOME/.xinitrc" USR1 ( trap '' USR1 exec X -ardelay 200 \ -arinterval 20 \ -dpms \ -nolisten tcp \ -noreset \ :0 vt1 ) & wait ``` This has sparked my interest and wanted to see what people thought of this. #### Edit: I tried this version, and it started Xorg without reading my `~/.xinitrc` so no window manager or wallpaper or any way of opening a terminal.

@"pfr"#p11049 Try: ##### runx.sh ```sh #!/bin/sh set -eu -o pipefail die() { echo >&2 "$0: $*"; exit 1; } test -e /tmp/.X0-lock && die "X server already running" export XAUTHORITY=$HOME/.Xauthority # judge dredd SA=$(mktemp $HOME/.serverauth.XXXXXX) # more XXXs => more hardcore CK=$(/usr/bin/openssl rand -hex 16) # make cookie test -z "$CK" && die "couldn't make cookie" trap "rm -f '$SA' '$XAUTHORITY'" EXIT HUP INT QUIT ILL TRAP BUS TERM # trap "" TSTP # no shell with Ctrl-Z echo "add :0 . $CK" | xauth -q -f $SA # guts unset CK # leave no cookie behind cp -fp $SA $XAUTHORITY # assume 1 server 1 session # off to the races xinit $HOME/.xinitrc -- /usr/X11R7/bin/X :0 -auth $SA ``` ##### runsafe.c ```C /** * Run a program/script safe from interruption by Ctrl-C: * - fork. * - do setsid() in the child (so signals are never delivered to it). * - ignore signals _only_ in parent (so that, if needed, the child * can still be killed by sending it a SIGINT). * * The program is meant to prevent scripts like startx(1) from granting * shell-access via Ctrl-C/Ctrl-Z (and you don't want, for some reason, * to run the X server with: Option DontVTSwitch "on"). * These shell scripts, or the xinit(1) program: * a) are long-running * b) are non-interactive, so the don't need a controlling terminal, * and * c) spawn many other processes which should not be run with those * signals ignored (as they would be (usually) if you ignored the * signals in the calling script using: trap '' TSTP). * * Usage: in startx(1), ignore SIGTSTP and add runsafe before the xinit(1) * command (full path required.): * * trap '' TSTP # ignore Ctrl-Z in shell & runsafe * runsafe /usr/X11R7/bin/xinit ~/.xinitrc -- /usr/X11R7/bin/X :0 */ #include #include #include #include #include #include #include static void dummy(int signo) { (void)signo; } int main(int argc, char* argv[]) { pid_t pid = -1; int status = 0; if (argc < 2) { fprintf(stderr, "Usage: %s prog args...\n", *argv); return EXIT_FAILURE; } switch (pid = fork()) { case -1: err(EXIT_FAILURE, "fork failed"); case 0: /* child */ signal(SIGINT, SIG_DFL); /* default SIGINT */ signal(SIGTSTP, SIG_DFL); /* default SIGTSTP */ if (setsid() == -1) err(EXIT_FAILURE, "setsid failed"); execv(argv[1], argv + 1); err(EXIT_FAILURE, "%s: execv failed", argv[1]); default: /* parent */ signal(SIGINT, SIG_IGN); /* ignore Ctrl-C */ /* * do an orderly shutdown on SIGTERM; * other signals deliberately left as-is. */ signal(SIGTERM, dummy); siginterrupt(SIGTERM, 1); if (wait(&status) != pid) warn("warning: wait failed"); if ((pid = getpgid(pid)) == 0) killpg(pid, SIGTERM); /* kill pgrp stragglers */ break; } return WIFEXITED(status) ? WEXITSTATUS(status) : 128+WTERMSIG(status); } ```

Interesting script. If I understand this correctly, trapping USR1 in the subshell makes the xinitrc script be executed only when the X server os ready to accept connections. Invoking @"rvp"#168 for confirmation. The script however completely bypasses setting an authentication cookie; feel like implementing a xauth line? > @"pfr"#p10966 I tried this version, and it started Xorg without reading my ~/.xinitrc so no window manager or wallpaper or any way of opening a terminal. Try making your .xinitrc script executable if it isn't already.

> @"JuvenalUrbino"#p10978 Invoking [@rvp](https://www.unitedbsd.com/u/rvp) for confirmation I actually emailed this to @"rvp"#168 prior to posting it and he'd also encouraged me to post it on the forums. I'm still hoping that he's reserving his big brain response for this thread 🧠 🎓️ > @"JuvenalUrbino"#p10978 Try making your .xinitrc script executable if it isn't already. Ahhh right, this certainly makes practical sense. Will try this out later. However I'm not sure how to implement an `xauth` line without simply taking it from the `startx` script. I'm waiting for much smarter people _(yourself included Dr. Urbino)_ to give their insight before I decide to ditch `startx` and rely on a custom script.

> @"pfr"#p10979 Ahhh right, this certainly makes practical sense (a good chance to sede what trap is used for)

@"pfr"#p10966 That `USR1` stuff is explained in the [Xserver](https://man.netbsd.org/amd64/Xserver.1) man-page (section SIGNALS) Try this out: ##### runx.sh ```sh #!/bin/sh trap 'DISPLAY=:0 sh $HOME/.xinitrc; kill %1' USR1 ( trap '' USR1 exec /usr/X11R7/bin/X :0 ) & wait ``` Explanation: * trap for `USR1` is registered (not _run_--yet) * The `USR1` signal is ignored * X server runs (in background), and as the man-page explains, notices that `USR1` is being ignored and so send a `USR1` back to its parent--ie. the script. * wait for X server job to terminate (cmd. will exit as soon as the script gets `USR1`). * now the `USR1` trap handler is run and control passes there (the `wait` cmd. gets interrupted). * handler runs the `~/.xinitrc` script * when that exits, the server job is killed, and the whole script finishes when the trap handler is done (as there are no more commands after `wait`).

@"rvp"#p10984 amazing! Honestly, I would have never expected that X had a special handling for SIGUSR1, much less to come across somebody implementing something like that in a script. This is not distant from what I had imagined (how the script worked), yet, I was missing 'why'.

@"JuvenalUrbino"#p10985 Yeah, that is an obscure interaction between the X server and `xdm`, and new to me too--which was why I asked @"pfr"#61 to post on the board. The other interesting thing is re: [this thread](https://www.unitedbsd.com/d/660-what-are-the-benefits-of-a-display-manager-over-startx). There, an `exec startx` is done to prevent someone from getting a shell by terminating `startx` with Ctrl-C. WIth this script, you can do something else (you can do this in `startx` too, but, we're rewriting that now aren't we...?): ignore the signals: ```sh trap '' INT QUIT TERM ``` But, a first attempt is producing some very odd results. Don't know if I've missed some shell subtlety in trap-handling, or if it's a bug in the shell...

@"JuvenalUrbino"#p10983 Making my `.xinitrc` executable didn't help. @"rvp"#p10984 I will give your script a try tonight 👍️ Do you recomend adding `trap '' INT QUIT TERM` and if so, where? #### Edit: Can confirm that your script does work @"rvp"#168 However, it seems that my `~/.Xresources` aren't being read.

> @"pfr"#p10989 Do you recomend adding `trap '' INT QUIT TERM` and if so, where? Not yet. Gotta do some debugging yet. > @"pfr"#p10989 However, it seems that my `~/.Xresources` aren't being read. Got this line in `~/.xinitrc`?: ```sh xrdb -merge ~/.Xresources ```

> @"rvp"#p10984 Try this out: > trap 'DISPLAY=:0 sh $HOME/.xinitrc; kill %1' USR1 Why is prepending `sh` mandatory here though? I've trapped scripts directly without problem and I wonder why this wasn't working. > @"rvp"#p10986 But, a first attempt is producing some very odd results. Don't know if I've missed some shell subtlety in trap-handling, or if it's a bug in the shell... What kind of results are you getting?

> @"JuvenalUrbino"#p11000 Why is prepending `sh` mandatory here though? I've trapped scripts directly without problem and I wonder why this wasn't working. `.xinitrc` will work if it is made executable. I put in the `sh` so it'll work even w/o an exec bit (as long as it is an `sh` script). > @"JuvenalUrbino"#p11000 What kind of results are you getting? I'll email you: Nobody likes watching sausages being made...

Alternative startx

rvp

pfr Hey, this works like a charm!

It should. It's just startx with all the generality ripped out.

pfr Is enabling DontVTSwitch necessary when using this script

Yes it is. The other way is to:
a) uncomment the trap "" TSTP line in the (updated) script, and
b) run xinit under the runsafe program.

pfr

rvp Option a seems... easier 😄

rvp

pfr The choice is either DontVTSwitch or a + b

pfr

rvp
Ahhhh gotcha.

So then, can you give me an example as to why someone would not want to turn on DontVTSwitch?

secondly, please remind me how to compile runsafe.c?
I'm getting some errors when I try to compile it with gcc

~ $ gcc -o runsafe runsafe.c
runsafe.c: In function ‘main’:
runsafe.c:43:2: error: ‘pid_tpid’ undeclared (first use in this function); did you mean ‘pid_t’?
  pid_tpid = -1;
  ^~~~~~~~
  pid_t
runsafe.c:43:2: note: each undeclared identifier is reported only once for each function it appears in
runsafe.c:51:10: error: ‘pid’ undeclared (first use in this function); did you mean ‘pipe’?
  switch (pid = fork()) {
          ^~~
          pipe
runsafe.c:72:24: error: expected ‘)’ before ‘;’ token
    killpg(pid, SIGTERM); /* kill pgrp stragglers */
                        ^
runsafe.c:74:2: error: expected expression before ‘}’ token
  }
  ^

Thanks!

Edit:

I just realised that I cant disable the ability to switch to TTY otherwise I wont be able to get back to my desktop after my laptop sleeps/hibernates

I set powerd LidSwitch action that allows me to put my laptop into hibernation after closing the lid and the only way to resume my session again after opening the laptop is to first switch to another TTYctrl-alt-F4 then back to my desktop ctrl-alt-F5

Any other solution here or is it now just a tradeoff?

rvp

pfr I'm getting some errors when I try to compile it with gcc

~ $ gcc -o runsafe runsafe.c
runsafe.c: In function ‘main’:
runsafe.c:43:2: error: ‘pid_tpid’ undeclared (first use in this function); did you mean ‘pid_t’?
  pid_tpid = -1;
  ^~~~~~~~
  pid_t

This looks lile a cut-'n-paste error: the code actually has a space between the type and the variable.

pfr secondly, please remind me how to compile runsafe.c?

Here.

pfr

rvp Cool, thank you.

Any thoughts on whether it's actualy worth it thouh, given my need to switch VT's in order to bring my laptop back from sleep?

rvp

pfr Any thoughts on whether it's actualy worth it thouh, given my need to switch VT's in order to bring my laptop back from sleep?

That's one of the reasons why I cooked up this method. I don't much like DontVTSwitch either.

pfr

Cool. Runsafe compiles and seems to work. I'm running it at the end of the runx script like so:
/usr/local/bin/runsafe $(xinit $HOME/.xinitrc -- /usr/X11R7/bin/X :0 -auth $SA)

rvp

pfr Not correct. Get rid of the $( ). See the example in the comment at the top of the source code.

pfr

rvp
I only added the $( ) because without it I get this error:

runsafe: xinit: execv failed: No such file or directory

rvp

pfr runsafe: xinit: execv failed: No such file or directory

For security reasons, PATH is not searched for the command to be executed. Give the full path to xinit. (This is mentioned in the comment in the C source, along with an example...)

pfr

rvp runsafe /usr/X11R7/bin/xinit ~/.xinitrc -- /usr/X11R7/bin/X :0

Gotcha. Here I was giving the full path to runsafe 🤦

« Previous Page