exec:start() throws error .../_build/default/lib/erlexec/priv/false: no such file or directory

[elsbiet]

hello,

when executing exec:start().

the following exception is thrown.

** exception throw: "Cannot find file .../_build/default/lib/erlexec/priv/false: no such file or directory"
in function exec:is_suid_and_root_owner/1 (.../_build/default/lib/erlexec/src/exec.erl, line 1111)
in call from exec:check_options/1 (.../_build/default/lib/erlexec/src/exec.erl, line 1124)
in call from exec:start/1 (.../_build/default/lib/erlexec/src/exec.erl, line 395)

what did i miss?

thanks in advance for your answer, g.

my environment:

Debian GNU/Linux 12 (bookworm)
Erlang/OTP 25 [erts-13.1.5] [source] [64-bit] [smp:12:12] [ds:12:12:10] [async-threads:1] [jit:ns]

[elsbiet]

likely exec-port was misplaced by the rebar generating process for an unknown reason.

after

mv _build/default/lib/erlexec/priv/exec-port _build/default/lib/erlexec/priv/x86_64-pc-linux-gnu

exec:start() , exec:run(..) are working fine.

[saleyn]

Could you pull the latest, and paste the output of:

make clean
make info
make

[elsbiet]

as i'm using rebar3 ( {erlexec, "~> 2.0"} to deps) i'm unsure how to do this in the rebar-environment.

i"m unsure how to do this in my rebar environment.

remark: the issue might be related to the fact that the os user is kind of limited. the problem does not exist for another user on the same machine.

[saleyn]

You can achieve that by modifying the pre_hooks in rebar.config in your local environment with:

{pre_hooks,  [
  {"(freebsd|netbsd|openbsd)", compile, "gmake -C c_src info all"},
  {"(linux|darwin|solaris)",   compile, "make -C c_src info all"}
]}.

and then pasting here the dependency build output for erlexec.

[elsbiet]

hello and thanks for your answer.

i've been trying very hard to reproduce the issue but did not succeed. i guess the issue was due to a partly temporary dysfunctional machine (which is beyond my beyond my capabilities).

sorry for any inconveniences.

[rNoz]

I have reached the same error. Any of the affected ones, did you find what was causing it?

[rNoz]

Hello @elsbiet,

Have been a couple of days investigating this case. And I have some proposals, let's see what do you think.

First of all, thanks for this cool library!

Cloned your repo, created the branch rnoz/erlexec-false and iterated there.

In this commit, we can replicate the failure we got at runtime.

In the final version of the branch I have added not only more tests showing other corner cases that would fail, but also modified the main default function to fix all those cases.

Another proposal to simplify testability. How do you think if we can pass as an option where to get the portexe dir, by default would be code:priv_dir but the user can specify as well. In that way, I can compact setup functions for each test.

I can create PRs for these things. Let me know what do you think.

Btw, what do you use to format the code? I don't get your style with my editor.

Thanks.

[saleyn]

@rNoz please submit a PR (note to normalize the indent width to 4 spaces)

[saleyn]

@rNoz with your commit I cannot reproduce the error on Linux. This is the compiled project structure I get, when running make test:

$ find _build/test/
_build/test/
_build/test/lib
_build/test/lib/.rebar3
_build/test/lib/.rebar3/rebar_compiler_erl
_build/test/lib/.rebar3/rebar_compiler_erl/source_project_apps.dag
_build/test/lib/erlexec
_build/test/lib/erlexec/include
_build/test/lib/erlexec/ebin
_build/test/lib/erlexec/ebin/exec_app.beam
_build/test/lib/erlexec/ebin/erlexec.app
_build/test/lib/erlexec/ebin/exec.beam
_build/test/lib/erlexec/priv
_build/test/lib/erlexec/src
_build/test/lib/erlexec/test
_build/test/lib/erlexec/test/test_exec.erl
_build/test/lib/erlexec/test/test_exec.beam
_build/test/plugins
_build/test/plugins/rebar3_ex_doc
_build/test/plugins/verl
_build/test/plugins/rebar3_hex
_build/test/plugins/hex_core

Note that priv directory is a link:

$ ls -l _build/test/lib/erlexec/
total 8
drwxr-xr-x 2 serge serge 4096 Jun 15 10:09 ebin
lrwxrwxrwx 1 serge serge   19 Jun 15 10:09 include -> ../../../../include
lrwxrwxrwx 1 serge serge   16 Jun 15 10:09 priv -> ../../../../priv
lrwxrwxrwx 1 serge serge   15 Jun 15 10:09 src -> ../../../../src
drwxr-xr-x 2 serge serge 4096 Jun 15 10:09 test

$ pwd
/home/serge/projects/erl-libs/erlexec
$ find priv
priv
priv/x86_64-pc-linux-gnu
priv/x86_64-pc-linux-gnu/exec-port

Notably the compiled port program exec-port stays under the project's ./priv directory with rebar3 creating links in all environments to this location. Can you provide the output in your case on a MacOS, as it is unclear where it fails to find the executable in your environment.

[rNoz]

Yes I use MacOS for development (yes, same files as you), however, the issue appeared in a unix machine (docker, aws). I have been investigating and couldn't get why it happened, but the priv directory was not copied to the image at that time. I use elixir, and that was part of mix release. This is quite tricky because same code was working in previous deployments, so priv was copied.

So, the fixes I am going to propose you are to improve the detection during initialization and errors emitted (because having priv/false was confusing and the pod was constantly restarting, so I couldn't verify if we got somehow that file instead of the exec-port or if was code-generated at runtime -the case-).

For instance, in that commit (I have just seen that I have to remove other unused funcs) it should pass the test, showing how it can generate the wrong path ending with priv/false. I'll prepare PRs along the day.

I iterated a bit more this morning, and as soon as I change the suffix, I get the error:

% simulating we don't have the priv dir
default_portexe_no_priv_test_() ->
    Priv = code:priv_dir(erlexec),
    Renamed = Priv ++ "_bak",
    {setup,
     fun() ->
         case filelib:is_dir(Priv) of
             true ->
               ok = file:rename(Priv, Renamed),
               renamed;
             false -> no_priv
         end
     end,
     fun(renamed) ->
         case filelib:is_dir(Renamed) of
             true -> file:rename(Renamed, Priv);
             false -> ok
         end
     end,
     fun() ->
       ?assert(not filelib:is_dir(Priv)),
       ?assert(filelib:is_dir(Renamed)),
       ?assert(lists:suffix("/priv/fals", exec:default(portexe))) % see the `fals` instead of false
     end}.

Failures:

  1) exec:-default_portexe_no_priv_test_/0-fun-3-/0
     Failure/Error: ?assert(lists : suffix ( "/priv/fals" , exec : default ( portexe ) ))
       expected: true
            got: false
     %% $$$/erlexec/src/exec.erl:1516:in `exec:-default_portexe_no_priv_test_/0-fun-3-/2`

Finished in 0.013 seconds
1 tests, 1 failures
===> Error running tests
```

Regarding format, I can do indent width to 4 spaces, as you see in above snippet. However, your format is slightly different (e.g., cases without indentation). Not sure if you recommend some erlang tooling to get that deterministically (and quickly).