RFR: 8364312: debug agent should set FD_CLOEXEC flag rather than explicitly closing every open file [v3]

Kevin Walls kevinw at openjdk.org
Wed Aug 6 10:58:04 UTC 2025


On Tue, 5 Aug 2025 10:32:42 GMT, Guanqiang Han <ghan at openjdk.org> wrote:

>> Create a new function that marks all file descriptors found in /proc/self/fd with the FD_CLOEXEC flag to ensure they are automatically closed upon execution of a new program via exec().
>
> Guanqiang Han has updated the pull request with a new target base due to a merge or a rebase. The incremental webrev excludes the unrelated changes brought in by the merge/rebase. The pull request contains four additional commits since the last revision:
> 
>  - a small fix
>  - Merge remote-tracking branch 'upstream/master' into 8364312
>  - Update exec_md.c
>    
>    correct an compilation error
>  - 8364312: debug agent should set FD_CLOEXEC flag rather than explicitly closing every open file
>    
>    Create a new function that marks all file descriptors found in /proc/self/fd with the FD_CLOEXEC flag to ensure they are automatically closed upon execution of a new program via exec().

Hi -

(The code duplication from src/java.base/unix/native/libjava/childproc.c is unfortunate but maybe it's hard to avoid.)

exec_md.c has the existing closeDescriptors() that "returns 0 on failure and 1 on success." which is unusual and unfortunate as zero is a typical success value.

As this is only used locally, this should be an opportunity to change that, and have the new markDescriptorsCloseOnExec() return more normal values, like 0 for success and -1 for failure (like in childproc.c which this borrows from).
(Unless Chris disagrees or sees wider usage...)


Another existing problem:
forkedChildProcess() comments that it will: "Close all file descriptors that have been copied over"
But it calls closeDescriptors() which says it: "Closes every file descriptor that is listed as a directory"

Can we make the comments consistent while we are here? 8-)

The comments before forkedChildProcess() at line 118 and within it at 124 talk about actual closing, not marking as close on exec.

-------------

PR Review: https://git.openjdk.org/jdk/pull/26568#pullrequestreview-3092000666


More information about the serviceability-dev mailing list