civodul pushed a commit to branch main
in repository shepherd.
commit 787d5a33aea061b5052faa0863c96be722440ce3
Author: Ludovic Courtès <[email protected]>
AuthorDate: Sun Apr 13 19:09:08 2025 +0200
shepherd: Exit if another instance is listening on the socket.
Partly fixes <https://issues.guix.gnu.org/76998>.
* modules/shepherd/comm.scm (open-server-socket): Add #:false-if-in-use?
and honor it.
* modules/shepherd.scm (call-with-server-socket): Pass #:false-if-in-use?
and exit when ‘open-server-socket’ returns #f.
* tests/basic.sh: Test it.
* NEWS: Update.
Reported-by: Daniel Littlewood <[email protected]>
Reported-by: Jake <[email protected]>
Reported-by: Danny Milosavljevic <[email protected]>
---
NEWS | 8 ++++++++
modules/shepherd.scm | 11 ++++++++++-
modules/shepherd/comm.scm | 34 ++++++++++++++++++++++++++++------
tests/basic.sh | 4 ++++
4 files changed, 50 insertions(+), 7 deletions(-)
diff --git a/NEWS b/NEWS
index f843b62..96bfd70 100644
--- a/NEWS
+++ b/NEWS
@@ -86,6 +86,14 @@ The ‘system*’ and ‘system’ replacements in the ‘shepherd’
process no
given command in the current directory rather than under
(default-service-directory).
+** Refuse to start when another shepherd is listening on the socket
+ (<https://issues.guix.gnu.org/76998>)
+
+Starting an additional ‘shepherd’ as a user used to lead it to take control of
+the socket (by default /run/user/UID/shepherd/socket) even though another
+instance was already running and listening to that socket. Since that
+behavior is undesirable, ‘shepherd’ now refuses to start in this situation.
+
* Changes in 1.0.3
** ‘spawn-command’ now honors #:log-file
diff --git a/modules/shepherd.scm b/modules/shepherd.scm
index a1bfebb..bf58199 100644
--- a/modules/shepherd.scm
+++ b/modules/shepherd.scm
@@ -55,7 +55,16 @@
socket file at FILE-NAME upon exit of PROC. Return the values of PROC."
(let ((sock (catch 'system-error
(lambda ()
- (open-server-socket file-name))
+ (or (open-server-socket file-name
+ #:false-if-in-use?
+ (not (= 1 (getpid))))
+ (begin
+ ;; Refuse to start when another shepherd is already
+ ;; listening on FILE-NAME.
+ (report-error (l10n "shepherd instance already \
+listening on '~a'")
+ file-name)
+ (exit 1))))
(lambda args
(match args
((key proc . _)
diff --git a/modules/shepherd/comm.scm b/modules/shepherd/comm.scm
index 8c1050e..6e65716 100644
--- a/modules/shepherd/comm.scm
+++ b/modules/shepherd/comm.scm
@@ -105,17 +105,39 @@ return the socket."
(list errno) rest)))
sock)))
-(define (open-server-socket file-name)
- "Open a socket at FILE-NAME, and listen for connections there."
+(define* (open-server-socket file-name #:key false-if-in-use?)
+ "Open a socket at @var{file-name}, and listen for connections there. When
+@var{false-if-in-use?} is true, attempt to connect to @var{file-name} first
+and return #f if the connection was successfully established--meaning a server
+is already listening on @var{file-name}."
(with-fluids ((%default-port-encoding "UTF-8"))
(let ((sock (socket PF_UNIX
(logior SOCK_STREAM SOCK_NONBLOCK SOCK_CLOEXEC)
0))
(address (make-socket-address AF_UNIX file-name)))
- (catch-system-error (delete-file file-name))
- (bind sock address)
- (listen sock 10)
- sock)))
+ (and (or (not false-if-in-use?)
+ ;; Try to connect to ADDRESS and return #f if that works.
+ ;; Since SOCK is non-blocking, retry a few times.
+ (catch 'system-error
+ (lambda ()
+ (let loop ((i 0))
+ (unless (or (connect sock address) ;non-blocking
+ (> i 5))
+ (sleep 1)
+ (loop (+ 1 i))))
+ (close-port sock)
+ #f)
+ (lambda args
+ (or (memv (system-error-errno args)
+ (list ENOENT ECONNREFUSED))
+ (begin
+ (close-port sock)
+ #t)))))
+ (begin
+ (catch-system-error (delete-file file-name))
+ (bind sock address)
+ (listen sock 10)
+ sock)))))
(define (read-command port)
"Receive a command from PORT; return the command of #f if something went
diff --git a/tests/basic.sh b/tests/basic.sh
index 2214e78..975f7e3 100644
--- a/tests/basic.sh
+++ b/tests/basic.sh
@@ -86,6 +86,10 @@ shepherd_pid="`cat $pid`"
kill -0 $shepherd_pid
test -S "$socket"
+
+# Attempt to reuse $socket should fail.
+shepherd -I -s "$socket" -c "$conf" && false
+
pristine_status=`$herd status -n 0` # Prep for 'reload' test.
echo $pristine_status | grep -E '(Start.*root|Stop.*test)'
