fix(browser): stop treating a missing cmdline as proof the daemon exited

_terminate_owned_daemon() read /proc/<pid>/cmdline and, on FileNotFoundError,
unlinked the pid file on the stated assumption that "the daemon may have
exited". Off Linux that file is always missing, so the branch always fired:
the pid file of a live daemon was deleted and the daemon itself never killed.
_owned_daemon_exists() swallowed the same error and therefore always returned
False, which is precisely the state its own docstring warns about, since a
close against an unrecognised session can bootstrap a fresh daemon and wait on
its browser indefinitely.

Demonstrated on macOS before the change: a pid file holding a live pid is
removed by _terminate_owned_daemon() and _owned_daemon_exists() reports False.
After it, the file survives and the daemon is reported present.

_process_command_line() now returns None for "this host cannot tell" and
_process_is_alive() answers the separate question of whether the pid exists.
Without procfs we decline to kill a process we cannot confirm is ours, and we
only forget a pid file once the pid is genuinely gone. Linux behaviour is
unchanged: the command-line identity check still gates both paths.
This commit is contained in:
Léo
2026-09-29 16:27:22 +02:00
parent 6105702901
commit 6cda92080c
2 changed files with 131 additions and 13 deletions
+53 -13
View File
@@ -125,6 +125,37 @@ def _browser_pid_file_candidates(
# host that has no procfs, and on one that does.
_PROC_ROOT = Path("/proc")
def _process_command_line(pid: int) -> str | None:
"""Command line of a running process, or ``None`` when it cannot be read.
``None`` means "this host cannot tell", not "the process is gone". Off
Linux there is no procfs to read a command line from, so callers must not
treat it as proof that the process exited.
"""
try:
return (_PROC_ROOT / str(pid) / "cmdline").read_bytes().replace(
b"\0", b" "
).decode("utf-8", errors="replace")
except (OSError, UnicodeError):
return None
def _process_is_alive(pid: int) -> bool:
"""Whether a pid currently exists. Signal 0 checks without delivering."""
try:
os.kill(pid, 0)
except ProcessLookupError:
return False
except PermissionError:
# Alive, owned by somebody else.
return True
except OSError:
return False
return True
_SCHOLARLY_METADATA_CUE_RE = re.compile(
r"\b(?:accept(?:ed|ance)?|publish(?:ed|ing|cation)?|venue|conference|"
r"journal|proceedings|doi)\b",
@@ -2359,16 +2390,18 @@ class PrivateBrowserTool:
for pid_file in pid_files:
try:
pid = int(pid_file.read_text().strip())
command_line = (Path("/proc") / str(pid) / "cmdline").read_bytes().replace(
b"\0", b" "
).decode("utf-8", errors="replace")
except FileNotFoundError:
# The daemon may have exited between writing its pid file and
# this cleanup pass. The exact file is still ours to remove.
with contextlib.suppress(FileNotFoundError, PermissionError, OSError):
pid_file.unlink()
except (OSError, ValueError):
continue
except (OSError, UnicodeError, ValueError):
command_line = _process_command_line(pid)
if command_line is None:
# Either the daemon exited between writing its pid file and
# this pass, or this host has no procfs to ask. Only the first
# justifies forgetting the pid file. Without procfs we cannot
# confirm the process is ours, so we neither kill it nor drop
# the record that would let a later pass find it.
if not _process_is_alive(pid):
with contextlib.suppress(FileNotFoundError, PermissionError, OSError):
pid_file.unlink()
continue
if "agent-browser" in command_line:
with contextlib.suppress(ProcessLookupError, PermissionError, OSError):
@@ -2395,10 +2428,17 @@ class PrivateBrowserTool:
for pid_file in _browser_pid_file_candidates(runtime_dir, namespace, session_id):
try:
pid = int(pid_file.read_text().strip())
command_line = (Path("/proc") / str(pid) / "cmdline").read_bytes().replace(
b"\0", b" "
).decode("utf-8", errors="replace")
except (FileNotFoundError, OSError, UnicodeError, ValueError):
except (OSError, ValueError):
continue
command_line = _process_command_line(pid)
if command_line is None:
# Without procfs we can only tell that something with this pid
# is alive, not that it is agent-browser. The pid file is our
# own namespaced one, so treat a live pid as a match: answering
# "no daemon" here is what lets `close` bootstrap a fresh one
# and wait on its browser forever.
if _process_is_alive(pid):
return True
continue
if "agent-browser" in command_line:
return True