| 1 |
/* SPDX-License-Identifier: LGPL-2.1-or-later */ |
| 2 |
/*** |
| 3 |
Copyright © 2010 ProFUSION embedded systems |
| 4 |
***/ |
| 5 |
|
| 6 |
#include <signal.h> |
| 7 |
#include <sys/wait.h> |
| 8 |
|
| 9 |
#include "alloc-util.h" |
| 10 |
#include "errno-util.h" |
| 11 |
#include "fd-util.h" |
| 12 |
#include "format-util.h" |
| 13 |
#include "initrd-util.h" |
| 14 |
#include "killall.h" |
| 15 |
#include "log.h" |
| 16 |
#include "pidref.h" |
| 17 |
#include "process-util.h" |
| 18 |
#include "set.h" |
| 19 |
#include "string-util.h" |
| 20 |
#include "terminal-util.h" |
| 21 |
#include "time-util.h" |
| 22 |
|
| 23 |
static int argv_has_at(const PidRef *pid) { |
| 24 |
int r; |
| 25 |
|
| 26 |
assert(pidref_is_set(pid)); |
| 27 |
assert(!pidref_is_remote(pid)); |
| 28 |
|
| 29 |
const char *p = procfs_file_alloca(pid->pid, "cmdline"); |
| 30 |
_cleanup_fclose_ FILE *f = fopen(p, "re"); |
| 31 |
if (!f) |
| 32 |
return log_debug_errno(errno, "Failed to open %s, ignoring: %m", p); |
| 33 |
|
| 34 |
/* Try to read the first character of the command line. If the cmdline is empty (which might be the case for |
| 35 |
* kernel threads but potentially also other stuff), this line won't do anything, but we don't care much, as |
| 36 |
* actual kernel threads are already filtered out above. */ |
| 37 |
char c = 0; |
| 38 |
(void) fread(&c, 1, 1, f); |
| 39 |
|
| 40 |
r = pidref_verify(pid); |
| 41 |
if (r < 0) |
| 42 |
return log_debug_errno(r, "Failed to verify pid " PID_FMT ", ignoring: %m", pid->pid); |
| 43 |
|
| 44 |
/* Processes with argv[0][0] = '@' we ignore from the killing spree. |
| 45 |
* |
| 46 |
* https://systemd.io/ROOT_STORAGE_DAEMONS */ |
| 47 |
return c == '@'; |
| 48 |
} |
| 49 |
|
| 50 |
static bool is_in_survivor_cgroup(const PidRef *pid) { |
| 51 |
_cleanup_free_ char *cgroup_path = NULL; |
| 52 |
int r; |
| 53 |
|
| 54 |
assert(pidref_is_set(pid)); |
| 55 |
|
| 56 |
r = cg_pidref_get_path(pid, &cgroup_path); |
| 57 |
if (r == -EUNATCH) { |
| 58 |
log_warning_errno(r, "Process " PID_FMT " appears to originate in foreign namespace, ignoring.", pid->pid); |
| 59 |
return true; |
| 60 |
} |
| 61 |
if (r < 0) { |
| 62 |
log_warning_errno(r, "Failed to get cgroup path of process " PID_FMT ", ignoring: %m", pid->pid); |
| 63 |
return false; |
| 64 |
} |
| 65 |
|
| 66 |
r = cg_get_xattr_bool(cgroup_path, "user.survive_final_kill_signal"); |
| 67 |
if (r < 0 && !ERRNO_IS_NEG_XATTR_ABSENT(r)) |
| 68 |
log_debug_errno(r, |
| 69 |
"Failed to get survive_final_kill_signal xattr of %s, ignoring: %m", |
| 70 |
cgroup_path); |
| 71 |
|
| 72 |
return r > 0; |
| 73 |
} |
| 74 |
|
| 75 |
static bool ignore_proc(PidRef *pid, bool warn_rootfs) { |
| 76 |
uid_t uid; |
| 77 |
|
| 78 |
assert(pidref_is_set(pid)); |
| 79 |
|
| 80 |
/* We are PID 1, let's not commit suicide */ |
| 81 |
if (pid->pid == 1) |
| 82 |
return true; |
| 83 |
|
| 84 |
/* Ignore kernel threads */ |
| 85 |
if (pidref_is_kernel_thread(pid) != 0) |
| 86 |
return true; /* also ignore processes where we can't determine this */ |
| 87 |
|
| 88 |
/* Ignore processes that are part of a cgroup marked with the user.survive_final_kill_signal xattr */ |
| 89 |
if (is_in_survivor_cgroup(pid)) |
| 90 |
return true; |
| 91 |
|
| 92 |
if (pidref_get_uid(pid, &uid) < 0) |
| 93 |
return true; /* not really, but better safe than sorry */ |
| 94 |
|
| 95 |
/* Non-root processes otherwise are always subject to be killed */ |
| 96 |
if (uid != 0) |
| 97 |
return false; |
| 98 |
|
| 99 |
if (argv_has_at(pid) == 0) |
| 100 |
return false; /* if this fails, ignore the process */ |
| 101 |
|
| 102 |
if (warn_rootfs && |
| 103 |
pidref_from_same_root_fs(pid, NULL) > 0) { |
| 104 |
_cleanup_free_ char *comm = NULL; |
| 105 |
|
| 106 |
(void) pidref_get_comm(pid, &comm); |
| 107 |
|
| 108 |
log_notice("Process " PID_FMT " (%s) has been marked to be excluded from killing. It is " |
| 109 |
"running from the root file system, and thus likely to block re-mounting of the " |
| 110 |
"root file system to read-only. Please consider moving it into an initrd file " |
| 111 |
"system instead.", pid->pid, strna(comm)); |
| 112 |
} |
| 113 |
|
| 114 |
return true; |
| 115 |
} |
| 116 |
|
| 117 |
static void log_children_not_yet_killed(Set *pids) { |
| 118 |
_cleanup_free_ char *lst_child = NULL; |
| 119 |
int r; |
| 120 |
|
| 121 |
void *p; |
| 122 |
SET_FOREACH(p, pids) { |
| 123 |
_cleanup_free_ char *s = NULL; |
| 124 |
|
| 125 |
if (pid_get_comm(PTR_TO_PID(p), &s) >= 0) |
| 126 |
r = strextendf_with_separator(&lst_child, ", ", PID_FMT " (%s)", PTR_TO_PID(p), s); |
| 127 |
else |
| 128 |
r = strextendf_with_separator(&lst_child, ", ", PID_FMT, PTR_TO_PID(p)); |
| 129 |
if (r < 0) |
| 130 |
return (void) log_oom_warning(); |
| 131 |
} |
| 132 |
|
| 133 |
if (isempty(lst_child)) |
| 134 |
return; |
| 135 |
|
| 136 |
log_warning("Waiting for process: %s", lst_child); |
| 137 |
} |
| 138 |
|
| 139 |
static int wait_for_children(Set *pids, sigset_t *mask, usec_t timeout) { |
| 140 |
usec_t until, date_log_child, n; |
| 141 |
|
| 142 |
assert(mask); |
| 143 |
|
| 144 |
/* Return the number of children remaining in the pids set: That correspond to the number |
| 145 |
* of processes still "alive" after the timeout */ |
| 146 |
|
| 147 |
if (set_isempty(pids)) |
| 148 |
return 0; |
| 149 |
|
| 150 |
n = now(CLOCK_MONOTONIC); |
| 151 |
until = usec_add(n, timeout); |
| 152 |
date_log_child = usec_add(n, 10u * USEC_PER_SEC); |
| 153 |
if (date_log_child > until) |
| 154 |
date_log_child = usec_add(n, timeout / 2u); |
| 155 |
|
| 156 |
for (;;) { |
| 157 |
struct timespec ts; |
| 158 |
int k; |
| 159 |
void *p; |
| 160 |
|
| 161 |
/* First, let the kernel inform us about killed |
| 162 |
* children. Most processes will probably be our |
| 163 |
* children, but some are not (might be our |
| 164 |
* grandchildren instead...). */ |
| 165 |
for (;;) { |
| 166 |
pid_t pid; |
| 167 |
|
| 168 |
pid = waitpid(-1, NULL, WNOHANG); |
| 169 |
if (pid == 0) |
| 170 |
break; |
| 171 |
if (pid < 0) { |
| 172 |
if (errno == ECHILD) |
| 173 |
break; |
| 174 |
|
| 175 |
return log_error_errno(errno, "waitpid() failed: %m"); |
| 176 |
} |
| 177 |
|
| 178 |
(void) set_remove(pids, PID_TO_PTR(pid)); |
| 179 |
} |
| 180 |
|
| 181 |
/* Now explicitly check who might be remaining, who |
| 182 |
* might not be our child. */ |
| 183 |
SET_FOREACH(p, pids) { |
| 184 |
|
| 185 |
/* kill(pid, 0) sends no signal, but it tells |
| 186 |
* us whether the process still exists. */ |
| 187 |
if (kill(PTR_TO_PID(p), 0) == 0) |
| 188 |
continue; |
| 189 |
|
| 190 |
if (errno != ESRCH) |
| 191 |
continue; |
| 192 |
|
| 193 |
set_remove(pids, p); |
| 194 |
} |
| 195 |
|
| 196 |
if (set_isempty(pids)) |
| 197 |
return 0; |
| 198 |
|
| 199 |
n = now(CLOCK_MONOTONIC); |
| 200 |
if (date_log_child > 0 && n >= date_log_child) { |
| 201 |
log_children_not_yet_killed(pids); |
| 202 |
/* Log the children not yet killed only once */ |
| 203 |
date_log_child = 0; |
| 204 |
} |
| 205 |
|
| 206 |
if (n >= until) |
| 207 |
return set_size(pids); |
| 208 |
|
| 209 |
if (date_log_child > 0) |
| 210 |
timespec_store(&ts, MIN(until - n, date_log_child - n)); |
| 211 |
else |
| 212 |
timespec_store(&ts, until - n); |
| 213 |
|
| 214 |
k = sigtimedwait(mask, NULL, &ts); |
| 215 |
if (k != SIGCHLD) { |
| 216 |
|
| 217 |
if (k < 0 && errno != EAGAIN) |
| 218 |
return log_error_errno(errno, "sigtimedwait() failed: %m"); |
| 219 |
|
| 220 |
if (k >= 0) |
| 221 |
log_warning("sigtimedwait() returned unexpected signal."); |
| 222 |
} |
| 223 |
} |
| 224 |
} |
| 225 |
|
| 226 |
static int killall(int sig, Set *pids, bool send_sighup) { |
| 227 |
_cleanup_closedir_ DIR *dir = NULL; |
| 228 |
int n_killed = 0, r; |
| 229 |
|
| 230 |
/* Send the specified signal to all remaining processes, if not excluded by ignore_proc(). |
| 231 |
* Returns the number of processes to which the specified signal was sent */ |
| 232 |
|
| 233 |
r = proc_dir_open(&dir); |
| 234 |
if (r < 0) |
| 235 |
return log_warning_errno(r, "Failed to open %s: %m", "/proc/"); |
| 236 |
|
| 237 |
for (;;) { |
| 238 |
_cleanup_(pidref_done) PidRef pidref = PIDREF_NULL; |
| 239 |
|
| 240 |
r = proc_dir_read_pidref(dir, &pidref); |
| 241 |
if (r < 0) |
| 242 |
return log_warning_errno(r, "Failed to enumerate /proc/: %m"); |
| 243 |
if (r == 0) |
| 244 |
break; |
| 245 |
|
| 246 |
if (ignore_proc(&pidref, sig == SIGKILL && !in_initrd())) |
| 247 |
continue; |
| 248 |
|
| 249 |
if (sig == SIGKILL) { |
| 250 |
_cleanup_free_ char *s = NULL; |
| 251 |
|
| 252 |
(void) pidref_get_comm(&pidref, &s); |
| 253 |
log_notice("Sending SIGKILL to PID "PID_FMT" (%s).", pidref.pid, strna(s)); |
| 254 |
} |
| 255 |
|
| 256 |
r = pidref_kill(&pidref, sig); |
| 257 |
if (r < 0) { |
| 258 |
if (r != -ESRCH) |
| 259 |
log_warning_errno(errno, "Could not kill " PID_FMT ", ignoring: %m", pidref.pid); |
| 260 |
} else { |
| 261 |
n_killed++; |
| 262 |
if (pids) { |
| 263 |
r = set_put(pids, PID_TO_PTR(pidref.pid)); |
| 264 |
if (r < 0) |
| 265 |
(void) log_oom_warning(); |
| 266 |
} |
| 267 |
} |
| 268 |
|
| 269 |
if (send_sighup) { |
| 270 |
/* Optionally, also send a SIGHUP signal, but only if the process has a controlling |
| 271 |
* tty. This is useful to allow handling of shells which ignore SIGTERM but react to |
| 272 |
* SIGHUP. We do not send this to processes that have no controlling TTY since we |
| 273 |
* don't want to trigger reloads of daemon processes. Also we make sure to only send |
| 274 |
* this after SIGTERM so that SIGTERM is always first in the queue. */ |
| 275 |
|
| 276 |
if (get_ctty_devnr(pidref.pid, NULL) >= 0) |
| 277 |
/* it's OK if the process is gone, just ignore the result */ |
| 278 |
(void) pidref_kill(&pidref, SIGHUP); |
| 279 |
} |
| 280 |
} |
| 281 |
|
| 282 |
return n_killed; |
| 283 |
} |
| 284 |
|
| 285 |
int broadcast_signal(int sig, bool wait_for_exit, bool send_sighup, usec_t timeout) { |
| 286 |
int n_children_left; |
| 287 |
sigset_t mask, oldmask; |
| 288 |
_cleanup_set_free_ Set *pids = NULL; |
| 289 |
|
| 290 |
/* Send the specified signal to all remaining processes, if not excluded by ignore_proc(). |
| 291 |
* Return: |
| 292 |
* - The number of processes still "alive" after the timeout (that should have been killed) |
| 293 |
* if the function needs to wait for the end of the processes (wait_for_exit). |
| 294 |
* - Otherwise, the number of processes to which the specified signal was sent */ |
| 295 |
|
| 296 |
if (wait_for_exit) |
| 297 |
pids = set_new(NULL); |
| 298 |
|
| 299 |
assert_se(sigemptyset(&mask) == 0); |
| 300 |
assert_se(sigaddset(&mask, SIGCHLD) == 0); |
| 301 |
assert_se(sigprocmask(SIG_BLOCK, &mask, &oldmask) == 0); |
| 302 |
|
| 303 |
if (kill(-1, SIGSTOP) < 0 && errno != ESRCH) |
| 304 |
log_warning_errno(errno, "kill(-1, SIGSTOP) failed: %m"); |
| 305 |
|
| 306 |
n_children_left = killall(sig, pids, send_sighup); |
| 307 |
|
| 308 |
if (kill(-1, SIGCONT) < 0 && errno != ESRCH) |
| 309 |
log_warning_errno(errno, "kill(-1, SIGCONT) failed: %m"); |
| 310 |
|
| 311 |
if (wait_for_exit && n_children_left > 0) |
| 312 |
n_children_left = wait_for_children(pids, &mask, timeout); |
| 313 |
|
| 314 |
assert_se(sigprocmask(SIG_SETMASK, &oldmask, NULL) == 0); |
| 315 |
|
| 316 |
return n_children_left; |
| 317 |
} |