lmkd: Use process_mrelease to reap the target process from a thread am: 7c3addb2a1 am: c4067fc9f2

Original change: https://android-review.googlesource.com/c/platform/system/memory/lmkd/+/1904583

Change-Id: I224057073a3fd0c569eb5803eb2c56710b01fc77

Android.bp

@@ -12,7 +12,10 @@ cc_defaults {
 cc_binary {
     name: "lmkd",
 
-    srcs: ["lmkd.cpp"],
+    srcs: [
+        "lmkd.cpp",
+        "reaper.cpp",
+    ],
     shared_libs: [
         "libcutils",
         "liblog",

lmkd.cpp

@@ -16,12 +16,10 @@
 
 #define LOG_TAG "lowmemorykiller"
 
-#include <dirent.h>
 #include <errno.h>
 #include <inttypes.h>
 #include <pwd.h>
 #include <sched.h>
-#include <signal.h>
 #include <stdbool.h>
 #include <stdlib.h>
 #include <string.h>
@@ -30,12 +28,9 @@
 #include <sys/eventfd.h>
 #include <sys/mman.h>
 #include <sys/pidfd.h>
-#include <sys/resource.h>
 #include <sys/socket.h>
 #include <sys/syscall.h>
 #include <sys/sysinfo.h>
-#include <sys/time.h>
-#include <sys/types.h>
 #include <time.h>
 #include <unistd.h>
 
@@ -47,10 +42,9 @@
 #include <log/log_event_list.h>
 #include <log/log_time.h>
 #include <private/android_filesystem_config.h>
-#include <processgroup/processgroup.h>
 #include <psi/psi.h>
-#include <system/thread_defs.h>
 
+#include "reaper.h"
 #include "statslog.h"
 
 #define BPF_FD_JUST_USE_INT
@@ -228,6 +222,8 @@ static struct psi_threshold psi_thresholds[VMPRESS_LEVEL_COUNT] = {
 };
 
 static android_log_context ctx;
+static Reaper reaper;
+static int reaper_comm_fd[2];
 
 enum polling_update {
     POLLING_DO_NOT_CHANGE,
@@ -277,9 +273,9 @@ static struct event_handler_info vmpressure_hinfo[VMPRESS_LEVEL_COUNT];
 
 /*
  * 1 ctrl listen socket, 3 ctrl data socket, 3 memory pressure levels,
- * 1 lmk events + 1 fd to wait for process death
+ * 1 lmk events + 1 fd to wait for process death + 1 fd to receive kill failure notifications
  */
-#define MAX_EPOLL_EVENTS (1 + MAX_DATA_CONN + VMPRESS_LEVEL_COUNT + 1 + 1)
+#define MAX_EPOLL_EVENTS (1 + MAX_DATA_CONN + VMPRESS_LEVEL_COUNT + 1 + 1 + 1)
 static int epollfd;
 static int maxevents;
 
@@ -2008,42 +2004,6 @@ static struct proc *proc_get_heaviest(int oomadj) {
     return maxprocp;
 }
 
-static void set_process_group_and_prio(int pid, const std::vector<std::string>& profiles,
-                                       int prio) {
-    DIR* d;
-    char proc_path[PATH_MAX];
-    struct dirent* de;
-
-    snprintf(proc_path, sizeof(proc_path), "/proc/%d/task", pid);
-    if (!(d = opendir(proc_path))) {
-        ALOGW("Failed to open %s; errno=%d: process pid(%d) might have died", proc_path, errno,
-              pid);
-        return;
-    }
-
-    while ((de = readdir(d))) {
-        int t_pid;
-
-        if (de->d_name[0] == '.') continue;
-        t_pid = atoi(de->d_name);
-
-        if (!t_pid) {
-            ALOGW("Failed to get t_pid for '%s' of pid(%d)", de->d_name, pid);
-            continue;
-        }
-
-        if (setpriority(PRIO_PROCESS, t_pid, prio) && errno != ESRCH) {
-            ALOGW("Unable to raise priority of killing t_pid (%d): errno=%d", t_pid, errno);
-        }
-
-        if (!SetTaskProfiles(t_pid, profiles, true)) {
-            ALOGW("Failed to set task_profiles on pid(%d) t_pid(%d)", pid, t_pid);
-            continue;
-        }
-    }
-    closedir(d);
-}
-
 static bool is_kill_pending(void) {
     char buf[24];
 
@@ -2114,6 +2074,19 @@ static void kill_done_handler(int data __unused, uint32_t events __unused,
     poll_params->update = POLLING_RESUME;
 }
 
+static void kill_fail_handler(int data __unused, uint32_t events __unused,
+                              struct polling_params *poll_params) {
+    int pid;
+
+    // Extract pid from the communication pipe. Clearing the pipe this way allows further
+    // epoll_wait calls to sleep until the next event.
+    if (TEMP_FAILURE_RETRY(read(reaper_comm_fd[0], &pid, sizeof(pid))) != sizeof(pid)) {
+        ALOGE("thread communication read failed: %s", strerror(errno));
+    }
+    stop_wait_for_proc_kill(false);
+    poll_params->update = POLLING_RESUME;
+}
+
 static void start_wait_for_proc_kill(int pid_or_fd) {
     static struct event_handler_info kill_done_hinfo = { 0, kill_done_handler };
     struct epoll_event epev;
@@ -2149,7 +2122,7 @@ static int kill_one_process(struct proc* procp, int min_oom_score, struct kill_i
     int pidfd = procp->pidfd;
     uid_t uid = procp->uid;
     char *taskname;
-    int r;
+    int kill_result;
     int result = -1;
     struct memory_stat *mem_st;
     struct kill_stat kill_st;
@@ -2188,29 +2161,21 @@ static int kill_one_process(struct proc* procp, int min_oom_score, struct kill_i
 
     snprintf(desc, sizeof(desc), "lmk,%d,%d,%d,%d,%d", pid, ki ? (int)ki->kill_reason : -1,
              procp->oomadj, min_oom_score, ki ? ki->max_thrashing : -1);
+
     trace_kill_start(pid, desc);
 
-    /* CAP_KILL required */
+    start_wait_for_proc_kill(pidfd < 0 ? pid : pidfd);
-    if (pidfd < 0) {
+    kill_result = reaper.kill({ pidfd, pid });
-        start_wait_for_proc_kill(pid);
-        r = kill(pid, SIGKILL);
-    } else {
-        start_wait_for_proc_kill(pidfd);
-        r = pidfd_send_signal(pidfd, SIGKILL, NULL, 0);
-    }
 
     trace_kill_end();
 
-    if (r) {
+    if (kill_result) {
         stop_wait_for_proc_kill(false);
         ALOGE("kill(%d): errno=%d", pid, errno);
         /* Delete process record even when we fail to kill so that we don't get stuck on it */
         goto out;
     }
 
-    set_process_group_and_prio(pid, {"CPUSET_SP_FOREGROUND", "SCHED_SP_FOREGROUND"},
-                               ANDROID_PRIORITY_HIGHEST);
-
     last_kill_tm = *tm;
 
     inc_killcnt(procp->oomadj);
@@ -3158,6 +3123,63 @@ static void destroy_monitors() {
     }
 }
 
+static void drop_reaper_comm() {
+    close(reaper_comm_fd[0]);
+    close(reaper_comm_fd[1]);
+}
+
+static bool setup_reaper_comm() {
+    if (pipe(reaper_comm_fd)) {
+        ALOGE("pipe failed: %s", strerror(errno));
+        return false;
+    }
+
+    // Ensure main thread never blocks on read
+    int flags = fcntl(reaper_comm_fd[0], F_GETFL);
+    if (fcntl(reaper_comm_fd[0], F_SETFL, flags | O_NONBLOCK)) {
+        ALOGE("fcntl failed: %s", strerror(errno));
+        drop_reaper_comm();
+        return false;
+    }
+
+    return true;
+}
+
+static bool init_reaper() {
+    if (!reaper.is_reaping_supported()) {
+        ALOGI("Process reaping is not supported");
+        return false;
+    }
+
+    if (!setup_reaper_comm()) {
+        ALOGE("Failed to create thread communication channel");
+        return false;
+    }
+
+    // Setup epoll handler
+    struct epoll_event epev;
+    static struct event_handler_info kill_failed_hinfo = { 0, kill_fail_handler };
+    epev.events = EPOLLIN;
+    epev.data.ptr = (void *)&kill_failed_hinfo;
+    if (epoll_ctl(epollfd, EPOLL_CTL_ADD, reaper_comm_fd[0], &epev)) {
+        ALOGE("epoll_ctl failed: %s", strerror(errno));
+        drop_reaper_comm();
+        return false;
+    }
+
+    if (!reaper.init(reaper_comm_fd[1])) {
+        ALOGE("Failed to initialize reaper object");
+        if (epoll_ctl(epollfd, EPOLL_CTL_DEL, reaper_comm_fd[0], &epev)) {
+            ALOGE("epoll_ctl failed: %s", strerror(errno));
+        }
+        drop_reaper_comm();
+        return false;
+    }
+    maxevents++;
+
+    return true;
+}
+
 static int init(void) {
     static struct event_handler_info kernel_poll_hinfo = { 0, kernel_event_handler };
     struct reread_data file_data = {
@@ -3480,6 +3502,8 @@ static void update_props() {
         thrashing_limit_pct * 2));
     swap_util_max = clamp(0, 100, GET_LMK_PROPERTY(int32, "swap_util_max", 100));
     filecache_min_kb = GET_LMK_PROPERTY(int64, "filecache_min_kb", 0);
+
+    reaper.enable_debug(debug_process_killing);
 }
 
 int main(int argc, char **argv) {
@@ -3521,6 +3545,11 @@ int main(int argc, char **argv) {
             }
         }
 
+        if (init_reaper()) {
+            ALOGI("Process reaper initialized with %d threads in the pool",
+                reaper.thread_cnt());
+        }
+
         mainloop();
     }
 

reaper.cpp

@@ -0,0 +1,248 @@
+/*
+ *  Copyright 2021 Google, Inc
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#define LOG_TAG "lowmemorykiller"
+
+#include <dirent.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <log/log.h>
+#include <signal.h>
+#include <string.h>
+#include <stdlib.h>
+#include <sys/epoll.h>
+#include <sys/pidfd.h>
+#include <sys/resource.h>
+#include <sys/sysinfo.h>
+#include <sys/types.h>
+#include <time.h>
+#include <unistd.h>
+
+#include <processgroup/processgroup.h>
+#include <system/thread_defs.h>
+
+#include "reaper.h"
+
+#define NS_PER_MS (NS_PER_SEC / MS_PER_SEC)
+#define THREAD_POOL_SIZE 2
+
+#ifndef __NR_process_mrelease
+#define __NR_process_mrelease 448
+#endif
+
+static int process_mrelease(int pidfd, unsigned int flags) {
+    return syscall(__NR_process_mrelease, pidfd, flags);
+}
+
+static inline long get_time_diff_ms(struct timespec *from,
+                                    struct timespec *to) {
+    return (to->tv_sec - from->tv_sec) * (long)MS_PER_SEC +
+           (to->tv_nsec - from->tv_nsec) / (long)NS_PER_MS;
+}
+
+static void* reaper_main(void* param) {
+    Reaper *reaper = static_cast<Reaper*>(param);
+    struct timespec start_tm, end_tm;
+    struct Reaper::target_proc target;
+    pid_t tid = gettid();
+
+    // Ensure the thread does not use little cores
+    if (!SetTaskProfiles(tid, {"CPUSET_SP_FOREGROUND"}, true)) {
+        ALOGE("Failed to assign cpuset to the reaper thread");
+    }
+
+    for (;;) {
+        target = reaper->dequeue_request();
+
+        if (reaper->debug_enabled()) {
+            clock_gettime(CLOCK_MONOTONIC_COARSE, &start_tm);
+        }
+
+        if (pidfd_send_signal(target.pidfd, SIGKILL, NULL, 0)) {
+            // Inform the main thread about failure to kill
+            reaper->notify_kill_failure(target.pid);
+            goto done;
+        }
+        if (process_mrelease(target.pidfd, 0)) {
+            ALOGE("process_mrelease %d failed: %s", target.pidfd, strerror(errno));
+            goto done;
+        }
+        if (reaper->debug_enabled()) {
+            clock_gettime(CLOCK_MONOTONIC_COARSE, &end_tm);
+            ALOGI("Process %d was reaped in %ldms", target.pid,
+                  get_time_diff_ms(&start_tm, &end_tm));
+        }
+done:
+        close(target.pidfd);
+        reaper->request_complete();
+    }
+
+    return NULL;
+}
+
+bool Reaper::is_reaping_supported() {
+    static enum {
+        UNKNOWN,
+        SUPPORTED,
+        UNSUPPORTED
+    } reap_support = UNKNOWN;
+
+    if (reap_support == UNKNOWN) {
+        if (process_mrelease(-1, 0) && errno == ENOSYS) {
+            reap_support = UNSUPPORTED;
+        } else {
+            reap_support = SUPPORTED;
+        }
+    }
+    return reap_support == SUPPORTED;
+}
+
+bool Reaper::init(int comm_fd) {
+    char name[16];
+
+    if (thread_cnt_ > 0) {
+        // init should not be called multiple times
+        return false;
+    }
+
+    thread_pool_ = new pthread_t[THREAD_POOL_SIZE];
+    for (int i = 0; i < THREAD_POOL_SIZE; i++) {
+        if (pthread_create(&thread_pool_[thread_cnt_], NULL, reaper_main, this)) {
+            ALOGE("pthread_create failed: %s", strerror(errno));
+            continue;
+        }
+        snprintf(name, sizeof(name), "lmkd_reaper%d", thread_cnt_);
+        if (pthread_setname_np(thread_pool_[thread_cnt_], name)) {
+            ALOGW("pthread_setname_np failed: %s", strerror(errno));
+        }
+        thread_cnt_++;
+    }
+
+    if (!thread_cnt_) {
+        delete[] thread_pool_;
+        return false;
+    }
+
+    queue_.reserve(thread_cnt_);
+    comm_fd_ = comm_fd;
+    return true;
+}
+
+static void set_process_group_and_prio(int pid, const std::vector<std::string>& profiles,
+                                       int prio) {
+    DIR* d;
+    char proc_path[PATH_MAX];
+    struct dirent* de;
+
+    snprintf(proc_path, sizeof(proc_path), "/proc/%d/task", pid);
+    if (!(d = opendir(proc_path))) {
+        ALOGW("Failed to open %s; errno=%d: process pid(%d) might have died", proc_path, errno,
+              pid);
+        return;
+    }
+
+    while ((de = readdir(d))) {
+        int t_pid;
+
+        if (de->d_name[0] == '.') continue;
+        t_pid = atoi(de->d_name);
+
+        if (!t_pid) {
+            ALOGW("Failed to get t_pid for '%s' of pid(%d)", de->d_name, pid);
+            continue;
+        }
+
+        if (setpriority(PRIO_PROCESS, t_pid, prio) && errno != ESRCH) {
+            ALOGW("Unable to raise priority of killing t_pid (%d): errno=%d", t_pid, errno);
+        }
+
+        if (!SetTaskProfiles(t_pid, profiles, true)) {
+            ALOGW("Failed to set task_profiles on pid(%d) t_pid(%d)", pid, t_pid);
+            continue;
+        }
+    }
+    closedir(d);
+}
+
+bool Reaper::async_kill(const struct target_proc& target) {
+    if (target.pidfd == -1) {
+        return false;
+    }
+
+    if (!thread_cnt_) {
+        return false;
+    }
+
+    mutex_.lock();
+    if (active_requests_ >= thread_cnt_) {
+        mutex_.unlock();
+        return false;
+    }
+    active_requests_++;
+
+    // Duplicate pidfd instead of reusing the original one to avoid synchronization and refcounting
+    // when both reaper and main threads are using or closing the pidfd
+    queue_.push_back({ dup(target.pidfd), target.pid });
+    // Wake up a reaper thread
+    cond_.notify_one();
+    mutex_.unlock();
+
+    set_process_group_and_prio(target.pid, {"CPUSET_SP_FOREGROUND", "SCHED_SP_FOREGROUND"},
+                               ANDROID_PRIORITY_HIGHEST);
+
+    return true;
+}
+
+int Reaper::kill(const struct target_proc& target) {
+    /* CAP_KILL required */
+    if (target.pidfd < 0) {
+        return ::kill(target.pid, SIGKILL);
+    }
+
+    if (async_kill(target)) {
+        // we assume the kill will be successful and if it fails we will be notified
+        return 0;
+    }
+
+    return pidfd_send_signal(target.pidfd, SIGKILL, NULL, 0);
+}
+
+Reaper::target_proc Reaper::dequeue_request() {
+    struct target_proc target;
+    std::unique_lock<std::mutex> lock(mutex_);
+
+    while (queue_.empty()) {
+        cond_.wait(lock);
+    }
+    target = queue_.back();
+    queue_.pop_back();
+
+    return target;
+}
+
+void Reaper::request_complete() {
+    std::scoped_lock<std::mutex> lock(mutex_);
+    active_requests_--;
+}
+
+void Reaper::notify_kill_failure(int pid) {
+    std::scoped_lock<std::mutex> lock(mutex_);
+
+    ALOGE("Failed to kill process %d", pid);
+    if (TEMP_FAILURE_RETRY(write(comm_fd_, &pid, sizeof(pid))) != sizeof(pid)) {
+        ALOGE("thread communication write failed: %s", strerror(errno));
+    }
+}

reaper.h

@@ -0,0 +1,59 @@
+/*
+ *  Copyright 2021 Google, Inc
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#pragma once
+
+#include <condition_variable>
+#include <mutex>
+#include <vector>
+
+class Reaper {
+public:
+    struct target_proc {
+        int pidfd;
+        int pid;
+    };
+private:
+    // mutex_ and cond_ are used to wakeup the reaper thread.
+    std::mutex mutex_;
+    std::condition_variable cond_;
+    // mutex_ protects queue_ and active_requests_ access.
+    std::vector<struct target_proc> queue_;
+    int active_requests_;
+    // write side of the pipe to communicate kill failures with the main thread
+    int comm_fd_;
+    int thread_cnt_;
+    pthread_t* thread_pool_;
+    bool debug_enabled_;
+
+    bool async_kill(const struct target_proc& target);
+public:
+    Reaper() : active_requests_(0), thread_cnt_(0), debug_enabled_(false) {}
+
+    static bool is_reaping_supported();
+
+    bool init(int comm_fd);
+    int thread_cnt() const { return thread_cnt_; }
+    void enable_debug(bool enable) { debug_enabled_ = enable; }
+    bool debug_enabled() const { return debug_enabled_; }
+
+    // return 0 on success or error code returned by the syscall
+    int kill(const struct target_proc& target);
+    // below members are used only by reaper_main
+    target_proc dequeue_request();
+    void request_complete();
+    void notify_kill_failure(int pid);
+};