修改容器判断存活的方式

src/main.rs

@@ -7,7 +7,7 @@ use path::{Path, PathBuf};
 
 use nix::sched::{clone, CloneCb, CloneFlags, setns};
 use nix::sys::{signal::{kill, Signal}, wait::{waitpid, WaitPidFlag}};
-use nix::unistd::{pause, dup2, pivot_root, setgid, setgroups, sethostname, setuid, Gid, Pid, Uid, User};
+use nix::unistd::{dup2, pivot_root, setgid, setgroups, sethostname, setuid, Gid, Pid, Uid, User};
 use nix::mount::{mount, MntFlags, MsFlags, umount2};
 use uuid;
 use toml;
@@ -16,6 +16,8 @@ use clap::Parser;
 
 use error::{Result, RockerError};
 mod error;
+use network::create_network;
+mod network;
 
 static WORKSPACE: &str = "/home/rocker";
 static USER_NAME: &str = "rocker";
@@ -318,13 +320,28 @@ fn init_exec_ns(pid: i32) -> Result<()>{
     Ok(())
 }
 
-fn start(is_wait: bool, cb: CloneCb, clong_flags: CloneFlags) -> Result<i32>{
+fn create_pause() -> Result<()> {
+    fs::OpenOptions::new()
+        .write(true)
+        .create(true)
+        .read(true)
+        .open("/pause")?;
+    Ok(())
+}
+
+fn check_pause<P: AsRef<Path>>(pause_path: P) -> bool {
+    // 判断该文件还在不在
+    pause_path.as_ref().exists()
+}
+
+fn start(is_wait: bool, cb: CloneCb, clong_flags: CloneFlags, container_id: &String, container_merged_pause_path: &PathBuf) -> Result<i32>{
     match unsafe {clone(cb, STACK.as_mut_slice(), clong_flags, None)} {
         Ok(child_pid) => {
             println!("clone ok: {child_pid:?}");
             // exec之前的步骤
-
-
+            create_network(container_id, child_pid.as_raw());
+            // 删除 container_merged_pause_path 解开阻塞
+            std::fs::remove_file(container_merged_pause_path)?;
 
             // 检查是否执行exec了
             let mut cnt = 0;
@@ -368,18 +385,28 @@ fn run_container(_container_id: &String, cmd: &String, args: &RockerArgs, volume
 
     let clone_flags;
 
+    let container_work_path = Path::new(WORKSPACE).join("containers").join(&_container_id);
+    let container_upper_path = container_work_path.join("upper");
+    let container_merged_path = container_work_path.join("merged");
+    let container_merged_pause_path = container_work_path.join("merged").join("pause");
+
     let rocker_user_info =  User::from_name(USER_NAME)?.ok_or(RockerError::OtherError(format!("没找到 用户: {USER_NAME}")))?;
     let rocker_uid = rocker_user_info.uid;
     let rocker_gid = rocker_user_info.gid;
     
     let _cb = if is_exec {
-        let _cb = move || {
+        let _cb = || {
+            dbg!(1);
             let container_info = get_container_info(_container_id).unwrap();
             init_exec_ns(container_info.pid).unwrap();
             init_container_env(None).unwrap();
             init_container_user(rocker_uid, rocker_gid).unwrap();
+            dbg!(2);
+            create_pause().unwrap();
+            while container_merged_pause_path.exists() {
+                std::thread::sleep(std::time::Duration::from_millis(10));
+            }
 
-            // pause();
             let cmd_vec = cmd.split(" ").collect::<Vec<&str>>();
             let err = process::Command::new(cmd_vec[0])
                 .args(&cmd_vec[1..])
@@ -392,14 +419,11 @@ fn run_container(_container_id: &String, cmd: &String, args: &RockerArgs, volume
         Box::new(_cb) as CloneCb
     } else {
         // 初始化容器工作目录
-        let container_work_path = Path::new(WORKSPACE).join("containers").join(&_container_id);
-        let container_upper_path = container_work_path.join("upper");
-        let container_merged_path = container_work_path.join("merged");
         create_dir(&container_work_path, true)?;
         create_dir(&container_upper_path, true)?;
         create_dir(&container_merged_path, true)?;
 
-        let _cb = move || {
+        let _cb = || {
             init_container_lock(&container_work_path).unwrap();
             init_container_overlay(volume_path, &container_upper_path, &container_merged_path).unwrap();
             if let Some(custom_volume) = &args.volume {
@@ -412,7 +436,11 @@ fn run_container(_container_id: &String, cmd: &String, args: &RockerArgs, volume
             init_container_log(args.log).unwrap();
             init_container_user(rocker_uid, rocker_gid).unwrap();
 
-            // pause();
+            create_pause().unwrap();
+            while container_merged_pause_path.exists() {
+                std::thread::sleep(std::time::Duration::from_millis(10));
+            }
+
             let cmd_vec = cmd.split(" ").collect::<Vec<&str>>();
             let err = process::Command::new(cmd_vec[0])
                 .args(&cmd_vec[1..])
@@ -424,7 +452,7 @@ fn run_container(_container_id: &String, cmd: &String, args: &RockerArgs, volume
         Box::new(_cb) as CloneCb
     };
 
-    start(args.wait, _cb, clone_flags)
+    start(args.wait, _cb, clone_flags, _container_id, &container_merged_pause_path)
 }
 
 #[derive(Deserialize, Serialize, Debug, PartialEq)]
@@ -490,12 +518,11 @@ fn get_container_info(container_id: &str) -> Result<ContainerInfo> {
     let info_str = fs::read_to_string(container_info_path)?;
     let mut container_info: ContainerInfo = toml::from_str(&info_str)?;
     
-    // 判断是否正在运行, 首先得到该容器进程对应的所有的fd
-    let proc_fd_path = Path::new("/proc").join(container_info.pid.to_string()).join("fd");
-    let is_running = if let Ok(fd_dir) = fs::read_dir(proc_fd_path) {
-        fd_dir.filter_map(|p|p.ok())
-            .filter_map(|f| fs::read_link(f.path()).ok())
-            .any(|p|p == lock_path)
+    // 判断是否正在运行, 进入proc/pid/判断当前进程网络中是否有下面的设备
+    let dev_path = Path::new("/proc").join(container_info.pid.to_string()).join("net").join("dev");
+    let is_running = if let Ok(dev_text) = fs::read_to_string(dev_path){
+        let slave_veth_name = format!("ro_{container_id}_3");
+        dev_text.lines().any(|l|l.starts_with(&slave_veth_name))
     } else {
         false
     };
@@ -542,10 +569,9 @@ fn stop_container(containers_id: &str, is_remove: bool) -> Result<()> {
         let container_work_path = Path::new(WORKSPACE).join("containers").join(container_id);
         let container_merged_path = container_work_path.join("merged");
         if let Ok(container_info) = get_container_info(container_id) {
-            println!("container_merged_path: {container_merged_path:?}");
             // 正在运行中的需要 kill
             if container_info.status == ContainerStatus::RUNNING {
-                let _ = kill(Pid::from_raw(container_info.pid), Signal::SIGTERM);
+                let _ = kill(Pid::from_raw(container_info.pid), Signal::SIGKILL);
                 let pid_path = Path::new("/proc").join(container_info.pid.to_string());
                 while pid_path.exists() {
                     std::thread::sleep(std::time::Duration::from_millis(10));
@@ -605,16 +631,21 @@ fn main() -> Result<()>{
         let cmd;
         match (&args.run, &args.image, &args.restart) {
             (Some(_cmd), Some(image_name), None) => {
+                dbg!(100);
                 volume_path = extend_image(image_name)?;
+                dbg!(101);
                 container_id = uuid::Uuid::new_v4().to_string()[0..8].to_string();
+                dbg!(102);
                 cmd = _cmd.clone();
             }
             (None, None, Some(_container_id)) => {
+                dbg!(211);
                 let container_info = get_container_info(_container_id)?;
+                dbg!(210);
                 volume_path = extend_image(&container_info.image)?;
                 container_id = _container_id.clone();
                 cmd = container_info.run;
-                
+                dbg!(200);
                 args.run = Some(cmd.clone());
                 args.image = Some(container_info.image);
                 args.log = container_info.log;
@@ -624,12 +655,14 @@ fn main() -> Result<()>{
                 if container_info.env.len() > 0 {
                     args.env = Some(container_info.env);
                 }
+                dbg!(201);
                 stop_container(&container_id, false)?;
             }
             _ => {
                 unreachable!()
             }
         }
+        dbg!(3);
         match run_container(&container_id, &cmd, &args, &volume_path, false) {
             Ok(child_pid) => {
                 save_container_info(&args, &container_id, child_pid)?;
@@ -642,21 +675,6 @@ fn main() -> Result<()>{
             }
         }
     }
-    // if let (Some(cmd), Some(image_name)) = (&args.run, &args.image) {
-    //     // run
-    //     let volume_path = extend_image(image_name)?;
-    //     let container_id = uuid::Uuid::new_v4().to_string()[0..8].to_string();
-    //     match run_container(&container_id, &cmd, &args, &volume_path, false) {
-    //         Ok(child_pid) => {
-    //             save_container_info(&args, &container_id, child_pid)?;
-    //         }
-    //         Err(e) => {
-    //             // clone 之后的错误 这里已经无法捕获了
-    //             println!("run_container失败: {e}");
-    //             stop_container(&container_id, true)?;
-    //         }
-    //     }
-    // } 
     else if args.ps || args.psa {
         // --ps
         show_containers(args.psa)?

src/network.rs

@@ -1,4 +1,4 @@
-use std::{fs::OpenOptions, io::Read};
+use std::{fmt::{Display, Formatter}, fs::OpenOptions, io::{Read, Write}};
 use rand::Rng;
 
 static NETWORK_FILE: &str = "/home/rocker/network";
@@ -7,7 +7,7 @@ fn get_gateway_addr(addr: &str) -> String {
     let prefix = &addr[..addr.len() - 1];
     format!("{}0", prefix)
 }
-fn create_bridge_dev(bridge_name: &str) {
+fn create_bridge_dev(bridge_name: &str) -> bool {
     // brctl addbr br0
     let args = ["addbr", bridge_name];
     let out = std::process::Command::new("brctl")
@@ -17,10 +17,12 @@ fn create_bridge_dev(bridge_name: &str) {
     if !out.status.success() {
         println!("create_bridge_dev: {args:?}");
         println!("{:?}", out);
+        return false
     }
+    true
 }
 
-fn set_bridge_ip(bridge_name: &str, addr: &str){
+fn set_bridge_ip(bridge_name: &str, addr: &str) -> bool {
     // 设置容器内从ip: ip addr add 192.168.124.1/24 dev bridge_name
     let args = ["addr", "add", &format!("{addr}/24"), "dev", bridge_name];
     let out = std::process::Command::new("ip")
@@ -30,10 +32,12 @@ fn set_bridge_ip(bridge_name: &str, addr: &str){
     if !out.status.success() {
         println!("set_bridge_ip: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
-fn set_slave_ip(slave_veth_name: &str, addr: &str, pid: &str) {
+fn set_slave_ip(slave_veth_name: &str, addr: &str, pid: &str) -> bool {
     // nsenter -t 10050 -n -- ip addr add <ip_address>/<subnet_mask> dev veth1
     let args = ["-t", pid, "-n", "--", "ip", "addr", "add", &format!("{addr}/24"), "dev", slave_veth_name];
     let out = std::process::Command::new("nsenter")
@@ -43,10 +47,12 @@ fn set_slave_ip(slave_veth_name: &str, addr: &str, pid: &str) {
     if !out.status.success() {
         println!("set_slave_ip: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
-fn set_snat(gateway_addr: &str, bridge_name: &str) {
+fn set_snat(gateway_addr: &str, bridge_name: &str) -> bool {
     // sudo iptables -t nat -A POSTROUTING -s 172.18.0.0/24 ! -o br0 -j MASQUERADE
     let args = ["-t", "nat", "-A", "POSTROUTING", "-s", &format!("{gateway_addr}/24"), "!", "-o", bridge_name, "-j", "MASQUERADE"];
     let out = std::process::Command::new("iptables")
@@ -56,10 +62,12 @@ fn set_snat(gateway_addr: &str, bridge_name: &str) {
     if !out.status.success() {
         println!("set_snat: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
-fn set_iptables_forward() {
+fn set_iptables_forward() -> bool {
     // sudo sysctl net.ipv4.conf.all.forwarding=1
     let args = ["net.ipv4.conf.all.forwarding=1"];
     let out = std::process::Command::new("sysctl")
@@ -69,6 +77,7 @@ fn set_iptables_forward() {
     if !out.status.success() {
         println!("set_iptables_forward: {args:?}");
         println!("{:?}", out);
+        return false;
     }
     // sudo iptables -t filter -P FORWARD ACCEPT
     let args = ["-t", "filter", "-P", "FORWARD", "ACCEPT"];
@@ -79,11 +88,13 @@ fn set_iptables_forward() {
     if !out.status.success() {
         println!("set_iptables_forward: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
 
-fn set_slave_route(slave_veth_name: &str, bridge_addr: &str, pid: &str) {
+fn set_slave_route(slave_veth_name: &str, bridge_addr: &str, pid: &str) -> bool {
     // sudo nsenter -t 1851 -n -- ip route add default via 172.18.0.1 dev veth3
     let args = ["-t", pid, "-n", "--", "ip", "route", "add", "default", "via", bridge_addr, "dev", slave_veth_name];
     let out = std::process::Command::new("nsenter")
@@ -93,10 +104,12 @@ fn set_slave_route(slave_veth_name: &str, bridge_addr: &str, pid: &str) {
     if !out.status.success() {
         println!("set_slave_route: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
-fn set_net_up(bridge_name: &str, master_veth_name: &str, pid: &str, slave_veth_name: &str) {
+fn set_net_up(bridge_name: &str, master_veth_name: &str, pid: &str, slave_veth_name: &str) -> bool {
     // sudo ip link set bridge_name up
     let args = ["link", "set", bridge_name, "up"];
     let out = std::process::Command::new("ip")
@@ -106,6 +119,7 @@ fn set_net_up(bridge_name: &str, master_veth_name: &str, pid: &str, slave_veth_n
     if !out.status.success() {
         println!("set_up: {args:?}");
         println!("{:?}", out);
+        return false;
     }
 
     // sudo ip link set master_veth_name up
@@ -117,6 +131,7 @@ fn set_net_up(bridge_name: &str, master_veth_name: &str, pid: &str, slave_veth_n
     if !out.status.success() {
         println!("set_up: {args:?}");
         println!("{:?}", out);
+        return false;
     }
 
     // nsenter -t 1970  -n -- ip link set dev slave_veth_name up
@@ -128,6 +143,7 @@ fn set_net_up(bridge_name: &str, master_veth_name: &str, pid: &str, slave_veth_n
     if !out.status.success() {
         println!("set_up: {args:?}");
         println!("{:?}", out);
+        return false;
     }
 
     // ip link set lo up
@@ -139,7 +155,9 @@ fn set_net_up(bridge_name: &str, master_veth_name: &str, pid: &str, slave_veth_n
     if !out.status.success() {
         println!("set_up: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
 fn set_up(dev_name: &str) {
@@ -211,7 +229,7 @@ fn del_bridge(bridge_name: &str) {
 }
 
 
-fn create_veth_pair(master_veth_name: &str, slave_veth_name: &str) {
+fn create_veth_pair(master_veth_name: &str, slave_veth_name: &str) -> bool {
     // sudo ip link add master_veth_name type veth peer name slave_veth_name
     let args = ["link", "add", master_veth_name, "type", "veth", "peer", "name", slave_veth_name];
     let out = std::process::Command::new("ip")
@@ -221,7 +239,9 @@ fn create_veth_pair(master_veth_name: &str, slave_veth_name: &str) {
     if !out.status.success() {
         println!("create_veth_pair: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    return true;
 }
 
 
@@ -236,6 +256,12 @@ struct NetWrok {
     slave_veth_name: String,   // ro_uuid[:8]_3  // ip 就是 192.168.124.3, 插在容器中
 }
 
+
+impl Display for NetWrok {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{},{},{},{},{},{}", self.gateway_addr, self.bridge_addr, self.slave_addr, self.bridge_name, self.master_veth_name, self.slave_veth_name)
+    }
+}
 fn get_all_network() -> Vec<NetWrok> {
     // 打开 NETWORK_FILE 所有已经配置的信息, 如果没有则创建
     let mut f = OpenOptions::new()
@@ -246,10 +272,24 @@ fn get_all_network() -> Vec<NetWrok> {
         .unwrap();
     let mut text = String::new();
     f.read_to_string(&mut text).unwrap();
-    vec![]
+
+    let mut all_network = vec![];
+    for line in text.lines() {
+        let network_info_vec = line.split(",").collect::<Vec<&str>>();
+        all_network.push(NetWrok {
+            gateway_addr: network_info_vec[0].to_string(),
+            bridge_addr: network_info_vec[1].to_string(),
+            slave_addr: network_info_vec[2].to_string(),
+            bridge_name: network_info_vec[3].to_string(),
+            master_veth_name: network_info_vec[4].to_string(),
+            slave_veth_name: network_info_vec[5].to_string(),
+        })
+
+    }
+    all_network
 }
 
-fn master_join_bridge(bridge_name: &str, master_veth_name: &str) {
+fn master_join_bridge(bridge_name: &str, master_veth_name: &str) -> bool {
     // 将主端链接到桥
     //sudo brctl addif bridge_name master_veth_name
     let args = ["addif", bridge_name, master_veth_name];
@@ -260,10 +300,23 @@ fn master_join_bridge(bridge_name: &str, master_veth_name: &str) {
     if !out.status.success() {
         println!("master_join_bridge: {args:?}");
         println!("{:?}", out);
+        return false;
+    }
+    true
+}
+
+fn write_network_info(all_network: Vec<NetWrok>) {
+    let mut f = OpenOptions::new()
+        .write(true)
+        .create(true)
+        .open(NETWORK_FILE)
+        .unwrap();
+    for n in all_network {
+        writeln!(f, "{}", n).unwrap();
     }
 }
 
-fn slave_join_bridge(slave_veth_name: &str, pid: &str) {
+fn slave_join_bridge(slave_veth_name: &str, pid: &str) -> bool {
     // 将从端 链接到容器
     // sudo ip link set slave_veth_name netns 1234
     let args = ["link", "set", slave_veth_name, "netns", pid];
@@ -274,10 +327,12 @@ fn slave_join_bridge(slave_veth_name: &str, pid: &str) {
     if !out.status.success() {
         println!("slave_join_bridge: {args:?}");
         println!("{:?}", out);
+        return false;
     }
+    true
 }
 
-fn create_network(uuid_name: &str, pid: i32) {
+pub fn create_network(uuid_name: &str, pid: i32) -> bool {
     // 创建一个随机地址段的 没分配过的ip
     let mut all_network = get_all_network();
     let mut rg = rand::thread_rng();
@@ -305,34 +360,40 @@ fn create_network(uuid_name: &str, pid: i32) {
         }
     };
 
+    println!("{network}");
     all_network.push(network.clone());
     // 写入到文件中
-    let line = format!("{},{},{},{},{},{}", network.gateway_addr, network.bridge_addr, network.slave_addr, network.bridge_name, network.master_veth_name, network.slave_veth_name);
-    println!("{line:?}");
+    write_network_info(all_network);
+    
 
     // 系统中创建桥
-    create_bridge_dev(&network.bridge_name);
+    if create_bridge_dev(&network.bridge_name) &&
     // 给桥分配ip
-    set_bridge_ip(&network.bridge_name, &network.bridge_addr);
+    set_bridge_ip(&network.bridge_name, &network.bridge_addr) &&
     // 创建veth pair
-    create_veth_pair(&network.master_veth_name, &network.slave_veth_name);
+    create_veth_pair(&network.master_veth_name, &network.slave_veth_name) &&
     // 宿主机 主pair 连接桥
-    master_join_bridge(&network.bridge_name, &network.master_veth_name);
+    master_join_bridge(&network.bridge_name, &network.master_veth_name) &&
     // 把从 pair 添加到容器内
-    slave_join_bridge(&network.slave_veth_name, pid.to_string().as_str());
+    slave_join_bridge(&network.slave_veth_name, pid.to_string().as_str()) &&
     // 给容器内 pair 分配ip
-    set_slave_ip(&network.slave_veth_name, &network.slave_addr, pid.to_string().as_str());
+    set_slave_ip(&network.slave_veth_name, &network.slave_addr, pid.to_string().as_str()) &&
     // 激活 上线
-    set_net_up(&network.bridge_name, &network.master_veth_name, pid.to_string().as_str(), &&network.slave_veth_name);
+    set_net_up(&network.bridge_name, &network.master_veth_name, pid.to_string().as_str(), &&network.slave_veth_name) &&
     // 访问公网设置
     // 配置路由
-    set_slave_route(&network.slave_veth_name, &network.bridge_addr, pid.to_string().as_str());
+    set_slave_route(&network.slave_veth_name, &network.bridge_addr, pid.to_string().as_str()) &&
     // 设置转发规则
-    set_iptables_forward();
+    set_iptables_forward() &&
     // 设置snat
-    set_snat(&network.gateway_addr, &network.bridge_name);
-
-    // 如果出错, 清理资源
+    set_snat(&network.gateway_addr, &network.bridge_name) {
+        println!("create_network success");
+        true
+    } else {
+        println!("create_network failed");
+        remove_network(&uuid_name);
+        false
+    }
 }
 
 
@@ -351,10 +412,8 @@ fn remove_network(uuid_name: &str,) {
 }
 
 fn main(){
-    // get_all_network();
-
     let container_id = uuid::Uuid::new_v4().to_string()[0..8].to_string();
-    create_network(&container_id, 15992);
+    create_network(&container_id, 17543);
     return;
     
     let addr = "192.168.124.1";