Linux signal 那些事儿（2）

    上一篇博文，基本算是给glibc的signal函数翻了个身。现在glibc的signal基本修正了传统的UNIX的一些弊端，我们说signal并没有我们想象的那么不堪。但是signal也有不尽人意的地方。比如信号处理期间，我们期望屏蔽某些信号，而不仅仅是屏蔽自身，这时候signal就不行了。信号既然是进程间通信IPC的一种机制，我们期望获取更多的信息，而不仅仅是signo，这时候signal/kill这个机制就基本不行了。
    上面所说的都是signal的一些毛病，但是这些都不是致命的，致命的问题在于老的signal机制的不可靠。信号分成可靠性信号和非可靠性信号，并不是说用sigaction安装，用sigqueue发送的信号就是可靠性性信号，用signal安装，kill/tkill发送的信号就是非可靠性信号。这种理解是错误的。这在Linux环境进程间通信（二）：信号（上）一文中讲的非常清楚了。
    信号值位于[SIGRTMIN,SIGRTMAX] 之间的信号，就是可靠信号，位于[SIGHUP,SIGSYS]之间信号，都是非可靠性信号，与安装函数是signal还是sigaction无关，与发送函数是kill还是sigqueue无关。

    1～31之间的所有信号都称为不可靠信号，原因就在于信号不可排队，如果kernel发现同一个信号已经有挂起信号，当前信号就会被丢弃，就好象从来没有被发送过一样，无法引起信号的传递，也无法让进程执行信号处理函数。这种实现的机理，造成了这些信号的不可靠。这正所谓：我本将心向明月，奈何明月照沟渠。
    为了解决这个问题，Linux引入了实时信号,信号值在[32~64]区间内，或者称之为可靠信号。这种信号，kernel不会ignore，哪怕已经有了好多同一个信号，kernel会把新收到信号放入queue之中，等待被传递出去。
    空口说白话，不是我们的风格，我现在用代码证明之。我参考了Linux Programming Interface 一书的例子，写了两个程序，一个是signal_receiver ,一个是signal_sender.
    先看signal_receiver的code：    

manu@manu-hacks:~/code/c/self/signal$ cat signal_receiver.c
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <signal.h>
#include <string.h>
#include <errno.h>


static int sig_cnt[NSIG];
static volatile sig_atomic_t get_SIGINT = 0;

void handler(int signo)
{
    if(signo == SIGINT)
        get_SIGINT = 1;
    else
        sig_cnt[signo]++;
}

int main(int argc,char* argv[])
{
    int i = 0;
    sigset_t blockall_mask ;
    sigset_t pending_mask ;
    sigset_t empty_mask ;
    printf("%s:PID is %ld\n",argv[0],getpid());


    for(i = 1; i < NSIG; i++)
    {
        if(i == SIGKILL || i == SIGSTOP)
            continue;

        if(signal(i,&handler) == SIG_ERR)
        {
            fprintf(stderr,"signal for signo(%d) failed (%s)\n",i,strerror(errno));
//            return -1;
        }
    }

    if(argc > 1)
    {
        int sleep_time = atoi(argv[1]);
        sigfillset(&blockall_mask);

        if(sigprocmask(SIG_SETMASK,&blockall_mask,NULL) == -1)
        {
            fprintf(stderr,"setprocmask to block all signal failed(%s)\n",strerror(errno));
            return -2;
        }

        printf("I will sleep %d second\n",sleep_time);

        sleep(sleep_time);
        if(sigpending(&pending_mask) == -1)
        {
            fprintf(stderr,"sigpending failed(%s)\n",strerror(errno));
            return -2;
        }

        for(i = 1 ; i < NSIG ; i++)
        {
            if(sigismember(&pending_mask,i))
            printf("signo(%d) :%s\n",i,strsignal(i));
        }

        sigemptyset(&empty_mask);
        if(sigprocmask(SIG_SETMASK,&empty_mask,NULL) == -1)
        {
            fprintf(stderr,"setprocmask to release all signal failed(%s)\n",strerror(errno));
            return -3;
        }

    }

    while(!get_SIGINT)
        continue ; //why not use pause ? I will explain later

    for(i = 1; i < NSIG ; i++)
    {
        if(sig_cnt[i] != 0 )
        {
            printf("%s:signal %d caught %d time%s\n",
                    argv[0],i,sig_cnt[i],(sig_cnt[i] >1)?"s":"");
        }
    }

    return 0;

}

     因为我们知道，SIGKILL和SIGSTOP这两个信号是不能够定制自己的信号处理函数的，当然也不能block，原因很简单，OS或者说root才是final boss，必须有稳定终结进程的办法。假如所有的信号，进程都能ignore，OS如何终结进程？
    这个signal_receiver会等待所有的信号，接收到某信号后，该信号的捕捉到的次数++，SIGINT会终结进程，进程退出前，会打印信号的捕捉统计。
    如果进程有参数，表示sleep时间，signal_receiver会先屏蔽所有信号（当然，SIGKILL和SIGSTOP并不能被真正屏蔽）。然后sleep 一段时间后，取消信号屏蔽。我们可以想象，在信号屏蔽期间，我们收到的信号，都会在kernel记录下来，但是并不能delivery，这种信号称之挂起信号。如果在sleep期间或者说信号屏蔽期间，我收到SIGUSR1 这个信号1次和10000次，对内核来说，都是没差别的，因为后面的9999次都会被ignore掉。SIGUSR1属于不可靠信号，位图表示有没有挂起信号，有的话，直接ignore，没有的话，则记录在kernel。
    然后我们看下，signal_sender： 

manu@manu-hacks:~/code/c/self/signal$ cat signal_sender.c
#include <stdio.h>
#include <stdlib.h>
#include <getopt.h>
#include <signal.h>
#include <string.h>
#include <errno.h>

void usage()
{
    fprintf(stderr,"USAGE:\n");
    fprintf(stderr,"--------------------------------\n");
    fprintf(stderr,"signal_sender  pid  signo  times\n");
}

int main(int argc,char* argv[])
{
    pid_t pid = -1 ;
    int signo = -1;
    int times = -1;
    int i ;


    if(argc < 4 )
    {
        usage();
        return -1;
    }

    pid = atol(argv[1]);
    signo = atoi(argv[2]);
    times = atoi(argv[3]);

    if(pid <= 0 || times < 0 || signo <1 ||signo >=64 ||signo == 32 || signo ==33)
    {
        usage();
        return -1;
    }

    printf("pid = %ld,signo = %d,times = %d\n",pid,signo,times);

    for( i = 0 ; i < times ; i++)
    {
        if(kill(pid,signo) == -1)
        {
            fprintf(stderr, "send signo(%d) to pid(%ld) failed,reason(%s)\n",signo,pid,strerror(errno));
            return -2;
        }
    }
    fprintf(stdout,"done\n");
    return 0;

}

     signal_sender需要三个参数，pid signo times，就是向拿个进程发送什么信号多少次的意思。如 signal_sender 1234 10 10000,含义是向pid=1234的 进程发送10号信号（SIGUSR1），连续发送10000次。
    有这两个进程，我们就可以实验了  。 

manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver &
[1] 23416
manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver:PID is 23416
signal for signo(32) failed (Invalid argument)
signal for signo(33) failed (Invalid argument)

manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 23416   10  10000
pid = 23416,signo = 10,times = 10000
done
manu@manu-hacks:~/code/c/self/signal$ sleep 20 ; ./signal_sender 23416   2  1
pid = 23416,signo = 2,times = 1
done
./signal_receiver:signal 10 caught 2507 times
[1]+ Done ./signal_receiver

    signal_receiver等待signal的来临，singal_sender向其发送SIGUSR1 10000次，然后sleep 20秒，确保sig_receiver处理完成。但是我们发现，其实一共才caught信号SIGUSR1  2507次，7000多次的发送都丢失了，所以我们称SIGUSR1 是非可靠信号，存在丢信号的问题。
    俗话说不怕不识货，就怕货比货 ，我们让可靠信号参战，看下效果：

manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver &
[1] 26067
./signal_receiver:PID is 26067
signal for signo(32) failed (Invalid argument)
signal for signo(33) failed (Invalid argument)
manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 26067   10  10000
pid = 26067,signo = 10,times = 10000
done
manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 26067   36  10000
pid = 26067,signo = 36,times = 10000
done
manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 26067   2  1
pid = 26067,signo = 2,times = 1
done
./signal_receiver:signal 10 caught 2879 times
./signal_receiver:signal 36 caught 10000 times
[1]+ Done ./signal_receiver

    可靠性信号36，发送10000次，signal_receiver全部收到，不可靠性信号10，共收到2879次。这个数字是不可预期的，取决于内核进程的调度。
    这个如果还不够直观，我们在比较一次，让signal_receiver先屏蔽所有信号一段时间，如30s，然后解除屏蔽。

manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver  30 &
[1] 27639
manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver:PID is 27639
signal for signo(32) failed (Invalid argument)
signal for signo(33) failed (Invalid argument)
I will sleep 30 second

manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 27639   10  10000
pid = 27639,signo = 10,times = 10000
done
manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 27639   36  10000
pid = 27639,signo = 36,times = 10000
done
manu@manu-hacks:~/code/c/self/signal$
manu@manu-hacks:~/code/c/self/signal$ signo(10) :User defined signal 1
signo(36) :Real-time signal 2

manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 27639   2 1
pid = 27639,signo = 2,times = 1
done
./signal_receiver:signal 10 caught 1 time
./signal_receiver:signal 36 caught 10000 times
[1]+ Done ./signal_receiver 30

      这个比较反差比较大，不可靠signal10 共收到1次，可靠性信号36 共caught到10000次。原因就在于sigprocmask将所有的信号都屏蔽了，造成所有的信号都不能delivery。对1～31的信号，内核发现已经有相应的挂起信号，则ignore到新来的信号。但是可靠性信号则不同，会添加队列中去，尽管已经有了相同的信号。需要注意的是，signal pending有上限，并不能无限制的发：

manu@manu-hacks:~/code/c/self/signal$ ulimit -a
core file size (blocks, -c) 0
data seg size (kbytes, -d) unlimited
scheduling priority (-e) 0
file size (blocks, -f) unlimited
pending signals (-i) 15408
max locked memory (kbytes, -l) 64
max memory size (kbytes, -m) unlimited
open files (-n) 1024
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
real-time priority (-r) 0
stack size (kbytes, -s) 8192
cpu time (seconds, -t) unlimited
max user processes (-u) 15408
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited

    我发送100万，最终会收到15408个可靠信号：

manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver  30 &
[1] 16488
manu@manu-hacks:~/code/c/self/signal$ ./signal_receiver:PID is 16488
signal for signo(32) failed (Invalid argument)
signal for signo(33) failed (Invalid argument)
I will sleep 30 second

manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 16488   36  1000000
pid = 16488,signo = 36,times = 1000000
done
manu@manu-hacks:~/code/c/self/signal$ signo(36) :Real-time signal 2

manu@manu-hacks:~/code/c/self/signal$ ./signal_sender 16488   2  1
pid = 16488,signo = 2,times = 1
done
./signal_receiver:signal 36 caught 15408 times
[1]+ Done ./signal_receiver 30

      内核是怎么做到的？
    
    上图是内核中signal相关的数据结构。其中task_struct中有sigpending类型的成员变量pending
    

struct task_struct {
    volatile long state; /* -1 unrunnable, 0 runnable, >0 stopped */
    void *stack;
    atomic_t usage;
    unsigned int flags; /* per process flags, defined below */
    unsigned int ptrace;
        ...
        ...
/* signal handlers */
    struct signal_struct *signal;
    struct sighand_struct *sighand;

    sigset_t blocked, real_blocked;
    sigset_t saved_sigmask; /* restored if set_restore_sigmask() was used */
    struct sigpending pending;

       ...
}

struct signal_struct {
      atomic_t     sigcnt;
      atomic_t     live;
      int     nr_threads;
      ...
      ...
      /* shared signal handling: */
      struct sigpending    shared_pending;
      ...
}

struct sigpending {
    struct list_head list;
    sigset_t signal;
};

#define _NSIG        64

#ifdef __i386__
# define _NSIG_BPW    32
#else
# define _NSIG_BPW    64
#endif

#define _NSIG_WORDS    (_NSIG / _NSIG_BPW)

typedef unsigned long old_sigset_t;        /* at least 32 bits */

typedef struct {
    unsigned long sig[_NSIG_WORDS];
} sigset_t;

      task_struct中的pending，和signal->shared_pending都是记录挂起信号的数据结构，读到此处，你可能会迷惑，为何有两个这样的结构。这牵扯到thread与信号的一些问题，我们此处简化，就认为是一个就好，后面讲述线程与信号关系的时候，再展开。
      
    我们看到了，kill也好，tkill也罢，最终都走到了_send_signal.当然了kill系统调用根据pid的情况会分成多个分支如pid >0 pid = 0 pid=-1;pid < 0&pid !=-1,总之了，我的图只绘制了pid >0 的分支。tkill也有类似情况。
    那么kernel是怎么做到的非可靠信号和可靠信号的的这些差别的呢？

static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
            int group, int from_ancestor_ns)
{
    struct sigpending *pending;
    struct sigqueue *q;
    int override_rlimit;
    int ret = 0, result;

    assert_spin_locked(&t->sighand->siglock);

    result = TRACE_SIGNAL_IGNORED;
    if (!prepare_signal(sig, t,
            from_ancestor_ns || (info == SEND_SIG_FORCED)))
        goto ret;

    pending = group ? &t->signal->shared_pending : &t->pending;
    /*
     * Short-circuit ignored signals and support queuing
     * exactly one non-rt signal, so that we can get more
     * detailed information about the cause of the signal.
     */
    result = TRACE_SIGNAL_ALREADY_PENDING;
    if (legacy_queue(pending, sig)) //如果是低于32的信号，并且已经在pending中出现了的信号，就直接返回了，ignore
        goto ret;

    result = TRACE_SIGNAL_DELIVERED;
    /*
     * fast-pathed signals for kernel-internal things like SIGSTOP
     * or SIGKILL.
     */
    if (info == SEND_SIG_FORCED)
        goto out_set;

    /*
     * Real-time signals must be queued if sent by sigqueue, or
     * some other real-time mechanism. It is implementation
     * defined whether kill() does so. We attempt to do so, on
     * the principle of least surprise, but since kill is not
     * allowed to fail with EAGAIN when low on memory we just
     * make sure at least one signal gets delivered and don't
     * pass on the info struct.
     */
    if (sig < SIGRTMIN)
        override_rlimit = (is_si_special(info) || info->si_code >= 0);
    else
        override_rlimit = 0;

      //分配sigqueue结构，并且链入到相应的pending。

    q = __sigqueue_alloc(sig, t, GFP_ATOMIC | __GFP_NOTRACK_FALSE_POSITIVE,
        override_rlimit);
    if (q) {
        list_add_tail(&q->list, &pending->list);
        switch ((unsigned long) info) {
        case (unsigned long) SEND_SIG_NOINFO:
            q->info.si_signo = sig;
            q->info.si_errno = 0;
            q->info.si_code = SI_USER;
            q->info.si_pid = task_tgid_nr_ns(current,
                            task_active_pid_ns(t));
            q->info.si_uid = from_kuid_munged(current_user_ns(), current_uid());
            break;
        case (unsigned long) SEND_SIG_PRIV:
            q->info.si_signo = sig;
            q->info.si_errno = 0;
            q->info.si_code = SI_KERNEL;
            q->info.si_pid = 0;
            q->info.si_uid = 0;
            break;
        default:
            copy_siginfo(&q->info, info);
            if (from_ancestor_ns)
                q->info.si_pid = 0;
            break;
        }

        userns_fixup_signal_uid(&q->info, t);

    } else if (!is_si_special(info)) {
        if (sig >= SIGRTMIN && info->si_code != SI_USER) {
            /*
             * Queue overflow, abort. We may abort if the
             * signal was rt and sent by user using something
             * other than kill().
             */
            result = TRACE_SIGNAL_OVERFLOW_FAIL;
            ret = -EAGAIN;
            goto ret;
        } else {
            /*
             * This is a silent loss of information. We still
             * send the signal, but the *info bits are lost.
             */
            result = TRACE_SIGNAL_LOSE_INFO;
        }
    }

out_set:
    signalfd_notify(t, sig);
    sigaddset(&pending->signal, sig);  //加入位图
    complete_signal(sig, t, group);
ret:
    trace_signal_generate(sig, info, t, group, result);
    return ret;
}

static inline int legacy_queue(struct sigpending *signals, int sig)
{
    return (sig < SIGRTMIN) && sigismember(&signals->signal, sig); //是不可靠信号，并且该信号已经存在挂起信号，

    那么15408的限制是在哪里呢？在__sigqueue_alloc 里面。

static struct sigqueue *
__sigqueue_alloc(int sig, struct task_struct *t, gfp_t flags, int override_rlimit)
{
    struct sigqueue *q = NULL;
    struct user_struct *user;

    /*
     * Protect access to @t credentials. This can go away when all
     * callers hold rcu read lock.
     */
    rcu_read_lock();
    user = get_uid(__task_cred(t)->user);
    atomic_inc(&user->sigpending);    //计数器+1
    rcu_read_unlock();

    if (override_rlimit ||
     atomic_read(&user->sigpending) <=
            task_rlimit(t, RLIMIT_SIGPENDING)) {
        q = kmem_cache_alloc(sigqueue_cachep, flags);
    } else {
        print_dropped_signal(sig);
    }

    if (unlikely(q == NULL)) {
        atomic_dec(&user->sigpending);
        free_uid(user);
    } else {
        INIT_LIST_HEAD(&q->list);
        q->flags = 0;
        q->user = user;
    }

    return q;
}

     我们看到，legacy_queue就是用来判断是否是非可靠信号（signo低于32），并且相同signo值已经存在在挂起信号之中，如果是，直接返回。
     而对于可靠信号，会分配一个sigqueue的结构，然后讲sigqueue链入到sigpending结构的中链表中。从而就不会丢失信号。当然对pending信号的总数作了限制，限制最多不可超过15408.当然了这个值是可以修改的：
    


参考文献：
1 Linux programming interface
2 深入理解linux内核
3 linux kernel 3.8.0 内核源码