Question

我有一个包含单个文件one.txt的目录。如果我运行ls | cat，它可以正常工作。但是，如果我尝试跟踪该管道的两端，那么我确实会看到命令和strace的输出，但是该过程并未完成。

strace ls 2> >(stdbuf -o 0 sed 's/^/command1:/') | strace cat 2> >(stdbuf -o 0 sed 's/^/command2:/')

我得到的输出是：

command2:execve("/usr/bin/cat", ["cat"], [/* 50 vars */]) = 0
command2:brk(0)                                  = 0x1938000
command2:mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7f87e5a93000
command2:access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file or directory)
<snip>
command2:open("/usr/lib/locale/locale-archive", O_RDONLY|O_CLOEXEC) = 3
command2:fstat(3, {st_mode=S_IFREG|0644, st_size=106070960, ...}) = 0
command2:mmap(NULL, 106070960, PROT_READ, MAP_PRIVATE, 3, 0) = 0x7f87def8a000
command2:close(3)                                = 0
command2:fstat(1, {st_mode=S_IFCHR|0620, st_rdev=makedev(136, 2), ...}) = 0
command2:fstat(0, {st_mode=S_IFIFO|0600, st_size=0, ...}) = 0
command2:fadvise64(0, 0, 0, POSIX_FADV_SEQUENTIAL) = -1 ESPIPE (Illegal seek)
command2:read(0, "command1:execve(\"/usr/bin/ls\", ["..., 65536) = 4985
command1:execve("/usr/bin/ls", ["ls"], [/* 50 vars */]) = 0
command1:brk(0)                                  = 0x1190000
command1:mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fae869c3000
command1:access("/etc/ld.so.preload", R_OK)      = -1 ENOENT (No such file or directory)
<snip>
command1:close(3)                                = 0
command1:fstat(1, {st_mode=S_IFIFO|0600, st_size=0, ...}) = 0
command2:write(1, "command1:close(3)               "..., 115) = 115
command2:read(0, "command1:mmap(NULL, 4096, PROT_R"..., 65536) = 160
command1:mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fae869c2000
one.txt
command1:write(1, "one.txt\n", 8)                = 8
command2:write(1, "command1:mmap(NULL, 4096, PROT_R"..., 160) = 160
command2:read(0, "command1:close(1)               "..., 65536) = 159
command1:close(1)                                = 0
command1:munmap(0x7fae869c2000, 4096)            = 0
command1:close(2)                                = 0
command2:write(1, "command1:close(1)               "..., 159) = 159
command2:read(0, "command1:exit_group(0)          "..., 65536) = 53
command1:exit_group(0)                           = ?
command2:write(1, "command1:exit_group(0)          "..., 53) = 53
command2:read(0, "command1:+++ exited with 0 +++\n", 65536) = 31
command1:+++ exited with 0 +++
command2:write(1, "command1:+++ exited with 0 +++\n", 31) = 31

，此后便挂起。 ps揭示了管道中的两个命令（此处为ls和cat）都在运行。

我正在运行Bash 4.2.46版本的RHEL7。

Answer 1

我在您的strace上放了一个strace：

$begin
S   23.1394048  -10.8890421    0.2596120
S   23.7763785   -9.4142178    0.2600115
S   24.3203694   -7.9027092    0.2596413
S   24.7669178   -6.3595161    0.2599540
S   25.1173247   -4.7917893    0.2597617
S   25.3678890   -3.2049607    0.2598870
S   25.5189950   -1.6056392    0.2599152
$end

    $input
    structre = ABC
    Condition1 = default
    Condition2 = default
    Condition3 = default
    Condition4 = default
    Condition5 = default
    Condition6 = default
    $end

它挂在strace bash -c 'strace true 2> >(cat > /dev/null)'上，表明它在等待孩子时卡住了。 wait4确认了这一点：

ps f

基于此，我的工作理论是这种效果是一个僵局，因为：

24740 pts/19 Ss 0:00 /bin/bash 24752 pts/19 S+ 0:00 \_ strace true 24753 pts/19 S+ 0:00 \_ /bin/bash 24755 pts/19 S+ 0:00 \_ cat等待所有孩子，即使不是直接产卵的孩子
Bash将流程替换作为流程的子代产生。由于进程替换已附加到stderr，因此它实际上等待父级退出。

这建议至少有两种解决方法，这两种方法似乎都可行：

strace

strace -D ls 2> >(nl) { strace ls; true; } 2> >(nl)，引用手册页，“ [作为独立的孙代而不是作为示踪对象的父级来运行]示踪剂进程”。第二个强制bash通过添加另一个在其后执行的命令来使bash再次执行-D以运行fork。

在这两种情况下，多余的分叉都意味着进程替换不会以strace的孩子身份结束，从而避免了该问题。

为什么管道上的此跟踪未完成

1 个答案: