尝试将进程添加到 cgroup v2 时 EBUSY

Question

我正在玩（内核cgroups v2）。Fedora 355.15.5-200.fc35.x86_64

我能够测试和控制器cpu，没有任何问题。但对于，或者我有麻烦了。以下是如何为控制器重现此问题的步骤列表（所有命令均以 root 用户身份执行）：cpusetpidsmemoryhugetlbiomemory

1. 确保在根 cgroup 中启用内存控制器：

   # cat cgroup.subtree_control
   cpu io memory hugetlb pids
   

2. 创建新的cgroup：

   # mkdir example
   # cd example/
   

3. 启用内存控制器：

   # echo "+memory" > cgroup.subtree_control 
   

4. 将进程添加到控制组：

   # echo $$ > cgroup.procs 
   bash: echo: write error: Device or resource busy
   

当我尝试首先将进程添加到cgroup控制器然后启用控制器时，最后一步会发生相同的错误。

我究竟做错了什么？

Answer 1

这是因为 cgroups v2“无内部进程”规则：

\n

   The "no internal processes" rule is in fact more subtle than\n   stated above.  More precisely, the rule is that a (nonroot)\n   cgroup can\'t both (1) have member processes, and (2) distribute\n   resources into child cgroups\xe2\x80\x94that is, have a nonempty\n   cgroup.subtree_control file.  Thus, it is possible for a cgroup\n   to have both member processes and child cgroups, but before\n   controllers can be enabled for that cgroup, the member processes\n   must be moved out of the cgroup (e.g., perhaps into the child\n   cgroups).\n\n   With the Linux 4.14 addition of "thread mode" (described below),\n   the "no internal processes" rule has been relaxed in some cases.\n

\n

来源：cgroups(7)

\n\n

另请参阅内核源代码：

\n

https://elixir.bootlin.com/linux/v5.19/source/kernel/cgroup/cgroup.c#L2586

\n

**\n * cgroup_migrate_vet_dst - verify whether a cgroup can be migration destination\n * @dst_cgrp: destination cgroup to test\n *\n * On the default hierarchy, except for the mixable, (possible) thread root\n * and threaded cgroups, subtree_control must be zero for migration\n * destination cgroups with tasks so that child cgroups don\'t compete\n * against tasks.\n */\nint cgroup_migrate_vet_dst(struct cgroup *dst_cgrp)\n{\n    // [..]\n\n    /* apply no-internal-process constraint */\n    if (dst_cgrp->subtree_control)\n        return -EBUSY;\n\n    return 0;\n}\n

\n\n

例如，这有效：

\n

# cd /sys/fs/cgroup\n/sys/fs/cgroup # rmdir deleteme\n/sys/fs/cgroup # mkdir deleteme\n/sys/fs/cgroup # cd deleteme \n/sys/fs/cgroup/deleteme # mkdir leaf\n/sys/fs/cgroup/deleteme # echo \'+memory\' > cgroup.subtree_control\n/sys/fs/cgroup/deleteme # echo $$ > leaf/cgroup.procs\n/sys/fs/cgroup/deleteme #\n

\n