proc_flush_task: flush /proc/tid/task/pid when a sub-thread exits
authorOleg Nesterov <oleg@redhat.com>
Tue, 22 Sep 2009 23:45:34 +0000 (16:45 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 23 Sep 2009 14:39:40 +0000 (07:39 -0700)
The exiting sub-thread flushes /proc/pid only, but this doesn't buy too
much: ps and friends mostly use /proc/tid/task/pid.

Remove "if (thread_group_leader())" checks from proc_flush_task() path,
this means we always remove /proc/tid/task/pid dentry on exit, and this
actually matches the comment above proc_flush_task().

The test-case:

static void* tfunc(void *arg)
{
char name[256];

sprintf(name, "/proc/%d/task/%ld/status", getpid(), gettid());
close(open(name, O_RDONLY));

return NULL;
}

int main(void)
{
pthread_t t;

for (;;) {
if (!pthread_create(&t, NULL, &tfunc, NULL))
pthread_join(t, NULL);
}
}

slabtop shows that pid/proc_inode_cache/etc grow quickly and
"indefinitely" until the task is killed or shrink_slab() is called, not
good.  And the main thread needs a lot of time to exit.

The same can happen if something like "ps -efL" runs continuously, while
some application spawns short-living threads.

Reported-by: "James M. Leddy" <jleddy@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Dominic Duval <dduval@redhat.com>
Cc: Frank Hirtz <fhirtz@redhat.com>
Cc: "Fuller, Johnray" <Johnray.Fuller@gs.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Paul Batkowski <pbatkowski@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/base.c

index 69bb70351b9b0ef3118ed62c19a3b6a69202b350..5bc587049b372d99254ff44e710f7445bfa70cd3 100644 (file)
@@ -2604,9 +2604,6 @@ static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
                dput(dentry);
        }
 
-       if (tgid == 0)
-               goto out;
-
        name.name = buf;
        name.len = snprintf(buf, sizeof(buf), "%d", tgid);
        leader = d_hash_and_lookup(mnt->mnt_root, &name);
@@ -2663,17 +2660,16 @@ out:
 void proc_flush_task(struct task_struct *task)
 {
        int i;
-       struct pid *pid, *tgid = NULL;
+       struct pid *pid, *tgid;
        struct upid *upid;
 
        pid = task_pid(task);
-       if (thread_group_leader(task))
-               tgid = task_tgid(task);
+       tgid = task_tgid(task);
 
        for (i = 0; i <= pid->level; i++) {
                upid = &pid->numbers[i];
                proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
-                       tgid ? tgid->numbers[i].nr : 0);
+                                       tgid->numbers[i].nr);
        }
 
        upid = &pid->numbers[pid->level];