]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
[PATCH] proc: convert do_task_stat() to use lock_task_sighand()
authorOleg Nesterov <oleg@tv-sign.ru>
Mon, 2 Oct 2006 09:18:53 +0000 (02:18 -0700)
committerLinus Torvalds <torvalds@g5.osdl.org>
Mon, 2 Oct 2006 14:57:24 +0000 (07:57 -0700)
Drop tasklist_lock. ->siglock protects almost all interesting data
(including sub-threads traversal) except:

->signal->tty
protected by tty_mutex

->real_parent
the task can't be unhashed while we are holding
->siglock, so ->real_parent can change from under us
but we can safely dereference it under rcu_read_lock()

->pgrp/->session
we can get inconsistent numbers if the task does
sys_setsid/daemonize at the same time. I hope this
is acceptable.

Signed-off-by: Oleg Nesterov <oleg@tv-sign.ru>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
fs/proc/array.c

index 279fbf542c88194886b3139db167baba84b97200..35bd39dac2470bd6df2bbc51cdb9ff6f0c153e14 100644 (file)
@@ -321,7 +321,7 @@ static int do_task_stat(struct task_struct *task, char * buffer, int whole)
        sigset_t sigign, sigcatch;
        char state;
        int res;
-       pid_t ppid, pgid = -1, sid = -1;
+       pid_t ppid = 0, pgid = -1, sid = -1;
        int num_threads = 0;
        struct mm_struct *mm;
        unsigned long long start_time;
@@ -329,8 +329,8 @@ static int do_task_stat(struct task_struct *task, char * buffer, int whole)
        unsigned long  min_flt = 0,  maj_flt = 0;
        cputime_t cutime, cstime, utime, stime;
        unsigned long rsslim = 0;
-       struct task_struct *t;
        char tcomm[sizeof(task->comm)];
+       unsigned long flags;
 
        state = *get_task_state(task);
        vsize = eip = esp = 0;
@@ -348,15 +348,33 @@ static int do_task_stat(struct task_struct *task, char * buffer, int whole)
        cutime = cstime = utime = stime = cputime_zero;
 
        mutex_lock(&tty_mutex);
-       read_lock(&tasklist_lock);
-       if (task->sighand) {
-               spin_lock_irq(&task->sighand->siglock);
-               num_threads = atomic_read(&task->signal->count);
+       rcu_read_lock();
+       if (lock_task_sighand(task, &flags)) {
+               struct signal_struct *sig = task->signal;
+               struct tty_struct *tty = sig->tty;
+
+               if (tty) {
+                       /*
+                        * sig->tty is not stable, but tty_mutex
+                        * protects us from release_dev(tty)
+                        */
+                       barrier();
+                       tty_pgrp = tty->pgrp;
+                       tty_nr = new_encode_dev(tty_devnum(tty));
+               }
+
+               num_threads = atomic_read(&sig->count);
                collect_sigign_sigcatch(task, &sigign, &sigcatch);
 
+               cmin_flt = sig->cmin_flt;
+               cmaj_flt = sig->cmaj_flt;
+               cutime = sig->cutime;
+               cstime = sig->cstime;
+               rsslim = sig->rlim[RLIMIT_RSS].rlim_cur;
+
                /* add up live thread stats at the group level */
                if (whole) {
-                       t = task;
+                       struct task_struct *t = task;
                        do {
                                min_flt += t->min_flt;
                                maj_flt += t->maj_flt;
@@ -364,31 +382,20 @@ static int do_task_stat(struct task_struct *task, char * buffer, int whole)
                                stime = cputime_add(stime, t->stime);
                                t = next_thread(t);
                        } while (t != task);
-               }
 
-               spin_unlock_irq(&task->sighand->siglock);
-       }
-       if (task->signal) {
-               if (task->signal->tty) {
-                       tty_pgrp = task->signal->tty->pgrp;
-                       tty_nr = new_encode_dev(tty_devnum(task->signal->tty));
+                       min_flt += sig->min_flt;
+                       maj_flt += sig->maj_flt;
+                       utime = cputime_add(utime, sig->utime);
+                       stime = cputime_add(stime, sig->stime);
                }
+
+               sid = sig->session;
                pgid = process_group(task);
-               sid = task->signal->session;
-               cmin_flt = task->signal->cmin_flt;
-               cmaj_flt = task->signal->cmaj_flt;
-               cutime = task->signal->cutime;
-               cstime = task->signal->cstime;
-               rsslim = task->signal->rlim[RLIMIT_RSS].rlim_cur;
-               if (whole) {
-                       min_flt += task->signal->min_flt;
-                       maj_flt += task->signal->maj_flt;
-                       utime = cputime_add(utime, task->signal->utime);
-                       stime = cputime_add(stime, task->signal->stime);
-               }
+               ppid = rcu_dereference(task->real_parent)->tgid;
+
+               unlock_task_sighand(task, &flags);
        }
-       ppid = pid_alive(task) ? task->group_leader->real_parent->tgid : 0;
-       read_unlock(&tasklist_lock);
+       rcu_read_unlock();
        mutex_unlock(&tty_mutex);
 
        if (!whole || num_threads<2)