proc: remove races from proc_id_readdir()
authorEric W. Biederman <ebiederm@xmission.com>
Thu, 29 Nov 2007 00:21:26 +0000 (16:21 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 29 Nov 2007 17:24:52 +0000 (09:24 -0800)
Oleg noticed that the call of task_pid_nr_ns() in proc_pid_readdir
is racy with respect to tasks exiting.

After a bit of examination it also appears that the call itself
is completely unnecessary.

So to fix the problem this patch modifies next_tgid() to return
both a tgid and the task struct in question.

A structure is introduced to return these values because it is
slightly cleaner and easier to optimize, and the resulting code
is a little shorter.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/proc/base.c

index a17c26859074e80f47088e4e5a44d900190ed66f..02a63ac04178e64c8f7e2a3bddee2fcadeaf8254 100644 (file)
@@ -2411,19 +2411,23 @@ out:
  * Find the first task with tgid >= tgid
  *
  */
-static struct task_struct *next_tgid(unsigned int tgid,
-               struct pid_namespace *ns)
-{
+struct tgid_iter {
+       unsigned int tgid;
        struct task_struct *task;
+};
+static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
+{
        struct pid *pid;
 
+       if (iter.task)
+               put_task_struct(iter.task);
        rcu_read_lock();
 retry:
-       task = NULL;
-       pid = find_ge_pid(tgid, ns);
+       iter.task = NULL;
+       pid = find_ge_pid(iter.tgid, ns);
        if (pid) {
-               tgid = pid_nr_ns(pid, ns) + 1;
-               task = pid_task(pid, PIDTYPE_PID);
+               iter.tgid = pid_nr_ns(pid, ns);
+               iter.task = pid_task(pid, PIDTYPE_PID);
                /* What we to know is if the pid we have find is the
                 * pid of a thread_group_leader.  Testing for task
                 * being a thread_group_leader is the obvious thing
@@ -2436,23 +2440,25 @@ retry:
                 * found doesn't happen to be a thread group leader.
                 * As we don't care in the case of readdir.
                 */
-               if (!task || !has_group_leader_pid(task))
+               if (!iter.task || !has_group_leader_pid(iter.task)) {
+                       iter.tgid += 1;
                        goto retry;
-               get_task_struct(task);
+               }
+               get_task_struct(iter.task);
        }
        rcu_read_unlock();
-       return task;
+       return iter;
 }
 
 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
 
 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
-       struct task_struct *task, int tgid)
+       struct tgid_iter iter)
 {
        char name[PROC_NUMBUF];
-       int len = snprintf(name, sizeof(name), "%d", tgid);
+       int len = snprintf(name, sizeof(name), "%d", iter.tgid);
        return proc_fill_cache(filp, dirent, filldir, name, len,
-                               proc_pid_instantiate, task, NULL);
+                               proc_pid_instantiate, iter.task, NULL);
 }
 
 /* for the /proc/ directory itself, after non-process stuff has been done */
@@ -2460,8 +2466,7 @@ int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
 {
        unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
        struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
-       struct task_struct *task;
-       int tgid;
+       struct tgid_iter iter;
        struct pid_namespace *ns;
 
        if (!reaper)
@@ -2474,14 +2479,14 @@ int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
        }
 
        ns = filp->f_dentry->d_sb->s_fs_info;
-       tgid = filp->f_pos - TGID_OFFSET;
-       for (task = next_tgid(tgid, ns);
-            task;
-            put_task_struct(task), task = next_tgid(tgid + 1, ns)) {
-               tgid = task_pid_nr_ns(task, ns);
-               filp->f_pos = tgid + TGID_OFFSET;
-               if (proc_pid_fill_cache(filp, dirent, filldir, task, tgid) < 0) {
-                       put_task_struct(task);
+       iter.task = NULL;
+       iter.tgid = filp->f_pos - TGID_OFFSET;
+       for (iter = next_tgid(ns, iter);
+            iter.task;
+            iter.tgid += 1, iter = next_tgid(ns, iter)) {
+               filp->f_pos = iter.tgid + TGID_OFFSET;
+               if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
+                       put_task_struct(iter.task);
                        goto out;
                }
        }