ipc,shm: fix shm_file deletion races
authorGreg Thelen <gthelen@google.com>
Thu, 21 Nov 2013 22:32:00 +0000 (14:32 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 22 Nov 2013 00:42:27 +0000 (16:42 -0800)
When IPC_RMID races with other shm operations there's potential for
use-after-free of the shm object's associated file (shm_file).

Here's the race before this patch:

  TASK 1                     TASK 2
  ------                     ------
  shm_rmid()
    ipc_lock_object()
                             shmctl()
                             shp = shm_obtain_object_check()

    shm_destroy()
      shum_unlock()
      fput(shp->shm_file)
                             ipc_lock_object()
                             shmem_lock(shp->shm_file)
                             <OOPS>

The oops is caused because shm_destroy() calls fput() after dropping the
ipc_lock.  fput() clears the file's f_inode, f_path.dentry, and
f_path.mnt, which causes various NULL pointer references in task 2.  I
reliably see the oops in task 2 if with shmlock, shmu

This patch fixes the races by:
1) set shm_file=NULL in shm_destroy() while holding ipc_object_lock().
2) modify at risk operations to check shm_file while holding
   ipc_object_lock().

Example workloads, which each trigger oops...

Workload 1:
  while true; do
    id=$(shmget 1 4096)
    shm_rmid $id &
    shmlock $id &
    wait
  done

  The oops stack shows accessing NULL f_inode due to racing fput:
    _raw_spin_lock
    shmem_lock
    SyS_shmctl

Workload 2:
  while true; do
    id=$(shmget 1 4096)
    shmat $id 4096 &
    shm_rmid $id &
    wait
  done

  The oops stack is similar to workload 1 due to NULL f_inode:
    touch_atime
    shmem_mmap
    shm_mmap
    mmap_region
    do_mmap_pgoff
    do_shmat
    SyS_shmat

Workload 3:
  while true; do
    id=$(shmget 1 4096)
    shmlock $id
    shm_rmid $id &
    shmunlock $id &
    wait
  done

  The oops stack shows second fput tripping on an NULL f_inode.  The
  first fput() completed via from shm_destroy(), but a racing thread did
  a get_file() and queued this fput():
    locks_remove_flock
    __fput
    ____fput
    task_work_run
    do_notify_resume
    int_signal

Fixes: c2c737a0461e ("ipc,shm: shorten critical region for shmat")
Fixes: 2caacaa82a51 ("ipc,shm: shorten critical region for shmctl")
Signed-off-by: Greg Thelen <gthelen@google.com>
Cc: Davidlohr Bueso <davidlohr@hp.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Manfred Spraul <manfred@colorfullife.com>
Cc: <stable@vger.kernel.org> # 3.10.17+ 3.11.6+
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
ipc/shm.c

index d69739610fd4384323004c46782116d54db6bbd5..0bdf21c6814ef8118be3ae56196b85f11b7b3ac2 100644 (file)
--- a/ipc/shm.c
+++ b/ipc/shm.c
@@ -208,15 +208,18 @@ static void shm_open(struct vm_area_struct *vma)
  */
 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 {
+       struct file *shm_file;
+
+       shm_file = shp->shm_file;
+       shp->shm_file = NULL;
        ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
        shm_rmid(ns, shp);
        shm_unlock(shp);
-       if (!is_file_hugepages(shp->shm_file))
-               shmem_lock(shp->shm_file, 0, shp->mlock_user);
+       if (!is_file_hugepages(shm_file))
+               shmem_lock(shm_file, 0, shp->mlock_user);
        else if (shp->mlock_user)
-               user_shm_unlock(file_inode(shp->shm_file)->i_size,
-                                               shp->mlock_user);
-       fput (shp->shm_file);
+               user_shm_unlock(file_inode(shm_file)->i_size, shp->mlock_user);
+       fput(shm_file);
        ipc_rcu_putref(shp, shm_rcu_free);
 }
 
@@ -983,6 +986,13 @@ SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
                }
 
                shm_file = shp->shm_file;
+
+               /* check if shm_destroy() is tearing down shp */
+               if (shm_file == NULL) {
+                       err = -EIDRM;
+                       goto out_unlock0;
+               }
+
                if (is_file_hugepages(shm_file))
                        goto out_unlock0;
 
@@ -1101,6 +1111,14 @@ long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
                goto out_unlock;
 
        ipc_lock_object(&shp->shm_perm);
+
+       /* check if shm_destroy() is tearing down shp */
+       if (shp->shm_file == NULL) {
+               ipc_unlock_object(&shp->shm_perm);
+               err = -EIDRM;
+               goto out_unlock;
+       }
+
        path = shp->shm_file->f_path;
        path_get(&path);
        shp->shm_nattch++;