[PATCH] fs/locks.c: Fix sys_flock() race
authorTrond Myklebust <Trond.Myklebust@netapp.com>
Fri, 31 Mar 2006 10:30:55 +0000 (02:30 -0800)
committerLinus Torvalds <torvalds@g5.osdl.org>
Fri, 31 Mar 2006 20:18:56 +0000 (12:18 -0800)
sys_flock() currently has a race which can result in a double free in the
multi-thread case.

Thread 1 Thread 2

sys_flock(file, LOCK_EX)
sys_flock(file, LOCK_UN)

If Thread 2 removes the lock from inode->i_lock before Thread 1 tests for
list_empty(&lock->fl_link) at the end of sys_flock, then both threads will
end up calling locks_free_lock for the same lock.

Fix is to make flock_lock_file() do the same as posix_lock_file(), namely
to make a copy of the request, so that the caller can always free the lock.

This also has the side-effect of fixing up a reference problem in the
lockd handling of flock.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
fs/locks.c

index 8fcfeb177a2ac89f5971ced7e2da7d2b726637f8..dda83d6cd48b17934797de3f6c9f65c52662f859 100644 (file)
@@ -726,8 +726,9 @@ EXPORT_SYMBOL(posix_locks_deadlock);
  * at the head of the list, but that's secret knowledge known only to
  * flock_lock_file and posix_lock_file.
  */
-static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
+static int flock_lock_file(struct file *filp, struct file_lock *request)
 {
+       struct file_lock *new_fl = NULL;
        struct file_lock **before;
        struct inode * inode = filp->f_dentry->d_inode;
        int error = 0;
@@ -742,17 +743,19 @@ static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
                        continue;
                if (filp != fl->fl_file)
                        continue;
-               if (new_fl->fl_type == fl->fl_type)
+               if (request->fl_type == fl->fl_type)
                        goto out;
                found = 1;
                locks_delete_lock(before);
                break;
        }
-       unlock_kernel();
 
-       if (new_fl->fl_type == F_UNLCK)
-               return 0;
+       if (request->fl_type == F_UNLCK)
+               goto out;
 
+       new_fl = locks_alloc_lock();
+       if (new_fl == NULL)
+               goto out;
        /*
         * If a higher-priority process was blocked on the old file lock,
         * give it the opportunity to lock the file.
@@ -760,26 +763,27 @@ static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
        if (found)
                cond_resched();
 
-       lock_kernel();
        for_each_lock(inode, before) {
                struct file_lock *fl = *before;
                if (IS_POSIX(fl))
                        break;
                if (IS_LEASE(fl))
                        continue;
-               if (!flock_locks_conflict(new_fl, fl))
+               if (!flock_locks_conflict(request, fl))
                        continue;
                error = -EAGAIN;
-               if (new_fl->fl_flags & FL_SLEEP) {
-                       locks_insert_block(fl, new_fl);
-               }
+               if (request->fl_flags & FL_SLEEP)
+                       locks_insert_block(fl, request);
                goto out;
        }
+       locks_copy_lock(new_fl, request);
        locks_insert_lock(&inode->i_flock, new_fl);
-       error = 0;
+       new_fl = NULL;
 
 out:
        unlock_kernel();
+       if (new_fl)
+               locks_free_lock(new_fl);
        return error;
 }
 
@@ -1560,9 +1564,7 @@ asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
                error = flock_lock_file_wait(filp, lock);
 
  out_free:
-       if (list_empty(&lock->fl_link)) {
-               locks_free_lock(lock);
-       }
+       locks_free_lock(lock);
 
  out_putf:
        fput(filp);