perf_counter: Fix output-sharing error path
authorIngo Molnar <mingo@elte.hu>
Thu, 3 Sep 2009 16:03:00 +0000 (18:03 +0200)
committerIngo Molnar <mingo@elte.hu>
Thu, 3 Sep 2009 16:46:23 +0000 (18:46 +0200)
We forget to release the fd in the PERF_FLAG_FD_OUTPUT
error path.

Reorganize the error flow here to be a clean fall-through
logic.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/perf_counter.c

index 0aa609f69103415f31cc42604793eee63928dff9..e0d91fdf0c3cfbb417c6fd94d38f2c5700942815 100644 (file)
@@ -4316,15 +4316,15 @@ SYSCALL_DEFINE5(perf_counter_open,
        struct file *group_file = NULL;
        int fput_needed = 0;
        int fput_needed2 = 0;
-       int ret;
+       int err;
 
        /* for future expandability... */
        if (flags & ~(PERF_FLAG_FD_NO_GROUP | PERF_FLAG_FD_OUTPUT))
                return -EINVAL;
 
-       ret = perf_copy_attr(attr_uptr, &attr);
-       if (ret)
-               return ret;
+       err = perf_copy_attr(attr_uptr, &attr);
+       if (err)
+               return err;
 
        if (!attr.exclude_kernel) {
                if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN))
@@ -4348,7 +4348,7 @@ SYSCALL_DEFINE5(perf_counter_open,
         */
        group_leader = NULL;
        if (group_fd != -1 && !(flags & PERF_FLAG_FD_NO_GROUP)) {
-               ret = -EINVAL;
+               err = -EINVAL;
                group_file = fget_light(group_fd, &fput_needed);
                if (!group_file)
                        goto err_put_context;
@@ -4377,22 +4377,22 @@ SYSCALL_DEFINE5(perf_counter_open,
 
        counter = perf_counter_alloc(&attr, cpu, ctx, group_leader,
                                     NULL, GFP_KERNEL);
-       ret = PTR_ERR(counter);
+       err = PTR_ERR(counter);
        if (IS_ERR(counter))
                goto err_put_context;
 
-       ret = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0);
-       if (ret < 0)
+       err = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0);
+       if (err < 0)
                goto err_free_put_context;
 
-       counter_file = fget_light(ret, &fput_needed2);
+       counter_file = fget_light(err, &fput_needed2);
        if (!counter_file)
                goto err_free_put_context;
 
        if (flags & PERF_FLAG_FD_OUTPUT) {
-               ret = perf_counter_set_output(counter, group_fd);
-               if (ret)
-                       goto err_free_put_context;
+               err = perf_counter_set_output(counter, group_fd);
+               if (err)
+                       goto err_fput_free_put_context;
        }
 
        counter->filp = counter_file;
@@ -4408,20 +4408,20 @@ SYSCALL_DEFINE5(perf_counter_open,
        list_add_tail(&counter->owner_entry, &current->perf_counter_list);
        mutex_unlock(&current->perf_counter_mutex);
 
+err_fput_free_put_context:
        fput_light(counter_file, fput_needed2);
 
-out_fput:
-       fput_light(group_file, fput_needed);
-
-       return ret;
-
 err_free_put_context:
-       kfree(counter);
+       if (err < 0)
+               kfree(counter);
 
 err_put_context:
-       put_ctx(ctx);
+       if (err < 0)
+               put_ctx(ctx);
+
+       fput_light(group_file, fput_needed);
 
-       goto out_fput;
+       return err;
 }
 
 /*