[PATCH] c/r: fix race of prepare_descendant() with an ongoing fork()

Oren Laadan orenl at librato.com
Tue Sep 22 13:47:43 PDT 2009

From: Oren Laadan <orenl at librato.edu>

If prepare_descendants() is walking a tree and one of the tasks is
forking, one of two bads can happen. If the child doesn't inherit the
->ctx, it breaks the assumption that the entire subtree is prepared.
If the child inherits the ->ctx, it will have one without having taken
a reference.

This patch closed this race by explicitly getting and referencing the
->ctx for a child process should the parent have one, atomically under
the tasklist_lock.

Signed-off-by: Oren Laadan <orenl at cs.columbia.edu>
 kernel/fork.c |   11 ++++++++---
 1 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/kernel/fork.c b/kernel/fork.c
index 9f13d7b..57118e4 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -62,6 +62,7 @@
 #include <linux/fs_struct.h>
 #include <linux/magic.h>
 #include <linux/perf_counter.h>
+#include <linux/checkpoint.h>
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -1148,9 +1149,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	p->pi_state_cache = NULL;
-	p->checkpoint_ctx = NULL;
 	 * sigaltstack should be cleared when sharing the same VM
@@ -1188,6 +1186,13 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	/* Need tasklist lock for parent etc handling! */
+	/* If parent is restarting, child should be too */
+	if (unlikely(current->checkpoint_ctx)) {
+		p->checkpoint_ctx = current->checkpoint_ctx;
+		ckpt_ctx_get(p->checkpoint_ctx);
+	}
 	 * The task hasn't been attached yet, so its cpus_allowed mask will
 	 * not be changed, nor will its assigned CPU.

More information about the Containers mailing list