A bit of defensive programming - during development, it ocassionally
happens that a call to init_new_context is missed, resulting in
context holding a host pid of zero. When that address space is torn
down, destroy_context does a kill(0), which instantly kills the whole
UML without any errors whatsoever.
This patch add a check for pids less than 2, to also catch 1 and
negative pids.
Signed-off-by: Jeff Dike <[email protected]>
---
arch/um/kernel/skas/mmu.c | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
Index: linux-2.6-git/arch/um/kernel/skas/mmu.c
===================================================================
--- linux-2.6-git.orig/arch/um/kernel/skas/mmu.c 2007-12-12 15:06:14.000000000 -0500
+++ linux-2.6-git/arch/um/kernel/skas/mmu.c 2007-12-12 15:12:11.000000000 -0500
@@ -164,8 +164,20 @@ void destroy_context(struct mm_struct *m
if (proc_mm)
os_close_file(mmu->id.u.mm_fd);
- else
+ else {
+ /*
+ * If init_new_context wasn't called, this will be
+ * zero, resulting in a kill(0), which will result in the
+ * whole UML suddenly dying. Also, cover negative and
+ * 1 cases, since they shouldn't happen either.
+ */
+ if (mmu->id.u.pid < 2) {
+ printk(KERN_ERR "corrupt mm_context - pid = %d\n",
+ mmu->id.u.pid);
+ return;
+ }
os_kill_ptraced_process(mmu->id.u.pid, 1);
+ }
if (skas_needs_stub)
free_page(mmu->id.stack);