diff mbox series

[RFC,v2] binfmt_elf: Protect mm_struct access with mmap_sem

Message ID 20190613104715.22367-1-mkoutny@suse.com (mailing list archive)
State New, archived
Headers show
Series [RFC,v2] binfmt_elf: Protect mm_struct access with mmap_sem | expand

Commit Message

Michal Koutný June 13, 2019, 10:47 a.m. UTC
find_extend_vma assumes the caller holds mmap_sem as a reader (explained
in expand_downwards()). The path when we are extending the stack VMA to
accommodate argv[] pointers happens without the lock.

I was not able to cause an mm_struct corruption but an inserted
BUG_ON(!rwsem_is_locked(&mm->mmap_sem)) in find_extend_vma could be
triggered as

    # <bigfile xargs echo
    xargs: echo: terminated by signal 11

(bigfile needs to have more than RLIMIT_STACK / sizeof(char *) rows)

Other accesses to mm_struct in exec path are protected by mmap_sem, so
conservatively, protect also this one.
Besides that, explain in comments why we omit mm_struct.arg_lock in the
exec(2) path and drop an obsolete comment about removed passed_fileno.

v2: Updated changelog

Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Matthew Wilcox <willy@infradead.org>
Reviewed-by: Cyrill Gorcunov <gorcunov@gmail.com>
Signed-off-by: Michal Koutný <mkoutny@suse.com>
---
 fs/binfmt_elf.c | 10 +++++++++-
 fs/exec.c       |  3 ++-
 2 files changed, 11 insertions(+), 2 deletions(-)
diff mbox series

Patch

diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index 8264b468f283..48e169760a9c 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -299,7 +299,11 @@  create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
 	 * Grow the stack manually; some architectures have a limit on how
 	 * far ahead a user-space access may be in order to grow the stack.
 	 */
+	if (down_read_killable(&current->mm->mmap_sem))
+		return -EINTR;
 	vma = find_extend_vma(current->mm, bprm->p);
+	up_read(&current->mm->mmap_sem);
+
 	if (!vma)
 		return -EFAULT;
 
@@ -1123,11 +1127,15 @@  static int load_elf_binary(struct linux_binprm *bprm)
 		goto out;
 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
 
+	/*
+	 * Don't take mm->arg_lock. The concurrent change might happen only
+	 * from prctl_set_mm but after de_thread we are certainly alone here.
+	 */
 	retval = create_elf_tables(bprm, &loc->elf_ex,
 			  load_addr, interp_load_addr);
 	if (retval < 0)
 		goto out;
-	/* N.B. passed_fileno might not be initialized? */
+
 	current->mm->end_code = end_code;
 	current->mm->start_code = start_code;
 	current->mm->start_data = start_data;
diff --git a/fs/exec.c b/fs/exec.c
index 89a500bb897a..d5b55c92019a 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -212,7 +212,8 @@  static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
 
 	/*
 	 * We are doing an exec().  'current' is the process
-	 * doing the exec and bprm->mm is the new process's mm.
+	 * doing the exec and bprm->mm is the new process's mm that is not
+	 * shared yet, so no synchronization on mmap_sem.
 	 */
 	ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
 			&page, NULL, NULL);