seq_buf: Make seq_buf_puts() null-terminate the buffer
authorMichael Ellerman <mpe@ellerman.id.au>
Fri, 19 Oct 2018 04:21:08 +0000 (15:21 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 12 Feb 2019 18:46:08 +0000 (19:46 +0100)
commit4a38ed76fbeca874c20914e0515f08c8fc7ee0af
tree31d3566be095c2599fde4b076f5a1b8e870c9210
parent1812be7e56c7c4447c2facdc52d81a872158a957
seq_buf: Make seq_buf_puts() null-terminate the buffer

[ Upstream commit 0464ed24380905d640030d368cd84a4e4d1e15e2 ]

Currently seq_buf_puts() will happily create a non null-terminated
string for you in the buffer. This is particularly dangerous if the
buffer is on the stack.

For example:

  char buf[8];
  char secret = "secret";
  struct seq_buf s;

  seq_buf_init(&s, buf, sizeof(buf));
  seq_buf_puts(&s, "foo");
  printk("Message is %s\n", buf);

Can result in:

  Message is fooªªªªªsecret

We could require all users to memset() their buffer to zero before
use. But that seems likely to be forgotten and lead to bugs.

Instead we can change seq_buf_puts() to always leave the buffer in a
null-terminated state.

The only downside is that this makes the buffer 1 character smaller
for seq_buf_puts(), but that seems like a good trade off.

Link: http://lkml.kernel.org/r/20181019042109.8064-1-mpe@ellerman.id.au
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
lib/seq_buf.c