Re: [PATCH] arch:powerpc simple_write_to_buffer return check

On 05/02/21 4:05 am, Oliver O'Halloran wrote:

On Fri, Feb 5, 2021 at 5:17 AM Mayank Suman [off-list ref] wrote:

quoted

Signed-off-by: Mayank Suman <redacted>

commit messages aren't optional

Sorry. I will include the commit message in PATCH v2.

quoted

---
 arch/powerpc/kernel/eeh.c                    | 8 ++++----
 arch/powerpc/platforms/powernv/eeh-powernv.c | 4 ++--
 2 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/arch/powerpc/kernel/eeh.c b/arch/powerpc/kernel/eeh.c
index 813713c9120c..2dbe1558a71f 100644
--- a/arch/powerpc/kernel/eeh.c
+++ b/arch/powerpc/kernel/eeh.c

@@ -1628,8 +1628,8 @@ static ssize_t eeh_force_recover_write(struct file *filp,
        char buf[20];
        int ret;

-       ret = simple_write_to_buffer(buf, sizeof(buf), ppos, user_buf, count);
-       if (!ret)
+       ret = simple_write_to_buffer(buf, sizeof(buf)-1, ppos, user_buf, count);

We should probably be zeroing the buffer. Reading to sizeof(buf) - 1
is done in a few places to guarantee that the string is nul
terminated, but without the preceeding memset() that isn't actually
guaranteed.

Yes, the buffer should be zeroed out first. I have included memset() in Patch v2.

quoted

+       if (ret <= 0)
                return -EFAULT;

EFAULT is supposed to be returned when the user supplies a buffer to
write(2) which is outside their address space. I figured letting the
sscanf() in the next step fail if the user passes writes a zero-length
buffer and returning EINVAL made more sense. That said, the exact
semantics around zero length writes are pretty handwavy so I guess
this isn't wrong, but I don't think it's better either.

simple_write_to_buffer may return negative value on fail.
So, -EFAULT should be return in case of negative return value. 
The conditional (!ret) was not sufficient to catch negative return value.