On Thu, Jul 17, 2025 at 4:40 PM Paolo Abeni [off-list ref] wrote:

On 7/15/25 10:11 AM, Pranav Tyagi wrote:

quoted

A NULL page from sg_page() in sk_msg_recvmsg() can reach
__kmap_local_page_prot() and crash the kernel. Add a check for the page
before calling copy_page_to_iter() and fail early with -EFAULT to
prevent the crash.

Interesting. I thought the sge in this case are build from the kernel, I
did not expect a null page to be possible. Can you describe in the
commit message how such bad sges are created?

quoted

Reported-by: syzbot+b18872ea9631b5dcef3b@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=b18872ea9631b5dcef3b
Fixes: 2bc793e3272a ("skmsg: Extract __tcp_bpf_recvmsg() and tcp_bpf_wait_data()")
Signed-off-by: Pranav Tyagi <redacted>

Does not apply to net. Please rebase and resend, adding the target tree
in the subj prefix and specifying a revision number.

Thanks,

Paolo

Hi Paolo,

Apologies for a delayed response. I will send a v2 of the patch with all the
required changes shortly. Also, as I am a beginner at kernel work, my initial
intention was to just fix the bug and (honestly) I did not think about
the cause of
bad sges. But, I will definitely take a deep dive and try my level best
to look for a suitable explanation for the same.

Thanks for the feedback.

Regards
Pranav Tyagi