On Wed, May 9, 2018 at 12:38 PM, Willem de Bruijn <willemdebruijn.ker...@gmail.com> wrote: >>> But a crash with the same signature is still occurring, so it should >>> eventually >>> get reported again. C reproducer is here, it works on Linus' tree (commit >>> 036db8bd963): https://syzkaller.appspot.com/text?tag=ReproC&x=105b1ae7800000 >> >> This appears to be a separate issue. >> >> This reproducer requires a setsockopt SOL_SOCKET/SO_TIMESTAMPING >> to trigger the use-after-free. And the freed path also points at a >> timestamping >> skb: >> >> [ 31.963619] Freed by task 2672: >> [ 31.964006] __kasan_slab_free+0x125/0x170 >> [ 31.964509] kfree+0x8b/0x1a0 >> [ 31.964875] skb_free_head+0x6f/0xa0 >> [ 31.965314] skb_release_data+0x420/0x5a0 >> [ 31.965802] skb_release_all+0x46/0x60 >> [ 31.966260] kfree_skb+0x91/0x1c0 >> [ 31.966669] __skb_complete_tx_timestamp+0x2e9/0x3d0 >> [ 31.967273] __skb_tstamp_tx+0x3b3/0x620 >> [ 31.967774] __dev_queue_xmit+0xed5/0x1a20 >> [ 31.968300] packet_sendmsg+0x36fd/0x5400 >> [ 31.968821] sock_sendmsg+0xc0/0x100 >> [ 31.969284] ___sys_sendmsg+0x367/0x880 >> [ 31.969777] __sys_sendmmsg+0x178/0x410 >> [ 31.970267] __x64_sys_sendmmsg+0x99/0x100 >> [ 31.970789] do_syscall_64+0x9a/0x2c0 >> [ 31.971260] entry_SYSCALL_64_after_hwframe+0x44/0xa9 > > This is a rare path taken when the timestamp skb cannot be queued > onto the socket (likely because of insufficient rcvbuf). > > Somehow, freeing the timestamp skb triggers this use-after-free in > the original skb from which the timestamp was cloned. As if there > is a bug in the shared info dataref.
Indeed. The skb shared info struct is zeroed by dev_validate_header as a result of dev->hard_header_len exceeding skb->end - skb->data. Not exactly sure yet how this can happen. The hard header length space is accounted for during allocation as reserved memory. But, packet_alloc_skb does call skb_reserve(), moving skb->data effectively beyond this reserved region. It may be incorrect to pass skb->data to dev_validate_header, as that does not point to the start of the ll_header anymore. Still figuring out what the right fix is..
