On Wed, Sep 11, 2013 at 12:45 PM, Joe Perches <j...@perches.com> wrote: > On Wed, 2013-09-11 at 12:25 -0700, Kees Cook wrote: >> On Wed, Sep 11, 2013 at 12:09 PM, Joe Perches <j...@perches.com> wrote: >> > On Wed, 2013-09-11 at 11:19 -0700, Kees Cook wrote: >> >> On Wed, Sep 11, 2013 at 2:31 AM, Dan Carpenter <dan.carpen...@oracle.com> >> >> wrote: >> >> > On Tue, Sep 10, 2013 at 10:19:17PM -0700, Kees Cook wrote: >> >> >> In the former case, format characters will get processed by the >> >> >> sprintf logic. In the latter, they are printed as-is. In this specific >> >> >> case, if there was a way to inject strings like "ohai %n" into the >> >> >> msgbuf string, the former would actually attempt to resolve the %n. In >> >> >> the simple case, this could lead to Oopses, and in the unlucky case, >> >> >> it could allow arbitrary memory writing and execution control. >> >> >> >> >> >> http://en.wikipedia.org/wiki/Uncontrolled_format_string >> >> > >> >> > The kernel ignores %n so hopefully it can't actually write to memory. >> >> >> >> I wish! This is not the case, though. See FORMAT_TYPE_NRCHARS in >> >> lib/vsprintf.c's vsnprintf(). >> >> >> >> $ git grep '%n' | wc -l >> >> 111 >> > >> > Umm. >> > >> > See: lib/vsprintf.c >> > >> > /** >> > * vsnprintf - Format a string and place it in a buffer >> > [...] >> > * %n is ignored >> > >> > %n does work for vsscanf though. >> >> The comment is a lie: >> >> int len = 0; >> printk("len:%d\n", len); >> printk("%s%n\n", "Ohai!", &len); >> printk("len:%d\n", len); >> >> [ 0.025930] len:0 >> [ 0.026003] Ohai! >> [ 0.026261] len:5 >> >> The functionality between scanf and printf was, I think, merged in >> 2009, if I'm reading the git blame correctly. > > Yeah. > > commit fef20d9c1380f04ba9492d6463148db07b413708 > Author: Frederic Weisbecker <fweis...@gmail.com> > Date: Fri Mar 6 17:21:50 2009 +0100 > > vsprintf: unify the format decoding layer for its 3 users > > Maybe it should be reignored... > > There are a few more in net/ though that may be pretty > easy to change to use the seq_printf return value.
I would love to remove %n. It's not clear to me how to re-split it from scanf, though. I'd need to study this code a bunch more. Dropping %n from all its non-scanf uses would be a good first-step, though. -Kees > > > net/ipv4/fib_trie.c- seq_printf(seq, > net/ipv4/fib_trie.c- > "%s\t%08X\t%08X\t%04X\t%d\t%u\t" > net/ipv4/fib_trie.c: > "%d\t%08X\t%d\t%u\t%u%n", > net/ipv4/fib_trie.c- fi->fib_dev ? > fi->fib_dev->name : "*", > -- > net/ipv4/fib_trie.c- seq_printf(seq, > net/ipv4/fib_trie.c- > "*\t%08X\t%08X\t%04X\t%d\t%u\t" > net/ipv4/fib_trie.c: > "%d\t%08X\t%d\t%u\t%u%n", > net/ipv4/fib_trie.c- prefix, 0, flags, 0, > 0, 0, > -- > net/ipv4/ping.c- seq_printf(f, "%5d: %08X:%04X %08X:%04X" > net/ipv4/ping.c: " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu > %d %pK %d%n", > net/ipv4/ping.c- bucket, src, srcp, dest, destp, sp->sk_state, > -- > net/ipv4/tcp_ipv4.c- seq_printf(f, "%4d: %08X:%04X %08X:%04X" > net/ipv4/tcp_ipv4.c: " %02X %08X:%08X %02X:%08lX %08X %5u %8d %u > %d %pK%n", > net/ipv4/tcp_ipv4.c- i, > -- > net/ipv4/tcp_ipv4.c- seq_printf(f, "%4d: %08X:%04X %08X:%04X %02X > %08X:%08X %02X:%08lX " > net/ipv4/tcp_ipv4.c: "%08X %5u %8d %lu %d %pK %lu %lu %u > %u %d%n", > net/ipv4/tcp_ipv4.c- i, src, srcp, dest, destp, sk->sk_state, > -- > net/ipv4/tcp_ipv4.c- seq_printf(f, "%4d: %08X:%04X %08X:%04X" > net/ipv4/tcp_ipv4.c: " %02X %08X:%08X %02X:%08lX %08X %5d %8d %d > %d %pK%n", > net/ipv4/tcp_ipv4.c- i, src, srcp, dest, destp, tw->tw_substate, > 0, 0, > -- > net/ipv4/udp.c- seq_printf(f, "%5d: %08X:%04X %08X:%04X" > net/ipv4/udp.c: " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK > %d%n", > net/ipv4/udp.c- bucket, src, srcp, dest, destp, sp->sk_state, > -- > net/sctp/objcnt.c: seq_printf(seq, "%s: %d%n", sctp_dbg_objcnt[i].label, > net/sctp/objcnt.c- > atomic_read(sctp_dbg_objcnt[i].counter), &len); > > > -- Kees Cook Chrome OS Security _______________________________________________ devel mailing list de...@linuxdriverproject.org http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
