Re: [PATCH] lib/vsprintf: add %pT[C012] format specifier

[Tetsuo Handa]

I'm planning to convert task_struct->comm to use RCU so that they always get
consistent result. Inconsistent result (e.g. trailing '\0' byte is emitted when
printing string argument) is caused by breaking a rule that the string argument
must not change during the function when it is passed as "const char *" (e.g.
strcmp() and printf("%s")).

Although task_struct->comm is not modified so frequently, task_struct->comm
passed as a "const char *" argument can change at any moment unless we pass
task_struct->comm via get_task_comm().

Since there are a lot of current->comm readers, Andrew Morton suggested that
we might be able to omit passing the argument and I wrote a patch that adds a
second format-start character for omit passing the argument.

Kees Cook wrote:
> I'm on board with the idea of embedding comm/pid/whatever, but I
> either missed or do not understand why a second format-start character
> is being added. I think this will complicate audits and maybe trigger
> weird info leaks (imagine printing a string that was %-escaped, but
> not 0x1A-escaped?)

The second format-start character applies to only format string.
If we use a dynamically generated string that was %-escaped at run-time as a
format string, we will leak globally accessible variables. But are there such
users? Such users are rare because they are already problematic since compilers
cannot check for safety using __printf() attributes at build-time.

> Hrmpf. Yeah, on the fence about this.

But I'm fine even if we cannot agree with the second format-start character.
Since my purpose is to make reading of task_struct->comm consistent, %pT-like
extension is what I want for centralizing pointer dereferences.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/