[Drbd-dev] Only 63 characters maximum allowed for shared secret (and other string values)

[Lars Ellenberg]

On Tue, Mar 12, 2013 at 11:29:10AM +0100, Tijs Van Buggenhout wrote:
> Hi,
> 
> In the online users-guide manual (and man 8 drbdsetup) I can read the 
> following for shared-secret keyword: "The shared secret used in peer 
> authentication. May be up to 64 characters."
> 
> This seems to be inaccurate, as only 63 characters can be defined as valid 
> value for the keyword, otherwise an error is raised. 64 bytes is the buffer 
> size for the value of the keyword (drbd/linux/drbd.h:#define SHARED_SECRET_MAX 
> 64) but it needs to be null terminated, hence one character is lost..

Right.  So the documentation is off by one.

> which would make one believe SHARED_SECRET_MAX is actually the maximum length 
> allowed for shared secret (SHARED_SECRET_MAX correspons with maxlen parameter 
> of __str_field_def macro).

Maximum payload including terminating NUL.

> Also in the same file, __str_field macro is defined as:
> where NLA_NUL_STRING is introduced as nla type for the field, meaning..
> 
> user/libgenl.h: *    NLA_NUL_STRING       Maximum length of string (excluding 
> NUL)

exactly, but what is described there is the .len member of the policy struct.

> #define __array(attr_nr, attr_flag, name, nla_type, _type, maxlen,      \
>                 __get, __put, __is_signed)                              \
>         [attr_nr] = { .type = nla_type,                                 \
>                       .len = maxlen - (nla_type == NLA_NUL_STRING) },
> 
> the (max) length for the value of the field is decreased to (maxlen - 1) when 
> nla_type equals NLA_NUL_STRING.

Yep, to make the value of the .len attribute of the policy struct match,
so validate_nla() will validate it to be <= that *including* the
terminating NUL.

> Did I misinterprete the manual? What is the intended behaviour?

See linux kernel source tree,
lib/nlattr.c, validate_nla, case NLA_NUL_STRING.

It validates the payload of that nla to contain a terminating NUL,
and contain that within the first pt->len + 1 byte
in case the attrlen should happen to be larger, even,
possibly due to padding.

That "+ 1" is why there is the "- (nla_type == NLA_NUL_STRING)"
in our macro.


-- 
: Lars Ellenberg
: LINBIT | Your Way to High Availability
: DRBD/HA support and consulting http://www.linbit.com

DRBD® and LINBIT® are registered trademarks of LINBIT, Austria.