Re: [PATCH v2 2/2] lsm: Add/fix return values in lsm_hooks.h and fix formatting

[Casey Schaufler <casey@schaufler-ca.com>]

On 12/8/2022 1:59 PM, Paul Moore wrote:
> On Thu, Dec 8, 2022 at 4:29 AM Roberto Sassu
> <roberto.sassu@huaweicloud.com> wrote:
>> On Wed, 2022-12-07 at 14:34 -0500, Paul Moore wrote:
>>> On Wed, Dec 7, 2022 at 4:18 AM Roberto Sassu
>>> <roberto.sassu@huaweicloud.com> wrote:
>>>> For this patch, I saw it is already in lsm/next. Paul, should I do an
>>>> incremental patch or change the one in the repo and you force push it?
>>>> I would just remove the three lines after the parameters description.
>>> Just send a patch against the current lsm/next branch to remove those
>>> lines, and please do it ASAP as the merge window opens this
>>> weekend/Monday.
>> Ok, was about to send but I would need a clarification first.
>>
>> In mount_api.rst, there is for security_fs_context_parse_param():
>>
>>      The value pointed to by param may be modified (if a string) or stolen
>>      (provided the value pointer is NULL'd out).  If it is stolen, 0 must be
>>      returned to prevent it being passed to the filesystem.
>>
>> Looking at security.c:
>>
>>         hlist_for_each_entry(hp, &security_hook_heads.fs_context_parse_param,
>>                              list) {
>>                 trc = hp->hook.fs_context_parse_param(fc, param);
>>                 if (trc == 0)
>>                         rc = 0;
>>                 else if (trc != -ENOPARAM)
>>                         return trc;
>>         }
>>
>> If, as mount_api.rst says, the value is modified by an LSM or stolen,
>> should it be passed to other LSMs too?
> All of the LSMs should be using fs_parse() in their
> fs_context_parse_param() hook to identify the mount options that they
> own, skipping those they do not (fs_parse() would return -ENOPARAM in
> those cases).  I don't believe we currently have any mount options
> that are shared across the different LSMs, so I believe this is a
> non-issue.

There aren't any today. SELinux and Smack are the only LSMs with
mount options. Smack mount options all begin with "smack", so it's
unlikely there is going to be a future overlap. I'd hate to do the
/proc/self/attr/current battle over again, so I recommend that any
new LSM that uses mount options be required to use an identifying
prefix. I don't see any way that using the same option name for
mounts, even if the use is the same, won't end in tears.

>
> In the future if we ever find the need to share mount options across
> different LSMs we will need some additional work to ensure it is
> handled properly, but I don't think we need to worry too much about
> that now.
>