Re: [PATCH] TOMOYO: Add garbage collector support. (v2)

From: Tetsuo Handa
Date: Sat May 16 2009 - 08:10:17 EST


James Morris wrote:
> On Thu, 14 May 2009, Tetsuo Handa wrote:
>
> > ---
> > security/tomoyo/common.c | 456 +++++++++++++++++++-----------------
> > security/tomoyo/common.h | 134 ++++++++--
> > security/tomoyo/domain.c | 437 +++++++++++++---------------------
> > security/tomoyo/file.c | 369 +++++++++++++++--------------
> > security/tomoyo/realpath.c | 567 ++++++++++++++++++++++++++++++++++-----------
> > security/tomoyo/realpath.h | 25 +
> > security/tomoyo/tomoyo.c | 40 +--
> > security/tomoyo/tomoyo.h | 13 -
> > 8 files changed, 1190 insertions(+), 851 deletions(-)
>
> This seems like an awfully large and invasive patch just to add GC support
> which you didn't originally didn't think was necessary.
>
> I think it needs review from people outside of your project.

I see. Would somebody please review?

For reviewers help, here is my thinking.



As of #15's posting ( http://lkml.org/lkml/2009/2/5/61 ), I was not aware that
use of kmalloc(GFP_KERNEL) with a lock held is not preferable.

For example, 2.6.30-rc6/security/tomoyo/tomoyo.c has below code.

static int tomoyo_update_single_path_acl(const u8 type, const char *filename,
struct tomoyo_domain_info *
const domain, const bool is_delete)
{
...(snipped)...
/***** EXCLUSIVE SECTION START *****/
down_write(&tomoyo_domain_acl_info_list_lock);
if (is_delete)
goto delete;
list_for_each_entry(ptr, &domain->acl_info_list, list) {
...(snipped)...
error = 0;
goto out;
}
/* Not found. Append it to the tail. */
acl = tomoyo_alloc_acl_element(TOMOYO_TYPE_SINGLE_PATH_ACL);
if (!acl)
goto out;
...(snipped)...
list_add_tail(&acl->head.list, &domain->acl_info_list);
error = 0;
goto out;
delete:
...(snipped)...
out:
up_write(&tomoyo_domain_acl_info_list_lock);
/***** EXCLUSIVE SECTION END *****/
return error;
}

tomoyo_alloc_acl_element() (which calls kmalloc(GFP_KERNEL)) is called with
tomoyo_domain_acl_info_list_lock lock held. And that lock is used like

static int tomoyo_check_single_path_acl2(const struct tomoyo_domain_info *
domain,
const struct tomoyo_path_info *
filename,
const u16 perm,
const bool may_use_pattern)
{
...(snipped)...
down_read(&tomoyo_domain_acl_info_list_lock);
list_for_each_entry(ptr, &domain->acl_info_list, list) {
...(snipped)...
error = 0;
break;
}
up_read(&tomoyo_domain_acl_info_list_lock);
return error;
}

This means that when we update TOMOYO's policy (via learning mode or via
/sys/kernel/security/tomoyo/ interface) we might experience a situation where
multiple processes are sleeping for undefined duration if one process went into
sleep state at kmalloc(GFP_KERNEL) in tomoyo_alloc_acl_element() and then
another process went into sleep state at down_read() in
tomoyo_check_single_path_acl2() because of down_write().

This is not a dead lock, but is not preferable.
Away from the discussion whether TOMOYO needs GC or not,
I think TOMOYO should not block reader processes for undefined duration.



TOMOYO will not block reader processes for undefined duration if we call
tomoyo_alloc_acl_element() before down_write(), but that memory will remain
unused if that memory was not added to the list. We cannot release memory
allocated by tomoyo_alloc_acl_element() because tomoyo_alloc_acl_element()
returns requested memory from memory block obtained by kmalloc(PATH_MAX).
tomoyo_alloc_acl_element() is good for minimizing memory usage, but is not
good for releasing unused memory.

TOMOYO will not block reader processes for undefined duration if a writer
process allocates memory before down_write() and releases after up_write()
if the allocated memory was not added to the list.
Thus, I replaced tomoyo_alloc_acl_element() with kmalloc()/kfree() in this
patch. And now, the only reason TOMOYO cannot release memory is that TOMOYO
cannot determine whether memory allocated by kmalloc() is referred by
other processes or not.

Since I replaced tomoyo_alloc_acl_element() with kmalloc()/kfree()
(in order to avoid sleeping operations within a semaphore protected code),
I think TOMOYO is now likely to be able to have GC support (and I also
included the GC in this patch).
--
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/