Message-ID: <202010071056.E4804235E@keescook>
Date:   Wed, 7 Oct 2020 11:11:28 -0700
From:   Kees Cook <keescook@...omium.org>
To:     Shuah Khan <skhan@...uxfoundation.org>
Cc:     corbet@....net, gregkh@...uxfoundation.org,
        linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 01/11] counters: Introduce counter_atomic* counters

On Tue, Oct 06, 2020 at 02:44:32PM -0600, Shuah Khan wrote:
> Introduce Simple atomic counters.
> 
> There are a number of atomic_t usages in the kernel where atomic_t api
> is used strictly for counting and not for managing object lifetime. In
> some cases, atomic_t might not even be needed.
> 
> The purpose of these counters is to clearly differentiate atomic_t
> counters from atomic_t usages that guard object lifetimes, hence prone
> to overflow and underflow errors. It allows tools that scan for underflow
> and overflow on atomic_t usages to detect overflow and underflows to scan
> just the cases that are prone to errors.
> 
> Simple atomic counters api provides interfaces for simple atomic counters
> that just count, and don't guard resource lifetimes. Counter will wrap
> around to 0 when it overflows and should not be used to guard resource
> lifetimes, device usage and open counts that control state changes, and
> pm states.
> 
> Using counter_atomic* to guard lifetimes could lead to use-after free
> when it overflows and undefined behavior when used to manage state
> changes and device usage/open states.
> 
> Reviewed-by: Greg Kroah-Hartman <gregkh@...uxfoundation.org>
> Signed-off-by: Shuah Khan <skhan@...uxfoundation.org>
> ---
>  Documentation/core-api/counters.rst | 103 +++++++++++++++++
>  MAINTAINERS                         |   7 ++
>  include/linux/counters.h            | 173 ++++++++++++++++++++++++++++
>  lib/Kconfig                         |  10 ++
>  lib/Makefile                        |   1 +
>  lib/test_counters.c                 | 157 +++++++++++++++++++++++++
>  6 files changed, 451 insertions(+)
>  create mode 100644 Documentation/core-api/counters.rst
>  create mode 100644 include/linux/counters.h
>  create mode 100644 lib/test_counters.c
> 
> diff --git a/Documentation/core-api/counters.rst b/Documentation/core-api/counters.rst
> new file mode 100644
> index 000000000000..ba1ce325b639
> --- /dev/null
> +++ b/Documentation/core-api/counters.rst
> @@ -0,0 +1,103 @@
> +.. SPDX-License-Identifier: GPL-2.0
> +
> +======================
> +Simple atomic counters
> +======================
> +
> +:Author: Shuah Khan
> +
> +There are a number of atomic_t usages in the kernel where atomic_t api
> +is used strictly for counting and not for managing object lifetime. In
> +some cases, atomic_t might not even be needed.
> +
> +The purpose of these counters is to clearly differentiate atomic_t counters
> +from atomic_t usages that guard object lifetimes, hence prone to overflow
> +and underflow errors. It allows tools that scan for underflow and overflow
> +on atomic_t usages to detect overflow and underflows to scan just the cases
> +that are prone to errors.
> +
> +Simple atomic counters api provides interfaces for simple atomic counters
> +that just count, and don't guard resource lifetimes. Counter will wrap
> +around to 0 when it overflows and should not be used to guard resource
> +lifetimes, device usage and open counts that control state changes, and
> +pm states.
> +
> +Using counter_atomic32_* to guard lifetimes could lead to use-after free
> +when it overflows and undefined behavior when used to manage state
> +changes and device usage/open states.
> +
> +Use refcount_t interfaces for guarding resources.
> +
> +.. warning::
> +        Counter will wrap around to 0 when it overflows.
> +        Should not be used to guard resource lifetimes.
> +        Should not be used to manage device state and pm state.
> +
> +Test Counters Module and selftest
> +---------------------------------
> +
> +Please see :ref:`lib/test_counters.c <Test Counters Module>` for how to
> +use these interfaces and also test them.
> +
> +Selftest for testing:
> +:ref:`testing/selftests/lib/test_counters.sh <selftest for counters>`
> +
> +Atomic counter interfaces
> +=========================
> +
> +counter_atomic32 and counter_atomic64 types use atomic_t and atomic64_t
> +underneath to leverage atomic_t api,  providing a small subset of atomic_t
> +interfaces necessary to support simple counters. ::
> +
> +        struct counter_atomic32 { atomic_t cnt; };
> +        struct counter_atomic64 { atomic64_t cnt; };
> +
> +Please see :ref:`Documentation/core-api/atomic_ops.rst <atomic_ops>` for
> +information on the Semantics and Behavior of Atomic operations.
> +
> +.. warning::
> +        It is important to keep the ops to a very small subset to ensure
> +        that the Counter API will never be used for guarding resource
> +        lifetimes and state management.
> +
> +        inc_return() is added to support current atomic_inc_return()
> +        usages and avoid forcing the use of _inc() followed by _read().
> +
> +Initializers
> +------------
> +
> +Interfaces for initializing counters are write operations which in turn
> +invoke their ``ATOMIC_INIT() and atomic_set()`` counterparts ::
> +
> +        #define COUNTER_ATOMIC_INIT(i)    { .cnt = ATOMIC_INIT(i) }
> +        counter_atomic32_set() --> atomic_set()
> +
> +        static struct counter_atomic32 acnt = COUNTER_ATOMIC_INIT(0);
> +        counter_atomic32_set(0);
> +
> +        static struct counter_atomic64 acnt = COUNTER_ATOMIC_INIT(0);
> +        counter_atomic64_set(0);
> +
> +Increment interface
> +-------------------
> +
> +Increments counter and doesn't return the new counter value. ::
> +
> +        counter_atomic32_inc() --> atomic_inc()
> +        counter_atomic64_inc() --> atomic64_inc()
> +
> +Increment and return new counter value interface
> +------------------------------------------------
> +
> +Increments counter and returns the new counter value. ::
> +
> +        counter_atomic32_inc_return() --> atomic_inc_return()
> +        counter_atomic64_inc_return() --> atomic64_inc_return()
> +
> +Decrement interface
> +-------------------
> +
> +Decrements counter and doesn't return the new counter value. ::
> +
> +        counter_atomic32_dec() --> atomic_dec()
> +        counter_atomic64_dec() --> atomic64_dec()
> diff --git a/MAINTAINERS b/MAINTAINERS
> index 33b27e62ce19..4e82d0ffcab0 100644
> --- a/MAINTAINERS
> +++ b/MAINTAINERS
> @@ -15839,6 +15839,13 @@ S:	Maintained
>  F:	Documentation/fb/sm712fb.rst
>  F:	drivers/video/fbdev/sm712*
>  
> +SIMPLE ATOMIC and NON-ATOMIC COUNTERS
> +M:	Shuah Khan <skhan@...uxfoundation.org>
> +L:	linux-kernel@...r.kernel.org
> +S:	Maintained
> +F:	include/linux/counters.h
> +F:	lib/test_counters.c
> +
>  SIMPLE FIRMWARE INTERFACE (SFI)
>  S:	Obsolete
>  W:	http://simplefirmware.org/
> diff --git a/include/linux/counters.h b/include/linux/counters.h
> new file mode 100644
> index 000000000000..c0c26a13f768
> --- /dev/null
> +++ b/include/linux/counters.h
> @@ -0,0 +1,173 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +/*
> + * Interface for simple atomic counters that just count.
> + *
> + * Counter will wrap around to 0 when it overflows and should not be
> + * used to guard resource lifetimes, device usage and open counts that
> + * control state changes, and pm states. Using counter_atomic to guard
> + * lifetimes could lead to use-after free when it overflows and undefined
> + * behavior when used to manage state changes and device usage/open states.
> + *
> + * Use refcount_t interfaces for guarding resources.
> + *
> + * The interface provides:
> + * atomic32 & atomic64 functions:
> + *	increment and no return
> + *	increment and return value
> + *	decrement and no return
> + *	read
> + *	set
> + *
> + * counter_atomic32 unctions leverage/use atomic_t interfaces.

typo: functions

> + * counter_atomic64 functions leverage/use atomic64_t interfaces.
> + * The counter will wrap around to 0 when it overflows.
> + * These interfaces should not be used to guard resource lifetimes.
> + *
> + * Reference and API guide:
> + *	Documentation/core-api/counters.rst for more information.
> + *
> + */
> +
> +#ifndef __LINUX_COUNTERS_H
> +#define __LINUX_COUNTERS_H
> +
> +#include <linux/atomic.h>
> +
> +/**
> + * struct counter_atomic32 - Simple atomic counter
> + * @cnt: int
> + *
> + * The counter wraps around to 0, when it overflows. Should not
> + * be used to guard object lifetimes.
> + **/
> +struct counter_atomic32 {
> +	atomic_t cnt;
> +};
> +
> +#define COUNTER_ATOMIC_INIT(i)		{ .cnt = ATOMIC_INIT(i) }
> +
> +/*
> + * counter_atomic32_inc() - increment counter value
> + * @cntr: struct counter_atomic32 pointer
> + *
> + */
> +static inline void counter_atomic32_inc(struct counter_atomic32 *cntr)
> +{
> +	atomic_inc(&cntr->cnt);
> +}
> +
> +/*
> + * counter_atomic32_inc_return() - increment counter value and return it
> + * @cntr: struct counter_atomic32 pointer
> + *
> + * Return: returns the new counter value after incrementing it
> + */
> +static inline int counter_atomic32_inc_return(struct counter_atomic32 *cntr)
> +{
> +	return atomic_inc_return(&cntr->cnt);
> +}

So, there's an issue here between the types and the documentation: while
this will eventually wrap around to 0, it will first go through the
negative value space (i.e. INT_MAX + 1 == INT_MIN, INT_MIN < 0).

Current users of atomic_t should already be expecting this, but does it
make sense?

i.e. should the documentation be updated to "wraps around to negative
values", or should the counter API be updated to force the unsigned
value:

+static inline u32 counter_atomic32_inc_return(struct counter_atomic32 *cntr)
+{
+	return (u32)atomic_inc_return(&cntr->cnt);
+}

I see many forcing the return type from atomic_*{read,return}*():

$ git grep -E '\((unsigned|unsigned int|u32)\).*\batomic.*(read|return)' | wc -l
67

My instinct is to say leave it "int" and adjust documentation, which is
the least disruptive, but I am enticed by the desire to make sure a
counter doesn't "misbehave" and go negative when the usage wants it
always positive.

> +static void test_counter_atomic32_overflow(void)
> +{
> +	static struct counter_atomic32 ucnt = COUNTER_ATOMIC_INIT(0);
> +	static struct counter_atomic32 ocnt = COUNTER_ATOMIC_INIT(INT_MAX);
> +	int start_val;
> +	int end_val;
> +
> +	start_val = counter_atomic32_read(&ucnt);
> +	counter_atomic32_dec(&ucnt);
> +	end_val = counter_atomic32_read(&ucnt);

This is testing that counter operations match native int operations,
which seems fine. I wonder if hard-coded values should be added too, to
just more directly map the explicit expectations? E.g. adding a second
test with each:

	test_counter_result_print32("Test underflow (int)",
				    start_val, end_val, start_val-1);
	test_counter_result_print32("Test underflow (-1)",
				    start_val, end_val, -1);


> +
> +	start_val = counter_atomic32_read(&ocnt);
> +	end_val = counter_atomic32_inc_return(&ocnt);

and:

	test_counter_result_print32("Test overflow (int)",
				    start_val, end_val, start_val+1);
	test_counter_result_print32("Test underflow (INT_MIN)",
				    start_val, end_val, INT_MIN);


Otherwise, yes, looks great; thank you!

-- 
Kees Cook

Powered by blists - more mailing lists