Re: [PATCH v4 0/3] Hardening perf subsystem

[Erick Archer]

Hi Andrew,

On Sat, Jun 01, 2024 at 06:56:15PM +0200, Erick Archer wrote:
> Hi everyone,
> 
> This is an effort to get rid of all multiplications from allocation
> functions in order to prevent integer overflows [1][2].
> 
> In the first patch, the "struct amd_uncore_ctx" can be refactored to
> use a flex array for the "events" member. This way, the allocation/
> freeing of the memory can be simplified. Then, the struct_size()
> helper can be used to do the arithmetic calculation for the memory
> to be allocated.
> 
> In the second patch, as the "struct intel_uncore_box" ends in a
> flexible array, the preferred way in the kernel is to use the
> struct_size() helper to do the arithmetic instead of the calculation
> "size + count * size" in the kzalloc_node() function.
> 
> In the third patch, as the "struct perf_buffer" also ends in a
> flexible array, the preferred way in the kernel is to use the
> struct_size() helper to do the arithmetic instead of the calculation
> "size + count * size" in the kzalloc_node() functions. At the same
> time, prepare for the coming implementation by GCC and Clang of the
> __counted_by attribute.
> 
> This time, I have decided to send these three patches in the same serie
> since all of them has been rejected by the maintainers. I have used
> the v4 tag since it is the latest iteration in one of the patches.
> 
> The reason these patches were rejected is that Peter Zijlstra detest
> the struct_size() helper [3][4]. However, Kees Cook and I consider that
> the use of this helper improves readability. But we can all say that it
> is a matter of preference.
> 
> Anyway, leaving aside personal preferences, these patches has the
> following pros:
> 
> - Code robustness improvement (__counted_by coverage).
> - Code robustness improvement (use of struct_size() to do calculations
>   on memory allocator functions).
> - Fewer lines of code.
> - Follow the recommendations made in "Deprecated Interfaces, Language
>   Features, Attributes, and Conventions" [5] as the preferred method
>   of doing things in the kernel.
> - Widely used in the kernel.
> - Widely accepted in the kernel.
> 
> There are also patches in this subsystem that use the struct_size()
> helper:
> 
> 6566f907bf31 ("Convert intel uncore to struct_size") by Matthew Wilcox
> dfbc411e0a5e ("perf/x86/rapl: Prefer struct_size() over open coded arithmetic") by me
> 
> Therefore, I would like these patches to be applied this time.

This is my last attemp to get these patches applied. I have decided to
send this mail to try to unjam this situation. I have folowed all the
reviewers comments and have no response from the maintainers other than
"I detest the struct_size() helper".

Therefore, I would like to know your opinion and that of other people
about these patches. If the final consensus is that the code has no real
benefit, I will stop insisting on it ;)

Regards,
Erick
> 
> Link: https://www.kernel.org/doc/html/latest/process/deprecated.html#open-coded-arithmetic-in-allocator-arguments [1]
> Link: https://github.com/KSPP/linux/issues/160 [2]
> Link: https://lore.kernel.org/linux-hardening/20240430091833.GD40213@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx [3]
> Link: https://lore.kernel.org/linux-hardening/20240430091504.GC40213@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx [4]
> Link: https://docs.kernel.org/process/deprecated.html [5]
> ---
> Changes in v4:
> 
> - Add the "Reviewed-by:" tag to the three patches.
> - Rebase against linux next (tag next-20240531).
> 
> Previous versions:
> 
> perf/x86/amd/uncore: Add flex array to struct amd_uncore_ctx
>   v1 -> https://lore.kernel.org/linux-hardening/AS8PR02MB7237E4848B44A5226BD3551C8BE02@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> 
> perf/x86/intel/uncore: Prefer struct_size over open coded arithmetic
>   v1 -> https://lore.kernel.org/linux-hardening/AS8PR02MB7237F4D39BF6AA0FF40E91638B392@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> 
> perf/ring_buffer: Prefer struct_size over open coded arithmetic
>   v3 -> https://lore.kernel.org/linux-hardening/AS8PR02MB72379B4807F3951A1B926BA58BE02@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
>   v2 -> https://lore.kernel.org/linux-hardening/AS8PR02MB7237569E4FBE0B26F62FDFDB8B1D2@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
>   v1 -> https://lore.kernel.org/linux-hardening/AS8PR02MB72372AB065EA8340D960CCC48B1B2@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
> 
>   Changes in v3:
>   - Refactor the logic, compared to the previous version, of the second
>     rb_alloc() function to gain __counted_by() coverage (Kees Cook and
>     Christophe JAILLET).
> 
>   Changes in v2:
>   - Annotate "struct perf_buffer" with __counted_by() attribute (Kees Cook).
>   - Refactor the logic to gain __counted_by() coverage (Kees Cook).
> ---
> Erick Archer (3):
>   perf/x86/amd/uncore: Add flex array to struct amd_uncore_ctx
>   perf/x86/intel/uncore: Prefer struct_size over open coded arithmetic
>   perf/ring_buffer: Prefer struct_size over open coded arithmetic
> 
>  arch/x86/events/amd/uncore.c   | 18 +++++-------------
>  arch/x86/events/intel/uncore.c |  7 +++----
>  kernel/events/internal.h       |  2 +-
>  kernel/events/ring_buffer.c    | 15 ++++-----------
>  4 files changed, 13 insertions(+), 29 deletions(-)
> 
> -- 
> 2.25.1
>