Message ID | 20191029154307.23053-1-bjorn.topel@gmail.com |
---|---|
State | Accepted |
Delegated to: | BPF Maintainers |
Headers | show |
Series | [bpf] bpf: change size to u64 for bpf_map_{area_alloc,charge_init}() | expand |
On Tue, 29 Oct 2019 16:43:07 +0100, Björn Töpel wrote: > From: Björn Töpel <bjorn.topel@intel.com> > > The functions bpf_map_area_alloc() and bpf_map_charge_init() prior > this commit passed the size parameter as size_t. In this commit this > is changed to u64. > > All users of these functions avoid size_t overflows on 32-bit systems, > by explicitly using u64 when calculating the allocation size and > memory charge cost. However, since the result was narrowed by the > size_t when passing size and cost to the functions, the overflow > handling was in vain. > > Instead of changing all call sites to size_t and handle overflow at > the call site, the parameter is changed to u64 and checked in the > functions above. > > Fixes: d407bd25a204 ("bpf: don't trigger OOM killer under pressure with map alloc") > Fixes: c85d69135a91 ("bpf: move memory size checks to bpf_map_charge_init()") > Signed-off-by: Björn Töpel <bjorn.topel@intel.com> Okay, I guess that's the smallest change we can make here. I'd prefer we went the way of using the standard overflow handling the kernel has, rather than proliferating this u64 + U32_MAX comparison stuff. But it's hard to argue with the patch length in light of the necessary backports.. Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
On Tue, 29 Oct 2019 at 17:12, Jakub Kicinski <jakub.kicinski@netronome.com> wrote: > > On Tue, 29 Oct 2019 16:43:07 +0100, Björn Töpel wrote: > > From: Björn Töpel <bjorn.topel@intel.com> > > > > The functions bpf_map_area_alloc() and bpf_map_charge_init() prior > > this commit passed the size parameter as size_t. In this commit this > > is changed to u64. > > > > All users of these functions avoid size_t overflows on 32-bit systems, > > by explicitly using u64 when calculating the allocation size and > > memory charge cost. However, since the result was narrowed by the > > size_t when passing size and cost to the functions, the overflow > > handling was in vain. > > > > Instead of changing all call sites to size_t and handle overflow at > > the call site, the parameter is changed to u64 and checked in the > > functions above. > > > > Fixes: d407bd25a204 ("bpf: don't trigger OOM killer under pressure with map alloc") > > Fixes: c85d69135a91 ("bpf: move memory size checks to bpf_map_charge_init()") > > Signed-off-by: Björn Töpel <bjorn.topel@intel.com> > > Okay, I guess that's the smallest change we can make here. > > I'd prefer we went the way of using the standard overflow handling the > kernel has, rather than proliferating this u64 + U32_MAX comparison > stuff. But it's hard to argue with the patch length in light of the > necessary backports.. > I agree with you, but this is a start, and then maps can gradually move over to standard overflow handling. > Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
On Tue, Oct 29, 2019 at 04:43:07PM +0100, Björn Töpel wrote: > From: Björn Töpel <bjorn.topel@intel.com> > > The functions bpf_map_area_alloc() and bpf_map_charge_init() prior > this commit passed the size parameter as size_t. In this commit this > is changed to u64. > > All users of these functions avoid size_t overflows on 32-bit systems, > by explicitly using u64 when calculating the allocation size and > memory charge cost. However, since the result was narrowed by the > size_t when passing size and cost to the functions, the overflow > handling was in vain. > > Instead of changing all call sites to size_t and handle overflow at > the call site, the parameter is changed to u64 and checked in the > functions above. > > Fixes: d407bd25a204 ("bpf: don't trigger OOM killer under pressure with map alloc") > Fixes: c85d69135a91 ("bpf: move memory size checks to bpf_map_charge_init()") > Signed-off-by: Björn Töpel <bjorn.topel@intel.com> Applied, thanks!
diff --git a/include/linux/bpf.h b/include/linux/bpf.h index 5b9d22338606..3bf3835d0e86 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -656,11 +656,11 @@ void bpf_map_put_with_uref(struct bpf_map *map); void bpf_map_put(struct bpf_map *map); int bpf_map_charge_memlock(struct bpf_map *map, u32 pages); void bpf_map_uncharge_memlock(struct bpf_map *map, u32 pages); -int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size); +int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size); void bpf_map_charge_finish(struct bpf_map_memory *mem); void bpf_map_charge_move(struct bpf_map_memory *dst, struct bpf_map_memory *src); -void *bpf_map_area_alloc(size_t size, int numa_node); +void *bpf_map_area_alloc(u64 size, int numa_node); void bpf_map_area_free(void *base); void bpf_map_init_from_attr(struct bpf_map *map, union bpf_attr *attr); diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c index 0937719b87e2..ace1cfaa24b6 100644 --- a/kernel/bpf/syscall.c +++ b/kernel/bpf/syscall.c @@ -126,7 +126,7 @@ static struct bpf_map *find_and_alloc_map(union bpf_attr *attr) return map; } -void *bpf_map_area_alloc(size_t size, int numa_node) +void *bpf_map_area_alloc(u64 size, int numa_node) { /* We really just want to fail instead of triggering OOM killer * under memory pressure, therefore we set __GFP_NORETRY to kmalloc, @@ -141,6 +141,9 @@ void *bpf_map_area_alloc(size_t size, int numa_node) const gfp_t flags = __GFP_NOWARN | __GFP_ZERO; void *area; + if (size >= SIZE_MAX) + return NULL; + if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { area = kmalloc_node(size, GFP_USER | __GFP_NORETRY | flags, numa_node); @@ -197,7 +200,7 @@ static void bpf_uncharge_memlock(struct user_struct *user, u32 pages) atomic_long_sub(pages, &user->locked_vm); } -int bpf_map_charge_init(struct bpf_map_memory *mem, size_t size) +int bpf_map_charge_init(struct bpf_map_memory *mem, u64 size) { u32 pages = round_up(size, PAGE_SIZE) >> PAGE_SHIFT; struct user_struct *user;