Message ID | 20241008091718.3797027-6-houtao@huaweicloud.com (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | BPF |
Headers | show |
Series | Misc fixes for bpf | expand |
On Tue, Oct 8, 2024 at 2:05 AM Hou Tao <houtao@huaweicloud.com> wrote: > > From: Hou Tao <houtao1@huawei.com> > > Under 32-bits host (e.g, arm32) , when a bpf program passes an u64 to > bpf_iter_bits_new(), bpf_iter_bits_new() will use bits_copy to save the > content of the u64, but the size of bits_copy is only 4-bytes, and there > will be stack corruption. > > Fix it by change the type of unsafe_ptr from u64 * to unsigned long *. > This will be confusing as BPF-side long is always 64-bit. So why not instead make sure it's u64 throughout (i.e., bits_copy is u64 explicitly), even on 32-bit architectures? > Signed-off-by: Hou Tao <houtao1@huawei.com> > --- > kernel/bpf/helpers.c | 18 ++++++++++-------- > 1 file changed, 10 insertions(+), 8 deletions(-) > > diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c > index 6c0205d5018c..dee69c3904a0 100644 > --- a/kernel/bpf/helpers.c > +++ b/kernel/bpf/helpers.c > @@ -2852,7 +2852,7 @@ struct bpf_iter_bits { > } __aligned(8); > > /* nr_bits only has 31 bits */ > -#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(u64)) > +#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(unsigned long)) > > struct bpf_iter_bits_kern { > union { > @@ -2868,8 +2868,9 @@ struct bpf_iter_bits_kern { > * bpf_iter_bits_new() - Initialize a new bits iterator for a given memory area > * @it: The new bpf_iter_bits to be created > * @unsafe_ptr__ign: A pointer pointing to a memory area to be iterated over > - * @nr_words: The size of the specified memory area, measured in 8-byte units. > - * Due to the limitation of memalloc, it can't be greater than 512. > + * @nr_words: The size of the specified memory area, measured in units of > + * sizeof(unsigned long). Due to the limitation of memalloc, it can't be > + * greater than 512. > * > * This function initializes a new bpf_iter_bits structure for iterating over > * a memory area which is specified by the @unsafe_ptr__ign and @nr_words. It > @@ -2879,17 +2880,18 @@ struct bpf_iter_bits_kern { > * On success, 0 is returned. On failure, ERR is returned. > */ > __bpf_kfunc int > -bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words) > +bpf_iter_bits_new(struct bpf_iter_bits *it, const unsigned long *unsafe_ptr__ign, u32 nr_words) > { > - struct bpf_iter_bits_kern *kit = (void *)it; > - u32 nr_bytes = nr_words * sizeof(u64); > + u32 nr_bytes = nr_words * sizeof(*unsafe_ptr__ign); > u32 nr_bits = BYTES_TO_BITS(nr_bytes); > + struct bpf_iter_bits_kern *kit; > int err; > > BUILD_BUG_ON(sizeof(struct bpf_iter_bits_kern) != sizeof(struct bpf_iter_bits)); > BUILD_BUG_ON(__alignof__(struct bpf_iter_bits_kern) != > __alignof__(struct bpf_iter_bits)); > > + kit = (void *)it; > kit->allocated = 0; > kit->nr_bits = 0; > kit->bits_copy = 0; > @@ -2900,8 +2902,8 @@ bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_w > if (nr_words > BITS_ITER_NR_WORDS_MAX) > return -E2BIG; > > - /* Optimization for u64 mask */ > - if (nr_bits == 64) { > + /* Optimization for unsigned long mask */ > + if (nr_words == 1) { > err = bpf_probe_read_kernel_common(&kit->bits_copy, nr_bytes, unsafe_ptr__ign); > if (err) > return -EFAULT; > -- > 2.29.2 >
On 10/9/2024 2:30 AM, Andrii Nakryiko wrote: > On Tue, Oct 8, 2024 at 2:05 AM Hou Tao <houtao@huaweicloud.com> wrote: >> From: Hou Tao <houtao1@huawei.com> >> >> Under 32-bits host (e.g, arm32) , when a bpf program passes an u64 to >> bpf_iter_bits_new(), bpf_iter_bits_new() will use bits_copy to save the >> content of the u64, but the size of bits_copy is only 4-bytes, and there >> will be stack corruption. >> >> Fix it by change the type of unsafe_ptr from u64 * to unsigned long *. >> > This will be confusing as BPF-side long is always 64-bit. So why not > instead make sure it's u64 throughout (i.e., bits_copy is u64 > explicitly), even on 32-bit architectures? Just learn about the size of BPF-side long is always 64-bits. I had considered to change bits_copy to u64. The main obstacle is that the pointer type of find_next_bit is unsigned long *, if it is used on an u64 under big-endian host, it may return invalid result. > >> Signed-off-by: Hou Tao <houtao1@huawei.com> >> --- >> kernel/bpf/helpers.c | 18 ++++++++++-------- >> 1 file changed, 10 insertions(+), 8 deletions(-) >> >> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c >> index 6c0205d5018c..dee69c3904a0 100644 >> --- a/kernel/bpf/helpers.c >> +++ b/kernel/bpf/helpers.c >> @@ -2852,7 +2852,7 @@ struct bpf_iter_bits { >> } __aligned(8); >> >> /* nr_bits only has 31 bits */ >> -#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(u64)) >> +#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(unsigned long)) >> >> struct bpf_iter_bits_kern { >> union { >> @@ -2868,8 +2868,9 @@ struct bpf_iter_bits_kern { >> * bpf_iter_bits_new() - Initialize a new bits iterator for a given memory area >> * @it: The new bpf_iter_bits to be created >> * @unsafe_ptr__ign: A pointer pointing to a memory area to be iterated over >> - * @nr_words: The size of the specified memory area, measured in 8-byte units. >> - * Due to the limitation of memalloc, it can't be greater than 512. >> + * @nr_words: The size of the specified memory area, measured in units of >> + * sizeof(unsigned long). Due to the limitation of memalloc, it can't be >> + * greater than 512. >> * >> * This function initializes a new bpf_iter_bits structure for iterating over >> * a memory area which is specified by the @unsafe_ptr__ign and @nr_words. It >> @@ -2879,17 +2880,18 @@ struct bpf_iter_bits_kern { >> * On success, 0 is returned. On failure, ERR is returned. >> */ >> __bpf_kfunc int >> -bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words) >> +bpf_iter_bits_new(struct bpf_iter_bits *it, const unsigned long *unsafe_ptr__ign, u32 nr_words) >> { >> - struct bpf_iter_bits_kern *kit = (void *)it; >> - u32 nr_bytes = nr_words * sizeof(u64); >> + u32 nr_bytes = nr_words * sizeof(*unsafe_ptr__ign); >> u32 nr_bits = BYTES_TO_BITS(nr_bytes); >> + struct bpf_iter_bits_kern *kit; >> int err; >> >> BUILD_BUG_ON(sizeof(struct bpf_iter_bits_kern) != sizeof(struct bpf_iter_bits)); >> BUILD_BUG_ON(__alignof__(struct bpf_iter_bits_kern) != >> __alignof__(struct bpf_iter_bits)); >> >> + kit = (void *)it; >> kit->allocated = 0; >> kit->nr_bits = 0; >> kit->bits_copy = 0; >> @@ -2900,8 +2902,8 @@ bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_w >> if (nr_words > BITS_ITER_NR_WORDS_MAX) >> return -E2BIG; >> >> - /* Optimization for u64 mask */ >> - if (nr_bits == 64) { >> + /* Optimization for unsigned long mask */ >> + if (nr_words == 1) { >> err = bpf_probe_read_kernel_common(&kit->bits_copy, nr_bytes, unsafe_ptr__ign); >> if (err) >> return -EFAULT; >> -- >> 2.29.2 >> > .
On Wed, Oct 9, 2024 at 10:45 AM Hou Tao <houtao@huaweicloud.com> wrote: > > > > On 10/9/2024 2:30 AM, Andrii Nakryiko wrote: > > On Tue, Oct 8, 2024 at 2:05 AM Hou Tao <houtao@huaweicloud.com> wrote: > >> From: Hou Tao <houtao1@huawei.com> > >> > >> Under 32-bits host (e.g, arm32) , when a bpf program passes an u64 to > >> bpf_iter_bits_new(), bpf_iter_bits_new() will use bits_copy to save the > >> content of the u64, but the size of bits_copy is only 4-bytes, and there > >> will be stack corruption. > >> > >> Fix it by change the type of unsafe_ptr from u64 * to unsigned long *. > >> > > This will be confusing as BPF-side long is always 64-bit. So why not > > instead make sure it's u64 throughout (i.e., bits_copy is u64 > > explicitly), even on 32-bit architectures? > > Just learn about the size of BPF-side long is always 64-bits. I had > considered to change bits_copy to u64. The main obstacle is that the > pointer type of find_next_bit is unsigned long *, if it is used on an > u64 under big-endian host, it may return invalid result. IIUC, BPF targets only 64-bit systems?
On Wed, Oct 9, 2024 at 4:40 AM Yafang Shao <laoar.shao@gmail.com> wrote: > > On Wed, Oct 9, 2024 at 10:45 AM Hou Tao <houtao@huaweicloud.com> wrote: > > > > > > > > On 10/9/2024 2:30 AM, Andrii Nakryiko wrote: > > > On Tue, Oct 8, 2024 at 2:05 AM Hou Tao <houtao@huaweicloud.com> wrote: > > >> From: Hou Tao <houtao1@huawei.com> > > >> > > >> Under 32-bits host (e.g, arm32) , when a bpf program passes an u64 to > > >> bpf_iter_bits_new(), bpf_iter_bits_new() will use bits_copy to save the > > >> content of the u64, but the size of bits_copy is only 4-bytes, and there > > >> will be stack corruption. > > >> > > >> Fix it by change the type of unsafe_ptr from u64 * to unsigned long *. > > >> > > > This will be confusing as BPF-side long is always 64-bit. So why not > > > instead make sure it's u64 throughout (i.e., bits_copy is u64 > > > explicitly), even on 32-bit architectures? > > > > Just learn about the size of BPF-side long is always 64-bits. I had > > considered to change bits_copy to u64. The main obstacle is that the > > pointer type of find_next_bit is unsigned long *, if it is used on an > > u64 under big-endian host, it may return invalid result. > > IIUC, BPF targets only 64-bit systems? 64-bit bpf programs run just fine on 32-bit systems.
On 10/9/2024 10:45 AM, Hou Tao wrote: > > On 10/9/2024 2:30 AM, Andrii Nakryiko wrote: >> On Tue, Oct 8, 2024 at 2:05 AM Hou Tao <houtao@huaweicloud.com> wrote: >>> From: Hou Tao <houtao1@huawei.com> >>> >>> Under 32-bits host (e.g, arm32) , when a bpf program passes an u64 to >>> bpf_iter_bits_new(), bpf_iter_bits_new() will use bits_copy to save the >>> content of the u64, but the size of bits_copy is only 4-bytes, and there >>> will be stack corruption. >>> >>> Fix it by change the type of unsafe_ptr from u64 * to unsigned long *. >>> >> This will be confusing as BPF-side long is always 64-bit. So why not >> instead make sure it's u64 throughout (i.e., bits_copy is u64 >> explicitly), even on 32-bit architectures? > Just learn about the size of BPF-side long is always 64-bits. I had > considered to change bits_copy to u64. The main obstacle is that the > pointer type of find_next_bit is unsigned long *, if it is used on an > u64 under big-endian host, it may return invalid result. I think doing the following swap for big endian and 32-bits host will let find_next_bit return the correct result: +static void swap_bits(u64 *bits, unsigned int nr) +{ +#if defined(__BIG_ENDIAN) && !defined(CONFIG_64BIT) + unsigned int i; + + for (i = 0; i < nr; i++) + bits[i] = (bits[i] >> 32) | ((u64)(u32)bits[i] << 32); +#endif +} + Will try to get some test environment to test it. >>> Signed-off-by: Hou Tao <houtao1@huawei.com> >>> --- >>> kernel/bpf/helpers.c | 18 ++++++++++-------- >>> 1 file changed, 10 insertions(+), 8 deletions(-) >>> >>> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c >>> index 6c0205d5018c..dee69c3904a0 100644 >>> --- a/kernel/bpf/helpers.c >>> +++ b/kernel/bpf/helpers.c >>> @@ -2852,7 +2852,7 @@ struct bpf_iter_bits { >>> } __aligned(8); >>> >>> /* nr_bits only has 31 bits */ >>> -#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(u64)) >>> +#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(unsigned long)) >>> >>> struct bpf_iter_bits_kern { >>> union { >>> @@ -2868,8 +2868,9 @@ struct bpf_iter_bits_kern { >>> * bpf_iter_bits_new() - Initialize a new bits iterator for a given memory area >>> * @it: The new bpf_iter_bits to be created >>> * @unsafe_ptr__ign: A pointer pointing to a memory area to be iterated over >>> - * @nr_words: The size of the specified memory area, measured in 8-byte units. >>> - * Due to the limitation of memalloc, it can't be greater than 512. >>> + * @nr_words: The size of the specified memory area, measured in units of >>> + * sizeof(unsigned long). Due to the limitation of memalloc, it can't be >>> + * greater than 512. >>> * >>> * This function initializes a new bpf_iter_bits structure for iterating over >>> * a memory area which is specified by the @unsafe_ptr__ign and @nr_words. It >>> @@ -2879,17 +2880,18 @@ struct bpf_iter_bits_kern { >>> * On success, 0 is returned. On failure, ERR is returned. >>> */ >>> __bpf_kfunc int >>> -bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words) >>> +bpf_iter_bits_new(struct bpf_iter_bits *it, const unsigned long *unsafe_ptr__ign, u32 nr_words) >>> { >>> - struct bpf_iter_bits_kern *kit = (void *)it; >>> - u32 nr_bytes = nr_words * sizeof(u64); >>> + u32 nr_bytes = nr_words * sizeof(*unsafe_ptr__ign); >>> u32 nr_bits = BYTES_TO_BITS(nr_bytes); >>> + struct bpf_iter_bits_kern *kit; >>> int err; >>> >>> BUILD_BUG_ON(sizeof(struct bpf_iter_bits_kern) != sizeof(struct bpf_iter_bits)); >>> BUILD_BUG_ON(__alignof__(struct bpf_iter_bits_kern) != >>> __alignof__(struct bpf_iter_bits)); >>> >>> + kit = (void *)it; >>> kit->allocated = 0; >>> kit->nr_bits = 0; >>> kit->bits_copy = 0; >>> @@ -2900,8 +2902,8 @@ bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_w >>> if (nr_words > BITS_ITER_NR_WORDS_MAX) >>> return -E2BIG; >>> >>> - /* Optimization for u64 mask */ >>> - if (nr_bits == 64) { >>> + /* Optimization for unsigned long mask */ >>> + if (nr_words == 1) { >>> err = bpf_probe_read_kernel_common(&kit->bits_copy, nr_bytes, unsafe_ptr__ign); >>> if (err) >>> return -EFAULT; >>> -- >>> 2.29.2 >>> >> . > .
diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c index 6c0205d5018c..dee69c3904a0 100644 --- a/kernel/bpf/helpers.c +++ b/kernel/bpf/helpers.c @@ -2852,7 +2852,7 @@ struct bpf_iter_bits { } __aligned(8); /* nr_bits only has 31 bits */ -#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(u64)) +#define BITS_ITER_NR_WORDS_MAX ((1U << 31) / BITS_PER_TYPE(unsigned long)) struct bpf_iter_bits_kern { union { @@ -2868,8 +2868,9 @@ struct bpf_iter_bits_kern { * bpf_iter_bits_new() - Initialize a new bits iterator for a given memory area * @it: The new bpf_iter_bits to be created * @unsafe_ptr__ign: A pointer pointing to a memory area to be iterated over - * @nr_words: The size of the specified memory area, measured in 8-byte units. - * Due to the limitation of memalloc, it can't be greater than 512. + * @nr_words: The size of the specified memory area, measured in units of + * sizeof(unsigned long). Due to the limitation of memalloc, it can't be + * greater than 512. * * This function initializes a new bpf_iter_bits structure for iterating over * a memory area which is specified by the @unsafe_ptr__ign and @nr_words. It @@ -2879,17 +2880,18 @@ struct bpf_iter_bits_kern { * On success, 0 is returned. On failure, ERR is returned. */ __bpf_kfunc int -bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_words) +bpf_iter_bits_new(struct bpf_iter_bits *it, const unsigned long *unsafe_ptr__ign, u32 nr_words) { - struct bpf_iter_bits_kern *kit = (void *)it; - u32 nr_bytes = nr_words * sizeof(u64); + u32 nr_bytes = nr_words * sizeof(*unsafe_ptr__ign); u32 nr_bits = BYTES_TO_BITS(nr_bytes); + struct bpf_iter_bits_kern *kit; int err; BUILD_BUG_ON(sizeof(struct bpf_iter_bits_kern) != sizeof(struct bpf_iter_bits)); BUILD_BUG_ON(__alignof__(struct bpf_iter_bits_kern) != __alignof__(struct bpf_iter_bits)); + kit = (void *)it; kit->allocated = 0; kit->nr_bits = 0; kit->bits_copy = 0; @@ -2900,8 +2902,8 @@ bpf_iter_bits_new(struct bpf_iter_bits *it, const u64 *unsafe_ptr__ign, u32 nr_w if (nr_words > BITS_ITER_NR_WORDS_MAX) return -E2BIG; - /* Optimization for u64 mask */ - if (nr_bits == 64) { + /* Optimization for unsigned long mask */ + if (nr_words == 1) { err = bpf_probe_read_kernel_common(&kit->bits_copy, nr_bytes, unsafe_ptr__ign); if (err) return -EFAULT;