Message ID | 20220302004339.3932356-3-song@kernel.org (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | BPF |
Headers | show |
Series | fixes for bpf_prog_pack | expand |
On 3/1/22 4:43 PM, Song Liu wrote: > On do_jit failure path, the header is freed by bpf_jit_binary_pack_free. > While bpf_jit_binary_pack_free doesn't require proper ro_header->size, > bpf_prog_pack_free still uses it. Set header->size in bpf_int_jit_compile > before calling bpf_jit_binary_pack_free. > > Fixes: 1022a5498f6f ("bpf, x86_64: Use bpf_jit_binary_pack_alloc") > Fixes: 33c9805860e5 ("bpf: Introduce bpf_jit_binary_pack_[alloc|finalize|free]") > Reported-by: Kui-Feng Lee <kuifeng@fb.com> > Signed-off-by: Song Liu <song@kernel.org> LGTM with a nit below related to comments. Acked-by: Yonghong Song <yhs@fb.com> > --- > arch/x86/net/bpf_jit_comp.c | 6 +++++- > kernel/bpf/core.c | 7 ++++--- > 2 files changed, 9 insertions(+), 4 deletions(-) > > diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c > index c7db0fe4de2f..b923d81ff6f9 100644 > --- a/arch/x86/net/bpf_jit_comp.c > +++ b/arch/x86/net/bpf_jit_comp.c > @@ -2330,8 +2330,12 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog) > if (proglen <= 0) { > out_image: > image = NULL; > - if (header) > + if (header) { > + /* set header->size for bpf_arch_text_copy */ This comment is confusing. Setting header->size is not for bpf_arch_text_copy. Probably you mean 'by bpf_arch_text_copy? I think this comment is not necessary. > + bpf_arch_text_copy(&header->size, &rw_header->size, > + sizeof(rw_header->size)); > bpf_jit_binary_pack_free(header, rw_header); > + } > prog = orig_prog; > goto out_addrs; > } > diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c > index ebb0193d07f0..da587e4619e0 100644 > --- a/kernel/bpf/core.c > +++ b/kernel/bpf/core.c > @@ -1112,13 +1112,14 @@ int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, > * 1) when the program is freed after; > * 2) when the JIT engine fails (before bpf_jit_binary_pack_finalize). > * For case 2), we need to free both the RO memory and the RW buffer. > - * Also, ro_header->size in 2) is not properly set yet, so rw_header->size > - * is used for uncharge. > + * > + * If bpf_jit_binary_pack_free is called before _finalize (jit error), Do you mean bpf_jit_binary_pack_free() is called before calling bpf_jit_binary_pack_finalize()? This seems not the case. > + * it is necessary to set ro_header->size properly before freeing it. > */ > void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, > struct bpf_binary_header *rw_header) > { > - u32 size = rw_header ? rw_header->size : ro_header->size; > + u32 size = ro_header->size; > > bpf_prog_pack_free(ro_header); > kvfree(rw_header);
> On Mar 1, 2022, at 11:12 PM, Yonghong Song <yhs@fb.com> wrote: > > > > On 3/1/22 4:43 PM, Song Liu wrote: >> On do_jit failure path, the header is freed by bpf_jit_binary_pack_free. >> While bpf_jit_binary_pack_free doesn't require proper ro_header->size, >> bpf_prog_pack_free still uses it. Set header->size in bpf_int_jit_compile >> before calling bpf_jit_binary_pack_free. >> Fixes: 1022a5498f6f ("bpf, x86_64: Use bpf_jit_binary_pack_alloc") >> Fixes: 33c9805860e5 ("bpf: Introduce bpf_jit_binary_pack_[alloc|finalize|free]") >> Reported-by: Kui-Feng Lee <kuifeng@fb.com> >> Signed-off-by: Song Liu <song@kernel.org> > > LGTM with a nit below related to comments. > > Acked-by: Yonghong Song <yhs@fb.com> > >> --- >> arch/x86/net/bpf_jit_comp.c | 6 +++++- >> kernel/bpf/core.c | 7 ++++--- >> 2 files changed, 9 insertions(+), 4 deletions(-) >> diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c >> index c7db0fe4de2f..b923d81ff6f9 100644 >> --- a/arch/x86/net/bpf_jit_comp.c >> +++ b/arch/x86/net/bpf_jit_comp.c >> @@ -2330,8 +2330,12 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog) >> if (proglen <= 0) { >> out_image: >> image = NULL; >> - if (header) >> + if (header) { >> + /* set header->size for bpf_arch_text_copy */ > > This comment is confusing. Setting header->size is not for bpf_arch_text_copy. Probably you mean 'by bpf_arch_text_copy? > I think this comment is not necessary. I meant to say set header->size for bpf_jit_binary_pack_free(). I guess it is not really necessary. Let me remove it. > >> + bpf_arch_text_copy(&header->size, &rw_header->size, >> + sizeof(rw_header->size)); >> bpf_jit_binary_pack_free(header, rw_header); >> + } >> prog = orig_prog; >> goto out_addrs; >> } >> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c >> index ebb0193d07f0..da587e4619e0 100644 >> --- a/kernel/bpf/core.c >> +++ b/kernel/bpf/core.c >> @@ -1112,13 +1112,14 @@ int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, >> * 1) when the program is freed after; >> * 2) when the JIT engine fails (before bpf_jit_binary_pack_finalize). >> * For case 2), we need to free both the RO memory and the RW buffer. >> - * Also, ro_header->size in 2) is not properly set yet, so rw_header->size >> - * is used for uncharge. >> + * >> + * If bpf_jit_binary_pack_free is called before _finalize (jit error), > > Do you mean bpf_jit_binary_pack_free() is called before calling > bpf_jit_binary_pack_finalize()? This seems not the case. Now I see this is a little confusing. Let me rephrase it in v2. Thanks, Song > >> + * it is necessary to set ro_header->size properly before freeing it. >> */ >> void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, >> struct bpf_binary_header *rw_header) >> { >> - u32 size = rw_header ? rw_header->size : ro_header->size; >> + u32 size = ro_header->size; >> bpf_prog_pack_free(ro_header); >> kvfree(rw_header);
diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index c7db0fe4de2f..b923d81ff6f9 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -2330,8 +2330,12 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog) if (proglen <= 0) { out_image: image = NULL; - if (header) + if (header) { + /* set header->size for bpf_arch_text_copy */ + bpf_arch_text_copy(&header->size, &rw_header->size, + sizeof(rw_header->size)); bpf_jit_binary_pack_free(header, rw_header); + } prog = orig_prog; goto out_addrs; } diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c index ebb0193d07f0..da587e4619e0 100644 --- a/kernel/bpf/core.c +++ b/kernel/bpf/core.c @@ -1112,13 +1112,14 @@ int bpf_jit_binary_pack_finalize(struct bpf_prog *prog, * 1) when the program is freed after; * 2) when the JIT engine fails (before bpf_jit_binary_pack_finalize). * For case 2), we need to free both the RO memory and the RW buffer. - * Also, ro_header->size in 2) is not properly set yet, so rw_header->size - * is used for uncharge. + * + * If bpf_jit_binary_pack_free is called before _finalize (jit error), + * it is necessary to set ro_header->size properly before freeing it. */ void bpf_jit_binary_pack_free(struct bpf_binary_header *ro_header, struct bpf_binary_header *rw_header) { - u32 size = rw_header ? rw_header->size : ro_header->size; + u32 size = ro_header->size; bpf_prog_pack_free(ro_header); kvfree(rw_header);
On do_jit failure path, the header is freed by bpf_jit_binary_pack_free. While bpf_jit_binary_pack_free doesn't require proper ro_header->size, bpf_prog_pack_free still uses it. Set header->size in bpf_int_jit_compile before calling bpf_jit_binary_pack_free. Fixes: 1022a5498f6f ("bpf, x86_64: Use bpf_jit_binary_pack_alloc") Fixes: 33c9805860e5 ("bpf: Introduce bpf_jit_binary_pack_[alloc|finalize|free]") Reported-by: Kui-Feng Lee <kuifeng@fb.com> Signed-off-by: Song Liu <song@kernel.org> --- arch/x86/net/bpf_jit_comp.c | 6 +++++- kernel/bpf/core.c | 7 ++++--- 2 files changed, 9 insertions(+), 4 deletions(-)