On Sat, 17 Nov 2018 at 23:47, Y Song [off-list ref] wrote:

On Sat, Nov 17, 2018 at 6:58 PM Ard Biesheuvel
[off-list ref] wrote:

quoted

All arch overrides of the __weak bpf_jit_free() amount to the same
thing: the allocated memory was never mapped read-only, and so
it does not have to be remapped to read-write before being freed.

So in preparation of permitting arches to serve allocations for BPF
JIT programs from other regions than the module region, refactor
the existing bpf_jit_free() implementations to use the shared code
where possible, and only specialize the remap and free operations.

Signed-off-by: Ard Biesheuvel <redacted>
---
 arch/mips/net/bpf_jit.c           |  7 ++-----
 arch/powerpc/net/bpf_jit_comp.c   |  7 ++-----
 arch/powerpc/net/bpf_jit_comp64.c |  9 +++------
 arch/sparc/net/bpf_jit_comp_32.c  |  7 ++-----
 kernel/bpf/core.c                 | 15 +++++----------
 5 files changed, 14 insertions(+), 31 deletions(-)

diff --git a/arch/mips/net/bpf_jit.c b/arch/mips/net/bpf_jit.c
index 1b69897274a1..5696bd7dccc7 100644
--- a/arch/mips/net/bpf_jit.c
+++ b/arch/mips/net/bpf_jit.c

@@ -1261,10 +1261,7 @@ void bpf_jit_compile(struct bpf_prog *fp)
        kfree(ctx.offsets);
 }

-void bpf_jit_free(struct bpf_prog *fp)
+void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 {
-       if (fp->jited)
-               bpf_jit_binary_free(bpf_jit_binary_hdr(fp));
-
-       bpf_prog_unlock_free(fp);
+       module_memfree(hdr);
 }

diff --git a/arch/powerpc/net/bpf_jit_comp.c b/arch/powerpc/net/bpf_jit_comp.c
index a1ea1ea6b40d..5b5ce4a1b44b 100644
--- a/arch/powerpc/net/bpf_jit_comp.c
+++ b/arch/powerpc/net/bpf_jit_comp.c

@@ -680,10 +680,7 @@ void bpf_jit_compile(struct bpf_prog *fp)
        return;
 }

-void bpf_jit_free(struct bpf_prog *fp)
+void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 {
-       if (fp->jited)
-               bpf_jit_binary_free(bpf_jit_binary_hdr(fp));
-
-       bpf_prog_unlock_free(fp);
+       module_memfree(hdr);
 }

diff --git a/arch/powerpc/net/bpf_jit_comp64.c b/arch/powerpc/net/bpf_jit_comp64.c
index 84c8f013a6c6..f64f1294bd62 100644
--- a/arch/powerpc/net/bpf_jit_comp64.c
+++ b/arch/powerpc/net/bpf_jit_comp64.c

@@ -1021,11 +1021,8 @@ struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *fp)
        return fp;
 }

-/* Overriding bpf_jit_free() as we don't set images read-only. */
-void bpf_jit_free(struct bpf_prog *fp)
+/* Overriding bpf_jit_binary_free() as we don't set images read-only. */
+void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 {
-       if (fp->jited)
-               bpf_jit_binary_free(bpf_jit_binary_hdr(fp));
-
-       bpf_prog_unlock_free(fp);
+       module_memfree(hdr);
 }

diff --git a/arch/sparc/net/bpf_jit_comp_32.c b/arch/sparc/net/bpf_jit_comp_32.c
index 01bda6bc9e7f..589950d152cc 100644
--- a/arch/sparc/net/bpf_jit_comp_32.c
+++ b/arch/sparc/net/bpf_jit_comp_32.c

@@ -756,10 +756,7 @@ cond_branch:                       f_offset = addrs[i + filter[i].jf];
        return;
 }

-void bpf_jit_free(struct bpf_prog *fp)
+void bpf_jit_binary_free(struct bpf_binary_header *hdr)
 {
-       if (fp->jited)
-               bpf_jit_binary_free(bpf_jit_binary_hdr(fp));
-
-       bpf_prog_unlock_free(fp);
+       module_memfree(hdr);
 }

diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
index 1a796e0799ec..29f766dac203 100644
--- a/kernel/bpf/core.c
+++ b/kernel/bpf/core.c

@@ -646,25 +646,20 @@ bpf_jit_binary_alloc(unsigned int proglen, u8 **image_ptr,
        return hdr;
 }

-void bpf_jit_binary_free(struct bpf_binary_header *hdr)
+void __weak bpf_jit_binary_free(struct bpf_binary_header *hdr)
 {
-       u32 pages = hdr->pages;
-
+       bpf_jit_binary_unlock_ro(hdr);
        module_memfree(hdr);
-       bpf_jit_uncharge_modmem(pages);
 }

-/* This symbol is only overridden by archs that have different
- * requirements than the usual eBPF JITs, f.e. when they only
- * implement cBPF JIT, do not set images read-only, etc.
- */

Do you want to move the above comments to
new weak function bpf_jit_binary_free?

Perhaps. But one thing I don't understand, looking at this again, is
why we have these overrides in the first place. module_memfree() just
calls vfree(), which takes down the mapping entirely (along with any
updated permissions), and so remapping it back to r/w right before
that seems rather pointless imo.

Can we get rid of bpf_jit_binary_unlock_ro() entirely, and along with
it, all these overrides for the free() path?