Merge https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf

Daniel Borkmann says:

====================
pull-request: bpf 2021-10-07

We've added 7 non-merge commits during the last 8 day(s) which contain
a total of 8 files changed, 38 insertions(+), 21 deletions(-).

The main changes are:

1) Fix ARM BPF JIT to preserve caller-saved regs for DIV/MOD JIT-internal
   helper call, from Johan Almbladh.

2) Fix integer overflow in BPF stack map element size calculation when
   used with preallocation, from Tatsuhiko Yasumatsu.

3) Fix an AF_UNIX regression due to added BPF sockmap support related
   to shutdown handling, from Jiang Wang.

4) Fix a segfault in libbpf when generating light skeletons from objects
   without BTF, from Kumar Kartikeya Dwivedi.

5) Fix a libbpf memory leak in strset to free the actual struct strset
   itself, from Andrii Nakryiko.

6) Dual-license bpf_insn.h similarly as we did for libbpf and bpftool,
   with ACKs from all contributors, from Luca Boccassi.
====================

Link: https://lore.kernel.org/r/20211007135010.21143-1-daniel@iogearbox.net
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 19 insertions, 0 deletions

diff --git a/arch/arm/net/bpf_jit_32.c b/arch/arm/net/bpf_jit_32.c
index a951276f0547..a903b26cde40 100644
--- a/arch/arm/net/bpf_jit_32.c
+++ b/arch/arm/net/bpf_jit_32.c
@@ -36,6 +36,10 @@
  *                        +-----+
  *                        |RSVD | JIT scratchpad
  * current ARM_SP =>      +-----+ <= (BPF_FP - STACK_SIZE + SCRATCH_SIZE)
+ *                        | ... | caller-saved registers
+ *                        +-----+
+ *                        | ... | arguments passed on stack
+ * ARM_SP during call =>  +-----|
  *                        |     |
  *                        | ... | Function call stack
  *                        |     |
@@ -63,6 +67,12 @@
  *
  * When popping registers off the stack at the end of a BPF function, we
  * reference them via the current ARM_FP register.
+ *
+ * Some eBPF operations are implemented via a call to a helper function.
+ * Such calls are "invisible" in the eBPF code, so it is up to the calling
+ * program to preserve any caller-saved ARM registers during the call. The
+ * JIT emits code to push and pop those registers onto the stack, immediately
+ * above the callee stack frame.
  */
 #define CALLEE_MASK	(1 << ARM_R4 | 1 << ARM_R5 | 1 << ARM_R6 | \
 			 1 << ARM_R7 | 1 << ARM_R8 | 1 << ARM_R9 | \
@@ -70,6 +80,8 @@
 #define CALLEE_PUSH_MASK (CALLEE_MASK | 1 << ARM_LR)
 #define CALLEE_POP_MASK  (CALLEE_MASK | 1 << ARM_PC)
 
+#define CALLER_MASK	(1 << ARM_R0 | 1 << ARM_R1 | 1 << ARM_R2 | 1 << ARM_R3)
+
 enum {
 	/* Stack layout - these are offsets from (top of stack - 4) */
 	BPF_R2_HI,
@@ -464,6 +476,7 @@ static inline int epilogue_offset(const struct jit_ctx *ctx)
 
 static inline void emit_udivmod(u8 rd, u8 rm, u8 rn, struct jit_ctx *ctx, u8 op)
 {
+	const int exclude_mask = BIT(ARM_R0) | BIT(ARM_R1);
 	const s8 *tmp = bpf2a32[TMP_REG_1];
 
 #if __LINUX_ARM_ARCH__ == 7
@@ -495,11 +508,17 @@ static inline void emit_udivmod(u8 rd, u8 rm, u8 rn, struct jit_ctx *ctx, u8 op)
 		emit(ARM_MOV_R(ARM_R0, rm), ctx);
 	}
 
+	/* Push caller-saved registers on stack */
+	emit(ARM_PUSH(CALLER_MASK & ~exclude_mask), ctx);
+
 	/* Call appropriate function */
 	emit_mov_i(ARM_IP, op == BPF_DIV ?
 		   (u32)jit_udiv32 : (u32)jit_mod32, ctx);
 	emit_blx_r(ARM_IP, ctx);
 
+	/* Restore caller-saved registers from stack */
+	emit(ARM_POP(CALLER_MASK & ~exclude_mask), ctx);
+
 	/* Save return value */
 	if (rd != ARM_R0)
 		emit(ARM_MOV_R(rd, ARM_R0), ctx);