bpf: add extra information for external symbol definitions
authorKonstantin Ananyev <konstantin.ananyev@intel.com>
Fri, 8 Jun 2018 08:42:33 +0000 (09:42 +0100)
committerThomas Monjalon <thomas@monjalon.net>
Thu, 12 Jul 2018 07:37:16 +0000 (09:37 +0200)
Extend struct rte_bpf_xsym with new fields to provide information about:
 - for variables - type and size
 - for functions - number of arguments and type/size of each argument
   and return value

Such information would allow validate code to perform
more extensive checking on input BPF program and catch
misbehaving BPF code.

That change would cause ABI/API breakage for librte_bpf.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
app/test-pmd/bpf_cmd.c
lib/librte_bpf/bpf_def.h
lib/librte_bpf/bpf_exec.c
lib/librte_bpf/bpf_impl.h
lib/librte_bpf/bpf_jit_x86.c
lib/librte_bpf/bpf_load.c
lib/librte_bpf/bpf_load_elf.c
lib/librte_bpf/rte_bpf.h
test/test/test_bpf.c

index 584fad9..830bfc1 100644 (file)
@@ -19,12 +19,35 @@ static const struct rte_bpf_xsym bpf_xsym[] = {
        {
                .name = RTE_STR(stdout),
                .type = RTE_BPF_XTYPE_VAR,
-               .var = &stdout,
+               .var = {
+                       .val = &stdout,
+                       .desc = {
+                               .type = RTE_BPF_ARG_PTR,
+                               .size = sizeof(stdout),
+                       },
+               },
        },
        {
                .name = RTE_STR(rte_pktmbuf_dump),
                .type = RTE_BPF_XTYPE_FUNC,
-               .func = (void *)rte_pktmbuf_dump,
+               .func = {
+                       .val = (void *)rte_pktmbuf_dump,
+                       .nb_args = 3,
+                       .args = {
+                               [0] = {
+                                       .type = RTE_BPF_ARG_RAW,
+                                       .size = sizeof(uintptr_t),
+                               },
+                               [1] = {
+                                       .type = RTE_BPF_ARG_PTR_MBUF,
+                                       .size = sizeof(struct rte_mbuf),
+                               },
+                               [2] = {
+                                       .type = RTE_BPF_ARG_RAW,
+                                       .size = sizeof(uint32_t),
+                               },
+                       },
+               },
        },
 };
 
index 6b69de3..c10f3ae 100644 (file)
@@ -131,6 +131,11 @@ struct ebpf_insn {
        int32_t imm;
 };
 
+/*
+ * eBPF allows functions with R1-R5 as arguments.
+ */
+#define        EBPF_FUNC_MAX_ARGS      (EBPF_REG_6 - EBPF_REG_1)
+
 #ifdef __cplusplus
 }
 #endif
index e373b1f..6a79139 100644 (file)
@@ -402,7 +402,7 @@ bpf_exec(const struct rte_bpf *bpf, uint64_t reg[EBPF_REG_NUM])
                        break;
                /* call instructions */
                case (BPF_JMP | EBPF_CALL):
-                       reg[EBPF_REG_0] = bpf->prm.xsym[ins->imm].func(
+                       reg[EBPF_REG_0] = bpf->prm.xsym[ins->imm].func.val(
                                reg[EBPF_REG_1], reg[EBPF_REG_2],
                                reg[EBPF_REG_3], reg[EBPF_REG_4],
                                reg[EBPF_REG_5]);
index 5d7e65c..b577e2c 100644 (file)
@@ -34,6 +34,20 @@ extern int rte_bpf_logtype;
 #define        RTE_BPF_LOG(lvl, fmt, args...) \
        rte_log(RTE_LOG_## lvl, rte_bpf_logtype, fmt, ##args)
 
+static inline size_t
+bpf_size(uint32_t bpf_op_sz)
+{
+       if (bpf_op_sz == BPF_B)
+               return sizeof(uint8_t);
+       else if (bpf_op_sz == BPF_H)
+               return sizeof(uint16_t);
+       else if (bpf_op_sz == BPF_W)
+               return sizeof(uint32_t);
+       else if (bpf_op_sz == EBPF_DW)
+               return sizeof(uint64_t);
+       return 0;
+}
+
 #ifdef __cplusplus
 }
 #endif
index 111e028..68ea389 100644 (file)
@@ -113,20 +113,6 @@ union bpf_jit_imm {
        uint8_t u8[4];
 };
 
-static size_t
-bpf_size(uint32_t bpf_op_sz)
-{
-       if (bpf_op_sz == BPF_B)
-               return sizeof(uint8_t);
-       else if (bpf_op_sz == BPF_H)
-               return sizeof(uint16_t);
-       else if (bpf_op_sz == BPF_W)
-               return sizeof(uint32_t);
-       else if (bpf_op_sz == EBPF_DW)
-               return sizeof(uint64_t);
-       return 0;
-}
-
 /*
  * In many cases for imm8 we can produce shorter code.
  */
@@ -1294,7 +1280,8 @@ emit(struct bpf_jit_state *st, const struct rte_bpf *bpf)
                        break;
                /* call instructions */
                case (BPF_JMP | EBPF_CALL):
-                       emit_call(st, (uintptr_t)bpf->prm.xsym[ins->imm].func);
+                       emit_call(st,
+                               (uintptr_t)bpf->prm.xsym[ins->imm].func.val);
                        break;
                /* return instruction */
                case (BPF_JMP | EBPF_EXIT):
index d1c9abd..2b84fe7 100644 (file)
@@ -51,17 +51,64 @@ bpf_load(const struct rte_bpf_prm *prm)
        return bpf;
 }
 
+/*
+ * Check that user provided external symbol.
+ */
+static int
+bpf_check_xsym(const struct rte_bpf_xsym *xsym)
+{
+       uint32_t i;
+
+       if (xsym->name == NULL)
+               return -EINVAL;
+
+       if (xsym->type == RTE_BPF_XTYPE_VAR) {
+               if (xsym->var.desc.type == RTE_BPF_ARG_UNDEF)
+                       return -EINVAL;
+       } else if (xsym->type == RTE_BPF_XTYPE_FUNC) {
+
+               if (xsym->func.nb_args > EBPF_FUNC_MAX_ARGS)
+                       return -EINVAL;
+
+               /* check function arguments */
+               for (i = 0; i != xsym->func.nb_args; i++) {
+                       if (xsym->func.args[i].type == RTE_BPF_ARG_UNDEF)
+                               return -EINVAL;
+               }
+
+               /* check return value info */
+               if (xsym->func.ret.type != RTE_BPF_ARG_UNDEF &&
+                               xsym->func.ret.size == 0)
+                       return -EINVAL;
+       } else
+               return -EINVAL;
+
+       return 0;
+}
+
 __rte_experimental struct rte_bpf *
 rte_bpf_load(const struct rte_bpf_prm *prm)
 {
        struct rte_bpf *bpf;
        int32_t rc;
+       uint32_t i;
 
-       if (prm == NULL || prm->ins == NULL) {
+       if (prm == NULL || prm->ins == NULL ||
+                       (prm->nb_xsym != 0 && prm->xsym == NULL)) {
                rte_errno = EINVAL;
                return NULL;
        }
 
+       rc = 0;
+       for (i = 0; i != prm->nb_xsym && rc == 0; i++)
+               rc = bpf_check_xsym(prm->xsym + i);
+
+       if (rc != 0) {
+               rte_errno = -rc;
+               RTE_BPF_LOG(ERR, "%s: %d-th xsym is invalid\n", __func__, i);
+               return NULL;
+       }
+
        bpf = bpf_load(prm);
        if (bpf == NULL) {
                rte_errno = ENOMEM;
index 6ab03d8..96d3630 100644 (file)
@@ -81,9 +81,9 @@ resolve_xsym(const char *sn, size_t ofs, struct ebpf_insn *ins, size_t ins_sz,
                ins[idx].imm = fidx;
        /* for variable we need to store its absolute address */
        else {
-               ins[idx].imm = (uintptr_t)prm->xsym[fidx].var;
+               ins[idx].imm = (uintptr_t)prm->xsym[fidx].var.val;
                ins[idx + 1].imm =
-                       (uint64_t)(uintptr_t)prm->xsym[fidx].var >> 32;
+                       (uint64_t)(uintptr_t)prm->xsym[fidx].var.val >> 32;
        }
 
        return 0;
index 1249a99..ad62ef2 100644 (file)
@@ -40,7 +40,11 @@ enum rte_bpf_arg_type {
  */
 struct rte_bpf_arg {
        enum rte_bpf_arg_type type;
-       size_t size;     /**< for pointer types, size of data it points to */
+       /**
+        * for ptr type - max size of data buffer it points to
+        * for raw type - the size (in bytes) of the value
+        */
+       size_t size;
        size_t buf_size;
        /**< for mbuf ptr type, max size of rte_mbuf data buffer */
 };
@@ -66,10 +70,19 @@ struct rte_bpf_xsym {
        const char *name;        /**< name */
        enum rte_bpf_xtype type; /**< type */
        union {
-               uint64_t (*func)(uint64_t, uint64_t, uint64_t,
+               struct {
+                       uint64_t (*val)(uint64_t, uint64_t, uint64_t,
                                uint64_t, uint64_t);
-               void *var;
-       }; /**< value */
+                       uint32_t nb_args;
+                       struct rte_bpf_arg args[EBPF_FUNC_MAX_ARGS];
+                       /**< Function arguments descriptions. */
+                       struct rte_bpf_arg ret; /**< function return value. */
+               } func;
+               struct {
+                       void *val; /**< actual memory location */
+                       struct rte_bpf_arg desc; /**< type, size, etc. */
+               } var; /**< external variable */
+       };
 };
 
 /**
index cbd6be6..1e9caef 100644 (file)
@@ -1530,7 +1530,24 @@ static const struct rte_bpf_xsym test_call1_xsym[] = {
        {
                .name = RTE_STR(dummy_func1),
                .type = RTE_BPF_XTYPE_FUNC,
-               .func = (void *)dummy_func1,
+               .func = {
+                       .val = (void *)dummy_func1,
+                       .nb_args = 3,
+                       .args = {
+                               [0] = {
+                                       .type = RTE_BPF_ARG_PTR,
+                                       .size = sizeof(struct dummy_offset),
+                               },
+                               [1] = {
+                                       .type = RTE_BPF_ARG_PTR,
+                                       .size = sizeof(uint32_t),
+                               },
+                               [2] = {
+                                       .type = RTE_BPF_ARG_PTR,
+                                       .size = sizeof(uint64_t),
+                               },
+                       },
+               },
        },
 };