moar code gengit st

global.c

@@ -14,6 +14,86 @@
 
 /* . */
 
+static int global_dot_ir_pre(const struct func *func, struct node *n,
+				struct prog *prog)
+{
+	struct node *sou = n->expr.args;
+
+	if (node_is(sou, ":deref")) {
+		/* (*ptr).member, if *ptr is not already loaded let it
+		 * know that we're only interested in one member */
+		sou->sym->irs.hint.dot = 1;
+
+		/* this also means we need to put ourselves on the
+		 * stack since data will be loaded via probe_read */
+		n->sym->irs.hint.stack = 1;
+	}
+	return 0;
+}
+
+static int global_dot_ir_post(const struct func *func, struct node *n,
+				struct prog *prog)
+{
+	struct node *sou, *member;
+	struct irstate *dst;
+	ssize_t offset;
+
+	sou = n->expr.args;
+	member = sou->next;
+	dst = &n->sym->irs;
+	
+	ir_init_sym(prog->ir, n->sym);
+
+	offset = type_offsetof(type_base(sou->sym->type), member->string.data);
+	assert(offset >= 0);
+
+	if (!sou->sym->irs.loc) {
+		/* sou is a :deref which wasn't loaded by child, just
+		 * read the member we're interested in. */
+		struct node *ptr = sou->expr.args;
+
+		ir_emit_sym_to_reg(prog->ir, BPF_REG_3, ptr->sym);
+		ir_emit_insn(prog->ir, ALU_IMM(BPF_ADD, 0, offset), BPF_REG_3, 0);
+		goto probe_read;
+	}
+
+	offset += sou->sym->irs.stack;
+
+	if (dst->loc == LOC_REG) {
+		switch (dst->size) {
+		case 1:
+			ir_emit_insn(prog->ir, LDXB(0, offset, 0),
+				     dst->reg, BPF_REG_BP);
+			break;
+		case 2:
+			ir_emit_insn(prog->ir, LDXH(0, offset, 0),
+				     dst->reg, BPF_REG_BP);
+			break;
+		case 4:
+			ir_emit_insn(prog->ir, LDXW(0, offset, 0),
+				     dst->reg, BPF_REG_BP);
+			break;
+		case 8:
+			ir_emit_insn(prog->ir, LDXDW(0, offset, 0),
+				     dst->reg, BPF_REG_BP);
+			break;
+		default:
+			assert(0);
+		}
+
+		return 0;
+	}
+
+	ir_emit_insn(prog->ir, ALU_IMM(BPF_ADD, 0, offset), BPF_REG_3, 0);
+probe_read:
+	ir_emit_insn(prog->ir, MOV_IMM(0, (int32_t)dst->size), BPF_REG_2, 0);
+	ir_emit_insn(prog->ir, MOV(0, 0), BPF_REG_1, BPF_REG_BP);
+	ir_emit_insn(prog->ir, ALU_IMM(BPF_ADD, 0, dst->stack), BPF_REG_1, 0);
+	ir_emit_insn(prog->ir, CALL(BPF_FUNC_probe_read), 0, 0);
+	/* TODO if (r0) exit(r0); */
+	return 0;
+}
+
 static int global_dot_type_infer(const struct func *func, struct node *n)
 {
 	struct node *sou, *member;
@@ -37,18 +117,16 @@ static int global_dot_type_infer(const struct func *func, struct node *n)
 		return -EINVAL;
 	}
 
-	tfields_foreach(f, t->sou.fields) {
-		if (!strcmp(f->name, member->string.data)) {
-			/* given `sou.member` where sou is a
-			 * struct/union, infer that the expression's
-			 * type is equal to member's type. */
-			n->sym->type = f->type;
-			return 0;
-		}
+	f = tfields_get(t->sou.fields, member->string.data);
+	if (!f) {
+		_e("%#N: type '%T' has no member named %N.\n", n, t, member);
+		return -EINVAL;
 	}
 
-	_e("%#N: type '%T' has no member named %N.\n", n, t, member);
-	return -EINVAL;
+	/* given `sou.member` where sou is a struct/union, infer that
+	 * the expression's type is equal to member's type. */
+	n->sym->type = f->type;
+	return 0;
 }
 
 
@@ -69,20 +147,19 @@ static int global_deref_ir_post(const struct func *func, struct node *n,
 		 * from our argument */
 		return 0;
 
-	if (dst->loc)
-		return 0;
+	ir_init_sym(prog->ir, n->sym);
 
-	size = type_sizeof(type_base(ptr->sym->type)->ptr.type);
-	/* parent might have already allocated stack for us. typically
-	 * when the value is one part of a map key or value. */
-	if (!dst->hint.stack)
-		irs_alloc_stack(dst, prog->ir, size);
+	if (dst->hint.lval)
+		/* *ptr = val, whatever is in our storage now it will
+                    be overwritten, so skip the load. */
+		return 0;
 
 	ir_emit_insn(prog->ir, MOV(0, 0), BPF_REG_1, BPF_REG_BP);
 	ir_emit_insn(prog->ir, ALU_IMM(BPF_ADD, 0, dst->stack), BPF_REG_1, 0);
-	ir_emit_insn(prog->ir, MOV_IMM(0, (int32_t)size), BPF_REG_2, 0);
-	ir_emit_mov_irs(prog->ir, BPF_REG_3, &ptr->sym->irs);
+	ir_emit_insn(prog->ir, MOV_IMM(0, (int32_t)dst->size), BPF_REG_2, 0);
+	ir_emit_sym_to_reg(prog->ir, BPF_REG_3, ptr->sym);
 	ir_emit_insn(prog->ir, CALL(BPF_FUNC_probe_read), 0, 0);
+	/* TODO if (r0) exit(r0); */
 	return 0;
 }
 
@@ -208,7 +285,7 @@ static int global_assign_type_infer(const struct func *func, struct node *n)
 		/* TODO do we need assignment expressions? */
 		n->sym->type = &t_void;
 		
-		if (node_is_map(lval))
+		if (node_is(lval, "{}"))
 			return global_assign_type_infer_map(lval);
 
 		return 0;
@@ -229,7 +306,7 @@ static int global_assign_static_validate(const struct func *func, struct node *n
 
 	lval = n->expr.args;
 
-	if (node_is_map(lval) || (lval->ntype == N_IDENT))
+	if (node_is(lval, "{}") || (lval->ntype == N_IDENT))
 		return 0;
 
 	_e("%#N: can't assign a value to %N.\n", n, lval);
@@ -285,6 +362,19 @@ static int global_quantize_type_infer(const struct func *func, struct node *n)
 
 /* pid */
 
+static int global_pid_ir_post(const struct func *func, struct node *n,
+				struct prog *prog)
+{
+	struct node *ptr = n->expr.args;
+
+	ir_init_sym(prog->ir, n->sym);
+
+	ir_emit_insn(prog->ir, CALL(BPF_FUNC_get_current_pid_tgid), 0, 0);
+	ir_emit_insn(prog->ir, ALU64_IMM(BPF_RSH, 0, 32), BPF_REG_0, 0);
+	ir_emit_reg_to_sym(prog->ir, n->sym, BPF_REG_0);
+	return 0;
+}
+
 struct type t_pid = {
 	.ttype = T_TYPEDEF,
 
@@ -299,6 +389,18 @@ struct type t_pid_func = {
 
 /* time */
 
+static int global_time_ir_post(const struct func *func, struct node *n,
+			       struct prog *prog)
+{
+	struct node *ptr = n->expr.args;
+
+	ir_init_sym(prog->ir, n->sym);
+
+	ir_emit_insn(prog->ir, CALL(BPF_FUNC_ktime_get_ns), 0, 0);
+	ir_emit_reg_to_sym(prog->ir, n->sym, BPF_REG_0);
+	return 0;
+}
+
 struct type t_time = {
 	.ttype = T_TYPEDEF,	/* TODO: should be a T_FUNC with a static
 				 * signature */
@@ -379,6 +481,9 @@ static const struct func global_funcs[] = {
 		.name = ".",
 		.type = &t_dot_func,
 		.type_infer = global_dot_type_infer,
+
+		.ir_pre  = global_dot_ir_pre,
+		.ir_post = global_dot_ir_post,
 	},
 	{
 		.name = ":deref",
@@ -416,11 +521,15 @@ static const struct func global_funcs[] = {
 		.name = "pid",
 		.type = &t_pid_func,
 		.static_ret = 1,
+
+		.ir_post = global_pid_ir_post,
 	},
 	{
 		.name = "time",
 		.type = &t_time_func,
 		.static_ret = 1,
+
+		.ir_post = global_time_ir_post,
 	},
 
 	{

ir.c

@@ -7,6 +7,7 @@
 
 #include "ir.h"
 #include "sym.h"
+#include "type.h"
 
 const uint16_t vreg_base = 0x8000;
 
@@ -331,24 +332,81 @@ void ir_emit_label  (struct ir *ir, int16_t label)
 /* 	ir_emit(ir, &vi); */
 /* } */
 
-void ir_emit_mov_irs(struct ir *ir, uint16_t dst, struct irstate *src)
+void ir_emit_sym_to_reg(struct ir *ir, uint16_t dst, struct sym *src)
 {
-	switch (src->loc) {
+	struct irstate *irs = &src->irs;
+
+	switch (irs->loc) {
 	case LOC_IMM:
-		ir_emit_insn(ir, MOV_IMM(0, src->imm), dst, 0);
+		ir_emit_insn(ir, MOV_IMM(0, irs->imm), dst, 0);
 		break;
 	case LOC_REG:
-		ir_emit_insn(ir, MOV(0, 0), dst, src->reg);
+		ir_emit_insn(ir, MOV(0, 0), dst, irs->reg);
 		break;
 	case LOC_STACK:
-		ir_emit_insn(ir, LDX(0, src->stack, 0), dst, BPF_REG_BP);
+		switch (irs->size) {
+		case 1:
+			ir_emit_insn(ir, LDXB(0, irs->stack, 0),
+				     dst, BPF_REG_BP);
+			break;
+		case 2:
+			ir_emit_insn(ir, LDXH(0, irs->stack, 0),
+				     dst, BPF_REG_BP);
+			break;
+		case 4:
+			ir_emit_insn(ir, LDXW(0, irs->stack, 0),
+				     dst, BPF_REG_BP);
+			break;
+		case 8:
+			ir_emit_insn(ir, LDXDW(0, irs->stack, 0),
+				     dst, BPF_REG_BP);
+			break;
+		default:
+			assert(0);
+		}
 		break;
+	default:
+		assert(0);
+	}
+}
 
+void ir_emit_reg_to_sym(struct ir *ir, struct sym *dst, uint16_t src)
+{
+	struct irstate *irs = &dst->irs;
+
+	switch (irs->loc) {
+	case LOC_REG:
+		ir_emit_insn(ir, MOV(0, 0), irs->reg, src);
+		break;
+	case LOC_STACK:
+		switch (irs->size) {
+		case 1:
+			ir_emit_insn(ir, STXB(0, irs->stack, 0),
+				     BPF_REG_BP, src);
+			break;
+		case 2:
+			ir_emit_insn(ir, STXH(0, irs->stack, 0),
+				     BPF_REG_BP, src);
+			break;
+		case 4:
+			ir_emit_insn(ir, STXW(0, irs->stack, 0),
+				     BPF_REG_BP, src);
+			break;
+		case 8:
+			ir_emit_insn(ir, STXDW(0, irs->stack, 0),
+				     BPF_REG_BP, src);
+			break;
+		default:
+			assert(0);
+		}
+		break;
 	default:
 		assert(0);
 	}
 }
 
+/* void ir_emit_xfer(struct ir *ir,  */
+
 int16_t ir_alloc_label (struct ir *ir)
 {
 	return ir->next_label--;
@@ -367,6 +425,33 @@ ssize_t ir_alloc_stack(struct ir *ir, size_t size)
 	return ir->sp;
 }
 
+void ir_init_sym(struct ir *ir, struct sym *sym)
+{
+	struct irstate *irs = &sym->irs;
+	struct type *t = type_base(sym->type);
+
+	if (irs->loc)
+		return;
+
+	irs->size = type_sizeof(t);
+
+	if (irs->hint.stack)
+		goto alloc_stack;
+
+	switch (t->ttype) {
+	case T_SCALAR:
+	case T_POINTER:
+		irs->loc = LOC_REG;
+		irs->reg = ir_alloc_register(ir);
+		return;
+
+	default:
+	alloc_stack:
+		irs->loc = LOC_STACK;
+		irs->stack = ir_alloc_stack(ir, irs->size);
+	}
+}
+
 struct ir *ir_new(void)
 {
 	struct ir *ir;

ir.h

@@ -106,12 +106,14 @@ enum irloc {
 struct irstate {
 	int loc;
 
-	uint16_t reg;
+	size_t   size;
 	int32_t  stack;
 	int32_t  imm;
+	uint16_t reg;
 
 	struct {
 		int dot:1;
+		int lval:1;
 		int stack:1;
 	} hint;
 };
@@ -124,27 +126,16 @@ int16_t  ir_alloc_label   (struct ir *ir);
 uint16_t ir_alloc_register(struct ir *ir);
 ssize_t  ir_alloc_stack   (struct ir *ir, size_t size);
 
+void ir_init_sym(struct ir *ir, struct sym *sym);
+
 void ir_emit_insn   (struct ir *ir, struct bpf_insn bpf, uint16_t dst, uint16_t src);
 void ir_emit_ldmap  (struct ir *ir, uint16_t dst, struct sym *map);
 void ir_emit_label  (struct ir *ir, int16_t label);
 /* void ir_emit_reg_get(struct ir *ir, uint16_t reg); */
 /* void ir_emit_reg_put(struct ir *ir, uint16_t reg); */
-void ir_emit_mov_irs(struct ir *ir, uint16_t dst, struct irstate *src);
+void ir_emit_sym_to_reg(struct ir *ir, uint16_t dst, struct sym *src);
+void ir_emit_reg_to_sym(struct ir *ir, struct sym *dst, uint16_t src);
 
 struct ir *ir_new(void);
 
-
-static inline void irs_alloc_register(struct irstate *irs, struct ir *ir)
-{
-	irs->loc = LOC_REG;
-	irs->reg = ir_alloc_register(ir);
-}
-
-static inline void irs_alloc_stack(struct irstate *irs, struct ir *ir,
-				   size_t size)
-{
-	irs->loc = LOC_STACK;
-	irs->stack = ir_alloc_stack(ir, size);
-}
-
 #endif	/* _PLY_IR_H */

kprobe.c

@@ -88,7 +88,7 @@ static int kprobe_ir_pre(struct prog *prog)
 
 	symtab_foreach(prog->locals, sym) {
 		if ((*sym)->name && !strcmp((*sym)->name, "regs")) {
-			irs_alloc_register(&(*sym)->irs, prog->ir);
+			ir_init_sym(prog->ir, *sym);
 
 			/* kernel sets r1 to the address of the
 			 * pt_regs struct, which ply denotes as

node.c

@@ -97,20 +97,13 @@ int node_replace(struct node *n, struct node *new)
 
 /* helpers */
 
-int node_is_block(struct node *n)
+int node_is(struct node *n, const char *func)
 {
 	if (!n || (n->ntype != N_EXPR))
 		return 0;
 
-	return !strcmp(n->expr.func, ":block");
-}
-
-int node_is_map(struct node *n)
-{
-	if (!n || (n->ntype != N_EXPR))
-		return 0;
+	return !strcmp(n->expr.func, func);
 
-	return !strcmp(n->expr.func, "{}");
 }
 
 

node.h

@@ -71,8 +71,7 @@ static inline int node_nargs(struct node *n)
 	return nargs;
 }
 
-int node_is_block(struct node *n);
-int node_is_map  (struct node *n);
+int node_is(struct node *n, const char *func);
 
 #define node_expr_foreach(_expr, _arg) \
 	for ((_arg) = (_expr)->expr.args; (_arg); (_arg) = (_arg)->next)

ply.c

@@ -71,17 +71,17 @@ struct ctx *ctx_get(void)
 	prog->probe = "k:SyS_read";
 
 	/* { 
-	 * 	us = pid();
+	 * 	# us = pid();
 	 * 	t[0, pid()] = time();
 	 * 	reads[pid()] = quantize(arg2);
 	 * }
 	 */
 	prog->ast =
 		node_expr(":block",
-			  node_expr("=",
-				    node_ident("us"),
-				    node_expr("pid", NULL),
-				    NULL),
+			  /* node_expr("=", */
+			  /* 	    node_ident("us"), */
+			  /* 	    node_expr("pid", NULL), */
+			  /* 	    NULL), */
 			  node_expr("=",
 				    node_expr("{}",
 					      node_ident("t"),

type.c

@@ -327,7 +327,17 @@ ssize_t type_sizeof(struct type *t)
 	return -EINVAL;
 }
 
+struct tfield *tfields_get(struct tfield *fields, const char *name)
+{
+	struct tfield *f;
 
+	tfields_foreach(f, fields) {
+		if (!strcmp(f->name, name))
+			return f;
+	}
+
+	return NULL;
+}
 
 int all_types_cmp(const void *_a, const void *_b)
 {

type.h

@@ -38,6 +38,8 @@ struct tfield {
 #define tfields_foreach(_f, _fields) \
 	for ((_f) = (_fields); (_f)->type; (_f)++)
 
+struct tfield *tfields_get(struct tfield *fields, const char *name);
+
 struct tstruct {
 	char *name;
 	struct tfield *fields;

utils.c

@@ -19,7 +19,7 @@ static int __ast_fprint_pre(struct node *n, void *_info)
 	if (n->prev)
 		fputc(' ', info->fp);
 
-	if (node_is_block(n->up)) {
+	if (node_is(n->up, ":block")) {
 		info->indent += 4;
 		fprintf(info->fp, "\n%*s", info->indent, "");
 	}
@@ -41,13 +41,13 @@ static int __ast_fprint_post(struct node *n, void *_info)
 	struct ast_fprint_info *info = _info;
 	struct node *c;
 
-	if (node_is_block(n))
+	if (node_is(n, ":block"))
 		fprintf(info->fp, "\n%*s", info->indent, "");
 
 	if (n->ntype == N_EXPR)
 		fputc(')', info->fp);
 
-	if (node_is_block(n->up))
+	if (node_is(n->up, ":block"))
 		info->indent -= 4;
 	
 	return 0;