rpnifs/rpn_jit.c

/*
 * Copyright (C) 2020 Weber Yann
 * 
 * This file is part of pyrpn.
 * 
 * pyrpn is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * any later version.
 * 
 * pyrpn is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 * 
 * You should have received a copy of the GNU General Public License
 * along with pyrpn.  If not, see <http://www.gnu.org/licenses/>.
 */
#include "rpn_jit.h"

int rpn_expr_init(rpn_expr_t* expr, const unsigned char stack_sz,
	  const size_t args_count)
{
#ifdef DEBUG
	if(!expr)
	{
		dprintf(2, "Error, NULL ptr given as expression to rpn_expr_init");
		errno = EINVAL;
		return -1;
	}
#endif
	bzero(expr, sizeof(rpn_expr_t));

	expr->stack_sz = stack_sz;
	expr->args_count = args_count;

	expr->state = RPN_SOURCE;
	memset(expr->err_reason, (int)'\0', 128);

	expr->stack = malloc(sizeof(unsigned long) * stack_sz);
	if(!expr->stack)
	{
		snprintf(expr->err_reason, 128,
			"Unable to malloc stack : %s", strerror(errno));
		expr->state = RPN_ERROR;
		return -1;
	}
	bzero(expr->stack, sizeof(unsigned long) * stack_sz);

	if(_rpn_expr_init_map(expr) < 0)
	{
		snprintf(expr->err_reason, 128, 
			"Unable to init code map : %s", strerror(errno));
		free(expr->expr);
		expr->state = RPN_ERROR;
		return -1;
	}
	return 0;
}

int rpn_expr_compile(rpn_expr_t *expr, const char *code)
{
#ifdef DEBUG
	if(!expr)
	{
		dprintf(2, "Error, NULL ptr given as expression to rpn_expr_compile");
		errno = EINVAL;
		return -1;
	}
#endif
	expr->expr = strdup(code);
	if(!expr->expr)
	{
		snprintf(expr->err_reason, 128, 
			"Unable to strdup expression : %s", strerror(errno));
		expr->state = RPN_ERROR;
		return -1;
	}
	return _rpn_expr_compile_expr(expr);
}

int rpn_expr_untokenize(rpn_expr_t *expr, rpn_tokenized_t *tokens, char long_op)
{
	int err;
	size_t i;


	errno = 0;
	#ifdef DEBUG
	if(!expr)
	{
		dprintf(2, "Error, NULL ptr given as expression to rpn_expr_untokenize");
		err = EINVAL;
		goto ret_err;
	}
	if(tokens->argc != expr->args_count)
	{
		/* even if it should work with tokens->argc < expr->args_count */
		snprintf(expr->err_reason, 128,
			"Expression argc differ from tokenized version");
		err = EINVAL;
		goto ret_err;
	}
	#endif
	if(!(expr->expr = rpn_tokenized_expr(tokens, long_op)))
	{
		err = errno;
		snprintf(expr->err_reason, 128,
			"Error reading tokenized expression : %s",
			strerror(err));
		goto ret_err;
	}

	for(i=0; i<tokens->tokens_sz; i++)
	{
		if(_rpn_expr_token_copy(expr, &(tokens->tokens[i])) < 0)
		{
			err = errno;
			if(errno == EINVAL)
			{
				dprintf(2,
"Fatal error, unknown token type : %d.\nMemory corruption ?\n",
					tokens->tokens[i].type);
				exit(1);
			}
			snprintf(expr->err_reason, 128,
				"Untokenize error : %s",
				strerror(err));
			goto ret_err;
		}
	}

	return 0;

	ret_err:
		errno = err;
		return -1;
}

char* rpn_random(size_t op_sz, size_t args_count)
{
	double step;
	size_t i, buff_sz, offset, rnd;
	char *buff, *cur;
	unsigned char op_n;
	unsigned long int seed, rnd_val;
	int nchr, err;

	buff_sz = offset = 0;
	buff = NULL;
	step = 1.0 / (rpn_op_sz() + (args_count>0?2:1)); // + args and values
	
	if(getrandom(&seed, sizeof(long int), 0) < 0)
	{
		err=errno;
		perror("Fails to get a random number from kernel");
		errno=err;
		return NULL;
	}
	srand48(seed);

	for(i=0; i<op_sz; i++)
	{
		if(buff_sz - offset < 21)
		{
			buff_sz += 40;
			cur = realloc(buff, sizeof(char) * buff_sz);
			if(!cur)
			{
				err=errno;
				perror("Error allocating random expression");
				errno=err;
				return NULL;
			}
			buff=cur;
		}
		cur = buff + offset;
		*cur = '\0';
		op_n = drand48() / step;
		if(op_n < rpn_op_sz())
		{
			cur[0] = rpn_ops[op_n].chr;
			cur[1] = ' ';
			cur[2] = '\0';
			offset += 2;
		}
		else if(op_n == rpn_op_sz())
		{

			if(getrandom(&rnd_val, sizeof(long int), 0) < 0)
			{
				err=errno;
				perror("Fails to get a random number for value");
				errno=err;
				return NULL;
			}
			// values
			if((nchr = sprintf(cur, "0x%lX ", rnd_val)) < 0)
			{
				err=errno;
				perror("Error while sprintf arguments in random generator");
				errno=err;
				return NULL;
			}
			offset += nchr;
		}
		else
		{
			rnd = drand48() / (1.0 / args_count);
			// arguments
			if((nchr = sprintf(cur, "A%ld ", rnd)) < 0)
			{
				err=errno;
				perror("Error while sprintf arguments in random generator");
				errno=err;
				return NULL;
			}
			offset += nchr;
		}
	}
	buff[offset] = '\0';
	return buff;
}

int _rpn_expr_compile_expr(rpn_expr_t* expr)
{
	rpn_tokenizer_t tokenizer;
	rpn_token_t *token;

	
	if(expr->state == RPN_ERROR)
	{
		goto err;
	}

	if(rpn_tokenizer_start(&tokenizer, &(expr->toks), expr->expr,
		expr->args_count) < 0)
	{
		snprintf(expr->err_reason, 128,
			"Error starting tokenizer : %s",
			tokenizer.err_reason);
		goto err;
	}

	while((token = rpn_tok(&tokenizer)))
	{
		if(_rpn_expr_token_copy(expr, token) < 0)
		{
			if(errno == EINVAL)
			{
				dprintf(2,
"Fatal error, unknown token type : %d chr %ld.\nMemory corruption ?\n",
					token->type, tokenizer.chr_no);
				exit(1);
			}
			snprintf(expr->err_reason, 128,
"Compilation error on chr %ld, unable to copy code part : %s",
				tokenizer.chr_no, strerror(errno));
			goto err;
		}
	}
	if(rpn_tokenizer_error(&tokenizer))
	{
		snprintf(expr->err_reason, 128,
			"Compilation error, chr %ld : %s",
			tokenizer.chr_no, tokenizer.err_reason);
		goto err;
	}

	if(_rpn_expr_end_map(expr))
	{
		snprintf(expr->err_reason, 128,
			"Error ending code map : %s",
			strerror(errno));
		expr->state = RPN_ERROR;
		return -1;
	}

	expr->state = RPN_READY;
	return 0;
	err:
		expr->state = RPN_ERROR;
		return -1;
}

int _rpn_expr_compile_tokens(rpn_expr_t* expr)
{
	size_t i;
	rpn_token_t *token;
	for(i=0; i<expr->toks.tokens_sz; i++)
	{
		token = &(expr->toks.tokens[i]);
		if(_rpn_expr_token_copy(expr, token) < 0)
		{
			if(errno == EINVAL)
			{
				dprintf(2,
"Fatal error, unknown token type : %d\nMemory corruption ?\n",
					token->type);
				exit(1);
			}
			snprintf(expr->err_reason, 128,
			"Compilation error, unable to copy code part : %s",
				strerror(errno));
			expr->state = RPN_ERROR;
			return -1;
		}

	}

	if(_rpn_expr_end_map(expr))
	{
		snprintf(expr->err_reason, 128,
			"Error ending code map : %s",
			strerror(errno));
		expr->state = RPN_ERROR;
		return -1;
	}

	expr->state = RPN_READY;
	return 0;
}

unsigned long rpn_expr_eval(rpn_expr_t *expr, unsigned long *args)
{
	rpn_run_f expr_run;
	unsigned long int res;
	if(expr->state == RPN_ERROR)
	{
		return 0;
	}
	expr_run = expr->code_map;
	res = expr_run(expr->stack_sz, args, expr->stack);
	return res;
}

void rpn_expr_close(rpn_expr_t* expr)
{
	if(expr->expr)
	{
		free(expr->expr);
		expr->expr = NULL;
	}
	if(expr->stack)
	{
		free(expr->stack);
		expr->stack = NULL;
	}
	if(expr->code_map)
	{
		if(munmap(expr->code_map, expr->code_map_sz))
		{
			perror("Unable to unmap code_map");
		}
		expr->code_map = NULL;
	}
}

void rpn_expr_reset_stack(rpn_expr_t *expr)
{
	bzero(expr->stack, sizeof(unsigned long) * expr->stack_sz);
}

int _rpn_expr_token_copy(rpn_expr_t *expr, rpn_token_t *token)
{
	unsigned long int *value;
	rpn_op_t local_op;
	value = NULL;
	switch(token->type)
	{
		case RPN_op:
			local_op = *(token->op);
			value = NULL;
			break;
		case RPN_arg:
			local_op.fun = &rpn_arg;
			local_op.fun_sz = &(CODE_SZ(rpn_arg));
			value = &(token->arg_n);
			break;
		case RPN_val:
			local_op.fun = &rpn_value;
			local_op.fun_sz = &(CODE_SZ(rpn_value));
			value = &(token->value);
			break;
		default:
			errno = EINVAL;
			return -1;
	}
	if(_rpn_code_part_cpy(expr, local_op.fun, *(local_op.fun_sz),
		value))
	{
		return -1;
	}
	return 0;
}

int _rpn_code_part_cpy(rpn_expr_t *expr, const void *code_part,
	unsigned long code_part_sz, const unsigned long *value)
{
	size_t old_sz, code_sz;
	void *new_ptr;
//printf("DEBUG _copy : %p %ld %p:%ld\n", code_part, code_part_sz, value, value?*value:0);
	code_sz = expr->code_map_ptr - expr->code_map;
	if(!expr->code_map_sz)
	{
		errno = EINVAL;
		return -1;
	}
	if(code_sz + code_part_sz >= expr->code_map_sz)
	{
		old_sz = expr->code_map_sz;
		expr->code_map_sz = (((code_sz + code_part_sz)>>9)+1)<<9;
		new_ptr = mremap(expr->code_map, old_sz, expr->code_map_sz,
			MREMAP_MAYMOVE);
		if(new_ptr == (void*)-1)
		{
			expr->code_map_sz = 0;
			return -1;
		}
		expr->code_map = new_ptr;
		expr->code_map_ptr = expr->code_map + code_sz;
	}
	memcpy(expr->code_map_ptr, code_part, code_part_sz);
	
	if(value)
	{
		// set 1st instruction argument
		*(unsigned long*)(expr->code_map_ptr + 2) = *value;
	}

	expr->code_map_ptr += code_part_sz;
	return 0;
}

int _rpn_expr_init_map(rpn_expr_t* expr)
{
	expr->code_map_sz = RPN_MAP_CHUNK;
	expr->code_map = mmap(NULL, expr->code_map_sz, PROT_READ | PROT_WRITE,
		MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
	if(!expr->code_map)
	{
		return -1;
	}
	expr->code_map_ptr = expr->code_map;
	if(CODE_PART_CPY(expr, rpn_exec))
	{
		return -1;
	}
	return 0;
}

int _rpn_expr_end_map(rpn_expr_t *expr)
{
	if(CODE_PART_CPY(expr, rpn_exec_ret))
	{
		return -1;
	}
	if(mprotect(expr->code_map, expr->code_map_ptr - expr->code_map,
		PROT_EXEC))
	{
		return -1;
	}
	return 0;
}