dt_parser.c

Sun Solaris 10 中的 DTrace 组件的源代码。请参看: http://www.sun.com/software/solaris/observability.jsp

/*
 * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License, Version 1.0 only.
 * See the file usr/src/LICENSING.NOTICE in this distribution or
 * http://www.opensolaris.org/license/ for details.
 */

#pragma ident	"@(#)dt_parser.c	1.11	04/12/17 SMI"

/*
 * DTrace D Language Parser
 *
 * The D Parser is a lex/yacc parser consisting of the lexer dt_lex.l, the
 * parsing grammar dt_grammar.y, and this file, dt_parser.c, which handles
 * the construction of the parse tree nodes and their syntactic validation.
 * The parse tree is constructed of dt_node_t structures (see <dt_parser.h>)
 * that are built in two passes: (1) the "create" pass, where the parse tree
 * nodes are allocated by calls from the grammar to dt_node_*() subroutines,
 * and (2) the "cook" pass, where nodes are coalesced, assigned D types, and
 * validated according to the syntactic rules of the language.
 *
 * All node allocations are performed using dt_node_alloc().  All node frees
 * during the parsing phase are performed by dt_node_free(), which frees node-
 * internal state but does not actually free the nodes.  All final node frees
 * are done as part of the end of dt_compile() or as part of destroying
 * persistent identifiers or translators which have embedded nodes.
 *
 * The dt_node_* routines that implement pass (1) may allocate new nodes.  The
 * dt_cook_* routines that implement pass (2) may *not* allocate new nodes.
 * They may free existing nodes using dt_node_free(), but they may not actually
 * deallocate any dt_node_t's.  Currently dt_cook_op2() is an exception to this
 * rule: see the comments therein for how this issue is resolved.
 *
 * The dt_cook_* routines are responsible for (at minimum) setting the final
 * node type (dn_ctfp/dn_type) and attributes (dn_attr).  If dn_ctfp/dn_type
 * are set manually (i.e. not by one of the type assignment functions), then
 * the DT_NF_COOKED flag must be set manually on the node.
 *
 * The cooking pass can be applied to the same parse tree more than once (used
 * in the case of a comma-separated list of probe descriptions).  As such, the
 * cook routines must not perform any parse tree transformations which would
 * be invalid if the tree were subsequently cooked using a different context.
 *
 * The dn_ctfp and dn_type fields form the type of the node.  This tuple can
 * take on the following set of values, which form our type invariants:
 *
 * 1. dn_ctfp = NULL, dn_type = CTF_ERR
 *
 *    In this state, the node has unknown type and is not yet cooked.  The
 *    DT_NF_COOKED flag is not yet set on the node.
 *
 * 2. dn_ctfp = DT_DYN_CTFP(dtp), dn_type = DT_DYN_TYPE(dtp)
 *
 *    In this state, the node is a dynamic D type.  This means that general
 *    operations are not valid on this node and only code that knows how to
 *    examine the inner details of the value can operate on it.  When a node is
 *    of type <DYN>, dn_ident *must* be valid on the node and it points to an
 *    identifier describing the object and its type.  The DT_NF_REF flag is
 *    set for all dt_node_t objects of type <DYN>.
 *
 * 3. dn_ctfp = DT_STR_CTFP(dtp), dn_type = DT_STR_TYPE(dtp)
 *
 *    In this state, the node is of type D string.  The string type is really
 *    a char[0] typedef, but requires special handling throughout the compiler.
 *
 * 4. dn_ctfp != NULL, dn_type = any other type ID
 *
 *    In this state, the node is of some known D/CTF type.  The normal libctf
 *    APIs can be used to learn more about the type name or structure.  When
 *    the type is assigned, the DT_NF_SIGNED, DT_NF_REF, and DT_NF_BITFIELD
 *    flags cache the corresponding attributes of the underlying CTF type.
 */

#include <sys/param.h>
#include <limits.h>
#include <setjmp.h>
#include <strings.h>
#include <assert.h>
#include <alloca.h>
#include <stdlib.h>
#include <stdarg.h>
#include <stdio.h>
#include <errno.h>
#include <ctype.h>

#include <dt_impl.h>
#include <dt_grammar.h>
#include <dt_module.h>
#include <dt_provider.h>
#include <dt_string.h>
#include <dt_as.h>

dt_pcb_t *yypcb;	/* current control block for parser */
dt_node_t *yypragma;	/* lex token list for control lines */
char yyintprefix;	/* int token macro prefix (+/-) */
char yyintsuffix[4];	/* int token suffix string [uU][lL] */
int yyintdecimal;	/* int token format flag (1=decimal, 0=octal/hex) */

static const char *
opstr(int op)
{
	switch (op) {
	case DT_TOK_COMMA:	return (",");
	case DT_TOK_ELLIPSIS:	return ("...");
	case DT_TOK_ASGN:	return ("=");
	case DT_TOK_ADD_EQ:	return ("+=");
	case DT_TOK_SUB_EQ:	return ("-=");
	case DT_TOK_MUL_EQ:	return ("*=");
	case DT_TOK_DIV_EQ:	return ("/=");
	case DT_TOK_MOD_EQ:	return ("%=");
	case DT_TOK_AND_EQ:	return ("&=");
	case DT_TOK_XOR_EQ:	return ("^=");
	case DT_TOK_OR_EQ:	return ("|=");
	case DT_TOK_LSH_EQ:	return ("<<=");
	case DT_TOK_RSH_EQ:	return (">>=");
	case DT_TOK_QUESTION:	return ("?");
	case DT_TOK_COLON:	return (":");
	case DT_TOK_LOR:	return ("||");
	case DT_TOK_LXOR:	return ("^^");
	case DT_TOK_LAND:	return ("&&");
	case DT_TOK_BOR:	return ("|");
	case DT_TOK_XOR:	return ("^");
	case DT_TOK_BAND:	return ("&");
	case DT_TOK_EQU:	return ("==");
	case DT_TOK_NEQ:	return ("!=");
	case DT_TOK_LT:		return ("<");
	case DT_TOK_LE:		return ("<=");
	case DT_TOK_GT:		return (">");
	case DT_TOK_GE:		return (">=");
	case DT_TOK_LSH:	return ("<<");
	case DT_TOK_RSH:	return (">>");
	case DT_TOK_ADD:	return ("+");
	case DT_TOK_SUB:	return ("-");
	case DT_TOK_MUL:	return ("*");
	case DT_TOK_DIV:	return ("/");
	case DT_TOK_MOD:	return ("%");
	case DT_TOK_LNEG:	return ("!");
	case DT_TOK_BNEG:	return ("~");
	case DT_TOK_ADDADD:	return ("++");
	case DT_TOK_PREINC:	return ("++");
	case DT_TOK_POSTINC:	return ("++");
	case DT_TOK_SUBSUB:	return ("--");
	case DT_TOK_PREDEC:	return ("--");
	case DT_TOK_POSTDEC:	return ("--");
	case DT_TOK_IPOS:	return ("+");
	case DT_TOK_INEG:	return ("-");
	case DT_TOK_DEREF:	return ("*");
	case DT_TOK_ADDROF:	return ("&");
	case DT_TOK_OFFSETOF:	return ("offsetof");
	case DT_TOK_SIZEOF:	return ("sizeof");
	case DT_TOK_STRINGOF:	return ("stringof");
	case DT_TOK_XLATE:	return ("xlate");
	case DT_TOK_LPAR:	return ("(");
	case DT_TOK_RPAR:	return (")");
	case DT_TOK_LBRAC:	return ("[");
	case DT_TOK_RBRAC:	return ("]");
	case DT_TOK_PTR:	return ("->");
	case DT_TOK_DOT:	return (".");
	case DT_TOK_STRING:	return ("<string>");
	case DT_TOK_IDENT:	return ("<ident>");
	case DT_TOK_TNAME:	return ("<type>");
	case DT_TOK_INT:	return ("<int>");
	default:		return ("<?>");
	}
}

int
dt_type_lookup(const char *s, dtrace_typeinfo_t *tip)
{
	static const char delimiters[] = " \t\n\r\v\f*`";
	dtrace_hdl_t *dtp = yypcb->pcb_hdl;
	const char *p, *q, *end, *obj;

	for (p = s, end = s + strlen(s); *p != '\0'; p = q) {
		while (isspace(*p))
			p++;	/* skip leading whitespace prior to token */

		if (p == end || (q = strpbrk(p + 1, delimiters)) == NULL)
			break;	/* empty string or single token remaining */

		if (*q == '`') {
			char *object = alloca((size_t)(q - p) + 1);
			char *type = alloca((size_t)(end - s) + 1);

			/*
			 * Copy from the start of the token (p) to the location
			 * backquote (q) to extract the nul-terminated object.
			 */
			bcopy(p, object, (size_t)(q - p));
			object[(size_t)(q - p)] = '\0';

			/*
			 * Copy the original string up to the start of this
			 * token (p) into type, and then concatenate everything
			 * after q.  This is the type name without the object.
			 */
			bcopy(s, type, (size_t)(p - s));
			bcopy(q + 1, type + (size_t)(p - s), strlen(q + 1) + 1);

			if (strchr(q + 1, '`') != NULL)
				return (dt_set_errno(dtp, EDT_BADSCOPE));

			return (dtrace_lookup_by_type(dtp, object, type, tip));
		}
	}

	if (yypcb->pcb_idepth != 0)
		obj = DTRACE_OBJ_CDEFS;
	else
		obj = DTRACE_OBJ_EVERY;

	return (dtrace_lookup_by_type(dtp, obj, s, tip));
}

/*
 * When we parse type expressions or parse an expression with unary "&", we
 * need to find a type that is a pointer to a previously known type.
 * Unfortunately CTF is limited to a per-container view, so ctf_type_pointer()
 * alone does not suffice for our needs.  We provide a more intelligent wrapper
 * for the compiler that attempts to compute a pointer to either the given type
 * or its base (that is, we try both "foo_t *" and "struct foo *"), and also
 * to potentially construct the required type on-the-fly.
 */
int
dt_type_pointer(dtrace_typeinfo_t *tip)
{
	dtrace_hdl_t *dtp = yypcb->pcb_hdl;
	ctf_file_t *ctfp = tip->dtt_ctfp;
	ctf_id_t type = tip->dtt_type;
	ctf_id_t base = ctf_type_resolve(ctfp, type);

	dt_module_t *dmp;
	ctf_id_t ptr;

	if ((ptr = ctf_type_pointer(ctfp, type)) != CTF_ERR ||
	    (ptr = ctf_type_pointer(ctfp, base)) != CTF_ERR) {
		tip->dtt_type = ptr;
		return (0);
	}

	if (yypcb->pcb_idepth != 0)
		dmp = dtp->dt_cdefs;
	else
		dmp = dtp->dt_ddefs;

	if (ctfp != dmp->dm_ctfp && ctfp != ctf_parent_file(dmp->dm_ctfp) &&
	    (type = ctf_add_type(dmp->dm_ctfp, ctfp, type)) == CTF_ERR)
		return (-1); /* errno is set for us */

	ptr = ctf_add_pointer(dmp->dm_ctfp, CTF_ADD_ROOT, type);

	if (ptr == CTF_ERR || ctf_update(dmp->dm_ctfp) == CTF_ERR)
		return (-1); /* errno is set for us */

	tip->dtt_object = dmp->dm_name;
	tip->dtt_ctfp = dmp->dm_ctfp;
	tip->dtt_type = ptr;

	return (0);
}

const char *
dt_type_name(ctf_file_t *ctfp, ctf_id_t type, char *buf, size_t len)
{
	dtrace_hdl_t *dtp = yypcb->pcb_hdl;

	if (ctfp == DT_FPTR_CTFP(dtp) && type == DT_FPTR_TYPE(dtp))
		(void) snprintf(buf, len, "function pointer");
	else if (ctfp == DT_FUNC_CTFP(dtp) && type == DT_FUNC_TYPE(dtp))
		(void) snprintf(buf, len, "function");
	else if (ctfp == DT_DYN_CTFP(dtp) && type == DT_DYN_TYPE(dtp))
		(void) snprintf(buf, len, "dynamic variable");
	else if (ctfp == NULL)
		(void) snprintf(buf, len, "<none>");
	else if (ctf_type_name(ctfp, type, buf, len) == NULL)
		(void) snprintf(buf, len, "unknown");

	return (buf);
}

/*
 * Perform the "usual arithmetic conversions" to determine which of the two
 * input operand types should be promoted and used as a result type.  The
 * rules for this are described in ISOC[6.3.1.8] and K&R[A6.5].
 */
static void
dt_type_promote(dt_node_t *lp, dt_node_t *rp, ctf_file_t **ofp, ctf_id_t *otype)
{
	ctf_file_t *lfp = lp->dn_ctfp;
	ctf_id_t ltype = lp->dn_type;

	ctf_file_t *rfp = rp->dn_ctfp;
	ctf_id_t rtype = rp->dn_type;

	ctf_id_t lbase = ctf_type_resolve(lfp, ltype);
	uint_t lkind = ctf_type_kind(lfp, lbase);

	ctf_id_t rbase = ctf_type_resolve(rfp, rtype);
	uint_t rkind = ctf_type_kind(rfp, rbase);

	dtrace_hdl_t *dtp = yypcb->pcb_hdl;
	ctf_encoding_t le, re;
	uint_t lrank, rrank;

	assert(lkind == CTF_K_INTEGER || lkind == CTF_K_ENUM);
	assert(rkind == CTF_K_INTEGER || rkind == CTF_K_ENUM);

	if (lkind == CTF_K_ENUM) {
		lfp = DT_INT_CTFP(dtp);
		ltype = lbase = DT_INT_TYPE(dtp);
	}

	if (rkind == CTF_K_ENUM) {
		rfp = DT_INT_CTFP(dtp);
		rtype = rbase = DT_INT_TYPE(dtp);
	}

	if (ctf_type_encoding(lfp, lbase, &le) == CTF_ERR) {
		yypcb->pcb_hdl->dt_ctferr = ctf_errno(lfp);
		longjmp(yypcb->pcb_jmpbuf, EDT_CTF);
	}

	if (ctf_type_encoding(rfp, rbase, &re) == CTF_ERR) {
		yypcb->pcb_hdl->dt_ctferr = ctf_errno(rfp);
		longjmp(yypcb->pcb_jmpbuf, EDT_CTF);
	}

	/*
	 * Compute an integer rank based on the size and unsigned status.
	 * If rank is identical, pick the "larger" of the equivalent types
	 * which we define as having a larger base ctf_id_t.  If rank is
	 * different, pick the type with the greater rank.
	 */
	lrank = le.cte_bits + ((le.cte_format & CTF_INT_SIGNED) == 0);
	rrank = re.cte_bits + ((re.cte_format & CTF_INT_SIGNED) == 0);

	if (lrank == rrank) {
		if (lbase - rbase < 0)
			goto return_rtype;
		else
			goto return_ltype;
	} else if (lrank > rrank) {
		goto return_ltype;
	} else
		goto return_rtype;

return_ltype:
	*ofp = lfp;
	*otype = ltype;
	return;

return_rtype:
	*ofp = rfp;
	*otype = rtype;
}

void
dt_node_promote(dt_node_t *lp, dt_node_t *rp, dt_node_t *dnp)
{
	dt_type_promote(lp, rp, &dnp->dn_ctfp, &dnp->dn_type);
	dt_node_type_assign(dnp, dnp->dn_ctfp, dnp->dn_type);
	dt_node_attr_assign(dnp, dt_attr_min(lp->dn_attr, rp->dn_attr));
}

const char *
dt_node_name(const dt_node_t *dnp, char *buf, size_t len)
{
	char n1[DT_TYPE_NAMELEN];
	char n2[DT_TYPE_NAMELEN];

	const char *prefix = "", *suffix = "";
	const dtrace_syminfo_t *dts;
	char *s;

	switch (dnp->dn_kind) {
	case DT_NODE_INT:
		(void) snprintf(buf, len, "integer constant 0x%llx",
		    (u_longlong_t)dnp->dn_value);
		break;
	case DT_NODE_STRING:
		s = strchr2esc(dnp->dn_string, strlen(dnp->dn_string));
		(void) snprintf(buf, len, "string constant \"%s\"",
		    s != NULL ? s : dnp->dn_string);
		free(s);
		break;
	case DT_NODE_IDENT:
		(void) snprintf(buf, len, "identifier %s", dnp->dn_string);
		break;
	case DT_NODE_VAR:
	case DT_NODE_FUNC:
	case DT_NODE_AGG:
	case DT_NODE_INLINE:
		switch (dnp->dn_ident->di_kind) {
		case DT_IDENT_FUNC:
		case DT_IDENT_AGGFUNC:
		case DT_IDENT_ACTFUNC:
			suffix = "( )";
			break;
		case DT_IDENT_AGG:
			prefix = "@";
			break;
		}
		(void) snprintf(buf, len, "%s %s%s%s",
		    dt_idkind_name(dnp->dn_ident->di_kind),
		    prefix, dnp->dn_ident->di_name, suffix);
		break;
	case DT_NODE_SYM:
		dts = dnp->dn_ident->di_data;
		(void) snprintf(buf, len, "symbol %s`%s",
		    dts->dts_object, dts->dts_name);
		break;
	case DT_NODE_TYPE:
		(void) snprintf(buf, len, "type %s",
		    dt_node_type_name(dnp, n1, sizeof (n1)));
		break;
	case DT_NODE_OP1:
	case DT_NODE_OP2:
	case DT_NODE_OP3:
		(void) snprintf(buf, len, "operator %s", opstr(dnp->dn_op));
		break;
	case DT_NODE_DEXPR:
	case DT_NODE_DFUNC:
		if (dnp->dn_expr)
			return (dt_node_name(dnp->dn_expr, buf, len));
		(void) snprintf(buf, len, "%s", "statement");
		break;
	case DT_NODE_PDESC:
		if (dnp->dn_desc->dtpd_id == 0) {
			(void) snprintf(buf, len,
			    "probe description %s:%s:%s:%s",
			    dnp->dn_desc->dtpd_provider, dnp->dn_desc->dtpd_mod,
			    dnp->dn_desc->dtpd_func, dnp->dn_desc->dtpd_name);
		} else {
			(void) snprintf(buf, len, "probe description %u",
			    dnp->dn_desc->dtpd_id);
		}
		break;
	case DT_NODE_CLAUSE:
		(void) snprintf(buf, len, "%s", "clause");
		break;
	case DT_NODE_MEMBER:
		(void) snprintf(buf, len, "member %s", dnp->dn_membname);
		break;
	case DT_NODE_XLATOR:
		(void) snprintf(buf, len, "translator <%s> (%s)",
		    dt_type_name(dnp->dn_xlator->dx_dst_ctfp,
			dnp->dn_xlator->dx_dst_type, n1, sizeof (n1)),
		    dt_type_name(dnp->dn_xlator->dx_src_ctfp,
			dnp->dn_xlator->dx_src_type, n2, sizeof (n2)));
		break;
	case DT_NODE_PROG:
		(void) snprintf(buf, len, "%s", "program");
		break;
	default:
		(void) snprintf(buf, len, "node <%u>", dnp->dn_kind);
		break;
	}

	return (buf);
}

static dt_node_t *
dt_node_alloc(int kind)
{
	dt_node_t *dnp = malloc(sizeof (dt_node_t));

	if (dnp == NULL)
		longjmp(yypcb->pcb_jmpbuf, EDT_NOMEM);

	dnp->dn_ctfp = NULL;
	dnp->dn_type = CTF_ERR;