dt_module.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_module.c	1.8	04/11/22 SMI"

#include <sys/types.h>
#include <sys/modctl.h>
#include <sys/kobj.h>
#include <sys/kobj_impl.h>
#include <sys/sysmacros.h>
#include <sys/elf.h>
#include <sys/task.h>

#include <unistd.h>
#include <project.h>
#include <strings.h>
#include <stdlib.h>
#include <libelf.h>
#include <limits.h>
#include <assert.h>
#include <errno.h>
#include <alloca.h>

#define	_POSIX_PTHREAD_SEMANTICS
#include <dirent.h>
#undef	_POSIX_PTHREAD_SEMANTICS

#include <dt_strtab.h>
#include <dt_module.h>
#include <dt_impl.h>

static const char *dt_module_strtab; /* active strtab for qsort callbacks */

static void
dt_module_symhash_insert(dt_module_t *dmp, const char *name, uint_t id)
{
	dt_sym_t *dsp = &dmp->dm_symchains[dmp->dm_symfree];
	uint_t h;

	assert(dmp->dm_symfree < dmp->dm_nsymelems + 1);

	dsp->ds_symid = id;
	h = dt_strtab_hash(name, NULL) % dmp->dm_nsymbuckets;
	dsp->ds_next = dmp->dm_symbuckets[h];
	dmp->dm_symbuckets[h] = dmp->dm_symfree++;
}

static uint_t
dt_module_syminit32(dt_module_t *dmp)
{
	const Elf32_Sym *sym = dmp->dm_symtab.cts_data;
	const char *base = dmp->dm_strtab.cts_data;
	size_t ss_size = dmp->dm_strtab.cts_size;
	uint_t i, n = dmp->dm_nsymelems;
	uint_t asrsv = 0;

	for (i = 0; i < n; i++, sym++) {
		const char *name = base + sym->st_name;
		uchar_t type = ELF32_ST_TYPE(sym->st_info);

		if (type >= STT_NUM || type == STT_SECTION)
			continue; /* skip sections and unknown types */

		if (sym->st_name == 0 || sym->st_name >= ss_size)
			continue; /* skip null or invalid names */

		if (sym->st_value != 0 &&
		    (ELF32_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size))
			asrsv++; /* reserve space in the address map */

		dt_module_symhash_insert(dmp, name, i);
	}

	return (asrsv);
}

static uint_t
dt_module_syminit64(dt_module_t *dmp)
{
	const Elf64_Sym *sym = dmp->dm_symtab.cts_data;
	const char *base = dmp->dm_strtab.cts_data;
	size_t ss_size = dmp->dm_strtab.cts_size;
	uint_t i, n = dmp->dm_nsymelems;
	uint_t asrsv = 0;

	for (i = 0; i < n; i++, sym++) {
		const char *name = base + sym->st_name;
		uchar_t type = ELF64_ST_TYPE(sym->st_info);

		if (type >= STT_NUM || type == STT_SECTION)
			continue; /* skip sections and unknown types */

		if (sym->st_name == 0 || sym->st_name >= ss_size)
			continue; /* skip null or invalid names */

		if (sym->st_value != 0 &&
		    (ELF64_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size))
			asrsv++; /* reserve space in the address map */

		dt_module_symhash_insert(dmp, name, i);
	}

	return (asrsv);
}

/*
 * Sort comparison function for 32-bit symbol address-to-name lookups.  We sort
 * symbols by value.  If values are equal, we prefer the symbol that is
 * non-zero sized, typed, not weak, or lexically first, in that order.
 */
static int
dt_module_symcomp32(const void *lp, const void *rp)
{
	Elf32_Sym *lhs = *((Elf32_Sym **)lp);
	Elf32_Sym *rhs = *((Elf32_Sym **)rp);

	if (lhs->st_value != rhs->st_value)
		return (lhs->st_value > rhs->st_value ? 1 : -1);

	if ((lhs->st_size == 0) != (rhs->st_size == 0))
		return (lhs->st_size == 0 ? 1 : -1);

	if ((ELF32_ST_TYPE(lhs->st_info) == STT_NOTYPE) !=
	    (ELF32_ST_TYPE(rhs->st_info) == STT_NOTYPE))
		return (ELF32_ST_TYPE(lhs->st_info) == STT_NOTYPE ? 1 : -1);

	if ((ELF32_ST_BIND(lhs->st_info) == STB_WEAK) !=
	    (ELF32_ST_BIND(rhs->st_info) == STB_WEAK))
		return (ELF32_ST_BIND(lhs->st_info) == STB_WEAK ? 1 : -1);

	return (strcmp(dt_module_strtab + lhs->st_name,
	    dt_module_strtab + rhs->st_name));
}

/*
 * Sort comparison function for 64-bit symbol address-to-name lookups.  We sort
 * symbols by value.  If values are equal, we prefer the symbol that is
 * non-zero sized, typed, not weak, or lexically first, in that order.
 */
static int
dt_module_symcomp64(const void *lp, const void *rp)
{
	Elf64_Sym *lhs = *((Elf64_Sym **)lp);
	Elf64_Sym *rhs = *((Elf64_Sym **)rp);

	if (lhs->st_value != rhs->st_value)
		return (lhs->st_value > rhs->st_value ? 1 : -1);

	if ((lhs->st_size == 0) != (rhs->st_size == 0))
		return (lhs->st_size == 0 ? 1 : -1);

	if ((ELF64_ST_TYPE(lhs->st_info) == STT_NOTYPE) !=
	    (ELF64_ST_TYPE(rhs->st_info) == STT_NOTYPE))
		return (ELF64_ST_TYPE(lhs->st_info) == STT_NOTYPE ? 1 : -1);

	if ((ELF64_ST_BIND(lhs->st_info) == STB_WEAK) !=
	    (ELF64_ST_BIND(rhs->st_info) == STB_WEAK))
		return (ELF64_ST_BIND(lhs->st_info) == STB_WEAK ? 1 : -1);

	return (strcmp(dt_module_strtab + lhs->st_name,
	    dt_module_strtab + rhs->st_name));
}

static void
dt_module_symsort32(dt_module_t *dmp)
{
	Elf32_Sym *symtab = (Elf32_Sym *)dmp->dm_symtab.cts_data;
	Elf32_Sym **sympp = (Elf32_Sym **)dmp->dm_asmap;
	const dt_sym_t *dsp = dmp->dm_symchains + 1;
	uint_t i, n = dmp->dm_symfree;

	for (i = 1; i < n; i++, dsp++) {
		Elf32_Sym *sym = symtab + dsp->ds_symid;
		if (sym->st_value != 0 &&
		    (ELF32_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size))
			*sympp++ = sym;
	}

	dmp->dm_aslen = (uint_t)(sympp - (Elf32_Sym **)dmp->dm_asmap);
	assert(dmp->dm_aslen <= dmp->dm_asrsv);

	dt_module_strtab = dmp->dm_strtab.cts_data;
	qsort(dmp->dm_asmap, dmp->dm_aslen,
	    sizeof (Elf32_Sym *), dt_module_symcomp32);
	dt_module_strtab = NULL;
}

static void
dt_module_symsort64(dt_module_t *dmp)
{
	Elf64_Sym *symtab = (Elf64_Sym *)dmp->dm_symtab.cts_data;
	Elf64_Sym **sympp = (Elf64_Sym **)dmp->dm_asmap;
	const dt_sym_t *dsp = dmp->dm_symchains + 1;
	uint_t i, n = dmp->dm_symfree;

	for (i = 1; i < n; i++, dsp++) {
		Elf64_Sym *sym = symtab + dsp->ds_symid;
		if (sym->st_value != 0 &&
		    (ELF64_ST_BIND(sym->st_info) != STB_LOCAL || sym->st_size))
			*sympp++ = sym;
	}

	dmp->dm_aslen = (uint_t)(sympp - (Elf64_Sym **)dmp->dm_asmap);
	assert(dmp->dm_aslen <= dmp->dm_asrsv);

	dt_module_strtab = dmp->dm_strtab.cts_data;
	qsort(dmp->dm_asmap, dmp->dm_aslen,
	    sizeof (Elf64_Sym *), dt_module_symcomp64);
	dt_module_strtab = NULL;
}

static GElf_Sym *
dt_module_symgelf32(const Elf32_Sym *src, GElf_Sym *dst)
{
	if (dst != NULL) {
		dst->st_name = src->st_name;
		dst->st_info = src->st_info;
		dst->st_other = src->st_other;
		dst->st_shndx = src->st_shndx;
		dst->st_value = src->st_value;
		dst->st_size = src->st_size;
	}

	return (dst);
}

static GElf_Sym *
dt_module_symgelf64(const Elf64_Sym *src, GElf_Sym *dst)
{
	if (dst != NULL)
		bcopy(src, dst, sizeof (GElf_Sym));

	return (dst);
}

static GElf_Sym *
dt_module_symname32(dt_module_t *dmp, const char *name,
    GElf_Sym *symp, uint_t *idp)
{
	const Elf32_Sym *symtab = dmp->dm_symtab.cts_data;
	const char *strtab = dmp->dm_strtab.cts_data;

	const Elf32_Sym *sym;
	const dt_sym_t *dsp;
	uint_t i, h;

	if (dmp->dm_nsymelems == 0)
		return (NULL);

	h = dt_strtab_hash(name, NULL) % dmp->dm_nsymbuckets;

	for (i = dmp->dm_symbuckets[h]; i != 0; i = dsp->ds_next) {
		dsp = &dmp->dm_symchains[i];
		sym = symtab + dsp->ds_symid;

		if (strcmp(name, strtab + sym->st_name) == 0) {
			if (idp != NULL)
				*idp = dsp->ds_symid;
			return (dt_module_symgelf32(sym, symp));
		}
	}

	return (NULL);
}

static GElf_Sym *
dt_module_symname64(dt_module_t *dmp, const char *name,
    GElf_Sym *symp, uint_t *idp)
{
	const Elf64_Sym *symtab = dmp->dm_symtab.cts_data;
	const char *strtab = dmp->dm_strtab.cts_data;

	const Elf64_Sym *sym;
	const dt_sym_t *dsp;
	uint_t i, h;

	if (dmp->dm_nsymelems == 0)
		return (NULL);

	h = dt_strtab_hash(name, NULL) % dmp->dm_nsymbuckets;

	for (i = dmp->dm_symbuckets[h]; i != 0; i = dsp->ds_next) {
		dsp = &dmp->dm_symchains[i];
		sym = symtab + dsp->ds_symid;

		if (strcmp(name, strtab + sym->st_name) == 0) {
			if (idp != NULL)
				*idp = dsp->ds_symid;
			return (dt_module_symgelf64(sym, symp));
		}
	}

	return (NULL);
}

static GElf_Sym *
dt_module_symaddr32(dt_module_t *dmp, GElf_Addr addr,
    GElf_Sym *symp, uint_t *idp)
{
	const Elf32_Sym **asmap = (const Elf32_Sym **)dmp->dm_asmap;
	const Elf32_Sym *symtab = dmp->dm_symtab.cts_data;
	const Elf32_Sym *sym;

	uint_t i, mid, lo = 0, hi = dmp->dm_aslen - 1;
	Elf32_Addr v;

	if (dmp->dm_aslen == 0)
		return (NULL);

	while (hi - lo > 1) {
		mid = (lo + hi) / 2;
		if (addr >= asmap[mid]->st_value)
			lo = mid;
		else
			hi = mid;
	}

	i = addr < asmap[hi]->st_value ? lo : hi;
	sym = asmap[i];
	v = sym->st_value;

	/*
	 * If the previous entry has the same value, improve our choice.  The
	 * order of equal-valued symbols is determined by the comparison func.
	 */
	while (i-- != 0 && asmap[i]->st_value == v)
		sym = asmap[i];

	if (addr - sym->st_value < MAX(sym->st_size, 1)) {
		if (idp != NULL)
			*idp = (uint_t)(sym - symtab);
		return (dt_module_symgelf32(sym, symp));
	}

	return (NULL);
}

static GElf_Sym *
dt_module_symaddr64(dt_module_t *dmp, GElf_Addr addr,
    GElf_Sym *symp, uint_t *idp)
{
	const Elf64_Sym **asmap = (const Elf64_Sym **)dmp->dm_asmap;
	const Elf64_Sym *symtab = dmp->dm_symtab.cts_data;
	const Elf64_Sym *sym;

	uint_t i, mid, lo = 0, hi = dmp->dm_aslen - 1;
	Elf64_Addr v;

	if (dmp->dm_aslen == 0)
		return (NULL);

	while (hi - lo > 1) {
		mid = (lo + hi) / 2;
		if (addr >= asmap[mid]->st_value)
			lo = mid;
		else
			hi = mid;
	}

	i = addr < asmap[hi]->st_value ? lo : hi;
	sym = asmap[i];
	v = sym->st_value;

	/*
	 * If the previous entry has the same value, improve our choice.  The
	 * order of equal-valued symbols is determined by the comparison func.
	 */
	while (i-- != 0 && asmap[i]->st_value == v)
		sym = asmap[i];

	if (addr - sym->st_value < MAX(sym->st_size, 1)) {
		if (idp != NULL)
			*idp = (uint_t)(sym - symtab);
		return (dt_module_symgelf64(sym, symp));
	}

	return (NULL);
}

static const dt_modops_t dt_modops_32 = {
	dt_module_syminit32,
	dt_module_symsort32,
	dt_module_symname32,
	dt_module_symaddr32
};

static const dt_modops_t dt_modops_64 = {
	dt_module_syminit64,
	dt_module_symsort64,
	dt_module_symname64,
	dt_module_symaddr64
};

dt_module_t *
dt_module_create(dtrace_hdl_t *dtp, const char *name)
{
	uint_t h = dt_strtab_hash(name, NULL) % dtp->dt_modbuckets;
	dt_module_t *dmp;

	for (dmp = dtp->dt_mods[h]; dmp != NULL; dmp = dmp->dm_next) {
		if (strcmp(dmp->dm_name, name) == 0)
			return (dmp);
	}

	if ((dmp = malloc(sizeof (dt_module_t))) == NULL)
		return (NULL); /* caller must handle allocation failure */

	bzero(dmp, sizeof (dt_module_t));
	(void) strlcpy(dmp->dm_name, name, sizeof (dmp->dm_name));
	dt_list_append(&dtp->dt_modlist, dmp);
	dmp->dm_next = dtp->dt_mods[h];
	dtp->dt_mods[h] = dmp;
	dtp->dt_nmods++;

	if (dtp->dt_conf.dtc_ctfmodel == CTF_MODEL_LP64)
		dmp->dm_ops = &dt_modops_64;
	else
		dmp->dm_ops = &dt_modops_32;

	return (dmp);
}