psymtab.c

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

		sp->sh_entsize = sizeof (Elf32_Dyn);

		(void) memcpy(&elfdata[off], dp, sp->sh_size);
		off += roundup(sp->sh_size, 4);
		sp++;

		/*
		 * Section Header[5]  sh_name: .plt
		 */
		if (pltsz != 0) {
			sp->sh_name = 35;
			sp->sh_type = SHT_PROGBITS;
			sp->sh_flags = SHF_WRITE | SHF_ALLOC | SHF_EXECINSTR;
			sp->sh_addr = d[DI_PLTGOT]->d_un.d_ptr;
			if (ehdr.e_type == ET_DYN)
				sp->sh_addr -= addr;
			sp->sh_offset = off;
			sp->sh_size = pltsz;
			sp->sh_link = 0;
			sp->sh_info = 0;
			sp->sh_addralign = 4;
			sp->sh_entsize = pltentsz;

			if (Pread(P, &elfdata[off], sp->sh_size,
			    d[DI_PLTGOT]->d_un.d_ptr) != sp->sh_size) {
				free(elfdata);
				goto bad32;
			}
			off += roundup(sp->sh_size, 4);
			sp++;
		}

		free(dp);
		goto good;

bad32:
		free(dp);
		return (NULL);
#ifdef _LP64
	} else if (P->status.pr_dmodel == PR_MODEL_LP64) {
		Elf64_Ehdr ehdr, *ep;
		Elf64_Phdr phdr;
		Elf64_Shdr *sp;
		Elf64_Dyn *dp;
		Elf64_Dyn *d[DI_NENT] = { 0 };
		uint_t i, dcount = 0;
		uint64_t off;
		size_t pltsz = 0, pltentsz;

		if (read_ehdr64(P, &ehdr, addr) != 0 ||
		    read_dynamic_phdr64(P, &ehdr, &phdr, addr) != 0)
			return (NULL);

		if (ehdr.e_type == ET_DYN)
			phdr.p_vaddr += addr;

		if ((dp = malloc(phdr.p_filesz)) == NULL)
			return (NULL);

		if (Pread(P, dp, phdr.p_filesz, phdr.p_vaddr) !=
		    phdr.p_filesz) {
			free(dp);
			return (NULL);
		}

		for (i = 0; i < phdr.p_filesz / sizeof (Elf64_Dyn); i++) {
			switch (dp[i].d_tag) {
			/*
			 * For the .plt section.
			 */
			case DT_PLTGOT:
				d[DI_PLTGOT] = &dp[i];
				continue;
			case DT_JMPREL:
				d[DI_JMPREL] = &dp[i];
				continue;
			case DT_PLTRELSZ:
				d[DI_PLTRELSZ] = &dp[i];
				continue;
			case DT_PLTREL:
				d[DI_PLTREL] = &dp[i];
				continue;
			default:
				continue;

			/*
			 * For the .dynsym section.
			 */
			case DT_SYMTAB:
				d[DI_SYMTAB] = &dp[i];
				break;
			case DT_HASH:
				d[DI_HASH] = &dp[i];
				break;
			case DT_SYMENT:
				d[DI_SYMENT] = &dp[i];
				break;

			/*
			 * For the .dynstr section.
			 */
			case DT_STRTAB:
				d[DI_STRTAB] = &dp[i];
				break;
			case DT_STRSZ:
				d[DI_STRSZ] = &dp[i];
				break;
			}

			dcount++;
		}

		/*
		 * We need all of those dynamic entries in order to put
		 * together a complete set of elf sections, but we'll
		 * let the PLT section slide if need be. The dynsym- and
		 * dynstr-related dynamic entries are mandatory in both
		 * executables and shared objects so if one of those is
		 * missing, we're in some trouble and should abort.
		 */
		if (dcount + 4 != DI_NENT) {
			dprintf("text section missing required dynamic "
			    "entries\n");
			return (NULL);
		}

		if (ehdr.e_type == ET_DYN) {
			if (d[DI_PLTGOT] != NULL)
				d[DI_PLTGOT]->d_un.d_ptr += addr;
			if (d[DI_JMPREL] != NULL)
				d[DI_JMPREL]->d_un.d_ptr += addr;
			d[DI_SYMTAB]->d_un.d_ptr += addr;
			d[DI_HASH]->d_un.d_ptr += addr;
			d[DI_STRTAB]->d_un.d_ptr += addr;
		}

		/* elf header */
		size = sizeof (Elf64_Ehdr);

		/* program headers from in-core elf fragment */
		size += ehdr.e_phnum * ehdr.e_phentsize;

		/* unused shdr, and .shstrtab section */
		size += sizeof (Elf64_Shdr);
		size += sizeof (Elf64_Shdr);
		size += roundup(sizeof (shstr), 8);

		/* .dynsym section */
		size += sizeof (Elf64_Shdr);
		if (Pread(P, &nchain, sizeof (nchain),
		    d[DI_HASH]->d_un.d_ptr + 4) != sizeof (nchain))
			goto bad64;
		size += sizeof (Elf64_Sym) * nchain;

		/* .dynstr section */
		size += sizeof (Elf64_Shdr);
		size += roundup(d[DI_STRSZ]->d_un.d_val, 8);

		/* .dynamic section */
		size += sizeof (Elf64_Shdr);
		size += roundup(phdr.p_filesz, 8);

		/* .plt section */
		if (d[DI_PLTGOT] != NULL && d[DI_JMPREL] != NULL &&
		    d[DI_PLTRELSZ] != NULL && d[DI_PLTREL] != NULL) {
			uintptr_t penult, ult;
			uintptr_t jmprel = d[DI_JMPREL]->d_un.d_ptr;
			size_t pltrelsz = d[DI_PLTRELSZ]->d_un.d_val;

			if (d[DI_PLTREL]->d_un.d_val == DT_RELA) {
				uint_t ndx = pltrelsz / sizeof (Elf64_Rela) - 2;
				Elf64_Rela r[2];

				if (Pread(P, r, sizeof (r), jmprel +
				    sizeof (r[0]) * ndx) != sizeof (r))
					goto bad64;

				penult = r[0].r_offset;
				ult = r[1].r_offset;

			} else if (d[DI_PLTREL]->d_un.d_val == DT_REL) {
				uint_t ndx = pltrelsz / sizeof (Elf64_Rel) - 2;
				Elf64_Rel r[2];

				if (Pread(P, r, sizeof (r), jmprel +
				    sizeof (r[0]) * ndx) != sizeof (r))
					goto bad64;

				penult = r[0].r_offset;
				ult = r[1].r_offset;
			} else {
				goto bad64;
			}

			pltentsz = ult - penult;

			if (ehdr.e_type == ET_DYN)
				ult += addr;

			pltsz = ult - d[DI_PLTGOT]->d_un.d_ptr + pltentsz;

			size += sizeof (Elf64_Shdr);
			size += roundup(pltsz, 8);
		}

		if ((elfdata = calloc(1, size)) == NULL)
			goto bad64;

		/* LINTED - alignment */
		ep = (Elf64_Ehdr *)elfdata;
		(void) memcpy(ep, &ehdr, offsetof(Elf64_Ehdr, e_phoff));

		ep->e_ehsize = sizeof (Elf64_Ehdr);
		ep->e_phoff = sizeof (Elf64_Ehdr);
		ep->e_phentsize = ehdr.e_phentsize;
		ep->e_phnum = ehdr.e_phnum;
		ep->e_shoff = ep->e_phoff + ep->e_phnum * ep->e_phentsize;
		ep->e_shentsize = sizeof (Elf64_Shdr);
		ep->e_shnum = (pltsz == 0) ? 5 : 6;
		ep->e_shstrndx = 1;

		/* LINTED - alignment */
		sp = (Elf64_Shdr *)(elfdata + ep->e_shoff);
		off = ep->e_shoff + ep->e_shentsize * ep->e_shnum;

		/*
		 * Copying the program headers directly from the process's
		 * address space is a little suspect, but since we only
		 * use them for their address and size values, this is fine.
		 */
		if (Pread(P, &elfdata[ep->e_phoff],
		    ep->e_phnum * ep->e_phentsize, addr + ehdr.e_phoff) !=
		    ep->e_phnum * ep->e_phentsize) {
			free(elfdata);
			goto bad64;
		}

		/*
		 * The first elf section is always skipped.
		 */
		sp++;

		/*
		 * Section Header[1]  sh_name: .shstrtab
		 */
		sp->sh_name = 0;
		sp->sh_type = SHT_STRTAB;
		sp->sh_flags = SHF_STRINGS;
		sp->sh_addr = 0;
		sp->sh_offset = off;
		sp->sh_size = sizeof (shstr);
		sp->sh_link = 0;
		sp->sh_info = 0;
		sp->sh_addralign = 1;
		sp->sh_entsize = 0;

		(void) memcpy(&elfdata[off], shstr, sizeof (shstr));
		off += roundup(sp->sh_size, 8);
		sp++;

		/*
		 * Section Header[2]  sh_name: .dynsym
		 */
		sp->sh_name = 10;
		sp->sh_type = SHT_DYNSYM;
		sp->sh_flags = SHF_ALLOC;
		sp->sh_addr = d[DI_SYMTAB]->d_un.d_ptr;
		if (ehdr.e_type == ET_DYN)
			sp->sh_addr -= addr;
		sp->sh_offset = off;
		sp->sh_size = nchain * sizeof (Elf64_Sym);
		sp->sh_link = 3;
		sp->sh_info = 1;
		sp->sh_addralign = 8;
		sp->sh_entsize = sizeof (Elf64_Sym);

		if (Pread(P, &elfdata[off], sp->sh_size,
		    d[DI_SYMTAB]->d_un.d_ptr) != sp->sh_size) {
			free(elfdata);
			goto bad64;
		}

		off += roundup(sp->sh_size, 8);
		sp++;

		/*
		 * Section Header[3]  sh_name: .dynstr
		 */
		sp->sh_name = 18;
		sp->sh_type = SHT_STRTAB;
		sp->sh_flags = SHF_ALLOC | SHF_STRINGS;
		sp->sh_addr = d[DI_STRTAB]->d_un.d_ptr;
		if (ehdr.e_type == ET_DYN)
			sp->sh_addr -= addr;
		sp->sh_offset = off;
		sp->sh_size = d[DI_STRSZ]->d_un.d_val;
		sp->sh_link = 0;
		sp->sh_info = 0;
		sp->sh_addralign = 1;
		sp->sh_entsize = 0;

		if (Pread(P, &elfdata[off], sp->sh_size,
		    d[DI_STRTAB]->d_un.d_ptr) != sp->sh_size) {
			free(elfdata);
			goto bad64;
		}
		off += roundup(sp->sh_size, 8);
		sp++;

		/*
		 * Section Header[4]  sh_name: .dynamic
		 */
		sp->sh_name = 26;
		sp->sh_type = SHT_DYNAMIC;
		sp->sh_flags = SHF_WRITE | SHF_ALLOC;
		sp->sh_addr = phdr.p_vaddr;
		if (ehdr.e_type == ET_DYN)
			sp->sh_addr -= addr;
		sp->sh_offset = off;
		sp->sh_size = phdr.p_filesz;
		sp->sh_link = 3;
		sp->sh_info = 0;
		sp->sh_addralign = 8;
		sp->sh_entsize = sizeof (Elf64_Dyn);

		(void) memcpy(&elfdata[off], dp, sp->sh_size);
		off += roundup(sp->sh_size, 8);
		sp++;

		/*
		 * Section Header[5]  sh_name: .plt
		 */
		if (pltsz != 0) {
			sp->sh_name = 35;
			sp->sh_type = SHT_PROGBITS;
			sp->sh_flags = SHF_WRITE | SHF_ALLOC | SHF_EXECINSTR;
			sp->sh_addr = d[DI_PLTGOT]->d_un.d_ptr;
			if (ehdr.e_type == ET_DYN)
				sp->sh_addr -= addr;
			sp->sh_offset = off;
			sp->sh_size = pltsz;
			sp->sh_link = 0;
			sp->sh_info = 0;
			sp->sh_addralign = 8;
			sp->sh_entsize = pltentsz;

			if (Pread(P, &elfdata[off], sp->sh_size,
			    d[DI_PLTGOT]->d_un.d_ptr) != sp->sh_size) {
				free(elfdata);
				goto bad64;
			}
			off += roundup(sp->sh_size, 8);
			sp++;
		}

		free(dp);
		goto good;

bad64:
		free(dp);
		return (NULL);
#endif	/* _LP64 */
	}
good:
	if ((elf = elf_memory(elfdata, size)) == NULL) {
		free(elfdata);
		return (NULL);
	}

	fptr->file_elfmem = elfdata;

	return (elf);
}

/*
 * We wouldn't need these if qsort(3C) took an argument for the callback...
 */
static mutex_t sort_mtx = DEFAULTMUTEX;
static char *sort_strs;
static GElf_Sym *sort_syms;

int
byaddr_cmp_common(GElf_Sym *a, char *aname, GElf_Sym *b, char *bname)
{
	if (a->st_value < b->st_value)
		return (-1);
	if (a->st_value > b->st_value)
		return (1);

	/*
	 * Prefer the function to the non-function.
	 */
	if (GELF_ST_TYPE(a->st_info) != GELF_ST_TYPE(b->st_info)) {
		if (GELF_ST_TYPE(a->st_info) == STT_FUNC)
			return (-1);
		if (GELF_ST_TYPE(b->st_info) == STT_FUNC)
			return (1);
	}

	/*
	 * Prefer the weak or strong global symbol to the local symbol.
	 */
	if (GELF_ST_BIND(a->st_info) != GELF_ST_BIND(b->st_info)) {
		if (GELF_ST_BIND(b->st_info) == STB_LOCAL)
			return (-1);
		if (GELF_ST_BIND(a->st_info) == STB_LOCAL)
			return (1);
	}

	/*
	 * Prefer the name with fewer leading underscores in the name.
	 */
	while (*aname == '_' && *bname == '_') {
		aname++;
		bname++;
	}

	if (*bname == '_')
		return (-1);
	if (*aname == '_')
		return (1);

	/*
	 * Prefer the symbol with the smaller size.
	 */
	if (a->st_size < b->st_size)
		return (-1);
	if (a->st_size > b->st_size)
		return (1);

	/*
	 * All other factors being equal, fall back to lexicographic order.
	 */
	return (strcmp(aname, bname));
}

static int
byaddr_cmp(const void *aa, const void *bb)
{
	GElf_Sym *a = &sort_syms[*(uint_t *)aa];
	GElf_Sym *b = &sort_syms[*(uint_t *)bb];
	char *aname = sort_strs + a->st_name;
	char *bname = sort_strs + b->st_name;

	return (byaddr_cmp_common(a, aname, b, bname));
}

static int
byname_cmp(const void *aa, const void *bb)
{
	GElf_Sym *a = &sort_syms[*(uint_t *)aa];
	GElf_Sym *b = &sort_syms[*(uint_t *)bb];
	char *aname = sort_strs + a->st_name;
	char *bname = sort_strs + b->st_name;

	return (strcmp(aname, bname));
}

void
optimize_symtab(sym_tbl_t *symtab)
{
	GElf_Sym *symp, *syms;
	uint_t i, *indexa, *indexb;
	Elf_Data *data;
	size_t symn, strsz, count;

	if (symtab == NULL || symtab->sym_data == NULL ||
	    symtab->sym_byaddr != NULL)
		return;

	data = symtab->sym_data;
	symn = symtab->sym_symn;
	strsz = symtab->sym_strsz;

	symp = syms = malloc(sizeof (GElf_Sym) * symn);

	/*
	 * First record all the symbols into a table and count up the ones
	 * that we're interested in. We mark symbols as invalid by setting
	 * the st_name to an illegal value.
	 */
	for (i = 0, count = 0; i < symn; i++, symp++) {
		if (gelf_getsym(data, i, symp) != NULL &&
		    symp->st_name < strsz &&
		    IS_DATA_TYPE(GELF_ST_TYPE(symp->st_info)))
			count++;
		else
			symp->st_name = strsz;
	}

	/*
	 * Allocate sufficient space for both tables and populate them
	 * with the same symbols we just counted.
	 */
	symtab->sym_count = count;
	indexa = symtab->sym_byaddr = calloc(sizeof (uint_t), count);
	indexb = symtab->sym_byname = calloc(sizeof (uint_t), count);

	for (i = 0, symp = syms; i < symn; i++, symp++) {
		if (symp->st_name < strsz)
			*indexa++ = *indexb++ = i;
	}

	/*
	 * Sort the two tables according to the appropriate criteria.
	 */
	(void) mutex_lock(&sort_mtx);
	sort_strs = symtab->sym_strs;
	sort_syms = syms;

	qsort(symtab->sym_byaddr, count, sizeof (uint_t), byaddr_cmp);
	qsort(symtab->sym_byname, count, sizeof (uint_t), byname_cmp);

	sort_strs = NULL;
	sort_syms = NULL;