pgcore.c

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

			    pgc->pgc_doff) != 0)
				return (1);
		}

	}
#ifdef __sparcv9
	if (P->status.pr_dmodel == PR_MODEL_LP64) {
		asrset_t asrs;
		if (Plwp_getasrs(P, lsp->pr_lwpid, asrs) == 0) {
			if (write_note(pgc->pgc_fd, NT_ASRS, &asrs,
			    sizeof (asrset_t), pgc->pgc_doff) != 0)
				return (1);
		}
	}
#endif	/* __sparcv9 */
#endif	/* sparc */

	return (0);
}

static uint_t
count_sections(pgcore_t *pgc)
{
	struct ps_prochandle *P = pgc->P;
	file_info_t *fptr;
	uint_t cnt;
	uint_t nshdrs = 0;

	if (!(pgc->pgc_content & (CC_CONTENT_CTF | CC_CONTENT_SYMTAB)))
		return (0);

	fptr = list_next(&P->file_head);
	for (cnt = P->num_files; cnt > 0; cnt--, fptr = list_next(fptr)) {
		int hit_symtab = 0;

		Pbuild_file_symtab(P, fptr);

		if ((pgc->pgc_content & CC_CONTENT_CTF) &&
		    Pbuild_file_ctf(P, fptr) != NULL) {
			sym_tbl_t *sym;

			nshdrs++;

			if (fptr->file_ctf_dyn) {
				sym = &fptr->file_dynsym;
			} else {
				sym = &fptr->file_symtab;
				hit_symtab = 1;
			}

			if (sym->sym_data != NULL && sym->sym_symn != 0 &&
			    sym->sym_strs != NULL)
				nshdrs += 2;
		}

		if ((pgc->pgc_content & CC_CONTENT_SYMTAB) && !hit_symtab &&
		    fptr->file_symtab.sym_data != NULL &&
		    fptr->file_symtab.sym_symn != 0 &&
		    fptr->file_symtab.sym_strs != NULL) {
			nshdrs += 2;
		}
	}

	return (nshdrs == 0 ? 0 : nshdrs + 2);
}

static int
write_shdr(pgcore_t *pgc, shstrtype_t name, uint_t type, ulong_t flags,
    uintptr_t addr, ulong_t offset, size_t size, uint_t link, uint_t info,
    uintptr_t addralign, uintptr_t entsize)
{
	if (pgc->P->status.pr_dmodel == PR_MODEL_ILP32) {
		Elf32_Shdr shdr;

		bzero(&shdr, sizeof (shdr));
		shdr.sh_name = shstrtab_ndx(&pgc->pgc_shstrtab, name);
		shdr.sh_type = type;
		shdr.sh_flags = flags;
		shdr.sh_addr = (Elf32_Addr)addr;
		shdr.sh_offset = offset;
		shdr.sh_size = size;
		shdr.sh_link = link;
		shdr.sh_info = info;
		shdr.sh_addralign = addralign;
		shdr.sh_entsize = entsize;

		if (pwrite64(pgc->pgc_fd, &shdr, sizeof (shdr),
		    *pgc->pgc_soff) != sizeof (shdr))
			return (-1);

		*pgc->pgc_soff += sizeof (shdr);
#ifdef _LP64
	} else {
		Elf64_Shdr shdr;

		bzero(&shdr, sizeof (shdr));
		shdr.sh_name = shstrtab_ndx(&pgc->pgc_shstrtab, name);
		shdr.sh_type = type;
		shdr.sh_flags = flags;
		shdr.sh_addr = addr;
		shdr.sh_offset = offset;
		shdr.sh_size = size;
		shdr.sh_link = link;
		shdr.sh_info = info;
		shdr.sh_addralign = addralign;
		shdr.sh_entsize = entsize;

		if (pwrite64(pgc->pgc_fd, &shdr, sizeof (shdr),
		    *pgc->pgc_soff) != sizeof (shdr))
			return (-1);

		*pgc->pgc_soff += sizeof (shdr);
#endif	/* _LP64 */
	}

	return (0);
}

static int
dump_symtab(pgcore_t *pgc, file_info_t *fptr, uint_t index, int dynsym)
{
	sym_tbl_t *sym = dynsym ? &fptr->file_dynsym : &fptr->file_symtab;
	shstrtype_t symname = dynsym ? STR_DYNSYM : STR_SYMTAB;
	shstrtype_t strname = dynsym ? STR_DYNSTR : STR_STRTAB;
	uint_t symtype = dynsym ? SHT_DYNSYM : SHT_SYMTAB;
	size_t size;
	uintptr_t addr = fptr->file_map->map_pmap.pr_vaddr;

	if (sym->sym_data == NULL || sym->sym_symn == 0 ||
	    sym->sym_strs == NULL)
		return (0);

	size = sym->sym_hdr.sh_size;
	if (pwrite64(pgc->pgc_fd, sym->sym_data->d_buf, size,
	    *pgc->pgc_doff) != size)
		return (-1);

	if (write_shdr(pgc, symname, symtype, 0, addr, *pgc->pgc_doff, size,
	    index + 1, sym->sym_hdr.sh_info, sym->sym_hdr.sh_addralign,
	    sym->sym_hdr.sh_entsize) != 0)
		return (-1);

	*pgc->pgc_doff += roundup(size, 8);

	size = sym->sym_strhdr.sh_size;
	if (pwrite64(pgc->pgc_fd, sym->sym_strs, size, *pgc->pgc_doff) != size)
		return (-1);

	if (write_shdr(pgc, strname, SHT_STRTAB, SHF_STRINGS, addr,
	    *pgc->pgc_doff, size, 0, 0, 1, 0) != 0)
		return (-1);

	*pgc->pgc_doff += roundup(size, 8);

	return (0);
}

static int
dump_sections(pgcore_t *pgc)
{
	struct ps_prochandle *P = pgc->P;
	file_info_t *fptr;
	uint_t cnt;
	uint_t index = 1;

	if (!(pgc->pgc_content & (CC_CONTENT_CTF | CC_CONTENT_SYMTAB)))
		return (0);

	fptr = list_next(&P->file_head);
	for (cnt = P->num_files; cnt > 0; cnt--, fptr = list_next(fptr)) {
		int hit_symtab = 0;

		Pbuild_file_symtab(P, fptr);

		if ((pgc->pgc_content & CC_CONTENT_CTF) &&
		    Pbuild_file_ctf(P, fptr) != NULL) {
			sym_tbl_t *sym;
			uint_t dynsym;
			uint_t symindex = 0;

			/*
			 * Write the symtab out first so we can correctly
			 * set the sh_link field in the CTF section header.
			 * symindex will be 0 if there is no corresponding
			 * symbol table section.
			 */
			if (fptr->file_ctf_dyn) {
				sym = &fptr->file_dynsym;
				dynsym = 1;
			} else {
				sym = &fptr->file_symtab;
				dynsym = 0;
				hit_symtab = 1;
			}

			if (sym->sym_data != NULL && sym->sym_symn != 0 &&
			    sym->sym_strs != NULL) {
				symindex = index;
				if (dump_symtab(pgc, fptr, index, dynsym) != 0)
				    return (-1);
				index += 2;
			}

			/*
			 * Write the CTF data that we've read out of the
			 * file itself into the core file.
			 */
			if (pwrite64(pgc->pgc_fd, fptr->file_ctf_buf,
			    fptr->file_ctf_size, *pgc->pgc_doff) !=
			    fptr->file_ctf_size)
				return (-1);

			if (write_shdr(pgc, STR_CTF, SHT_PROGBITS, 0,
			    fptr->file_map->map_pmap.pr_vaddr, *pgc->pgc_doff,
			    fptr->file_ctf_size, symindex, 0, 4, 0) != 0)
				return (-1);

			index++;
			*pgc->pgc_doff += roundup(fptr->file_ctf_size, 8);
		}

		if ((pgc->pgc_content & CC_CONTENT_SYMTAB) && !hit_symtab &&
		    fptr->file_symtab.sym_data != NULL &&
		    fptr->file_symtab.sym_symn != 0 &&
		    fptr->file_symtab.sym_strs != NULL) {
			if (dump_symtab(pgc, fptr, index, 0) != 0)
				return (-1);
			index += 2;
		}
	}

	return (0);
}

/*ARGSUSED*/
static int
dump_map(void *data, const prmap_t *pmp, const char *name)
{
	pgcore_t *pgc = data;
	struct ps_prochandle *P = pgc->P;
#ifdef _LP64
	Elf64_Phdr phdr;
#else
	Elf32_Phdr phdr;
#endif
	size_t n;

	bzero(&phdr, sizeof (phdr));
	phdr.p_type = PT_LOAD;
	phdr.p_vaddr = pmp->pr_vaddr;
	phdr.p_memsz = pmp->pr_size;
	if (pmp->pr_mflags & MA_READ)
		phdr.p_flags |= PF_R;
	if (pmp->pr_mflags & MA_WRITE)
		phdr.p_flags |= PF_W;
	if (pmp->pr_mflags & MA_EXEC)
		phdr.p_flags |= PF_X;

	if (pmp->pr_vaddr + pmp->pr_size > P->status.pr_stkbase &&
	    pmp->pr_vaddr < P->status.pr_stkbase + P->status.pr_stksize) {
		if (!(pgc->pgc_content & CC_CONTENT_STACK))
			goto exclude;

	} else if ((pmp->pr_mflags & MA_ANON) &&
	    pmp->pr_vaddr + pmp->pr_size > P->status.pr_brkbase &&
	    pmp->pr_vaddr < P->status.pr_brkbase + P->status.pr_brksize) {
		if (!(pgc->pgc_content & CC_CONTENT_HEAP))
			goto exclude;

	} else if (pmp->pr_mflags & MA_ISM) {
		if (pmp->pr_mflags & MA_NORESERVE) {
			if (!(pgc->pgc_content & CC_CONTENT_DISM))
				goto exclude;
		} else {
			if (!(pgc->pgc_content & CC_CONTENT_ISM))
				goto exclude;
		}

	} else if (pmp->pr_mflags & MA_SHM) {
		if (!(pgc->pgc_content & CC_CONTENT_SHM))
			goto exclude;

	} else if (pmp->pr_mflags & MA_SHARED) {
		if (pmp->pr_mflags & MA_ANON) {
			if (!(pgc->pgc_content & CC_CONTENT_SHANON))
				goto exclude;
		} else {
			if (!(pgc->pgc_content & CC_CONTENT_SHFILE))
				goto exclude;
		}

	} else if (pmp->pr_mflags & MA_ANON) {
		if (!(pgc->pgc_content & CC_CONTENT_ANON))
			goto exclude;

	} else if (phdr.p_flags == (PF_R | PF_X)) {
		if (!(pgc->pgc_content & CC_CONTENT_TEXT))
			goto exclude;

	} else if (phdr.p_flags == PF_R) {
		if (!(pgc->pgc_content & CC_CONTENT_RODATA))
			goto exclude;

	} else {
		if (!(pgc->pgc_content & CC_CONTENT_DATA))
			goto exclude;
	}

	n = 0;
	while (n < pmp->pr_size) {
		size_t csz = MIN(pmp->pr_size - n, pgc->pgc_chunksz);

		/*
		 * If we can't read out part of the victim's address
		 * space for some reason ignore that failure and try to
		 * emit a partial core file without that mapping's data.
		 * As in the kernel, we mark these failures with the
		 * PF_SUNW_FAILURE flag and store the errno where the
		 * mapping would have been.
		 */
		if (Pread(P, pgc->pgc_chunk, csz, pmp->pr_vaddr + n) != csz ||
		    pwrite64(pgc->pgc_fd, pgc->pgc_chunk, csz,
		    *pgc->pgc_doff + n) != csz) {
			int err = errno;
			(void) pwrite64(pgc->pgc_fd, &err, sizeof (err),
			    *pgc->pgc_doff);
			*pgc->pgc_doff += roundup(sizeof (err), 8);

			phdr.p_flags |= PF_SUNW_FAILURE;
			(void) ftruncate64(pgc->pgc_fd, *pgc->pgc_doff);
			goto exclude;
		}

		n += csz;
	}

	phdr.p_offset = *pgc->pgc_doff;
	phdr.p_filesz = pmp->pr_size;
	*pgc->pgc_doff += roundup(phdr.p_filesz, 8);

exclude:
	if (P->status.pr_dmodel == PR_MODEL_NATIVE) {
		if (pwrite64(pgc->pgc_fd, &phdr, sizeof (phdr),
		    *pgc->pgc_poff) != sizeof (phdr))
			return (1);

		*pgc->pgc_poff += sizeof (phdr);
#ifdef _LP64
	} else {
		Elf32_Phdr phdr32;

		bzero(&phdr32, sizeof (phdr32));
		phdr32.p_type = phdr.p_type;
		phdr32.p_vaddr = (Elf32_Addr)phdr.p_vaddr;
		phdr32.p_memsz = (Elf32_Word)phdr.p_memsz;
		phdr32.p_flags = phdr.p_flags;
		phdr32.p_offset = (Elf32_Off)phdr.p_offset;
		phdr32.p_filesz = (Elf32_Word)phdr.p_filesz;

		if (pwrite64(pgc->pgc_fd, &phdr32, sizeof (phdr32),
		    *pgc->pgc_poff) != sizeof (phdr32))
			return (1);

		*pgc->pgc_poff += sizeof (phdr32);
#endif	/* _LP64 */
	}

	return (0);
}

int
write_shstrtab(struct ps_prochandle *P, pgcore_t *pgc)
{
	off64_t off = *pgc->pgc_doff;
	size_t size = 0;
	shstrtab_t *s = &pgc->pgc_shstrtab;
	int i, ndx;

	if (shstrtab_size(s) == 1)
		return (0);

	/*
	 * Preemptively stick the name of the shstrtab in the string table.
	 */
	(void) shstrtab_ndx(&pgc->pgc_shstrtab, STR_SHSTRTAB);
	size = shstrtab_size(s);

	if (pwrite64(pgc->pgc_fd, "", 1, off) != 1)
		return (1);

	/*
	 * Dump all the strings that we used being sure we include the
	 * terminating null character.
	 */
	for (i = 0; i < STR_NUM; i++) {
		if ((ndx = s->sst_ndx[i]) != 0) {
			const char *str = shstrtab_data[i];
			size_t len = strlen(str) + 1;
			if (pwrite64(pgc->pgc_fd, str, len, off + ndx) != len)
				return (1);
		}
	}

	if (P->status.pr_dmodel == PR_MODEL_ILP32) {
		Elf32_Shdr shdr;

		bzero(&shdr, sizeof (shdr));
		shdr.sh_name = shstrtab_ndx(&pgc->pgc_shstrtab, STR_SHSTRTAB);
		shdr.sh_size = size;
		shdr.sh_offset = *pgc->pgc_doff;
		shdr.sh_addralign = 1;
		shdr.sh_flags = SHF_STRINGS;
		shdr.sh_type = SHT_STRTAB;

		if (pwrite64(pgc->pgc_fd, &shdr, sizeof (shdr),
		    *pgc->pgc_soff) != sizeof (shdr))
			return (1);

		*pgc->pgc_soff += sizeof (shdr);
#ifdef _LP64
	} else {
		Elf64_Shdr shdr;

		bzero(&shdr, sizeof (shdr));
		shdr.sh_name = shstrtab_ndx(&pgc->pgc_shstrtab, STR_SHSTRTAB);
		shdr.sh_size = size;
		shdr.sh_offset = *pgc->pgc_doff;
		shdr.sh_addralign = 1;
		shdr.sh_flags = SHF_STRINGS;
		shdr.sh_type = SHT_STRTAB;

		if (pwrite64(pgc->pgc_fd, &shdr, sizeof (shdr),
		    *pgc->pgc_soff) != sizeof (shdr))
			return (1);

		*pgc->pgc_soff += sizeof (shdr);
#endif	/* _LP64 */
	}

	*pgc->pgc_doff += roundup(size, 8);

	return (0);
}

/*
 * Don't explicity stop the process; that's up to the consumer.
 */
int
Pfgcore(struct ps_prochandle *P, int fd, core_content_t content)
{
	char plat[SYS_NMLN];
	char zonename[ZONENAME_MAX];
	int platlen = -1;
	pgcore_t pgc;
	off64_t poff, soff, doff, boff;
	struct utsname uts;
	uint_t nphdrs, nshdrs;

	if (ftruncate64(fd, 0) != 0)
		return (-1);

	if (content == CC_CONTENT_INVALID) {
		errno = EINVAL;
		return (-1);
	}

	/*
	 * Cache the mappings and other useful data.
	 */
	(void) Prd_agent(P);
	(void) Ppsinfo(P);

	pgc.P = P;
	pgc.pgc_fd = fd;
	pgc.pgc_poff = &poff;
	pgc.pgc_soff = &soff;
	pgc.pgc_doff = &doff;
	pgc.pgc_content = content;
	pgc.pgc_chunksz = PAGESIZE;
	if ((pgc.pgc_chunk = malloc(pgc.pgc_chunksz)) == NULL)
		return (-1);

	shstrtab_init(&pgc.pgc_shstrtab);

	/*
	 * There are two PT_NOTE program headers for ancillary data, and
	 * one for each mapping.
	 */
	nphdrs = 2 + P->map_count;
	nshdrs = count_sections(&pgc);

	(void) Pplatform(P, plat, sizeof (plat));
	platlen = strlen(plat) + 1;
	Preadauxvec(P);
	(void) Puname(P, &uts);
	if (Pzonename(P, zonename, sizeof (zonename)) == NULL)
		zonename[0] = '\0';

	/*
	 * Set up the ELF header.
	 */
	if (P->status.pr_dmodel == PR_MODEL_ILP32) {
		Elf32_Ehdr ehdr;

		bzero(&ehdr, sizeof (ehdr));
		ehdr.e_ident[EI_MAG0] = ELFMAG0;