4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #pragma ident "%Z%%M% %I% %E% SMI"
29 * Routines for preparing tdata trees for conversion into CTF data, and
30 * for placing the resulting data into an output file.
36 #include <sys/types.h>
49 typedef struct iidesc_match {
58 burst_iitypes(void *data, void *arg)
61 iiburst_t *iiburst = arg;
63 switch (ii->ii_type) {
68 if (!(ii->ii_flags & IIDESC_F_USED))
75 ii->ii_dtype->t_flags |= TDESC_F_ISROOT;
76 (void) iitraverse_td(ii, iiburst->iib_tdtd);
82 save_type_by_id(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private)
84 iiburst_t *iiburst = private;
87 * Doing this on every node is horribly inefficient, but given that
88 * we may be suppressing some types, we can't trust nextid in the
91 if (tdp->t_id > iiburst->iib_maxtypeid)
92 iiburst->iib_maxtypeid = tdp->t_id;
94 slist_add(&iiburst->iib_types, tdp, tdesc_idcmp);
99 static tdtrav_cb_f burst_types_cbs[] = {
101 save_type_by_id, /* intrinsic */
102 save_type_by_id, /* pointer */
103 save_type_by_id, /* array */
104 save_type_by_id, /* function */
105 save_type_by_id, /* struct */
106 save_type_by_id, /* union */
107 save_type_by_id, /* enum */
108 save_type_by_id, /* forward */
109 save_type_by_id, /* typedef */
110 tdtrav_assert, /* typedef_unres */
111 save_type_by_id, /* volatile */
112 save_type_by_id, /* const */
113 save_type_by_id /* restrict */
118 iiburst_new(tdata_t *td, int max)
120 iiburst_t *iiburst = xcalloc(sizeof (iiburst_t));
121 iiburst->iib_td = td;
122 iiburst->iib_funcs = xcalloc(sizeof (iidesc_t *) * max);
123 iiburst->iib_nfuncs = 0;
124 iiburst->iib_objts = xcalloc(sizeof (iidesc_t *) * max);
125 iiburst->iib_nobjts = 0;
130 iiburst_types(iiburst_t *iiburst)
134 tdtrav_init(&tdtd, &iiburst->iib_td->td_curvgen, NULL, burst_types_cbs,
135 NULL, (void *)iiburst);
137 iiburst->iib_tdtd = &tdtd;
139 (void) hash_iter(iiburst->iib_td->td_iihash, burst_iitypes, iiburst);
143 iiburst_free(iiburst_t *iiburst)
145 free(iiburst->iib_funcs);
146 free(iiburst->iib_objts);
147 list_free(iiburst->iib_types, NULL, NULL);
152 * See if this iidesc matches the ELF symbol data we pass in.
154 * A fuzzy match is where we have a local symbol matching the name of a
155 * global type description. This is common when a mapfile is used for a
156 * DSO, but we don't accept it by default.
158 * A weak fuzzy match is when a weak symbol was resolved and matched to
159 * a global type description.
162 matching_iidesc(void *arg1, void *arg2)
164 iidesc_t *iidesc = arg1;
165 iidesc_match_t *match = arg2;
166 if (streq(iidesc->ii_name, match->iim_name) == 0)
169 switch (iidesc->ii_type) {
172 if (match->iim_bind == STB_GLOBAL) {
173 match->iim_ret = iidesc;
175 } else if (match->iim_fuzzy && match->iim_ret == NULL) {
176 match->iim_ret = iidesc;
177 /* continue to look for strong match */
183 if (match->iim_bind == STB_LOCAL &&
184 match->iim_file != NULL &&
185 streq(iidesc->ii_owner, match->iim_file)) {
186 match->iim_ret = iidesc;
197 find_iidesc(tdata_t *td, iidesc_match_t *match)
199 match->iim_ret = NULL;
200 iter_iidescs_by_name(td, match->iim_name,
201 matching_iidesc, match);
202 return (match->iim_ret);
206 * If we have a weak symbol, attempt to find the strong symbol it will
207 * resolve to. Note: the code where this actually happens is in
208 * sym_process() in cmd/sgs/libld/common/syms.c
210 * Finding the matching symbol is unfortunately not trivial. For a
211 * symbol to be a candidate, it must:
213 * - have the same type (function, object)
214 * - have the same value (address)
215 * - have the same size
216 * - not be another weak symbol
217 * - belong to the same section (checked via section index)
219 * If such a candidate is global, then we assume we've found it. The
220 * linker generates the symbol table such that the curfile might be
221 * incorrect; this is OK for global symbols, since find_iidesc() doesn't
222 * need to check for the source file for the symbol.
224 * We might have found a strong local symbol, where the curfile is
225 * accurate and matches that of the weak symbol. We assume this is a
228 * If we've got a local symbol with a non-matching curfile, there are
229 * two possibilities. Either this is a completely different symbol, or
230 * it's a once-global symbol that was scoped to local via a mapfile. In
231 * the latter case, curfile is likely inaccurate since the linker does
232 * not preserve the needed curfile in the order of the symbol table (see
233 * the comments about locally scoped symbols in libld's update_osym()).
234 * As we can't tell this case from the former one, we use this symbol
235 * iff no other matching symbol is found.
237 * What we really need here is a SUNW section containing weak<->strong
238 * mappings that we can consume.
241 check_for_weak(GElf_Sym *weak, char const *weakfile,
242 Elf_Data *data, int nent, Elf_Data *strdata,
243 GElf_Sym *retsym, char **curfilep)
245 char *curfile = NULL;
246 char *tmpfile1 = NULL;
253 if (GELF_ST_BIND(weak->st_info) != STB_WEAK)
256 for (i = 0; i < nent; i++) {
260 if (gelf_getsym(data, i, &sym) == NULL)
263 type = GELF_ST_TYPE(sym.st_info);
265 if (type == STT_FILE)
266 curfile = (char *)strdata->d_buf + sym.st_name;
268 if (GELF_ST_TYPE(weak->st_info) != type ||
269 weak->st_value != sym.st_value)
272 if (weak->st_size != sym.st_size)
275 if (GELF_ST_BIND(sym.st_info) == STB_WEAK)
278 if (sym.st_shndx != weak->st_shndx)
281 if (GELF_ST_BIND(sym.st_info) == STB_LOCAL &&
282 (curfile == NULL || weakfile == NULL ||
283 strcmp(curfile, weakfile) != 0)) {
296 *curfilep = tmpfile1;
305 * When we've found the underlying symbol's type description
306 * for a weak symbol, we need to copy it and rename it to match
307 * the weak symbol. We also need to add it to the td so it's
308 * handled along with the others later.
311 copy_from_strong(tdata_t *td, GElf_Sym *sym, iidesc_t *strongdesc,
312 const char *weakname, const char *weakfile)
314 iidesc_t *new = iidesc_dup_rename(strongdesc, weakname, weakfile);
315 uchar_t type = GELF_ST_TYPE(sym->st_info);
319 new->ii_type = II_GVAR;
322 new->ii_type = II_GFUN;
326 hash_add(td->td_iihash, new);
332 * Process the symbol table of the output file, associating each symbol
333 * with a type description if possible, and sorting them into functions
334 * and data, maintaining symbol table order.
337 sort_iidescs(Elf *elf, const char *file, tdata_t *td, int fuzzymatch,
343 Elf_Data *data, *strdata;
346 iidesc_match_t match;
348 match.iim_fuzzy = fuzzymatch;
349 match.iim_file = NULL;
351 if ((stidx = findelfsecidx(elf, file,
352 dynsym ? ".dynsym" : ".symtab")) < 0)
353 terminate("%s: Can't open symbol table\n", file);
354 scn = elf_getscn(elf, stidx);
355 data = elf_getdata(scn, NULL);
356 gelf_getshdr(scn, &shdr);
357 nent = shdr.sh_size / shdr.sh_entsize;
359 scn = elf_getscn(elf, shdr.sh_link);
360 strdata = elf_getdata(scn, NULL);
362 iiburst = iiburst_new(td, nent);
364 for (i = 0; i < nent; i++) {
368 iidesc_match_t smatch;
372 if (gelf_getsym(data, i, &sym) == NULL)
373 elfterminate(file, "Couldn't read symbol %d", i);
375 match.iim_name = (char *)strdata->d_buf + sym.st_name;
376 match.iim_bind = GELF_ST_BIND(sym.st_info);
378 switch (GELF_ST_TYPE(sym.st_info)) {
380 match.iim_file = match.iim_name;
383 tolist = iiburst->iib_objts;
384 curr = &iiburst->iib_nobjts;
387 tolist = iiburst->iib_funcs;
388 curr = &iiburst->iib_nfuncs;
394 if (ignore_symbol(&sym, match.iim_name))
397 iidesc = find_iidesc(td, &match);
399 if (iidesc != NULL) {
400 tolist[*curr] = iidesc;
401 iidesc->ii_flags |= IIDESC_F_USED;
406 if (!check_for_weak(&sym, match.iim_file, data, nent, strdata,
407 &ssym, &smatch.iim_file)) {
412 smatch.iim_fuzzy = fuzzymatch;
413 smatch.iim_name = (char *)strdata->d_buf + ssym.st_name;
414 smatch.iim_bind = GELF_ST_BIND(ssym.st_info);
416 debug(3, "Weak symbol %s resolved to %s\n", match.iim_name,
419 iidesc = find_iidesc(td, &smatch);
421 if (iidesc != NULL) {
422 tolist[*curr] = copy_from_strong(td, &sym,
423 iidesc, match.iim_name, match.iim_file);
424 tolist[*curr]->ii_flags |= IIDESC_F_USED;
431 * Stabs are generated for every function declared in a given C source
432 * file. When converting an object file, we may encounter a stab that
433 * has no symbol table entry because the optimizer has decided to omit
434 * that item (for example, an unreferenced static function). We may
435 * see iidescs that do not have an associated symtab entry, and so
436 * we do not write records for those functions into the CTF data.
437 * All others get marked as a root by this function.
439 iiburst_types(iiburst);
442 * By not adding some of the functions and/or objects, we may have
443 * caused some types that were referenced solely by those
444 * functions/objects to be suppressed. This could cause a label,
445 * generated prior to the evisceration, to be incorrect. Find the
446 * highest type index, and change the label indicies to be no higher
449 tdata_label_newmax(td, iiburst->iib_maxtypeid);
455 write_file(Elf *src, const char *srcname, Elf *dst, const char *dstname,
456 caddr_t ctfdata, size_t ctfsize, int flags)
458 GElf_Ehdr sehdr, dehdr;
459 Elf_Scn *sscn, *dscn;
460 Elf_Data *sdata, *ddata;
462 GElf_Word symtab_type;
464 off_t new_offset = 0;
465 off_t ctfnameoff = 0;
466 int dynsym = (flags & CTF_USE_DYNSYM);
467 int keep_stabs = (flags & CTF_KEEP_STABS);
475 if (gelf_newehdr(dst, gelf_getclass(src)) == NULL)
476 elfterminate(dstname, "Cannot copy ehdr to temp file");
477 gelf_getehdr(src, &sehdr);
478 memcpy(&dehdr, &sehdr, sizeof (GElf_Ehdr));
479 gelf_update_ehdr(dst, &dehdr);
481 symtab_type = dynsym ? SHT_DYNSYM : SHT_SYMTAB;
484 * Neither the existing stab sections nor the SUNW_ctf sections (new or
485 * existing) are SHF_ALLOC'd, so they won't be in areas referenced by
486 * program headers. As such, we can just blindly copy the program
487 * headers from the existing file to the new file.
489 if (sehdr.e_phnum != 0) {
490 (void) elf_flagelf(dst, ELF_C_SET, ELF_F_LAYOUT);
491 if (gelf_newphdr(dst, sehdr.e_phnum) == NULL)
492 elfterminate(dstname, "Cannot make phdrs in temp file");
494 for (i = 0; i < sehdr.e_phnum; i++) {
497 gelf_getphdr(src, i, &phdr);
498 gelf_update_phdr(dst, i, &phdr);
502 secxlate = xmalloc(sizeof (int) * sehdr.e_shnum);
503 for (srcidx = dstidx = 0; srcidx < sehdr.e_shnum; srcidx++) {
504 Elf_Scn *scn = elf_getscn(src, srcidx);
508 gelf_getshdr(scn, &shdr1);
509 sname = elf_strptr(src, sehdr.e_shstrndx, shdr1.sh_name);
511 elfterminate(srcname, "Can't find string at %u",
515 if (strcmp(sname, CTF_ELF_SCN_NAME) == 0) {
516 secxlate[srcidx] = -1;
517 } else if (!keep_stabs &&
518 (strncmp(sname, ".stab", 5) == 0 ||
519 strncmp(sname, ".debug", 6) == 0 ||
520 strncmp(sname, ".rel.debug", 10) == 0 ||
521 strncmp(sname, ".rela.debug", 11) == 0)) {
522 secxlate[srcidx] = -1;
523 } else if (dynsym && shdr1.sh_type == SHT_SYMTAB) {
525 * If we're building CTF against the dynsym,
526 * we'll rip out the symtab so debuggers aren't
529 secxlate[srcidx] = -1;
531 secxlate[srcidx] = dstidx++;
532 curnmoff += strlen(sname) + 1;
535 new_offset = (off_t)dehdr.e_phoff;
538 for (srcidx = 1; srcidx < sehdr.e_shnum; srcidx++) {
541 sscn = elf_getscn(src, srcidx);
542 gelf_getshdr(sscn, &shdr);
544 if (secxlate[srcidx] == -1) {
549 dscn = elf_newscn(dst);
552 * If this file has program headers, we need to explicitly lay
553 * out sections. If none of the sections prior to this one have
554 * been removed, then we can just use the existing location. If
555 * one or more sections have been changed, then we need to
556 * adjust this one to avoid holes.
558 if (changing && sehdr.e_phnum != 0) {
559 pad = new_offset % shdr.sh_addralign;
562 new_offset += shdr.sh_addralign - pad;
563 shdr.sh_offset = new_offset;
566 shdr.sh_link = secxlate[shdr.sh_link];
568 if (shdr.sh_type == SHT_REL || shdr.sh_type == SHT_RELA)
569 shdr.sh_info = secxlate[shdr.sh_info];
571 sname = elf_strptr(src, sehdr.e_shstrndx, shdr.sh_name);
573 elfterminate(srcname, "Can't find string at %u",
578 if (gelf_update_shdr(dscn, &shdr) == 0)
579 elfterminate(dstname, "Cannot update sect %s", sname);
582 if ((sdata = elf_getdata(sscn, NULL)) == NULL)
583 elfterminate(srcname, "Cannot get sect %s data", sname);
584 if ((ddata = elf_newdata(dscn)) == NULL)
585 elfterminate(dstname, "Can't make sect %s data", sname);
587 bcopy(sdata, ddata, sizeof (Elf_Data));
590 * FreeBSD's Elf_Data has private fields which the
591 * elf_* routines manage. Simply copying the
592 * entire structure corrupts the data. So we need
593 * to copy the public fields explictly.
595 ddata->d_align = sdata->d_align;
596 ddata->d_off = sdata->d_off;
597 ddata->d_size = sdata->d_size;
598 ddata->d_type = sdata->d_type;
599 ddata->d_version = sdata->d_version;
602 if (srcidx == sehdr.e_shstrndx) {
603 char seclen = strlen(CTF_ELF_SCN_NAME);
605 ddata->d_buf = xmalloc(ddata->d_size + shdr.sh_size +
607 bcopy(sdata->d_buf, ddata->d_buf, shdr.sh_size);
608 strcpy((caddr_t)ddata->d_buf + shdr.sh_size,
610 ctfnameoff = (off_t)shdr.sh_size;
611 shdr.sh_size += seclen + 1;
612 ddata->d_size += seclen + 1;
614 if (sehdr.e_phnum != 0)
618 if (shdr.sh_type == symtab_type && shdr.sh_entsize != 0) {
619 int nsym = shdr.sh_size / shdr.sh_entsize;
621 symtab_idx = secxlate[srcidx];
623 ddata->d_buf = xmalloc(shdr.sh_size);
624 bcopy(sdata->d_buf, ddata->d_buf, shdr.sh_size);
626 for (i = 0; i < nsym; i++) {
630 if (gelf_getsym(ddata, i, &sym) == NULL)
631 printf("Could not get symbol %d\n",i);
633 if (sym.st_shndx >= SHN_LORESERVE)
636 if ((newscn = secxlate[sym.st_shndx]) !=
639 (newscn == -1 ? 1 : newscn);
641 gelf_update_sym(ddata, i, &sym);
647 if (ddata->d_buf == NULL) {
648 ddata->d_buf = xmalloc(shdr.sh_size);
649 bcopy(sdata->d_buf, ddata->d_buf, shdr.sh_size);
653 if (gelf_update_shdr(dscn, &shdr) == 0)
654 elfterminate(dstname, "Cannot update sect %s", sname);
656 new_offset = (off_t)shdr.sh_offset;
657 if (shdr.sh_type != SHT_NOBITS)
658 new_offset += shdr.sh_size;
661 if (symtab_idx == -1) {
662 terminate("%s: Cannot find %s section\n", srcname,
663 dynsym ? "SHT_DYNSYM" : "SHT_SYMTAB");
666 /* Add the ctf section */
667 dscn = elf_newscn(dst);
668 gelf_getshdr(dscn, &shdr);
669 shdr.sh_name = ctfnameoff;
670 shdr.sh_type = SHT_PROGBITS;
671 shdr.sh_size = ctfsize;
672 shdr.sh_link = symtab_idx;
673 shdr.sh_addralign = 4;
674 if (changing && sehdr.e_phnum != 0) {
675 pad = new_offset % shdr.sh_addralign;
678 new_offset += shdr.sh_addralign - pad;
680 shdr.sh_offset = new_offset;
681 new_offset += shdr.sh_size;
684 ddata = elf_newdata(dscn);
685 ddata->d_buf = ctfdata;
686 ddata->d_size = ctfsize;
687 ddata->d_align = shdr.sh_addralign;
690 gelf_update_shdr(dscn, &shdr);
692 /* update the section header location */
693 if (sehdr.e_phnum != 0) {
694 size_t align = gelf_fsize(dst, ELF_T_ADDR, 1, EV_CURRENT);
695 size_t r = new_offset % align;
698 new_offset += align - r;
700 dehdr.e_shoff = new_offset;
704 dehdr.e_shstrndx = secxlate[sehdr.e_shstrndx];
705 gelf_update_ehdr(dst, &dehdr);
706 if (elf_update(dst, ELF_C_WRITE) < 0)
707 elfterminate(dstname, "Cannot finalize temp file");
713 make_ctf_data(tdata_t *td, Elf *elf, const char *file, size_t *lenp, int flags)
718 iiburst = sort_iidescs(elf, file, td, flags & CTF_FUZZY_MATCH,
719 flags & CTF_USE_DYNSYM);
720 data = ctf_gen(iiburst, lenp, flags & CTF_COMPRESS);
722 iiburst_free(iiburst);
728 write_ctf(tdata_t *td, const char *curname, const char *newname, int flags)
738 (void) elf_version(EV_CURRENT);
739 if ((fd = open(curname, O_RDONLY)) < 0 || fstat(fd, &st) < 0)
740 terminate("%s: Cannot open for re-reading", curname);
741 if ((elf = elf_begin(fd, ELF_C_READ, NULL)) == NULL)
742 elfterminate(curname, "Cannot re-read");
744 if ((tfd = open(newname, O_RDWR | O_CREAT | O_TRUNC, st.st_mode)) < 0)
745 terminate("Cannot open temp file %s for writing", newname);
746 if ((telf = elf_begin(tfd, ELF_C_WRITE, NULL)) == NULL)
747 elfterminate(curname, "Cannot write");
749 data = make_ctf_data(td, elf, curname, &len, flags);
750 write_file(elf, curname, telf, newname, data, len, flags);