11673e404SJohn Birrell /* 21673e404SJohn Birrell * CDDL HEADER START 31673e404SJohn Birrell * 41673e404SJohn Birrell * The contents of this file are subject to the terms of the 51673e404SJohn Birrell * Common Development and Distribution License (the "License"). 61673e404SJohn Birrell * You may not use this file except in compliance with the License. 71673e404SJohn Birrell * 81673e404SJohn Birrell * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 91673e404SJohn Birrell * or http://www.opensolaris.org/os/licensing. 101673e404SJohn Birrell * See the License for the specific language governing permissions 111673e404SJohn Birrell * and limitations under the License. 121673e404SJohn Birrell * 131673e404SJohn Birrell * When distributing Covered Code, include this CDDL HEADER in each 141673e404SJohn Birrell * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 151673e404SJohn Birrell * If applicable, add the following below this CDDL HEADER, with the 161673e404SJohn Birrell * fields enclosed by brackets "[]" replaced with your own identifying 171673e404SJohn Birrell * information: Portions Copyright [yyyy] [name of copyright owner] 181673e404SJohn Birrell * 191673e404SJohn Birrell * CDDL HEADER END 201673e404SJohn Birrell */ 211673e404SJohn Birrell /* 221673e404SJohn Birrell * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 231673e404SJohn Birrell * Use is subject to license terms. 241673e404SJohn Birrell */ 251673e404SJohn Birrell 261673e404SJohn Birrell #pragma ident "%Z%%M% %I% %E% SMI" 271673e404SJohn Birrell 281673e404SJohn Birrell /* 291673e404SJohn Birrell * Routines for preparing tdata trees for conversion into CTF data, and 301673e404SJohn Birrell * for placing the resulting data into an output file. 311673e404SJohn Birrell */ 321673e404SJohn Birrell 331673e404SJohn Birrell #include <stdio.h> 341673e404SJohn Birrell #include <stdlib.h> 351673e404SJohn Birrell #include <strings.h> 361673e404SJohn Birrell #include <sys/types.h> 371673e404SJohn Birrell #include <sys/stat.h> 381673e404SJohn Birrell #include <fcntl.h> 391673e404SJohn Birrell #include <libelf.h> 401673e404SJohn Birrell #include <gelf.h> 411673e404SJohn Birrell #include <unistd.h> 421673e404SJohn Birrell 431673e404SJohn Birrell #include "ctftools.h" 441673e404SJohn Birrell #include "list.h" 451673e404SJohn Birrell #include "memory.h" 461673e404SJohn Birrell #include "traverse.h" 471673e404SJohn Birrell #include "symbol.h" 481673e404SJohn Birrell 491673e404SJohn Birrell typedef struct iidesc_match { 501673e404SJohn Birrell int iim_fuzzy; 511673e404SJohn Birrell iidesc_t *iim_ret; 521673e404SJohn Birrell char *iim_name; 531673e404SJohn Birrell char *iim_file; 541673e404SJohn Birrell uchar_t iim_bind; 551673e404SJohn Birrell } iidesc_match_t; 561673e404SJohn Birrell 571673e404SJohn Birrell static int 581673e404SJohn Birrell burst_iitypes(void *data, void *arg) 591673e404SJohn Birrell { 601673e404SJohn Birrell iidesc_t *ii = data; 611673e404SJohn Birrell iiburst_t *iiburst = arg; 621673e404SJohn Birrell 631673e404SJohn Birrell switch (ii->ii_type) { 641673e404SJohn Birrell case II_GFUN: 651673e404SJohn Birrell case II_SFUN: 661673e404SJohn Birrell case II_GVAR: 671673e404SJohn Birrell case II_SVAR: 681673e404SJohn Birrell if (!(ii->ii_flags & IIDESC_F_USED)) 691673e404SJohn Birrell return (0); 701673e404SJohn Birrell break; 711673e404SJohn Birrell default: 721673e404SJohn Birrell break; 731673e404SJohn Birrell } 741673e404SJohn Birrell 751673e404SJohn Birrell ii->ii_dtype->t_flags |= TDESC_F_ISROOT; 761673e404SJohn Birrell (void) iitraverse_td(ii, iiburst->iib_tdtd); 771673e404SJohn Birrell return (1); 781673e404SJohn Birrell } 791673e404SJohn Birrell 801673e404SJohn Birrell /*ARGSUSED1*/ 811673e404SJohn Birrell static int 824cc75139SJohn Birrell save_type_by_id(tdesc_t *tdp, tdesc_t **tdpp __unused, void *private) 831673e404SJohn Birrell { 841673e404SJohn Birrell iiburst_t *iiburst = private; 851673e404SJohn Birrell 861673e404SJohn Birrell /* 871673e404SJohn Birrell * Doing this on every node is horribly inefficient, but given that 881673e404SJohn Birrell * we may be suppressing some types, we can't trust nextid in the 891673e404SJohn Birrell * tdata_t. 901673e404SJohn Birrell */ 911673e404SJohn Birrell if (tdp->t_id > iiburst->iib_maxtypeid) 921673e404SJohn Birrell iiburst->iib_maxtypeid = tdp->t_id; 931673e404SJohn Birrell 941673e404SJohn Birrell slist_add(&iiburst->iib_types, tdp, tdesc_idcmp); 951673e404SJohn Birrell 961673e404SJohn Birrell return (1); 971673e404SJohn Birrell } 981673e404SJohn Birrell 991673e404SJohn Birrell static tdtrav_cb_f burst_types_cbs[] = { 1001673e404SJohn Birrell NULL, 1011673e404SJohn Birrell save_type_by_id, /* intrinsic */ 1021673e404SJohn Birrell save_type_by_id, /* pointer */ 1031673e404SJohn Birrell save_type_by_id, /* array */ 1041673e404SJohn Birrell save_type_by_id, /* function */ 1051673e404SJohn Birrell save_type_by_id, /* struct */ 1061673e404SJohn Birrell save_type_by_id, /* union */ 1071673e404SJohn Birrell save_type_by_id, /* enum */ 1081673e404SJohn Birrell save_type_by_id, /* forward */ 1091673e404SJohn Birrell save_type_by_id, /* typedef */ 1101673e404SJohn Birrell tdtrav_assert, /* typedef_unres */ 1111673e404SJohn Birrell save_type_by_id, /* volatile */ 1121673e404SJohn Birrell save_type_by_id, /* const */ 1131673e404SJohn Birrell save_type_by_id /* restrict */ 1141673e404SJohn Birrell }; 1151673e404SJohn Birrell 1161673e404SJohn Birrell 1171673e404SJohn Birrell static iiburst_t * 1181673e404SJohn Birrell iiburst_new(tdata_t *td, int max) 1191673e404SJohn Birrell { 1201673e404SJohn Birrell iiburst_t *iiburst = xcalloc(sizeof (iiburst_t)); 1211673e404SJohn Birrell iiburst->iib_td = td; 1221673e404SJohn Birrell iiburst->iib_funcs = xcalloc(sizeof (iidesc_t *) * max); 1231673e404SJohn Birrell iiburst->iib_nfuncs = 0; 1241673e404SJohn Birrell iiburst->iib_objts = xcalloc(sizeof (iidesc_t *) * max); 1251673e404SJohn Birrell iiburst->iib_nobjts = 0; 1261673e404SJohn Birrell return (iiburst); 1271673e404SJohn Birrell } 1281673e404SJohn Birrell 1291673e404SJohn Birrell static void 1301673e404SJohn Birrell iiburst_types(iiburst_t *iiburst) 1311673e404SJohn Birrell { 1321673e404SJohn Birrell tdtrav_data_t tdtd; 1331673e404SJohn Birrell 1341673e404SJohn Birrell tdtrav_init(&tdtd, &iiburst->iib_td->td_curvgen, NULL, burst_types_cbs, 1351673e404SJohn Birrell NULL, (void *)iiburst); 1361673e404SJohn Birrell 1371673e404SJohn Birrell iiburst->iib_tdtd = &tdtd; 1381673e404SJohn Birrell 1391673e404SJohn Birrell (void) hash_iter(iiburst->iib_td->td_iihash, burst_iitypes, iiburst); 1401673e404SJohn Birrell } 1411673e404SJohn Birrell 1421673e404SJohn Birrell static void 1431673e404SJohn Birrell iiburst_free(iiburst_t *iiburst) 1441673e404SJohn Birrell { 1451673e404SJohn Birrell free(iiburst->iib_funcs); 1461673e404SJohn Birrell free(iiburst->iib_objts); 1471673e404SJohn Birrell list_free(iiburst->iib_types, NULL, NULL); 1481673e404SJohn Birrell free(iiburst); 1491673e404SJohn Birrell } 1501673e404SJohn Birrell 1511673e404SJohn Birrell /* 1521673e404SJohn Birrell * See if this iidesc matches the ELF symbol data we pass in. 1531673e404SJohn Birrell * 1541673e404SJohn Birrell * A fuzzy match is where we have a local symbol matching the name of a 1551673e404SJohn Birrell * global type description. This is common when a mapfile is used for a 1561673e404SJohn Birrell * DSO, but we don't accept it by default. 1571673e404SJohn Birrell * 1581673e404SJohn Birrell * A weak fuzzy match is when a weak symbol was resolved and matched to 1591673e404SJohn Birrell * a global type description. 1601673e404SJohn Birrell */ 1611673e404SJohn Birrell static int 1624cc75139SJohn Birrell matching_iidesc(void *arg1, void *arg2) 1631673e404SJohn Birrell { 1644cc75139SJohn Birrell iidesc_t *iidesc = arg1; 1654cc75139SJohn Birrell iidesc_match_t *match = arg2; 1661673e404SJohn Birrell if (streq(iidesc->ii_name, match->iim_name) == 0) 1671673e404SJohn Birrell return (0); 1681673e404SJohn Birrell 1691673e404SJohn Birrell switch (iidesc->ii_type) { 1701673e404SJohn Birrell case II_GFUN: 1711673e404SJohn Birrell case II_GVAR: 1721673e404SJohn Birrell if (match->iim_bind == STB_GLOBAL) { 1731673e404SJohn Birrell match->iim_ret = iidesc; 1741673e404SJohn Birrell return (-1); 1751673e404SJohn Birrell } else if (match->iim_fuzzy && match->iim_ret == NULL) { 1761673e404SJohn Birrell match->iim_ret = iidesc; 1771673e404SJohn Birrell /* continue to look for strong match */ 1781673e404SJohn Birrell return (0); 1791673e404SJohn Birrell } 1801673e404SJohn Birrell break; 1811673e404SJohn Birrell case II_SFUN: 1821673e404SJohn Birrell case II_SVAR: 1831673e404SJohn Birrell if (match->iim_bind == STB_LOCAL && 1841673e404SJohn Birrell match->iim_file != NULL && 1851673e404SJohn Birrell streq(iidesc->ii_owner, match->iim_file)) { 1861673e404SJohn Birrell match->iim_ret = iidesc; 1871673e404SJohn Birrell return (-1); 1881673e404SJohn Birrell } 1891673e404SJohn Birrell break; 1904cc75139SJohn Birrell default: 1914cc75139SJohn Birrell break; 1921673e404SJohn Birrell } 1931673e404SJohn Birrell return (0); 1941673e404SJohn Birrell } 1951673e404SJohn Birrell 1961673e404SJohn Birrell static iidesc_t * 1971673e404SJohn Birrell find_iidesc(tdata_t *td, iidesc_match_t *match) 1981673e404SJohn Birrell { 1991673e404SJohn Birrell match->iim_ret = NULL; 2001673e404SJohn Birrell iter_iidescs_by_name(td, match->iim_name, 2014cc75139SJohn Birrell matching_iidesc, match); 2021673e404SJohn Birrell return (match->iim_ret); 2031673e404SJohn Birrell } 2041673e404SJohn Birrell 2051673e404SJohn Birrell /* 2061673e404SJohn Birrell * If we have a weak symbol, attempt to find the strong symbol it will 2071673e404SJohn Birrell * resolve to. Note: the code where this actually happens is in 2081673e404SJohn Birrell * sym_process() in cmd/sgs/libld/common/syms.c 2091673e404SJohn Birrell * 2101673e404SJohn Birrell * Finding the matching symbol is unfortunately not trivial. For a 2111673e404SJohn Birrell * symbol to be a candidate, it must: 2121673e404SJohn Birrell * 2131673e404SJohn Birrell * - have the same type (function, object) 2141673e404SJohn Birrell * - have the same value (address) 2151673e404SJohn Birrell * - have the same size 2161673e404SJohn Birrell * - not be another weak symbol 2171673e404SJohn Birrell * - belong to the same section (checked via section index) 2181673e404SJohn Birrell * 2191673e404SJohn Birrell * If such a candidate is global, then we assume we've found it. The 2201673e404SJohn Birrell * linker generates the symbol table such that the curfile might be 2211673e404SJohn Birrell * incorrect; this is OK for global symbols, since find_iidesc() doesn't 2221673e404SJohn Birrell * need to check for the source file for the symbol. 2231673e404SJohn Birrell * 2241673e404SJohn Birrell * We might have found a strong local symbol, where the curfile is 2251673e404SJohn Birrell * accurate and matches that of the weak symbol. We assume this is a 2261673e404SJohn Birrell * reasonable match. 2271673e404SJohn Birrell * 2281673e404SJohn Birrell * If we've got a local symbol with a non-matching curfile, there are 2291673e404SJohn Birrell * two possibilities. Either this is a completely different symbol, or 2301673e404SJohn Birrell * it's a once-global symbol that was scoped to local via a mapfile. In 2311673e404SJohn Birrell * the latter case, curfile is likely inaccurate since the linker does 2321673e404SJohn Birrell * not preserve the needed curfile in the order of the symbol table (see 2331673e404SJohn Birrell * the comments about locally scoped symbols in libld's update_osym()). 2341673e404SJohn Birrell * As we can't tell this case from the former one, we use this symbol 2351673e404SJohn Birrell * iff no other matching symbol is found. 2361673e404SJohn Birrell * 2371673e404SJohn Birrell * What we really need here is a SUNW section containing weak<->strong 2381673e404SJohn Birrell * mappings that we can consume. 2391673e404SJohn Birrell */ 2401673e404SJohn Birrell static int 2411673e404SJohn Birrell check_for_weak(GElf_Sym *weak, char const *weakfile, 2421673e404SJohn Birrell Elf_Data *data, int nent, Elf_Data *strdata, 2431673e404SJohn Birrell GElf_Sym *retsym, char **curfilep) 2441673e404SJohn Birrell { 2451673e404SJohn Birrell char *curfile = NULL; 2464cc75139SJohn Birrell char *tmpfile1 = NULL; 2471673e404SJohn Birrell GElf_Sym tmpsym; 2481673e404SJohn Birrell int candidate = 0; 2491673e404SJohn Birrell int i; 2504cc75139SJohn Birrell tmpsym.st_info = 0; 2514cc75139SJohn Birrell tmpsym.st_name = 0; 2521673e404SJohn Birrell 2531673e404SJohn Birrell if (GELF_ST_BIND(weak->st_info) != STB_WEAK) 2541673e404SJohn Birrell return (0); 2551673e404SJohn Birrell 2561673e404SJohn Birrell for (i = 0; i < nent; i++) { 2571673e404SJohn Birrell GElf_Sym sym; 2581673e404SJohn Birrell uchar_t type; 2591673e404SJohn Birrell 2601673e404SJohn Birrell if (gelf_getsym(data, i, &sym) == NULL) 2611673e404SJohn Birrell continue; 2621673e404SJohn Birrell 2631673e404SJohn Birrell type = GELF_ST_TYPE(sym.st_info); 2641673e404SJohn Birrell 2651673e404SJohn Birrell if (type == STT_FILE) 2661673e404SJohn Birrell curfile = (char *)strdata->d_buf + sym.st_name; 2671673e404SJohn Birrell 2681673e404SJohn Birrell if (GELF_ST_TYPE(weak->st_info) != type || 2691673e404SJohn Birrell weak->st_value != sym.st_value) 2701673e404SJohn Birrell continue; 2711673e404SJohn Birrell 2721673e404SJohn Birrell if (weak->st_size != sym.st_size) 2731673e404SJohn Birrell continue; 2741673e404SJohn Birrell 2751673e404SJohn Birrell if (GELF_ST_BIND(sym.st_info) == STB_WEAK) 2761673e404SJohn Birrell continue; 2771673e404SJohn Birrell 2781673e404SJohn Birrell if (sym.st_shndx != weak->st_shndx) 2791673e404SJohn Birrell continue; 2801673e404SJohn Birrell 2811673e404SJohn Birrell if (GELF_ST_BIND(sym.st_info) == STB_LOCAL && 2821673e404SJohn Birrell (curfile == NULL || weakfile == NULL || 2831673e404SJohn Birrell strcmp(curfile, weakfile) != 0)) { 2841673e404SJohn Birrell candidate = 1; 2854cc75139SJohn Birrell tmpfile1 = curfile; 2861673e404SJohn Birrell tmpsym = sym; 2871673e404SJohn Birrell continue; 2881673e404SJohn Birrell } 2891673e404SJohn Birrell 2901673e404SJohn Birrell *curfilep = curfile; 2911673e404SJohn Birrell *retsym = sym; 2921673e404SJohn Birrell return (1); 2931673e404SJohn Birrell } 2941673e404SJohn Birrell 2951673e404SJohn Birrell if (candidate) { 2964cc75139SJohn Birrell *curfilep = tmpfile1; 2971673e404SJohn Birrell *retsym = tmpsym; 2981673e404SJohn Birrell return (1); 2991673e404SJohn Birrell } 3001673e404SJohn Birrell 3011673e404SJohn Birrell return (0); 3021673e404SJohn Birrell } 3031673e404SJohn Birrell 3041673e404SJohn Birrell /* 3051673e404SJohn Birrell * When we've found the underlying symbol's type description 3061673e404SJohn Birrell * for a weak symbol, we need to copy it and rename it to match 3071673e404SJohn Birrell * the weak symbol. We also need to add it to the td so it's 3081673e404SJohn Birrell * handled along with the others later. 3091673e404SJohn Birrell */ 3101673e404SJohn Birrell static iidesc_t * 3111673e404SJohn Birrell copy_from_strong(tdata_t *td, GElf_Sym *sym, iidesc_t *strongdesc, 3121673e404SJohn Birrell const char *weakname, const char *weakfile) 3131673e404SJohn Birrell { 3141673e404SJohn Birrell iidesc_t *new = iidesc_dup_rename(strongdesc, weakname, weakfile); 3151673e404SJohn Birrell uchar_t type = GELF_ST_TYPE(sym->st_info); 3161673e404SJohn Birrell 3171673e404SJohn Birrell switch (type) { 3181673e404SJohn Birrell case STT_OBJECT: 3191673e404SJohn Birrell new->ii_type = II_GVAR; 3201673e404SJohn Birrell break; 3211673e404SJohn Birrell case STT_FUNC: 3221673e404SJohn Birrell new->ii_type = II_GFUN; 3231673e404SJohn Birrell break; 3241673e404SJohn Birrell } 3251673e404SJohn Birrell 3261673e404SJohn Birrell hash_add(td->td_iihash, new); 3271673e404SJohn Birrell 3281673e404SJohn Birrell return (new); 3291673e404SJohn Birrell } 3301673e404SJohn Birrell 3311673e404SJohn Birrell /* 3321673e404SJohn Birrell * Process the symbol table of the output file, associating each symbol 3331673e404SJohn Birrell * with a type description if possible, and sorting them into functions 3341673e404SJohn Birrell * and data, maintaining symbol table order. 3351673e404SJohn Birrell */ 3361673e404SJohn Birrell static iiburst_t * 3371673e404SJohn Birrell sort_iidescs(Elf *elf, const char *file, tdata_t *td, int fuzzymatch, 3381673e404SJohn Birrell int dynsym) 3391673e404SJohn Birrell { 3401673e404SJohn Birrell iiburst_t *iiburst; 3411673e404SJohn Birrell Elf_Scn *scn; 3421673e404SJohn Birrell GElf_Shdr shdr; 3431673e404SJohn Birrell Elf_Data *data, *strdata; 3441673e404SJohn Birrell int i, stidx; 3451673e404SJohn Birrell int nent; 3461673e404SJohn Birrell iidesc_match_t match; 3471673e404SJohn Birrell 3481673e404SJohn Birrell match.iim_fuzzy = fuzzymatch; 3491673e404SJohn Birrell match.iim_file = NULL; 3501673e404SJohn Birrell 3511673e404SJohn Birrell if ((stidx = findelfsecidx(elf, file, 3521673e404SJohn Birrell dynsym ? ".dynsym" : ".symtab")) < 0) 3531673e404SJohn Birrell terminate("%s: Can't open symbol table\n", file); 3541673e404SJohn Birrell scn = elf_getscn(elf, stidx); 3551673e404SJohn Birrell data = elf_getdata(scn, NULL); 3561673e404SJohn Birrell gelf_getshdr(scn, &shdr); 3571673e404SJohn Birrell nent = shdr.sh_size / shdr.sh_entsize; 3581673e404SJohn Birrell 3591673e404SJohn Birrell scn = elf_getscn(elf, shdr.sh_link); 3601673e404SJohn Birrell strdata = elf_getdata(scn, NULL); 3611673e404SJohn Birrell 3621673e404SJohn Birrell iiburst = iiburst_new(td, nent); 3631673e404SJohn Birrell 3641673e404SJohn Birrell for (i = 0; i < nent; i++) { 3651673e404SJohn Birrell GElf_Sym sym; 3668b0d8e9fSDimitry Andric char *bname; 3671673e404SJohn Birrell iidesc_t **tolist; 3681673e404SJohn Birrell GElf_Sym ssym; 3691673e404SJohn Birrell iidesc_match_t smatch; 3701673e404SJohn Birrell int *curr; 3711673e404SJohn Birrell iidesc_t *iidesc; 3721673e404SJohn Birrell 3731673e404SJohn Birrell if (gelf_getsym(data, i, &sym) == NULL) 3741673e404SJohn Birrell elfterminate(file, "Couldn't read symbol %d", i); 3751673e404SJohn Birrell 3761673e404SJohn Birrell match.iim_name = (char *)strdata->d_buf + sym.st_name; 3771673e404SJohn Birrell match.iim_bind = GELF_ST_BIND(sym.st_info); 3781673e404SJohn Birrell 3791673e404SJohn Birrell switch (GELF_ST_TYPE(sym.st_info)) { 3801673e404SJohn Birrell case STT_FILE: 3818b0d8e9fSDimitry Andric bname = strrchr(match.iim_name, '/'); 382e86b103cSDimitry Andric match.iim_file = bname == NULL ? match.iim_name : bname + 1; 3831673e404SJohn Birrell continue; 3841673e404SJohn Birrell case STT_OBJECT: 3851673e404SJohn Birrell tolist = iiburst->iib_objts; 3861673e404SJohn Birrell curr = &iiburst->iib_nobjts; 3871673e404SJohn Birrell break; 3881673e404SJohn Birrell case STT_FUNC: 3891673e404SJohn Birrell tolist = iiburst->iib_funcs; 3901673e404SJohn Birrell curr = &iiburst->iib_nfuncs; 3911673e404SJohn Birrell break; 3921673e404SJohn Birrell default: 3931673e404SJohn Birrell continue; 3941673e404SJohn Birrell } 3951673e404SJohn Birrell 3961673e404SJohn Birrell if (ignore_symbol(&sym, match.iim_name)) 3971673e404SJohn Birrell continue; 3981673e404SJohn Birrell 3991673e404SJohn Birrell iidesc = find_iidesc(td, &match); 4001673e404SJohn Birrell 4011673e404SJohn Birrell if (iidesc != NULL) { 4021673e404SJohn Birrell tolist[*curr] = iidesc; 4031673e404SJohn Birrell iidesc->ii_flags |= IIDESC_F_USED; 4041673e404SJohn Birrell (*curr)++; 4051673e404SJohn Birrell continue; 4061673e404SJohn Birrell } 4071673e404SJohn Birrell 4081673e404SJohn Birrell if (!check_for_weak(&sym, match.iim_file, data, nent, strdata, 4091673e404SJohn Birrell &ssym, &smatch.iim_file)) { 4101673e404SJohn Birrell (*curr)++; 4111673e404SJohn Birrell continue; 4121673e404SJohn Birrell } 4131673e404SJohn Birrell 4141673e404SJohn Birrell smatch.iim_fuzzy = fuzzymatch; 4151673e404SJohn Birrell smatch.iim_name = (char *)strdata->d_buf + ssym.st_name; 4161673e404SJohn Birrell smatch.iim_bind = GELF_ST_BIND(ssym.st_info); 4171673e404SJohn Birrell 4181673e404SJohn Birrell debug(3, "Weak symbol %s resolved to %s\n", match.iim_name, 4191673e404SJohn Birrell smatch.iim_name); 4201673e404SJohn Birrell 4211673e404SJohn Birrell iidesc = find_iidesc(td, &smatch); 4221673e404SJohn Birrell 4231673e404SJohn Birrell if (iidesc != NULL) { 4241673e404SJohn Birrell tolist[*curr] = copy_from_strong(td, &sym, 4251673e404SJohn Birrell iidesc, match.iim_name, match.iim_file); 4261673e404SJohn Birrell tolist[*curr]->ii_flags |= IIDESC_F_USED; 4271673e404SJohn Birrell } 4281673e404SJohn Birrell 4291673e404SJohn Birrell (*curr)++; 4301673e404SJohn Birrell } 4311673e404SJohn Birrell 4321673e404SJohn Birrell /* 4331673e404SJohn Birrell * Stabs are generated for every function declared in a given C source 4341673e404SJohn Birrell * file. When converting an object file, we may encounter a stab that 4351673e404SJohn Birrell * has no symbol table entry because the optimizer has decided to omit 4361673e404SJohn Birrell * that item (for example, an unreferenced static function). We may 4371673e404SJohn Birrell * see iidescs that do not have an associated symtab entry, and so 4381673e404SJohn Birrell * we do not write records for those functions into the CTF data. 4391673e404SJohn Birrell * All others get marked as a root by this function. 4401673e404SJohn Birrell */ 4411673e404SJohn Birrell iiburst_types(iiburst); 4421673e404SJohn Birrell 4431673e404SJohn Birrell /* 4441673e404SJohn Birrell * By not adding some of the functions and/or objects, we may have 4451673e404SJohn Birrell * caused some types that were referenced solely by those 4461673e404SJohn Birrell * functions/objects to be suppressed. This could cause a label, 4471673e404SJohn Birrell * generated prior to the evisceration, to be incorrect. Find the 4481673e404SJohn Birrell * highest type index, and change the label indicies to be no higher 4491673e404SJohn Birrell * than this value. 4501673e404SJohn Birrell */ 4511673e404SJohn Birrell tdata_label_newmax(td, iiburst->iib_maxtypeid); 4521673e404SJohn Birrell 4531673e404SJohn Birrell return (iiburst); 4541673e404SJohn Birrell } 4551673e404SJohn Birrell 4561673e404SJohn Birrell static void 4571673e404SJohn Birrell write_file(Elf *src, const char *srcname, Elf *dst, const char *dstname, 4581673e404SJohn Birrell caddr_t ctfdata, size_t ctfsize, int flags) 4591673e404SJohn Birrell { 4601673e404SJohn Birrell GElf_Ehdr sehdr, dehdr; 4611673e404SJohn Birrell Elf_Scn *sscn, *dscn; 4621673e404SJohn Birrell Elf_Data *sdata, *ddata; 4631673e404SJohn Birrell GElf_Shdr shdr; 4641673e404SJohn Birrell GElf_Word symtab_type; 4651673e404SJohn Birrell int symtab_idx = -1; 4661673e404SJohn Birrell off_t new_offset = 0; 4671673e404SJohn Birrell off_t ctfnameoff = 0; 4681673e404SJohn Birrell int dynsym = (flags & CTF_USE_DYNSYM); 4691673e404SJohn Birrell int keep_stabs = (flags & CTF_KEEP_STABS); 4701673e404SJohn Birrell int *secxlate; 4711673e404SJohn Birrell int srcidx, dstidx; 4721673e404SJohn Birrell int curnmoff = 0; 4731673e404SJohn Birrell int changing = 0; 4741673e404SJohn Birrell int pad; 4751673e404SJohn Birrell int i; 4761673e404SJohn Birrell 4771673e404SJohn Birrell if (gelf_newehdr(dst, gelf_getclass(src)) == NULL) 4781673e404SJohn Birrell elfterminate(dstname, "Cannot copy ehdr to temp file"); 4791673e404SJohn Birrell gelf_getehdr(src, &sehdr); 4801673e404SJohn Birrell memcpy(&dehdr, &sehdr, sizeof (GElf_Ehdr)); 4811673e404SJohn Birrell gelf_update_ehdr(dst, &dehdr); 4821673e404SJohn Birrell 4831673e404SJohn Birrell symtab_type = dynsym ? SHT_DYNSYM : SHT_SYMTAB; 4841673e404SJohn Birrell 4851673e404SJohn Birrell /* 4861673e404SJohn Birrell * Neither the existing stab sections nor the SUNW_ctf sections (new or 4871673e404SJohn Birrell * existing) are SHF_ALLOC'd, so they won't be in areas referenced by 4881673e404SJohn Birrell * program headers. As such, we can just blindly copy the program 4891673e404SJohn Birrell * headers from the existing file to the new file. 4901673e404SJohn Birrell */ 4911673e404SJohn Birrell if (sehdr.e_phnum != 0) { 4921673e404SJohn Birrell (void) elf_flagelf(dst, ELF_C_SET, ELF_F_LAYOUT); 4931673e404SJohn Birrell if (gelf_newphdr(dst, sehdr.e_phnum) == NULL) 4941673e404SJohn Birrell elfterminate(dstname, "Cannot make phdrs in temp file"); 4951673e404SJohn Birrell 4961673e404SJohn Birrell for (i = 0; i < sehdr.e_phnum; i++) { 4971673e404SJohn Birrell GElf_Phdr phdr; 4981673e404SJohn Birrell 4991673e404SJohn Birrell gelf_getphdr(src, i, &phdr); 5001673e404SJohn Birrell gelf_update_phdr(dst, i, &phdr); 5011673e404SJohn Birrell } 5021673e404SJohn Birrell } 5031673e404SJohn Birrell 5041673e404SJohn Birrell secxlate = xmalloc(sizeof (int) * sehdr.e_shnum); 5051673e404SJohn Birrell for (srcidx = dstidx = 0; srcidx < sehdr.e_shnum; srcidx++) { 5061673e404SJohn Birrell Elf_Scn *scn = elf_getscn(src, srcidx); 5074cc75139SJohn Birrell GElf_Shdr shdr1; 5081673e404SJohn Birrell char *sname; 5091673e404SJohn Birrell 5104cc75139SJohn Birrell gelf_getshdr(scn, &shdr1); 5114cc75139SJohn Birrell sname = elf_strptr(src, sehdr.e_shstrndx, shdr1.sh_name); 5121673e404SJohn Birrell if (sname == NULL) { 5131673e404SJohn Birrell elfterminate(srcname, "Can't find string at %u", 5144cc75139SJohn Birrell shdr1.sh_name); 5151673e404SJohn Birrell } 5161673e404SJohn Birrell 5171673e404SJohn Birrell if (strcmp(sname, CTF_ELF_SCN_NAME) == 0) { 5181673e404SJohn Birrell secxlate[srcidx] = -1; 5191673e404SJohn Birrell } else if (!keep_stabs && 5201673e404SJohn Birrell (strncmp(sname, ".stab", 5) == 0 || 5211673e404SJohn Birrell strncmp(sname, ".debug", 6) == 0 || 5221673e404SJohn Birrell strncmp(sname, ".rel.debug", 10) == 0 || 5231673e404SJohn Birrell strncmp(sname, ".rela.debug", 11) == 0)) { 5241673e404SJohn Birrell secxlate[srcidx] = -1; 5254cc75139SJohn Birrell } else if (dynsym && shdr1.sh_type == SHT_SYMTAB) { 5261673e404SJohn Birrell /* 5271673e404SJohn Birrell * If we're building CTF against the dynsym, 5281673e404SJohn Birrell * we'll rip out the symtab so debuggers aren't 5291673e404SJohn Birrell * confused. 5301673e404SJohn Birrell */ 5311673e404SJohn Birrell secxlate[srcidx] = -1; 5321673e404SJohn Birrell } else { 5331673e404SJohn Birrell secxlate[srcidx] = dstidx++; 5341673e404SJohn Birrell curnmoff += strlen(sname) + 1; 5351673e404SJohn Birrell } 5361673e404SJohn Birrell 5371673e404SJohn Birrell new_offset = (off_t)dehdr.e_phoff; 5381673e404SJohn Birrell } 5391673e404SJohn Birrell 5401673e404SJohn Birrell for (srcidx = 1; srcidx < sehdr.e_shnum; srcidx++) { 5411673e404SJohn Birrell char *sname; 5421673e404SJohn Birrell 5431673e404SJohn Birrell sscn = elf_getscn(src, srcidx); 5441673e404SJohn Birrell gelf_getshdr(sscn, &shdr); 5451673e404SJohn Birrell 5461673e404SJohn Birrell if (secxlate[srcidx] == -1) { 5471673e404SJohn Birrell changing = 1; 5481673e404SJohn Birrell continue; 5491673e404SJohn Birrell } 5501673e404SJohn Birrell 5511673e404SJohn Birrell dscn = elf_newscn(dst); 5521673e404SJohn Birrell 5531673e404SJohn Birrell /* 5541673e404SJohn Birrell * If this file has program headers, we need to explicitly lay 5551673e404SJohn Birrell * out sections. If none of the sections prior to this one have 5561673e404SJohn Birrell * been removed, then we can just use the existing location. If 5571673e404SJohn Birrell * one or more sections have been changed, then we need to 5581673e404SJohn Birrell * adjust this one to avoid holes. 5591673e404SJohn Birrell */ 5601673e404SJohn Birrell if (changing && sehdr.e_phnum != 0) { 5611673e404SJohn Birrell pad = new_offset % shdr.sh_addralign; 5621673e404SJohn Birrell 5631673e404SJohn Birrell if (pad) 5641673e404SJohn Birrell new_offset += shdr.sh_addralign - pad; 5651673e404SJohn Birrell shdr.sh_offset = new_offset; 5661673e404SJohn Birrell } 5671673e404SJohn Birrell 5681673e404SJohn Birrell shdr.sh_link = secxlate[shdr.sh_link]; 5691673e404SJohn Birrell 5701673e404SJohn Birrell if (shdr.sh_type == SHT_REL || shdr.sh_type == SHT_RELA) 5711673e404SJohn Birrell shdr.sh_info = secxlate[shdr.sh_info]; 5721673e404SJohn Birrell 5731673e404SJohn Birrell sname = elf_strptr(src, sehdr.e_shstrndx, shdr.sh_name); 5741673e404SJohn Birrell if (sname == NULL) { 5751673e404SJohn Birrell elfterminate(srcname, "Can't find string at %u", 5761673e404SJohn Birrell shdr.sh_name); 5771673e404SJohn Birrell } 5784cc75139SJohn Birrell 579*bc96366cSSteven Hartland #ifndef illumos 5804cc75139SJohn Birrell if (gelf_update_shdr(dscn, &shdr) == 0) 5814cc75139SJohn Birrell elfterminate(dstname, "Cannot update sect %s", sname); 5824cc75139SJohn Birrell #endif 5834cc75139SJohn Birrell 5841673e404SJohn Birrell if ((sdata = elf_getdata(sscn, NULL)) == NULL) 5851673e404SJohn Birrell elfterminate(srcname, "Cannot get sect %s data", sname); 5861673e404SJohn Birrell if ((ddata = elf_newdata(dscn)) == NULL) 5871673e404SJohn Birrell elfterminate(dstname, "Can't make sect %s data", sname); 588*bc96366cSSteven Hartland #ifdef illumos 5891673e404SJohn Birrell bcopy(sdata, ddata, sizeof (Elf_Data)); 5904cc75139SJohn Birrell #else 5914cc75139SJohn Birrell /* 5924cc75139SJohn Birrell * FreeBSD's Elf_Data has private fields which the 5934cc75139SJohn Birrell * elf_* routines manage. Simply copying the 5944cc75139SJohn Birrell * entire structure corrupts the data. So we need 5954cc75139SJohn Birrell * to copy the public fields explictly. 5964cc75139SJohn Birrell */ 5974cc75139SJohn Birrell ddata->d_align = sdata->d_align; 5984cc75139SJohn Birrell ddata->d_off = sdata->d_off; 5994cc75139SJohn Birrell ddata->d_size = sdata->d_size; 6004cc75139SJohn Birrell ddata->d_type = sdata->d_type; 6014cc75139SJohn Birrell ddata->d_version = sdata->d_version; 6024cc75139SJohn Birrell #endif 6031673e404SJohn Birrell 6041673e404SJohn Birrell if (srcidx == sehdr.e_shstrndx) { 6051673e404SJohn Birrell char seclen = strlen(CTF_ELF_SCN_NAME); 6061673e404SJohn Birrell 6071673e404SJohn Birrell ddata->d_buf = xmalloc(ddata->d_size + shdr.sh_size + 6081673e404SJohn Birrell seclen + 1); 6091673e404SJohn Birrell bcopy(sdata->d_buf, ddata->d_buf, shdr.sh_size); 6101673e404SJohn Birrell strcpy((caddr_t)ddata->d_buf + shdr.sh_size, 6111673e404SJohn Birrell CTF_ELF_SCN_NAME); 6121673e404SJohn Birrell ctfnameoff = (off_t)shdr.sh_size; 6131673e404SJohn Birrell shdr.sh_size += seclen + 1; 6141673e404SJohn Birrell ddata->d_size += seclen + 1; 6151673e404SJohn Birrell 6161673e404SJohn Birrell if (sehdr.e_phnum != 0) 6171673e404SJohn Birrell changing = 1; 6181673e404SJohn Birrell } 6191673e404SJohn Birrell 6201673e404SJohn Birrell if (shdr.sh_type == symtab_type && shdr.sh_entsize != 0) { 6211673e404SJohn Birrell int nsym = shdr.sh_size / shdr.sh_entsize; 6221673e404SJohn Birrell 6231673e404SJohn Birrell symtab_idx = secxlate[srcidx]; 6241673e404SJohn Birrell 6251673e404SJohn Birrell ddata->d_buf = xmalloc(shdr.sh_size); 6261673e404SJohn Birrell bcopy(sdata->d_buf, ddata->d_buf, shdr.sh_size); 6271673e404SJohn Birrell 6281673e404SJohn Birrell for (i = 0; i < nsym; i++) { 6291673e404SJohn Birrell GElf_Sym sym; 6301673e404SJohn Birrell short newscn; 6311673e404SJohn Birrell 6324cc75139SJohn Birrell if (gelf_getsym(ddata, i, &sym) == NULL) 6334cc75139SJohn Birrell printf("Could not get symbol %d\n",i); 6341673e404SJohn Birrell 6351673e404SJohn Birrell if (sym.st_shndx >= SHN_LORESERVE) 6361673e404SJohn Birrell continue; 6371673e404SJohn Birrell 6381673e404SJohn Birrell if ((newscn = secxlate[sym.st_shndx]) != 6391673e404SJohn Birrell sym.st_shndx) { 6401673e404SJohn Birrell sym.st_shndx = 6411673e404SJohn Birrell (newscn == -1 ? 1 : newscn); 6421673e404SJohn Birrell 6431673e404SJohn Birrell gelf_update_sym(ddata, i, &sym); 6441673e404SJohn Birrell } 6451673e404SJohn Birrell } 6461673e404SJohn Birrell } 6471673e404SJohn Birrell 648*bc96366cSSteven Hartland #ifndef illumos 64996d1cd18SNavdeep Parhar if (ddata->d_buf == NULL && sdata->d_buf != NULL) { 6504cc75139SJohn Birrell ddata->d_buf = xmalloc(shdr.sh_size); 6514cc75139SJohn Birrell bcopy(sdata->d_buf, ddata->d_buf, shdr.sh_size); 6524cc75139SJohn Birrell } 6534cc75139SJohn Birrell #endif 6544cc75139SJohn Birrell 6554cc75139SJohn Birrell if (gelf_update_shdr(dscn, &shdr) == 0) 6561673e404SJohn Birrell elfterminate(dstname, "Cannot update sect %s", sname); 6571673e404SJohn Birrell 6581673e404SJohn Birrell new_offset = (off_t)shdr.sh_offset; 6591673e404SJohn Birrell if (shdr.sh_type != SHT_NOBITS) 6601673e404SJohn Birrell new_offset += shdr.sh_size; 6611673e404SJohn Birrell } 6621673e404SJohn Birrell 6631673e404SJohn Birrell if (symtab_idx == -1) { 6641673e404SJohn Birrell terminate("%s: Cannot find %s section\n", srcname, 6651673e404SJohn Birrell dynsym ? "SHT_DYNSYM" : "SHT_SYMTAB"); 6661673e404SJohn Birrell } 6671673e404SJohn Birrell 6681673e404SJohn Birrell /* Add the ctf section */ 6691673e404SJohn Birrell dscn = elf_newscn(dst); 6701673e404SJohn Birrell gelf_getshdr(dscn, &shdr); 6711673e404SJohn Birrell shdr.sh_name = ctfnameoff; 6721673e404SJohn Birrell shdr.sh_type = SHT_PROGBITS; 6731673e404SJohn Birrell shdr.sh_size = ctfsize; 6741673e404SJohn Birrell shdr.sh_link = symtab_idx; 6751673e404SJohn Birrell shdr.sh_addralign = 4; 6761673e404SJohn Birrell if (changing && sehdr.e_phnum != 0) { 6771673e404SJohn Birrell pad = new_offset % shdr.sh_addralign; 6781673e404SJohn Birrell 6791673e404SJohn Birrell if (pad) 6801673e404SJohn Birrell new_offset += shdr.sh_addralign - pad; 6811673e404SJohn Birrell 6821673e404SJohn Birrell shdr.sh_offset = new_offset; 6831673e404SJohn Birrell new_offset += shdr.sh_size; 6841673e404SJohn Birrell } 6851673e404SJohn Birrell 6861673e404SJohn Birrell ddata = elf_newdata(dscn); 6871673e404SJohn Birrell ddata->d_buf = ctfdata; 6881673e404SJohn Birrell ddata->d_size = ctfsize; 6891673e404SJohn Birrell ddata->d_align = shdr.sh_addralign; 6904cc75139SJohn Birrell ddata->d_off = 0; 6911673e404SJohn Birrell 6921673e404SJohn Birrell gelf_update_shdr(dscn, &shdr); 6931673e404SJohn Birrell 6941673e404SJohn Birrell /* update the section header location */ 6951673e404SJohn Birrell if (sehdr.e_phnum != 0) { 6961673e404SJohn Birrell size_t align = gelf_fsize(dst, ELF_T_ADDR, 1, EV_CURRENT); 6971673e404SJohn Birrell size_t r = new_offset % align; 6981673e404SJohn Birrell 6991673e404SJohn Birrell if (r) 7001673e404SJohn Birrell new_offset += align - r; 7011673e404SJohn Birrell 7021673e404SJohn Birrell dehdr.e_shoff = new_offset; 7031673e404SJohn Birrell } 7041673e404SJohn Birrell 7051673e404SJohn Birrell /* commit to disk */ 7061673e404SJohn Birrell dehdr.e_shstrndx = secxlate[sehdr.e_shstrndx]; 7071673e404SJohn Birrell gelf_update_ehdr(dst, &dehdr); 7081673e404SJohn Birrell if (elf_update(dst, ELF_C_WRITE) < 0) 7091673e404SJohn Birrell elfterminate(dstname, "Cannot finalize temp file"); 7101673e404SJohn Birrell 7111673e404SJohn Birrell free(secxlate); 7121673e404SJohn Birrell } 7131673e404SJohn Birrell 7141673e404SJohn Birrell static caddr_t 7151673e404SJohn Birrell make_ctf_data(tdata_t *td, Elf *elf, const char *file, size_t *lenp, int flags) 7161673e404SJohn Birrell { 7171673e404SJohn Birrell iiburst_t *iiburst; 7181673e404SJohn Birrell caddr_t data; 7191673e404SJohn Birrell 7201673e404SJohn Birrell iiburst = sort_iidescs(elf, file, td, flags & CTF_FUZZY_MATCH, 7211673e404SJohn Birrell flags & CTF_USE_DYNSYM); 722a6425ab5SOleksandr Tymoshenko data = ctf_gen(iiburst, lenp, flags & (CTF_COMPRESS | CTF_SWAP_BYTES)); 7231673e404SJohn Birrell 7241673e404SJohn Birrell iiburst_free(iiburst); 7251673e404SJohn Birrell 7261673e404SJohn Birrell return (data); 7271673e404SJohn Birrell } 7281673e404SJohn Birrell 7291673e404SJohn Birrell void 7301673e404SJohn Birrell write_ctf(tdata_t *td, const char *curname, const char *newname, int flags) 7311673e404SJohn Birrell { 7321673e404SJohn Birrell struct stat st; 7331673e404SJohn Birrell Elf *elf = NULL; 7341673e404SJohn Birrell Elf *telf = NULL; 735a6425ab5SOleksandr Tymoshenko GElf_Ehdr ehdr; 7361673e404SJohn Birrell caddr_t data; 7371673e404SJohn Birrell size_t len; 7381673e404SJohn Birrell int fd = -1; 7391673e404SJohn Birrell int tfd = -1; 740a6425ab5SOleksandr Tymoshenko int byteorder; 7411673e404SJohn Birrell 7421673e404SJohn Birrell (void) elf_version(EV_CURRENT); 7431673e404SJohn Birrell if ((fd = open(curname, O_RDONLY)) < 0 || fstat(fd, &st) < 0) 7441673e404SJohn Birrell terminate("%s: Cannot open for re-reading", curname); 7451673e404SJohn Birrell if ((elf = elf_begin(fd, ELF_C_READ, NULL)) == NULL) 7461673e404SJohn Birrell elfterminate(curname, "Cannot re-read"); 7471673e404SJohn Birrell 7481673e404SJohn Birrell if ((tfd = open(newname, O_RDWR | O_CREAT | O_TRUNC, st.st_mode)) < 0) 7491673e404SJohn Birrell terminate("Cannot open temp file %s for writing", newname); 7501673e404SJohn Birrell if ((telf = elf_begin(tfd, ELF_C_WRITE, NULL)) == NULL) 7511673e404SJohn Birrell elfterminate(curname, "Cannot write"); 7521673e404SJohn Birrell 753a6425ab5SOleksandr Tymoshenko if (gelf_getehdr(elf, &ehdr)) { 754a6425ab5SOleksandr Tymoshenko #if BYTE_ORDER == _BIG_ENDIAN 755a6425ab5SOleksandr Tymoshenko byteorder = ELFDATA2MSB; 756a6425ab5SOleksandr Tymoshenko #else 757a6425ab5SOleksandr Tymoshenko byteorder = ELFDATA2LSB; 758a6425ab5SOleksandr Tymoshenko #endif 759a6425ab5SOleksandr Tymoshenko /* 760a6425ab5SOleksandr Tymoshenko * If target and host has the same byte order 761a6425ab5SOleksandr Tymoshenko * clear byte swapping request 762a6425ab5SOleksandr Tymoshenko */ 763a6425ab5SOleksandr Tymoshenko if (ehdr.e_ident[EI_DATA] == byteorder) 764a6425ab5SOleksandr Tymoshenko flags &= ~CTF_SWAP_BYTES; 765a6425ab5SOleksandr Tymoshenko } 766a6425ab5SOleksandr Tymoshenko else 767a6425ab5SOleksandr Tymoshenko elfterminate(curname, "Failed to get EHDR"); 768a6425ab5SOleksandr Tymoshenko 7691673e404SJohn Birrell data = make_ctf_data(td, elf, curname, &len, flags); 7701673e404SJohn Birrell write_file(elf, curname, telf, newname, data, len, flags); 7711673e404SJohn Birrell free(data); 7721673e404SJohn Birrell 7731673e404SJohn Birrell elf_end(telf); 7741673e404SJohn Birrell elf_end(elf); 7751673e404SJohn Birrell (void) close(fd); 7761673e404SJohn Birrell (void) close(tfd); 7771673e404SJohn Birrell } 778