1 /* Find debugging and symbol information for a module in libdwfl.
2    Copyright (C) 2005-2013 Red Hat, Inc.
3    This file is part of elfutils.
4 
5    This file is free software; you can redistribute it and/or modify
6    it under the terms of either
7 
8      * the GNU Lesser General Public License as published by the Free
9        Software Foundation; either version 3 of the License, or (at
10        your option) any later version
11 
12    or
13 
14      * the GNU General Public License as published by the Free
15        Software Foundation; either version 2 of the License, or (at
16        your option) any later version
17 
18    or both in parallel, as here.
19 
20    elfutils is distributed in the hope that it will be useful, but
21    WITHOUT ANY WARRANTY; without even the implied warranty of
22    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
23    General Public License for more details.
24 
25    You should have received copies of the GNU General Public License and
26    the GNU Lesser General Public License along with this program.  If
27    not, see <http://www.gnu.org/licenses/>.  */
28 
29 #ifdef HAVE_CONFIG_H
30 # include <config.h>
31 #endif
32 
33 #include "libdwflP.h"
34 
35 struct search_state
36 {
37   Dwfl_Module *mod;
38   GElf_Addr addr;
39 
40   GElf_Sym *closest_sym;
41   bool adjust_st_value;
42   GElf_Word addr_shndx;
43   Elf *addr_symelf;
44 
45   /* Keep track of the closest symbol we have seen so far.
46      Here we store only symbols with nonzero st_size.  */
47   const char *closest_name;
48   GElf_Addr closest_value;
49   GElf_Word closest_shndx;
50   Elf *closest_elf;
51 
52   /* Keep track of an eligible symbol with st_size == 0 as a fallback.  */
53   const char *sizeless_name;
54   GElf_Sym sizeless_sym;
55   GElf_Addr sizeless_value;
56   GElf_Word sizeless_shndx;
57   Elf *sizeless_elf;
58 
59   /* Keep track of the lowest address a relevant sizeless symbol could have.  */
60   GElf_Addr min_label;
61 };
62 
63 /* Return true iff we consider ADDR to lie in the same section as SYM.  */
64 static inline bool
same_section(struct search_state * state,GElf_Addr value,Elf * symelf,GElf_Word shndx)65 same_section (struct search_state *state,
66 	      GElf_Addr value, Elf *symelf, GElf_Word shndx)
67 {
68   /* For absolute symbols and the like, only match exactly.  */
69   if (shndx >= SHN_LORESERVE)
70     return value == state->addr;
71 
72   /* If value might not be st_value, the shndx of the symbol might
73       not match the section of the value. Explicitly look both up.  */
74   if (! state->adjust_st_value)
75     {
76       Dwarf_Addr v;
77       if (state->addr_shndx == SHN_UNDEF)
78         {
79           v = state->addr;
80           state->addr_shndx = __libdwfl_find_section_ndx (state->mod, &v);
81         }
82 
83       v = value;
84       return state->addr_shndx == __libdwfl_find_section_ndx (state->mod, &v);
85     }
86 
87   /* Figure out what section ADDR lies in.  */
88   if (state->addr_shndx == SHN_UNDEF || state->addr_symelf != symelf)
89     {
90       GElf_Addr mod_addr = dwfl_deadjust_st_value (state->mod, symelf,
91 						   state->addr);
92       Elf_Scn *scn = NULL;
93       state->addr_shndx = SHN_ABS;
94       state->addr_symelf = symelf;
95       while ((scn = elf_nextscn (symelf, scn)) != NULL)
96         {
97           GElf_Shdr shdr_mem;
98           GElf_Shdr *shdr = gelf_getshdr (scn, &shdr_mem);
99           if (likely (shdr != NULL)
100               && mod_addr >= shdr->sh_addr
101               && mod_addr < shdr->sh_addr + shdr->sh_size)
102             {
103               state->addr_shndx = elf_ndxscn (scn);
104               break;
105             }
106         }
107     }
108 
109   return shndx == state->addr_shndx && state->addr_symelf == symelf;
110 }
111 
112 /* Return GELF_ST_BIND as higher-is-better integer.  */
113 static inline int
binding_value(const GElf_Sym * symp)114 binding_value (const GElf_Sym *symp)
115 {
116   switch (GELF_ST_BIND (symp->st_info))
117     {
118     case STB_GLOBAL:
119       return 3;
120     case STB_WEAK:
121       return 2;
122     case STB_LOCAL:
123       return 1;
124     default:
125       return 0;
126     }
127 }
128 
129 /* Try one symbol and associated value from the search table.  */
130 static inline void
try_sym_value(struct search_state * state,GElf_Addr value,GElf_Sym * sym,const char * name,GElf_Word shndx,Elf * elf,bool resolved)131 try_sym_value (struct search_state *state,
132                GElf_Addr value, GElf_Sym *sym,
133                const char *name, GElf_Word shndx,
134                Elf *elf, bool resolved)
135 {
136     /* Even if we don't choose this symbol, its existence excludes
137        any sizeless symbol (assembly label) that is below its upper
138        bound.  */
139     if (value + sym->st_size > state->min_label)
140       state->min_label = value + sym->st_size;
141 
142     if (sym->st_size == 0 || state->addr - value < sym->st_size)
143       {
144 	/* This symbol is a better candidate than the current one
145 	   if it's closer to ADDR or is global when it was local.  */
146 	if (state->closest_name == NULL
147 	    || state->closest_value < value
148 	    || binding_value (state->closest_sym) < binding_value (sym))
149 	  {
150 	    if (sym->st_size != 0)
151 	      {
152 		*state->closest_sym = *sym;
153 		state->closest_value = value;
154 		state->closest_shndx = shndx;
155 		state->closest_elf = elf;
156 		state->closest_name = name;
157 	      }
158 	    else if (state->closest_name == NULL
159 		     && value >= state->min_label
160 		     && same_section (state, value,
161 				      resolved ? state->mod->main.elf : elf,
162 				      shndx))
163 	      {
164 		/* Handwritten assembly symbols sometimes have no
165 		   st_size.  If no symbol with proper size includes
166 		   the address, we'll use the closest one that is in
167 		   the same section as ADDR.  */
168 		state->sizeless_sym = *sym;
169 		state->sizeless_value = value;
170 		state->sizeless_shndx = shndx;
171 		state->sizeless_elf = elf;
172 		state->sizeless_name = name;
173 	      }
174 	  }
175 	/* When the beginning of its range is no closer,
176 	   the end of its range might be.  Otherwise follow
177 	   GELF_ST_BIND preference.  If all are equal prefer
178 	   the first symbol found.  */
179 	else if (sym->st_size != 0
180 		 && state->closest_value == value
181 		 && ((state->closest_sym->st_size > sym->st_size
182 		      && (binding_value (state->closest_sym)
183 			  <= binding_value (sym)))
184 		     || (state->closest_sym->st_size >= sym->st_size
185 			 && (binding_value (state->closest_sym)
186 			     < binding_value (sym)))))
187 	  {
188 	    *state->closest_sym = *sym;
189 	    state->closest_value = value;
190 	    state->closest_shndx = shndx;
191 	    state->closest_elf = elf;
192 	    state->closest_name = name;
193 	  }
194       }
195 }
196 
197 /* Look through the symbol table for a matching symbol.  */
198 static inline void
search_table(struct search_state * state,int start,int end)199 search_table (struct search_state *state, int start, int end)
200 {
201       for (int i = start; i < end; ++i)
202 	{
203 	  GElf_Sym sym;
204 	  GElf_Addr value;
205 	  GElf_Word shndx;
206 	  Elf *elf;
207 	  bool resolved;
208 	  const char *name = __libdwfl_getsym (state->mod, i, &sym, &value,
209 					       &shndx, &elf, NULL,
210 					       &resolved,
211 					       state->adjust_st_value);
212 	  if (name != NULL && name[0] != '\0'
213 	      && sym.st_shndx != SHN_UNDEF
214 	      && value <= state->addr
215 	      && GELF_ST_TYPE (sym.st_info) != STT_SECTION
216 	      && GELF_ST_TYPE (sym.st_info) != STT_FILE
217 	      && GELF_ST_TYPE (sym.st_info) != STT_TLS)
218 	    {
219 	      try_sym_value (state, value, &sym, name, shndx, elf, resolved);
220 
221 	      /* If this is an addrinfo variant and the value could be
222 		 resolved then also try matching the (adjusted) st_value.  */
223 	      if (resolved && state->mod->e_type != ET_REL)
224 		{
225 		  GElf_Addr adjusted_st_value;
226 		  adjusted_st_value = dwfl_adjusted_st_value (state->mod, elf,
227 							      sym.st_value);
228 		  if (value != adjusted_st_value
229 		      && adjusted_st_value <= state->addr)
230 		    try_sym_value (state, adjusted_st_value, &sym, name, shndx,
231 				   elf, false);
232 		}
233 	    }
234 	}
235 }
236 
237 /* Returns the name of the symbol "closest" to ADDR.
238    Never returns symbols at addresses above ADDR.
239 
240    Wrapper for old dwfl_module_addrsym and new dwfl_module_addrinfo.
241    adjust_st_value set to true returns adjusted SYM st_value, set to false
242    it will not adjust SYM at all, but does match against resolved values.   */
243 static const char *
__libdwfl_addrsym(Dwfl_Module * _mod,GElf_Addr _addr,GElf_Off * off,GElf_Sym * _closest_sym,GElf_Word * shndxp,Elf ** elfp,Dwarf_Addr * biasp,bool _adjust_st_value)244 __libdwfl_addrsym (Dwfl_Module *_mod, GElf_Addr _addr, GElf_Off *off,
245 		   GElf_Sym *_closest_sym, GElf_Word *shndxp,
246 		   Elf **elfp, Dwarf_Addr *biasp, bool _adjust_st_value)
247 {
248   int syments = INTUSE(dwfl_module_getsymtab) (_mod);
249   if (syments < 0)
250     return NULL;
251 
252   struct search_state state =
253     {
254       .addr = _addr,
255       .mod = _mod,
256       .closest_sym = _closest_sym,
257       .adjust_st_value = _adjust_st_value,
258       .addr_shndx = SHN_UNDEF,
259       .addr_symelf = NULL,
260       .closest_name = NULL,
261       .closest_value = 0,
262       .closest_shndx = SHN_UNDEF,
263       .closest_elf = NULL,
264       .sizeless_name = NULL,
265       .sizeless_sym = { 0, 0, 0, 0, 0, SHN_UNDEF },
266       .sizeless_value = 0,
267       .sizeless_shndx = SHN_UNDEF,
268       .sizeless_elf = NULL,
269       .min_label = 0
270     };
271 
272   /* First go through global symbols.  mod->first_global and
273      mod->aux_first_global are setup by dwfl_module_getsymtab to the
274      index of the first global symbol in those symbol tables.  Both
275      are non-zero when the table exist, except when there is only a
276      dynsym table loaded through phdrs, then first_global is zero and
277      there will be no auxiliary table.  All symbols with local binding
278      come first in the symbol table, then all globals.  The zeroth,
279      null entry, in the auxiliary table is skipped if there is a main
280      table.  */
281   int first_global = INTUSE (dwfl_module_getsymtab_first_global) (state.mod);
282   if (first_global < 0)
283     return NULL;
284   search_table (&state, first_global == 0 ? 1 : first_global, syments);
285 
286   /* If we found nothing searching the global symbols, then try the locals.
287      Unless we have a global sizeless symbol that matches exactly.  */
288   if (state.closest_name == NULL && first_global > 1
289       && (state.sizeless_name == NULL || state.sizeless_value != state.addr))
290     search_table (&state, 1, first_global);
291 
292   /* If we found no proper sized symbol to use, fall back to the best
293      candidate sizeless symbol we found, if any.  */
294   if (state.closest_name == NULL
295       && state.sizeless_name != NULL
296       && state.sizeless_value >= state.min_label)
297     {
298       *state.closest_sym = state.sizeless_sym;
299       state.closest_value = state.sizeless_value;
300       state.closest_shndx = state.sizeless_shndx;
301       state.closest_elf = state.sizeless_elf;
302       state.closest_name = state.sizeless_name;
303     }
304 
305   *off = state.addr - state.closest_value;
306 
307   if (shndxp != NULL)
308     *shndxp = state.closest_shndx;
309   if (elfp != NULL)
310     *elfp = state.closest_elf;
311   if (biasp != NULL)
312     *biasp = dwfl_adjusted_st_value (state.mod, state.closest_elf, 0);
313   return state.closest_name;
314 }
315 
316 
317 const char *
dwfl_module_addrsym(Dwfl_Module * mod,GElf_Addr addr,GElf_Sym * closest_sym,GElf_Word * shndxp)318 dwfl_module_addrsym (Dwfl_Module *mod, GElf_Addr addr,
319 		     GElf_Sym *closest_sym, GElf_Word *shndxp)
320 {
321   GElf_Off off;
322   return __libdwfl_addrsym (mod, addr, &off, closest_sym, shndxp,
323 			    NULL, NULL, true);
324 }
INTDEF(dwfl_module_addrsym)325 INTDEF (dwfl_module_addrsym)
326 
327 const char
328 *dwfl_module_addrinfo (Dwfl_Module *mod, GElf_Addr address,
329 		       GElf_Off *offset, GElf_Sym *sym,
330 		       GElf_Word *shndxp, Elf **elfp, Dwarf_Addr *bias)
331 {
332   return __libdwfl_addrsym (mod, address, offset, sym, shndxp, elfp, bias,
333 			    false);
334 }
335 INTDEF (dwfl_module_addrinfo)
336