1 /* ELF strtab with GC and suffix merging support.
2    Copyright (C) 2001-2016 Free Software Foundation, Inc.
3    Written by Jakub Jelinek <jakub@redhat.com>.
4 
5    This file is part of BFD, the Binary File Descriptor library.
6 
7    This program is free software; you can redistribute it and/or modify
8    it under the terms of the GNU General Public License as published by
9    the Free Software Foundation; either version 3 of the License, or
10    (at your option) any later version.
11 
12    This program is distributed in the hope that it will be useful,
13    but WITHOUT ANY WARRANTY; without even the implied warranty of
14    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15    GNU General Public License for more details.
16 
17    You should have received a copy of the GNU General Public License
18    along with this program; if not, write to the Free Software
19    Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston,
20    MA 02110-1301, USA.  */
21 
22 #include "sysdep.h"
23 #include "bfd.h"
24 #include "libbfd.h"
25 #include "elf-bfd.h"
26 #include "hashtab.h"
27 #include "libiberty.h"
28 
29 /* An entry in the strtab hash table.  */
30 
31 struct elf_strtab_hash_entry
32 {
33   struct bfd_hash_entry root;
34   /* Length of this entry.  This includes the zero terminator.  */
35   int len;
36   unsigned int refcount;
37   union {
38     /* Index within the merged section.  */
39     bfd_size_type index;
40     /* Entry this is a suffix of (if len < 0).  */
41     struct elf_strtab_hash_entry *suffix;
42   } u;
43 };
44 
45 /* The strtab hash table.  */
46 
47 struct elf_strtab_hash
48 {
49   struct bfd_hash_table table;
50   /* Next available index.  */
51   size_t size;
52   /* Number of array entries alloced.  */
53   size_t alloced;
54   /* Final strtab size.  */
55   bfd_size_type sec_size;
56   /* Array of pointers to strtab entries.  */
57   struct elf_strtab_hash_entry **array;
58 };
59 
60 /* Routine to create an entry in a section merge hashtab.  */
61 
62 static struct bfd_hash_entry *
63 elf_strtab_hash_newfunc (struct bfd_hash_entry *entry,
64 			 struct bfd_hash_table *table,
65 			 const char *string)
66 {
67   /* Allocate the structure if it has not already been allocated by a
68      subclass.  */
69   if (entry == NULL)
70     entry = (struct bfd_hash_entry *)
71         bfd_hash_allocate (table, sizeof (struct elf_strtab_hash_entry));
72   if (entry == NULL)
73     return NULL;
74 
75   /* Call the allocation method of the superclass.  */
76   entry = bfd_hash_newfunc (entry, table, string);
77 
78   if (entry)
79     {
80       /* Initialize the local fields.  */
81       struct elf_strtab_hash_entry *ret;
82 
83       ret = (struct elf_strtab_hash_entry *) entry;
84       ret->u.index = -1;
85       ret->refcount = 0;
86       ret->len = 0;
87     }
88 
89   return entry;
90 }
91 
92 /* Create a new hash table.  */
93 
94 struct elf_strtab_hash *
95 _bfd_elf_strtab_init (void)
96 {
97   struct elf_strtab_hash *table;
98   bfd_size_type amt = sizeof (struct elf_strtab_hash);
99 
100   table = (struct elf_strtab_hash *) bfd_malloc (amt);
101   if (table == NULL)
102     return NULL;
103 
104   if (!bfd_hash_table_init (&table->table, elf_strtab_hash_newfunc,
105 			    sizeof (struct elf_strtab_hash_entry)))
106     {
107       free (table);
108       return NULL;
109     }
110 
111   table->sec_size = 0;
112   table->size = 1;
113   table->alloced = 64;
114   amt = sizeof (struct elf_strtab_hasn_entry *);
115   table->array = ((struct elf_strtab_hash_entry **)
116 		  bfd_malloc (table->alloced * amt));
117   if (table->array == NULL)
118     {
119       free (table);
120       return NULL;
121     }
122 
123   table->array[0] = NULL;
124 
125   return table;
126 }
127 
128 /* Free a strtab.  */
129 
130 void
131 _bfd_elf_strtab_free (struct elf_strtab_hash *tab)
132 {
133   bfd_hash_table_free (&tab->table);
134   free (tab->array);
135   free (tab);
136 }
137 
138 /* Get the index of an entity in a hash table, adding it if it is not
139    already present.  */
140 
141 size_t
142 _bfd_elf_strtab_add (struct elf_strtab_hash *tab,
143 		     const char *str,
144 		     bfd_boolean copy)
145 {
146   register struct elf_strtab_hash_entry *entry;
147 
148   /* We handle this specially, since we don't want to do refcounting
149      on it.  */
150   if (*str == '\0')
151     return 0;
152 
153   BFD_ASSERT (tab->sec_size == 0);
154   entry = (struct elf_strtab_hash_entry *)
155 	  bfd_hash_lookup (&tab->table, str, TRUE, copy);
156 
157   if (entry == NULL)
158     return (size_t) -1;
159 
160   entry->refcount++;
161   if (entry->len == 0)
162     {
163       entry->len = strlen (str) + 1;
164       /* 2G strings lose.  */
165       BFD_ASSERT (entry->len > 0);
166       if (tab->size == tab->alloced)
167 	{
168 	  bfd_size_type amt = sizeof (struct elf_strtab_hash_entry *);
169 	  tab->alloced *= 2;
170 	  tab->array = (struct elf_strtab_hash_entry **)
171               bfd_realloc_or_free (tab->array, tab->alloced * amt);
172 	  if (tab->array == NULL)
173 	    return (size_t) -1;
174 	}
175 
176       entry->u.index = tab->size++;
177       tab->array[entry->u.index] = entry;
178     }
179   return entry->u.index;
180 }
181 
182 void
183 _bfd_elf_strtab_addref (struct elf_strtab_hash *tab, size_t idx)
184 {
185   if (idx == 0 || idx == (size_t) -1)
186     return;
187   BFD_ASSERT (tab->sec_size == 0);
188   BFD_ASSERT (idx < tab->size);
189   ++tab->array[idx]->refcount;
190 }
191 
192 void
193 _bfd_elf_strtab_delref (struct elf_strtab_hash *tab, size_t idx)
194 {
195   if (idx == 0 || idx == (size_t) -1)
196     return;
197   BFD_ASSERT (tab->sec_size == 0);
198   BFD_ASSERT (idx < tab->size);
199   BFD_ASSERT (tab->array[idx]->refcount > 0);
200   --tab->array[idx]->refcount;
201 }
202 
203 unsigned int
204 _bfd_elf_strtab_refcount (struct elf_strtab_hash *tab, size_t idx)
205 {
206   return tab->array[idx]->refcount;
207 }
208 
209 void
210 _bfd_elf_strtab_clear_all_refs (struct elf_strtab_hash *tab)
211 {
212   size_t idx;
213 
214   for (idx = 1; idx < tab->size; idx++)
215     tab->array[idx]->refcount = 0;
216 }
217 
218 /* Save strtab refcounts prior to adding --as-needed library.  */
219 
220 struct strtab_save
221 {
222   size_t size;
223   unsigned int refcount[1];
224 };
225 
226 void *
227 _bfd_elf_strtab_save (struct elf_strtab_hash *tab)
228 {
229   struct strtab_save *save;
230   size_t idx, size;
231 
232   size = sizeof (*save) + (tab->size - 1) * sizeof (save->refcount[0]);
233   save = bfd_malloc (size);
234   if (save == NULL)
235     return save;
236 
237   save->size = tab->size;
238   for (idx = 1; idx < tab->size; idx++)
239     save->refcount[idx] = tab->array[idx]->refcount;
240   return save;
241 }
242 
243 /* Restore strtab refcounts on finding --as-needed library not needed.  */
244 
245 void
246 _bfd_elf_strtab_restore (struct elf_strtab_hash *tab, void *buf)
247 {
248   size_t idx, curr_size = tab->size;
249   struct strtab_save *save = (struct strtab_save *) buf;
250 
251   BFD_ASSERT (tab->sec_size == 0);
252   BFD_ASSERT (save->size <= curr_size);
253   tab->size = save->size;
254   for (idx = 1; idx < save->size; ++idx)
255     tab->array[idx]->refcount = save->refcount[idx];
256 
257   for (; idx < curr_size; ++idx)
258     {
259       /* We don't remove entries from the hash table, just set their
260 	 REFCOUNT to zero.  Setting LEN zero will result in the size
261 	 growing if the entry is added again.  See _bfd_elf_strtab_add.  */
262       tab->array[idx]->refcount = 0;
263       tab->array[idx]->len = 0;
264     }
265 }
266 
267 bfd_size_type
268 _bfd_elf_strtab_size (struct elf_strtab_hash *tab)
269 {
270   return tab->sec_size ? tab->sec_size : tab->size;
271 }
272 
273 bfd_size_type
274 _bfd_elf_strtab_offset (struct elf_strtab_hash *tab, size_t idx)
275 {
276   struct elf_strtab_hash_entry *entry;
277 
278   if (idx == 0)
279     return 0;
280   BFD_ASSERT (idx < tab->size);
281   BFD_ASSERT (tab->sec_size);
282   entry = tab->array[idx];
283   BFD_ASSERT (entry->refcount > 0);
284   entry->refcount--;
285   return tab->array[idx]->u.index;
286 }
287 
288 bfd_boolean
289 _bfd_elf_strtab_emit (register bfd *abfd, struct elf_strtab_hash *tab)
290 {
291   bfd_size_type off = 1;
292   size_t i;
293 
294   if (bfd_bwrite ("", 1, abfd) != 1)
295     return FALSE;
296 
297   for (i = 1; i < tab->size; ++i)
298     {
299       register const char *str;
300       register unsigned int len;
301 
302       BFD_ASSERT (tab->array[i]->refcount == 0);
303       len = tab->array[i]->len;
304       if ((int) len < 0)
305 	continue;
306 
307       str = tab->array[i]->root.string;
308       if (bfd_bwrite (str, len, abfd) != len)
309 	return FALSE;
310 
311       off += len;
312     }
313 
314   BFD_ASSERT (off == tab->sec_size);
315   return TRUE;
316 }
317 
318 /* Compare two elf_strtab_hash_entry structures.  Called via qsort.  */
319 
320 static int
321 strrevcmp (const void *a, const void *b)
322 {
323   struct elf_strtab_hash_entry *A = *(struct elf_strtab_hash_entry **) a;
324   struct elf_strtab_hash_entry *B = *(struct elf_strtab_hash_entry **) b;
325   unsigned int lenA = A->len;
326   unsigned int lenB = B->len;
327   const unsigned char *s = (const unsigned char *) A->root.string + lenA - 1;
328   const unsigned char *t = (const unsigned char *) B->root.string + lenB - 1;
329   int l = lenA < lenB ? lenA : lenB;
330 
331   while (l)
332     {
333       if (*s != *t)
334 	return (int) *s - (int) *t;
335       s--;
336       t--;
337       l--;
338     }
339   return lenA - lenB;
340 }
341 
342 static inline int
343 is_suffix (const struct elf_strtab_hash_entry *A,
344 	   const struct elf_strtab_hash_entry *B)
345 {
346   if (A->len <= B->len)
347     /* B cannot be a suffix of A unless A is equal to B, which is guaranteed
348        not to be equal by the hash table.  */
349     return 0;
350 
351   return memcmp (A->root.string + (A->len - B->len),
352 		 B->root.string, B->len - 1) == 0;
353 }
354 
355 /* This function assigns final string table offsets for used strings,
356    merging strings matching suffixes of longer strings if possible.  */
357 
358 void
359 _bfd_elf_strtab_finalize (struct elf_strtab_hash *tab)
360 {
361   struct elf_strtab_hash_entry **array, **a, *e;
362   bfd_size_type amt, sec_size;
363   size_t size, i;
364 
365   /* Sort the strings by suffix and length.  */
366   amt = tab->size;
367   amt *= sizeof (struct elf_strtab_hash_entry *);
368   array = (struct elf_strtab_hash_entry **) bfd_malloc (amt);
369   if (array == NULL)
370     goto alloc_failure;
371 
372   for (i = 1, a = array; i < tab->size; ++i)
373     {
374       e = tab->array[i];
375       if (e->refcount)
376 	{
377 	  *a++ = e;
378 	  /* Adjust the length to not include the zero terminator.  */
379 	  e->len -= 1;
380 	}
381       else
382 	e->len = 0;
383     }
384 
385   size = a - array;
386   if (size != 0)
387     {
388       qsort (array, size, sizeof (struct elf_strtab_hash_entry *), strrevcmp);
389 
390       /* Loop over the sorted array and merge suffixes.  Start from the
391 	 end because we want eg.
392 
393 	 s1 -> "d"
394 	 s2 -> "bcd"
395 	 s3 -> "abcd"
396 
397 	 to end up as
398 
399 	 s3 -> "abcd"
400 	 s2 _____^
401 	 s1 _______^
402 
403 	 ie. we don't want s1 pointing into the old s2.  */
404       e = *--a;
405       e->len += 1;
406       while (--a >= array)
407 	{
408 	  struct elf_strtab_hash_entry *cmp = *a;
409 
410 	  cmp->len += 1;
411 	  if (is_suffix (e, cmp))
412 	    {
413 	      cmp->u.suffix = e;
414 	      cmp->len = -cmp->len;
415 	    }
416 	  else
417 	    e = cmp;
418 	}
419     }
420 
421 alloc_failure:
422   if (array)
423     free (array);
424 
425   /* Assign positions to the strings we want to keep.  */
426   sec_size = 1;
427   for (i = 1; i < tab->size; ++i)
428     {
429       e = tab->array[i];
430       if (e->refcount && e->len > 0)
431 	{
432 	  e->u.index = sec_size;
433 	  sec_size += e->len;
434 	}
435     }
436 
437   tab->sec_size = sec_size;
438 
439   /* Adjust the rest.  */
440   for (i = 1; i < tab->size; ++i)
441     {
442       e = tab->array[i];
443       if (e->refcount && e->len < 0)
444 	e->u.index = e->u.suffix->u.index + (e->u.suffix->len + e->len);
445     }
446 }
447