1 /*
2  * Copyright (C) 2008, 2009 The Android Open Source Project
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  *  * Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  *  * Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in
12  *    the documentation and/or other materials provided with the
13  *    distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19  * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22  * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <dlfcn.h>
30 #include <errno.h>
31 #include <fcntl.h>
32 #include <inttypes.h>
33 #include <pthread.h>
34 #include <stdio.h>
35 #include <stdlib.h>
36 #include <string.h>
37 #include <sys/mman.h>
38 #include <unistd.h>
39 
40 #include <new>
41 
42 // Private C library headers.
43 #include "private/bionic_tls.h"
44 #include "private/KernelArgumentBlock.h"
45 #include "private/ScopedPthreadMutexLocker.h"
46 #include "private/ScopedFd.h"
47 #include "private/ScopeGuard.h"
48 #include "private/UniquePtr.h"
49 
50 #include "linker.h"
51 #include "linker_debug.h"
52 #include "linker_environ.h"
53 #include "linker_phdr.h"
54 #include "linker_allocator.h"
55 
56 /* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
57  *
58  * Do NOT use malloc() and friends or pthread_*() code here.
59  * Don't use printf() either; it's caused mysterious memory
60  * corruption in the past.
61  * The linker runs before we bring up libc and it's easiest
62  * to make sure it does not depend on any complex libc features
63  *
64  * open issues / todo:
65  *
66  * - cleaner error reporting
67  * - after linking, set as much stuff as possible to READONLY
68  *   and NOEXEC
69  */
70 
71 #if defined(__LP64__)
72 #define SEARCH_NAME(x) x
73 #else
74 // Nvidia drivers are relying on the bug:
75 // http://code.google.com/p/android/issues/detail?id=6670
76 // so we continue to use base-name lookup for lp32
get_base_name(const char * name)77 static const char* get_base_name(const char* name) {
78   const char* bname = strrchr(name, '/');
79   return bname ? bname + 1 : name;
80 }
81 #define SEARCH_NAME(x) get_base_name(x)
82 #endif
83 
84 static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
85 
86 static LinkerAllocator<soinfo> g_soinfo_allocator;
87 static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
88 
89 static soinfo* solist;
90 static soinfo* sonext;
91 static soinfo* somain; // main process, always the one after libdl_info
92 
93 static const char* const kDefaultLdPaths[] = {
94 #if defined(__LP64__)
95   "/vendor/lib64",
96   "/system/lib64",
97 #else
98   "/vendor/lib",
99   "/system/lib",
100 #endif
101   nullptr
102 };
103 
104 #define LDPATH_BUFSIZE (LDPATH_MAX*64)
105 #define LDPATH_MAX 8
106 
107 #define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
108 #define LDPRELOAD_MAX 8
109 
110 static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
111 static const char* g_ld_library_paths[LDPATH_MAX + 1];
112 
113 static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
114 static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
115 
116 static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
117 
118 __LIBC_HIDDEN__ int g_ld_debug_verbosity;
119 
120 __LIBC_HIDDEN__ abort_msg_t* g_abort_message = nullptr; // For debuggerd.
121 
122 enum RelocationKind {
123   kRelocAbsolute = 0,
124   kRelocRelative,
125   kRelocCopy,
126   kRelocSymbol,
127   kRelocMax
128 };
129 
130 #if STATS
131 struct linker_stats_t {
132   int count[kRelocMax];
133 };
134 
135 static linker_stats_t linker_stats;
136 
count_relocation(RelocationKind kind)137 static void count_relocation(RelocationKind kind) {
138   ++linker_stats.count[kind];
139 }
140 #else
count_relocation(RelocationKind)141 static void count_relocation(RelocationKind) {
142 }
143 #endif
144 
145 #if COUNT_PAGES
146 static unsigned bitmask[4096];
147 #if defined(__LP64__)
148 #define MARK(offset) \
149     do { \
150       if ((((offset) >> 12) >> 5) < 4096) \
151           bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
152     } while (0)
153 #else
154 #define MARK(offset) \
155     do { \
156       bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
157     } while (0)
158 #endif
159 #else
160 #define MARK(x) do {} while (0)
161 #endif
162 
163 // You shouldn't try to call memory-allocating functions in the dynamic linker.
164 // Guard against the most obvious ones.
165 #define DISALLOW_ALLOCATION(return_type, name, ...) \
166     return_type name __VA_ARGS__ \
167     { \
168       __libc_fatal("ERROR: " #name " called from the dynamic linker!\n"); \
169     }
170 DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
171 DISALLOW_ALLOCATION(void, free, (void* u __unused));
172 DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
173 DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
174 
175 static char __linker_dl_err_buf[768];
176 
linker_get_error_buffer()177 char* linker_get_error_buffer() {
178   return &__linker_dl_err_buf[0];
179 }
180 
linker_get_error_buffer_size()181 size_t linker_get_error_buffer_size() {
182   return sizeof(__linker_dl_err_buf);
183 }
184 
185 // This function is an empty stub where GDB locates a breakpoint to get notified
186 // about linker activity.
187 extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
188 
189 static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
190 static r_debug _r_debug = {1, nullptr, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
191 static link_map* r_debug_tail = 0;
192 
insert_soinfo_into_debug_map(soinfo * info)193 static void insert_soinfo_into_debug_map(soinfo* info) {
194   // Copy the necessary fields into the debug structure.
195   link_map* map = &(info->link_map_head);
196   map->l_addr = info->load_bias;
197   map->l_name = reinterpret_cast<char*>(info->name);
198   map->l_ld = info->dynamic;
199 
200   // Stick the new library at the end of the list.
201   // gdb tends to care more about libc than it does
202   // about leaf libraries, and ordering it this way
203   // reduces the back-and-forth over the wire.
204   if (r_debug_tail) {
205     r_debug_tail->l_next = map;
206     map->l_prev = r_debug_tail;
207     map->l_next = 0;
208   } else {
209     _r_debug.r_map = map;
210     map->l_prev = 0;
211     map->l_next = 0;
212   }
213   r_debug_tail = map;
214 }
215 
remove_soinfo_from_debug_map(soinfo * info)216 static void remove_soinfo_from_debug_map(soinfo* info) {
217   link_map* map = &(info->link_map_head);
218 
219   if (r_debug_tail == map) {
220     r_debug_tail = map->l_prev;
221   }
222 
223   if (map->l_prev) {
224     map->l_prev->l_next = map->l_next;
225   }
226   if (map->l_next) {
227     map->l_next->l_prev = map->l_prev;
228   }
229 }
230 
notify_gdb_of_load(soinfo * info)231 static void notify_gdb_of_load(soinfo* info) {
232   if (info->flags & FLAG_EXE) {
233     // GDB already knows about the main executable
234     return;
235   }
236 
237   ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
238 
239   _r_debug.r_state = r_debug::RT_ADD;
240   rtld_db_dlactivity();
241 
242   insert_soinfo_into_debug_map(info);
243 
244   _r_debug.r_state = r_debug::RT_CONSISTENT;
245   rtld_db_dlactivity();
246 }
247 
notify_gdb_of_unload(soinfo * info)248 static void notify_gdb_of_unload(soinfo* info) {
249   if (info->flags & FLAG_EXE) {
250     // GDB already knows about the main executable
251     return;
252   }
253 
254   ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
255 
256   _r_debug.r_state = r_debug::RT_DELETE;
257   rtld_db_dlactivity();
258 
259   remove_soinfo_from_debug_map(info);
260 
261   _r_debug.r_state = r_debug::RT_CONSISTENT;
262   rtld_db_dlactivity();
263 }
264 
notify_gdb_of_libraries()265 void notify_gdb_of_libraries() {
266   _r_debug.r_state = r_debug::RT_ADD;
267   rtld_db_dlactivity();
268   _r_debug.r_state = r_debug::RT_CONSISTENT;
269   rtld_db_dlactivity();
270 }
271 
alloc()272 LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
273   return g_soinfo_links_allocator.alloc();
274 }
275 
free(LinkedListEntry<soinfo> * entry)276 void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
277   g_soinfo_links_allocator.free(entry);
278 }
279 
protect_data(int protection)280 static void protect_data(int protection) {
281   g_soinfo_allocator.protect_all(protection);
282   g_soinfo_links_allocator.protect_all(protection);
283 }
284 
soinfo_alloc(const char * name,struct stat * file_stat,off64_t file_offset)285 static soinfo* soinfo_alloc(const char* name, struct stat* file_stat, off64_t file_offset) {
286   if (strlen(name) >= SOINFO_NAME_LEN) {
287     DL_ERR("library name \"%s\" too long", name);
288     return nullptr;
289   }
290 
291   soinfo* si = new (g_soinfo_allocator.alloc()) soinfo(name, file_stat, file_offset);
292 
293   sonext->next = si;
294   sonext = si;
295 
296   TRACE("name %s: allocated soinfo @ %p", name, si);
297   return si;
298 }
299 
soinfo_free(soinfo * si)300 static void soinfo_free(soinfo* si) {
301   if (si == nullptr) {
302     return;
303   }
304 
305   if (si->base != 0 && si->size != 0) {
306     munmap(reinterpret_cast<void*>(si->base), si->size);
307   }
308 
309   soinfo *prev = nullptr, *trav;
310 
311   TRACE("name %s: freeing soinfo @ %p", si->name, si);
312 
313   for (trav = solist; trav != nullptr; trav = trav->next) {
314     if (trav == si) {
315       break;
316     }
317     prev = trav;
318   }
319   if (trav == nullptr) {
320     // si was not in solist
321     DL_ERR("name \"%s\" is not in solist!", si->name);
322     return;
323   }
324 
325   // clear links to/from si
326   si->remove_all_links();
327 
328   // prev will never be null, because the first entry in solist is
329   // always the static libdl_info.
330   prev->next = si->next;
331   if (si == sonext) {
332     sonext = prev;
333   }
334 
335   g_soinfo_allocator.free(si);
336 }
337 
338 
parse_path(const char * path,const char * delimiters,const char ** array,char * buf,size_t buf_size,size_t max_count)339 static void parse_path(const char* path, const char* delimiters,
340                        const char** array, char* buf, size_t buf_size, size_t max_count) {
341   if (path == nullptr) {
342     return;
343   }
344 
345   size_t len = strlcpy(buf, path, buf_size);
346 
347   size_t i = 0;
348   char* buf_p = buf;
349   while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
350     if (*array[i] != '\0') {
351       ++i;
352     }
353   }
354 
355   // Forget the last path if we had to truncate; this occurs if the 2nd to
356   // last char isn't '\0' (i.e. wasn't originally a delimiter).
357   if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
358     array[i - 1] = nullptr;
359   } else {
360     array[i] = nullptr;
361   }
362 }
363 
parse_LD_LIBRARY_PATH(const char * path)364 static void parse_LD_LIBRARY_PATH(const char* path) {
365   parse_path(path, ":", g_ld_library_paths,
366              g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
367 }
368 
parse_LD_PRELOAD(const char * path)369 static void parse_LD_PRELOAD(const char* path) {
370   // We have historically supported ':' as well as ' ' in LD_PRELOAD.
371   parse_path(path, " :", g_ld_preload_names,
372              g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
373 }
374 
375 #if defined(__arm__)
376 
377 // For a given PC, find the .so that it belongs to.
378 // Returns the base address of the .ARM.exidx section
379 // for that .so, and the number of 8-byte entries
380 // in that section (via *pcount).
381 //
382 // Intended to be called by libc's __gnu_Unwind_Find_exidx().
383 //
384 // This function is exposed via dlfcn.cpp and libdl.so.
dl_unwind_find_exidx(_Unwind_Ptr pc,int * pcount)385 _Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
386   unsigned addr = (unsigned)pc;
387 
388   for (soinfo* si = solist; si != 0; si = si->next) {
389     if ((addr >= si->base) && (addr < (si->base + si->size))) {
390         *pcount = si->ARM_exidx_count;
391         return (_Unwind_Ptr)si->ARM_exidx;
392     }
393   }
394   *pcount = 0;
395   return nullptr;
396 }
397 
398 #endif
399 
400 // Here, we only have to provide a callback to iterate across all the
401 // loaded libraries. gcc_eh does the rest.
dl_iterate_phdr(int (* cb)(dl_phdr_info * info,size_t size,void * data),void * data)402 int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
403   int rv = 0;
404   for (soinfo* si = solist; si != nullptr; si = si->next) {
405     dl_phdr_info dl_info;
406     dl_info.dlpi_addr = si->link_map_head.l_addr;
407     dl_info.dlpi_name = si->link_map_head.l_name;
408     dl_info.dlpi_phdr = si->phdr;
409     dl_info.dlpi_phnum = si->phnum;
410     rv = cb(&dl_info, sizeof(dl_phdr_info), data);
411     if (rv != 0) {
412       break;
413     }
414   }
415   return rv;
416 }
417 
ElfW(Sym)418 static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
419   ElfW(Sym)* symtab = si->symtab;
420 
421   TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
422              name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
423 
424   for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
425     ElfW(Sym)* s = symtab + n;
426     if (strcmp(si->get_string(s->st_name), name)) continue;
427 
428     // only concern ourselves with global and weak symbol definitions
429     switch (ELF_ST_BIND(s->st_info)) {
430       case STB_GLOBAL:
431       case STB_WEAK:
432         if (s->st_shndx == SHN_UNDEF) {
433           continue;
434         }
435 
436         TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
437                  name, si->name, reinterpret_cast<void*>(s->st_value),
438                  static_cast<size_t>(s->st_size));
439         return s;
440       case STB_LOCAL:
441         continue;
442       default:
443         __libc_fatal("ERROR: Unexpected ST_BIND value: %d for '%s' in '%s'",
444             ELF_ST_BIND(s->st_info), name, si->name);
445     }
446   }
447 
448   TRACE_TYPE(LOOKUP, "NOT FOUND %s in %s@%p %x %zd",
449              name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
450 
451 
452   return nullptr;
453 }
454 
soinfo(const char * name,const struct stat * file_stat,off64_t file_offset)455 soinfo::soinfo(const char* name, const struct stat* file_stat, off64_t file_offset) {
456   memset(this, 0, sizeof(*this));
457 
458   strlcpy(this->name, name, sizeof(this->name));
459   flags = FLAG_NEW_SOINFO;
460   version = SOINFO_VERSION;
461 
462   if (file_stat != nullptr) {
463     this->st_dev = file_stat->st_dev;
464     this->st_ino = file_stat->st_ino;
465     this->file_offset = file_offset;
466   }
467 }
468 
elfhash(const char * _name)469 static unsigned elfhash(const char* _name) {
470   const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
471   unsigned h = 0, g;
472 
473   while (*name) {
474     h = (h << 4) + *name++;
475     g = h & 0xf0000000;
476     h ^= g;
477     h ^= g >> 24;
478   }
479   return h;
480 }
481 
ElfW(Sym)482 static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi) {
483   unsigned elf_hash = elfhash(name);
484   ElfW(Sym)* s = nullptr;
485 
486   /* "This element's presence in a shared object library alters the dynamic linker's
487    * symbol resolution algorithm for references within the library. Instead of starting
488    * a symbol search with the executable file, the dynamic linker starts from the shared
489    * object itself. If the shared object fails to supply the referenced symbol, the
490    * dynamic linker then searches the executable file and other shared objects as usual."
491    *
492    * http://www.sco.com/developers/gabi/2012-12-31/ch5.dynamic.html
493    *
494    * Note that this is unlikely since static linker avoids generating
495    * relocations for -Bsymbolic linked dynamic executables.
496    */
497   if (si->has_DT_SYMBOLIC) {
498     DEBUG("%s: looking up %s in local scope (DT_SYMBOLIC)", si->name, name);
499     s = soinfo_elf_lookup(si, elf_hash, name);
500     if (s != nullptr) {
501       *lsi = si;
502     }
503   }
504 
505   if (s == nullptr && somain != nullptr) {
506     // 1. Look for it in the main executable unless we already did.
507     if (si != somain || !si->has_DT_SYMBOLIC) {
508       DEBUG("%s: looking up %s in executable %s",
509             si->name, name, somain->name);
510       s = soinfo_elf_lookup(somain, elf_hash, name);
511       if (s != nullptr) {
512         *lsi = somain;
513       }
514     }
515 
516     // 2. Look for it in the ld_preloads
517     if (s == nullptr) {
518       for (int i = 0; g_ld_preloads[i] != NULL; i++) {
519         s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
520         if (s != nullptr) {
521           *lsi = g_ld_preloads[i];
522           break;
523         }
524       }
525     }
526   }
527 
528   /* Look for symbols in the local scope (the object who is
529    * searching). This happens with C++ templates on x86 for some
530    * reason.
531    *
532    * Notes on weak symbols:
533    * The ELF specs are ambiguous about treatment of weak definitions in
534    * dynamic linking.  Some systems return the first definition found
535    * and some the first non-weak definition.   This is system dependent.
536    * Here we return the first definition found for simplicity.  */
537 
538   if (s == nullptr && !si->has_DT_SYMBOLIC) {
539     DEBUG("%s: looking up %s in local scope", si->name, name);
540     s = soinfo_elf_lookup(si, elf_hash, name);
541     if (s != nullptr) {
542       *lsi = si;
543     }
544   }
545 
546   if (s == nullptr) {
547     si->get_children().visit([&](soinfo* child) {
548       DEBUG("%s: looking up %s in %s", si->name, name, child->name);
549       s = soinfo_elf_lookup(child, elf_hash, name);
550       if (s != nullptr) {
551         *lsi = child;
552         return false;
553       }
554       return true;
555     });
556   }
557 
558   if (s != nullptr) {
559     TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
560                "found in %s, base = %p, load bias = %p",
561                si->name, name, reinterpret_cast<void*>(s->st_value),
562                (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
563                reinterpret_cast<void*>((*lsi)->load_bias));
564   }
565 
566   return s;
567 }
568 
569 // Each size has it's own allocator.
570 template<size_t size>
571 class SizeBasedAllocator {
572  public:
alloc()573   static void* alloc() {
574     return allocator_.alloc();
575   }
576 
free(void * ptr)577   static void free(void* ptr) {
578     allocator_.free(ptr);
579   }
580 
581  private:
582   static LinkerBlockAllocator allocator_;
583 };
584 
585 template<size_t size>
586 LinkerBlockAllocator SizeBasedAllocator<size>::allocator_(size);
587 
588 template<typename T>
589 class TypeBasedAllocator {
590  public:
alloc()591   static T* alloc() {
592     return reinterpret_cast<T*>(SizeBasedAllocator<sizeof(T)>::alloc());
593   }
594 
free(T * ptr)595   static void free(T* ptr) {
596     SizeBasedAllocator<sizeof(T)>::free(ptr);
597   }
598 };
599 
600 class LoadTask {
601  public:
602   struct deleter_t {
operator ()LoadTask::deleter_t603     void operator()(LoadTask* t) {
604       TypeBasedAllocator<LoadTask>::free(t);
605     }
606   };
607 
608   typedef UniquePtr<LoadTask, deleter_t> unique_ptr;
609 
610   static deleter_t deleter;
611 
create(const char * name,soinfo * needed_by)612   static LoadTask* create(const char* name, soinfo* needed_by) {
613     LoadTask* ptr = TypeBasedAllocator<LoadTask>::alloc();
614     return new (ptr) LoadTask(name, needed_by);
615   }
616 
get_name() const617   const char* get_name() const {
618     return name_;
619   }
620 
get_needed_by() const621   soinfo* get_needed_by() const {
622     return needed_by_;
623   }
624  private:
LoadTask(const char * name,soinfo * needed_by)625   LoadTask(const char* name, soinfo* needed_by)
626     : name_(name), needed_by_(needed_by) {}
627 
628   const char* name_;
629   soinfo* needed_by_;
630 
631   DISALLOW_IMPLICIT_CONSTRUCTORS(LoadTask);
632 };
633 
634 LoadTask::deleter_t LoadTask::deleter;
635 
636 template <typename T>
637 using linked_list_t = LinkedList<T, TypeBasedAllocator<LinkedListEntry<T>>>;
638 
639 typedef linked_list_t<soinfo> SoinfoLinkedList;
640 typedef linked_list_t<const char> StringLinkedList;
641 typedef linked_list_t<LoadTask> LoadTaskList;
642 
643 
644 // This is used by dlsym(3).  It performs symbol lookup only within the
645 // specified soinfo object and its dependencies in breadth first order.
ElfW(Sym)646 ElfW(Sym)* dlsym_handle_lookup(soinfo* si, soinfo** found, const char* name) {
647   SoinfoLinkedList visit_list;
648   SoinfoLinkedList visited;
649 
650   visit_list.push_back(si);
651   soinfo* current_soinfo;
652   while ((current_soinfo = visit_list.pop_front()) != nullptr) {
653     if (visited.contains(current_soinfo)) {
654       continue;
655     }
656 
657     ElfW(Sym)* result = soinfo_elf_lookup(current_soinfo, elfhash(name), name);
658 
659     if (result != nullptr) {
660       *found = current_soinfo;
661       return result;
662     }
663     visited.push_back(current_soinfo);
664 
665     current_soinfo->get_children().for_each([&](soinfo* child) {
666       visit_list.push_back(child);
667     });
668   }
669 
670   return nullptr;
671 }
672 
673 /* This is used by dlsym(3) to performs a global symbol lookup. If the
674    start value is null (for RTLD_DEFAULT), the search starts at the
675    beginning of the global solist. Otherwise the search starts at the
676    specified soinfo (for RTLD_NEXT).
677  */
ElfW(Sym)678 ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
679   unsigned elf_hash = elfhash(name);
680 
681   if (start == nullptr) {
682     start = solist;
683   }
684 
685   ElfW(Sym)* s = nullptr;
686   for (soinfo* si = start; (s == nullptr) && (si != nullptr); si = si->next) {
687     s = soinfo_elf_lookup(si, elf_hash, name);
688     if (s != nullptr) {
689       *found = si;
690       break;
691     }
692   }
693 
694   if (s != nullptr) {
695     TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
696                name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
697   }
698 
699   return s;
700 }
701 
find_containing_library(const void * p)702 soinfo* find_containing_library(const void* p) {
703   ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
704   for (soinfo* si = solist; si != nullptr; si = si->next) {
705     if (address >= si->base && address - si->base < si->size) {
706       return si;
707     }
708   }
709   return nullptr;
710 }
711 
ElfW(Sym)712 ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
713   ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
714 
715   // Search the library's symbol table for any defined symbol which
716   // contains this address.
717   for (size_t i = 0; i < si->nchain; ++i) {
718     ElfW(Sym)* sym = &si->symtab[i];
719     if (sym->st_shndx != SHN_UNDEF &&
720         soaddr >= sym->st_value &&
721         soaddr < sym->st_value + sym->st_size) {
722       return sym;
723     }
724   }
725 
726   return nullptr;
727 }
728 
open_library_on_path(const char * name,const char * const paths[])729 static int open_library_on_path(const char* name, const char* const paths[]) {
730   char buf[512];
731   for (size_t i = 0; paths[i] != nullptr; ++i) {
732     int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
733     if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
734       PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
735       continue;
736     }
737     int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
738     if (fd != -1) {
739       return fd;
740     }
741   }
742   return -1;
743 }
744 
open_library(const char * name)745 static int open_library(const char* name) {
746   TRACE("[ opening %s ]", name);
747 
748   // If the name contains a slash, we should attempt to open it directly and not search the paths.
749   if (strchr(name, '/') != nullptr) {
750     int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
751     if (fd != -1) {
752       return fd;
753     }
754     // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
755 #if defined(__LP64__)
756     return -1;
757 #endif
758   }
759 
760   // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
761   int fd = open_library_on_path(name, g_ld_library_paths);
762   if (fd == -1) {
763     fd = open_library_on_path(name, kDefaultLdPaths);
764   }
765   return fd;
766 }
767 
768 template<typename F>
for_each_dt_needed(const soinfo * si,F action)769 static void for_each_dt_needed(const soinfo* si, F action) {
770   for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
771     if (d->d_tag == DT_NEEDED) {
772       action(si->get_string(d->d_un.d_val));
773     }
774   }
775 }
776 
load_library(LoadTaskList & load_tasks,const char * name,int dlflags,const android_dlextinfo * extinfo)777 static soinfo* load_library(LoadTaskList& load_tasks, const char* name, int dlflags, const android_dlextinfo* extinfo) {
778   int fd = -1;
779   off64_t file_offset = 0;
780   ScopedFd file_guard(-1);
781 
782   if (extinfo != nullptr && (extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD) != 0) {
783     fd = extinfo->library_fd;
784     if ((extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD_OFFSET) != 0) {
785       file_offset = extinfo->library_fd_offset;
786     }
787   } else {
788     // Open the file.
789     fd = open_library(name);
790     if (fd == -1) {
791       DL_ERR("library \"%s\" not found", name);
792       return nullptr;
793     }
794 
795     file_guard.reset(fd);
796   }
797 
798   if ((file_offset % PAGE_SIZE) != 0) {
799     DL_ERR("file offset for the library \"%s\" is not page-aligned: %" PRId64, name, file_offset);
800     return nullptr;
801   }
802 
803   struct stat file_stat;
804   if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
805     DL_ERR("unable to stat file for the library \"%s\": %s", name, strerror(errno));
806     return nullptr;
807   }
808 
809   // Check for symlink and other situations where
810   // file can have different names.
811   for (soinfo* si = solist; si != nullptr; si = si->next) {
812     if (si->get_st_dev() != 0 &&
813         si->get_st_ino() != 0 &&
814         si->get_st_dev() == file_stat.st_dev &&
815         si->get_st_ino() == file_stat.st_ino &&
816         si->get_file_offset() == file_offset) {
817       TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
818       return si;
819     }
820   }
821 
822   if ((dlflags & RTLD_NOLOAD) != 0) {
823     DL_ERR("library \"%s\" wasn't loaded and RTLD_NOLOAD prevented it", name);
824     return nullptr;
825   }
826 
827   // Read the ELF header and load the segments.
828   ElfReader elf_reader(name, fd, file_offset);
829   if (!elf_reader.Load(extinfo)) {
830     return nullptr;
831   }
832 
833   soinfo* si = soinfo_alloc(SEARCH_NAME(name), &file_stat, file_offset);
834   if (si == nullptr) {
835     return nullptr;
836   }
837   si->base = elf_reader.load_start();
838   si->size = elf_reader.load_size();
839   si->load_bias = elf_reader.load_bias();
840   si->phnum = elf_reader.phdr_count();
841   si->phdr = elf_reader.loaded_phdr();
842 
843   if (!si->PrelinkImage()) {
844     soinfo_free(si);
845     return nullptr;
846   }
847 
848   for_each_dt_needed(si, [&] (const char* name) {
849     load_tasks.push_back(LoadTask::create(name, si));
850   });
851 
852   return si;
853 }
854 
find_loaded_library_by_name(const char * name)855 static soinfo *find_loaded_library_by_name(const char* name) {
856   const char* search_name = SEARCH_NAME(name);
857   for (soinfo* si = solist; si != nullptr; si = si->next) {
858     if (!strcmp(search_name, si->name)) {
859       return si;
860     }
861   }
862   return nullptr;
863 }
864 
find_library_internal(LoadTaskList & load_tasks,const char * name,int dlflags,const android_dlextinfo * extinfo)865 static soinfo* find_library_internal(LoadTaskList& load_tasks, const char* name, int dlflags, const android_dlextinfo* extinfo) {
866 
867   soinfo* si = find_loaded_library_by_name(name);
868 
869   // Library might still be loaded, the accurate detection
870   // of this fact is done by load_library.
871   if (si == nullptr) {
872     TRACE("[ '%s' has not been found by name.  Trying harder...]", name);
873     si = load_library(load_tasks, name, dlflags, extinfo);
874   }
875 
876   return si;
877 }
878 
879 static void soinfo_unload(soinfo* si);
880 
is_recursive(soinfo * si,soinfo * parent)881 static bool is_recursive(soinfo* si, soinfo* parent) {
882   if (parent == nullptr) {
883     return false;
884   }
885 
886   if (si == parent) {
887     DL_ERR("recursive link to \"%s\"", si->name);
888     return true;
889   }
890 
891   return !parent->get_parents().visit([&](soinfo* grandparent) {
892     return !is_recursive(si, grandparent);
893   });
894 }
895 
find_libraries(const char * const library_names[],size_t library_names_size,soinfo * soinfos[],soinfo * ld_preloads[],size_t ld_preloads_size,int dlflags,const android_dlextinfo * extinfo)896 static bool find_libraries(const char* const library_names[], size_t library_names_size, soinfo* soinfos[],
897     soinfo* ld_preloads[], size_t ld_preloads_size, int dlflags, const android_dlextinfo* extinfo) {
898   // Step 0: prepare.
899   LoadTaskList load_tasks;
900   for (size_t i = 0; i < library_names_size; ++i) {
901     const char* name = library_names[i];
902     load_tasks.push_back(LoadTask::create(name, nullptr));
903   }
904 
905   // Libraries added to this list in reverse order so that we can
906   // start linking from bottom-up - see step 2.
907   SoinfoLinkedList found_libs;
908   size_t soinfos_size = 0;
909 
910   auto failure_guard = make_scope_guard([&]() {
911     // Housekeeping
912     load_tasks.for_each([] (LoadTask* t) {
913       LoadTask::deleter(t);
914     });
915 
916     for (size_t i = 0; i<soinfos_size; ++i) {
917       soinfo_unload(soinfos[i]);
918     }
919   });
920 
921   // Step 1: load and pre-link all DT_NEEDED libraries in breadth first order.
922   for (LoadTask::unique_ptr task(load_tasks.pop_front()); task.get() != nullptr; task.reset(load_tasks.pop_front())) {
923     soinfo* si = find_library_internal(load_tasks, task->get_name(), dlflags, extinfo);
924     if (si == nullptr) {
925       return false;
926     }
927 
928     soinfo* needed_by = task->get_needed_by();
929 
930     if (is_recursive(si, needed_by)) {
931       return false;
932     }
933 
934     si->ref_count++;
935     if (needed_by != nullptr) {
936       needed_by->add_child(si);
937     }
938     found_libs.push_front(si);
939 
940     // When ld_preloads is not null first
941     // ld_preloads_size libs are in fact ld_preloads.
942     if (ld_preloads != nullptr && soinfos_size < ld_preloads_size) {
943       ld_preloads[soinfos_size] = si;
944     }
945 
946     if (soinfos_size<library_names_size) {
947       soinfos[soinfos_size++] = si;
948     }
949   }
950 
951   // Step 2: link libraries.
952   soinfo* si;
953   while ((si = found_libs.pop_front()) != nullptr) {
954     if ((si->flags & FLAG_LINKED) == 0) {
955       if (!si->LinkImage(extinfo)) {
956         return false;
957       }
958       si->flags |= FLAG_LINKED;
959     }
960   }
961 
962   // All is well - found_libs and load_tasks are empty at this point
963   // and all libs are successfully linked.
964   failure_guard.disable();
965   return true;
966 }
967 
find_library(const char * name,int dlflags,const android_dlextinfo * extinfo)968 static soinfo* find_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
969   if (name == nullptr) {
970     somain->ref_count++;
971     return somain;
972   }
973 
974   soinfo* si;
975 
976   if (!find_libraries(&name, 1, &si, nullptr, 0, dlflags, extinfo)) {
977     return nullptr;
978   }
979 
980   return si;
981 }
982 
soinfo_unload(soinfo * si)983 static void soinfo_unload(soinfo* si) {
984   if (si->ref_count == 1) {
985     TRACE("unloading '%s'", si->name);
986     si->CallDestructors();
987 
988     if (si->has_min_version(0)) {
989       soinfo* child = nullptr;
990       while ((child = si->get_children().pop_front()) != nullptr) {
991         TRACE("%s needs to unload %s", si->name, child->name);
992         soinfo_unload(child);
993       }
994     } else {
995       for_each_dt_needed(si, [&] (const char* library_name) {
996         TRACE("deprecated (old format of soinfo): %s needs to unload %s", si->name, library_name);
997         soinfo* needed = find_library(library_name, RTLD_NOLOAD, nullptr);
998         if (needed != nullptr) {
999           soinfo_unload(needed);
1000         } else {
1001           // Not found: for example if symlink was deleted between dlopen and dlclose
1002           // Since we cannot really handle errors at this point - print and continue.
1003           PRINT("warning: couldn't find %s needed by %s on unload.", library_name, si->name);
1004         }
1005       });
1006     }
1007 
1008     notify_gdb_of_unload(si);
1009     si->ref_count = 0;
1010     soinfo_free(si);
1011   } else {
1012     si->ref_count--;
1013     TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
1014   }
1015 }
1016 
do_android_get_LD_LIBRARY_PATH(char * buffer,size_t buffer_size)1017 void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
1018   // Use basic string manipulation calls to avoid snprintf.
1019   // snprintf indirectly calls pthread_getspecific to get the size of a buffer.
1020   // When debug malloc is enabled, this call returns 0. This in turn causes
1021   // snprintf to do nothing, which causes libraries to fail to load.
1022   // See b/17302493 for further details.
1023   // Once the above bug is fixed, this code can be modified to use
1024   // snprintf again.
1025   size_t required_len = strlen(kDefaultLdPaths[0]) + strlen(kDefaultLdPaths[1]) + 2;
1026   if (buffer_size < required_len) {
1027     __libc_fatal("android_get_LD_LIBRARY_PATH failed, buffer too small: buffer len %zu, required len %zu",
1028                  buffer_size, required_len);
1029   }
1030   char* end = stpcpy(buffer, kDefaultLdPaths[0]);
1031   *end = ':';
1032   strcpy(end + 1, kDefaultLdPaths[1]);
1033 }
1034 
do_android_update_LD_LIBRARY_PATH(const char * ld_library_path)1035 void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
1036   if (!get_AT_SECURE()) {
1037     parse_LD_LIBRARY_PATH(ld_library_path);
1038   }
1039 }
1040 
do_dlopen(const char * name,int flags,const android_dlextinfo * extinfo)1041 soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
1042   if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL|RTLD_NOLOAD)) != 0) {
1043     DL_ERR("invalid flags to dlopen: %x", flags);
1044     return nullptr;
1045   }
1046   if (extinfo != nullptr) {
1047     if ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0) {
1048       DL_ERR("invalid extended flags to android_dlopen_ext: 0x%" PRIx64, extinfo->flags);
1049       return nullptr;
1050     }
1051     if ((extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD) == 0 &&
1052         (extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD_OFFSET) != 0) {
1053       DL_ERR("invalid extended flag combination (ANDROID_DLEXT_USE_LIBRARY_FD_OFFSET without ANDROID_DLEXT_USE_LIBRARY_FD): 0x%" PRIx64, extinfo->flags);
1054       return nullptr;
1055     }
1056   }
1057   protect_data(PROT_READ | PROT_WRITE);
1058   soinfo* si = find_library(name, flags, extinfo);
1059   if (si != nullptr) {
1060     si->CallConstructors();
1061   }
1062   protect_data(PROT_READ);
1063   return si;
1064 }
1065 
do_dlclose(soinfo * si)1066 void do_dlclose(soinfo* si) {
1067   protect_data(PROT_READ | PROT_WRITE);
1068   soinfo_unload(si);
1069   protect_data(PROT_READ);
1070 }
1071 
call_ifunc_resolver(ElfW (Addr)resolver_addr)1072 static ElfW(Addr) call_ifunc_resolver(ElfW(Addr) resolver_addr) {
1073   typedef ElfW(Addr) (*ifunc_resolver_t)(void);
1074   ifunc_resolver_t ifunc_resolver = reinterpret_cast<ifunc_resolver_t>(resolver_addr);
1075   ElfW(Addr) ifunc_addr = ifunc_resolver();
1076   TRACE_TYPE(RELO, "Called ifunc_resolver@%p. The result is %p", ifunc_resolver, reinterpret_cast<void*>(ifunc_addr));
1077 
1078   return ifunc_addr;
1079 }
1080 
1081 #if defined(USE_RELA)
Relocate(ElfW (Rela)* rela,unsigned count)1082 int soinfo::Relocate(ElfW(Rela)* rela, unsigned count) {
1083   for (size_t idx = 0; idx < count; ++idx, ++rela) {
1084     unsigned type = ELFW(R_TYPE)(rela->r_info);
1085     unsigned sym = ELFW(R_SYM)(rela->r_info);
1086     ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + load_bias);
1087     ElfW(Addr) sym_addr = 0;
1088     const char* sym_name = nullptr;
1089 
1090     DEBUG("Processing '%s' relocation at index %zd", name, idx);
1091     if (type == 0) { // R_*_NONE
1092       continue;
1093     }
1094 
1095     ElfW(Sym)* s = nullptr;
1096     soinfo* lsi = nullptr;
1097 
1098     if (sym != 0) {
1099       sym_name = get_string(symtab[sym].st_name);
1100       s = soinfo_do_lookup(this, sym_name, &lsi);
1101       if (s == nullptr) {
1102         // We only allow an undefined symbol if this is a weak reference...
1103         s = &symtab[sym];
1104         if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1105           DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, name);
1106           return -1;
1107         }
1108 
1109         /* IHI0044C AAELF 4.5.1.1:
1110 
1111            Libraries are not searched to resolve weak references.
1112            It is not an error for a weak reference to remain unsatisfied.
1113 
1114            During linking, the value of an undefined weak reference is:
1115            - Zero if the relocation type is absolute
1116            - The address of the place if the relocation is pc-relative
1117            - The address of nominal base address if the relocation
1118              type is base-relative.
1119          */
1120 
1121         switch (type) {
1122 #if defined(__aarch64__)
1123           case R_AARCH64_JUMP_SLOT:
1124           case R_AARCH64_GLOB_DAT:
1125           case R_AARCH64_ABS64:
1126           case R_AARCH64_ABS32:
1127           case R_AARCH64_ABS16:
1128           case R_AARCH64_RELATIVE:
1129           case R_AARCH64_IRELATIVE:
1130             /*
1131              * The sym_addr was initialized to be zero above, or the relocation
1132              * code below does not care about value of sym_addr.
1133              * No need to do anything.
1134              */
1135             break;
1136 #elif defined(__x86_64__)
1137           case R_X86_64_JUMP_SLOT:
1138           case R_X86_64_GLOB_DAT:
1139           case R_X86_64_32:
1140           case R_X86_64_64:
1141           case R_X86_64_RELATIVE:
1142           case R_X86_64_IRELATIVE:
1143             // No need to do anything.
1144             break;
1145           case R_X86_64_PC32:
1146             sym_addr = reloc;
1147             break;
1148 #endif
1149           default:
1150             DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
1151             return -1;
1152         }
1153       } else {
1154         // We got a definition.
1155         sym_addr = lsi->resolve_symbol_address(s);
1156       }
1157       count_relocation(kRelocSymbol);
1158     }
1159 
1160     switch (type) {
1161 #if defined(__aarch64__)
1162       case R_AARCH64_JUMP_SLOT:
1163         count_relocation(kRelocAbsolute);
1164         MARK(rela->r_offset);
1165         TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
1166                    reloc, (sym_addr + rela->r_addend), sym_name);
1167         *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
1168         break;
1169       case R_AARCH64_GLOB_DAT:
1170         count_relocation(kRelocAbsolute);
1171         MARK(rela->r_offset);
1172         TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
1173                    reloc, (sym_addr + rela->r_addend), sym_name);
1174         *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
1175         break;
1176       case R_AARCH64_ABS64:
1177         count_relocation(kRelocAbsolute);
1178         MARK(rela->r_offset);
1179         TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
1180                    reloc, (sym_addr + rela->r_addend), sym_name);
1181         *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1182         break;
1183       case R_AARCH64_ABS32:
1184         count_relocation(kRelocAbsolute);
1185         MARK(rela->r_offset);
1186         TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
1187                    reloc, (sym_addr + rela->r_addend), sym_name);
1188         if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1189             ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1190           *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1191         } else {
1192           DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1193                  (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
1194                  static_cast<ElfW(Addr)>(INT32_MIN),
1195                  static_cast<ElfW(Addr)>(UINT32_MAX));
1196           return -1;
1197         }
1198         break;
1199       case R_AARCH64_ABS16:
1200         count_relocation(kRelocAbsolute);
1201         MARK(rela->r_offset);
1202         TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
1203                    reloc, (sym_addr + rela->r_addend), sym_name);
1204         if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1205             ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1206           *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1207         } else {
1208           DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1209                  (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
1210                  static_cast<ElfW(Addr)>(INT16_MIN),
1211                  static_cast<ElfW(Addr)>(UINT16_MAX));
1212           return -1;
1213         }
1214         break;
1215       case R_AARCH64_PREL64:
1216         count_relocation(kRelocRelative);
1217         MARK(rela->r_offset);
1218         TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
1219                    reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1220         *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
1221         break;
1222       case R_AARCH64_PREL32:
1223         count_relocation(kRelocRelative);
1224         MARK(rela->r_offset);
1225         TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
1226                    reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1227         if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1228             ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1229           *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1230         } else {
1231           DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1232                  (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1233                  static_cast<ElfW(Addr)>(INT32_MIN),
1234                  static_cast<ElfW(Addr)>(UINT32_MAX));
1235           return -1;
1236         }
1237         break;
1238       case R_AARCH64_PREL16:
1239         count_relocation(kRelocRelative);
1240         MARK(rela->r_offset);
1241         TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1242                    reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1243         if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1244             ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1245           *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1246         } else {
1247           DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1248                  (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1249                  static_cast<ElfW(Addr)>(INT16_MIN),
1250                  static_cast<ElfW(Addr)>(UINT16_MAX));
1251           return -1;
1252         }
1253         break;
1254 
1255       case R_AARCH64_RELATIVE:
1256         count_relocation(kRelocRelative);
1257         MARK(rela->r_offset);
1258         if (sym) {
1259           DL_ERR("odd RELATIVE form...");
1260           return -1;
1261         }
1262         TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1263                    reloc, (base + rela->r_addend));
1264         *reinterpret_cast<ElfW(Addr)*>(reloc) = (base + rela->r_addend);
1265         break;
1266 
1267       case R_AARCH64_IRELATIVE:
1268         count_relocation(kRelocRelative);
1269         MARK(rela->r_offset);
1270         TRACE_TYPE(RELO, "RELO IRELATIVE %16llx <- %16llx\n", reloc, (base + rela->r_addend));
1271         *reinterpret_cast<ElfW(Addr)*>(reloc) = call_ifunc_resolver(base + rela->r_addend);
1272         break;
1273 
1274       case R_AARCH64_COPY:
1275         /*
1276          * ET_EXEC is not supported so this should not happen.
1277          *
1278          * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1279          *
1280          * Section 4.7.1.10 "Dynamic relocations"
1281          * R_AARCH64_COPY may only appear in executable objects where e_type is
1282          * set to ET_EXEC.
1283          */
1284         DL_ERR("%s R_AARCH64_COPY relocations are not supported", name);
1285         return -1;
1286       case R_AARCH64_TLS_TPREL64:
1287         TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1288                    reloc, (sym_addr + rela->r_addend), rela->r_offset);
1289         break;
1290       case R_AARCH64_TLS_DTPREL32:
1291         TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1292                    reloc, (sym_addr + rela->r_addend), rela->r_offset);
1293         break;
1294 #elif defined(__x86_64__)
1295       case R_X86_64_JUMP_SLOT:
1296         count_relocation(kRelocAbsolute);
1297         MARK(rela->r_offset);
1298         TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1299                    static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1300         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1301         break;
1302       case R_X86_64_GLOB_DAT:
1303         count_relocation(kRelocAbsolute);
1304         MARK(rela->r_offset);
1305         TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1306                    static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1307         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1308         break;
1309       case R_X86_64_RELATIVE:
1310         count_relocation(kRelocRelative);
1311         MARK(rela->r_offset);
1312         if (sym) {
1313           DL_ERR("odd RELATIVE form...");
1314           return -1;
1315         }
1316         TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1317                    static_cast<size_t>(base));
1318         *reinterpret_cast<ElfW(Addr)*>(reloc) = base + rela->r_addend;
1319         break;
1320       case R_X86_64_IRELATIVE:
1321         count_relocation(kRelocRelative);
1322         MARK(rela->r_offset);
1323         TRACE_TYPE(RELO, "RELO IRELATIVE %16llx <- %16llx\n", reloc, (base + rela->r_addend));
1324         *reinterpret_cast<ElfW(Addr)*>(reloc) = call_ifunc_resolver(base + rela->r_addend);
1325         break;
1326       case R_X86_64_32:
1327         count_relocation(kRelocRelative);
1328         MARK(rela->r_offset);
1329         TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1330                    static_cast<size_t>(sym_addr), sym_name);
1331         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1332         break;
1333       case R_X86_64_64:
1334         count_relocation(kRelocRelative);
1335         MARK(rela->r_offset);
1336         TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1337                    static_cast<size_t>(sym_addr), sym_name);
1338         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1339         break;
1340       case R_X86_64_PC32:
1341         count_relocation(kRelocRelative);
1342         MARK(rela->r_offset);
1343         TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1344                    static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1345                    static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1346         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1347         break;
1348 #endif
1349 
1350       default:
1351         DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1352         return -1;
1353     }
1354   }
1355   return 0;
1356 }
1357 
1358 #else // REL, not RELA.
Relocate(ElfW (Rel)* rel,unsigned count)1359 int soinfo::Relocate(ElfW(Rel)* rel, unsigned count) {
1360   for (size_t idx = 0; idx < count; ++idx, ++rel) {
1361     unsigned type = ELFW(R_TYPE)(rel->r_info);
1362     // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1363     unsigned sym = ELFW(R_SYM)(rel->r_info);
1364     ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + load_bias);
1365     ElfW(Addr) sym_addr = 0;
1366     const char* sym_name = nullptr;
1367 
1368     DEBUG("Processing '%s' relocation at index %zd", name, idx);
1369     if (type == 0) { // R_*_NONE
1370       continue;
1371     }
1372 
1373     ElfW(Sym)* s = nullptr;
1374     soinfo* lsi = nullptr;
1375 
1376     if (sym != 0) {
1377       sym_name = get_string(symtab[sym].st_name);
1378       s = soinfo_do_lookup(this, sym_name, &lsi);
1379       if (s == nullptr) {
1380         // We only allow an undefined symbol if this is a weak reference...
1381         s = &symtab[sym];
1382         if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1383           DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, name);
1384           return -1;
1385         }
1386 
1387         /* IHI0044C AAELF 4.5.1.1:
1388 
1389            Libraries are not searched to resolve weak references.
1390            It is not an error for a weak reference to remain
1391            unsatisfied.
1392 
1393            During linking, the value of an undefined weak reference is:
1394            - Zero if the relocation type is absolute
1395            - The address of the place if the relocation is pc-relative
1396            - The address of nominal base address if the relocation
1397              type is base-relative.
1398         */
1399 
1400         switch (type) {
1401 #if defined(__arm__)
1402           case R_ARM_JUMP_SLOT:
1403           case R_ARM_GLOB_DAT:
1404           case R_ARM_ABS32:
1405           case R_ARM_RELATIVE:    /* Don't care. */
1406             // sym_addr was initialized to be zero above or relocation
1407             // code below does not care about value of sym_addr.
1408             // No need to do anything.
1409             break;
1410 #elif defined(__i386__)
1411           case R_386_JMP_SLOT:
1412           case R_386_GLOB_DAT:
1413           case R_386_32:
1414           case R_386_RELATIVE:    /* Don't care. */
1415           case R_386_IRELATIVE:
1416             // sym_addr was initialized to be zero above or relocation
1417             // code below does not care about value of sym_addr.
1418             // No need to do anything.
1419             break;
1420           case R_386_PC32:
1421             sym_addr = reloc;
1422             break;
1423 #endif
1424 
1425 #if defined(__arm__)
1426           case R_ARM_COPY:
1427             // Fall through. Can't really copy if weak symbol is not found at run-time.
1428 #endif
1429           default:
1430             DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1431             return -1;
1432         }
1433       } else {
1434         // We got a definition.
1435         sym_addr = lsi->resolve_symbol_address(s);
1436       }
1437       count_relocation(kRelocSymbol);
1438     }
1439 
1440     switch (type) {
1441 #if defined(__arm__)
1442       case R_ARM_JUMP_SLOT:
1443         count_relocation(kRelocAbsolute);
1444         MARK(rel->r_offset);
1445         TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1446         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1447         break;
1448       case R_ARM_GLOB_DAT:
1449         count_relocation(kRelocAbsolute);
1450         MARK(rel->r_offset);
1451         TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1452         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1453         break;
1454       case R_ARM_ABS32:
1455         count_relocation(kRelocAbsolute);
1456         MARK(rel->r_offset);
1457         TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1458         *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1459         break;
1460       case R_ARM_REL32:
1461         count_relocation(kRelocRelative);
1462         MARK(rel->r_offset);
1463         TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1464                    reloc, sym_addr, rel->r_offset, sym_name);
1465         *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1466         break;
1467       case R_ARM_COPY:
1468         /*
1469          * ET_EXEC is not supported so this should not happen.
1470          *
1471          * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1472          *
1473          * Section 4.7.1.10 "Dynamic relocations"
1474          * R_ARM_COPY may only appear in executable objects where e_type is
1475          * set to ET_EXEC.
1476          */
1477         DL_ERR("%s R_ARM_COPY relocations are not supported", name);
1478         return -1;
1479 #elif defined(__i386__)
1480       case R_386_JMP_SLOT:
1481         count_relocation(kRelocAbsolute);
1482         MARK(rel->r_offset);
1483         TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1484         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1485         break;
1486       case R_386_GLOB_DAT:
1487         count_relocation(kRelocAbsolute);
1488         MARK(rel->r_offset);
1489         TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1490         *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1491         break;
1492       case R_386_32:
1493         count_relocation(kRelocRelative);
1494         MARK(rel->r_offset);
1495         TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1496         *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1497         break;
1498       case R_386_PC32:
1499         count_relocation(kRelocRelative);
1500         MARK(rel->r_offset);
1501         TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1502                    reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1503         *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1504         break;
1505 #elif defined(__mips__)
1506       case R_MIPS_REL32:
1507 #if defined(__LP64__)
1508         // MIPS Elf64_Rel entries contain compound relocations
1509         // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1510         if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1511             ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1512           DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1513                  type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1514                  (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1515           return -1;
1516         }
1517 #endif
1518         count_relocation(kRelocAbsolute);
1519         MARK(rel->r_offset);
1520         TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1521                    static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1522         if (s) {
1523           *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1524         } else {
1525           *reinterpret_cast<ElfW(Addr)*>(reloc) += base;
1526         }
1527         break;
1528 #endif
1529 
1530 #if defined(__arm__)
1531       case R_ARM_RELATIVE:
1532 #elif defined(__i386__)
1533       case R_386_RELATIVE:
1534 #endif
1535         count_relocation(kRelocRelative);
1536         MARK(rel->r_offset);
1537         if (sym) {
1538           DL_ERR("odd RELATIVE form...");
1539           return -1;
1540         }
1541         TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1542                    reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(base));
1543         *reinterpret_cast<ElfW(Addr)*>(reloc) += base;
1544         break;
1545 #if defined(__i386__)
1546       case R_386_IRELATIVE:
1547         count_relocation(kRelocRelative);
1548         MARK(rel->r_offset);
1549         TRACE_TYPE(RELO, "RELO IRELATIVE %p <- %p", reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(base));
1550         *reinterpret_cast<ElfW(Addr)*>(reloc) = call_ifunc_resolver(base + *reinterpret_cast<ElfW(Addr)*>(reloc));
1551         break;
1552 #endif
1553 
1554       default:
1555         DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1556         return -1;
1557     }
1558   }
1559   return 0;
1560 }
1561 #endif
1562 
1563 #if defined(__mips__)
mips_relocate_got(soinfo * si)1564 static bool mips_relocate_got(soinfo* si) {
1565   ElfW(Addr)** got = si->plt_got;
1566   if (got == nullptr) {
1567     return true;
1568   }
1569   unsigned local_gotno = si->mips_local_gotno;
1570   unsigned gotsym = si->mips_gotsym;
1571   unsigned symtabno = si->mips_symtabno;
1572   ElfW(Sym)* symtab = si->symtab;
1573 
1574   // got[0] is the address of the lazy resolver function.
1575   // got[1] may be used for a GNU extension.
1576   // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1577   // FIXME: maybe this should be in a separate routine?
1578   if ((si->flags & FLAG_LINKER) == 0) {
1579     size_t g = 0;
1580     got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1581     if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1582       got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1583     }
1584     // Relocate the local GOT entries.
1585     for (; g < local_gotno; g++) {
1586       got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1587     }
1588   }
1589 
1590   // Now for the global GOT entries...
1591   ElfW(Sym)* sym = symtab + gotsym;
1592   got = si->plt_got + local_gotno;
1593   for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1594     // This is an undefined reference... try to locate it.
1595     const char* sym_name = si->get_string(sym->st_name);
1596     soinfo* lsi = nullptr;
1597     ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi);
1598     if (s == nullptr) {
1599       // We only allow an undefined symbol if this is a weak reference.
1600       s = &symtab[g];
1601       if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1602         DL_ERR("cannot locate \"%s\"...", sym_name);
1603         return false;
1604       }
1605       *got = 0;
1606     } else {
1607       // FIXME: is this sufficient?
1608       // For reference see NetBSD link loader
1609       // http://cvsweb.netbsd.org/bsdweb.cgi/src/libexec/ld.elf_so/arch/mips/mips_reloc.c?rev=1.53&content-type=text/x-cvsweb-markup
1610       *got = reinterpret_cast<ElfW(Addr)*>(lsi->resolve_symbol_address(s));
1611     }
1612   }
1613   return true;
1614 }
1615 #endif
1616 
CallArray(const char * array_name __unused,linker_function_t * functions,size_t count,bool reverse)1617 void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1618   if (functions == nullptr) {
1619     return;
1620   }
1621 
1622   TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1623 
1624   int begin = reverse ? (count - 1) : 0;
1625   int end = reverse ? -1 : count;
1626   int step = reverse ? -1 : 1;
1627 
1628   for (int i = begin; i != end; i += step) {
1629     TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1630     CallFunction("function", functions[i]);
1631   }
1632 
1633   TRACE("[ Done calling %s for '%s' ]", array_name, name);
1634 }
1635 
CallFunction(const char * function_name __unused,linker_function_t function)1636 void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1637   if (function == nullptr || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1638     return;
1639   }
1640 
1641   TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1642   function();
1643   TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1644 
1645   // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1646   // are still writable. This happens with our debug malloc (see http://b/7941716).
1647   protect_data(PROT_READ | PROT_WRITE);
1648 }
1649 
CallPreInitConstructors()1650 void soinfo::CallPreInitConstructors() {
1651   // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1652   // but ignored in a shared library.
1653   CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1654 }
1655 
CallConstructors()1656 void soinfo::CallConstructors() {
1657   if (constructors_called) {
1658     return;
1659   }
1660 
1661   // We set constructors_called before actually calling the constructors, otherwise it doesn't
1662   // protect against recursive constructor calls. One simple example of constructor recursion
1663   // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1664   // 1. The program depends on libc, so libc's constructor is called here.
1665   // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1666   // 3. dlopen() calls the constructors on the newly created
1667   //    soinfo for libc_malloc_debug_leak.so.
1668   // 4. The debug .so depends on libc, so CallConstructors is
1669   //    called again with the libc soinfo. If it doesn't trigger the early-
1670   //    out above, the libc constructor will be called again (recursively!).
1671   constructors_called = true;
1672 
1673   if ((flags & FLAG_EXE) == 0 && preinit_array != nullptr) {
1674     // The GNU dynamic linker silently ignores these, but we warn the developer.
1675     PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1676           name, preinit_array_count);
1677   }
1678 
1679   get_children().for_each([] (soinfo* si) {
1680     si->CallConstructors();
1681   });
1682 
1683   TRACE("\"%s\": calling constructors", name);
1684 
1685   // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1686   CallFunction("DT_INIT", init_func);
1687   CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1688 }
1689 
CallDestructors()1690 void soinfo::CallDestructors() {
1691   if (!constructors_called) {
1692     return;
1693   }
1694   TRACE("\"%s\": calling destructors", name);
1695 
1696   // DT_FINI_ARRAY must be parsed in reverse order.
1697   CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1698 
1699   // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1700   CallFunction("DT_FINI", fini_func);
1701 
1702   // This is needed on second call to dlopen
1703   // after library has been unloaded with RTLD_NODELETE
1704   constructors_called = false;
1705 }
1706 
add_child(soinfo * child)1707 void soinfo::add_child(soinfo* child) {
1708   if (has_min_version(0)) {
1709     child->parents.push_back(this);
1710     this->children.push_back(child);
1711   }
1712 }
1713 
remove_all_links()1714 void soinfo::remove_all_links() {
1715   if (!has_min_version(0)) {
1716     return;
1717   }
1718 
1719   // 1. Untie connected soinfos from 'this'.
1720   children.for_each([&] (soinfo* child) {
1721     child->parents.remove_if([&] (const soinfo* parent) {
1722       return parent == this;
1723     });
1724   });
1725 
1726   parents.for_each([&] (soinfo* parent) {
1727     parent->children.remove_if([&] (const soinfo* child) {
1728       return child == this;
1729     });
1730   });
1731 
1732   // 2. Once everything untied - clear local lists.
1733   parents.clear();
1734   children.clear();
1735 }
1736 
get_st_dev()1737 dev_t soinfo::get_st_dev() {
1738   if (has_min_version(0)) {
1739     return st_dev;
1740   }
1741 
1742   return 0;
1743 };
1744 
get_st_ino()1745 ino_t soinfo::get_st_ino() {
1746   if (has_min_version(0)) {
1747     return st_ino;
1748   }
1749 
1750   return 0;
1751 }
1752 
get_file_offset()1753 off64_t soinfo::get_file_offset() {
1754   if (has_min_version(1)) {
1755     return file_offset;
1756   }
1757 
1758   return 0;
1759 }
1760 
1761 // This is a return on get_children()/get_parents() if
1762 // 'this->flags' does not have FLAG_NEW_SOINFO set.
1763 static soinfo::soinfo_list_t g_empty_list;
1764 
get_children()1765 soinfo::soinfo_list_t& soinfo::get_children() {
1766   if (has_min_version(0)) {
1767     return this->children;
1768   }
1769 
1770   return g_empty_list;
1771 }
1772 
get_parents()1773 soinfo::soinfo_list_t& soinfo::get_parents() {
1774   if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1775     return g_empty_list;
1776   }
1777 
1778   return this->parents;
1779 }
1780 
ElfW(Addr)1781 ElfW(Addr) soinfo::resolve_symbol_address(ElfW(Sym)* s) {
1782   if (ELF_ST_TYPE(s->st_info) == STT_GNU_IFUNC) {
1783     return call_ifunc_resolver(s->st_value + load_bias);
1784   }
1785 
1786   return static_cast<ElfW(Addr)>(s->st_value + load_bias);
1787 }
1788 
get_string(ElfW (Word)index) const1789 const char* soinfo::get_string(ElfW(Word) index) const {
1790   if (has_min_version(1) && (index >= strtab_size)) {
1791     __libc_fatal("%s: strtab out of bounds error; STRSZ=%zd, name=%d", name, strtab_size, index);
1792   }
1793 
1794   return strtab + index;
1795 }
1796 
1797 /* Force any of the closed stdin, stdout and stderr to be associated with
1798    /dev/null. */
nullify_closed_stdio()1799 static int nullify_closed_stdio() {
1800   int dev_null, i, status;
1801   int return_value = 0;
1802 
1803   dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1804   if (dev_null < 0) {
1805     DL_ERR("cannot open /dev/null: %s", strerror(errno));
1806     return -1;
1807   }
1808   TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1809 
1810   /* If any of the stdio file descriptors is valid and not associated
1811      with /dev/null, dup /dev/null to it.  */
1812   for (i = 0; i < 3; i++) {
1813     /* If it is /dev/null already, we are done. */
1814     if (i == dev_null) {
1815       continue;
1816     }
1817 
1818     TRACE("[ Nullifying stdio file descriptor %d]", i);
1819     status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1820 
1821     /* If file is opened, we are good. */
1822     if (status != -1) {
1823       continue;
1824     }
1825 
1826     /* The only error we allow is that the file descriptor does not
1827        exist, in which case we dup /dev/null to it. */
1828     if (errno != EBADF) {
1829       DL_ERR("fcntl failed: %s", strerror(errno));
1830       return_value = -1;
1831       continue;
1832     }
1833 
1834     /* Try dupping /dev/null to this stdio file descriptor and
1835        repeat if there is a signal.  Note that any errors in closing
1836        the stdio descriptor are lost.  */
1837     status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1838     if (status < 0) {
1839       DL_ERR("dup2 failed: %s", strerror(errno));
1840       return_value = -1;
1841       continue;
1842     }
1843   }
1844 
1845   /* If /dev/null is not one of the stdio file descriptors, close it. */
1846   if (dev_null > 2) {
1847     TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1848     status = TEMP_FAILURE_RETRY(close(dev_null));
1849     if (status == -1) {
1850       DL_ERR("close failed: %s", strerror(errno));
1851       return_value = -1;
1852     }
1853   }
1854 
1855   return return_value;
1856 }
1857 
PrelinkImage()1858 bool soinfo::PrelinkImage() {
1859   /* Extract dynamic section */
1860   ElfW(Word) dynamic_flags = 0;
1861   phdr_table_get_dynamic_section(phdr, phnum, load_bias, &dynamic, &dynamic_flags);
1862 
1863   /* We can't log anything until the linker is relocated */
1864   bool relocating_linker = (flags & FLAG_LINKER) != 0;
1865   if (!relocating_linker) {
1866     INFO("[ linking %s ]", name);
1867     DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(base), flags);
1868   }
1869 
1870   if (dynamic == nullptr) {
1871     if (!relocating_linker) {
1872       DL_ERR("missing PT_DYNAMIC in \"%s\"", name);
1873     }
1874     return false;
1875   } else {
1876     if (!relocating_linker) {
1877       DEBUG("dynamic = %p", dynamic);
1878     }
1879   }
1880 
1881 #if defined(__arm__)
1882   (void) phdr_table_get_arm_exidx(phdr, phnum, load_bias,
1883                                   &ARM_exidx, &ARM_exidx_count);
1884 #endif
1885 
1886   // Extract useful information from dynamic section.
1887   uint32_t needed_count = 0;
1888   for (ElfW(Dyn)* d = dynamic; d->d_tag != DT_NULL; ++d) {
1889     DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1890           d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1891     switch (d->d_tag) {
1892       case DT_SONAME:
1893         // TODO: glibc dynamic linker uses this name for
1894         // initial library lookup; consider doing the same here.
1895         break;
1896 
1897       case DT_HASH:
1898         nbucket = reinterpret_cast<uint32_t*>(load_bias + d->d_un.d_ptr)[0];
1899         nchain = reinterpret_cast<uint32_t*>(load_bias + d->d_un.d_ptr)[1];
1900         bucket = reinterpret_cast<uint32_t*>(load_bias + d->d_un.d_ptr + 8);
1901         chain = reinterpret_cast<uint32_t*>(load_bias + d->d_un.d_ptr + 8 + nbucket * 4);
1902         break;
1903 
1904       case DT_STRTAB:
1905         strtab = reinterpret_cast<const char*>(load_bias + d->d_un.d_ptr);
1906         break;
1907 
1908       case DT_STRSZ:
1909         strtab_size = d->d_un.d_val;
1910         break;
1911 
1912       case DT_SYMTAB:
1913         symtab = reinterpret_cast<ElfW(Sym)*>(load_bias + d->d_un.d_ptr);
1914         break;
1915 
1916       case DT_SYMENT:
1917         if (d->d_un.d_val != sizeof(ElfW(Sym))) {
1918           DL_ERR("invalid DT_SYMENT: %zd", static_cast<size_t>(d->d_un.d_val));
1919           return false;
1920         }
1921         break;
1922 
1923       case DT_PLTREL:
1924 #if defined(USE_RELA)
1925         if (d->d_un.d_val != DT_RELA) {
1926           DL_ERR("unsupported DT_PLTREL in \"%s\"; expected DT_RELA", name);
1927           return false;
1928         }
1929 #else
1930         if (d->d_un.d_val != DT_REL) {
1931           DL_ERR("unsupported DT_PLTREL in \"%s\"; expected DT_REL", name);
1932           return false;
1933         }
1934 #endif
1935         break;
1936 
1937       case DT_JMPREL:
1938 #if defined(USE_RELA)
1939         plt_rela = reinterpret_cast<ElfW(Rela)*>(load_bias + d->d_un.d_ptr);
1940 #else
1941         plt_rel = reinterpret_cast<ElfW(Rel)*>(load_bias + d->d_un.d_ptr);
1942 #endif
1943         break;
1944 
1945       case DT_PLTRELSZ:
1946 #if defined(USE_RELA)
1947         plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1948 #else
1949         plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1950 #endif
1951         break;
1952 
1953       case DT_PLTGOT:
1954 #if defined(__mips__)
1955         // Used by mips and mips64.
1956         plt_got = reinterpret_cast<ElfW(Addr)**>(load_bias + d->d_un.d_ptr);
1957 #endif
1958         // Ignore for other platforms... (because RTLD_LAZY is not supported)
1959         break;
1960 
1961       case DT_DEBUG:
1962         // Set the DT_DEBUG entry to the address of _r_debug for GDB
1963         // if the dynamic table is writable
1964 // FIXME: not working currently for N64
1965 // The flags for the LOAD and DYNAMIC program headers do not agree.
1966 // The LOAD section containing the dynamic table has been mapped as
1967 // read-only, but the DYNAMIC header claims it is writable.
1968 #if !(defined(__mips__) && defined(__LP64__))
1969         if ((dynamic_flags & PF_W) != 0) {
1970           d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1971         }
1972         break;
1973 #endif
1974 #if defined(USE_RELA)
1975       case DT_RELA:
1976         rela = reinterpret_cast<ElfW(Rela)*>(load_bias + d->d_un.d_ptr);
1977         break;
1978 
1979       case DT_RELASZ:
1980         rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1981         break;
1982 
1983       case DT_RELAENT:
1984         if (d->d_un.d_val != sizeof(ElfW(Rela))) {
1985           DL_ERR("invalid DT_RELAENT: %zd", static_cast<size_t>(d->d_un.d_val));
1986           return false;
1987         }
1988         break;
1989 
1990       // ignored (see DT_RELCOUNT comments for details)
1991       case DT_RELACOUNT:
1992         break;
1993 
1994       case DT_REL:
1995         DL_ERR("unsupported DT_REL in \"%s\"", name);
1996         return false;
1997 
1998       case DT_RELSZ:
1999         DL_ERR("unsupported DT_RELSZ in \"%s\"", name);
2000         return false;
2001 #else
2002       case DT_REL:
2003         rel = reinterpret_cast<ElfW(Rel)*>(load_bias + d->d_un.d_ptr);
2004         break;
2005 
2006       case DT_RELSZ:
2007         rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
2008         break;
2009 
2010       case DT_RELENT:
2011         if (d->d_un.d_val != sizeof(ElfW(Rel))) {
2012           DL_ERR("invalid DT_RELENT: %zd", static_cast<size_t>(d->d_un.d_val));
2013           return false;
2014         }
2015         break;
2016 
2017       // "Indicates that all RELATIVE relocations have been concatenated together,
2018       // and specifies the RELATIVE relocation count."
2019       //
2020       // TODO: Spec also mentions that this can be used to optimize relocation process;
2021       // Not currently used by bionic linker - ignored.
2022       case DT_RELCOUNT:
2023         break;
2024       case DT_RELA:
2025         DL_ERR("unsupported DT_RELA in \"%s\"", name);
2026         return false;
2027 #endif
2028       case DT_INIT:
2029         init_func = reinterpret_cast<linker_function_t>(load_bias + d->d_un.d_ptr);
2030         DEBUG("%s constructors (DT_INIT) found at %p", name, init_func);
2031         break;
2032 
2033       case DT_FINI:
2034         fini_func = reinterpret_cast<linker_function_t>(load_bias + d->d_un.d_ptr);
2035         DEBUG("%s destructors (DT_FINI) found at %p", name, fini_func);
2036         break;
2037 
2038       case DT_INIT_ARRAY:
2039         init_array = reinterpret_cast<linker_function_t*>(load_bias + d->d_un.d_ptr);
2040         DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", name, init_array);
2041         break;
2042 
2043       case DT_INIT_ARRAYSZ:
2044         init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
2045         break;
2046 
2047       case DT_FINI_ARRAY:
2048         fini_array = reinterpret_cast<linker_function_t*>(load_bias + d->d_un.d_ptr);
2049         DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", name, fini_array);
2050         break;
2051 
2052       case DT_FINI_ARRAYSZ:
2053         fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
2054         break;
2055 
2056       case DT_PREINIT_ARRAY:
2057         preinit_array = reinterpret_cast<linker_function_t*>(load_bias + d->d_un.d_ptr);
2058         DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", name, preinit_array);
2059         break;
2060 
2061       case DT_PREINIT_ARRAYSZ:
2062         preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
2063         break;
2064 
2065       case DT_TEXTREL:
2066 #if defined(__LP64__)
2067         DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", name);
2068         return false;
2069 #else
2070         has_text_relocations = true;
2071         break;
2072 #endif
2073 
2074       case DT_SYMBOLIC:
2075         has_DT_SYMBOLIC = true;
2076         break;
2077 
2078       case DT_NEEDED:
2079         ++needed_count;
2080         break;
2081 
2082       case DT_FLAGS:
2083         if (d->d_un.d_val & DF_TEXTREL) {
2084 #if defined(__LP64__)
2085           DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", name);
2086           return false;
2087 #else
2088           has_text_relocations = true;
2089 #endif
2090         }
2091         if (d->d_un.d_val & DF_SYMBOLIC) {
2092           has_DT_SYMBOLIC = true;
2093         }
2094         break;
2095 
2096       case DT_FLAGS_1:
2097         if ((d->d_un.d_val & DF_1_GLOBAL) != 0) {
2098           rtld_flags |= RTLD_GLOBAL;
2099         }
2100         // TODO: Implement other flags
2101 
2102         if ((d->d_un.d_val & ~(DF_1_NOW | DF_1_GLOBAL)) != 0) {
2103           DL_WARN("Unsupported flags DT_FLAGS_1=%p", reinterpret_cast<void*>(d->d_un.d_val));
2104         }
2105         break;
2106 #if defined(__mips__)
2107       case DT_MIPS_RLD_MAP:
2108         // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
2109         {
2110           r_debug** dp = reinterpret_cast<r_debug**>(load_bias + d->d_un.d_ptr);
2111           *dp = &_r_debug;
2112         }
2113         break;
2114 
2115       case DT_MIPS_RLD_VERSION:
2116       case DT_MIPS_FLAGS:
2117       case DT_MIPS_BASE_ADDRESS:
2118       case DT_MIPS_UNREFEXTNO:
2119         break;
2120 
2121       case DT_MIPS_SYMTABNO:
2122         mips_symtabno = d->d_un.d_val;
2123         break;
2124 
2125       case DT_MIPS_LOCAL_GOTNO:
2126         mips_local_gotno = d->d_un.d_val;
2127         break;
2128 
2129       case DT_MIPS_GOTSYM:
2130         mips_gotsym = d->d_un.d_val;
2131         break;
2132 #endif
2133       // Ignored: "Its use has been superseded by the DF_BIND_NOW flag"
2134       case DT_BIND_NOW:
2135         break;
2136 
2137       // Ignore: bionic does not support symbol versioning...
2138       case DT_VERSYM:
2139       case DT_VERDEF:
2140       case DT_VERDEFNUM:
2141         break;
2142 
2143       default:
2144         if (!relocating_linker) {
2145           DL_WARN("%s: unused DT entry: type %p arg %p", name,
2146               reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
2147         }
2148         break;
2149     }
2150   }
2151 
2152   DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
2153         reinterpret_cast<void*>(base), strtab, symtab);
2154 
2155   // Sanity checks.
2156   if (relocating_linker && needed_count != 0) {
2157     DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
2158     return false;
2159   }
2160   if (nbucket == 0) {
2161     DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", name);
2162     return false;
2163   }
2164   if (strtab == 0) {
2165     DL_ERR("empty/missing DT_STRTAB in \"%s\"", name);
2166     return false;
2167   }
2168   if (symtab == 0) {
2169     DL_ERR("empty/missing DT_SYMTAB in \"%s\"", name);
2170     return false;
2171   }
2172   return true;
2173 }
2174 
LinkImage(const android_dlextinfo * extinfo)2175 bool soinfo::LinkImage(const android_dlextinfo* extinfo) {
2176 
2177 #if !defined(__LP64__)
2178   if (has_text_relocations) {
2179     // Make segments writable to allow text relocations to work properly. We will later call
2180     // phdr_table_protect_segments() after all of them are applied and all constructors are run.
2181     DL_WARN("%s has text relocations. This is wasting memory and prevents "
2182             "security hardening. Please fix.", name);
2183     if (phdr_table_unprotect_segments(phdr, phnum, load_bias) < 0) {
2184       DL_ERR("can't unprotect loadable segments for \"%s\": %s",
2185              name, strerror(errno));
2186       return false;
2187     }
2188   }
2189 #endif
2190 
2191 #if defined(USE_RELA)
2192   if (rela != nullptr) {
2193     DEBUG("[ relocating %s ]", name);
2194     if (Relocate(rela, rela_count)) {
2195       return false;
2196     }
2197   }
2198   if (plt_rela != nullptr) {
2199     DEBUG("[ relocating %s plt ]", name);
2200     if (Relocate(plt_rela, plt_rela_count)) {
2201       return false;
2202     }
2203   }
2204 #else
2205   if (rel != nullptr) {
2206     DEBUG("[ relocating %s ]", name);
2207     if (Relocate(rel, rel_count)) {
2208       return false;
2209     }
2210   }
2211   if (plt_rel != nullptr) {
2212     DEBUG("[ relocating %s plt ]", name);
2213     if (Relocate(plt_rel, plt_rel_count)) {
2214       return false;
2215     }
2216   }
2217 #endif
2218 
2219 #if defined(__mips__)
2220   if (!mips_relocate_got(this)) {
2221     return false;
2222   }
2223 #endif
2224 
2225   DEBUG("[ finished linking %s ]", name);
2226 
2227 #if !defined(__LP64__)
2228   if (has_text_relocations) {
2229     // All relocations are done, we can protect our segments back to read-only.
2230     if (phdr_table_protect_segments(phdr, phnum, load_bias) < 0) {
2231       DL_ERR("can't protect segments for \"%s\": %s",
2232              name, strerror(errno));
2233       return false;
2234     }
2235   }
2236 #endif
2237 
2238   /* We can also turn on GNU RELRO protection */
2239   if (phdr_table_protect_gnu_relro(phdr, phnum, load_bias) < 0) {
2240     DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
2241            name, strerror(errno));
2242     return false;
2243   }
2244 
2245   /* Handle serializing/sharing the RELRO segment */
2246   if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
2247     if (phdr_table_serialize_gnu_relro(phdr, phnum, load_bias,
2248                                        extinfo->relro_fd) < 0) {
2249       DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
2250              name, strerror(errno));
2251       return false;
2252     }
2253   } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
2254     if (phdr_table_map_gnu_relro(phdr, phnum, load_bias,
2255                                  extinfo->relro_fd) < 0) {
2256       DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
2257              name, strerror(errno));
2258       return false;
2259     }
2260   }
2261 
2262   notify_gdb_of_load(this);
2263   return true;
2264 }
2265 
2266 /*
2267  * This function add vdso to internal dso list.
2268  * It helps to stack unwinding through signal handlers.
2269  * Also, it makes bionic more like glibc.
2270  */
add_vdso(KernelArgumentBlock & args __unused)2271 static void add_vdso(KernelArgumentBlock& args __unused) {
2272 #if defined(AT_SYSINFO_EHDR)
2273   ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
2274   if (ehdr_vdso == nullptr) {
2275     return;
2276   }
2277 
2278   soinfo* si = soinfo_alloc("[vdso]", nullptr, 0);
2279 
2280   si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
2281   si->phnum = ehdr_vdso->e_phnum;
2282   si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
2283   si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2284   si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
2285 
2286   si->PrelinkImage();
2287   si->LinkImage(nullptr);
2288 #endif
2289 }
2290 
2291 /*
2292  * This is linker soinfo for GDB. See details below.
2293  */
2294 #if defined(__LP64__)
2295 #define LINKER_PATH "/system/bin/linker64"
2296 #else
2297 #define LINKER_PATH "/system/bin/linker"
2298 #endif
2299 static soinfo linker_soinfo_for_gdb(LINKER_PATH, nullptr, 0);
2300 
2301 /* gdb expects the linker to be in the debug shared object list.
2302  * Without this, gdb has trouble locating the linker's ".text"
2303  * and ".plt" sections. Gdb could also potentially use this to
2304  * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2305  * Don't use soinfo_alloc(), because the linker shouldn't
2306  * be on the soinfo list.
2307  */
init_linker_info_for_gdb(ElfW (Addr)linker_base)2308 static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
2309   linker_soinfo_for_gdb.base = linker_base;
2310 
2311   /*
2312    * Set the dynamic field in the link map otherwise gdb will complain with
2313    * the following:
2314    *   warning: .dynamic section for "/system/bin/linker" is not at the
2315    *   expected address (wrong library or version mismatch?)
2316    */
2317   ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2318   ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2319   phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2320                                  &linker_soinfo_for_gdb.dynamic, nullptr);
2321   insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2322 }
2323 
2324 /*
2325  * This code is called after the linker has linked itself and
2326  * fixed it's own GOT. It is safe to make references to externs
2327  * and other non-local data at this point.
2328  */
__linker_init_post_relocation(KernelArgumentBlock & args,ElfW (Addr)linker_base)2329 static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2330 #if TIMING
2331   struct timeval t0, t1;
2332   gettimeofday(&t0, 0);
2333 #endif
2334 
2335   // Initialize environment functions, and get to the ELF aux vectors table.
2336   linker_env_init(args);
2337 
2338   // If this is a setuid/setgid program, close the security hole described in
2339   // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2340   if (get_AT_SECURE()) {
2341     nullify_closed_stdio();
2342   }
2343 
2344   debuggerd_init();
2345 
2346   // Get a few environment variables.
2347   const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2348   if (LD_DEBUG != nullptr) {
2349     g_ld_debug_verbosity = atoi(LD_DEBUG);
2350   }
2351 
2352   // Normally, these are cleaned by linker_env_init, but the test
2353   // doesn't cost us anything.
2354   const char* ldpath_env = nullptr;
2355   const char* ldpreload_env = nullptr;
2356   if (!get_AT_SECURE()) {
2357     ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2358     ldpreload_env = linker_env_get("LD_PRELOAD");
2359   }
2360 
2361   INFO("[ android linker & debugger ]");
2362 
2363   soinfo* si = soinfo_alloc(args.argv[0], nullptr, 0);
2364   if (si == nullptr) {
2365     exit(EXIT_FAILURE);
2366   }
2367 
2368   /* bootstrap the link map, the main exe always needs to be first */
2369   si->flags |= FLAG_EXE;
2370   link_map* map = &(si->link_map_head);
2371 
2372   map->l_addr = 0;
2373   map->l_name = args.argv[0];
2374   map->l_prev = nullptr;
2375   map->l_next = nullptr;
2376 
2377   _r_debug.r_map = map;
2378   r_debug_tail = map;
2379 
2380   init_linker_info_for_gdb(linker_base);
2381 
2382   // Extract information passed from the kernel.
2383   si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2384   si->phnum = args.getauxval(AT_PHNUM);
2385   si->entry = args.getauxval(AT_ENTRY);
2386 
2387   /* Compute the value of si->base. We can't rely on the fact that
2388    * the first entry is the PHDR because this will not be true
2389    * for certain executables (e.g. some in the NDK unit test suite)
2390    */
2391   si->base = 0;
2392   si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2393   si->load_bias = 0;
2394   for (size_t i = 0; i < si->phnum; ++i) {
2395     if (si->phdr[i].p_type == PT_PHDR) {
2396       si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2397       si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2398       break;
2399     }
2400   }
2401   si->dynamic = nullptr;
2402   si->ref_count = 1;
2403 
2404   ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2405   if (elf_hdr->e_type != ET_DYN) {
2406     __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2407     exit(EXIT_FAILURE);
2408   }
2409 
2410   // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2411   parse_LD_LIBRARY_PATH(ldpath_env);
2412   parse_LD_PRELOAD(ldpreload_env);
2413 
2414   somain = si;
2415 
2416   if (!si->PrelinkImage()) {
2417     __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2418     exit(EXIT_FAILURE);
2419   }
2420 
2421   // Load ld_preloads and dependencies.
2422   StringLinkedList needed_library_name_list;
2423   size_t needed_libraries_count = 0;
2424   size_t ld_preloads_count = 0;
2425   while (g_ld_preload_names[ld_preloads_count] != nullptr) {
2426     needed_library_name_list.push_back(g_ld_preload_names[ld_preloads_count++]);
2427     ++needed_libraries_count;
2428   }
2429 
2430   for_each_dt_needed(si, [&](const char* name) {
2431     needed_library_name_list.push_back(name);
2432     ++needed_libraries_count;
2433   });
2434 
2435   const char* needed_library_names[needed_libraries_count];
2436   soinfo* needed_library_si[needed_libraries_count];
2437 
2438   memset(needed_library_names, 0, sizeof(needed_library_names));
2439   needed_library_name_list.copy_to_array(needed_library_names, needed_libraries_count);
2440 
2441   if (needed_libraries_count > 0 && !find_libraries(needed_library_names, needed_libraries_count, needed_library_si, g_ld_preloads, ld_preloads_count, 0, nullptr)) {
2442     __libc_format_fd(2, "CANNOT LINK EXECUTABLE DEPENDENCIES: %s\n", linker_get_error_buffer());
2443     exit(EXIT_FAILURE);
2444   }
2445 
2446   for (size_t i = 0; i<needed_libraries_count; ++i) {
2447     si->add_child(needed_library_si[i]);
2448   }
2449 
2450   if (!si->LinkImage(nullptr)) {
2451     __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2452     exit(EXIT_FAILURE);
2453   }
2454 
2455   add_vdso(args);
2456 
2457   si->CallPreInitConstructors();
2458 
2459   /* After the PrelinkImage, the si->load_bias is initialized.
2460    * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2461    * We need to update this value for so exe here. So Unwind_Backtrace
2462    * for some arch like x86 could work correctly within so exe.
2463    */
2464   map->l_addr = si->load_bias;
2465   si->CallConstructors();
2466 
2467 #if TIMING
2468   gettimeofday(&t1, nullptr);
2469   PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2470            (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2471            (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2472 #endif
2473 #if STATS
2474   PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2475          linker_stats.count[kRelocAbsolute],
2476          linker_stats.count[kRelocRelative],
2477          linker_stats.count[kRelocCopy],
2478          linker_stats.count[kRelocSymbol]);
2479 #endif
2480 #if COUNT_PAGES
2481   {
2482     unsigned n;
2483     unsigned i;
2484     unsigned count = 0;
2485     for (n = 0; n < 4096; n++) {
2486       if (bitmask[n]) {
2487         unsigned x = bitmask[n];
2488 #if defined(__LP64__)
2489         for (i = 0; i < 32; i++) {
2490 #else
2491         for (i = 0; i < 8; i++) {
2492 #endif
2493           if (x & 1) {
2494             count++;
2495           }
2496           x >>= 1;
2497         }
2498       }
2499     }
2500     PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2501   }
2502 #endif
2503 
2504 #if TIMING || STATS || COUNT_PAGES
2505   fflush(stdout);
2506 #endif
2507 
2508   TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2509   return si->entry;
2510 }
2511 
2512 /* Compute the load-bias of an existing executable. This shall only
2513  * be used to compute the load bias of an executable or shared library
2514  * that was loaded by the kernel itself.
2515  *
2516  * Input:
2517  *    elf    -> address of ELF header, assumed to be at the start of the file.
2518  * Return:
2519  *    load bias, i.e. add the value of any p_vaddr in the file to get
2520  *    the corresponding address in memory.
2521  */
2522 static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2523   ElfW(Addr) offset = elf->e_phoff;
2524   const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2525   const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2526 
2527   for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2528     if (phdr->p_type == PT_LOAD) {
2529       return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2530     }
2531   }
2532   return 0;
2533 }
2534 
2535 extern "C" void _start();
2536 
2537 /*
2538  * This is the entry point for the linker, called from begin.S. This
2539  * method is responsible for fixing the linker's own relocations, and
2540  * then calling __linker_init_post_relocation().
2541  *
2542  * Because this method is called before the linker has fixed it's own
2543  * relocations, any attempt to reference an extern variable, extern
2544  * function, or other GOT reference will generate a segfault.
2545  */
2546 extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2547   KernelArgumentBlock args(raw_args);
2548 
2549   ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2550   ElfW(Addr) entry_point = args.getauxval(AT_ENTRY);
2551   ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2552   ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2553 
2554   soinfo linker_so("[dynamic linker]", nullptr, 0);
2555 
2556   // If the linker is not acting as PT_INTERP entry_point is equal to
2557   // _start. Which means that the linker is running as an executable and
2558   // already linked by PT_INTERP.
2559   //
2560   // This happens when user tries to run 'adb shell /system/bin/linker'
2561   // see also https://code.google.com/p/android/issues/detail?id=63174
2562   if (reinterpret_cast<ElfW(Addr)>(&_start) == entry_point) {
2563     __libc_fatal("This is %s, the helper program for shared library executables.\n", args.argv[0]);
2564   }
2565 
2566   linker_so.base = linker_addr;
2567   linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2568   linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2569   linker_so.dynamic = nullptr;
2570   linker_so.phdr = phdr;
2571   linker_so.phnum = elf_hdr->e_phnum;
2572   linker_so.flags |= FLAG_LINKER;
2573 
2574   if (!(linker_so.PrelinkImage() && linker_so.LinkImage(nullptr))) {
2575     // It would be nice to print an error message, but if the linker
2576     // can't link itself, there's no guarantee that we'll be able to
2577     // call write() (because it involves a GOT reference). We may as
2578     // well try though...
2579     const char* msg = "CANNOT LINK EXECUTABLE: ";
2580     write(2, msg, strlen(msg));
2581     write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2582     write(2, "\n", 1);
2583     _exit(EXIT_FAILURE);
2584   }
2585 
2586   __libc_init_tls(args);
2587 
2588   // Initialize the linker's own global variables
2589   linker_so.CallConstructors();
2590 
2591   // Initialize static variables. Note that in order to
2592   // get correct libdl_info we need to call constructors
2593   // before get_libdl_info().
2594   solist = get_libdl_info();
2595   sonext = get_libdl_info();
2596 
2597   // We have successfully fixed our own relocations. It's safe to run
2598   // the main part of the linker now.
2599   args.abort_message_ptr = &g_abort_message;
2600   ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2601 
2602   protect_data(PROT_READ);
2603 
2604   // Return the address that the calling assembly stub should jump to.
2605   return start_address;
2606 }
2607