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 <sys/stat.h>
39#include <unistd.h>
40
41// Private C library headers.
42#include "private/bionic_tls.h"
43#include "private/KernelArgumentBlock.h"
44#include "private/ScopedPthreadMutexLocker.h"
45#include "private/ScopedFd.h"
46
47#include "linker.h"
48#include "linker_debug.h"
49#include "linker_environ.h"
50#include "linker_phdr.h"
51#include "linker_allocator.h"
52
53/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
54 *
55 * Do NOT use malloc() and friends or pthread_*() code here.
56 * Don't use printf() either; it's caused mysterious memory
57 * corruption in the past.
58 * The linker runs before we bring up libc and it's easiest
59 * to make sure it does not depend on any complex libc features
60 *
61 * open issues / todo:
62 *
63 * - cleaner error reporting
64 * - after linking, set as much stuff as possible to READONLY
65 *   and NOEXEC
66 */
67
68#if defined(__LP64__)
69#define SEARCH_NAME(x) x
70#else
71// Nvidia drivers are relying on the bug:
72// http://code.google.com/p/android/issues/detail?id=6670
73// so we continue to use base-name lookup for lp32
74static const char* get_base_name(const char* name) {
75  const char* bname = strrchr(name, '/');
76  return bname ? bname + 1 : name;
77}
78#define SEARCH_NAME(x) get_base_name(x)
79#endif
80
81static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
82static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
83
84static LinkerAllocator<soinfo> g_soinfo_allocator;
85static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
86
87static soinfo* solist;
88static soinfo* sonext;
89static soinfo* somain; /* main process, always the one after libdl_info */
90
91static const char* const kDefaultLdPaths[] = {
92#if defined(__LP64__)
93  "/vendor/lib64",
94  "/system/lib64",
95#else
96  "/vendor/lib",
97  "/system/lib",
98#endif
99  NULL
100};
101
102#define LDPATH_BUFSIZE (LDPATH_MAX*64)
103#define LDPATH_MAX 8
104
105#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
106#define LDPRELOAD_MAX 8
107
108static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
109static const char* g_ld_library_paths[LDPATH_MAX + 1];
110
111static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
112static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
113
114static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
115
116__LIBC_HIDDEN__ int g_ld_debug_verbosity;
117
118__LIBC_HIDDEN__ abort_msg_t* g_abort_message = NULL; // For debuggerd.
119
120enum RelocationKind {
121    kRelocAbsolute = 0,
122    kRelocRelative,
123    kRelocCopy,
124    kRelocSymbol,
125    kRelocMax
126};
127
128#if STATS
129struct linker_stats_t {
130    int count[kRelocMax];
131};
132
133static linker_stats_t linker_stats;
134
135static void count_relocation(RelocationKind kind) {
136    ++linker_stats.count[kind];
137}
138#else
139static void count_relocation(RelocationKind) {
140}
141#endif
142
143#if COUNT_PAGES
144static unsigned bitmask[4096];
145#if defined(__LP64__)
146#define MARK(offset) \
147    do { \
148        if ((((offset) >> 12) >> 5) < 4096) \
149            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
150    } while (0)
151#else
152#define MARK(offset) \
153    do { \
154        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
155    } while (0)
156#endif
157#else
158#define MARK(x) do {} while (0)
159#endif
160
161// You shouldn't try to call memory-allocating functions in the dynamic linker.
162// Guard against the most obvious ones.
163#define DISALLOW_ALLOCATION(return_type, name, ...) \
164    return_type name __VA_ARGS__ \
165    { \
166        __libc_fatal("ERROR: " #name " called from the dynamic linker!\n"); \
167    }
168DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
169DISALLOW_ALLOCATION(void, free, (void* u __unused));
170DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
171DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
172
173static char tmp_err_buf[768];
174static char __linker_dl_err_buf[768];
175
176char* linker_get_error_buffer() {
177  return &__linker_dl_err_buf[0];
178}
179
180size_t linker_get_error_buffer_size() {
181  return sizeof(__linker_dl_err_buf);
182}
183
184/*
185 * This function is an empty stub where GDB locates a breakpoint to get notified
186 * about linker activity.
187 */
188extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
189
190static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
191static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
192static link_map* r_debug_tail = 0;
193
194static void insert_soinfo_into_debug_map(soinfo* info) {
195    // Copy the necessary fields into the debug structure.
196    link_map* map = &(info->link_map_head);
197    map->l_addr = info->load_bias;
198    map->l_name = reinterpret_cast<char*>(info->name);
199    map->l_ld = info->dynamic;
200
201    /* Stick the new library at the end of the list.
202     * gdb tends to care more about libc than it does
203     * about leaf libraries, and ordering it this way
204     * reduces the back-and-forth over the wire.
205     */
206    if (r_debug_tail) {
207        r_debug_tail->l_next = map;
208        map->l_prev = r_debug_tail;
209        map->l_next = 0;
210    } else {
211        _r_debug.r_map = map;
212        map->l_prev = 0;
213        map->l_next = 0;
214    }
215    r_debug_tail = map;
216}
217
218static void remove_soinfo_from_debug_map(soinfo* info) {
219    link_map* map = &(info->link_map_head);
220
221    if (r_debug_tail == map) {
222        r_debug_tail = map->l_prev;
223    }
224
225    if (map->l_prev) {
226        map->l_prev->l_next = map->l_next;
227    }
228    if (map->l_next) {
229        map->l_next->l_prev = map->l_prev;
230    }
231}
232
233static void notify_gdb_of_load(soinfo* info) {
234    if (info->flags & FLAG_EXE) {
235        // GDB already knows about the main executable
236        return;
237    }
238
239    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
240
241    _r_debug.r_state = r_debug::RT_ADD;
242    rtld_db_dlactivity();
243
244    insert_soinfo_into_debug_map(info);
245
246    _r_debug.r_state = r_debug::RT_CONSISTENT;
247    rtld_db_dlactivity();
248}
249
250static void notify_gdb_of_unload(soinfo* info) {
251    if (info->flags & FLAG_EXE) {
252        // GDB already knows about the main executable
253        return;
254    }
255
256    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
257
258    _r_debug.r_state = r_debug::RT_DELETE;
259    rtld_db_dlactivity();
260
261    remove_soinfo_from_debug_map(info);
262
263    _r_debug.r_state = r_debug::RT_CONSISTENT;
264    rtld_db_dlactivity();
265}
266
267void notify_gdb_of_libraries() {
268  _r_debug.r_state = r_debug::RT_ADD;
269  rtld_db_dlactivity();
270  _r_debug.r_state = r_debug::RT_CONSISTENT;
271  rtld_db_dlactivity();
272}
273
274LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
275  return g_soinfo_links_allocator.alloc();
276}
277
278void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
279  g_soinfo_links_allocator.free(entry);
280}
281
282static void protect_data(int protection) {
283  g_soinfo_allocator.protect_all(protection);
284  g_soinfo_links_allocator.protect_all(protection);
285}
286
287static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
288  if (strlen(name) >= SOINFO_NAME_LEN) {
289    DL_ERR("library name \"%s\" too long", name);
290    return NULL;
291  }
292
293  soinfo* si = g_soinfo_allocator.alloc();
294
295  // Initialize the new element.
296  memset(si, 0, sizeof(soinfo));
297  strlcpy(si->name, name, sizeof(si->name));
298  si->flags = FLAG_NEW_SOINFO;
299
300  if (file_stat != NULL) {
301    si->set_st_dev(file_stat->st_dev);
302    si->set_st_ino(file_stat->st_ino);
303  }
304
305  sonext->next = si;
306  sonext = si;
307
308  TRACE("name %s: allocated soinfo @ %p", name, si);
309  return si;
310}
311
312static void soinfo_free(soinfo* si) {
313    if (si == NULL) {
314        return;
315    }
316
317    if (si->base != 0 && si->size != 0) {
318      munmap(reinterpret_cast<void*>(si->base), si->size);
319    }
320
321    soinfo *prev = NULL, *trav;
322
323    TRACE("name %s: freeing soinfo @ %p", si->name, si);
324
325    for (trav = solist; trav != NULL; trav = trav->next) {
326        if (trav == si)
327            break;
328        prev = trav;
329    }
330    if (trav == NULL) {
331        /* si was not in solist */
332        DL_ERR("name \"%s\" is not in solist!", si->name);
333        return;
334    }
335
336    // clear links to/from si
337    si->remove_all_links();
338
339    /* prev will never be NULL, because the first entry in solist is
340       always the static libdl_info.
341    */
342    prev->next = si->next;
343    if (si == sonext) {
344        sonext = prev;
345    }
346
347    g_soinfo_allocator.free(si);
348}
349
350
351static void parse_path(const char* path, const char* delimiters,
352                       const char** array, char* buf, size_t buf_size, size_t max_count) {
353  if (path == NULL) {
354    return;
355  }
356
357  size_t len = strlcpy(buf, path, buf_size);
358
359  size_t i = 0;
360  char* buf_p = buf;
361  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
362    if (*array[i] != '\0') {
363      ++i;
364    }
365  }
366
367  // Forget the last path if we had to truncate; this occurs if the 2nd to
368  // last char isn't '\0' (i.e. wasn't originally a delimiter).
369  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
370    array[i - 1] = NULL;
371  } else {
372    array[i] = NULL;
373  }
374}
375
376static void parse_LD_LIBRARY_PATH(const char* path) {
377  parse_path(path, ":", g_ld_library_paths,
378             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
379}
380
381static void parse_LD_PRELOAD(const char* path) {
382  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
383  parse_path(path, " :", g_ld_preload_names,
384             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
385}
386
387#if defined(__arm__)
388
389/* For a given PC, find the .so that it belongs to.
390 * Returns the base address of the .ARM.exidx section
391 * for that .so, and the number of 8-byte entries
392 * in that section (via *pcount).
393 *
394 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
395 *
396 * This function is exposed via dlfcn.cpp and libdl.so.
397 */
398_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
399    unsigned addr = (unsigned)pc;
400
401    for (soinfo* si = solist; si != 0; si = si->next) {
402        if ((addr >= si->base) && (addr < (si->base + si->size))) {
403            *pcount = si->ARM_exidx_count;
404            return (_Unwind_Ptr)si->ARM_exidx;
405        }
406    }
407    *pcount = 0;
408    return NULL;
409}
410
411#endif
412
413/* Here, we only have to provide a callback to iterate across all the
414 * loaded libraries. gcc_eh does the rest. */
415int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
416    int rv = 0;
417    for (soinfo* si = solist; si != NULL; si = si->next) {
418        dl_phdr_info dl_info;
419        dl_info.dlpi_addr = si->link_map_head.l_addr;
420        dl_info.dlpi_name = si->link_map_head.l_name;
421        dl_info.dlpi_phdr = si->phdr;
422        dl_info.dlpi_phnum = si->phnum;
423        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
424        if (rv != 0) {
425            break;
426        }
427    }
428    return rv;
429}
430
431static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
432  ElfW(Sym)* symtab = si->symtab;
433  const char* strtab = si->strtab;
434
435  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
436             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
437
438  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
439    ElfW(Sym)* s = symtab + n;
440    if (strcmp(strtab + s->st_name, name)) continue;
441
442    /* only concern ourselves with global and weak symbol definitions */
443    switch (ELF_ST_BIND(s->st_info)) {
444      case STB_GLOBAL:
445      case STB_WEAK:
446        if (s->st_shndx == SHN_UNDEF) {
447          continue;
448        }
449
450        TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
451                 name, si->name, reinterpret_cast<void*>(s->st_value),
452                 static_cast<size_t>(s->st_size));
453        return s;
454      case STB_LOCAL:
455        continue;
456      default:
457        __libc_fatal("ERROR: Unexpected ST_BIND value: %d for '%s' in '%s'",
458            ELF_ST_BIND(s->st_info), name, si->name);
459    }
460  }
461
462  TRACE_TYPE(LOOKUP, "NOT FOUND %s in %s@%p %x %zd",
463             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
464
465
466  return NULL;
467}
468
469static 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
482static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
483    unsigned elf_hash = elfhash(name);
484    ElfW(Sym)* s = NULL;
485
486    if (si != NULL && somain != NULL) {
487        /*
488         * Local scope is executable scope. Just start looking into it right away
489         * for the shortcut.
490         */
491
492        if (si == somain) {
493            s = soinfo_elf_lookup(si, elf_hash, name);
494            if (s != NULL) {
495                *lsi = si;
496                goto done;
497            }
498
499            /* Next, look for it in the preloads list */
500            for (int i = 0; g_ld_preloads[i] != NULL; i++) {
501                s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
502                if (s != NULL) {
503                    *lsi = g_ld_preloads[i];
504                    goto done;
505                }
506            }
507        } else {
508            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
509
510            /*
511             * If this object was built with symbolic relocations disabled, the
512             * first place to look to resolve external references is the main
513             * executable.
514             */
515
516            if (!si->has_DT_SYMBOLIC) {
517                DEBUG("%s: looking up %s in executable %s",
518                      si->name, name, somain->name);
519                s = soinfo_elf_lookup(somain, elf_hash, name);
520                if (s != NULL) {
521                    *lsi = somain;
522                    goto done;
523                }
524
525                /* Next, look for it in the preloads list */
526                for (int i = 0; g_ld_preloads[i] != NULL; i++) {
527                    s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
528                    if (s != NULL) {
529                        *lsi = g_ld_preloads[i];
530                        goto done;
531                    }
532                }
533            }
534
535            /* Look for symbols in the local scope (the object who is
536             * searching). This happens with C++ templates on x86 for some
537             * reason.
538             *
539             * Notes on weak symbols:
540             * The ELF specs are ambiguous about treatment of weak definitions in
541             * dynamic linking.  Some systems return the first definition found
542             * and some the first non-weak definition.   This is system dependent.
543             * Here we return the first definition found for simplicity.  */
544
545            s = soinfo_elf_lookup(si, elf_hash, name);
546            if (s != NULL) {
547                *lsi = si;
548                goto done;
549            }
550
551            /*
552             * If this object was built with -Bsymbolic and symbol is not found
553             * in the local scope, try to find the symbol in the main executable.
554             */
555
556            if (si->has_DT_SYMBOLIC) {
557                DEBUG("%s: looking up %s in executable %s after local scope",
558                      si->name, name, somain->name);
559                s = soinfo_elf_lookup(somain, elf_hash, name);
560                if (s != NULL) {
561                    *lsi = somain;
562                    goto done;
563                }
564
565                /* Next, look for it in the preloads list */
566                for (int i = 0; g_ld_preloads[i] != NULL; i++) {
567                    s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
568                    if (s != NULL) {
569                        *lsi = g_ld_preloads[i];
570                        goto done;
571                    }
572                }
573            }
574        }
575    }
576
577    for (int i = 0; needed[i] != NULL; i++) {
578        DEBUG("%s: looking up %s in %s",
579              si->name, name, needed[i]->name);
580        s = soinfo_elf_lookup(needed[i], elf_hash, name);
581        if (s != NULL) {
582            *lsi = needed[i];
583            goto done;
584        }
585    }
586
587done:
588    if (s != NULL) {
589        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
590                   "found in %s, base = %p, load bias = %p",
591                   si->name, name, reinterpret_cast<void*>(s->st_value),
592                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
593                   reinterpret_cast<void*>((*lsi)->load_bias));
594        return s;
595    }
596
597    return NULL;
598}
599
600// Another soinfo list allocator to use in dlsym. We don't reuse
601// SoinfoListAllocator because it is write-protected most of the time.
602static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_list_allocator_rw;
603class SoinfoListAllocatorRW {
604 public:
605  static LinkedListEntry<soinfo>* alloc() {
606    return g_soinfo_list_allocator_rw.alloc();
607  }
608
609  static void free(LinkedListEntry<soinfo>* ptr) {
610    g_soinfo_list_allocator_rw.free(ptr);
611  }
612};
613
614// This is used by dlsym(3).  It performs symbol lookup only within the
615// specified soinfo object and its dependencies in breadth first order.
616ElfW(Sym)* dlsym_handle_lookup(soinfo* si, soinfo** found, const char* name) {
617  LinkedList<soinfo, SoinfoListAllocatorRW> visit_list;
618  LinkedList<soinfo, SoinfoListAllocatorRW> visited;
619  visit_list.push_back(si);
620  soinfo* current_soinfo;
621  while ((current_soinfo = visit_list.pop_front()) != nullptr) {
622    if (visited.contains(current_soinfo)) {
623      continue;
624    }
625
626    ElfW(Sym)* result = soinfo_elf_lookup(current_soinfo, elfhash(name), name);
627
628    if (result != nullptr) {
629      *found = current_soinfo;
630      visit_list.clear();
631      visited.clear();
632      return result;
633    }
634    visited.push_back(current_soinfo);
635
636    current_soinfo->get_children().for_each([&](soinfo* child) {
637      visit_list.push_back(child);
638    });
639  }
640
641  visit_list.clear();
642  visited.clear();
643  return nullptr;
644}
645
646/* This is used by dlsym(3) to performs a global symbol lookup. If the
647   start value is null (for RTLD_DEFAULT), the search starts at the
648   beginning of the global solist. Otherwise the search starts at the
649   specified soinfo (for RTLD_NEXT).
650 */
651ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
652  unsigned elf_hash = elfhash(name);
653
654  if (start == NULL) {
655    start = solist;
656  }
657
658  ElfW(Sym)* s = NULL;
659  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
660    s = soinfo_elf_lookup(si, elf_hash, name);
661    if (s != NULL) {
662      *found = si;
663      break;
664    }
665  }
666
667  if (s != NULL) {
668    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
669               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
670  }
671
672  return s;
673}
674
675soinfo* find_containing_library(const void* p) {
676  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
677  for (soinfo* si = solist; si != NULL; si = si->next) {
678    if (address >= si->base && address - si->base < si->size) {
679      return si;
680    }
681  }
682  return NULL;
683}
684
685ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
686  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
687
688  // Search the library's symbol table for any defined symbol which
689  // contains this address.
690  for (size_t i = 0; i < si->nchain; ++i) {
691    ElfW(Sym)* sym = &si->symtab[i];
692    if (sym->st_shndx != SHN_UNDEF &&
693        soaddr >= sym->st_value &&
694        soaddr < sym->st_value + sym->st_size) {
695      return sym;
696    }
697  }
698
699  return NULL;
700}
701
702static int open_library_on_path(const char* name, const char* const paths[]) {
703  char buf[512];
704  for (size_t i = 0; paths[i] != NULL; ++i) {
705    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
706    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
707      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
708      continue;
709    }
710    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
711    if (fd != -1) {
712      return fd;
713    }
714  }
715  return -1;
716}
717
718static int open_library(const char* name) {
719  TRACE("[ opening %s ]", name);
720
721  // If the name contains a slash, we should attempt to open it directly and not search the paths.
722  if (strchr(name, '/') != NULL) {
723    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
724    if (fd != -1) {
725      return fd;
726    }
727    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
728#if defined(__LP64__)
729    return -1;
730#endif
731  }
732
733  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
734  int fd = open_library_on_path(name, g_ld_library_paths);
735  if (fd == -1) {
736    fd = open_library_on_path(name, kDefaultLdPaths);
737  }
738  return fd;
739}
740
741static soinfo* load_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
742    int fd = -1;
743    ScopedFd file_guard(-1);
744
745    if (extinfo != NULL && (extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD) != 0) {
746      fd = extinfo->library_fd;
747    } else {
748      // Open the file.
749      fd = open_library(name);
750      if (fd == -1) {
751        DL_ERR("library \"%s\" not found", name);
752        return NULL;
753      }
754
755      file_guard.reset(fd);
756    }
757
758    ElfReader elf_reader(name, fd);
759
760    struct stat file_stat;
761    if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
762      DL_ERR("unable to stat file for the library %s: %s", name, strerror(errno));
763      return NULL;
764    }
765
766    // Check for symlink and other situations where
767    // file can have different names.
768    for (soinfo* si = solist; si != NULL; si = si->next) {
769      if (si->get_st_dev() != 0 &&
770          si->get_st_ino() != 0 &&
771          si->get_st_dev() == file_stat.st_dev &&
772          si->get_st_ino() == file_stat.st_ino) {
773        TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
774        return si;
775      }
776    }
777
778    if ((dlflags & RTLD_NOLOAD) != 0) {
779      return NULL;
780    }
781
782    // Read the ELF header and load the segments.
783    if (!elf_reader.Load(extinfo)) {
784        return NULL;
785    }
786
787    soinfo* si = soinfo_alloc(SEARCH_NAME(name), &file_stat);
788    if (si == NULL) {
789        return NULL;
790    }
791    si->base = elf_reader.load_start();
792    si->size = elf_reader.load_size();
793    si->load_bias = elf_reader.load_bias();
794    si->phnum = elf_reader.phdr_count();
795    si->phdr = elf_reader.loaded_phdr();
796
797    // At this point we know that whatever is loaded @ base is a valid ELF
798    // shared library whose segments are properly mapped in.
799    TRACE("[ load_library base=%p size=%zu name='%s' ]",
800          reinterpret_cast<void*>(si->base), si->size, si->name);
801
802    if (!soinfo_link_image(si, extinfo)) {
803      soinfo_free(si);
804      return NULL;
805    }
806
807    return si;
808}
809
810static soinfo *find_loaded_library_by_name(const char* name) {
811  const char* search_name = SEARCH_NAME(name);
812  for (soinfo* si = solist; si != NULL; si = si->next) {
813    if (!strcmp(search_name, si->name)) {
814      return si;
815    }
816  }
817  return NULL;
818}
819
820static soinfo* find_library_internal(const char* name, int dlflags, const android_dlextinfo* extinfo) {
821  if (name == NULL) {
822    return somain;
823  }
824
825  soinfo* si = find_loaded_library_by_name(name);
826
827  // Library might still be loaded, the accurate detection
828  // of this fact is done by load_library
829  if (si == NULL) {
830    TRACE("[ '%s' has not been found by name.  Trying harder...]", name);
831    si = load_library(name, dlflags, extinfo);
832  }
833
834  if (si != NULL && (si->flags & FLAG_LINKED) == 0) {
835    DL_ERR("recursive link to \"%s\"", si->name);
836    return NULL;
837  }
838
839  return si;
840}
841
842static soinfo* find_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
843  soinfo* si = find_library_internal(name, dlflags, extinfo);
844  if (si != NULL) {
845    si->ref_count++;
846  }
847  return si;
848}
849
850static void soinfo_unload(soinfo* si) {
851  if (si->ref_count == 1) {
852    TRACE("unloading '%s'", si->name);
853    si->CallDestructors();
854
855    if ((si->flags | FLAG_NEW_SOINFO) != 0) {
856      si->get_children().for_each([&] (soinfo* child) {
857        TRACE("%s needs to unload %s", si->name, child->name);
858        soinfo_unload(child);
859      });
860    } else {
861      for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
862        if (d->d_tag == DT_NEEDED) {
863          const char* library_name = si->strtab + d->d_un.d_val;
864          TRACE("%s needs to unload %s", si->name, library_name);
865          soinfo* needed = find_library(library_name, RTLD_NOLOAD, NULL);
866          if (needed != NULL) {
867            soinfo_unload(needed);
868          } else {
869            // Not found: for example if symlink was deleted between dlopen and dlclose
870            // Since we cannot really handle errors at this point - print and continue.
871            PRINT("warning: couldn't find %s needed by %s on unload.", library_name, si->name);
872          }
873        }
874      }
875    }
876
877    notify_gdb_of_unload(si);
878    si->ref_count = 0;
879    soinfo_free(si);
880  } else {
881    si->ref_count--;
882    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
883  }
884}
885
886void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
887  // Use basic string manipulation calls to avoid snprintf.
888  // snprintf indirectly calls pthread_getspecific to get the size of a buffer.
889  // When debug malloc is enabled, this call returns 0. This in turn causes
890  // snprintf to do nothing, which causes libraries to fail to load.
891  // See b/17302493 for further details.
892  // Once the above bug is fixed, this code can be modified to use
893  // snprintf again.
894  size_t required_len = strlen(kDefaultLdPaths[0]) + strlen(kDefaultLdPaths[1]) + 2;
895  if (buffer_size < required_len) {
896    __libc_fatal("android_get_LD_LIBRARY_PATH failed, buffer too small: buffer len %zu, required len %zu",
897                 buffer_size, required_len);
898  }
899  char* end = stpcpy(buffer, kDefaultLdPaths[0]);
900  *end = ':';
901  strcpy(end + 1, kDefaultLdPaths[1]);
902}
903
904void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
905  if (!get_AT_SECURE()) {
906    parse_LD_LIBRARY_PATH(ld_library_path);
907  }
908}
909
910soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
911  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL|RTLD_NOLOAD)) != 0) {
912    DL_ERR("invalid flags to dlopen: %x", flags);
913    return NULL;
914  }
915  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
916    DL_ERR("invalid extended flags to android_dlopen_ext: %" PRIx64, extinfo->flags);
917    return NULL;
918  }
919  protect_data(PROT_READ | PROT_WRITE);
920  soinfo* si = find_library(name, flags, extinfo);
921  if (si != NULL) {
922    si->CallConstructors();
923  }
924  protect_data(PROT_READ);
925  return si;
926}
927
928void do_dlclose(soinfo* si) {
929  protect_data(PROT_READ | PROT_WRITE);
930  soinfo_unload(si);
931  protect_data(PROT_READ);
932}
933
934#if defined(USE_RELA)
935static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
936  ElfW(Sym)* s;
937  soinfo* lsi;
938
939  for (size_t idx = 0; idx < count; ++idx, ++rela) {
940    unsigned type = ELFW(R_TYPE)(rela->r_info);
941    unsigned sym = ELFW(R_SYM)(rela->r_info);
942    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
943    ElfW(Addr) sym_addr = 0;
944    const char* sym_name = NULL;
945
946    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
947    if (type == 0) { // R_*_NONE
948      continue;
949    }
950    if (sym != 0) {
951      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
952      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
953      if (s == NULL) {
954        // We only allow an undefined symbol if this is a weak reference...
955        s = &si->symtab[sym];
956        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
957          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
958          return -1;
959        }
960
961        /* IHI0044C AAELF 4.5.1.1:
962
963           Libraries are not searched to resolve weak references.
964           It is not an error for a weak reference to remain unsatisfied.
965
966           During linking, the value of an undefined weak reference is:
967           - Zero if the relocation type is absolute
968           - The address of the place if the relocation is pc-relative
969           - The address of nominal base address if the relocation
970             type is base-relative.
971         */
972
973        switch (type) {
974#if defined(__aarch64__)
975        case R_AARCH64_JUMP_SLOT:
976        case R_AARCH64_GLOB_DAT:
977        case R_AARCH64_ABS64:
978        case R_AARCH64_ABS32:
979        case R_AARCH64_ABS16:
980        case R_AARCH64_RELATIVE:
981          /*
982           * The sym_addr was initialized to be zero above, or the relocation
983           * code below does not care about value of sym_addr.
984           * No need to do anything.
985           */
986          break;
987#elif defined(__x86_64__)
988        case R_X86_64_JUMP_SLOT:
989        case R_X86_64_GLOB_DAT:
990        case R_X86_64_32:
991        case R_X86_64_64:
992        case R_X86_64_RELATIVE:
993          // No need to do anything.
994          break;
995        case R_X86_64_PC32:
996          sym_addr = reloc;
997          break;
998#endif
999        default:
1000          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
1001          return -1;
1002        }
1003      } else {
1004        // We got a definition.
1005        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1006      }
1007      count_relocation(kRelocSymbol);
1008    } else {
1009      s = NULL;
1010    }
1011
1012    switch (type) {
1013#if defined(__aarch64__)
1014    case R_AARCH64_JUMP_SLOT:
1015        count_relocation(kRelocAbsolute);
1016        MARK(rela->r_offset);
1017        TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
1018                   reloc, (sym_addr + rela->r_addend), sym_name);
1019        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
1020        break;
1021    case R_AARCH64_GLOB_DAT:
1022        count_relocation(kRelocAbsolute);
1023        MARK(rela->r_offset);
1024        TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
1025                   reloc, (sym_addr + rela->r_addend), sym_name);
1026        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
1027        break;
1028    case R_AARCH64_ABS64:
1029        count_relocation(kRelocAbsolute);
1030        MARK(rela->r_offset);
1031        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
1032                   reloc, (sym_addr + rela->r_addend), sym_name);
1033        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1034        break;
1035    case R_AARCH64_ABS32:
1036        count_relocation(kRelocAbsolute);
1037        MARK(rela->r_offset);
1038        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
1039                   reloc, (sym_addr + rela->r_addend), sym_name);
1040        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1041            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1042            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1043        } else {
1044            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1045                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
1046                   static_cast<ElfW(Addr)>(INT32_MIN),
1047                   static_cast<ElfW(Addr)>(UINT32_MAX));
1048            return -1;
1049        }
1050        break;
1051    case R_AARCH64_ABS16:
1052        count_relocation(kRelocAbsolute);
1053        MARK(rela->r_offset);
1054        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
1055                   reloc, (sym_addr + rela->r_addend), sym_name);
1056        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1057            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1058            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1059        } else {
1060            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1061                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
1062                   static_cast<ElfW(Addr)>(INT16_MIN),
1063                   static_cast<ElfW(Addr)>(UINT16_MAX));
1064            return -1;
1065        }
1066        break;
1067    case R_AARCH64_PREL64:
1068        count_relocation(kRelocRelative);
1069        MARK(rela->r_offset);
1070        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
1071                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1072        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
1073        break;
1074    case R_AARCH64_PREL32:
1075        count_relocation(kRelocRelative);
1076        MARK(rela->r_offset);
1077        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
1078                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1079        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1080            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1081            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1082        } else {
1083            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1084                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1085                   static_cast<ElfW(Addr)>(INT32_MIN),
1086                   static_cast<ElfW(Addr)>(UINT32_MAX));
1087            return -1;
1088        }
1089        break;
1090    case R_AARCH64_PREL16:
1091        count_relocation(kRelocRelative);
1092        MARK(rela->r_offset);
1093        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1094                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1095        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1096            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1097            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1098        } else {
1099            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1100                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1101                   static_cast<ElfW(Addr)>(INT16_MIN),
1102                   static_cast<ElfW(Addr)>(UINT16_MAX));
1103            return -1;
1104        }
1105        break;
1106
1107    case R_AARCH64_RELATIVE:
1108        count_relocation(kRelocRelative);
1109        MARK(rela->r_offset);
1110        if (sym) {
1111            DL_ERR("odd RELATIVE form...");
1112            return -1;
1113        }
1114        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1115                   reloc, (si->base + rela->r_addend));
1116        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1117        break;
1118
1119    case R_AARCH64_COPY:
1120        /*
1121         * ET_EXEC is not supported so this should not happen.
1122         *
1123         * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1124         *
1125         * Section 4.7.1.10 "Dynamic relocations"
1126         * R_AARCH64_COPY may only appear in executable objects where e_type is
1127         * set to ET_EXEC.
1128         */
1129        DL_ERR("%s R_AARCH64_COPY relocations are not supported", si->name);
1130        return -1;
1131    case R_AARCH64_TLS_TPREL64:
1132        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1133                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1134        break;
1135    case R_AARCH64_TLS_DTPREL32:
1136        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1137                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1138        break;
1139#elif defined(__x86_64__)
1140    case R_X86_64_JUMP_SLOT:
1141      count_relocation(kRelocAbsolute);
1142      MARK(rela->r_offset);
1143      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1144                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1145      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1146      break;
1147    case R_X86_64_GLOB_DAT:
1148      count_relocation(kRelocAbsolute);
1149      MARK(rela->r_offset);
1150      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1151                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1152      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1153      break;
1154    case R_X86_64_RELATIVE:
1155      count_relocation(kRelocRelative);
1156      MARK(rela->r_offset);
1157      if (sym) {
1158        DL_ERR("odd RELATIVE form...");
1159        return -1;
1160      }
1161      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1162                 static_cast<size_t>(si->base));
1163      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1164      break;
1165    case R_X86_64_32:
1166      count_relocation(kRelocRelative);
1167      MARK(rela->r_offset);
1168      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1169                 static_cast<size_t>(sym_addr), sym_name);
1170      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1171      break;
1172    case R_X86_64_64:
1173      count_relocation(kRelocRelative);
1174      MARK(rela->r_offset);
1175      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1176                 static_cast<size_t>(sym_addr), sym_name);
1177      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1178      break;
1179    case R_X86_64_PC32:
1180      count_relocation(kRelocRelative);
1181      MARK(rela->r_offset);
1182      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1183                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1184                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1185      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1186      break;
1187#endif
1188
1189    default:
1190      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1191      return -1;
1192    }
1193  }
1194  return 0;
1195}
1196
1197#else // REL, not RELA.
1198
1199static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1200    ElfW(Sym)* s;
1201    soinfo* lsi;
1202
1203    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1204        unsigned type = ELFW(R_TYPE)(rel->r_info);
1205        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1206        unsigned sym = ELFW(R_SYM)(rel->r_info);
1207        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1208        ElfW(Addr) sym_addr = 0;
1209        const char* sym_name = NULL;
1210
1211        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1212        if (type == 0) { // R_*_NONE
1213            continue;
1214        }
1215        if (sym != 0) {
1216            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1217            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1218            if (s == NULL) {
1219                // We only allow an undefined symbol if this is a weak reference...
1220                s = &si->symtab[sym];
1221                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1222                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1223                    return -1;
1224                }
1225
1226                /* IHI0044C AAELF 4.5.1.1:
1227
1228                   Libraries are not searched to resolve weak references.
1229                   It is not an error for a weak reference to remain
1230                   unsatisfied.
1231
1232                   During linking, the value of an undefined weak reference is:
1233                   - Zero if the relocation type is absolute
1234                   - The address of the place if the relocation is pc-relative
1235                   - The address of nominal base address if the relocation
1236                     type is base-relative.
1237                  */
1238
1239                switch (type) {
1240#if defined(__arm__)
1241                case R_ARM_JUMP_SLOT:
1242                case R_ARM_GLOB_DAT:
1243                case R_ARM_ABS32:
1244                case R_ARM_RELATIVE:    /* Don't care. */
1245                    // sym_addr was initialized to be zero above or relocation
1246                    // code below does not care about value of sym_addr.
1247                    // No need to do anything.
1248                    break;
1249#elif defined(__i386__)
1250                case R_386_JMP_SLOT:
1251                case R_386_GLOB_DAT:
1252                case R_386_32:
1253                case R_386_RELATIVE:    /* Don't care. */
1254                    // sym_addr was initialized to be zero above or relocation
1255                    // code below does not care about value of sym_addr.
1256                    // No need to do anything.
1257                    break;
1258                case R_386_PC32:
1259                    sym_addr = reloc;
1260                    break;
1261#endif
1262
1263#if defined(__arm__)
1264                case R_ARM_COPY:
1265                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1266#endif
1267                default:
1268                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1269                    return -1;
1270                }
1271            } else {
1272                // We got a definition.
1273                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1274            }
1275            count_relocation(kRelocSymbol);
1276        } else {
1277            s = NULL;
1278        }
1279
1280        switch (type) {
1281#if defined(__arm__)
1282        case R_ARM_JUMP_SLOT:
1283            count_relocation(kRelocAbsolute);
1284            MARK(rel->r_offset);
1285            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1286            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1287            break;
1288        case R_ARM_GLOB_DAT:
1289            count_relocation(kRelocAbsolute);
1290            MARK(rel->r_offset);
1291            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1292            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1293            break;
1294        case R_ARM_ABS32:
1295            count_relocation(kRelocAbsolute);
1296            MARK(rel->r_offset);
1297            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1298            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1299            break;
1300        case R_ARM_REL32:
1301            count_relocation(kRelocRelative);
1302            MARK(rel->r_offset);
1303            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1304                       reloc, sym_addr, rel->r_offset, sym_name);
1305            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1306            break;
1307        case R_ARM_COPY:
1308            /*
1309             * ET_EXEC is not supported so this should not happen.
1310             *
1311             * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1312             *
1313             * Section 4.7.1.10 "Dynamic relocations"
1314             * R_ARM_COPY may only appear in executable objects where e_type is
1315             * set to ET_EXEC.
1316             */
1317            DL_ERR("%s R_ARM_COPY relocations are not supported", si->name);
1318            return -1;
1319#elif defined(__i386__)
1320        case R_386_JMP_SLOT:
1321            count_relocation(kRelocAbsolute);
1322            MARK(rel->r_offset);
1323            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1324            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1325            break;
1326        case R_386_GLOB_DAT:
1327            count_relocation(kRelocAbsolute);
1328            MARK(rel->r_offset);
1329            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1330            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1331            break;
1332        case R_386_32:
1333            count_relocation(kRelocRelative);
1334            MARK(rel->r_offset);
1335            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1336            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1337            break;
1338        case R_386_PC32:
1339            count_relocation(kRelocRelative);
1340            MARK(rel->r_offset);
1341            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1342                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1343            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1344            break;
1345#elif defined(__mips__)
1346        case R_MIPS_REL32:
1347#if defined(__LP64__)
1348            // MIPS Elf64_Rel entries contain compound relocations
1349            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1350            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1351                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1352                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1353                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1354                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1355                return -1;
1356            }
1357#endif
1358            count_relocation(kRelocAbsolute);
1359            MARK(rel->r_offset);
1360            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1361                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1362            if (s) {
1363                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1364            } else {
1365                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1366            }
1367            break;
1368#endif
1369
1370#if defined(__arm__)
1371        case R_ARM_RELATIVE:
1372#elif defined(__i386__)
1373        case R_386_RELATIVE:
1374#endif
1375            count_relocation(kRelocRelative);
1376            MARK(rel->r_offset);
1377            if (sym) {
1378                DL_ERR("odd RELATIVE form...");
1379                return -1;
1380            }
1381            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1382                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1383            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1384            break;
1385
1386        default:
1387            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1388            return -1;
1389        }
1390    }
1391    return 0;
1392}
1393#endif
1394
1395#if defined(__mips__)
1396static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1397    ElfW(Addr)** got = si->plt_got;
1398    if (got == NULL) {
1399        return true;
1400    }
1401    unsigned local_gotno = si->mips_local_gotno;
1402    unsigned gotsym = si->mips_gotsym;
1403    unsigned symtabno = si->mips_symtabno;
1404    ElfW(Sym)* symtab = si->symtab;
1405
1406    // got[0] is the address of the lazy resolver function.
1407    // got[1] may be used for a GNU extension.
1408    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1409    // FIXME: maybe this should be in a separate routine?
1410    if ((si->flags & FLAG_LINKER) == 0) {
1411        size_t g = 0;
1412        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1413        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1414            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1415        }
1416        // Relocate the local GOT entries.
1417        for (; g < local_gotno; g++) {
1418            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1419        }
1420    }
1421
1422    // Now for the global GOT entries...
1423    ElfW(Sym)* sym = symtab + gotsym;
1424    got = si->plt_got + local_gotno;
1425    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1426        // This is an undefined reference... try to locate it.
1427        const char* sym_name = si->strtab + sym->st_name;
1428        soinfo* lsi;
1429        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1430        if (s == NULL) {
1431            // We only allow an undefined symbol if this is a weak reference.
1432            s = &symtab[g];
1433            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1434                DL_ERR("cannot locate \"%s\"...", sym_name);
1435                return false;
1436            }
1437            *got = 0;
1438        } else {
1439            // FIXME: is this sufficient?
1440            // For reference see NetBSD link loader
1441            // 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
1442            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1443        }
1444    }
1445    return true;
1446}
1447#endif
1448
1449void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1450  if (functions == NULL) {
1451    return;
1452  }
1453
1454  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1455
1456  int begin = reverse ? (count - 1) : 0;
1457  int end = reverse ? -1 : count;
1458  int step = reverse ? -1 : 1;
1459
1460  for (int i = begin; i != end; i += step) {
1461    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1462    CallFunction("function", functions[i]);
1463  }
1464
1465  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1466}
1467
1468void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1469  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1470    return;
1471  }
1472
1473  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1474  function();
1475  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1476
1477  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1478  // are still writable. This happens with our debug malloc (see http://b/7941716).
1479  protect_data(PROT_READ | PROT_WRITE);
1480}
1481
1482void soinfo::CallPreInitConstructors() {
1483  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1484  // but ignored in a shared library.
1485  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1486}
1487
1488void soinfo::CallConstructors() {
1489  if (constructors_called) {
1490    return;
1491  }
1492
1493  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1494  // protect against recursive constructor calls. One simple example of constructor recursion
1495  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1496  // 1. The program depends on libc, so libc's constructor is called here.
1497  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1498  // 3. dlopen() calls the constructors on the newly created
1499  //    soinfo for libc_malloc_debug_leak.so.
1500  // 4. The debug .so depends on libc, so CallConstructors is
1501  //    called again with the libc soinfo. If it doesn't trigger the early-
1502  //    out above, the libc constructor will be called again (recursively!).
1503  constructors_called = true;
1504
1505  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1506    // The GNU dynamic linker silently ignores these, but we warn the developer.
1507    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1508          name, preinit_array_count);
1509  }
1510
1511  get_children().for_each([] (soinfo* si) {
1512    si->CallConstructors();
1513  });
1514
1515  TRACE("\"%s\": calling constructors", name);
1516
1517  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1518  CallFunction("DT_INIT", init_func);
1519  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1520}
1521
1522void soinfo::CallDestructors() {
1523  TRACE("\"%s\": calling destructors", name);
1524
1525  // DT_FINI_ARRAY must be parsed in reverse order.
1526  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1527
1528  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1529  CallFunction("DT_FINI", fini_func);
1530
1531  // This is needed on second call to dlopen
1532  // after library has been unloaded with RTLD_NODELETE
1533  constructors_called = false;
1534}
1535
1536void soinfo::add_child(soinfo* child) {
1537  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1538    return;
1539  }
1540
1541  this->children.push_front(child);
1542  child->parents.push_front(this);
1543}
1544
1545void soinfo::remove_all_links() {
1546  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1547    return;
1548  }
1549
1550  // 1. Untie connected soinfos from 'this'.
1551  children.for_each([&] (soinfo* child) {
1552    child->parents.remove_if([&] (const soinfo* parent) {
1553      return parent == this;
1554    });
1555  });
1556
1557  parents.for_each([&] (soinfo* parent) {
1558    parent->children.for_each([&] (const soinfo* child) {
1559      return child == this;
1560    });
1561  });
1562
1563  // 2. Once everything untied - clear local lists.
1564  parents.clear();
1565  children.clear();
1566}
1567
1568void soinfo::set_st_dev(dev_t dev) {
1569  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1570    return;
1571  }
1572
1573  st_dev = dev;
1574}
1575
1576void soinfo::set_st_ino(ino_t ino) {
1577  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1578    return;
1579  }
1580
1581  st_ino = ino;
1582}
1583
1584dev_t soinfo::get_st_dev() {
1585  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1586    return 0;
1587  }
1588
1589  return st_dev;
1590};
1591
1592ino_t soinfo::get_st_ino() {
1593  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1594    return 0;
1595  }
1596
1597  return st_ino;
1598}
1599
1600// This is a return on get_children() in case
1601// 'this->flags' does not have FLAG_NEW_SOINFO set.
1602static soinfo::soinfo_list_t g_empty_list;
1603
1604soinfo::soinfo_list_t& soinfo::get_children() {
1605  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1606    return g_empty_list;
1607  }
1608
1609  return this->children;
1610}
1611
1612/* Force any of the closed stdin, stdout and stderr to be associated with
1613   /dev/null. */
1614static int nullify_closed_stdio() {
1615    int dev_null, i, status;
1616    int return_value = 0;
1617
1618    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1619    if (dev_null < 0) {
1620        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1621        return -1;
1622    }
1623    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1624
1625    /* If any of the stdio file descriptors is valid and not associated
1626       with /dev/null, dup /dev/null to it.  */
1627    for (i = 0; i < 3; i++) {
1628        /* If it is /dev/null already, we are done. */
1629        if (i == dev_null) {
1630            continue;
1631        }
1632
1633        TRACE("[ Nullifying stdio file descriptor %d]", i);
1634        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1635
1636        /* If file is opened, we are good. */
1637        if (status != -1) {
1638            continue;
1639        }
1640
1641        /* The only error we allow is that the file descriptor does not
1642           exist, in which case we dup /dev/null to it. */
1643        if (errno != EBADF) {
1644            DL_ERR("fcntl failed: %s", strerror(errno));
1645            return_value = -1;
1646            continue;
1647        }
1648
1649        /* Try dupping /dev/null to this stdio file descriptor and
1650           repeat if there is a signal.  Note that any errors in closing
1651           the stdio descriptor are lost.  */
1652        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1653        if (status < 0) {
1654            DL_ERR("dup2 failed: %s", strerror(errno));
1655            return_value = -1;
1656            continue;
1657        }
1658    }
1659
1660    /* If /dev/null is not one of the stdio file descriptors, close it. */
1661    if (dev_null > 2) {
1662        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1663        status = TEMP_FAILURE_RETRY(close(dev_null));
1664        if (status == -1) {
1665            DL_ERR("close failed: %s", strerror(errno));
1666            return_value = -1;
1667        }
1668    }
1669
1670    return return_value;
1671}
1672
1673static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1674    /* "base" might wrap around UINT32_MAX. */
1675    ElfW(Addr) base = si->load_bias;
1676    const ElfW(Phdr)* phdr = si->phdr;
1677    int phnum = si->phnum;
1678    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1679
1680    /* We can't debug anything until the linker is relocated */
1681    if (!relocating_linker) {
1682        INFO("[ linking %s ]", si->name);
1683        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1684    }
1685
1686    /* Extract dynamic section */
1687    size_t dynamic_count;
1688    ElfW(Word) dynamic_flags;
1689    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1690                                   &dynamic_count, &dynamic_flags);
1691    if (si->dynamic == NULL) {
1692        if (!relocating_linker) {
1693            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1694        }
1695        return false;
1696    } else {
1697        if (!relocating_linker) {
1698            DEBUG("dynamic = %p", si->dynamic);
1699        }
1700    }
1701
1702#if defined(__arm__)
1703    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1704                                    &si->ARM_exidx, &si->ARM_exidx_count);
1705#endif
1706
1707    // Extract useful information from dynamic section.
1708    uint32_t needed_count = 0;
1709    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1710        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1711              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1712        switch (d->d_tag) {
1713        case DT_HASH:
1714            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1715            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1716            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1717            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1718            break;
1719        case DT_STRTAB:
1720            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1721            break;
1722        case DT_SYMTAB:
1723            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1724            break;
1725#if !defined(__LP64__)
1726        case DT_PLTREL:
1727            if (d->d_un.d_val != DT_REL) {
1728                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1729                return false;
1730            }
1731            break;
1732#endif
1733        case DT_JMPREL:
1734#if defined(USE_RELA)
1735            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1736#else
1737            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1738#endif
1739            break;
1740        case DT_PLTRELSZ:
1741#if defined(USE_RELA)
1742            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1743#else
1744            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1745#endif
1746            break;
1747#if defined(__mips__)
1748        case DT_PLTGOT:
1749            // Used by mips and mips64.
1750            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1751            break;
1752#endif
1753        case DT_DEBUG:
1754            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1755            // if the dynamic table is writable
1756// FIXME: not working currently for N64
1757// The flags for the LOAD and DYNAMIC program headers do not agree.
1758// The LOAD section containng the dynamic table has been mapped as
1759// read-only, but the DYNAMIC header claims it is writable.
1760#if !(defined(__mips__) && defined(__LP64__))
1761            if ((dynamic_flags & PF_W) != 0) {
1762                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1763            }
1764            break;
1765#endif
1766#if defined(USE_RELA)
1767         case DT_RELA:
1768            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1769            break;
1770         case DT_RELASZ:
1771            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1772            break;
1773        case DT_REL:
1774            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1775            return false;
1776        case DT_RELSZ:
1777            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1778            return false;
1779#else
1780        case DT_REL:
1781            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1782            break;
1783        case DT_RELSZ:
1784            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1785            break;
1786         case DT_RELA:
1787            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1788            return false;
1789#endif
1790        case DT_INIT:
1791            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1792            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1793            break;
1794        case DT_FINI:
1795            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1796            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1797            break;
1798        case DT_INIT_ARRAY:
1799            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1800            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1801            break;
1802        case DT_INIT_ARRAYSZ:
1803            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1804            break;
1805        case DT_FINI_ARRAY:
1806            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1807            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1808            break;
1809        case DT_FINI_ARRAYSZ:
1810            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1811            break;
1812        case DT_PREINIT_ARRAY:
1813            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1814            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1815            break;
1816        case DT_PREINIT_ARRAYSZ:
1817            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1818            break;
1819        case DT_TEXTREL:
1820#if defined(__LP64__)
1821            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1822            return false;
1823#else
1824            si->has_text_relocations = true;
1825            break;
1826#endif
1827        case DT_SYMBOLIC:
1828            si->has_DT_SYMBOLIC = true;
1829            break;
1830        case DT_NEEDED:
1831            ++needed_count;
1832            break;
1833        case DT_FLAGS:
1834            if (d->d_un.d_val & DF_TEXTREL) {
1835#if defined(__LP64__)
1836                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1837                return false;
1838#else
1839                si->has_text_relocations = true;
1840#endif
1841            }
1842            if (d->d_un.d_val & DF_SYMBOLIC) {
1843                si->has_DT_SYMBOLIC = true;
1844            }
1845            break;
1846#if defined(__mips__)
1847        case DT_STRSZ:
1848        case DT_SYMENT:
1849        case DT_RELENT:
1850             break;
1851        case DT_MIPS_RLD_MAP:
1852            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1853            {
1854              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1855              *dp = &_r_debug;
1856            }
1857            break;
1858        case DT_MIPS_RLD_VERSION:
1859        case DT_MIPS_FLAGS:
1860        case DT_MIPS_BASE_ADDRESS:
1861        case DT_MIPS_UNREFEXTNO:
1862            break;
1863
1864        case DT_MIPS_SYMTABNO:
1865            si->mips_symtabno = d->d_un.d_val;
1866            break;
1867
1868        case DT_MIPS_LOCAL_GOTNO:
1869            si->mips_local_gotno = d->d_un.d_val;
1870            break;
1871
1872        case DT_MIPS_GOTSYM:
1873            si->mips_gotsym = d->d_un.d_val;
1874            break;
1875#endif
1876
1877        default:
1878            DEBUG("Unused DT entry: type %p arg %p",
1879                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1880            break;
1881        }
1882    }
1883
1884    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1885          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1886
1887    // Sanity checks.
1888    if (relocating_linker && needed_count != 0) {
1889        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1890        return false;
1891    }
1892    if (si->nbucket == 0) {
1893        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1894        return false;
1895    }
1896    if (si->strtab == 0) {
1897        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1898        return false;
1899    }
1900    if (si->symtab == 0) {
1901        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1902        return false;
1903    }
1904
1905    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1906    if (si->flags & FLAG_EXE) {
1907        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1908        size_t preload_count = 0;
1909        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1910            soinfo* lsi = find_library(g_ld_preload_names[i], 0, NULL);
1911            if (lsi != NULL) {
1912                g_ld_preloads[preload_count++] = lsi;
1913            } else {
1914                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1915                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1916                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1917            }
1918        }
1919    }
1920
1921    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1922    soinfo** pneeded = needed;
1923
1924    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1925        if (d->d_tag == DT_NEEDED) {
1926            const char* library_name = si->strtab + d->d_un.d_val;
1927            DEBUG("%s needs %s", si->name, library_name);
1928            soinfo* lsi = find_library(library_name, 0, NULL);
1929            if (lsi == NULL) {
1930                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1931                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1932                       library_name, si->name, tmp_err_buf);
1933                return false;
1934            }
1935
1936            si->add_child(lsi);
1937            *pneeded++ = lsi;
1938        }
1939    }
1940    *pneeded = NULL;
1941
1942#if !defined(__LP64__)
1943    if (si->has_text_relocations) {
1944        // Make segments writable to allow text relocations to work properly. We will later call
1945        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1946        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1947                "security hardening. Please fix.", si->name);
1948        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1949            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1950                   si->name, strerror(errno));
1951            return false;
1952        }
1953    }
1954#endif
1955
1956#if defined(USE_RELA)
1957    if (si->plt_rela != NULL) {
1958        DEBUG("[ relocating %s plt ]\n", si->name);
1959        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1960            return false;
1961        }
1962    }
1963    if (si->rela != NULL) {
1964        DEBUG("[ relocating %s ]\n", si->name);
1965        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1966            return false;
1967        }
1968    }
1969#else
1970    if (si->plt_rel != NULL) {
1971        DEBUG("[ relocating %s plt ]", si->name);
1972        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1973            return false;
1974        }
1975    }
1976    if (si->rel != NULL) {
1977        DEBUG("[ relocating %s ]", si->name);
1978        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1979            return false;
1980        }
1981    }
1982#endif
1983
1984#if defined(__mips__)
1985    if (!mips_relocate_got(si, needed)) {
1986        return false;
1987    }
1988#endif
1989
1990    si->flags |= FLAG_LINKED;
1991    DEBUG("[ finished linking %s ]", si->name);
1992
1993#if !defined(__LP64__)
1994    if (si->has_text_relocations) {
1995        // All relocations are done, we can protect our segments back to read-only.
1996        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1997            DL_ERR("can't protect segments for \"%s\": %s",
1998                   si->name, strerror(errno));
1999            return false;
2000        }
2001    }
2002#endif
2003
2004    /* We can also turn on GNU RELRO protection */
2005    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
2006        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
2007               si->name, strerror(errno));
2008        return false;
2009    }
2010
2011    /* Handle serializing/sharing the RELRO segment */
2012    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
2013      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
2014                                         extinfo->relro_fd) < 0) {
2015        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
2016               si->name, strerror(errno));
2017        return false;
2018      }
2019    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
2020      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
2021                                   extinfo->relro_fd) < 0) {
2022        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
2023               si->name, strerror(errno));
2024        return false;
2025      }
2026    }
2027
2028    notify_gdb_of_load(si);
2029    return true;
2030}
2031
2032/*
2033 * This function add vdso to internal dso list.
2034 * It helps to stack unwinding through signal handlers.
2035 * Also, it makes bionic more like glibc.
2036 */
2037static void add_vdso(KernelArgumentBlock& args __unused) {
2038#if defined(AT_SYSINFO_EHDR)
2039  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
2040  if (ehdr_vdso == NULL) {
2041    return;
2042  }
2043
2044  soinfo* si = soinfo_alloc("[vdso]", NULL);
2045
2046  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
2047  si->phnum = ehdr_vdso->e_phnum;
2048  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
2049  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2050  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
2051
2052  soinfo_link_image(si, NULL);
2053#endif
2054}
2055
2056/*
2057 * This is linker soinfo for GDB. See details below.
2058 */
2059static soinfo linker_soinfo_for_gdb;
2060
2061/* gdb expects the linker to be in the debug shared object list.
2062 * Without this, gdb has trouble locating the linker's ".text"
2063 * and ".plt" sections. Gdb could also potentially use this to
2064 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2065 * Don't use soinfo_alloc(), because the linker shouldn't
2066 * be on the soinfo list.
2067 */
2068static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
2069#if defined(__LP64__)
2070  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
2071#else
2072  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
2073#endif
2074  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
2075  linker_soinfo_for_gdb.base = linker_base;
2076
2077  /*
2078   * Set the dynamic field in the link map otherwise gdb will complain with
2079   * the following:
2080   *   warning: .dynamic section for "/system/bin/linker" is not at the
2081   *   expected address (wrong library or version mismatch?)
2082   */
2083  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2084  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2085  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2086                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
2087  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2088}
2089
2090/*
2091 * This code is called after the linker has linked itself and
2092 * fixed it's own GOT. It is safe to make references to externs
2093 * and other non-local data at this point.
2094 */
2095static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2096    /* NOTE: we store the args pointer on a special location
2097     *       of the temporary TLS area in order to pass it to
2098     *       the C Library's runtime initializer.
2099     *
2100     *       The initializer must clear the slot and reset the TLS
2101     *       to point to a different location to ensure that no other
2102     *       shared library constructor can access it.
2103     */
2104  __libc_init_tls(args);
2105
2106#if TIMING
2107    struct timeval t0, t1;
2108    gettimeofday(&t0, 0);
2109#endif
2110
2111    // Initialize environment functions, and get to the ELF aux vectors table.
2112    linker_env_init(args);
2113
2114    // If this is a setuid/setgid program, close the security hole described in
2115    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2116    if (get_AT_SECURE()) {
2117        nullify_closed_stdio();
2118    }
2119
2120    debuggerd_init();
2121
2122    // Get a few environment variables.
2123    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2124    if (LD_DEBUG != NULL) {
2125      g_ld_debug_verbosity = atoi(LD_DEBUG);
2126    }
2127
2128    // Normally, these are cleaned by linker_env_init, but the test
2129    // doesn't cost us anything.
2130    const char* ldpath_env = NULL;
2131    const char* ldpreload_env = NULL;
2132    if (!get_AT_SECURE()) {
2133      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2134      ldpreload_env = linker_env_get("LD_PRELOAD");
2135    }
2136
2137    INFO("[ android linker & debugger ]");
2138
2139    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2140    if (si == NULL) {
2141        exit(EXIT_FAILURE);
2142    }
2143
2144    /* bootstrap the link map, the main exe always needs to be first */
2145    si->flags |= FLAG_EXE;
2146    link_map* map = &(si->link_map_head);
2147
2148    map->l_addr = 0;
2149    map->l_name = args.argv[0];
2150    map->l_prev = NULL;
2151    map->l_next = NULL;
2152
2153    _r_debug.r_map = map;
2154    r_debug_tail = map;
2155
2156    init_linker_info_for_gdb(linker_base);
2157
2158    // Extract information passed from the kernel.
2159    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2160    si->phnum = args.getauxval(AT_PHNUM);
2161    si->entry = args.getauxval(AT_ENTRY);
2162
2163    /* Compute the value of si->base. We can't rely on the fact that
2164     * the first entry is the PHDR because this will not be true
2165     * for certain executables (e.g. some in the NDK unit test suite)
2166     */
2167    si->base = 0;
2168    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2169    si->load_bias = 0;
2170    for (size_t i = 0; i < si->phnum; ++i) {
2171      if (si->phdr[i].p_type == PT_PHDR) {
2172        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2173        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2174        break;
2175      }
2176    }
2177    si->dynamic = NULL;
2178    si->ref_count = 1;
2179
2180    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2181    if (elf_hdr->e_type != ET_DYN) {
2182        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2183        exit(EXIT_FAILURE);
2184    }
2185
2186    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2187    parse_LD_LIBRARY_PATH(ldpath_env);
2188    parse_LD_PRELOAD(ldpreload_env);
2189
2190    somain = si;
2191
2192    if (!soinfo_link_image(si, NULL)) {
2193        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2194        exit(EXIT_FAILURE);
2195    }
2196
2197    add_vdso(args);
2198
2199    si->CallPreInitConstructors();
2200
2201    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2202        g_ld_preloads[i]->CallConstructors();
2203    }
2204
2205    /* After the link_image, the si->load_bias is initialized.
2206     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2207     * We need to update this value for so exe here. So Unwind_Backtrace
2208     * for some arch like x86 could work correctly within so exe.
2209     */
2210    map->l_addr = si->load_bias;
2211    si->CallConstructors();
2212
2213#if TIMING
2214    gettimeofday(&t1, NULL);
2215    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2216               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2217               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2218#endif
2219#if STATS
2220    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2221           linker_stats.count[kRelocAbsolute],
2222           linker_stats.count[kRelocRelative],
2223           linker_stats.count[kRelocCopy],
2224           linker_stats.count[kRelocSymbol]);
2225#endif
2226#if COUNT_PAGES
2227    {
2228        unsigned n;
2229        unsigned i;
2230        unsigned count = 0;
2231        for (n = 0; n < 4096; n++) {
2232            if (bitmask[n]) {
2233                unsigned x = bitmask[n];
2234#if defined(__LP64__)
2235                for (i = 0; i < 32; i++) {
2236#else
2237                for (i = 0; i < 8; i++) {
2238#endif
2239                    if (x & 1) {
2240                        count++;
2241                    }
2242                    x >>= 1;
2243                }
2244            }
2245        }
2246        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2247    }
2248#endif
2249
2250#if TIMING || STATS || COUNT_PAGES
2251    fflush(stdout);
2252#endif
2253
2254    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2255    return si->entry;
2256}
2257
2258/* Compute the load-bias of an existing executable. This shall only
2259 * be used to compute the load bias of an executable or shared library
2260 * that was loaded by the kernel itself.
2261 *
2262 * Input:
2263 *    elf    -> address of ELF header, assumed to be at the start of the file.
2264 * Return:
2265 *    load bias, i.e. add the value of any p_vaddr in the file to get
2266 *    the corresponding address in memory.
2267 */
2268static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2269  ElfW(Addr) offset = elf->e_phoff;
2270  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2271  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2272
2273  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2274    if (phdr->p_type == PT_LOAD) {
2275      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2276    }
2277  }
2278  return 0;
2279}
2280
2281extern "C" void _start();
2282
2283/*
2284 * This is the entry point for the linker, called from begin.S. This
2285 * method is responsible for fixing the linker's own relocations, and
2286 * then calling __linker_init_post_relocation().
2287 *
2288 * Because this method is called before the linker has fixed it's own
2289 * relocations, any attempt to reference an extern variable, extern
2290 * function, or other GOT reference will generate a segfault.
2291 */
2292extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2293  // Initialize static variables.
2294  solist = get_libdl_info();
2295  sonext = get_libdl_info();
2296
2297  KernelArgumentBlock args(raw_args);
2298
2299  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2300  ElfW(Addr) entry_point = args.getauxval(AT_ENTRY);
2301  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2302  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2303
2304  soinfo linker_so;
2305  memset(&linker_so, 0, sizeof(soinfo));
2306
2307  // If the linker is not acting as PT_INTERP entry_point is equal to
2308  // _start. Which means that the linker is running as an executable and
2309  // already linked by PT_INTERP.
2310  //
2311  // This happens when user tries to run 'adb shell /system/bin/linker'
2312  // see also https://code.google.com/p/android/issues/detail?id=63174
2313  if (reinterpret_cast<ElfW(Addr)>(&_start) == entry_point) {
2314    __libc_fatal("This is %s, the helper program for shared library executables.\n", args.argv[0]);
2315  }
2316
2317  strcpy(linker_so.name, "[dynamic linker]");
2318  linker_so.base = linker_addr;
2319  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2320  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2321  linker_so.dynamic = NULL;
2322  linker_so.phdr = phdr;
2323  linker_so.phnum = elf_hdr->e_phnum;
2324  linker_so.flags |= FLAG_LINKER;
2325
2326  if (!soinfo_link_image(&linker_so, NULL)) {
2327    // It would be nice to print an error message, but if the linker
2328    // can't link itself, there's no guarantee that we'll be able to
2329    // call write() (because it involves a GOT reference). We may as
2330    // well try though...
2331    const char* msg = "CANNOT LINK EXECUTABLE: ";
2332    write(2, msg, strlen(msg));
2333    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2334    write(2, "\n", 1);
2335    _exit(EXIT_FAILURE);
2336  }
2337
2338  // Initialize the linker's own global variables
2339  linker_so.CallConstructors();
2340
2341  // We have successfully fixed our own relocations. It's safe to run
2342  // the main part of the linker now.
2343  args.abort_message_ptr = &g_abort_message;
2344  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2345
2346  protect_data(PROT_READ);
2347
2348  // Return the address that the calling assembly stub should jump to.
2349  return start_address;
2350}
2351