linker.cpp revision d286796fce1e6fff407b719a53eb2afbeb96c327
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
46#include "linker.h"
47#include "linker_debug.h"
48#include "linker_environ.h"
49#include "linker_phdr.h"
50#include "linker_allocator.h"
51
52/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
53 *
54 * Do NOT use malloc() and friends or pthread_*() code here.
55 * Don't use printf() either; it's caused mysterious memory
56 * corruption in the past.
57 * The linker runs before we bring up libc and it's easiest
58 * to make sure it does not depend on any complex libc features
59 *
60 * open issues / todo:
61 *
62 * - cleaner error reporting
63 * - after linking, set as much stuff as possible to READONLY
64 *   and NOEXEC
65 */
66
67#if defined(__LP64__)
68#define SEARCH_NAME(x) x
69#else
70// Nvidia drivers are relying on the bug:
71// http://code.google.com/p/android/issues/detail?id=6670
72// so we continue to use base-name lookup for lp32
73static const char* get_base_name(const char* name) {
74  const char* bname = strrchr(name, '/');
75  return bname ? bname + 1 : name;
76}
77#define SEARCH_NAME(x) get_base_name(x)
78#endif
79
80static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
81static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
82
83static LinkerAllocator<soinfo> g_soinfo_allocator;
84static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
85
86static soinfo* solist;
87static soinfo* sonext;
88static soinfo* somain; /* main process, always the one after libdl_info */
89
90static const char* const kDefaultLdPaths[] = {
91#if defined(__LP64__)
92  "/vendor/lib64",
93  "/system/lib64",
94#else
95  "/vendor/lib",
96  "/system/lib",
97#endif
98  NULL
99};
100
101#define LDPATH_BUFSIZE (LDPATH_MAX*64)
102#define LDPATH_MAX 8
103
104#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
105#define LDPRELOAD_MAX 8
106
107static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
108static const char* g_ld_library_paths[LDPATH_MAX + 1];
109
110static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
111static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
112
113static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
114
115__LIBC_HIDDEN__ int g_ld_debug_verbosity;
116
117__LIBC_HIDDEN__ abort_msg_t* g_abort_message = NULL; // For debuggerd.
118
119enum RelocationKind {
120    kRelocAbsolute = 0,
121    kRelocRelative,
122    kRelocCopy,
123    kRelocSymbol,
124    kRelocMax
125};
126
127#if STATS
128struct linker_stats_t {
129    int count[kRelocMax];
130};
131
132static linker_stats_t linker_stats;
133
134static void count_relocation(RelocationKind kind) {
135    ++linker_stats.count[kind];
136}
137#else
138static void count_relocation(RelocationKind) {
139}
140#endif
141
142#if COUNT_PAGES
143static unsigned bitmask[4096];
144#if defined(__LP64__)
145#define MARK(offset) \
146    do { \
147        if ((((offset) >> 12) >> 5) < 4096) \
148            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
149    } while (0)
150#else
151#define MARK(offset) \
152    do { \
153        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
154    } while (0)
155#endif
156#else
157#define MARK(x) do {} while (0)
158#endif
159
160// You shouldn't try to call memory-allocating functions in the dynamic linker.
161// Guard against the most obvious ones.
162#define DISALLOW_ALLOCATION(return_type, name, ...) \
163    return_type name __VA_ARGS__ \
164    { \
165        const char* msg = "ERROR: " #name " called from the dynamic linker!\n"; \
166        __libc_format_log(ANDROID_LOG_FATAL, "linker", "%s", msg); \
167        write(2, msg, strlen(msg)); \
168        abort(); \
169    }
170DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
171DISALLOW_ALLOCATION(void, free, (void* u __unused));
172DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
173DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
174
175static char tmp_err_buf[768];
176static char __linker_dl_err_buf[768];
177
178char* linker_get_error_buffer() {
179  return &__linker_dl_err_buf[0];
180}
181
182size_t linker_get_error_buffer_size() {
183  return sizeof(__linker_dl_err_buf);
184}
185
186/*
187 * This function is an empty stub where GDB locates a breakpoint to get notified
188 * about linker activity.
189 */
190extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
191
192static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
193static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
194static link_map* r_debug_tail = 0;
195
196static void insert_soinfo_into_debug_map(soinfo* info) {
197    // Copy the necessary fields into the debug structure.
198    link_map* map = &(info->link_map_head);
199    map->l_addr = info->load_bias;
200    map->l_name = reinterpret_cast<char*>(info->name);
201    map->l_ld = info->dynamic;
202
203    /* Stick the new library at the end of the list.
204     * gdb tends to care more about libc than it does
205     * about leaf libraries, and ordering it this way
206     * reduces the back-and-forth over the wire.
207     */
208    if (r_debug_tail) {
209        r_debug_tail->l_next = map;
210        map->l_prev = r_debug_tail;
211        map->l_next = 0;
212    } else {
213        _r_debug.r_map = map;
214        map->l_prev = 0;
215        map->l_next = 0;
216    }
217    r_debug_tail = map;
218}
219
220static void remove_soinfo_from_debug_map(soinfo* info) {
221    link_map* map = &(info->link_map_head);
222
223    if (r_debug_tail == map) {
224        r_debug_tail = map->l_prev;
225    }
226
227    if (map->l_prev) {
228        map->l_prev->l_next = map->l_next;
229    }
230    if (map->l_next) {
231        map->l_next->l_prev = map->l_prev;
232    }
233}
234
235static void notify_gdb_of_load(soinfo* info) {
236    if (info->flags & FLAG_EXE) {
237        // GDB already knows about the main executable
238        return;
239    }
240
241    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
242
243    _r_debug.r_state = r_debug::RT_ADD;
244    rtld_db_dlactivity();
245
246    insert_soinfo_into_debug_map(info);
247
248    _r_debug.r_state = r_debug::RT_CONSISTENT;
249    rtld_db_dlactivity();
250}
251
252static void notify_gdb_of_unload(soinfo* info) {
253    if (info->flags & FLAG_EXE) {
254        // GDB already knows about the main executable
255        return;
256    }
257
258    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
259
260    _r_debug.r_state = r_debug::RT_DELETE;
261    rtld_db_dlactivity();
262
263    remove_soinfo_from_debug_map(info);
264
265    _r_debug.r_state = r_debug::RT_CONSISTENT;
266    rtld_db_dlactivity();
267}
268
269void notify_gdb_of_libraries() {
270  _r_debug.r_state = r_debug::RT_ADD;
271  rtld_db_dlactivity();
272  _r_debug.r_state = r_debug::RT_CONSISTENT;
273  rtld_db_dlactivity();
274}
275
276LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
277  return g_soinfo_links_allocator.alloc();
278}
279
280void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
281  g_soinfo_links_allocator.free(entry);
282}
283
284static void protect_data(int protection) {
285  g_soinfo_allocator.protect_all(protection);
286  g_soinfo_links_allocator.protect_all(protection);
287}
288
289static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
290  if (strlen(name) >= SOINFO_NAME_LEN) {
291    DL_ERR("library name \"%s\" too long", name);
292    return NULL;
293  }
294
295  soinfo* si = g_soinfo_allocator.alloc();
296
297  // Initialize the new element.
298  memset(si, 0, sizeof(soinfo));
299  strlcpy(si->name, name, sizeof(si->name));
300  si->flags = FLAG_NEW_SOINFO;
301
302  if (file_stat != NULL) {
303    si->set_st_dev(file_stat->st_dev);
304    si->set_st_ino(file_stat->st_ino);
305  }
306
307  sonext->next = si;
308  sonext = si;
309
310  TRACE("name %s: allocated soinfo @ %p", name, si);
311  return si;
312}
313
314static void soinfo_free(soinfo* si) {
315    if (si == NULL) {
316        return;
317    }
318
319    if (si->base != 0 && si->size != 0) {
320      munmap(reinterpret_cast<void*>(si->base), si->size);
321    }
322
323    soinfo *prev = NULL, *trav;
324
325    TRACE("name %s: freeing soinfo @ %p", si->name, si);
326
327    for (trav = solist; trav != NULL; trav = trav->next) {
328        if (trav == si)
329            break;
330        prev = trav;
331    }
332    if (trav == NULL) {
333        /* si was not in solist */
334        DL_ERR("name \"%s\" is not in solist!", si->name);
335        return;
336    }
337
338    // clear links to/from si
339    si->remove_all_links();
340
341    /* prev will never be NULL, because the first entry in solist is
342       always the static libdl_info.
343    */
344    prev->next = si->next;
345    if (si == sonext) {
346        sonext = prev;
347    }
348
349    g_soinfo_allocator.free(si);
350}
351
352
353static void parse_path(const char* path, const char* delimiters,
354                       const char** array, char* buf, size_t buf_size, size_t max_count) {
355  if (path == NULL) {
356    return;
357  }
358
359  size_t len = strlcpy(buf, path, buf_size);
360
361  size_t i = 0;
362  char* buf_p = buf;
363  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
364    if (*array[i] != '\0') {
365      ++i;
366    }
367  }
368
369  // Forget the last path if we had to truncate; this occurs if the 2nd to
370  // last char isn't '\0' (i.e. wasn't originally a delimiter).
371  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
372    array[i - 1] = NULL;
373  } else {
374    array[i] = NULL;
375  }
376}
377
378static void parse_LD_LIBRARY_PATH(const char* path) {
379  parse_path(path, ":", g_ld_library_paths,
380             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
381}
382
383static void parse_LD_PRELOAD(const char* path) {
384  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
385  parse_path(path, " :", g_ld_preload_names,
386             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
387}
388
389#if defined(__arm__)
390
391/* For a given PC, find the .so that it belongs to.
392 * Returns the base address of the .ARM.exidx section
393 * for that .so, and the number of 8-byte entries
394 * in that section (via *pcount).
395 *
396 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
397 *
398 * This function is exposed via dlfcn.cpp and libdl.so.
399 */
400_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
401    unsigned addr = (unsigned)pc;
402
403    for (soinfo* si = solist; si != 0; si = si->next) {
404        if ((addr >= si->base) && (addr < (si->base + si->size))) {
405            *pcount = si->ARM_exidx_count;
406            return (_Unwind_Ptr)si->ARM_exidx;
407        }
408    }
409    *pcount = 0;
410    return NULL;
411}
412
413#endif
414
415/* Here, we only have to provide a callback to iterate across all the
416 * loaded libraries. gcc_eh does the rest. */
417int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
418    int rv = 0;
419    for (soinfo* si = solist; si != NULL; si = si->next) {
420        dl_phdr_info dl_info;
421        dl_info.dlpi_addr = si->link_map_head.l_addr;
422        dl_info.dlpi_name = si->link_map_head.l_name;
423        dl_info.dlpi_phdr = si->phdr;
424        dl_info.dlpi_phnum = si->phnum;
425        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
426        if (rv != 0) {
427            break;
428        }
429    }
430    return rv;
431}
432
433static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
434  ElfW(Sym)* symtab = si->symtab;
435  const char* strtab = si->strtab;
436
437  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
438             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
439
440  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
441    ElfW(Sym)* s = symtab + n;
442    if (strcmp(strtab + s->st_name, name)) continue;
443
444    /* only concern ourselves with global and weak symbol definitions */
445    switch (ELF_ST_BIND(s->st_info)) {
446      case STB_GLOBAL:
447      case STB_WEAK:
448        if (s->st_shndx == SHN_UNDEF) {
449        continue;
450      }
451
452      TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
453                 name, si->name, reinterpret_cast<void*>(s->st_value),
454                 static_cast<size_t>(s->st_size));
455      return s;
456    }
457  }
458
459  return NULL;
460}
461
462static unsigned elfhash(const char* _name) {
463    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
464    unsigned h = 0, g;
465
466    while (*name) {
467        h = (h << 4) + *name++;
468        g = h & 0xf0000000;
469        h ^= g;
470        h ^= g >> 24;
471    }
472    return h;
473}
474
475static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
476    unsigned elf_hash = elfhash(name);
477    ElfW(Sym)* s = NULL;
478
479    if (si != NULL && somain != NULL) {
480        /*
481         * Local scope is executable scope. Just start looking into it right away
482         * for the shortcut.
483         */
484
485        if (si == somain) {
486            s = soinfo_elf_lookup(si, elf_hash, name);
487            if (s != NULL) {
488                *lsi = si;
489                goto done;
490            }
491        } else {
492            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
493
494            /*
495             * If this object was built with symbolic relocations disabled, the
496             * first place to look to resolve external references is the main
497             * executable.
498             */
499
500            if (!si->has_DT_SYMBOLIC) {
501                DEBUG("%s: looking up %s in executable %s",
502                      si->name, name, somain->name);
503                s = soinfo_elf_lookup(somain, elf_hash, name);
504                if (s != NULL) {
505                    *lsi = somain;
506                    goto done;
507                }
508            }
509
510            /* Look for symbols in the local scope (the object who is
511             * searching). This happens with C++ templates on x86 for some
512             * reason.
513             *
514             * Notes on weak symbols:
515             * The ELF specs are ambiguous about treatment of weak definitions in
516             * dynamic linking.  Some systems return the first definition found
517             * and some the first non-weak definition.   This is system dependent.
518             * Here we return the first definition found for simplicity.  */
519
520            s = soinfo_elf_lookup(si, elf_hash, name);
521            if (s != NULL) {
522                *lsi = si;
523                goto done;
524            }
525
526            /*
527             * If this object was built with -Bsymbolic and symbol is not found
528             * in the local scope, try to find the symbol in the main executable.
529             */
530
531            if (si->has_DT_SYMBOLIC) {
532                DEBUG("%s: looking up %s in executable %s after local scope",
533                      si->name, name, somain->name);
534                s = soinfo_elf_lookup(somain, elf_hash, name);
535                if (s != NULL) {
536                    *lsi = somain;
537                    goto done;
538                }
539            }
540        }
541    }
542
543    /* Next, look for it in the preloads list */
544    for (int i = 0; g_ld_preloads[i] != NULL; i++) {
545        s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
546        if (s != NULL) {
547            *lsi = g_ld_preloads[i];
548            goto done;
549        }
550    }
551
552    for (int i = 0; needed[i] != NULL; i++) {
553        DEBUG("%s: looking up %s in %s",
554              si->name, name, needed[i]->name);
555        s = soinfo_elf_lookup(needed[i], elf_hash, name);
556        if (s != NULL) {
557            *lsi = needed[i];
558            goto done;
559        }
560    }
561
562done:
563    if (s != NULL) {
564        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
565                   "found in %s, base = %p, load bias = %p",
566                   si->name, name, reinterpret_cast<void*>(s->st_value),
567                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
568                   reinterpret_cast<void*>((*lsi)->load_bias));
569        return s;
570    }
571
572    return NULL;
573}
574
575/* This is used by dlsym(3).  It performs symbol lookup only within the
576   specified soinfo object and not in any of its dependencies.
577
578   TODO: Only looking in the specified soinfo seems wrong. dlsym(3) says
579   that it should do a breadth first search through the dependency
580   tree. This agrees with the ELF spec (aka System V Application
581   Binary Interface) where in Chapter 5 it discuss resolving "Shared
582   Object Dependencies" in breadth first search order.
583 */
584ElfW(Sym)* dlsym_handle_lookup(soinfo* si, const char* name) {
585    return soinfo_elf_lookup(si, elfhash(name), name);
586}
587
588/* This is used by dlsym(3) to performs a global symbol lookup. If the
589   start value is null (for RTLD_DEFAULT), the search starts at the
590   beginning of the global solist. Otherwise the search starts at the
591   specified soinfo (for RTLD_NEXT).
592 */
593ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
594  unsigned elf_hash = elfhash(name);
595
596  if (start == NULL) {
597    start = solist;
598  }
599
600  ElfW(Sym)* s = NULL;
601  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
602    s = soinfo_elf_lookup(si, elf_hash, name);
603    if (s != NULL) {
604      *found = si;
605      break;
606    }
607  }
608
609  if (s != NULL) {
610    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
611               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
612  }
613
614  return s;
615}
616
617soinfo* find_containing_library(const void* p) {
618  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
619  for (soinfo* si = solist; si != NULL; si = si->next) {
620    if (address >= si->base && address - si->base < si->size) {
621      return si;
622    }
623  }
624  return NULL;
625}
626
627ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
628  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
629
630  // Search the library's symbol table for any defined symbol which
631  // contains this address.
632  for (size_t i = 0; i < si->nchain; ++i) {
633    ElfW(Sym)* sym = &si->symtab[i];
634    if (sym->st_shndx != SHN_UNDEF &&
635        soaddr >= sym->st_value &&
636        soaddr < sym->st_value + sym->st_size) {
637      return sym;
638    }
639  }
640
641  return NULL;
642}
643
644static int open_library_on_path(const char* name, const char* const paths[]) {
645  char buf[512];
646  for (size_t i = 0; paths[i] != NULL; ++i) {
647    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
648    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
649      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
650      continue;
651    }
652    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
653    if (fd != -1) {
654      return fd;
655    }
656  }
657  return -1;
658}
659
660static int open_library(const char* name) {
661  TRACE("[ opening %s ]", name);
662
663  // If the name contains a slash, we should attempt to open it directly and not search the paths.
664  if (strchr(name, '/') != NULL) {
665    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
666    if (fd != -1) {
667      return fd;
668    }
669    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
670#if defined(__LP64__)
671    // TODO: uncomment this after bug b/7465467 is fixed.
672    // return -1;
673#endif
674  }
675
676  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
677  int fd = open_library_on_path(name, g_ld_library_paths);
678  if (fd == -1) {
679    fd = open_library_on_path(name, kDefaultLdPaths);
680  }
681  return fd;
682}
683
684static soinfo* load_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
685    // Open the file.
686    int fd = open_library(name);
687    if (fd == -1) {
688        DL_ERR("library \"%s\" not found", name);
689        return NULL;
690    }
691
692    ElfReader elf_reader(name, fd);
693
694    struct stat file_stat;
695    if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
696      DL_ERR("unable to stat file for the library %s: %s", name, strerror(errno));
697      return NULL;
698    }
699
700    // Check for symlink and other situations where
701    // file can have different names.
702    for (soinfo* si = solist; si != NULL; si = si->next) {
703      if (si->get_st_dev() != 0 &&
704          si->get_st_ino() != 0 &&
705          si->get_st_dev() == file_stat.st_dev &&
706          si->get_st_ino() == file_stat.st_ino) {
707        TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
708        return si;
709      }
710    }
711
712    if ((dlflags & RTLD_NOLOAD) != 0) {
713      return NULL;
714    }
715
716    // Read the ELF header and load the segments.
717    if (!elf_reader.Load(extinfo)) {
718        return NULL;
719    }
720
721    soinfo* si = soinfo_alloc(SEARCH_NAME(name), &file_stat);
722    if (si == NULL) {
723        return NULL;
724    }
725    si->base = elf_reader.load_start();
726    si->size = elf_reader.load_size();
727    si->load_bias = elf_reader.load_bias();
728    si->phnum = elf_reader.phdr_count();
729    si->phdr = elf_reader.loaded_phdr();
730
731    // At this point we know that whatever is loaded @ base is a valid ELF
732    // shared library whose segments are properly mapped in.
733    TRACE("[ find_library_internal base=%p size=%zu name='%s' ]",
734          reinterpret_cast<void*>(si->base), si->size, si->name);
735
736    if (!soinfo_link_image(si, extinfo)) {
737      soinfo_free(si);
738      return NULL;
739    }
740
741    return si;
742}
743
744static soinfo *find_loaded_library_by_name(const char* name) {
745  const char* search_name = SEARCH_NAME(name);
746  for (soinfo* si = solist; si != NULL; si = si->next) {
747    if (!strcmp(search_name, si->name)) {
748      return si;
749    }
750  }
751  return NULL;
752}
753
754static soinfo* find_library_internal(const char* name, int dlflags, const android_dlextinfo* extinfo) {
755  if (name == NULL) {
756    return somain;
757  }
758
759  soinfo* si = find_loaded_library_by_name(name);
760
761  // Library might still be loaded, the accurate detection
762  // of this fact is done by load_library
763  if (si == NULL) {
764    TRACE("[ '%s' has not been found by name.  Trying harder...]", name);
765    si = load_library(name, dlflags, extinfo);
766  }
767
768  if (si != NULL && (si->flags & FLAG_LINKED) == 0) {
769    DL_ERR("recursive link to \"%s\"", si->name);
770    return NULL;
771  }
772
773  return si;
774}
775
776static soinfo* find_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
777  soinfo* si = find_library_internal(name, dlflags, extinfo);
778  if (si != NULL) {
779    si->ref_count++;
780  }
781  return si;
782}
783
784static void soinfo_unload(soinfo* si) {
785  if (si->ref_count == 1) {
786    TRACE("unloading '%s'", si->name);
787    si->CallDestructors();
788
789    if ((si->flags | FLAG_NEW_SOINFO) != 0) {
790      si->get_children().for_each([&] (soinfo* child) {
791        TRACE("%s needs to unload %s", si->name, child->name);
792        soinfo_unload(child);
793      });
794    } else {
795      for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
796        if (d->d_tag == DT_NEEDED) {
797          const char* library_name = si->strtab + d->d_un.d_val;
798          TRACE("%s needs to unload %s", si->name, library_name);
799          soinfo* needed = find_library(library_name, RTLD_NOLOAD, NULL);
800          if (needed != NULL) {
801            soinfo_unload(needed);
802          } else {
803            // Not found: for example if symlink was deleted between dlopen and dlclose
804            // Since we cannot really handle errors at this point - print and continue.
805            PRINT("warning: couldn't find %s needed by %s on unload.", library_name, si->name);
806          }
807        }
808      }
809    }
810
811    notify_gdb_of_unload(si);
812    si->ref_count = 0;
813    soinfo_free(si);
814  } else {
815    si->ref_count--;
816    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
817  }
818}
819
820void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
821  snprintf(buffer, buffer_size, "%s:%s", kDefaultLdPaths[0], kDefaultLdPaths[1]);
822}
823
824void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
825  if (!get_AT_SECURE()) {
826    parse_LD_LIBRARY_PATH(ld_library_path);
827  }
828}
829
830soinfo* do_dlopen(const char* name, int flags, soinfo* /*caller*/, const android_dlextinfo* extinfo) {
831  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL|RTLD_NOLOAD)) != 0) {
832    DL_ERR("invalid flags to dlopen: %x", flags);
833    return NULL;
834  }
835  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
836    DL_ERR("invalid extended flags to android_dlopen_ext: %x", extinfo->flags);
837    return NULL;
838  }
839  protect_data(PROT_READ | PROT_WRITE);
840  soinfo* si = find_library(name, flags, extinfo);
841  if (si != NULL) {
842    si->CallConstructors();
843  }
844  protect_data(PROT_READ);
845  return si;
846}
847
848void do_dlclose(soinfo* si) {
849  protect_data(PROT_READ | PROT_WRITE);
850  soinfo_unload(si);
851  protect_data(PROT_READ);
852}
853
854#if defined(USE_RELA)
855static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
856  ElfW(Sym)* s;
857  soinfo* lsi;
858
859  for (size_t idx = 0; idx < count; ++idx, ++rela) {
860    unsigned type = ELFW(R_TYPE)(rela->r_info);
861    unsigned sym = ELFW(R_SYM)(rela->r_info);
862    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
863    ElfW(Addr) sym_addr = 0;
864    const char* sym_name = NULL;
865
866    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
867    if (type == 0) { // R_*_NONE
868      continue;
869    }
870    if (sym != 0) {
871      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
872      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
873      if (s == NULL) {
874        // We only allow an undefined symbol if this is a weak reference...
875        s = &si->symtab[sym];
876        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
877          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
878          return -1;
879        }
880
881        /* IHI0044C AAELF 4.5.1.1:
882
883           Libraries are not searched to resolve weak references.
884           It is not an error for a weak reference to remain unsatisfied.
885
886           During linking, the value of an undefined weak reference is:
887           - Zero if the relocation type is absolute
888           - The address of the place if the relocation is pc-relative
889           - The address of nominal base address if the relocation
890             type is base-relative.
891         */
892
893        switch (type) {
894#if defined(__aarch64__)
895        case R_AARCH64_JUMP_SLOT:
896        case R_AARCH64_GLOB_DAT:
897        case R_AARCH64_ABS64:
898        case R_AARCH64_ABS32:
899        case R_AARCH64_ABS16:
900        case R_AARCH64_RELATIVE:
901          /*
902           * The sym_addr was initialized to be zero above, or the relocation
903           * code below does not care about value of sym_addr.
904           * No need to do anything.
905           */
906          break;
907#elif defined(__x86_64__)
908        case R_X86_64_JUMP_SLOT:
909        case R_X86_64_GLOB_DAT:
910        case R_X86_64_32:
911        case R_X86_64_64:
912        case R_X86_64_RELATIVE:
913          // No need to do anything.
914          break;
915        case R_X86_64_PC32:
916          sym_addr = reloc;
917          break;
918#endif
919        default:
920          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
921          return -1;
922        }
923      } else {
924        // We got a definition.
925        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
926      }
927      count_relocation(kRelocSymbol);
928    } else {
929      s = NULL;
930    }
931
932    switch (type) {
933#if defined(__aarch64__)
934    case R_AARCH64_JUMP_SLOT:
935        count_relocation(kRelocAbsolute);
936        MARK(rela->r_offset);
937        TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
938                   reloc, (sym_addr + rela->r_addend), sym_name);
939        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
940        break;
941    case R_AARCH64_GLOB_DAT:
942        count_relocation(kRelocAbsolute);
943        MARK(rela->r_offset);
944        TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
945                   reloc, (sym_addr + rela->r_addend), sym_name);
946        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
947        break;
948    case R_AARCH64_ABS64:
949        count_relocation(kRelocAbsolute);
950        MARK(rela->r_offset);
951        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
952                   reloc, (sym_addr + rela->r_addend), sym_name);
953        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
954        break;
955    case R_AARCH64_ABS32:
956        count_relocation(kRelocAbsolute);
957        MARK(rela->r_offset);
958        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
959                   reloc, (sym_addr + rela->r_addend), sym_name);
960        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
961            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
962            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
963        } else {
964            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
965                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
966                   static_cast<ElfW(Addr)>(INT32_MIN),
967                   static_cast<ElfW(Addr)>(UINT32_MAX));
968            return -1;
969        }
970        break;
971    case R_AARCH64_ABS16:
972        count_relocation(kRelocAbsolute);
973        MARK(rela->r_offset);
974        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
975                   reloc, (sym_addr + rela->r_addend), sym_name);
976        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
977            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
978            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
979        } else {
980            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
981                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
982                   static_cast<ElfW(Addr)>(INT16_MIN),
983                   static_cast<ElfW(Addr)>(UINT16_MAX));
984            return -1;
985        }
986        break;
987    case R_AARCH64_PREL64:
988        count_relocation(kRelocRelative);
989        MARK(rela->r_offset);
990        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
991                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
992        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
993        break;
994    case R_AARCH64_PREL32:
995        count_relocation(kRelocRelative);
996        MARK(rela->r_offset);
997        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
998                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
999        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1000            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1001            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1002        } else {
1003            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1004                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1005                   static_cast<ElfW(Addr)>(INT32_MIN),
1006                   static_cast<ElfW(Addr)>(UINT32_MAX));
1007            return -1;
1008        }
1009        break;
1010    case R_AARCH64_PREL16:
1011        count_relocation(kRelocRelative);
1012        MARK(rela->r_offset);
1013        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1014                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1015        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1016            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1017            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1018        } else {
1019            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1020                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1021                   static_cast<ElfW(Addr)>(INT16_MIN),
1022                   static_cast<ElfW(Addr)>(UINT16_MAX));
1023            return -1;
1024        }
1025        break;
1026
1027    case R_AARCH64_RELATIVE:
1028        count_relocation(kRelocRelative);
1029        MARK(rela->r_offset);
1030        if (sym) {
1031            DL_ERR("odd RELATIVE form...");
1032            return -1;
1033        }
1034        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1035                   reloc, (si->base + rela->r_addend));
1036        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1037        break;
1038
1039    case R_AARCH64_COPY:
1040        /*
1041         * ET_EXEC is not supported so this should not happen.
1042         *
1043         * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1044         *
1045         * Section 4.7.1.10 "Dynamic relocations"
1046         * R_AARCH64_COPY may only appear in executable objects where e_type is
1047         * set to ET_EXEC.
1048         */
1049        DL_ERR("%s R_AARCH64_COPY relocations are not supported", si->name);
1050        return -1;
1051    case R_AARCH64_TLS_TPREL64:
1052        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1053                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1054        break;
1055    case R_AARCH64_TLS_DTPREL32:
1056        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1057                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1058        break;
1059#elif defined(__x86_64__)
1060    case R_X86_64_JUMP_SLOT:
1061      count_relocation(kRelocAbsolute);
1062      MARK(rela->r_offset);
1063      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1064                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1065      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1066      break;
1067    case R_X86_64_GLOB_DAT:
1068      count_relocation(kRelocAbsolute);
1069      MARK(rela->r_offset);
1070      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1071                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1072      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1073      break;
1074    case R_X86_64_RELATIVE:
1075      count_relocation(kRelocRelative);
1076      MARK(rela->r_offset);
1077      if (sym) {
1078        DL_ERR("odd RELATIVE form...");
1079        return -1;
1080      }
1081      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1082                 static_cast<size_t>(si->base));
1083      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1084      break;
1085    case R_X86_64_32:
1086      count_relocation(kRelocRelative);
1087      MARK(rela->r_offset);
1088      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1089                 static_cast<size_t>(sym_addr), sym_name);
1090      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1091      break;
1092    case R_X86_64_64:
1093      count_relocation(kRelocRelative);
1094      MARK(rela->r_offset);
1095      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1096                 static_cast<size_t>(sym_addr), sym_name);
1097      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1098      break;
1099    case R_X86_64_PC32:
1100      count_relocation(kRelocRelative);
1101      MARK(rela->r_offset);
1102      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1103                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1104                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1105      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1106      break;
1107#endif
1108
1109    default:
1110      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1111      return -1;
1112    }
1113  }
1114  return 0;
1115}
1116
1117#else // REL, not RELA.
1118
1119static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1120    ElfW(Sym)* s;
1121    soinfo* lsi;
1122
1123    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1124        unsigned type = ELFW(R_TYPE)(rel->r_info);
1125        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1126        unsigned sym = ELFW(R_SYM)(rel->r_info);
1127        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1128        ElfW(Addr) sym_addr = 0;
1129        const char* sym_name = NULL;
1130
1131        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1132        if (type == 0) { // R_*_NONE
1133            continue;
1134        }
1135        if (sym != 0) {
1136            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1137            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1138            if (s == NULL) {
1139                // We only allow an undefined symbol if this is a weak reference...
1140                s = &si->symtab[sym];
1141                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1142                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1143                    return -1;
1144                }
1145
1146                /* IHI0044C AAELF 4.5.1.1:
1147
1148                   Libraries are not searched to resolve weak references.
1149                   It is not an error for a weak reference to remain
1150                   unsatisfied.
1151
1152                   During linking, the value of an undefined weak reference is:
1153                   - Zero if the relocation type is absolute
1154                   - The address of the place if the relocation is pc-relative
1155                   - The address of nominal base address if the relocation
1156                     type is base-relative.
1157                  */
1158
1159                switch (type) {
1160#if defined(__arm__)
1161                case R_ARM_JUMP_SLOT:
1162                case R_ARM_GLOB_DAT:
1163                case R_ARM_ABS32:
1164                case R_ARM_RELATIVE:    /* Don't care. */
1165                    // sym_addr was initialized to be zero above or relocation
1166                    // code below does not care about value of sym_addr.
1167                    // No need to do anything.
1168                    break;
1169#elif defined(__i386__)
1170                case R_386_JMP_SLOT:
1171                case R_386_GLOB_DAT:
1172                case R_386_32:
1173                case R_386_RELATIVE:    /* Don't care. */
1174                    // sym_addr was initialized to be zero above or relocation
1175                    // code below does not care about value of sym_addr.
1176                    // No need to do anything.
1177                    break;
1178                case R_386_PC32:
1179                    sym_addr = reloc;
1180                    break;
1181#endif
1182
1183#if defined(__arm__)
1184                case R_ARM_COPY:
1185                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1186#endif
1187                default:
1188                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1189                    return -1;
1190                }
1191            } else {
1192                // We got a definition.
1193                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1194            }
1195            count_relocation(kRelocSymbol);
1196        } else {
1197            s = NULL;
1198        }
1199
1200        switch (type) {
1201#if defined(__arm__)
1202        case R_ARM_JUMP_SLOT:
1203            count_relocation(kRelocAbsolute);
1204            MARK(rel->r_offset);
1205            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1206            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1207            break;
1208        case R_ARM_GLOB_DAT:
1209            count_relocation(kRelocAbsolute);
1210            MARK(rel->r_offset);
1211            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1212            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1213            break;
1214        case R_ARM_ABS32:
1215            count_relocation(kRelocAbsolute);
1216            MARK(rel->r_offset);
1217            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1218            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1219            break;
1220        case R_ARM_REL32:
1221            count_relocation(kRelocRelative);
1222            MARK(rel->r_offset);
1223            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1224                       reloc, sym_addr, rel->r_offset, sym_name);
1225            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1226            break;
1227        case R_ARM_COPY:
1228            /*
1229             * ET_EXEC is not supported so this should not happen.
1230             *
1231             * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1232             *
1233             * Section 4.7.1.10 "Dynamic relocations"
1234             * R_ARM_COPY may only appear in executable objects where e_type is
1235             * set to ET_EXEC.
1236             */
1237            DL_ERR("%s R_ARM_COPY relocations are not supported", si->name);
1238            return -1;
1239#elif defined(__i386__)
1240        case R_386_JMP_SLOT:
1241            count_relocation(kRelocAbsolute);
1242            MARK(rel->r_offset);
1243            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1244            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1245            break;
1246        case R_386_GLOB_DAT:
1247            count_relocation(kRelocAbsolute);
1248            MARK(rel->r_offset);
1249            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1250            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1251            break;
1252        case R_386_32:
1253            count_relocation(kRelocRelative);
1254            MARK(rel->r_offset);
1255            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1256            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1257            break;
1258        case R_386_PC32:
1259            count_relocation(kRelocRelative);
1260            MARK(rel->r_offset);
1261            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1262                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1263            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1264            break;
1265#elif defined(__mips__)
1266        case R_MIPS_REL32:
1267#if defined(__LP64__)
1268            // MIPS Elf64_Rel entries contain compound relocations
1269            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1270            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1271                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1272                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1273                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1274                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1275                return -1;
1276            }
1277#endif
1278            count_relocation(kRelocAbsolute);
1279            MARK(rel->r_offset);
1280            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1281                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1282            if (s) {
1283                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1284            } else {
1285                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1286            }
1287            break;
1288#endif
1289
1290#if defined(__arm__)
1291        case R_ARM_RELATIVE:
1292#elif defined(__i386__)
1293        case R_386_RELATIVE:
1294#endif
1295            count_relocation(kRelocRelative);
1296            MARK(rel->r_offset);
1297            if (sym) {
1298                DL_ERR("odd RELATIVE form...");
1299                return -1;
1300            }
1301            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1302                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1303            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1304            break;
1305
1306        default:
1307            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1308            return -1;
1309        }
1310    }
1311    return 0;
1312}
1313#endif
1314
1315#if defined(__mips__)
1316static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1317    ElfW(Addr)** got = si->plt_got;
1318    if (got == NULL) {
1319        return true;
1320    }
1321    unsigned local_gotno = si->mips_local_gotno;
1322    unsigned gotsym = si->mips_gotsym;
1323    unsigned symtabno = si->mips_symtabno;
1324    ElfW(Sym)* symtab = si->symtab;
1325
1326    // got[0] is the address of the lazy resolver function.
1327    // got[1] may be used for a GNU extension.
1328    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1329    // FIXME: maybe this should be in a separate routine?
1330    if ((si->flags & FLAG_LINKER) == 0) {
1331        size_t g = 0;
1332        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1333        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1334            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1335        }
1336        // Relocate the local GOT entries.
1337        for (; g < local_gotno; g++) {
1338            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1339        }
1340    }
1341
1342    // Now for the global GOT entries...
1343    ElfW(Sym)* sym = symtab + gotsym;
1344    got = si->plt_got + local_gotno;
1345    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1346        // This is an undefined reference... try to locate it.
1347        const char* sym_name = si->strtab + sym->st_name;
1348        soinfo* lsi;
1349        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1350        if (s == NULL) {
1351            // We only allow an undefined symbol if this is a weak reference.
1352            s = &symtab[g];
1353            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1354                DL_ERR("cannot locate \"%s\"...", sym_name);
1355                return false;
1356            }
1357            *got = 0;
1358        } else {
1359            // FIXME: is this sufficient?
1360            // For reference see NetBSD link loader
1361            // 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
1362            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1363        }
1364    }
1365    return true;
1366}
1367#endif
1368
1369void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1370  if (functions == NULL) {
1371    return;
1372  }
1373
1374  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1375
1376  int begin = reverse ? (count - 1) : 0;
1377  int end = reverse ? -1 : count;
1378  int step = reverse ? -1 : 1;
1379
1380  for (int i = begin; i != end; i += step) {
1381    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1382    CallFunction("function", functions[i]);
1383  }
1384
1385  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1386}
1387
1388void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1389  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1390    return;
1391  }
1392
1393  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1394  function();
1395  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1396
1397  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1398  // are still writable. This happens with our debug malloc (see http://b/7941716).
1399  protect_data(PROT_READ | PROT_WRITE);
1400}
1401
1402void soinfo::CallPreInitConstructors() {
1403  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1404  // but ignored in a shared library.
1405  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1406}
1407
1408void soinfo::CallConstructors() {
1409  if (constructors_called) {
1410    return;
1411  }
1412
1413  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1414  // protect against recursive constructor calls. One simple example of constructor recursion
1415  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1416  // 1. The program depends on libc, so libc's constructor is called here.
1417  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1418  // 3. dlopen() calls the constructors on the newly created
1419  //    soinfo for libc_malloc_debug_leak.so.
1420  // 4. The debug .so depends on libc, so CallConstructors is
1421  //    called again with the libc soinfo. If it doesn't trigger the early-
1422  //    out above, the libc constructor will be called again (recursively!).
1423  constructors_called = true;
1424
1425  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1426    // The GNU dynamic linker silently ignores these, but we warn the developer.
1427    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1428          name, preinit_array_count);
1429  }
1430
1431  get_children().for_each([] (soinfo* si) {
1432    si->CallConstructors();
1433  });
1434
1435  TRACE("\"%s\": calling constructors", name);
1436
1437  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1438  CallFunction("DT_INIT", init_func);
1439  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1440}
1441
1442void soinfo::CallDestructors() {
1443  TRACE("\"%s\": calling destructors", name);
1444
1445  // DT_FINI_ARRAY must be parsed in reverse order.
1446  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1447
1448  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1449  CallFunction("DT_FINI", fini_func);
1450
1451  // This is needed on second call to dlopen
1452  // after library has been unloaded with RTLD_NODELETE
1453  constructors_called = false;
1454}
1455
1456void soinfo::add_child(soinfo* child) {
1457  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1458    return;
1459  }
1460
1461  this->children.push_front(child);
1462  child->parents.push_front(this);
1463}
1464
1465void soinfo::remove_all_links() {
1466  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1467    return;
1468  }
1469
1470  // 1. Untie connected soinfos from 'this'.
1471  children.for_each([&] (soinfo* child) {
1472    child->parents.remove_if([&] (const soinfo* parent) {
1473      return parent == this;
1474    });
1475  });
1476
1477  parents.for_each([&] (soinfo* parent) {
1478    parent->children.for_each([&] (const soinfo* child) {
1479      return child == this;
1480    });
1481  });
1482
1483  // 2. Once everything untied - clear local lists.
1484  parents.clear();
1485  children.clear();
1486}
1487
1488void soinfo::set_st_dev(dev_t dev) {
1489  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1490    return;
1491  }
1492
1493  st_dev = dev;
1494}
1495
1496void soinfo::set_st_ino(ino_t ino) {
1497  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1498    return;
1499  }
1500
1501  st_ino = ino;
1502}
1503
1504dev_t soinfo::get_st_dev() {
1505  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1506    return 0;
1507  }
1508
1509  return st_dev;
1510};
1511
1512ino_t soinfo::get_st_ino() {
1513  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1514    return 0;
1515  }
1516
1517  return st_ino;
1518}
1519
1520// This is a return on get_children() in case
1521// 'this->flags' does not have FLAG_NEW_SOINFO set.
1522static soinfo::soinfo_list_t g_empty_list;
1523
1524soinfo::soinfo_list_t& soinfo::get_children() {
1525  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1526    return g_empty_list;
1527  }
1528
1529  return this->children;
1530}
1531
1532/* Force any of the closed stdin, stdout and stderr to be associated with
1533   /dev/null. */
1534static int nullify_closed_stdio() {
1535    int dev_null, i, status;
1536    int return_value = 0;
1537
1538    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1539    if (dev_null < 0) {
1540        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1541        return -1;
1542    }
1543    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1544
1545    /* If any of the stdio file descriptors is valid and not associated
1546       with /dev/null, dup /dev/null to it.  */
1547    for (i = 0; i < 3; i++) {
1548        /* If it is /dev/null already, we are done. */
1549        if (i == dev_null) {
1550            continue;
1551        }
1552
1553        TRACE("[ Nullifying stdio file descriptor %d]", i);
1554        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1555
1556        /* If file is opened, we are good. */
1557        if (status != -1) {
1558            continue;
1559        }
1560
1561        /* The only error we allow is that the file descriptor does not
1562           exist, in which case we dup /dev/null to it. */
1563        if (errno != EBADF) {
1564            DL_ERR("fcntl failed: %s", strerror(errno));
1565            return_value = -1;
1566            continue;
1567        }
1568
1569        /* Try dupping /dev/null to this stdio file descriptor and
1570           repeat if there is a signal.  Note that any errors in closing
1571           the stdio descriptor are lost.  */
1572        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1573        if (status < 0) {
1574            DL_ERR("dup2 failed: %s", strerror(errno));
1575            return_value = -1;
1576            continue;
1577        }
1578    }
1579
1580    /* If /dev/null is not one of the stdio file descriptors, close it. */
1581    if (dev_null > 2) {
1582        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1583        status = TEMP_FAILURE_RETRY(close(dev_null));
1584        if (status == -1) {
1585            DL_ERR("close failed: %s", strerror(errno));
1586            return_value = -1;
1587        }
1588    }
1589
1590    return return_value;
1591}
1592
1593static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1594    /* "base" might wrap around UINT32_MAX. */
1595    ElfW(Addr) base = si->load_bias;
1596    const ElfW(Phdr)* phdr = si->phdr;
1597    int phnum = si->phnum;
1598    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1599
1600    /* We can't debug anything until the linker is relocated */
1601    if (!relocating_linker) {
1602        INFO("[ linking %s ]", si->name);
1603        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1604    }
1605
1606    /* Extract dynamic section */
1607    size_t dynamic_count;
1608    ElfW(Word) dynamic_flags;
1609    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1610                                   &dynamic_count, &dynamic_flags);
1611    if (si->dynamic == NULL) {
1612        if (!relocating_linker) {
1613            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1614        }
1615        return false;
1616    } else {
1617        if (!relocating_linker) {
1618            DEBUG("dynamic = %p", si->dynamic);
1619        }
1620    }
1621
1622#if defined(__arm__)
1623    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1624                                    &si->ARM_exidx, &si->ARM_exidx_count);
1625#endif
1626
1627    // Extract useful information from dynamic section.
1628    uint32_t needed_count = 0;
1629    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1630        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1631              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1632        switch (d->d_tag) {
1633        case DT_HASH:
1634            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1635            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1636            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1637            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1638            break;
1639        case DT_STRTAB:
1640            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1641            break;
1642        case DT_SYMTAB:
1643            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1644            break;
1645#if !defined(__LP64__)
1646        case DT_PLTREL:
1647            if (d->d_un.d_val != DT_REL) {
1648                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1649                return false;
1650            }
1651            break;
1652#endif
1653        case DT_JMPREL:
1654#if defined(USE_RELA)
1655            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1656#else
1657            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1658#endif
1659            break;
1660        case DT_PLTRELSZ:
1661#if defined(USE_RELA)
1662            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1663#else
1664            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1665#endif
1666            break;
1667#if defined(__mips__)
1668        case DT_PLTGOT:
1669            // Used by mips and mips64.
1670            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1671            break;
1672#endif
1673        case DT_DEBUG:
1674            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1675            // if the dynamic table is writable
1676// FIXME: not working currently for N64
1677// The flags for the LOAD and DYNAMIC program headers do not agree.
1678// The LOAD section containng the dynamic table has been mapped as
1679// read-only, but the DYNAMIC header claims it is writable.
1680#if !(defined(__mips__) && defined(__LP64__))
1681            if ((dynamic_flags & PF_W) != 0) {
1682                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1683            }
1684            break;
1685#endif
1686#if defined(USE_RELA)
1687         case DT_RELA:
1688            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1689            break;
1690         case DT_RELASZ:
1691            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1692            break;
1693        case DT_REL:
1694            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1695            return false;
1696        case DT_RELSZ:
1697            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1698            return false;
1699#else
1700        case DT_REL:
1701            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1702            break;
1703        case DT_RELSZ:
1704            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1705            break;
1706         case DT_RELA:
1707            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1708            return false;
1709#endif
1710        case DT_INIT:
1711            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1712            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1713            break;
1714        case DT_FINI:
1715            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1716            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1717            break;
1718        case DT_INIT_ARRAY:
1719            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1720            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1721            break;
1722        case DT_INIT_ARRAYSZ:
1723            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1724            break;
1725        case DT_FINI_ARRAY:
1726            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1727            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1728            break;
1729        case DT_FINI_ARRAYSZ:
1730            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1731            break;
1732        case DT_PREINIT_ARRAY:
1733            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1734            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1735            break;
1736        case DT_PREINIT_ARRAYSZ:
1737            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1738            break;
1739        case DT_TEXTREL:
1740#if defined(__LP64__)
1741            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1742            return false;
1743#else
1744            si->has_text_relocations = true;
1745            break;
1746#endif
1747        case DT_SYMBOLIC:
1748            si->has_DT_SYMBOLIC = true;
1749            break;
1750        case DT_NEEDED:
1751            ++needed_count;
1752            break;
1753        case DT_FLAGS:
1754            if (d->d_un.d_val & DF_TEXTREL) {
1755#if defined(__LP64__)
1756                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1757                return false;
1758#else
1759                si->has_text_relocations = true;
1760#endif
1761            }
1762            if (d->d_un.d_val & DF_SYMBOLIC) {
1763                si->has_DT_SYMBOLIC = true;
1764            }
1765            break;
1766#if defined(__mips__)
1767        case DT_STRSZ:
1768        case DT_SYMENT:
1769        case DT_RELENT:
1770             break;
1771        case DT_MIPS_RLD_MAP:
1772            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1773            {
1774              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1775              *dp = &_r_debug;
1776            }
1777            break;
1778        case DT_MIPS_RLD_VERSION:
1779        case DT_MIPS_FLAGS:
1780        case DT_MIPS_BASE_ADDRESS:
1781        case DT_MIPS_UNREFEXTNO:
1782            break;
1783
1784        case DT_MIPS_SYMTABNO:
1785            si->mips_symtabno = d->d_un.d_val;
1786            break;
1787
1788        case DT_MIPS_LOCAL_GOTNO:
1789            si->mips_local_gotno = d->d_un.d_val;
1790            break;
1791
1792        case DT_MIPS_GOTSYM:
1793            si->mips_gotsym = d->d_un.d_val;
1794            break;
1795#endif
1796
1797        default:
1798            DEBUG("Unused DT entry: type %p arg %p",
1799                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1800            break;
1801        }
1802    }
1803
1804    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1805          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1806
1807    // Sanity checks.
1808    if (relocating_linker && needed_count != 0) {
1809        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1810        return false;
1811    }
1812    if (si->nbucket == 0) {
1813        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1814        return false;
1815    }
1816    if (si->strtab == 0) {
1817        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1818        return false;
1819    }
1820    if (si->symtab == 0) {
1821        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1822        return false;
1823    }
1824
1825    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1826    if (si->flags & FLAG_EXE) {
1827        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1828        size_t preload_count = 0;
1829        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1830            soinfo* lsi = find_library(g_ld_preload_names[i], 0, NULL);
1831            if (lsi != NULL) {
1832                g_ld_preloads[preload_count++] = lsi;
1833            } else {
1834                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1835                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1836                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1837            }
1838        }
1839    }
1840
1841    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1842    soinfo** pneeded = needed;
1843
1844    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1845        if (d->d_tag == DT_NEEDED) {
1846            const char* library_name = si->strtab + d->d_un.d_val;
1847            DEBUG("%s needs %s", si->name, library_name);
1848            soinfo* lsi = find_library(library_name, 0, NULL);
1849            if (lsi == NULL) {
1850                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1851                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1852                       library_name, si->name, tmp_err_buf);
1853                return false;
1854            }
1855
1856            si->add_child(lsi);
1857            *pneeded++ = lsi;
1858        }
1859    }
1860    *pneeded = NULL;
1861
1862#if !defined(__LP64__)
1863    if (si->has_text_relocations) {
1864        // Make segments writable to allow text relocations to work properly. We will later call
1865        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1866#if !defined(__i386__) // The platform itself has too many text relocations on x86.
1867        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1868                "security hardening. Please fix.", si->name);
1869#endif
1870        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1871            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1872                   si->name, strerror(errno));
1873            return false;
1874        }
1875    }
1876#endif
1877
1878#if defined(USE_RELA)
1879    if (si->plt_rela != NULL) {
1880        DEBUG("[ relocating %s plt ]\n", si->name);
1881        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1882            return false;
1883        }
1884    }
1885    if (si->rela != NULL) {
1886        DEBUG("[ relocating %s ]\n", si->name);
1887        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1888            return false;
1889        }
1890    }
1891#else
1892    if (si->plt_rel != NULL) {
1893        DEBUG("[ relocating %s plt ]", si->name);
1894        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1895            return false;
1896        }
1897    }
1898    if (si->rel != NULL) {
1899        DEBUG("[ relocating %s ]", si->name);
1900        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1901            return false;
1902        }
1903    }
1904#endif
1905
1906#if defined(__mips__)
1907    if (!mips_relocate_got(si, needed)) {
1908        return false;
1909    }
1910#endif
1911
1912    si->flags |= FLAG_LINKED;
1913    DEBUG("[ finished linking %s ]", si->name);
1914
1915#if !defined(__LP64__)
1916    if (si->has_text_relocations) {
1917        // All relocations are done, we can protect our segments back to read-only.
1918        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1919            DL_ERR("can't protect segments for \"%s\": %s",
1920                   si->name, strerror(errno));
1921            return false;
1922        }
1923    }
1924#endif
1925
1926    /* We can also turn on GNU RELRO protection */
1927    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1928        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1929               si->name, strerror(errno));
1930        return false;
1931    }
1932
1933    /* Handle serializing/sharing the RELRO segment */
1934    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1935      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
1936                                         extinfo->relro_fd) < 0) {
1937        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
1938               si->name, strerror(errno));
1939        return false;
1940      }
1941    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
1942      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
1943                                   extinfo->relro_fd) < 0) {
1944        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
1945               si->name, strerror(errno));
1946        return false;
1947      }
1948    }
1949
1950    notify_gdb_of_load(si);
1951    return true;
1952}
1953
1954/*
1955 * This function add vdso to internal dso list.
1956 * It helps to stack unwinding through signal handlers.
1957 * Also, it makes bionic more like glibc.
1958 */
1959static void add_vdso(KernelArgumentBlock& args __unused) {
1960#if defined(AT_SYSINFO_EHDR)
1961  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
1962  if (ehdr_vdso == NULL) {
1963    return;
1964  }
1965
1966  soinfo* si = soinfo_alloc("[vdso]", NULL);
1967
1968  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
1969  si->phnum = ehdr_vdso->e_phnum;
1970  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
1971  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
1972  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
1973
1974  soinfo_link_image(si, NULL);
1975#endif
1976}
1977
1978/*
1979 * This is linker soinfo for GDB. See details below.
1980 */
1981static soinfo linker_soinfo_for_gdb;
1982
1983/* gdb expects the linker to be in the debug shared object list.
1984 * Without this, gdb has trouble locating the linker's ".text"
1985 * and ".plt" sections. Gdb could also potentially use this to
1986 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
1987 * Don't use soinfo_alloc(), because the linker shouldn't
1988 * be on the soinfo list.
1989 */
1990static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
1991#if defined(__LP64__)
1992  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
1993#else
1994  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
1995#endif
1996  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
1997  linker_soinfo_for_gdb.base = linker_base;
1998
1999  /*
2000   * Set the dynamic field in the link map otherwise gdb will complain with
2001   * the following:
2002   *   warning: .dynamic section for "/system/bin/linker" is not at the
2003   *   expected address (wrong library or version mismatch?)
2004   */
2005  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2006  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2007  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2008                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
2009  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2010}
2011
2012/*
2013 * This code is called after the linker has linked itself and
2014 * fixed it's own GOT. It is safe to make references to externs
2015 * and other non-local data at this point.
2016 */
2017static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2018    /* NOTE: we store the args pointer on a special location
2019     *       of the temporary TLS area in order to pass it to
2020     *       the C Library's runtime initializer.
2021     *
2022     *       The initializer must clear the slot and reset the TLS
2023     *       to point to a different location to ensure that no other
2024     *       shared library constructor can access it.
2025     */
2026  __libc_init_tls(args);
2027
2028#if TIMING
2029    struct timeval t0, t1;
2030    gettimeofday(&t0, 0);
2031#endif
2032
2033    // Initialize environment functions, and get to the ELF aux vectors table.
2034    linker_env_init(args);
2035
2036    // If this is a setuid/setgid program, close the security hole described in
2037    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2038    if (get_AT_SECURE()) {
2039        nullify_closed_stdio();
2040    }
2041
2042    debuggerd_init();
2043
2044    // Get a few environment variables.
2045    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2046    if (LD_DEBUG != NULL) {
2047      g_ld_debug_verbosity = atoi(LD_DEBUG);
2048    }
2049
2050    // Normally, these are cleaned by linker_env_init, but the test
2051    // doesn't cost us anything.
2052    const char* ldpath_env = NULL;
2053    const char* ldpreload_env = NULL;
2054    if (!get_AT_SECURE()) {
2055      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2056      ldpreload_env = linker_env_get("LD_PRELOAD");
2057    }
2058
2059    // Linker does not call constructors for its own
2060    // global variables so we need to initialize
2061    // the allocators explicitly.
2062    g_soinfo_allocator.init();
2063    g_soinfo_links_allocator.init();
2064
2065    INFO("[ android linker & debugger ]");
2066
2067    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2068    if (si == NULL) {
2069        exit(EXIT_FAILURE);
2070    }
2071
2072    /* bootstrap the link map, the main exe always needs to be first */
2073    si->flags |= FLAG_EXE;
2074    link_map* map = &(si->link_map_head);
2075
2076    map->l_addr = 0;
2077    map->l_name = args.argv[0];
2078    map->l_prev = NULL;
2079    map->l_next = NULL;
2080
2081    _r_debug.r_map = map;
2082    r_debug_tail = map;
2083
2084    init_linker_info_for_gdb(linker_base);
2085
2086    // Extract information passed from the kernel.
2087    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2088    si->phnum = args.getauxval(AT_PHNUM);
2089    si->entry = args.getauxval(AT_ENTRY);
2090
2091    /* Compute the value of si->base. We can't rely on the fact that
2092     * the first entry is the PHDR because this will not be true
2093     * for certain executables (e.g. some in the NDK unit test suite)
2094     */
2095    si->base = 0;
2096    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2097    si->load_bias = 0;
2098    for (size_t i = 0; i < si->phnum; ++i) {
2099      if (si->phdr[i].p_type == PT_PHDR) {
2100        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2101        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2102        break;
2103      }
2104    }
2105    si->dynamic = NULL;
2106    si->ref_count = 1;
2107
2108    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2109    if (elf_hdr->e_type != ET_DYN) {
2110        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2111        exit(EXIT_FAILURE);
2112    }
2113
2114    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2115    parse_LD_LIBRARY_PATH(ldpath_env);
2116    parse_LD_PRELOAD(ldpreload_env);
2117
2118    somain = si;
2119
2120    if (!soinfo_link_image(si, NULL)) {
2121        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2122        exit(EXIT_FAILURE);
2123    }
2124
2125    add_vdso(args);
2126
2127    si->CallPreInitConstructors();
2128
2129    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2130        g_ld_preloads[i]->CallConstructors();
2131    }
2132
2133    /* After the link_image, the si->load_bias is initialized.
2134     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2135     * We need to update this value for so exe here. So Unwind_Backtrace
2136     * for some arch like x86 could work correctly within so exe.
2137     */
2138    map->l_addr = si->load_bias;
2139    si->CallConstructors();
2140
2141#if TIMING
2142    gettimeofday(&t1, NULL);
2143    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2144               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2145               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2146#endif
2147#if STATS
2148    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2149           linker_stats.count[kRelocAbsolute],
2150           linker_stats.count[kRelocRelative],
2151           linker_stats.count[kRelocCopy],
2152           linker_stats.count[kRelocSymbol]);
2153#endif
2154#if COUNT_PAGES
2155    {
2156        unsigned n;
2157        unsigned i;
2158        unsigned count = 0;
2159        for (n = 0; n < 4096; n++) {
2160            if (bitmask[n]) {
2161                unsigned x = bitmask[n];
2162#if defined(__LP64__)
2163                for (i = 0; i < 32; i++) {
2164#else
2165                for (i = 0; i < 8; i++) {
2166#endif
2167                    if (x & 1) {
2168                        count++;
2169                    }
2170                    x >>= 1;
2171                }
2172            }
2173        }
2174        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2175    }
2176#endif
2177
2178#if TIMING || STATS || COUNT_PAGES
2179    fflush(stdout);
2180#endif
2181
2182    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2183    return si->entry;
2184}
2185
2186/* Compute the load-bias of an existing executable. This shall only
2187 * be used to compute the load bias of an executable or shared library
2188 * that was loaded by the kernel itself.
2189 *
2190 * Input:
2191 *    elf    -> address of ELF header, assumed to be at the start of the file.
2192 * Return:
2193 *    load bias, i.e. add the value of any p_vaddr in the file to get
2194 *    the corresponding address in memory.
2195 */
2196static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2197  ElfW(Addr) offset = elf->e_phoff;
2198  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2199  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2200
2201  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2202    if (phdr->p_type == PT_LOAD) {
2203      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2204    }
2205  }
2206  return 0;
2207}
2208
2209/*
2210 * This is the entry point for the linker, called from begin.S. This
2211 * method is responsible for fixing the linker's own relocations, and
2212 * then calling __linker_init_post_relocation().
2213 *
2214 * Because this method is called before the linker has fixed it's own
2215 * relocations, any attempt to reference an extern variable, extern
2216 * function, or other GOT reference will generate a segfault.
2217 */
2218extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2219  // Initialize static variables.
2220  solist = get_libdl_info();
2221  sonext = get_libdl_info();
2222
2223  KernelArgumentBlock args(raw_args);
2224
2225  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2226  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2227  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2228
2229  soinfo linker_so;
2230  memset(&linker_so, 0, sizeof(soinfo));
2231
2232  strcpy(linker_so.name, "[dynamic linker]");
2233  linker_so.base = linker_addr;
2234  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2235  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2236  linker_so.dynamic = NULL;
2237  linker_so.phdr = phdr;
2238  linker_so.phnum = elf_hdr->e_phnum;
2239  linker_so.flags |= FLAG_LINKER;
2240
2241  if (!soinfo_link_image(&linker_so, NULL)) {
2242    // It would be nice to print an error message, but if the linker
2243    // can't link itself, there's no guarantee that we'll be able to
2244    // call write() (because it involves a GOT reference). We may as
2245    // well try though...
2246    const char* msg = "CANNOT LINK EXECUTABLE: ";
2247    write(2, msg, strlen(msg));
2248    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2249    write(2, "\n", 1);
2250    _exit(EXIT_FAILURE);
2251  }
2252
2253  // We have successfully fixed our own relocations. It's safe to run
2254  // the main part of the linker now.
2255  args.abort_message_ptr = &g_abort_message;
2256  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2257
2258  protect_data(PROT_READ);
2259
2260  // Return the address that the calling assembly stub should jump to.
2261  return start_address;
2262}
2263