linker.cpp revision aa836f731016f7f1099d6a6ce47a35f1ec17dd4e
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 * - are we doing everything we should for ARM_COPY relocations?
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        const char* msg = "ERROR: " #name " called from the dynamic linker!\n"; \
167        __libc_format_log(ANDROID_LOG_FATAL, "linker", "%s", msg); \
168        write(2, msg, strlen(msg)); \
169        abort(); \
170    }
171DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
172DISALLOW_ALLOCATION(void, free, (void* u __unused));
173DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
174DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
175
176static char tmp_err_buf[768];
177static char __linker_dl_err_buf[768];
178
179char* linker_get_error_buffer() {
180  return &__linker_dl_err_buf[0];
181}
182
183size_t linker_get_error_buffer_size() {
184  return sizeof(__linker_dl_err_buf);
185}
186
187/*
188 * This function is an empty stub where GDB locates a breakpoint to get notified
189 * about linker activity.
190 */
191extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
192
193static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
194static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
195static link_map* r_debug_tail = 0;
196
197static void insert_soinfo_into_debug_map(soinfo* info) {
198    // Copy the necessary fields into the debug structure.
199    link_map* map = &(info->link_map_head);
200    map->l_addr = info->load_bias;
201    map->l_name = reinterpret_cast<char*>(info->name);
202    map->l_ld = info->dynamic;
203
204    /* Stick the new library at the end of the list.
205     * gdb tends to care more about libc than it does
206     * about leaf libraries, and ordering it this way
207     * reduces the back-and-forth over the wire.
208     */
209    if (r_debug_tail) {
210        r_debug_tail->l_next = map;
211        map->l_prev = r_debug_tail;
212        map->l_next = 0;
213    } else {
214        _r_debug.r_map = map;
215        map->l_prev = 0;
216        map->l_next = 0;
217    }
218    r_debug_tail = map;
219}
220
221static void remove_soinfo_from_debug_map(soinfo* info) {
222    link_map* map = &(info->link_map_head);
223
224    if (r_debug_tail == map) {
225        r_debug_tail = map->l_prev;
226    }
227
228    if (map->l_prev) {
229        map->l_prev->l_next = map->l_next;
230    }
231    if (map->l_next) {
232        map->l_next->l_prev = map->l_prev;
233    }
234}
235
236static void notify_gdb_of_load(soinfo* info) {
237    if (info->flags & FLAG_EXE) {
238        // GDB already knows about the main executable
239        return;
240    }
241
242    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
243
244    _r_debug.r_state = r_debug::RT_ADD;
245    rtld_db_dlactivity();
246
247    insert_soinfo_into_debug_map(info);
248
249    _r_debug.r_state = r_debug::RT_CONSISTENT;
250    rtld_db_dlactivity();
251}
252
253static void notify_gdb_of_unload(soinfo* info) {
254    if (info->flags & FLAG_EXE) {
255        // GDB already knows about the main executable
256        return;
257    }
258
259    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
260
261    _r_debug.r_state = r_debug::RT_DELETE;
262    rtld_db_dlactivity();
263
264    remove_soinfo_from_debug_map(info);
265
266    _r_debug.r_state = r_debug::RT_CONSISTENT;
267    rtld_db_dlactivity();
268}
269
270void notify_gdb_of_libraries() {
271  _r_debug.r_state = r_debug::RT_ADD;
272  rtld_db_dlactivity();
273  _r_debug.r_state = r_debug::RT_CONSISTENT;
274  rtld_db_dlactivity();
275}
276
277LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
278  return g_soinfo_links_allocator.alloc();
279}
280
281void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
282  g_soinfo_links_allocator.free(entry);
283}
284
285static void protect_data(int protection) {
286  g_soinfo_allocator.protect_all(protection);
287  g_soinfo_links_allocator.protect_all(protection);
288}
289
290static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
291  if (strlen(name) >= SOINFO_NAME_LEN) {
292    DL_ERR("library name \"%s\" too long", name);
293    return NULL;
294  }
295
296  soinfo* si = g_soinfo_allocator.alloc();
297
298  // Initialize the new element.
299  memset(si, 0, sizeof(soinfo));
300  strlcpy(si->name, name, sizeof(si->name));
301  si->flags = FLAG_NEW_SOINFO;
302
303  if (file_stat != NULL) {
304    si->set_st_dev(file_stat->st_dev);
305    si->set_st_ino(file_stat->st_ino);
306  }
307
308  sonext->next = si;
309  sonext = si;
310
311  TRACE("name %s: allocated soinfo @ %p", name, si);
312  return si;
313}
314
315static void soinfo_free(soinfo* si) {
316    if (si == NULL) {
317        return;
318    }
319
320    if (si->base != 0 && si->size != 0) {
321      munmap(reinterpret_cast<void*>(si->base), si->size);
322    }
323
324    soinfo *prev = NULL, *trav;
325
326    TRACE("name %s: freeing soinfo @ %p", si->name, si);
327
328    for (trav = solist; trav != NULL; trav = trav->next) {
329        if (trav == si)
330            break;
331        prev = trav;
332    }
333    if (trav == NULL) {
334        /* si was not in solist */
335        DL_ERR("name \"%s\" is not in solist!", si->name);
336        return;
337    }
338
339    // clear links to/from si
340    si->remove_all_links();
341
342    /* prev will never be NULL, because the first entry in solist is
343       always the static libdl_info.
344    */
345    prev->next = si->next;
346    if (si == sonext) {
347        sonext = prev;
348    }
349
350    g_soinfo_allocator.free(si);
351}
352
353
354static void parse_path(const char* path, const char* delimiters,
355                       const char** array, char* buf, size_t buf_size, size_t max_count) {
356  if (path == NULL) {
357    return;
358  }
359
360  size_t len = strlcpy(buf, path, buf_size);
361
362  size_t i = 0;
363  char* buf_p = buf;
364  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
365    if (*array[i] != '\0') {
366      ++i;
367    }
368  }
369
370  // Forget the last path if we had to truncate; this occurs if the 2nd to
371  // last char isn't '\0' (i.e. wasn't originally a delimiter).
372  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
373    array[i - 1] = NULL;
374  } else {
375    array[i] = NULL;
376  }
377}
378
379static void parse_LD_LIBRARY_PATH(const char* path) {
380  parse_path(path, ":", g_ld_library_paths,
381             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
382}
383
384static void parse_LD_PRELOAD(const char* path) {
385  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
386  parse_path(path, " :", g_ld_preload_names,
387             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
388}
389
390#if defined(__arm__)
391
392/* For a given PC, find the .so that it belongs to.
393 * Returns the base address of the .ARM.exidx section
394 * for that .so, and the number of 8-byte entries
395 * in that section (via *pcount).
396 *
397 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
398 *
399 * This function is exposed via dlfcn.cpp and libdl.so.
400 */
401_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
402    unsigned addr = (unsigned)pc;
403
404    for (soinfo* si = solist; si != 0; si = si->next) {
405        if ((addr >= si->base) && (addr < (si->base + si->size))) {
406            *pcount = si->ARM_exidx_count;
407            return (_Unwind_Ptr)si->ARM_exidx;
408        }
409    }
410    *pcount = 0;
411    return NULL;
412}
413
414#endif
415
416/* Here, we only have to provide a callback to iterate across all the
417 * loaded libraries. gcc_eh does the rest. */
418int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
419    int rv = 0;
420    for (soinfo* si = solist; si != NULL; si = si->next) {
421        dl_phdr_info dl_info;
422        dl_info.dlpi_addr = si->link_map_head.l_addr;
423        dl_info.dlpi_name = si->link_map_head.l_name;
424        dl_info.dlpi_phdr = si->phdr;
425        dl_info.dlpi_phnum = si->phnum;
426        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
427        if (rv != 0) {
428            break;
429        }
430    }
431    return rv;
432}
433
434static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
435  ElfW(Sym)* symtab = si->symtab;
436  const char* strtab = si->strtab;
437
438  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
439             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
440
441  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
442    ElfW(Sym)* s = symtab + n;
443    if (strcmp(strtab + s->st_name, name)) continue;
444
445    /* only concern ourselves with global and weak symbol definitions */
446    switch (ELF_ST_BIND(s->st_info)) {
447      case STB_GLOBAL:
448      case STB_WEAK:
449        if (s->st_shndx == SHN_UNDEF) {
450        continue;
451      }
452
453      TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
454                 name, si->name, reinterpret_cast<void*>(s->st_value),
455                 static_cast<size_t>(s->st_size));
456      return s;
457    }
458  }
459
460  return NULL;
461}
462
463static unsigned elfhash(const char* _name) {
464    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
465    unsigned h = 0, g;
466
467    while (*name) {
468        h = (h << 4) + *name++;
469        g = h & 0xf0000000;
470        h ^= g;
471        h ^= g >> 24;
472    }
473    return h;
474}
475
476static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
477    unsigned elf_hash = elfhash(name);
478    ElfW(Sym)* s = NULL;
479
480    if (si != NULL && somain != NULL) {
481        /*
482         * Local scope is executable scope. Just start looking into it right away
483         * for the shortcut.
484         */
485
486        if (si == somain) {
487            s = soinfo_elf_lookup(si, elf_hash, name);
488            if (s != NULL) {
489                *lsi = si;
490                goto done;
491            }
492        } else {
493            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
494
495            /*
496             * If this object was built with symbolic relocations disabled, the
497             * first place to look to resolve external references is the main
498             * executable.
499             */
500
501            if (!si->has_DT_SYMBOLIC) {
502                DEBUG("%s: looking up %s in executable %s",
503                      si->name, name, somain->name);
504                s = soinfo_elf_lookup(somain, elf_hash, name);
505                if (s != NULL) {
506                    *lsi = somain;
507                    goto done;
508                }
509            }
510
511            /* Look for symbols in the local scope (the object who is
512             * searching). This happens with C++ templates on x86 for some
513             * reason.
514             *
515             * Notes on weak symbols:
516             * The ELF specs are ambiguous about treatment of weak definitions in
517             * dynamic linking.  Some systems return the first definition found
518             * and some the first non-weak definition.   This is system dependent.
519             * Here we return the first definition found for simplicity.  */
520
521            s = soinfo_elf_lookup(si, elf_hash, name);
522            if (s != NULL) {
523                *lsi = si;
524                goto done;
525            }
526
527            /*
528             * If this object was built with -Bsymbolic and symbol is not found
529             * in the local scope, try to find the symbol in the main executable.
530             */
531
532            if (si->has_DT_SYMBOLIC) {
533                DEBUG("%s: looking up %s in executable %s after local scope",
534                      si->name, name, somain->name);
535                s = soinfo_elf_lookup(somain, elf_hash, name);
536                if (s != NULL) {
537                    *lsi = somain;
538                    goto done;
539                }
540            }
541        }
542    }
543
544    /* Next, look for it in the preloads list */
545    for (int i = 0; g_ld_preloads[i] != NULL; i++) {
546        s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
547        if (s != NULL) {
548            *lsi = g_ld_preloads[i];
549            goto done;
550        }
551    }
552
553    for (int i = 0; needed[i] != NULL; i++) {
554        DEBUG("%s: looking up %s in %s",
555              si->name, name, needed[i]->name);
556        s = soinfo_elf_lookup(needed[i], elf_hash, name);
557        if (s != NULL) {
558            *lsi = needed[i];
559            goto done;
560        }
561    }
562
563done:
564    if (s != NULL) {
565        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
566                   "found in %s, base = %p, load bias = %p",
567                   si->name, name, reinterpret_cast<void*>(s->st_value),
568                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
569                   reinterpret_cast<void*>((*lsi)->load_bias));
570        return s;
571    }
572
573    return NULL;
574}
575
576/* This is used by dlsym(3).  It performs symbol lookup only within the
577   specified soinfo object and not in any of its dependencies.
578
579   TODO: Only looking in the specified soinfo seems wrong. dlsym(3) says
580   that it should do a breadth first search through the dependency
581   tree. This agrees with the ELF spec (aka System V Application
582   Binary Interface) where in Chapter 5 it discuss resolving "Shared
583   Object Dependencies" in breadth first search order.
584 */
585ElfW(Sym)* dlsym_handle_lookup(soinfo* si, const char* name) {
586    return soinfo_elf_lookup(si, elfhash(name), name);
587}
588
589/* This is used by dlsym(3) to performs a global symbol lookup. If the
590   start value is null (for RTLD_DEFAULT), the search starts at the
591   beginning of the global solist. Otherwise the search starts at the
592   specified soinfo (for RTLD_NEXT).
593 */
594ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
595  unsigned elf_hash = elfhash(name);
596
597  if (start == NULL) {
598    start = solist;
599  }
600
601  ElfW(Sym)* s = NULL;
602  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
603    s = soinfo_elf_lookup(si, elf_hash, name);
604    if (s != NULL) {
605      *found = si;
606      break;
607    }
608  }
609
610  if (s != NULL) {
611    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
612               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
613  }
614
615  return s;
616}
617
618soinfo* find_containing_library(const void* p) {
619  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
620  for (soinfo* si = solist; si != NULL; si = si->next) {
621    if (address >= si->base && address - si->base < si->size) {
622      return si;
623    }
624  }
625  return NULL;
626}
627
628ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
629  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
630
631  // Search the library's symbol table for any defined symbol which
632  // contains this address.
633  for (size_t i = 0; i < si->nchain; ++i) {
634    ElfW(Sym)* sym = &si->symtab[i];
635    if (sym->st_shndx != SHN_UNDEF &&
636        soaddr >= sym->st_value &&
637        soaddr < sym->st_value + sym->st_size) {
638      return sym;
639    }
640  }
641
642  return NULL;
643}
644
645static int open_library_on_path(const char* name, const char* const paths[]) {
646  char buf[512];
647  for (size_t i = 0; paths[i] != NULL; ++i) {
648    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
649    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
650      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
651      continue;
652    }
653    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
654    if (fd != -1) {
655      return fd;
656    }
657  }
658  return -1;
659}
660
661static int open_library(const char* name) {
662  TRACE("[ opening %s ]", name);
663
664  // If the name contains a slash, we should attempt to open it directly and not search the paths.
665  if (strchr(name, '/') != NULL) {
666    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
667    if (fd != -1) {
668      return fd;
669    }
670    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
671#if defined(__LP64__)
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, 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        if ((si->flags & FLAG_EXE) == 0) {
1041            /*
1042              * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1043              *
1044              * Section 4.7.1.10 "Dynamic relocations"
1045              * R_AARCH64_COPY may only appear in executable objects where e_type is
1046              * set to ET_EXEC.
1047              *
1048              * FLAG_EXE is set for both ET_DYN and ET_EXEC executables.
1049              * We should explicitly disallow ET_DYN executables from having
1050              * R_AARCH64_COPY relocations.
1051              */
1052            DL_ERR("%s R_AARCH64_COPY relocations only supported for ET_EXEC", si->name);
1053            return -1;
1054        }
1055        count_relocation(kRelocCopy);
1056        MARK(rela->r_offset);
1057        TRACE_TYPE(RELO, "RELO COPY %16llx <- %lld @ %16llx %s\n",
1058                   reloc,
1059                   s->st_size,
1060                   (sym_addr + rela->r_addend),
1061                   sym_name);
1062        if (reloc == (sym_addr + rela->r_addend)) {
1063            ElfW(Sym)* src = soinfo_do_lookup(NULL, sym_name, &lsi, needed);
1064
1065            if (src == NULL) {
1066                DL_ERR("%s R_AARCH64_COPY relocation source cannot be resolved", si->name);
1067                return -1;
1068            }
1069            if (lsi->has_DT_SYMBOLIC) {
1070                DL_ERR("%s invalid R_AARCH64_COPY relocation against DT_SYMBOLIC shared "
1071                       "library %s (built with -Bsymbolic?)", si->name, lsi->name);
1072                return -1;
1073            }
1074            if (s->st_size < src->st_size) {
1075                DL_ERR("%s R_AARCH64_COPY relocation size mismatch (%lld < %lld)",
1076                       si->name, s->st_size, src->st_size);
1077                return -1;
1078            }
1079            memcpy(reinterpret_cast<void*>(reloc),
1080                   reinterpret_cast<void*>(src->st_value + lsi->load_bias), src->st_size);
1081        } else {
1082            DL_ERR("%s R_AARCH64_COPY relocation target cannot be resolved", si->name);
1083            return -1;
1084        }
1085        break;
1086    case R_AARCH64_TLS_TPREL64:
1087        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1088                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1089        break;
1090    case R_AARCH64_TLS_DTPREL32:
1091        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1092                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1093        break;
1094#elif defined(__x86_64__)
1095    case R_X86_64_JUMP_SLOT:
1096      count_relocation(kRelocAbsolute);
1097      MARK(rela->r_offset);
1098      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1099                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1100      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1101      break;
1102    case R_X86_64_GLOB_DAT:
1103      count_relocation(kRelocAbsolute);
1104      MARK(rela->r_offset);
1105      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1106                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1107      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1108      break;
1109    case R_X86_64_RELATIVE:
1110      count_relocation(kRelocRelative);
1111      MARK(rela->r_offset);
1112      if (sym) {
1113        DL_ERR("odd RELATIVE form...");
1114        return -1;
1115      }
1116      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1117                 static_cast<size_t>(si->base));
1118      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1119      break;
1120    case R_X86_64_32:
1121      count_relocation(kRelocRelative);
1122      MARK(rela->r_offset);
1123      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1124                 static_cast<size_t>(sym_addr), sym_name);
1125      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1126      break;
1127    case R_X86_64_64:
1128      count_relocation(kRelocRelative);
1129      MARK(rela->r_offset);
1130      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1131                 static_cast<size_t>(sym_addr), sym_name);
1132      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1133      break;
1134    case R_X86_64_PC32:
1135      count_relocation(kRelocRelative);
1136      MARK(rela->r_offset);
1137      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1138                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1139                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1140      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1141      break;
1142#endif
1143
1144    default:
1145      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1146      return -1;
1147    }
1148  }
1149  return 0;
1150}
1151
1152#else // REL, not RELA.
1153
1154static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1155    ElfW(Sym)* s;
1156    soinfo* lsi;
1157
1158    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1159        unsigned type = ELFW(R_TYPE)(rel->r_info);
1160        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1161        unsigned sym = ELFW(R_SYM)(rel->r_info);
1162        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1163        ElfW(Addr) sym_addr = 0;
1164        const char* sym_name = NULL;
1165
1166        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1167        if (type == 0) { // R_*_NONE
1168            continue;
1169        }
1170        if (sym != 0) {
1171            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1172            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1173            if (s == NULL) {
1174                // We only allow an undefined symbol if this is a weak reference...
1175                s = &si->symtab[sym];
1176                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1177                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1178                    return -1;
1179                }
1180
1181                /* IHI0044C AAELF 4.5.1.1:
1182
1183                   Libraries are not searched to resolve weak references.
1184                   It is not an error for a weak reference to remain
1185                   unsatisfied.
1186
1187                   During linking, the value of an undefined weak reference is:
1188                   - Zero if the relocation type is absolute
1189                   - The address of the place if the relocation is pc-relative
1190                   - The address of nominal base address if the relocation
1191                     type is base-relative.
1192                  */
1193
1194                switch (type) {
1195#if defined(__arm__)
1196                case R_ARM_JUMP_SLOT:
1197                case R_ARM_GLOB_DAT:
1198                case R_ARM_ABS32:
1199                case R_ARM_RELATIVE:    /* Don't care. */
1200                    // sym_addr was initialized to be zero above or relocation
1201                    // code below does not care about value of sym_addr.
1202                    // No need to do anything.
1203                    break;
1204#elif defined(__i386__)
1205                case R_386_JMP_SLOT:
1206                case R_386_GLOB_DAT:
1207                case R_386_32:
1208                case R_386_RELATIVE:    /* Don't care. */
1209                    // sym_addr was initialized to be zero above or relocation
1210                    // code below does not care about value of sym_addr.
1211                    // No need to do anything.
1212                    break;
1213                case R_386_PC32:
1214                    sym_addr = reloc;
1215                    break;
1216#endif
1217
1218#if defined(__arm__)
1219                case R_ARM_COPY:
1220                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1221#endif
1222                default:
1223                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1224                    return -1;
1225                }
1226            } else {
1227                // We got a definition.
1228                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1229            }
1230            count_relocation(kRelocSymbol);
1231        } else {
1232            s = NULL;
1233        }
1234
1235        switch (type) {
1236#if defined(__arm__)
1237        case R_ARM_JUMP_SLOT:
1238            count_relocation(kRelocAbsolute);
1239            MARK(rel->r_offset);
1240            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1241            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1242            break;
1243        case R_ARM_GLOB_DAT:
1244            count_relocation(kRelocAbsolute);
1245            MARK(rel->r_offset);
1246            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1247            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1248            break;
1249        case R_ARM_ABS32:
1250            count_relocation(kRelocAbsolute);
1251            MARK(rel->r_offset);
1252            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1253            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1254            break;
1255        case R_ARM_REL32:
1256            count_relocation(kRelocRelative);
1257            MARK(rel->r_offset);
1258            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1259                       reloc, sym_addr, rel->r_offset, sym_name);
1260            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1261            break;
1262        case R_ARM_COPY:
1263            if ((si->flags & FLAG_EXE) == 0) {
1264                /*
1265                 * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1266                 *
1267                 * Section 4.7.1.10 "Dynamic relocations"
1268                 * R_ARM_COPY may only appear in executable objects where e_type is
1269                 * set to ET_EXEC.
1270                 *
1271                 * TODO: FLAG_EXE is set for both ET_DYN and ET_EXEC executables.
1272                 * We should explicitly disallow ET_DYN executables from having
1273                 * R_ARM_COPY relocations.
1274                 */
1275                DL_ERR("%s R_ARM_COPY relocations only supported for ET_EXEC", si->name);
1276                return -1;
1277            }
1278            count_relocation(kRelocCopy);
1279            MARK(rel->r_offset);
1280            TRACE_TYPE(RELO, "RELO %08x <- %d @ %08x %s", reloc, s->st_size, sym_addr, sym_name);
1281            if (reloc == sym_addr) {
1282                ElfW(Sym)* src = soinfo_do_lookup(NULL, sym_name, &lsi, needed);
1283
1284                if (src == NULL) {
1285                    DL_ERR("%s R_ARM_COPY relocation source cannot be resolved", si->name);
1286                    return -1;
1287                }
1288                if (lsi->has_DT_SYMBOLIC) {
1289                    DL_ERR("%s invalid R_ARM_COPY relocation against DT_SYMBOLIC shared "
1290                           "library %s (built with -Bsymbolic?)", si->name, lsi->name);
1291                    return -1;
1292                }
1293                if (s->st_size < src->st_size) {
1294                    DL_ERR("%s R_ARM_COPY relocation size mismatch (%d < %d)",
1295                           si->name, s->st_size, src->st_size);
1296                    return -1;
1297                }
1298                memcpy(reinterpret_cast<void*>(reloc),
1299                       reinterpret_cast<void*>(src->st_value + lsi->load_bias), src->st_size);
1300            } else {
1301                DL_ERR("%s R_ARM_COPY relocation target cannot be resolved", si->name);
1302                return -1;
1303            }
1304            break;
1305#elif defined(__i386__)
1306        case R_386_JMP_SLOT:
1307            count_relocation(kRelocAbsolute);
1308            MARK(rel->r_offset);
1309            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1310            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1311            break;
1312        case R_386_GLOB_DAT:
1313            count_relocation(kRelocAbsolute);
1314            MARK(rel->r_offset);
1315            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1316            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1317            break;
1318        case R_386_32:
1319            count_relocation(kRelocRelative);
1320            MARK(rel->r_offset);
1321            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1322            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1323            break;
1324        case R_386_PC32:
1325            count_relocation(kRelocRelative);
1326            MARK(rel->r_offset);
1327            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1328                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1329            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1330            break;
1331#elif defined(__mips__)
1332        case R_MIPS_REL32:
1333#if defined(__LP64__)
1334            // MIPS Elf64_Rel entries contain compound relocations
1335            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1336            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1337                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1338                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1339                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1340                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1341                return -1;
1342            }
1343#endif
1344            count_relocation(kRelocAbsolute);
1345            MARK(rel->r_offset);
1346            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1347                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1348            if (s) {
1349                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1350            } else {
1351                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1352            }
1353            break;
1354#endif
1355
1356#if defined(__arm__)
1357        case R_ARM_RELATIVE:
1358#elif defined(__i386__)
1359        case R_386_RELATIVE:
1360#endif
1361            count_relocation(kRelocRelative);
1362            MARK(rel->r_offset);
1363            if (sym) {
1364                DL_ERR("odd RELATIVE form...");
1365                return -1;
1366            }
1367            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1368                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1369            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1370            break;
1371
1372        default:
1373            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1374            return -1;
1375        }
1376    }
1377    return 0;
1378}
1379#endif
1380
1381#if defined(__mips__)
1382static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1383    ElfW(Addr)** got = si->plt_got;
1384    if (got == NULL) {
1385        return true;
1386    }
1387    unsigned local_gotno = si->mips_local_gotno;
1388    unsigned gotsym = si->mips_gotsym;
1389    unsigned symtabno = si->mips_symtabno;
1390    ElfW(Sym)* symtab = si->symtab;
1391
1392    // got[0] is the address of the lazy resolver function.
1393    // got[1] may be used for a GNU extension.
1394    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1395    // FIXME: maybe this should be in a separate routine?
1396    if ((si->flags & FLAG_LINKER) == 0) {
1397        size_t g = 0;
1398        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1399        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1400            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1401        }
1402        // Relocate the local GOT entries.
1403        for (; g < local_gotno; g++) {
1404            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1405        }
1406    }
1407
1408    // Now for the global GOT entries...
1409    ElfW(Sym)* sym = symtab + gotsym;
1410    got = si->plt_got + local_gotno;
1411    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1412        // This is an undefined reference... try to locate it.
1413        const char* sym_name = si->strtab + sym->st_name;
1414        soinfo* lsi;
1415        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1416        if (s == NULL) {
1417            // We only allow an undefined symbol if this is a weak reference.
1418            s = &symtab[g];
1419            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1420                DL_ERR("cannot locate \"%s\"...", sym_name);
1421                return false;
1422            }
1423            *got = 0;
1424        } else {
1425            // FIXME: is this sufficient?
1426            // For reference see NetBSD link loader
1427            // 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
1428            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1429        }
1430    }
1431    return true;
1432}
1433#endif
1434
1435void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1436  if (functions == NULL) {
1437    return;
1438  }
1439
1440  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1441
1442  int begin = reverse ? (count - 1) : 0;
1443  int end = reverse ? -1 : count;
1444  int step = reverse ? -1 : 1;
1445
1446  for (int i = begin; i != end; i += step) {
1447    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1448    CallFunction("function", functions[i]);
1449  }
1450
1451  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1452}
1453
1454void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1455  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1456    return;
1457  }
1458
1459  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1460  function();
1461  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1462
1463  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1464  // are still writable. This happens with our debug malloc (see http://b/7941716).
1465  protect_data(PROT_READ | PROT_WRITE);
1466}
1467
1468void soinfo::CallPreInitConstructors() {
1469  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1470  // but ignored in a shared library.
1471  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1472}
1473
1474void soinfo::CallConstructors() {
1475  if (constructors_called) {
1476    return;
1477  }
1478
1479  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1480  // protect against recursive constructor calls. One simple example of constructor recursion
1481  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1482  // 1. The program depends on libc, so libc's constructor is called here.
1483  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1484  // 3. dlopen() calls the constructors on the newly created
1485  //    soinfo for libc_malloc_debug_leak.so.
1486  // 4. The debug .so depends on libc, so CallConstructors is
1487  //    called again with the libc soinfo. If it doesn't trigger the early-
1488  //    out above, the libc constructor will be called again (recursively!).
1489  constructors_called = true;
1490
1491  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1492    // The GNU dynamic linker silently ignores these, but we warn the developer.
1493    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1494          name, preinit_array_count);
1495  }
1496
1497  get_children().for_each([] (soinfo* si) {
1498    si->CallConstructors();
1499  });
1500
1501  TRACE("\"%s\": calling constructors", name);
1502
1503  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1504  CallFunction("DT_INIT", init_func);
1505  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1506}
1507
1508void soinfo::CallDestructors() {
1509  TRACE("\"%s\": calling destructors", name);
1510
1511  // DT_FINI_ARRAY must be parsed in reverse order.
1512  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1513
1514  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1515  CallFunction("DT_FINI", fini_func);
1516
1517  // This is needed on second call to dlopen
1518  // after library has been unloaded with RTLD_NODELETE
1519  constructors_called = false;
1520}
1521
1522void soinfo::add_child(soinfo* child) {
1523  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1524    return;
1525  }
1526
1527  this->children.push_front(child);
1528  child->parents.push_front(this);
1529}
1530
1531void soinfo::remove_all_links() {
1532  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1533    return;
1534  }
1535
1536  // 1. Untie connected soinfos from 'this'.
1537  children.for_each([&] (soinfo* child) {
1538    child->parents.remove_if([&] (const soinfo* parent) {
1539      return parent == this;
1540    });
1541  });
1542
1543  parents.for_each([&] (soinfo* parent) {
1544    parent->children.for_each([&] (const soinfo* child) {
1545      return child == this;
1546    });
1547  });
1548
1549  // 2. Once everything untied - clear local lists.
1550  parents.clear();
1551  children.clear();
1552}
1553
1554void soinfo::set_st_dev(dev_t dev) {
1555  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1556    return;
1557  }
1558
1559  st_dev = dev;
1560}
1561
1562void soinfo::set_st_ino(ino_t ino) {
1563  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1564    return;
1565  }
1566
1567  st_ino = ino;
1568}
1569
1570dev_t soinfo::get_st_dev() {
1571  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1572    return 0;
1573  }
1574
1575  return st_dev;
1576};
1577
1578ino_t soinfo::get_st_ino() {
1579  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1580    return 0;
1581  }
1582
1583  return st_ino;
1584}
1585
1586// This is a return on get_children() in case
1587// 'this->flags' does not have FLAG_NEW_SOINFO set.
1588static soinfo::soinfo_list_t g_empty_list;
1589
1590soinfo::soinfo_list_t& soinfo::get_children() {
1591  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1592    return g_empty_list;
1593  }
1594
1595  return this->children;
1596}
1597
1598/* Force any of the closed stdin, stdout and stderr to be associated with
1599   /dev/null. */
1600static int nullify_closed_stdio() {
1601    int dev_null, i, status;
1602    int return_value = 0;
1603
1604    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1605    if (dev_null < 0) {
1606        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1607        return -1;
1608    }
1609    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1610
1611    /* If any of the stdio file descriptors is valid and not associated
1612       with /dev/null, dup /dev/null to it.  */
1613    for (i = 0; i < 3; i++) {
1614        /* If it is /dev/null already, we are done. */
1615        if (i == dev_null) {
1616            continue;
1617        }
1618
1619        TRACE("[ Nullifying stdio file descriptor %d]", i);
1620        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1621
1622        /* If file is opened, we are good. */
1623        if (status != -1) {
1624            continue;
1625        }
1626
1627        /* The only error we allow is that the file descriptor does not
1628           exist, in which case we dup /dev/null to it. */
1629        if (errno != EBADF) {
1630            DL_ERR("fcntl failed: %s", strerror(errno));
1631            return_value = -1;
1632            continue;
1633        }
1634
1635        /* Try dupping /dev/null to this stdio file descriptor and
1636           repeat if there is a signal.  Note that any errors in closing
1637           the stdio descriptor are lost.  */
1638        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1639        if (status < 0) {
1640            DL_ERR("dup2 failed: %s", strerror(errno));
1641            return_value = -1;
1642            continue;
1643        }
1644    }
1645
1646    /* If /dev/null is not one of the stdio file descriptors, close it. */
1647    if (dev_null > 2) {
1648        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1649        status = TEMP_FAILURE_RETRY(close(dev_null));
1650        if (status == -1) {
1651            DL_ERR("close failed: %s", strerror(errno));
1652            return_value = -1;
1653        }
1654    }
1655
1656    return return_value;
1657}
1658
1659static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1660    /* "base" might wrap around UINT32_MAX. */
1661    ElfW(Addr) base = si->load_bias;
1662    const ElfW(Phdr)* phdr = si->phdr;
1663    int phnum = si->phnum;
1664    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1665
1666    /* We can't debug anything until the linker is relocated */
1667    if (!relocating_linker) {
1668        INFO("[ linking %s ]", si->name);
1669        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1670    }
1671
1672    /* Extract dynamic section */
1673    size_t dynamic_count;
1674    ElfW(Word) dynamic_flags;
1675    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1676                                   &dynamic_count, &dynamic_flags);
1677    if (si->dynamic == NULL) {
1678        if (!relocating_linker) {
1679            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1680        }
1681        return false;
1682    } else {
1683        if (!relocating_linker) {
1684            DEBUG("dynamic = %p", si->dynamic);
1685        }
1686    }
1687
1688#if defined(__arm__)
1689    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1690                                    &si->ARM_exidx, &si->ARM_exidx_count);
1691#endif
1692
1693    // Extract useful information from dynamic section.
1694    uint32_t needed_count = 0;
1695    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1696        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1697              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1698        switch (d->d_tag) {
1699        case DT_HASH:
1700            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1701            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1702            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1703            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1704            break;
1705        case DT_STRTAB:
1706            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1707            break;
1708        case DT_SYMTAB:
1709            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1710            break;
1711#if !defined(__LP64__)
1712        case DT_PLTREL:
1713            if (d->d_un.d_val != DT_REL) {
1714                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1715                return false;
1716            }
1717            break;
1718#endif
1719        case DT_JMPREL:
1720#if defined(USE_RELA)
1721            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1722#else
1723            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1724#endif
1725            break;
1726        case DT_PLTRELSZ:
1727#if defined(USE_RELA)
1728            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1729#else
1730            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1731#endif
1732            break;
1733#if defined(__mips__)
1734        case DT_PLTGOT:
1735            // Used by mips and mips64.
1736            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1737            break;
1738#endif
1739        case DT_DEBUG:
1740            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1741            // if the dynamic table is writable
1742// FIXME: not working currently for N64
1743// The flags for the LOAD and DYNAMIC program headers do not agree.
1744// The LOAD section containng the dynamic table has been mapped as
1745// read-only, but the DYNAMIC header claims it is writable.
1746#if !(defined(__mips__) && defined(__LP64__))
1747            if ((dynamic_flags & PF_W) != 0) {
1748                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1749            }
1750            break;
1751#endif
1752#if defined(USE_RELA)
1753         case DT_RELA:
1754            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1755            break;
1756         case DT_RELASZ:
1757            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1758            break;
1759        case DT_REL:
1760            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1761            return false;
1762        case DT_RELSZ:
1763            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1764            return false;
1765#else
1766        case DT_REL:
1767            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1768            break;
1769        case DT_RELSZ:
1770            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1771            break;
1772         case DT_RELA:
1773            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1774            return false;
1775#endif
1776        case DT_INIT:
1777            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1778            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1779            break;
1780        case DT_FINI:
1781            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1782            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1783            break;
1784        case DT_INIT_ARRAY:
1785            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1786            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1787            break;
1788        case DT_INIT_ARRAYSZ:
1789            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1790            break;
1791        case DT_FINI_ARRAY:
1792            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1793            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1794            break;
1795        case DT_FINI_ARRAYSZ:
1796            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1797            break;
1798        case DT_PREINIT_ARRAY:
1799            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1800            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1801            break;
1802        case DT_PREINIT_ARRAYSZ:
1803            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1804            break;
1805        case DT_TEXTREL:
1806#if defined(__LP64__)
1807            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1808            return false;
1809#else
1810            si->has_text_relocations = true;
1811            break;
1812#endif
1813        case DT_SYMBOLIC:
1814            si->has_DT_SYMBOLIC = true;
1815            break;
1816        case DT_NEEDED:
1817            ++needed_count;
1818            break;
1819        case DT_FLAGS:
1820            if (d->d_un.d_val & DF_TEXTREL) {
1821#if defined(__LP64__)
1822                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1823                return false;
1824#else
1825                si->has_text_relocations = true;
1826#endif
1827            }
1828            if (d->d_un.d_val & DF_SYMBOLIC) {
1829                si->has_DT_SYMBOLIC = true;
1830            }
1831            break;
1832#if defined(__mips__)
1833        case DT_STRSZ:
1834        case DT_SYMENT:
1835        case DT_RELENT:
1836             break;
1837        case DT_MIPS_RLD_MAP:
1838            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1839            {
1840              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1841              *dp = &_r_debug;
1842            }
1843            break;
1844        case DT_MIPS_RLD_VERSION:
1845        case DT_MIPS_FLAGS:
1846        case DT_MIPS_BASE_ADDRESS:
1847        case DT_MIPS_UNREFEXTNO:
1848            break;
1849
1850        case DT_MIPS_SYMTABNO:
1851            si->mips_symtabno = d->d_un.d_val;
1852            break;
1853
1854        case DT_MIPS_LOCAL_GOTNO:
1855            si->mips_local_gotno = d->d_un.d_val;
1856            break;
1857
1858        case DT_MIPS_GOTSYM:
1859            si->mips_gotsym = d->d_un.d_val;
1860            break;
1861#endif
1862
1863        default:
1864            DEBUG("Unused DT entry: type %p arg %p",
1865                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1866            break;
1867        }
1868    }
1869
1870    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1871          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1872
1873    // Sanity checks.
1874    if (relocating_linker && needed_count != 0) {
1875        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1876        return false;
1877    }
1878    if (si->nbucket == 0) {
1879        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1880        return false;
1881    }
1882    if (si->strtab == 0) {
1883        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1884        return false;
1885    }
1886    if (si->symtab == 0) {
1887        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1888        return false;
1889    }
1890
1891    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1892    if (si->flags & FLAG_EXE) {
1893        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1894        size_t preload_count = 0;
1895        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1896            soinfo* lsi = find_library(g_ld_preload_names[i], 0, NULL);
1897            if (lsi != NULL) {
1898                g_ld_preloads[preload_count++] = lsi;
1899            } else {
1900                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1901                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1902                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1903            }
1904        }
1905    }
1906
1907    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1908    soinfo** pneeded = needed;
1909
1910    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1911        if (d->d_tag == DT_NEEDED) {
1912            const char* library_name = si->strtab + d->d_un.d_val;
1913            DEBUG("%s needs %s", si->name, library_name);
1914            soinfo* lsi = find_library(library_name, 0, NULL);
1915            if (lsi == NULL) {
1916                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1917                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1918                       library_name, si->name, tmp_err_buf);
1919                return false;
1920            }
1921
1922            si->add_child(lsi);
1923            *pneeded++ = lsi;
1924        }
1925    }
1926    *pneeded = NULL;
1927
1928#if !defined(__LP64__)
1929    if (si->has_text_relocations) {
1930        // Make segments writable to allow text relocations to work properly. We will later call
1931        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1932        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1933                "security hardening. Please fix.", si->name);
1934        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1935            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1936                   si->name, strerror(errno));
1937            return false;
1938        }
1939    }
1940#endif
1941
1942#if defined(USE_RELA)
1943    if (si->plt_rela != NULL) {
1944        DEBUG("[ relocating %s plt ]\n", si->name);
1945        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1946            return false;
1947        }
1948    }
1949    if (si->rela != NULL) {
1950        DEBUG("[ relocating %s ]\n", si->name);
1951        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1952            return false;
1953        }
1954    }
1955#else
1956    if (si->plt_rel != NULL) {
1957        DEBUG("[ relocating %s plt ]", si->name);
1958        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1959            return false;
1960        }
1961    }
1962    if (si->rel != NULL) {
1963        DEBUG("[ relocating %s ]", si->name);
1964        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1965            return false;
1966        }
1967    }
1968#endif
1969
1970#if defined(__mips__)
1971    if (!mips_relocate_got(si, needed)) {
1972        return false;
1973    }
1974#endif
1975
1976    si->flags |= FLAG_LINKED;
1977    DEBUG("[ finished linking %s ]", si->name);
1978
1979#if !defined(__LP64__)
1980    if (si->has_text_relocations) {
1981        // All relocations are done, we can protect our segments back to read-only.
1982        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1983            DL_ERR("can't protect segments for \"%s\": %s",
1984                   si->name, strerror(errno));
1985            return false;
1986        }
1987    }
1988#endif
1989
1990    /* We can also turn on GNU RELRO protection */
1991    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1992        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1993               si->name, strerror(errno));
1994        return false;
1995    }
1996
1997    /* Handle serializing/sharing the RELRO segment */
1998    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1999      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
2000                                         extinfo->relro_fd) < 0) {
2001        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
2002               si->name, strerror(errno));
2003        return false;
2004      }
2005    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
2006      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
2007                                   extinfo->relro_fd) < 0) {
2008        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
2009               si->name, strerror(errno));
2010        return false;
2011      }
2012    }
2013
2014    notify_gdb_of_load(si);
2015    return true;
2016}
2017
2018/*
2019 * This function add vdso to internal dso list.
2020 * It helps to stack unwinding through signal handlers.
2021 * Also, it makes bionic more like glibc.
2022 */
2023static void add_vdso(KernelArgumentBlock& args __unused) {
2024#if defined(AT_SYSINFO_EHDR)
2025  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
2026  if (ehdr_vdso == NULL) {
2027    return;
2028  }
2029
2030  soinfo* si = soinfo_alloc("[vdso]", NULL);
2031
2032  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
2033  si->phnum = ehdr_vdso->e_phnum;
2034  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
2035  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2036  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
2037
2038  soinfo_link_image(si, NULL);
2039#endif
2040}
2041
2042/*
2043 * This is linker soinfo for GDB. See details below.
2044 */
2045static soinfo linker_soinfo_for_gdb;
2046
2047/* gdb expects the linker to be in the debug shared object list.
2048 * Without this, gdb has trouble locating the linker's ".text"
2049 * and ".plt" sections. Gdb could also potentially use this to
2050 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2051 * Don't use soinfo_alloc(), because the linker shouldn't
2052 * be on the soinfo list.
2053 */
2054static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
2055#if defined(__LP64__)
2056  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
2057#else
2058  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
2059#endif
2060  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
2061  linker_soinfo_for_gdb.base = linker_base;
2062
2063  /*
2064   * Set the dynamic field in the link map otherwise gdb will complain with
2065   * the following:
2066   *   warning: .dynamic section for "/system/bin/linker" is not at the
2067   *   expected address (wrong library or version mismatch?)
2068   */
2069  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2070  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2071  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2072                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
2073  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2074}
2075
2076/*
2077 * This code is called after the linker has linked itself and
2078 * fixed it's own GOT. It is safe to make references to externs
2079 * and other non-local data at this point.
2080 */
2081static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2082    /* NOTE: we store the args pointer on a special location
2083     *       of the temporary TLS area in order to pass it to
2084     *       the C Library's runtime initializer.
2085     *
2086     *       The initializer must clear the slot and reset the TLS
2087     *       to point to a different location to ensure that no other
2088     *       shared library constructor can access it.
2089     */
2090  __libc_init_tls(args);
2091
2092#if TIMING
2093    struct timeval t0, t1;
2094    gettimeofday(&t0, 0);
2095#endif
2096
2097    // Initialize environment functions, and get to the ELF aux vectors table.
2098    linker_env_init(args);
2099
2100    // If this is a setuid/setgid program, close the security hole described in
2101    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2102    if (get_AT_SECURE()) {
2103        nullify_closed_stdio();
2104    }
2105
2106    debuggerd_init();
2107
2108    // Get a few environment variables.
2109    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2110    if (LD_DEBUG != NULL) {
2111      g_ld_debug_verbosity = atoi(LD_DEBUG);
2112    }
2113
2114    // Normally, these are cleaned by linker_env_init, but the test
2115    // doesn't cost us anything.
2116    const char* ldpath_env = NULL;
2117    const char* ldpreload_env = NULL;
2118    if (!get_AT_SECURE()) {
2119      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2120      ldpreload_env = linker_env_get("LD_PRELOAD");
2121    }
2122
2123    // Linker does not call constructors for its own
2124    // global variables so we need to initialize
2125    // the allocators explicitly.
2126    g_soinfo_allocator.init();
2127    g_soinfo_links_allocator.init();
2128
2129    INFO("[ android linker & debugger ]");
2130
2131    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2132    if (si == NULL) {
2133        exit(EXIT_FAILURE);
2134    }
2135
2136    /* bootstrap the link map, the main exe always needs to be first */
2137    si->flags |= FLAG_EXE;
2138    link_map* map = &(si->link_map_head);
2139
2140    map->l_addr = 0;
2141    map->l_name = args.argv[0];
2142    map->l_prev = NULL;
2143    map->l_next = NULL;
2144
2145    _r_debug.r_map = map;
2146    r_debug_tail = map;
2147
2148    init_linker_info_for_gdb(linker_base);
2149
2150    // Extract information passed from the kernel.
2151    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2152    si->phnum = args.getauxval(AT_PHNUM);
2153    si->entry = args.getauxval(AT_ENTRY);
2154
2155    /* Compute the value of si->base. We can't rely on the fact that
2156     * the first entry is the PHDR because this will not be true
2157     * for certain executables (e.g. some in the NDK unit test suite)
2158     */
2159    si->base = 0;
2160    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2161    si->load_bias = 0;
2162    for (size_t i = 0; i < si->phnum; ++i) {
2163      if (si->phdr[i].p_type == PT_PHDR) {
2164        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2165        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2166        break;
2167      }
2168    }
2169    si->dynamic = NULL;
2170    si->ref_count = 1;
2171
2172#if defined(__LP64__)
2173    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2174    if (elf_hdr->e_type != ET_DYN) {
2175        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2176        exit(EXIT_FAILURE);
2177    }
2178#endif
2179
2180    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2181    parse_LD_LIBRARY_PATH(ldpath_env);
2182    parse_LD_PRELOAD(ldpreload_env);
2183
2184    somain = si;
2185
2186    if (!soinfo_link_image(si, NULL)) {
2187        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2188        exit(EXIT_FAILURE);
2189    }
2190
2191    add_vdso(args);
2192
2193    si->CallPreInitConstructors();
2194
2195    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2196        g_ld_preloads[i]->CallConstructors();
2197    }
2198
2199    /* After the link_image, the si->load_bias is initialized.
2200     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2201     * We need to update this value for so exe here. So Unwind_Backtrace
2202     * for some arch like x86 could work correctly within so exe.
2203     */
2204    map->l_addr = si->load_bias;
2205    si->CallConstructors();
2206
2207#if TIMING
2208    gettimeofday(&t1, NULL);
2209    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2210               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2211               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2212#endif
2213#if STATS
2214    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2215           linker_stats.count[kRelocAbsolute],
2216           linker_stats.count[kRelocRelative],
2217           linker_stats.count[kRelocCopy],
2218           linker_stats.count[kRelocSymbol]);
2219#endif
2220#if COUNT_PAGES
2221    {
2222        unsigned n;
2223        unsigned i;
2224        unsigned count = 0;
2225        for (n = 0; n < 4096; n++) {
2226            if (bitmask[n]) {
2227                unsigned x = bitmask[n];
2228#if defined(__LP64__)
2229                for (i = 0; i < 32; i++) {
2230#else
2231                for (i = 0; i < 8; i++) {
2232#endif
2233                    if (x & 1) {
2234                        count++;
2235                    }
2236                    x >>= 1;
2237                }
2238            }
2239        }
2240        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2241    }
2242#endif
2243
2244#if TIMING || STATS || COUNT_PAGES
2245    fflush(stdout);
2246#endif
2247
2248    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2249    return si->entry;
2250}
2251
2252/* Compute the load-bias of an existing executable. This shall only
2253 * be used to compute the load bias of an executable or shared library
2254 * that was loaded by the kernel itself.
2255 *
2256 * Input:
2257 *    elf    -> address of ELF header, assumed to be at the start of the file.
2258 * Return:
2259 *    load bias, i.e. add the value of any p_vaddr in the file to get
2260 *    the corresponding address in memory.
2261 */
2262static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2263  ElfW(Addr) offset = elf->e_phoff;
2264  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2265  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2266
2267  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2268    if (phdr->p_type == PT_LOAD) {
2269      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2270    }
2271  }
2272  return 0;
2273}
2274
2275/*
2276 * This is the entry point for the linker, called from begin.S. This
2277 * method is responsible for fixing the linker's own relocations, and
2278 * then calling __linker_init_post_relocation().
2279 *
2280 * Because this method is called before the linker has fixed it's own
2281 * relocations, any attempt to reference an extern variable, extern
2282 * function, or other GOT reference will generate a segfault.
2283 */
2284extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2285  // Initialize static variables.
2286  solist = get_libdl_info();
2287  sonext = get_libdl_info();
2288
2289  KernelArgumentBlock args(raw_args);
2290
2291  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2292  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2293  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2294
2295  soinfo linker_so;
2296  memset(&linker_so, 0, sizeof(soinfo));
2297
2298  strcpy(linker_so.name, "[dynamic linker]");
2299  linker_so.base = linker_addr;
2300  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2301  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2302  linker_so.dynamic = NULL;
2303  linker_so.phdr = phdr;
2304  linker_so.phnum = elf_hdr->e_phnum;
2305  linker_so.flags |= FLAG_LINKER;
2306
2307  if (!soinfo_link_image(&linker_so, NULL)) {
2308    // It would be nice to print an error message, but if the linker
2309    // can't link itself, there's no guarantee that we'll be able to
2310    // call write() (because it involves a GOT reference). We may as
2311    // well try though...
2312    const char* msg = "CANNOT LINK EXECUTABLE: ";
2313    write(2, msg, strlen(msg));
2314    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2315    write(2, "\n", 1);
2316    _exit(EXIT_FAILURE);
2317  }
2318
2319  // We have successfully fixed our own relocations. It's safe to run
2320  // the main part of the linker now.
2321  args.abort_message_ptr = &g_abort_message;
2322  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2323
2324  protect_data(PROT_READ);
2325
2326  // Return the address that the calling assembly stub should jump to.
2327  return start_address;
2328}
2329