linker.cpp revision 65549969488bbce7f5d1b57714ba32c466943470
1/*
2 * Copyright (C) 2008, 2009 The Android Open Source Project
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *  * Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 *  * Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in
12 *    the documentation and/or other materials provided with the
13 *    distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <dlfcn.h>
30#include <errno.h>
31#include <fcntl.h>
32#include <inttypes.h>
33#include <pthread.h>
34#include <stdio.h>
35#include <stdlib.h>
36#include <string.h>
37#include <sys/mman.h>
38#include <sys/stat.h>
39#include <unistd.h>
40
41// Private C library headers.
42#include "private/bionic_tls.h"
43#include "private/KernelArgumentBlock.h"
44#include "private/ScopedPthreadMutexLocker.h"
45#include "private/ScopedFd.h"
46
47#include "linker.h"
48#include "linker_debug.h"
49#include "linker_environ.h"
50#include "linker_phdr.h"
51#include "linker_allocator.h"
52
53/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
54 *
55 * Do NOT use malloc() and friends or pthread_*() code here.
56 * Don't use printf() either; it's caused mysterious memory
57 * corruption in the past.
58 * The linker runs before we bring up libc and it's easiest
59 * to make sure it does not depend on any complex libc features
60 *
61 * open issues / todo:
62 *
63 * - cleaner error reporting
64 * - after linking, set as much stuff as possible to READONLY
65 *   and NOEXEC
66 */
67
68#if defined(__LP64__)
69#define SEARCH_NAME(x) x
70#else
71// Nvidia drivers are relying on the bug:
72// http://code.google.com/p/android/issues/detail?id=6670
73// so we continue to use base-name lookup for lp32
74static const char* get_base_name(const char* name) {
75  const char* bname = strrchr(name, '/');
76  return bname ? bname + 1 : name;
77}
78#define SEARCH_NAME(x) get_base_name(x)
79#endif
80
81static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
82static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
83
84static LinkerAllocator<soinfo> g_soinfo_allocator;
85static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
86
87static soinfo* solist;
88static soinfo* sonext;
89static soinfo* somain; /* main process, always the one after libdl_info */
90
91static const char* const kDefaultLdPaths[] = {
92#if defined(__LP64__)
93  "/vendor/lib64",
94  "/system/lib64",
95#else
96  "/vendor/lib",
97  "/system/lib",
98#endif
99  NULL
100};
101
102#define LDPATH_BUFSIZE (LDPATH_MAX*64)
103#define LDPATH_MAX 8
104
105#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
106#define LDPRELOAD_MAX 8
107
108static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
109static const char* g_ld_library_paths[LDPATH_MAX + 1];
110
111static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
112static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
113
114static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
115
116__LIBC_HIDDEN__ int g_ld_debug_verbosity;
117
118__LIBC_HIDDEN__ abort_msg_t* g_abort_message = NULL; // For debuggerd.
119
120enum RelocationKind {
121    kRelocAbsolute = 0,
122    kRelocRelative,
123    kRelocCopy,
124    kRelocSymbol,
125    kRelocMax
126};
127
128enum class SymbolLookupScope {
129  kAllowLocal,
130  kExcludeLocal,
131};
132
133#if STATS
134struct linker_stats_t {
135    int count[kRelocMax];
136};
137
138static linker_stats_t linker_stats;
139
140static void count_relocation(RelocationKind kind) {
141    ++linker_stats.count[kind];
142}
143#else
144static void count_relocation(RelocationKind) {
145}
146#endif
147
148#if COUNT_PAGES
149static unsigned bitmask[4096];
150#if defined(__LP64__)
151#define MARK(offset) \
152    do { \
153        if ((((offset) >> 12) >> 5) < 4096) \
154            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
155    } while (0)
156#else
157#define MARK(offset) \
158    do { \
159        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
160    } while (0)
161#endif
162#else
163#define MARK(x) do {} while (0)
164#endif
165
166// You shouldn't try to call memory-allocating functions in the dynamic linker.
167// Guard against the most obvious ones.
168#define DISALLOW_ALLOCATION(return_type, name, ...) \
169    return_type name __VA_ARGS__ \
170    { \
171        __libc_fatal("ERROR: " #name " called from the dynamic linker!\n"); \
172    }
173DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
174DISALLOW_ALLOCATION(void, free, (void* u __unused));
175DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
176DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
177
178static char tmp_err_buf[768];
179static char __linker_dl_err_buf[768];
180
181char* linker_get_error_buffer() {
182  return &__linker_dl_err_buf[0];
183}
184
185size_t linker_get_error_buffer_size() {
186  return sizeof(__linker_dl_err_buf);
187}
188
189/*
190 * This function is an empty stub where GDB locates a breakpoint to get notified
191 * about linker activity.
192 */
193extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
194
195static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
196static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
197static link_map* r_debug_tail = 0;
198
199static void insert_soinfo_into_debug_map(soinfo* info) {
200    // Copy the necessary fields into the debug structure.
201    link_map* map = &(info->link_map_head);
202    map->l_addr = info->load_bias;
203    map->l_name = reinterpret_cast<char*>(info->name);
204    map->l_ld = info->dynamic;
205
206    /* Stick the new library at the end of the list.
207     * gdb tends to care more about libc than it does
208     * about leaf libraries, and ordering it this way
209     * reduces the back-and-forth over the wire.
210     */
211    if (r_debug_tail) {
212        r_debug_tail->l_next = map;
213        map->l_prev = r_debug_tail;
214        map->l_next = 0;
215    } else {
216        _r_debug.r_map = map;
217        map->l_prev = 0;
218        map->l_next = 0;
219    }
220    r_debug_tail = map;
221}
222
223static void remove_soinfo_from_debug_map(soinfo* info) {
224    link_map* map = &(info->link_map_head);
225
226    if (r_debug_tail == map) {
227        r_debug_tail = map->l_prev;
228    }
229
230    if (map->l_prev) {
231        map->l_prev->l_next = map->l_next;
232    }
233    if (map->l_next) {
234        map->l_next->l_prev = map->l_prev;
235    }
236}
237
238static void notify_gdb_of_load(soinfo* info) {
239    if (info->flags & FLAG_EXE) {
240        // GDB already knows about the main executable
241        return;
242    }
243
244    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
245
246    _r_debug.r_state = r_debug::RT_ADD;
247    rtld_db_dlactivity();
248
249    insert_soinfo_into_debug_map(info);
250
251    _r_debug.r_state = r_debug::RT_CONSISTENT;
252    rtld_db_dlactivity();
253}
254
255static void notify_gdb_of_unload(soinfo* info) {
256    if (info->flags & FLAG_EXE) {
257        // GDB already knows about the main executable
258        return;
259    }
260
261    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
262
263    _r_debug.r_state = r_debug::RT_DELETE;
264    rtld_db_dlactivity();
265
266    remove_soinfo_from_debug_map(info);
267
268    _r_debug.r_state = r_debug::RT_CONSISTENT;
269    rtld_db_dlactivity();
270}
271
272void notify_gdb_of_libraries() {
273  _r_debug.r_state = r_debug::RT_ADD;
274  rtld_db_dlactivity();
275  _r_debug.r_state = r_debug::RT_CONSISTENT;
276  rtld_db_dlactivity();
277}
278
279LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
280  return g_soinfo_links_allocator.alloc();
281}
282
283void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
284  g_soinfo_links_allocator.free(entry);
285}
286
287static void protect_data(int protection) {
288  g_soinfo_allocator.protect_all(protection);
289  g_soinfo_links_allocator.protect_all(protection);
290}
291
292static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
293  if (strlen(name) >= SOINFO_NAME_LEN) {
294    DL_ERR("library name \"%s\" too long", name);
295    return NULL;
296  }
297
298  soinfo* si = g_soinfo_allocator.alloc();
299
300  // Initialize the new element.
301  memset(si, 0, sizeof(soinfo));
302  strlcpy(si->name, name, sizeof(si->name));
303  si->flags = FLAG_NEW_SOINFO;
304
305  if (file_stat != NULL) {
306    si->set_st_dev(file_stat->st_dev);
307    si->set_st_ino(file_stat->st_ino);
308  }
309
310  sonext->next = si;
311  sonext = si;
312
313  TRACE("name %s: allocated soinfo @ %p", name, si);
314  return si;
315}
316
317static void soinfo_free(soinfo* si) {
318    if (si == NULL) {
319        return;
320    }
321
322    if (si->base != 0 && si->size != 0) {
323      munmap(reinterpret_cast<void*>(si->base), si->size);
324    }
325
326    soinfo *prev = NULL, *trav;
327
328    TRACE("name %s: freeing soinfo @ %p", si->name, si);
329
330    for (trav = solist; trav != NULL; trav = trav->next) {
331        if (trav == si)
332            break;
333        prev = trav;
334    }
335    if (trav == NULL) {
336        /* si was not in solist */
337        DL_ERR("name \"%s\" is not in solist!", si->name);
338        return;
339    }
340
341    // clear links to/from si
342    si->remove_all_links();
343
344    /* prev will never be NULL, because the first entry in solist is
345       always the static libdl_info.
346    */
347    prev->next = si->next;
348    if (si == sonext) {
349        sonext = prev;
350    }
351
352    g_soinfo_allocator.free(si);
353}
354
355
356static void parse_path(const char* path, const char* delimiters,
357                       const char** array, char* buf, size_t buf_size, size_t max_count) {
358  if (path == NULL) {
359    return;
360  }
361
362  size_t len = strlcpy(buf, path, buf_size);
363
364  size_t i = 0;
365  char* buf_p = buf;
366  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
367    if (*array[i] != '\0') {
368      ++i;
369    }
370  }
371
372  // Forget the last path if we had to truncate; this occurs if the 2nd to
373  // last char isn't '\0' (i.e. wasn't originally a delimiter).
374  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
375    array[i - 1] = NULL;
376  } else {
377    array[i] = NULL;
378  }
379}
380
381static void parse_LD_LIBRARY_PATH(const char* path) {
382  parse_path(path, ":", g_ld_library_paths,
383             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
384}
385
386static void parse_LD_PRELOAD(const char* path) {
387  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
388  parse_path(path, " :", g_ld_preload_names,
389             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
390}
391
392#if defined(__arm__)
393
394/* For a given PC, find the .so that it belongs to.
395 * Returns the base address of the .ARM.exidx section
396 * for that .so, and the number of 8-byte entries
397 * in that section (via *pcount).
398 *
399 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
400 *
401 * This function is exposed via dlfcn.cpp and libdl.so.
402 */
403_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
404    unsigned addr = (unsigned)pc;
405
406    for (soinfo* si = solist; si != 0; si = si->next) {
407        if ((addr >= si->base) && (addr < (si->base + si->size))) {
408            *pcount = si->ARM_exidx_count;
409            return (_Unwind_Ptr)si->ARM_exidx;
410        }
411    }
412    *pcount = 0;
413    return NULL;
414}
415
416#endif
417
418/* Here, we only have to provide a callback to iterate across all the
419 * loaded libraries. gcc_eh does the rest. */
420int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
421    int rv = 0;
422    for (soinfo* si = solist; si != NULL; si = si->next) {
423        dl_phdr_info dl_info;
424        dl_info.dlpi_addr = si->link_map_head.l_addr;
425        dl_info.dlpi_name = si->link_map_head.l_name;
426        dl_info.dlpi_phdr = si->phdr;
427        dl_info.dlpi_phnum = si->phnum;
428        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
429        if (rv != 0) {
430            break;
431        }
432    }
433    return rv;
434}
435
436static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name, const SymbolLookupScope& lookup_scope) {
437  ElfW(Sym)* symtab = si->symtab;
438  const char* strtab = si->strtab;
439
440  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
441             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
442
443  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
444    ElfW(Sym)* s = symtab + n;
445    if (strcmp(strtab + s->st_name, name)) continue;
446
447    switch (ELF_ST_BIND(s->st_info)) {
448      case STB_GLOBAL:
449      case STB_WEAK:
450        if (s->st_shndx == SHN_UNDEF) {
451          continue;
452        }
453
454        TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
455                 name, si->name, reinterpret_cast<void*>(s->st_value),
456                 static_cast<size_t>(s->st_size));
457        return s;
458      case STB_LOCAL:
459        if (lookup_scope != SymbolLookupScope::kAllowLocal) {
460          continue;
461        }
462        TRACE_TYPE(LOOKUP, "FOUND LOCAL %s in %s (%p) %zd",
463                name, si->name, reinterpret_cast<void*>(s->st_value),
464                static_cast<size_t>(s->st_size));
465        return s;
466      default:
467        __libc_fatal("ERROR: Unexpected ST_BIND value: %d for '%s' in '%s'",
468            ELF_ST_BIND(s->st_info), name, si->name);
469    }
470  }
471
472  return NULL;
473}
474
475static unsigned elfhash(const char* _name) {
476    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
477    unsigned h = 0, g;
478
479    while (*name) {
480        h = (h << 4) + *name++;
481        g = h & 0xf0000000;
482        h ^= g;
483        h ^= g >> 24;
484    }
485    return h;
486}
487
488static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
489    unsigned elf_hash = elfhash(name);
490    ElfW(Sym)* s = NULL;
491
492    if (si != NULL && somain != NULL) {
493        /*
494         * Local scope is executable scope. Just start looking into it right away
495         * for the shortcut.
496         */
497
498        if (si == somain) {
499            s = soinfo_elf_lookup(si, elf_hash, name, SymbolLookupScope::kAllowLocal);
500            if (s != NULL) {
501                *lsi = si;
502                goto done;
503            }
504        } else {
505            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
506
507            /*
508             * If this object was built with symbolic relocations disabled, the
509             * first place to look to resolve external references is the main
510             * executable.
511             */
512
513            if (!si->has_DT_SYMBOLIC) {
514                DEBUG("%s: looking up %s in executable %s",
515                      si->name, name, somain->name);
516                s = soinfo_elf_lookup(somain, elf_hash, name, SymbolLookupScope::kExcludeLocal);
517                if (s != NULL) {
518                    *lsi = somain;
519                    goto done;
520                }
521            }
522
523            /* Look for symbols in the local scope (the object who is
524             * searching). This happens with C++ templates on x86 for some
525             * reason.
526             *
527             * Notes on weak symbols:
528             * The ELF specs are ambiguous about treatment of weak definitions in
529             * dynamic linking.  Some systems return the first definition found
530             * and some the first non-weak definition.   This is system dependent.
531             * Here we return the first definition found for simplicity.  */
532
533            s = soinfo_elf_lookup(si, elf_hash, name, SymbolLookupScope::kAllowLocal);
534            if (s != NULL) {
535                *lsi = si;
536                goto done;
537            }
538
539            /*
540             * If this object was built with -Bsymbolic and symbol is not found
541             * in the local scope, try to find the symbol in the main executable.
542             */
543
544            if (si->has_DT_SYMBOLIC) {
545                DEBUG("%s: looking up %s in executable %s after local scope",
546                      si->name, name, somain->name);
547                s = soinfo_elf_lookup(somain, elf_hash, name, SymbolLookupScope::kExcludeLocal);
548                if (s != NULL) {
549                    *lsi = somain;
550                    goto done;
551                }
552            }
553        }
554    }
555
556    /* Next, look for it in the preloads list */
557    for (int i = 0; g_ld_preloads[i] != NULL; i++) {
558        s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name, SymbolLookupScope::kExcludeLocal);
559        if (s != NULL) {
560            *lsi = g_ld_preloads[i];
561            goto done;
562        }
563    }
564
565    for (int i = 0; needed[i] != NULL; i++) {
566        DEBUG("%s: looking up %s in %s",
567              si->name, name, needed[i]->name);
568        s = soinfo_elf_lookup(needed[i], elf_hash, name, SymbolLookupScope::kExcludeLocal);
569        if (s != NULL) {
570            *lsi = needed[i];
571            goto done;
572        }
573    }
574
575done:
576    if (s != NULL) {
577        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
578                   "found in %s, base = %p, load bias = %p",
579                   si->name, name, reinterpret_cast<void*>(s->st_value),
580                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
581                   reinterpret_cast<void*>((*lsi)->load_bias));
582        return s;
583    }
584
585    return NULL;
586}
587
588/* This is used by dlsym(3).  It performs symbol lookup only within the
589   specified soinfo object and not in any of its dependencies.
590
591   TODO: Only looking in the specified soinfo seems wrong. dlsym(3) says
592   that it should do a breadth first search through the dependency
593   tree. This agrees with the ELF spec (aka System V Application
594   Binary Interface) where in Chapter 5 it discuss resolving "Shared
595   Object Dependencies" in breadth first search order.
596 */
597ElfW(Sym)* dlsym_handle_lookup(soinfo* si, const char* name, soinfo* caller) {
598    return soinfo_elf_lookup(si, elfhash(name), name,
599        caller == si ? SymbolLookupScope::kAllowLocal : SymbolLookupScope::kExcludeLocal);
600}
601
602/* This is used by dlsym(3) to performs a global symbol lookup. If the
603   start value is null (for RTLD_DEFAULT), the search starts at the
604   beginning of the global solist. Otherwise the search starts at the
605   specified soinfo (for RTLD_NEXT).
606 */
607ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start, soinfo* caller) {
608  unsigned elf_hash = elfhash(name);
609
610  if (start == NULL) {
611    start = solist;
612  }
613
614  ElfW(Sym)* s = NULL;
615  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
616    s = soinfo_elf_lookup(si, elf_hash, name,
617        caller == si ? SymbolLookupScope::kAllowLocal : SymbolLookupScope::kExcludeLocal);
618    if (s != NULL) {
619      *found = si;
620      break;
621    }
622  }
623
624  if (s != NULL) {
625    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
626               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
627  }
628
629  return s;
630}
631
632soinfo* find_containing_library(const void* p) {
633  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
634  for (soinfo* si = solist; si != NULL; si = si->next) {
635    if (address >= si->base && address - si->base < si->size) {
636      return si;
637    }
638  }
639  return NULL;
640}
641
642ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
643  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
644
645  // Search the library's symbol table for any defined symbol which
646  // contains this address.
647  for (size_t i = 0; i < si->nchain; ++i) {
648    ElfW(Sym)* sym = &si->symtab[i];
649    if (sym->st_shndx != SHN_UNDEF &&
650        soaddr >= sym->st_value &&
651        soaddr < sym->st_value + sym->st_size) {
652      return sym;
653    }
654  }
655
656  return NULL;
657}
658
659static int open_library_on_path(const char* name, const char* const paths[]) {
660  char buf[512];
661  for (size_t i = 0; paths[i] != NULL; ++i) {
662    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
663    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
664      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
665      continue;
666    }
667    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
668    if (fd != -1) {
669      return fd;
670    }
671  }
672  return -1;
673}
674
675static int open_library(const char* name) {
676  TRACE("[ opening %s ]", name);
677
678  // If the name contains a slash, we should attempt to open it directly and not search the paths.
679  if (strchr(name, '/') != NULL) {
680    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
681    if (fd != -1) {
682      return fd;
683    }
684    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
685#if defined(__LP64__)
686    return -1;
687#endif
688  }
689
690  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
691  int fd = open_library_on_path(name, g_ld_library_paths);
692  if (fd == -1) {
693    fd = open_library_on_path(name, kDefaultLdPaths);
694  }
695  return fd;
696}
697
698static soinfo* load_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
699    int fd = -1;
700    ScopedFd file_guard(-1);
701
702    if (extinfo != NULL && (extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD) != 0) {
703      fd = extinfo->library_fd;
704    } else {
705      // Open the file.
706      fd = open_library(name);
707      if (fd == -1) {
708        DL_ERR("library \"%s\" not found", name);
709        return NULL;
710      }
711
712      file_guard.reset(fd);
713    }
714
715    ElfReader elf_reader(name, fd);
716
717    struct stat file_stat;
718    if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
719      DL_ERR("unable to stat file for the library %s: %s", name, strerror(errno));
720      return NULL;
721    }
722
723    // Check for symlink and other situations where
724    // file can have different names.
725    for (soinfo* si = solist; si != NULL; si = si->next) {
726      if (si->get_st_dev() != 0 &&
727          si->get_st_ino() != 0 &&
728          si->get_st_dev() == file_stat.st_dev &&
729          si->get_st_ino() == file_stat.st_ino) {
730        TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
731        return si;
732      }
733    }
734
735    if ((dlflags & RTLD_NOLOAD) != 0) {
736      return NULL;
737    }
738
739    // Read the ELF header and load the segments.
740    if (!elf_reader.Load(extinfo)) {
741        return NULL;
742    }
743
744    soinfo* si = soinfo_alloc(SEARCH_NAME(name), &file_stat);
745    if (si == NULL) {
746        return NULL;
747    }
748    si->base = elf_reader.load_start();
749    si->size = elf_reader.load_size();
750    si->load_bias = elf_reader.load_bias();
751    si->phnum = elf_reader.phdr_count();
752    si->phdr = elf_reader.loaded_phdr();
753
754    // At this point we know that whatever is loaded @ base is a valid ELF
755    // shared library whose segments are properly mapped in.
756    TRACE("[ load_library base=%p size=%zu name='%s' ]",
757          reinterpret_cast<void*>(si->base), si->size, si->name);
758
759    if (!soinfo_link_image(si, extinfo)) {
760      soinfo_free(si);
761      return NULL;
762    }
763
764    return si;
765}
766
767static soinfo *find_loaded_library_by_name(const char* name) {
768  const char* search_name = SEARCH_NAME(name);
769  for (soinfo* si = solist; si != NULL; si = si->next) {
770    if (!strcmp(search_name, si->name)) {
771      return si;
772    }
773  }
774  return NULL;
775}
776
777static soinfo* find_library_internal(const char* name, int dlflags, const android_dlextinfo* extinfo) {
778  if (name == NULL) {
779    return somain;
780  }
781
782  soinfo* si = find_loaded_library_by_name(name);
783
784  // Library might still be loaded, the accurate detection
785  // of this fact is done by load_library
786  if (si == NULL) {
787    TRACE("[ '%s' has not been found by name.  Trying harder...]", name);
788    si = load_library(name, dlflags, extinfo);
789  }
790
791  if (si != NULL && (si->flags & FLAG_LINKED) == 0) {
792    DL_ERR("recursive link to \"%s\"", si->name);
793    return NULL;
794  }
795
796  return si;
797}
798
799static soinfo* find_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
800  soinfo* si = find_library_internal(name, dlflags, extinfo);
801  if (si != NULL) {
802    si->ref_count++;
803  }
804  return si;
805}
806
807static void soinfo_unload(soinfo* si) {
808  if (si->ref_count == 1) {
809    TRACE("unloading '%s'", si->name);
810    si->CallDestructors();
811
812    if ((si->flags | FLAG_NEW_SOINFO) != 0) {
813      si->get_children().for_each([&] (soinfo* child) {
814        TRACE("%s needs to unload %s", si->name, child->name);
815        soinfo_unload(child);
816      });
817    } else {
818      for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
819        if (d->d_tag == DT_NEEDED) {
820          const char* library_name = si->strtab + d->d_un.d_val;
821          TRACE("%s needs to unload %s", si->name, library_name);
822          soinfo* needed = find_library(library_name, RTLD_NOLOAD, NULL);
823          if (needed != NULL) {
824            soinfo_unload(needed);
825          } else {
826            // Not found: for example if symlink was deleted between dlopen and dlclose
827            // Since we cannot really handle errors at this point - print and continue.
828            PRINT("warning: couldn't find %s needed by %s on unload.", library_name, si->name);
829          }
830        }
831      }
832    }
833
834    notify_gdb_of_unload(si);
835    si->ref_count = 0;
836    soinfo_free(si);
837  } else {
838    si->ref_count--;
839    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
840  }
841}
842
843void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
844  snprintf(buffer, buffer_size, "%s:%s", kDefaultLdPaths[0], kDefaultLdPaths[1]);
845}
846
847void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
848  if (!get_AT_SECURE()) {
849    parse_LD_LIBRARY_PATH(ld_library_path);
850  }
851}
852
853soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
854  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL|RTLD_NOLOAD)) != 0) {
855    DL_ERR("invalid flags to dlopen: %x", flags);
856    return NULL;
857  }
858  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
859    DL_ERR("invalid extended flags to android_dlopen_ext: %" PRIx64, extinfo->flags);
860    return NULL;
861  }
862  protect_data(PROT_READ | PROT_WRITE);
863  soinfo* si = find_library(name, flags, extinfo);
864  if (si != NULL) {
865    si->CallConstructors();
866  }
867  protect_data(PROT_READ);
868  return si;
869}
870
871void do_dlclose(soinfo* si) {
872  protect_data(PROT_READ | PROT_WRITE);
873  soinfo_unload(si);
874  protect_data(PROT_READ);
875}
876
877#if defined(USE_RELA)
878static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
879  ElfW(Sym)* s;
880  soinfo* lsi;
881
882  for (size_t idx = 0; idx < count; ++idx, ++rela) {
883    unsigned type = ELFW(R_TYPE)(rela->r_info);
884    unsigned sym = ELFW(R_SYM)(rela->r_info);
885    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
886    ElfW(Addr) sym_addr = 0;
887    const char* sym_name = NULL;
888
889    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
890    if (type == 0) { // R_*_NONE
891      continue;
892    }
893    if (sym != 0) {
894      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
895      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
896      if (s == NULL) {
897        // We only allow an undefined symbol if this is a weak reference...
898        s = &si->symtab[sym];
899        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
900          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
901          return -1;
902        }
903
904        /* IHI0044C AAELF 4.5.1.1:
905
906           Libraries are not searched to resolve weak references.
907           It is not an error for a weak reference to remain unsatisfied.
908
909           During linking, the value of an undefined weak reference is:
910           - Zero if the relocation type is absolute
911           - The address of the place if the relocation is pc-relative
912           - The address of nominal base address if the relocation
913             type is base-relative.
914         */
915
916        switch (type) {
917#if defined(__aarch64__)
918        case R_AARCH64_JUMP_SLOT:
919        case R_AARCH64_GLOB_DAT:
920        case R_AARCH64_ABS64:
921        case R_AARCH64_ABS32:
922        case R_AARCH64_ABS16:
923        case R_AARCH64_RELATIVE:
924          /*
925           * The sym_addr was initialized to be zero above, or the relocation
926           * code below does not care about value of sym_addr.
927           * No need to do anything.
928           */
929          break;
930#elif defined(__x86_64__)
931        case R_X86_64_JUMP_SLOT:
932        case R_X86_64_GLOB_DAT:
933        case R_X86_64_32:
934        case R_X86_64_64:
935        case R_X86_64_RELATIVE:
936          // No need to do anything.
937          break;
938        case R_X86_64_PC32:
939          sym_addr = reloc;
940          break;
941#endif
942        default:
943          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
944          return -1;
945        }
946      } else {
947        // We got a definition.
948        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
949      }
950      count_relocation(kRelocSymbol);
951    } else {
952      s = NULL;
953    }
954
955    switch (type) {
956#if defined(__aarch64__)
957    case R_AARCH64_JUMP_SLOT:
958        count_relocation(kRelocAbsolute);
959        MARK(rela->r_offset);
960        TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
961                   reloc, (sym_addr + rela->r_addend), sym_name);
962        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
963        break;
964    case R_AARCH64_GLOB_DAT:
965        count_relocation(kRelocAbsolute);
966        MARK(rela->r_offset);
967        TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
968                   reloc, (sym_addr + rela->r_addend), sym_name);
969        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
970        break;
971    case R_AARCH64_ABS64:
972        count_relocation(kRelocAbsolute);
973        MARK(rela->r_offset);
974        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
975                   reloc, (sym_addr + rela->r_addend), sym_name);
976        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
977        break;
978    case R_AARCH64_ABS32:
979        count_relocation(kRelocAbsolute);
980        MARK(rela->r_offset);
981        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
982                   reloc, (sym_addr + rela->r_addend), sym_name);
983        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
984            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
985            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
986        } else {
987            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
988                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
989                   static_cast<ElfW(Addr)>(INT32_MIN),
990                   static_cast<ElfW(Addr)>(UINT32_MAX));
991            return -1;
992        }
993        break;
994    case R_AARCH64_ABS16:
995        count_relocation(kRelocAbsolute);
996        MARK(rela->r_offset);
997        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
998                   reloc, (sym_addr + rela->r_addend), sym_name);
999        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1000            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1001            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
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)),
1005                   static_cast<ElfW(Addr)>(INT16_MIN),
1006                   static_cast<ElfW(Addr)>(UINT16_MAX));
1007            return -1;
1008        }
1009        break;
1010    case R_AARCH64_PREL64:
1011        count_relocation(kRelocRelative);
1012        MARK(rela->r_offset);
1013        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
1014                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1015        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
1016        break;
1017    case R_AARCH64_PREL32:
1018        count_relocation(kRelocRelative);
1019        MARK(rela->r_offset);
1020        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
1021                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1022        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1023            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1024            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1025        } else {
1026            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1027                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1028                   static_cast<ElfW(Addr)>(INT32_MIN),
1029                   static_cast<ElfW(Addr)>(UINT32_MAX));
1030            return -1;
1031        }
1032        break;
1033    case R_AARCH64_PREL16:
1034        count_relocation(kRelocRelative);
1035        MARK(rela->r_offset);
1036        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1037                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1038        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1039            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1040            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1041        } else {
1042            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1043                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1044                   static_cast<ElfW(Addr)>(INT16_MIN),
1045                   static_cast<ElfW(Addr)>(UINT16_MAX));
1046            return -1;
1047        }
1048        break;
1049
1050    case R_AARCH64_RELATIVE:
1051        count_relocation(kRelocRelative);
1052        MARK(rela->r_offset);
1053        if (sym) {
1054            DL_ERR("odd RELATIVE form...");
1055            return -1;
1056        }
1057        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1058                   reloc, (si->base + rela->r_addend));
1059        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1060        break;
1061
1062    case R_AARCH64_COPY:
1063        /*
1064         * ET_EXEC is not supported so this should not happen.
1065         *
1066         * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1067         *
1068         * Section 4.7.1.10 "Dynamic relocations"
1069         * R_AARCH64_COPY may only appear in executable objects where e_type is
1070         * set to ET_EXEC.
1071         */
1072        DL_ERR("%s R_AARCH64_COPY relocations are not supported", si->name);
1073        return -1;
1074    case R_AARCH64_TLS_TPREL64:
1075        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1076                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1077        break;
1078    case R_AARCH64_TLS_DTPREL32:
1079        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1080                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1081        break;
1082#elif defined(__x86_64__)
1083    case R_X86_64_JUMP_SLOT:
1084      count_relocation(kRelocAbsolute);
1085      MARK(rela->r_offset);
1086      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1087                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1088      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1089      break;
1090    case R_X86_64_GLOB_DAT:
1091      count_relocation(kRelocAbsolute);
1092      MARK(rela->r_offset);
1093      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1094                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1095      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1096      break;
1097    case R_X86_64_RELATIVE:
1098      count_relocation(kRelocRelative);
1099      MARK(rela->r_offset);
1100      if (sym) {
1101        DL_ERR("odd RELATIVE form...");
1102        return -1;
1103      }
1104      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1105                 static_cast<size_t>(si->base));
1106      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1107      break;
1108    case R_X86_64_32:
1109      count_relocation(kRelocRelative);
1110      MARK(rela->r_offset);
1111      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1112                 static_cast<size_t>(sym_addr), sym_name);
1113      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1114      break;
1115    case R_X86_64_64:
1116      count_relocation(kRelocRelative);
1117      MARK(rela->r_offset);
1118      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1119                 static_cast<size_t>(sym_addr), sym_name);
1120      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1121      break;
1122    case R_X86_64_PC32:
1123      count_relocation(kRelocRelative);
1124      MARK(rela->r_offset);
1125      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1126                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1127                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1128      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1129      break;
1130#endif
1131
1132    default:
1133      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1134      return -1;
1135    }
1136  }
1137  return 0;
1138}
1139
1140#else // REL, not RELA.
1141
1142static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1143    ElfW(Sym)* s;
1144    soinfo* lsi;
1145
1146    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1147        unsigned type = ELFW(R_TYPE)(rel->r_info);
1148        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1149        unsigned sym = ELFW(R_SYM)(rel->r_info);
1150        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1151        ElfW(Addr) sym_addr = 0;
1152        const char* sym_name = NULL;
1153
1154        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1155        if (type == 0) { // R_*_NONE
1156            continue;
1157        }
1158        if (sym != 0) {
1159            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1160            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1161            if (s == NULL) {
1162                // We only allow an undefined symbol if this is a weak reference...
1163                s = &si->symtab[sym];
1164                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1165                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1166                    return -1;
1167                }
1168
1169                /* IHI0044C AAELF 4.5.1.1:
1170
1171                   Libraries are not searched to resolve weak references.
1172                   It is not an error for a weak reference to remain
1173                   unsatisfied.
1174
1175                   During linking, the value of an undefined weak reference is:
1176                   - Zero if the relocation type is absolute
1177                   - The address of the place if the relocation is pc-relative
1178                   - The address of nominal base address if the relocation
1179                     type is base-relative.
1180                  */
1181
1182                switch (type) {
1183#if defined(__arm__)
1184                case R_ARM_JUMP_SLOT:
1185                case R_ARM_GLOB_DAT:
1186                case R_ARM_ABS32:
1187                case R_ARM_RELATIVE:    /* Don't care. */
1188                    // sym_addr was initialized to be zero above or relocation
1189                    // code below does not care about value of sym_addr.
1190                    // No need to do anything.
1191                    break;
1192#elif defined(__i386__)
1193                case R_386_JMP_SLOT:
1194                case R_386_GLOB_DAT:
1195                case R_386_32:
1196                case R_386_RELATIVE:    /* Don't care. */
1197                    // sym_addr was initialized to be zero above or relocation
1198                    // code below does not care about value of sym_addr.
1199                    // No need to do anything.
1200                    break;
1201                case R_386_PC32:
1202                    sym_addr = reloc;
1203                    break;
1204#endif
1205
1206#if defined(__arm__)
1207                case R_ARM_COPY:
1208                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1209#endif
1210                default:
1211                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1212                    return -1;
1213                }
1214            } else {
1215                // We got a definition.
1216                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1217            }
1218            count_relocation(kRelocSymbol);
1219        } else {
1220            s = NULL;
1221        }
1222
1223        switch (type) {
1224#if defined(__arm__)
1225        case R_ARM_JUMP_SLOT:
1226            count_relocation(kRelocAbsolute);
1227            MARK(rel->r_offset);
1228            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1229            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1230            break;
1231        case R_ARM_GLOB_DAT:
1232            count_relocation(kRelocAbsolute);
1233            MARK(rel->r_offset);
1234            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1235            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1236            break;
1237        case R_ARM_ABS32:
1238            count_relocation(kRelocAbsolute);
1239            MARK(rel->r_offset);
1240            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1241            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1242            break;
1243        case R_ARM_REL32:
1244            count_relocation(kRelocRelative);
1245            MARK(rel->r_offset);
1246            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1247                       reloc, sym_addr, rel->r_offset, sym_name);
1248            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1249            break;
1250        case R_ARM_COPY:
1251            /*
1252             * ET_EXEC is not supported so this should not happen.
1253             *
1254             * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1255             *
1256             * Section 4.7.1.10 "Dynamic relocations"
1257             * R_ARM_COPY may only appear in executable objects where e_type is
1258             * set to ET_EXEC.
1259             */
1260            DL_ERR("%s R_ARM_COPY relocations are not supported", si->name);
1261            return -1;
1262#elif defined(__i386__)
1263        case R_386_JMP_SLOT:
1264            count_relocation(kRelocAbsolute);
1265            MARK(rel->r_offset);
1266            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1267            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1268            break;
1269        case R_386_GLOB_DAT:
1270            count_relocation(kRelocAbsolute);
1271            MARK(rel->r_offset);
1272            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1273            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1274            break;
1275        case R_386_32:
1276            count_relocation(kRelocRelative);
1277            MARK(rel->r_offset);
1278            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1279            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1280            break;
1281        case R_386_PC32:
1282            count_relocation(kRelocRelative);
1283            MARK(rel->r_offset);
1284            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1285                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1286            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1287            break;
1288#elif defined(__mips__)
1289        case R_MIPS_REL32:
1290#if defined(__LP64__)
1291            // MIPS Elf64_Rel entries contain compound relocations
1292            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1293            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1294                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1295                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1296                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1297                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1298                return -1;
1299            }
1300#endif
1301            count_relocation(kRelocAbsolute);
1302            MARK(rel->r_offset);
1303            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1304                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1305            if (s) {
1306                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1307            } else {
1308                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1309            }
1310            break;
1311#endif
1312
1313#if defined(__arm__)
1314        case R_ARM_RELATIVE:
1315#elif defined(__i386__)
1316        case R_386_RELATIVE:
1317#endif
1318            count_relocation(kRelocRelative);
1319            MARK(rel->r_offset);
1320            if (sym) {
1321                DL_ERR("odd RELATIVE form...");
1322                return -1;
1323            }
1324            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1325                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1326            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1327            break;
1328
1329        default:
1330            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1331            return -1;
1332        }
1333    }
1334    return 0;
1335}
1336#endif
1337
1338#if defined(__mips__)
1339static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1340    ElfW(Addr)** got = si->plt_got;
1341    if (got == NULL) {
1342        return true;
1343    }
1344    unsigned local_gotno = si->mips_local_gotno;
1345    unsigned gotsym = si->mips_gotsym;
1346    unsigned symtabno = si->mips_symtabno;
1347    ElfW(Sym)* symtab = si->symtab;
1348
1349    // got[0] is the address of the lazy resolver function.
1350    // got[1] may be used for a GNU extension.
1351    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1352    // FIXME: maybe this should be in a separate routine?
1353    if ((si->flags & FLAG_LINKER) == 0) {
1354        size_t g = 0;
1355        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1356        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1357            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1358        }
1359        // Relocate the local GOT entries.
1360        for (; g < local_gotno; g++) {
1361            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1362        }
1363    }
1364
1365    // Now for the global GOT entries...
1366    ElfW(Sym)* sym = symtab + gotsym;
1367    got = si->plt_got + local_gotno;
1368    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1369        // This is an undefined reference... try to locate it.
1370        const char* sym_name = si->strtab + sym->st_name;
1371        soinfo* lsi;
1372        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1373        if (s == NULL) {
1374            // We only allow an undefined symbol if this is a weak reference.
1375            s = &symtab[g];
1376            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1377                DL_ERR("cannot locate \"%s\"...", sym_name);
1378                return false;
1379            }
1380            *got = 0;
1381        } else {
1382            // FIXME: is this sufficient?
1383            // For reference see NetBSD link loader
1384            // 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
1385            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1386        }
1387    }
1388    return true;
1389}
1390#endif
1391
1392void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1393  if (functions == NULL) {
1394    return;
1395  }
1396
1397  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1398
1399  int begin = reverse ? (count - 1) : 0;
1400  int end = reverse ? -1 : count;
1401  int step = reverse ? -1 : 1;
1402
1403  for (int i = begin; i != end; i += step) {
1404    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1405    CallFunction("function", functions[i]);
1406  }
1407
1408  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1409}
1410
1411void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1412  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1413    return;
1414  }
1415
1416  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1417  function();
1418  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1419
1420  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1421  // are still writable. This happens with our debug malloc (see http://b/7941716).
1422  protect_data(PROT_READ | PROT_WRITE);
1423}
1424
1425void soinfo::CallPreInitConstructors() {
1426  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1427  // but ignored in a shared library.
1428  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1429}
1430
1431void soinfo::CallConstructors() {
1432  if (constructors_called) {
1433    return;
1434  }
1435
1436  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1437  // protect against recursive constructor calls. One simple example of constructor recursion
1438  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1439  // 1. The program depends on libc, so libc's constructor is called here.
1440  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1441  // 3. dlopen() calls the constructors on the newly created
1442  //    soinfo for libc_malloc_debug_leak.so.
1443  // 4. The debug .so depends on libc, so CallConstructors is
1444  //    called again with the libc soinfo. If it doesn't trigger the early-
1445  //    out above, the libc constructor will be called again (recursively!).
1446  constructors_called = true;
1447
1448  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1449    // The GNU dynamic linker silently ignores these, but we warn the developer.
1450    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1451          name, preinit_array_count);
1452  }
1453
1454  get_children().for_each([] (soinfo* si) {
1455    si->CallConstructors();
1456  });
1457
1458  TRACE("\"%s\": calling constructors", name);
1459
1460  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1461  CallFunction("DT_INIT", init_func);
1462  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1463}
1464
1465void soinfo::CallDestructors() {
1466  TRACE("\"%s\": calling destructors", name);
1467
1468  // DT_FINI_ARRAY must be parsed in reverse order.
1469  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1470
1471  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1472  CallFunction("DT_FINI", fini_func);
1473
1474  // This is needed on second call to dlopen
1475  // after library has been unloaded with RTLD_NODELETE
1476  constructors_called = false;
1477}
1478
1479void soinfo::add_child(soinfo* child) {
1480  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1481    return;
1482  }
1483
1484  this->children.push_front(child);
1485  child->parents.push_front(this);
1486}
1487
1488void soinfo::remove_all_links() {
1489  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1490    return;
1491  }
1492
1493  // 1. Untie connected soinfos from 'this'.
1494  children.for_each([&] (soinfo* child) {
1495    child->parents.remove_if([&] (const soinfo* parent) {
1496      return parent == this;
1497    });
1498  });
1499
1500  parents.for_each([&] (soinfo* parent) {
1501    parent->children.for_each([&] (const soinfo* child) {
1502      return child == this;
1503    });
1504  });
1505
1506  // 2. Once everything untied - clear local lists.
1507  parents.clear();
1508  children.clear();
1509}
1510
1511void soinfo::set_st_dev(dev_t dev) {
1512  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1513    return;
1514  }
1515
1516  st_dev = dev;
1517}
1518
1519void soinfo::set_st_ino(ino_t ino) {
1520  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1521    return;
1522  }
1523
1524  st_ino = ino;
1525}
1526
1527dev_t soinfo::get_st_dev() {
1528  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1529    return 0;
1530  }
1531
1532  return st_dev;
1533};
1534
1535ino_t soinfo::get_st_ino() {
1536  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1537    return 0;
1538  }
1539
1540  return st_ino;
1541}
1542
1543// This is a return on get_children() in case
1544// 'this->flags' does not have FLAG_NEW_SOINFO set.
1545static soinfo::soinfo_list_t g_empty_list;
1546
1547soinfo::soinfo_list_t& soinfo::get_children() {
1548  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1549    return g_empty_list;
1550  }
1551
1552  return this->children;
1553}
1554
1555/* Force any of the closed stdin, stdout and stderr to be associated with
1556   /dev/null. */
1557static int nullify_closed_stdio() {
1558    int dev_null, i, status;
1559    int return_value = 0;
1560
1561    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1562    if (dev_null < 0) {
1563        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1564        return -1;
1565    }
1566    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1567
1568    /* If any of the stdio file descriptors is valid and not associated
1569       with /dev/null, dup /dev/null to it.  */
1570    for (i = 0; i < 3; i++) {
1571        /* If it is /dev/null already, we are done. */
1572        if (i == dev_null) {
1573            continue;
1574        }
1575
1576        TRACE("[ Nullifying stdio file descriptor %d]", i);
1577        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1578
1579        /* If file is opened, we are good. */
1580        if (status != -1) {
1581            continue;
1582        }
1583
1584        /* The only error we allow is that the file descriptor does not
1585           exist, in which case we dup /dev/null to it. */
1586        if (errno != EBADF) {
1587            DL_ERR("fcntl failed: %s", strerror(errno));
1588            return_value = -1;
1589            continue;
1590        }
1591
1592        /* Try dupping /dev/null to this stdio file descriptor and
1593           repeat if there is a signal.  Note that any errors in closing
1594           the stdio descriptor are lost.  */
1595        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1596        if (status < 0) {
1597            DL_ERR("dup2 failed: %s", strerror(errno));
1598            return_value = -1;
1599            continue;
1600        }
1601    }
1602
1603    /* If /dev/null is not one of the stdio file descriptors, close it. */
1604    if (dev_null > 2) {
1605        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1606        status = TEMP_FAILURE_RETRY(close(dev_null));
1607        if (status == -1) {
1608            DL_ERR("close failed: %s", strerror(errno));
1609            return_value = -1;
1610        }
1611    }
1612
1613    return return_value;
1614}
1615
1616static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1617    /* "base" might wrap around UINT32_MAX. */
1618    ElfW(Addr) base = si->load_bias;
1619    const ElfW(Phdr)* phdr = si->phdr;
1620    int phnum = si->phnum;
1621    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1622
1623    /* We can't debug anything until the linker is relocated */
1624    if (!relocating_linker) {
1625        INFO("[ linking %s ]", si->name);
1626        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1627    }
1628
1629    /* Extract dynamic section */
1630    size_t dynamic_count;
1631    ElfW(Word) dynamic_flags;
1632    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1633                                   &dynamic_count, &dynamic_flags);
1634    if (si->dynamic == NULL) {
1635        if (!relocating_linker) {
1636            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1637        }
1638        return false;
1639    } else {
1640        if (!relocating_linker) {
1641            DEBUG("dynamic = %p", si->dynamic);
1642        }
1643    }
1644
1645#if defined(__arm__)
1646    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1647                                    &si->ARM_exidx, &si->ARM_exidx_count);
1648#endif
1649
1650    // Extract useful information from dynamic section.
1651    uint32_t needed_count = 0;
1652    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1653        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1654              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1655        switch (d->d_tag) {
1656        case DT_HASH:
1657            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1658            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1659            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1660            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1661            break;
1662        case DT_STRTAB:
1663            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1664            break;
1665        case DT_SYMTAB:
1666            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1667            break;
1668#if !defined(__LP64__)
1669        case DT_PLTREL:
1670            if (d->d_un.d_val != DT_REL) {
1671                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1672                return false;
1673            }
1674            break;
1675#endif
1676        case DT_JMPREL:
1677#if defined(USE_RELA)
1678            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1679#else
1680            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1681#endif
1682            break;
1683        case DT_PLTRELSZ:
1684#if defined(USE_RELA)
1685            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1686#else
1687            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1688#endif
1689            break;
1690#if defined(__mips__)
1691        case DT_PLTGOT:
1692            // Used by mips and mips64.
1693            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1694            break;
1695#endif
1696        case DT_DEBUG:
1697            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1698            // if the dynamic table is writable
1699// FIXME: not working currently for N64
1700// The flags for the LOAD and DYNAMIC program headers do not agree.
1701// The LOAD section containng the dynamic table has been mapped as
1702// read-only, but the DYNAMIC header claims it is writable.
1703#if !(defined(__mips__) && defined(__LP64__))
1704            if ((dynamic_flags & PF_W) != 0) {
1705                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1706            }
1707            break;
1708#endif
1709#if defined(USE_RELA)
1710         case DT_RELA:
1711            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1712            break;
1713         case DT_RELASZ:
1714            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1715            break;
1716        case DT_REL:
1717            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1718            return false;
1719        case DT_RELSZ:
1720            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1721            return false;
1722#else
1723        case DT_REL:
1724            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1725            break;
1726        case DT_RELSZ:
1727            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1728            break;
1729         case DT_RELA:
1730            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1731            return false;
1732#endif
1733        case DT_INIT:
1734            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1735            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1736            break;
1737        case DT_FINI:
1738            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1739            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1740            break;
1741        case DT_INIT_ARRAY:
1742            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1743            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1744            break;
1745        case DT_INIT_ARRAYSZ:
1746            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1747            break;
1748        case DT_FINI_ARRAY:
1749            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1750            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1751            break;
1752        case DT_FINI_ARRAYSZ:
1753            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1754            break;
1755        case DT_PREINIT_ARRAY:
1756            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1757            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1758            break;
1759        case DT_PREINIT_ARRAYSZ:
1760            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1761            break;
1762        case DT_TEXTREL:
1763#if defined(__LP64__)
1764            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1765            return false;
1766#else
1767            si->has_text_relocations = true;
1768            break;
1769#endif
1770        case DT_SYMBOLIC:
1771            si->has_DT_SYMBOLIC = true;
1772            break;
1773        case DT_NEEDED:
1774            ++needed_count;
1775            break;
1776        case DT_FLAGS:
1777            if (d->d_un.d_val & DF_TEXTREL) {
1778#if defined(__LP64__)
1779                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1780                return false;
1781#else
1782                si->has_text_relocations = true;
1783#endif
1784            }
1785            if (d->d_un.d_val & DF_SYMBOLIC) {
1786                si->has_DT_SYMBOLIC = true;
1787            }
1788            break;
1789#if defined(__mips__)
1790        case DT_STRSZ:
1791        case DT_SYMENT:
1792        case DT_RELENT:
1793             break;
1794        case DT_MIPS_RLD_MAP:
1795            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1796            {
1797              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1798              *dp = &_r_debug;
1799            }
1800            break;
1801        case DT_MIPS_RLD_VERSION:
1802        case DT_MIPS_FLAGS:
1803        case DT_MIPS_BASE_ADDRESS:
1804        case DT_MIPS_UNREFEXTNO:
1805            break;
1806
1807        case DT_MIPS_SYMTABNO:
1808            si->mips_symtabno = d->d_un.d_val;
1809            break;
1810
1811        case DT_MIPS_LOCAL_GOTNO:
1812            si->mips_local_gotno = d->d_un.d_val;
1813            break;
1814
1815        case DT_MIPS_GOTSYM:
1816            si->mips_gotsym = d->d_un.d_val;
1817            break;
1818#endif
1819
1820        default:
1821            DEBUG("Unused DT entry: type %p arg %p",
1822                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1823            break;
1824        }
1825    }
1826
1827    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1828          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1829
1830    // Sanity checks.
1831    if (relocating_linker && needed_count != 0) {
1832        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1833        return false;
1834    }
1835    if (si->nbucket == 0) {
1836        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1837        return false;
1838    }
1839    if (si->strtab == 0) {
1840        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1841        return false;
1842    }
1843    if (si->symtab == 0) {
1844        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1845        return false;
1846    }
1847
1848    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1849    if (si->flags & FLAG_EXE) {
1850        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1851        size_t preload_count = 0;
1852        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1853            soinfo* lsi = find_library(g_ld_preload_names[i], 0, NULL);
1854            if (lsi != NULL) {
1855                g_ld_preloads[preload_count++] = lsi;
1856            } else {
1857                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1858                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1859                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1860            }
1861        }
1862    }
1863
1864    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1865    soinfo** pneeded = needed;
1866
1867    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1868        if (d->d_tag == DT_NEEDED) {
1869            const char* library_name = si->strtab + d->d_un.d_val;
1870            DEBUG("%s needs %s", si->name, library_name);
1871            soinfo* lsi = find_library(library_name, 0, NULL);
1872            if (lsi == NULL) {
1873                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1874                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1875                       library_name, si->name, tmp_err_buf);
1876                return false;
1877            }
1878
1879            si->add_child(lsi);
1880            *pneeded++ = lsi;
1881        }
1882    }
1883    *pneeded = NULL;
1884
1885#if !defined(__LP64__)
1886    if (si->has_text_relocations) {
1887        // Make segments writable to allow text relocations to work properly. We will later call
1888        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1889        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1890                "security hardening. Please fix.", si->name);
1891        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1892            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1893                   si->name, strerror(errno));
1894            return false;
1895        }
1896    }
1897#endif
1898
1899#if defined(USE_RELA)
1900    if (si->plt_rela != NULL) {
1901        DEBUG("[ relocating %s plt ]\n", si->name);
1902        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1903            return false;
1904        }
1905    }
1906    if (si->rela != NULL) {
1907        DEBUG("[ relocating %s ]\n", si->name);
1908        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1909            return false;
1910        }
1911    }
1912#else
1913    if (si->plt_rel != NULL) {
1914        DEBUG("[ relocating %s plt ]", si->name);
1915        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1916            return false;
1917        }
1918    }
1919    if (si->rel != NULL) {
1920        DEBUG("[ relocating %s ]", si->name);
1921        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1922            return false;
1923        }
1924    }
1925#endif
1926
1927#if defined(__mips__)
1928    if (!mips_relocate_got(si, needed)) {
1929        return false;
1930    }
1931#endif
1932
1933    si->flags |= FLAG_LINKED;
1934    DEBUG("[ finished linking %s ]", si->name);
1935
1936#if !defined(__LP64__)
1937    if (si->has_text_relocations) {
1938        // All relocations are done, we can protect our segments back to read-only.
1939        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1940            DL_ERR("can't protect segments for \"%s\": %s",
1941                   si->name, strerror(errno));
1942            return false;
1943        }
1944    }
1945#endif
1946
1947    /* We can also turn on GNU RELRO protection */
1948    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1949        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1950               si->name, strerror(errno));
1951        return false;
1952    }
1953
1954    /* Handle serializing/sharing the RELRO segment */
1955    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1956      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
1957                                         extinfo->relro_fd) < 0) {
1958        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
1959               si->name, strerror(errno));
1960        return false;
1961      }
1962    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
1963      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
1964                                   extinfo->relro_fd) < 0) {
1965        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
1966               si->name, strerror(errno));
1967        return false;
1968      }
1969    }
1970
1971    notify_gdb_of_load(si);
1972    return true;
1973}
1974
1975/*
1976 * This function add vdso to internal dso list.
1977 * It helps to stack unwinding through signal handlers.
1978 * Also, it makes bionic more like glibc.
1979 */
1980static void add_vdso(KernelArgumentBlock& args __unused) {
1981#if defined(AT_SYSINFO_EHDR)
1982  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
1983  if (ehdr_vdso == NULL) {
1984    return;
1985  }
1986
1987  soinfo* si = soinfo_alloc("[vdso]", NULL);
1988
1989  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
1990  si->phnum = ehdr_vdso->e_phnum;
1991  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
1992  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
1993  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
1994
1995  soinfo_link_image(si, NULL);
1996#endif
1997}
1998
1999/*
2000 * This is linker soinfo for GDB. See details below.
2001 */
2002static soinfo linker_soinfo_for_gdb;
2003
2004/* gdb expects the linker to be in the debug shared object list.
2005 * Without this, gdb has trouble locating the linker's ".text"
2006 * and ".plt" sections. Gdb could also potentially use this to
2007 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2008 * Don't use soinfo_alloc(), because the linker shouldn't
2009 * be on the soinfo list.
2010 */
2011static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
2012#if defined(__LP64__)
2013  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
2014#else
2015  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
2016#endif
2017  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
2018  linker_soinfo_for_gdb.base = linker_base;
2019
2020  /*
2021   * Set the dynamic field in the link map otherwise gdb will complain with
2022   * the following:
2023   *   warning: .dynamic section for "/system/bin/linker" is not at the
2024   *   expected address (wrong library or version mismatch?)
2025   */
2026  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2027  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2028  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2029                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
2030  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2031}
2032
2033/*
2034 * This code is called after the linker has linked itself and
2035 * fixed it's own GOT. It is safe to make references to externs
2036 * and other non-local data at this point.
2037 */
2038static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2039    /* NOTE: we store the args pointer on a special location
2040     *       of the temporary TLS area in order to pass it to
2041     *       the C Library's runtime initializer.
2042     *
2043     *       The initializer must clear the slot and reset the TLS
2044     *       to point to a different location to ensure that no other
2045     *       shared library constructor can access it.
2046     */
2047  __libc_init_tls(args);
2048
2049#if TIMING
2050    struct timeval t0, t1;
2051    gettimeofday(&t0, 0);
2052#endif
2053
2054    // Initialize environment functions, and get to the ELF aux vectors table.
2055    linker_env_init(args);
2056
2057    // If this is a setuid/setgid program, close the security hole described in
2058    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2059    if (get_AT_SECURE()) {
2060        nullify_closed_stdio();
2061    }
2062
2063    debuggerd_init();
2064
2065    // Get a few environment variables.
2066    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2067    if (LD_DEBUG != NULL) {
2068      g_ld_debug_verbosity = atoi(LD_DEBUG);
2069    }
2070
2071    // Normally, these are cleaned by linker_env_init, but the test
2072    // doesn't cost us anything.
2073    const char* ldpath_env = NULL;
2074    const char* ldpreload_env = NULL;
2075    if (!get_AT_SECURE()) {
2076      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2077      ldpreload_env = linker_env_get("LD_PRELOAD");
2078    }
2079
2080    INFO("[ android linker & debugger ]");
2081
2082    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2083    if (si == NULL) {
2084        exit(EXIT_FAILURE);
2085    }
2086
2087    /* bootstrap the link map, the main exe always needs to be first */
2088    si->flags |= FLAG_EXE;
2089    link_map* map = &(si->link_map_head);
2090
2091    map->l_addr = 0;
2092    map->l_name = args.argv[0];
2093    map->l_prev = NULL;
2094    map->l_next = NULL;
2095
2096    _r_debug.r_map = map;
2097    r_debug_tail = map;
2098
2099    init_linker_info_for_gdb(linker_base);
2100
2101    // Extract information passed from the kernel.
2102    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2103    si->phnum = args.getauxval(AT_PHNUM);
2104    si->entry = args.getauxval(AT_ENTRY);
2105
2106    /* Compute the value of si->base. We can't rely on the fact that
2107     * the first entry is the PHDR because this will not be true
2108     * for certain executables (e.g. some in the NDK unit test suite)
2109     */
2110    si->base = 0;
2111    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2112    si->load_bias = 0;
2113    for (size_t i = 0; i < si->phnum; ++i) {
2114      if (si->phdr[i].p_type == PT_PHDR) {
2115        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2116        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2117        break;
2118      }
2119    }
2120    si->dynamic = NULL;
2121    si->ref_count = 1;
2122
2123    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2124    if (elf_hdr->e_type != ET_DYN) {
2125        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2126        exit(EXIT_FAILURE);
2127    }
2128
2129    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2130    parse_LD_LIBRARY_PATH(ldpath_env);
2131    parse_LD_PRELOAD(ldpreload_env);
2132
2133    somain = si;
2134
2135    if (!soinfo_link_image(si, NULL)) {
2136        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2137        exit(EXIT_FAILURE);
2138    }
2139
2140    add_vdso(args);
2141
2142    si->CallPreInitConstructors();
2143
2144    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2145        g_ld_preloads[i]->CallConstructors();
2146    }
2147
2148    /* After the link_image, the si->load_bias is initialized.
2149     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2150     * We need to update this value for so exe here. So Unwind_Backtrace
2151     * for some arch like x86 could work correctly within so exe.
2152     */
2153    map->l_addr = si->load_bias;
2154    si->CallConstructors();
2155
2156#if TIMING
2157    gettimeofday(&t1, NULL);
2158    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2159               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2160               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2161#endif
2162#if STATS
2163    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2164           linker_stats.count[kRelocAbsolute],
2165           linker_stats.count[kRelocRelative],
2166           linker_stats.count[kRelocCopy],
2167           linker_stats.count[kRelocSymbol]);
2168#endif
2169#if COUNT_PAGES
2170    {
2171        unsigned n;
2172        unsigned i;
2173        unsigned count = 0;
2174        for (n = 0; n < 4096; n++) {
2175            if (bitmask[n]) {
2176                unsigned x = bitmask[n];
2177#if defined(__LP64__)
2178                for (i = 0; i < 32; i++) {
2179#else
2180                for (i = 0; i < 8; i++) {
2181#endif
2182                    if (x & 1) {
2183                        count++;
2184                    }
2185                    x >>= 1;
2186                }
2187            }
2188        }
2189        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2190    }
2191#endif
2192
2193#if TIMING || STATS || COUNT_PAGES
2194    fflush(stdout);
2195#endif
2196
2197    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2198    return si->entry;
2199}
2200
2201/* Compute the load-bias of an existing executable. This shall only
2202 * be used to compute the load bias of an executable or shared library
2203 * that was loaded by the kernel itself.
2204 *
2205 * Input:
2206 *    elf    -> address of ELF header, assumed to be at the start of the file.
2207 * Return:
2208 *    load bias, i.e. add the value of any p_vaddr in the file to get
2209 *    the corresponding address in memory.
2210 */
2211static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2212  ElfW(Addr) offset = elf->e_phoff;
2213  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2214  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2215
2216  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2217    if (phdr->p_type == PT_LOAD) {
2218      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2219    }
2220  }
2221  return 0;
2222}
2223
2224/*
2225 * This is the entry point for the linker, called from begin.S. This
2226 * method is responsible for fixing the linker's own relocations, and
2227 * then calling __linker_init_post_relocation().
2228 *
2229 * Because this method is called before the linker has fixed it's own
2230 * relocations, any attempt to reference an extern variable, extern
2231 * function, or other GOT reference will generate a segfault.
2232 */
2233extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2234  // Initialize static variables.
2235  solist = get_libdl_info();
2236  sonext = get_libdl_info();
2237
2238  KernelArgumentBlock args(raw_args);
2239
2240  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2241  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2242  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2243
2244  soinfo linker_so;
2245  memset(&linker_so, 0, sizeof(soinfo));
2246
2247  strcpy(linker_so.name, "[dynamic linker]");
2248  linker_so.base = linker_addr;
2249  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2250  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2251  linker_so.dynamic = NULL;
2252  linker_so.phdr = phdr;
2253  linker_so.phnum = elf_hdr->e_phnum;
2254  linker_so.flags |= FLAG_LINKER;
2255
2256  if (!soinfo_link_image(&linker_so, NULL)) {
2257    // It would be nice to print an error message, but if the linker
2258    // can't link itself, there's no guarantee that we'll be able to
2259    // call write() (because it involves a GOT reference). We may as
2260    // well try though...
2261    const char* msg = "CANNOT LINK EXECUTABLE: ";
2262    write(2, msg, strlen(msg));
2263    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2264    write(2, "\n", 1);
2265    _exit(EXIT_FAILURE);
2266  }
2267
2268  // lets properly initialize global variables
2269  linker_so.CallConstructors();
2270
2271  // We have successfully fixed our own relocations. It's safe to run
2272  // the main part of the linker now.
2273  args.abort_message_ptr = &g_abort_message;
2274  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2275
2276  protect_data(PROT_READ);
2277
2278  // Return the address that the calling assembly stub should jump to.
2279  return start_address;
2280}
2281