linker.cpp revision 172955a4e30b88ce8239a7ef426b4e8903e9923c
1/*
2 * Copyright (C) 2008, 2009 The Android Open Source Project
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *  * Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 *  * Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in
12 *    the documentation and/or other materials provided with the
13 *    distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <dlfcn.h>
30#include <errno.h>
31#include <fcntl.h>
32#include <inttypes.h>
33#include <pthread.h>
34#include <stdio.h>
35#include <stdlib.h>
36#include <string.h>
37#include <sys/mman.h>
38#include <sys/stat.h>
39#include <unistd.h>
40
41// Private C library headers.
42#include "private/bionic_tls.h"
43#include "private/KernelArgumentBlock.h"
44#include "private/ScopedPthreadMutexLocker.h"
45#include "private/ScopedFd.h"
46
47#include "linker.h"
48#include "linker_debug.h"
49#include "linker_environ.h"
50#include "linker_phdr.h"
51#include "linker_allocator.h"
52
53/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
54 *
55 * Do NOT use malloc() and friends or pthread_*() code here.
56 * Don't use printf() either; it's caused mysterious memory
57 * corruption in the past.
58 * The linker runs before we bring up libc and it's easiest
59 * to make sure it does not depend on any complex libc features
60 *
61 * open issues / todo:
62 *
63 * - cleaner error reporting
64 * - after linking, set as much stuff as possible to READONLY
65 *   and NOEXEC
66 */
67
68#if defined(__LP64__)
69#define SEARCH_NAME(x) x
70#else
71// Nvidia drivers are relying on the bug:
72// http://code.google.com/p/android/issues/detail?id=6670
73// so we continue to use base-name lookup for lp32
74static const char* get_base_name(const char* name) {
75  const char* bname = strrchr(name, '/');
76  return bname ? bname + 1 : name;
77}
78#define SEARCH_NAME(x) get_base_name(x)
79#endif
80
81static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
82static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
83
84static LinkerAllocator<soinfo> g_soinfo_allocator;
85static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
86
87static soinfo* solist;
88static soinfo* sonext;
89static soinfo* somain; /* main process, always the one after libdl_info */
90
91static const char* const kDefaultLdPaths[] = {
92#if defined(__LP64__)
93  "/vendor/lib64",
94  "/system/lib64",
95#else
96  "/vendor/lib",
97  "/system/lib",
98#endif
99  NULL
100};
101
102#define LDPATH_BUFSIZE (LDPATH_MAX*64)
103#define LDPATH_MAX 8
104
105#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
106#define LDPRELOAD_MAX 8
107
108static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
109static const char* g_ld_library_paths[LDPATH_MAX + 1];
110
111static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
112static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
113
114static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
115
116__LIBC_HIDDEN__ int g_ld_debug_verbosity;
117
118__LIBC_HIDDEN__ abort_msg_t* g_abort_message = NULL; // For debuggerd.
119
120enum RelocationKind {
121    kRelocAbsolute = 0,
122    kRelocRelative,
123    kRelocCopy,
124    kRelocSymbol,
125    kRelocMax
126};
127
128#if STATS
129struct linker_stats_t {
130    int count[kRelocMax];
131};
132
133static linker_stats_t linker_stats;
134
135static void count_relocation(RelocationKind kind) {
136    ++linker_stats.count[kind];
137}
138#else
139static void count_relocation(RelocationKind) {
140}
141#endif
142
143#if COUNT_PAGES
144static unsigned bitmask[4096];
145#if defined(__LP64__)
146#define MARK(offset) \
147    do { \
148        if ((((offset) >> 12) >> 5) < 4096) \
149            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
150    } while (0)
151#else
152#define MARK(offset) \
153    do { \
154        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
155    } while (0)
156#endif
157#else
158#define MARK(x) do {} while (0)
159#endif
160
161// You shouldn't try to call memory-allocating functions in the dynamic linker.
162// Guard against the most obvious ones.
163#define DISALLOW_ALLOCATION(return_type, name, ...) \
164    return_type name __VA_ARGS__ \
165    { \
166        __libc_fatal("ERROR: " #name " called from the dynamic linker!\n"); \
167    }
168DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
169DISALLOW_ALLOCATION(void, free, (void* u __unused));
170DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
171DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
172
173static char tmp_err_buf[768];
174static char __linker_dl_err_buf[768];
175
176char* linker_get_error_buffer() {
177  return &__linker_dl_err_buf[0];
178}
179
180size_t linker_get_error_buffer_size() {
181  return sizeof(__linker_dl_err_buf);
182}
183
184/*
185 * This function is an empty stub where GDB locates a breakpoint to get notified
186 * about linker activity.
187 */
188extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
189
190static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
191static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
192static link_map* r_debug_tail = 0;
193
194static void insert_soinfo_into_debug_map(soinfo* info) {
195    // Copy the necessary fields into the debug structure.
196    link_map* map = &(info->link_map_head);
197    map->l_addr = info->load_bias;
198    map->l_name = reinterpret_cast<char*>(info->name);
199    map->l_ld = info->dynamic;
200
201    /* Stick the new library at the end of the list.
202     * gdb tends to care more about libc than it does
203     * about leaf libraries, and ordering it this way
204     * reduces the back-and-forth over the wire.
205     */
206    if (r_debug_tail) {
207        r_debug_tail->l_next = map;
208        map->l_prev = r_debug_tail;
209        map->l_next = 0;
210    } else {
211        _r_debug.r_map = map;
212        map->l_prev = 0;
213        map->l_next = 0;
214    }
215    r_debug_tail = map;
216}
217
218static void remove_soinfo_from_debug_map(soinfo* info) {
219    link_map* map = &(info->link_map_head);
220
221    if (r_debug_tail == map) {
222        r_debug_tail = map->l_prev;
223    }
224
225    if (map->l_prev) {
226        map->l_prev->l_next = map->l_next;
227    }
228    if (map->l_next) {
229        map->l_next->l_prev = map->l_prev;
230    }
231}
232
233static void notify_gdb_of_load(soinfo* info) {
234    if (info->flags & FLAG_EXE) {
235        // GDB already knows about the main executable
236        return;
237    }
238
239    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
240
241    _r_debug.r_state = r_debug::RT_ADD;
242    rtld_db_dlactivity();
243
244    insert_soinfo_into_debug_map(info);
245
246    _r_debug.r_state = r_debug::RT_CONSISTENT;
247    rtld_db_dlactivity();
248}
249
250static void notify_gdb_of_unload(soinfo* info) {
251    if (info->flags & FLAG_EXE) {
252        // GDB already knows about the main executable
253        return;
254    }
255
256    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
257
258    _r_debug.r_state = r_debug::RT_DELETE;
259    rtld_db_dlactivity();
260
261    remove_soinfo_from_debug_map(info);
262
263    _r_debug.r_state = r_debug::RT_CONSISTENT;
264    rtld_db_dlactivity();
265}
266
267void notify_gdb_of_libraries() {
268  _r_debug.r_state = r_debug::RT_ADD;
269  rtld_db_dlactivity();
270  _r_debug.r_state = r_debug::RT_CONSISTENT;
271  rtld_db_dlactivity();
272}
273
274LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
275  return g_soinfo_links_allocator.alloc();
276}
277
278void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
279  g_soinfo_links_allocator.free(entry);
280}
281
282static void protect_data(int protection) {
283  g_soinfo_allocator.protect_all(protection);
284  g_soinfo_links_allocator.protect_all(protection);
285}
286
287static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
288  if (strlen(name) >= SOINFO_NAME_LEN) {
289    DL_ERR("library name \"%s\" too long", name);
290    return NULL;
291  }
292
293  soinfo* si = g_soinfo_allocator.alloc();
294
295  // Initialize the new element.
296  memset(si, 0, sizeof(soinfo));
297  strlcpy(si->name, name, sizeof(si->name));
298  si->flags = FLAG_NEW_SOINFO;
299
300  if (file_stat != NULL) {
301    si->set_st_dev(file_stat->st_dev);
302    si->set_st_ino(file_stat->st_ino);
303  }
304
305  sonext->next = si;
306  sonext = si;
307
308  TRACE("name %s: allocated soinfo @ %p", name, si);
309  return si;
310}
311
312static void soinfo_free(soinfo* si) {
313    if (si == NULL) {
314        return;
315    }
316
317    if (si->base != 0 && si->size != 0) {
318      munmap(reinterpret_cast<void*>(si->base), si->size);
319    }
320
321    soinfo *prev = NULL, *trav;
322
323    TRACE("name %s: freeing soinfo @ %p", si->name, si);
324
325    for (trav = solist; trav != NULL; trav = trav->next) {
326        if (trav == si)
327            break;
328        prev = trav;
329    }
330    if (trav == NULL) {
331        /* si was not in solist */
332        DL_ERR("name \"%s\" is not in solist!", si->name);
333        return;
334    }
335
336    // clear links to/from si
337    si->remove_all_links();
338
339    /* prev will never be NULL, because the first entry in solist is
340       always the static libdl_info.
341    */
342    prev->next = si->next;
343    if (si == sonext) {
344        sonext = prev;
345    }
346
347    g_soinfo_allocator.free(si);
348}
349
350
351static void parse_path(const char* path, const char* delimiters,
352                       const char** array, char* buf, size_t buf_size, size_t max_count) {
353  if (path == NULL) {
354    return;
355  }
356
357  size_t len = strlcpy(buf, path, buf_size);
358
359  size_t i = 0;
360  char* buf_p = buf;
361  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
362    if (*array[i] != '\0') {
363      ++i;
364    }
365  }
366
367  // Forget the last path if we had to truncate; this occurs if the 2nd to
368  // last char isn't '\0' (i.e. wasn't originally a delimiter).
369  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
370    array[i - 1] = NULL;
371  } else {
372    array[i] = NULL;
373  }
374}
375
376static void parse_LD_LIBRARY_PATH(const char* path) {
377  parse_path(path, ":", g_ld_library_paths,
378             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
379}
380
381static void parse_LD_PRELOAD(const char* path) {
382  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
383  parse_path(path, " :", g_ld_preload_names,
384             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
385}
386
387#if defined(__arm__)
388
389/* For a given PC, find the .so that it belongs to.
390 * Returns the base address of the .ARM.exidx section
391 * for that .so, and the number of 8-byte entries
392 * in that section (via *pcount).
393 *
394 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
395 *
396 * This function is exposed via dlfcn.cpp and libdl.so.
397 */
398_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
399    unsigned addr = (unsigned)pc;
400
401    for (soinfo* si = solist; si != 0; si = si->next) {
402        if ((addr >= si->base) && (addr < (si->base + si->size))) {
403            *pcount = si->ARM_exidx_count;
404            return (_Unwind_Ptr)si->ARM_exidx;
405        }
406    }
407    *pcount = 0;
408    return NULL;
409}
410
411#endif
412
413/* Here, we only have to provide a callback to iterate across all the
414 * loaded libraries. gcc_eh does the rest. */
415int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
416    int rv = 0;
417    for (soinfo* si = solist; si != NULL; si = si->next) {
418        dl_phdr_info dl_info;
419        dl_info.dlpi_addr = si->link_map_head.l_addr;
420        dl_info.dlpi_name = si->link_map_head.l_name;
421        dl_info.dlpi_phdr = si->phdr;
422        dl_info.dlpi_phnum = si->phnum;
423        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
424        if (rv != 0) {
425            break;
426        }
427    }
428    return rv;
429}
430
431static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
432  ElfW(Sym)* symtab = si->symtab;
433  const char* strtab = si->strtab;
434
435  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
436             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
437
438  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
439    ElfW(Sym)* s = symtab + n;
440    if (strcmp(strtab + s->st_name, name)) continue;
441
442    /* only concern ourselves with global and weak symbol definitions */
443    switch (ELF_ST_BIND(s->st_info)) {
444      case STB_GLOBAL:
445      case STB_WEAK:
446        if (s->st_shndx == SHN_UNDEF) {
447          continue;
448        }
449
450        TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
451                 name, si->name, reinterpret_cast<void*>(s->st_value),
452                 static_cast<size_t>(s->st_size));
453        return s;
454      case STB_LOCAL:
455        continue;
456      default:
457        __libc_fatal("ERROR: Unexpected ST_BIND value: %d for '%s' in '%s'",
458            ELF_ST_BIND(s->st_info), name, si->name);
459    }
460  }
461
462  TRACE_TYPE(LOOKUP, "NOT FOUND %s in %s@%p %x %zd",
463             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
464
465
466  return NULL;
467}
468
469static unsigned elfhash(const char* _name) {
470    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
471    unsigned h = 0, g;
472
473    while (*name) {
474        h = (h << 4) + *name++;
475        g = h & 0xf0000000;
476        h ^= g;
477        h ^= g >> 24;
478    }
479    return h;
480}
481
482static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
483    unsigned elf_hash = elfhash(name);
484    ElfW(Sym)* s = NULL;
485
486    if (si != NULL && somain != NULL) {
487        /*
488         * Local scope is executable scope. Just start looking into it right away
489         * for the shortcut.
490         */
491
492        if (si == somain) {
493            s = soinfo_elf_lookup(si, elf_hash, name);
494            if (s != NULL) {
495                *lsi = si;
496                goto done;
497            }
498        } else {
499            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
500
501            /*
502             * If this object was built with symbolic relocations disabled, the
503             * first place to look to resolve external references is the main
504             * executable.
505             */
506
507            if (!si->has_DT_SYMBOLIC) {
508                DEBUG("%s: looking up %s in executable %s",
509                      si->name, name, somain->name);
510                s = soinfo_elf_lookup(somain, elf_hash, name);
511                if (s != NULL) {
512                    *lsi = somain;
513                    goto done;
514                }
515            }
516
517            /* Look for symbols in the local scope (the object who is
518             * searching). This happens with C++ templates on x86 for some
519             * reason.
520             *
521             * Notes on weak symbols:
522             * The ELF specs are ambiguous about treatment of weak definitions in
523             * dynamic linking.  Some systems return the first definition found
524             * and some the first non-weak definition.   This is system dependent.
525             * Here we return the first definition found for simplicity.  */
526
527            s = soinfo_elf_lookup(si, elf_hash, name);
528            if (s != NULL) {
529                *lsi = si;
530                goto done;
531            }
532
533            /*
534             * If this object was built with -Bsymbolic and symbol is not found
535             * in the local scope, try to find the symbol in the main executable.
536             */
537
538            if (si->has_DT_SYMBOLIC) {
539                DEBUG("%s: looking up %s in executable %s after local scope",
540                      si->name, name, somain->name);
541                s = soinfo_elf_lookup(somain, elf_hash, name);
542                if (s != NULL) {
543                    *lsi = somain;
544                    goto done;
545                }
546            }
547        }
548    }
549
550    /* Next, look for it in the preloads list */
551    for (int i = 0; g_ld_preloads[i] != NULL; i++) {
552        s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
553        if (s != NULL) {
554            *lsi = g_ld_preloads[i];
555            goto done;
556        }
557    }
558
559    for (int i = 0; needed[i] != NULL; i++) {
560        DEBUG("%s: looking up %s in %s",
561              si->name, name, needed[i]->name);
562        s = soinfo_elf_lookup(needed[i], elf_hash, name);
563        if (s != NULL) {
564            *lsi = needed[i];
565            goto done;
566        }
567    }
568
569done:
570    if (s != NULL) {
571        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
572                   "found in %s, base = %p, load bias = %p",
573                   si->name, name, reinterpret_cast<void*>(s->st_value),
574                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
575                   reinterpret_cast<void*>((*lsi)->load_bias));
576        return s;
577    }
578
579    return NULL;
580}
581
582// Another soinfo list allocator to use in dlsym. We don't reuse
583// SoinfoListAllocator because it is write-protected most of the time.
584static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_list_allocator_rw;
585class SoinfoListAllocatorRW {
586 public:
587  static LinkedListEntry<soinfo>* alloc() {
588    return g_soinfo_list_allocator_rw.alloc();
589  }
590
591  static void free(LinkedListEntry<soinfo>* ptr) {
592    g_soinfo_list_allocator_rw.free(ptr);
593  }
594};
595
596// This is used by dlsym(3).  It performs symbol lookup only within the
597// specified soinfo object and its dependencies in breadth first order.
598ElfW(Sym)* dlsym_handle_lookup(soinfo* si, soinfo** found, const char* name) {
599  LinkedList<soinfo, SoinfoListAllocatorRW> visit_list;
600  LinkedList<soinfo, SoinfoListAllocatorRW> visited;
601  visit_list.push_back(si);
602  soinfo* current_soinfo;
603  while ((current_soinfo = visit_list.pop_front()) != nullptr) {
604    if (visited.contains(current_soinfo)) {
605      continue;
606    }
607
608    ElfW(Sym)* result = soinfo_elf_lookup(current_soinfo, elfhash(name), name);
609
610    if (result != nullptr) {
611      *found = current_soinfo;
612      visit_list.clear();
613      visited.clear();
614      return result;
615    }
616    visited.push_back(current_soinfo);
617
618    current_soinfo->get_children().for_each([&](soinfo* child) {
619      visit_list.push_back(child);
620    });
621  }
622
623  visit_list.clear();
624  visited.clear();
625  return nullptr;
626}
627
628/* This is used by dlsym(3) to performs a global symbol lookup. If the
629   start value is null (for RTLD_DEFAULT), the search starts at the
630   beginning of the global solist. Otherwise the search starts at the
631   specified soinfo (for RTLD_NEXT).
632 */
633ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
634  unsigned elf_hash = elfhash(name);
635
636  if (start == NULL) {
637    start = solist;
638  }
639
640  ElfW(Sym)* s = NULL;
641  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
642    s = soinfo_elf_lookup(si, elf_hash, name);
643    if (s != NULL) {
644      *found = si;
645      break;
646    }
647  }
648
649  if (s != NULL) {
650    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
651               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
652  }
653
654  return s;
655}
656
657soinfo* find_containing_library(const void* p) {
658  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
659  for (soinfo* si = solist; si != NULL; si = si->next) {
660    if (address >= si->base && address - si->base < si->size) {
661      return si;
662    }
663  }
664  return NULL;
665}
666
667ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
668  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
669
670  // Search the library's symbol table for any defined symbol which
671  // contains this address.
672  for (size_t i = 0; i < si->nchain; ++i) {
673    ElfW(Sym)* sym = &si->symtab[i];
674    if (sym->st_shndx != SHN_UNDEF &&
675        soaddr >= sym->st_value &&
676        soaddr < sym->st_value + sym->st_size) {
677      return sym;
678    }
679  }
680
681  return NULL;
682}
683
684static int open_library_on_path(const char* name, const char* const paths[]) {
685  char buf[512];
686  for (size_t i = 0; paths[i] != NULL; ++i) {
687    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
688    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
689      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
690      continue;
691    }
692    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
693    if (fd != -1) {
694      return fd;
695    }
696  }
697  return -1;
698}
699
700static int open_library(const char* name) {
701  TRACE("[ opening %s ]", name);
702
703  // If the name contains a slash, we should attempt to open it directly and not search the paths.
704  if (strchr(name, '/') != NULL) {
705    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
706    if (fd != -1) {
707      return fd;
708    }
709    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
710#if defined(__LP64__)
711    return -1;
712#endif
713  }
714
715  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
716  int fd = open_library_on_path(name, g_ld_library_paths);
717  if (fd == -1) {
718    fd = open_library_on_path(name, kDefaultLdPaths);
719  }
720  return fd;
721}
722
723static soinfo* load_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
724    int fd = -1;
725    ScopedFd file_guard(-1);
726
727    if (extinfo != NULL && (extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD) != 0) {
728      fd = extinfo->library_fd;
729    } else {
730      // Open the file.
731      fd = open_library(name);
732      if (fd == -1) {
733        DL_ERR("library \"%s\" not found", name);
734        return NULL;
735      }
736
737      file_guard.reset(fd);
738    }
739
740    ElfReader elf_reader(name, fd);
741
742    struct stat file_stat;
743    if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
744      DL_ERR("unable to stat file for the library %s: %s", name, strerror(errno));
745      return NULL;
746    }
747
748    // Check for symlink and other situations where
749    // file can have different names.
750    for (soinfo* si = solist; si != NULL; si = si->next) {
751      if (si->get_st_dev() != 0 &&
752          si->get_st_ino() != 0 &&
753          si->get_st_dev() == file_stat.st_dev &&
754          si->get_st_ino() == file_stat.st_ino) {
755        TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
756        return si;
757      }
758    }
759
760    if ((dlflags & RTLD_NOLOAD) != 0) {
761      return NULL;
762    }
763
764    // Read the ELF header and load the segments.
765    if (!elf_reader.Load(extinfo)) {
766        return NULL;
767    }
768
769    soinfo* si = soinfo_alloc(SEARCH_NAME(name), &file_stat);
770    if (si == NULL) {
771        return NULL;
772    }
773    si->base = elf_reader.load_start();
774    si->size = elf_reader.load_size();
775    si->load_bias = elf_reader.load_bias();
776    si->phnum = elf_reader.phdr_count();
777    si->phdr = elf_reader.loaded_phdr();
778
779    // At this point we know that whatever is loaded @ base is a valid ELF
780    // shared library whose segments are properly mapped in.
781    TRACE("[ load_library base=%p size=%zu name='%s' ]",
782          reinterpret_cast<void*>(si->base), si->size, si->name);
783
784    if (!soinfo_link_image(si, extinfo)) {
785      soinfo_free(si);
786      return NULL;
787    }
788
789    return si;
790}
791
792static soinfo *find_loaded_library_by_name(const char* name) {
793  const char* search_name = SEARCH_NAME(name);
794  for (soinfo* si = solist; si != NULL; si = si->next) {
795    if (!strcmp(search_name, si->name)) {
796      return si;
797    }
798  }
799  return NULL;
800}
801
802static soinfo* find_library_internal(const char* name, int dlflags, const android_dlextinfo* extinfo) {
803  if (name == NULL) {
804    return somain;
805  }
806
807  soinfo* si = find_loaded_library_by_name(name);
808
809  // Library might still be loaded, the accurate detection
810  // of this fact is done by load_library
811  if (si == NULL) {
812    TRACE("[ '%s' has not been found by name.  Trying harder...]", name);
813    si = load_library(name, dlflags, extinfo);
814  }
815
816  if (si != NULL && (si->flags & FLAG_LINKED) == 0) {
817    DL_ERR("recursive link to \"%s\"", si->name);
818    return NULL;
819  }
820
821  return si;
822}
823
824static soinfo* find_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
825  soinfo* si = find_library_internal(name, dlflags, extinfo);
826  if (si != NULL) {
827    si->ref_count++;
828  }
829  return si;
830}
831
832static void soinfo_unload(soinfo* si) {
833  if (si->ref_count == 1) {
834    TRACE("unloading '%s'", si->name);
835    si->CallDestructors();
836
837    if ((si->flags | FLAG_NEW_SOINFO) != 0) {
838      si->get_children().for_each([&] (soinfo* child) {
839        TRACE("%s needs to unload %s", si->name, child->name);
840        soinfo_unload(child);
841      });
842    } else {
843      for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
844        if (d->d_tag == DT_NEEDED) {
845          const char* library_name = si->strtab + d->d_un.d_val;
846          TRACE("%s needs to unload %s", si->name, library_name);
847          soinfo* needed = find_library(library_name, RTLD_NOLOAD, NULL);
848          if (needed != NULL) {
849            soinfo_unload(needed);
850          } else {
851            // Not found: for example if symlink was deleted between dlopen and dlclose
852            // Since we cannot really handle errors at this point - print and continue.
853            PRINT("warning: couldn't find %s needed by %s on unload.", library_name, si->name);
854          }
855        }
856      }
857    }
858
859    notify_gdb_of_unload(si);
860    si->ref_count = 0;
861    soinfo_free(si);
862  } else {
863    si->ref_count--;
864    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
865  }
866}
867
868void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
869  // Use basic string manipulation calls to avoid snprintf.
870  // snprintf indirectly calls pthread_getspecific to get the size of a buffer.
871  // When debug malloc is enabled, this call returns 0. This in turn causes
872  // snprintf to do nothing, which causes libraries to fail to load.
873  // See b/17302493 for further details.
874  // Once the above bug is fixed, this code can be modified to use
875  // snprintf again.
876  size_t required_len = strlen(kDefaultLdPaths[0]) + strlen(kDefaultLdPaths[1]) + 2;
877  if (buffer_size < required_len) {
878    __libc_fatal("android_get_LD_LIBRARY_PATH failed, buffer too small: buffer len %zu, required len %zu",
879                 buffer_size, required_len);
880  }
881  char* end = stpcpy(buffer, kDefaultLdPaths[0]);
882  *end = ':';
883  strcpy(end + 1, kDefaultLdPaths[1]);
884}
885
886void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
887  if (!get_AT_SECURE()) {
888    parse_LD_LIBRARY_PATH(ld_library_path);
889  }
890}
891
892soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
893  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL|RTLD_NOLOAD)) != 0) {
894    DL_ERR("invalid flags to dlopen: %x", flags);
895    return NULL;
896  }
897  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
898    DL_ERR("invalid extended flags to android_dlopen_ext: %" PRIx64, extinfo->flags);
899    return NULL;
900  }
901  protect_data(PROT_READ | PROT_WRITE);
902  soinfo* si = find_library(name, flags, extinfo);
903  if (si != NULL) {
904    si->CallConstructors();
905  }
906  protect_data(PROT_READ);
907  return si;
908}
909
910void do_dlclose(soinfo* si) {
911  protect_data(PROT_READ | PROT_WRITE);
912  soinfo_unload(si);
913  protect_data(PROT_READ);
914}
915
916#if defined(USE_RELA)
917static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
918  ElfW(Sym)* s;
919  soinfo* lsi;
920
921  for (size_t idx = 0; idx < count; ++idx, ++rela) {
922    unsigned type = ELFW(R_TYPE)(rela->r_info);
923    unsigned sym = ELFW(R_SYM)(rela->r_info);
924    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
925    ElfW(Addr) sym_addr = 0;
926    const char* sym_name = NULL;
927
928    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
929    if (type == 0) { // R_*_NONE
930      continue;
931    }
932    if (sym != 0) {
933      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
934      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
935      if (s == NULL) {
936        // We only allow an undefined symbol if this is a weak reference...
937        s = &si->symtab[sym];
938        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
939          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
940          return -1;
941        }
942
943        /* IHI0044C AAELF 4.5.1.1:
944
945           Libraries are not searched to resolve weak references.
946           It is not an error for a weak reference to remain unsatisfied.
947
948           During linking, the value of an undefined weak reference is:
949           - Zero if the relocation type is absolute
950           - The address of the place if the relocation is pc-relative
951           - The address of nominal base address if the relocation
952             type is base-relative.
953         */
954
955        switch (type) {
956#if defined(__aarch64__)
957        case R_AARCH64_JUMP_SLOT:
958        case R_AARCH64_GLOB_DAT:
959        case R_AARCH64_ABS64:
960        case R_AARCH64_ABS32:
961        case R_AARCH64_ABS16:
962        case R_AARCH64_RELATIVE:
963          /*
964           * The sym_addr was initialized to be zero above, or the relocation
965           * code below does not care about value of sym_addr.
966           * No need to do anything.
967           */
968          break;
969#elif defined(__x86_64__)
970        case R_X86_64_JUMP_SLOT:
971        case R_X86_64_GLOB_DAT:
972        case R_X86_64_32:
973        case R_X86_64_64:
974        case R_X86_64_RELATIVE:
975          // No need to do anything.
976          break;
977        case R_X86_64_PC32:
978          sym_addr = reloc;
979          break;
980#endif
981        default:
982          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
983          return -1;
984        }
985      } else {
986        // We got a definition.
987        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
988      }
989      count_relocation(kRelocSymbol);
990    } else {
991      s = NULL;
992    }
993
994    switch (type) {
995#if defined(__aarch64__)
996    case R_AARCH64_JUMP_SLOT:
997        count_relocation(kRelocAbsolute);
998        MARK(rela->r_offset);
999        TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
1000                   reloc, (sym_addr + rela->r_addend), sym_name);
1001        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
1002        break;
1003    case R_AARCH64_GLOB_DAT:
1004        count_relocation(kRelocAbsolute);
1005        MARK(rela->r_offset);
1006        TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
1007                   reloc, (sym_addr + rela->r_addend), sym_name);
1008        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
1009        break;
1010    case R_AARCH64_ABS64:
1011        count_relocation(kRelocAbsolute);
1012        MARK(rela->r_offset);
1013        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
1014                   reloc, (sym_addr + rela->r_addend), sym_name);
1015        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1016        break;
1017    case R_AARCH64_ABS32:
1018        count_relocation(kRelocAbsolute);
1019        MARK(rela->r_offset);
1020        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
1021                   reloc, (sym_addr + rela->r_addend), sym_name);
1022        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1023            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1024            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
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)),
1028                   static_cast<ElfW(Addr)>(INT32_MIN),
1029                   static_cast<ElfW(Addr)>(UINT32_MAX));
1030            return -1;
1031        }
1032        break;
1033    case R_AARCH64_ABS16:
1034        count_relocation(kRelocAbsolute);
1035        MARK(rela->r_offset);
1036        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
1037                   reloc, (sym_addr + rela->r_addend), sym_name);
1038        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1039            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1040            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
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)),
1044                   static_cast<ElfW(Addr)>(INT16_MIN),
1045                   static_cast<ElfW(Addr)>(UINT16_MAX));
1046            return -1;
1047        }
1048        break;
1049    case R_AARCH64_PREL64:
1050        count_relocation(kRelocRelative);
1051        MARK(rela->r_offset);
1052        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
1053                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1054        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
1055        break;
1056    case R_AARCH64_PREL32:
1057        count_relocation(kRelocRelative);
1058        MARK(rela->r_offset);
1059        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
1060                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1061        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1062            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1063            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1064        } else {
1065            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1066                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1067                   static_cast<ElfW(Addr)>(INT32_MIN),
1068                   static_cast<ElfW(Addr)>(UINT32_MAX));
1069            return -1;
1070        }
1071        break;
1072    case R_AARCH64_PREL16:
1073        count_relocation(kRelocRelative);
1074        MARK(rela->r_offset);
1075        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1076                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1077        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1078            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1079            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1080        } else {
1081            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1082                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1083                   static_cast<ElfW(Addr)>(INT16_MIN),
1084                   static_cast<ElfW(Addr)>(UINT16_MAX));
1085            return -1;
1086        }
1087        break;
1088
1089    case R_AARCH64_RELATIVE:
1090        count_relocation(kRelocRelative);
1091        MARK(rela->r_offset);
1092        if (sym) {
1093            DL_ERR("odd RELATIVE form...");
1094            return -1;
1095        }
1096        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1097                   reloc, (si->base + rela->r_addend));
1098        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1099        break;
1100
1101    case R_AARCH64_COPY:
1102        /*
1103         * ET_EXEC is not supported so this should not happen.
1104         *
1105         * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1106         *
1107         * Section 4.7.1.10 "Dynamic relocations"
1108         * R_AARCH64_COPY may only appear in executable objects where e_type is
1109         * set to ET_EXEC.
1110         */
1111        DL_ERR("%s R_AARCH64_COPY relocations are not supported", si->name);
1112        return -1;
1113    case R_AARCH64_TLS_TPREL64:
1114        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1115                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1116        break;
1117    case R_AARCH64_TLS_DTPREL32:
1118        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1119                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1120        break;
1121#elif defined(__x86_64__)
1122    case R_X86_64_JUMP_SLOT:
1123      count_relocation(kRelocAbsolute);
1124      MARK(rela->r_offset);
1125      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1126                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1127      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1128      break;
1129    case R_X86_64_GLOB_DAT:
1130      count_relocation(kRelocAbsolute);
1131      MARK(rela->r_offset);
1132      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1133                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1134      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1135      break;
1136    case R_X86_64_RELATIVE:
1137      count_relocation(kRelocRelative);
1138      MARK(rela->r_offset);
1139      if (sym) {
1140        DL_ERR("odd RELATIVE form...");
1141        return -1;
1142      }
1143      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1144                 static_cast<size_t>(si->base));
1145      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1146      break;
1147    case R_X86_64_32:
1148      count_relocation(kRelocRelative);
1149      MARK(rela->r_offset);
1150      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1151                 static_cast<size_t>(sym_addr), sym_name);
1152      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1153      break;
1154    case R_X86_64_64:
1155      count_relocation(kRelocRelative);
1156      MARK(rela->r_offset);
1157      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1158                 static_cast<size_t>(sym_addr), sym_name);
1159      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1160      break;
1161    case R_X86_64_PC32:
1162      count_relocation(kRelocRelative);
1163      MARK(rela->r_offset);
1164      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1165                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1166                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1167      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1168      break;
1169#endif
1170
1171    default:
1172      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1173      return -1;
1174    }
1175  }
1176  return 0;
1177}
1178
1179#else // REL, not RELA.
1180
1181static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1182    ElfW(Sym)* s;
1183    soinfo* lsi;
1184
1185    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1186        unsigned type = ELFW(R_TYPE)(rel->r_info);
1187        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1188        unsigned sym = ELFW(R_SYM)(rel->r_info);
1189        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1190        ElfW(Addr) sym_addr = 0;
1191        const char* sym_name = NULL;
1192
1193        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1194        if (type == 0) { // R_*_NONE
1195            continue;
1196        }
1197        if (sym != 0) {
1198            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1199            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1200            if (s == NULL) {
1201                // We only allow an undefined symbol if this is a weak reference...
1202                s = &si->symtab[sym];
1203                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1204                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1205                    return -1;
1206                }
1207
1208                /* IHI0044C AAELF 4.5.1.1:
1209
1210                   Libraries are not searched to resolve weak references.
1211                   It is not an error for a weak reference to remain
1212                   unsatisfied.
1213
1214                   During linking, the value of an undefined weak reference is:
1215                   - Zero if the relocation type is absolute
1216                   - The address of the place if the relocation is pc-relative
1217                   - The address of nominal base address if the relocation
1218                     type is base-relative.
1219                  */
1220
1221                switch (type) {
1222#if defined(__arm__)
1223                case R_ARM_JUMP_SLOT:
1224                case R_ARM_GLOB_DAT:
1225                case R_ARM_ABS32:
1226                case R_ARM_RELATIVE:    /* Don't care. */
1227                    // sym_addr was initialized to be zero above or relocation
1228                    // code below does not care about value of sym_addr.
1229                    // No need to do anything.
1230                    break;
1231#elif defined(__i386__)
1232                case R_386_JMP_SLOT:
1233                case R_386_GLOB_DAT:
1234                case R_386_32:
1235                case R_386_RELATIVE:    /* Don't care. */
1236                    // sym_addr was initialized to be zero above or relocation
1237                    // code below does not care about value of sym_addr.
1238                    // No need to do anything.
1239                    break;
1240                case R_386_PC32:
1241                    sym_addr = reloc;
1242                    break;
1243#endif
1244
1245#if defined(__arm__)
1246                case R_ARM_COPY:
1247                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1248#endif
1249                default:
1250                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1251                    return -1;
1252                }
1253            } else {
1254                // We got a definition.
1255                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1256            }
1257            count_relocation(kRelocSymbol);
1258        } else {
1259            s = NULL;
1260        }
1261
1262        switch (type) {
1263#if defined(__arm__)
1264        case R_ARM_JUMP_SLOT:
1265            count_relocation(kRelocAbsolute);
1266            MARK(rel->r_offset);
1267            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1268            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1269            break;
1270        case R_ARM_GLOB_DAT:
1271            count_relocation(kRelocAbsolute);
1272            MARK(rel->r_offset);
1273            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1274            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1275            break;
1276        case R_ARM_ABS32:
1277            count_relocation(kRelocAbsolute);
1278            MARK(rel->r_offset);
1279            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1280            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1281            break;
1282        case R_ARM_REL32:
1283            count_relocation(kRelocRelative);
1284            MARK(rel->r_offset);
1285            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1286                       reloc, sym_addr, rel->r_offset, sym_name);
1287            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1288            break;
1289        case R_ARM_COPY:
1290            /*
1291             * ET_EXEC is not supported so this should not happen.
1292             *
1293             * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1294             *
1295             * Section 4.7.1.10 "Dynamic relocations"
1296             * R_ARM_COPY may only appear in executable objects where e_type is
1297             * set to ET_EXEC.
1298             */
1299            DL_ERR("%s R_ARM_COPY relocations are not supported", si->name);
1300            return -1;
1301#elif defined(__i386__)
1302        case R_386_JMP_SLOT:
1303            count_relocation(kRelocAbsolute);
1304            MARK(rel->r_offset);
1305            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1306            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1307            break;
1308        case R_386_GLOB_DAT:
1309            count_relocation(kRelocAbsolute);
1310            MARK(rel->r_offset);
1311            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1312            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1313            break;
1314        case R_386_32:
1315            count_relocation(kRelocRelative);
1316            MARK(rel->r_offset);
1317            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1318            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1319            break;
1320        case R_386_PC32:
1321            count_relocation(kRelocRelative);
1322            MARK(rel->r_offset);
1323            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1324                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1325            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1326            break;
1327#elif defined(__mips__)
1328        case R_MIPS_REL32:
1329#if defined(__LP64__)
1330            // MIPS Elf64_Rel entries contain compound relocations
1331            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1332            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1333                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1334                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1335                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1336                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1337                return -1;
1338            }
1339#endif
1340            count_relocation(kRelocAbsolute);
1341            MARK(rel->r_offset);
1342            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1343                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1344            if (s) {
1345                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1346            } else {
1347                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1348            }
1349            break;
1350#endif
1351
1352#if defined(__arm__)
1353        case R_ARM_RELATIVE:
1354#elif defined(__i386__)
1355        case R_386_RELATIVE:
1356#endif
1357            count_relocation(kRelocRelative);
1358            MARK(rel->r_offset);
1359            if (sym) {
1360                DL_ERR("odd RELATIVE form...");
1361                return -1;
1362            }
1363            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1364                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1365            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1366            break;
1367
1368        default:
1369            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1370            return -1;
1371        }
1372    }
1373    return 0;
1374}
1375#endif
1376
1377#if defined(__mips__)
1378static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1379    ElfW(Addr)** got = si->plt_got;
1380    if (got == NULL) {
1381        return true;
1382    }
1383    unsigned local_gotno = si->mips_local_gotno;
1384    unsigned gotsym = si->mips_gotsym;
1385    unsigned symtabno = si->mips_symtabno;
1386    ElfW(Sym)* symtab = si->symtab;
1387
1388    // got[0] is the address of the lazy resolver function.
1389    // got[1] may be used for a GNU extension.
1390    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1391    // FIXME: maybe this should be in a separate routine?
1392    if ((si->flags & FLAG_LINKER) == 0) {
1393        size_t g = 0;
1394        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1395        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1396            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1397        }
1398        // Relocate the local GOT entries.
1399        for (; g < local_gotno; g++) {
1400            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1401        }
1402    }
1403
1404    // Now for the global GOT entries...
1405    ElfW(Sym)* sym = symtab + gotsym;
1406    got = si->plt_got + local_gotno;
1407    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1408        // This is an undefined reference... try to locate it.
1409        const char* sym_name = si->strtab + sym->st_name;
1410        soinfo* lsi;
1411        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1412        if (s == NULL) {
1413            // We only allow an undefined symbol if this is a weak reference.
1414            s = &symtab[g];
1415            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1416                DL_ERR("cannot locate \"%s\"...", sym_name);
1417                return false;
1418            }
1419            *got = 0;
1420        } else {
1421            // FIXME: is this sufficient?
1422            // For reference see NetBSD link loader
1423            // 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
1424            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1425        }
1426    }
1427    return true;
1428}
1429#endif
1430
1431void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1432  if (functions == NULL) {
1433    return;
1434  }
1435
1436  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1437
1438  int begin = reverse ? (count - 1) : 0;
1439  int end = reverse ? -1 : count;
1440  int step = reverse ? -1 : 1;
1441
1442  for (int i = begin; i != end; i += step) {
1443    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1444    CallFunction("function", functions[i]);
1445  }
1446
1447  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1448}
1449
1450void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1451  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1452    return;
1453  }
1454
1455  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1456  function();
1457  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1458
1459  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1460  // are still writable. This happens with our debug malloc (see http://b/7941716).
1461  protect_data(PROT_READ | PROT_WRITE);
1462}
1463
1464void soinfo::CallPreInitConstructors() {
1465  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1466  // but ignored in a shared library.
1467  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1468}
1469
1470void soinfo::CallConstructors() {
1471  if (constructors_called) {
1472    return;
1473  }
1474
1475  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1476  // protect against recursive constructor calls. One simple example of constructor recursion
1477  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1478  // 1. The program depends on libc, so libc's constructor is called here.
1479  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1480  // 3. dlopen() calls the constructors on the newly created
1481  //    soinfo for libc_malloc_debug_leak.so.
1482  // 4. The debug .so depends on libc, so CallConstructors is
1483  //    called again with the libc soinfo. If it doesn't trigger the early-
1484  //    out above, the libc constructor will be called again (recursively!).
1485  constructors_called = true;
1486
1487  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1488    // The GNU dynamic linker silently ignores these, but we warn the developer.
1489    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1490          name, preinit_array_count);
1491  }
1492
1493  get_children().for_each([] (soinfo* si) {
1494    si->CallConstructors();
1495  });
1496
1497  TRACE("\"%s\": calling constructors", name);
1498
1499  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1500  CallFunction("DT_INIT", init_func);
1501  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1502}
1503
1504void soinfo::CallDestructors() {
1505  TRACE("\"%s\": calling destructors", name);
1506
1507  // DT_FINI_ARRAY must be parsed in reverse order.
1508  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1509
1510  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1511  CallFunction("DT_FINI", fini_func);
1512
1513  // This is needed on second call to dlopen
1514  // after library has been unloaded with RTLD_NODELETE
1515  constructors_called = false;
1516}
1517
1518void soinfo::add_child(soinfo* child) {
1519  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1520    return;
1521  }
1522
1523  this->children.push_front(child);
1524  child->parents.push_front(this);
1525}
1526
1527void soinfo::remove_all_links() {
1528  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1529    return;
1530  }
1531
1532  // 1. Untie connected soinfos from 'this'.
1533  children.for_each([&] (soinfo* child) {
1534    child->parents.remove_if([&] (const soinfo* parent) {
1535      return parent == this;
1536    });
1537  });
1538
1539  parents.for_each([&] (soinfo* parent) {
1540    parent->children.for_each([&] (const soinfo* child) {
1541      return child == this;
1542    });
1543  });
1544
1545  // 2. Once everything untied - clear local lists.
1546  parents.clear();
1547  children.clear();
1548}
1549
1550void soinfo::set_st_dev(dev_t dev) {
1551  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1552    return;
1553  }
1554
1555  st_dev = dev;
1556}
1557
1558void soinfo::set_st_ino(ino_t ino) {
1559  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1560    return;
1561  }
1562
1563  st_ino = ino;
1564}
1565
1566dev_t soinfo::get_st_dev() {
1567  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1568    return 0;
1569  }
1570
1571  return st_dev;
1572};
1573
1574ino_t soinfo::get_st_ino() {
1575  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1576    return 0;
1577  }
1578
1579  return st_ino;
1580}
1581
1582// This is a return on get_children() in case
1583// 'this->flags' does not have FLAG_NEW_SOINFO set.
1584static soinfo::soinfo_list_t g_empty_list;
1585
1586soinfo::soinfo_list_t& soinfo::get_children() {
1587  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1588    return g_empty_list;
1589  }
1590
1591  return this->children;
1592}
1593
1594/* Force any of the closed stdin, stdout and stderr to be associated with
1595   /dev/null. */
1596static int nullify_closed_stdio() {
1597    int dev_null, i, status;
1598    int return_value = 0;
1599
1600    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1601    if (dev_null < 0) {
1602        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1603        return -1;
1604    }
1605    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1606
1607    /* If any of the stdio file descriptors is valid and not associated
1608       with /dev/null, dup /dev/null to it.  */
1609    for (i = 0; i < 3; i++) {
1610        /* If it is /dev/null already, we are done. */
1611        if (i == dev_null) {
1612            continue;
1613        }
1614
1615        TRACE("[ Nullifying stdio file descriptor %d]", i);
1616        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1617
1618        /* If file is opened, we are good. */
1619        if (status != -1) {
1620            continue;
1621        }
1622
1623        /* The only error we allow is that the file descriptor does not
1624           exist, in which case we dup /dev/null to it. */
1625        if (errno != EBADF) {
1626            DL_ERR("fcntl failed: %s", strerror(errno));
1627            return_value = -1;
1628            continue;
1629        }
1630
1631        /* Try dupping /dev/null to this stdio file descriptor and
1632           repeat if there is a signal.  Note that any errors in closing
1633           the stdio descriptor are lost.  */
1634        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1635        if (status < 0) {
1636            DL_ERR("dup2 failed: %s", strerror(errno));
1637            return_value = -1;
1638            continue;
1639        }
1640    }
1641
1642    /* If /dev/null is not one of the stdio file descriptors, close it. */
1643    if (dev_null > 2) {
1644        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1645        status = TEMP_FAILURE_RETRY(close(dev_null));
1646        if (status == -1) {
1647            DL_ERR("close failed: %s", strerror(errno));
1648            return_value = -1;
1649        }
1650    }
1651
1652    return return_value;
1653}
1654
1655static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1656    /* "base" might wrap around UINT32_MAX. */
1657    ElfW(Addr) base = si->load_bias;
1658    const ElfW(Phdr)* phdr = si->phdr;
1659    int phnum = si->phnum;
1660    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1661
1662    /* We can't debug anything until the linker is relocated */
1663    if (!relocating_linker) {
1664        INFO("[ linking %s ]", si->name);
1665        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1666    }
1667
1668    /* Extract dynamic section */
1669    size_t dynamic_count;
1670    ElfW(Word) dynamic_flags;
1671    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1672                                   &dynamic_count, &dynamic_flags);
1673    if (si->dynamic == NULL) {
1674        if (!relocating_linker) {
1675            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1676        }
1677        return false;
1678    } else {
1679        if (!relocating_linker) {
1680            DEBUG("dynamic = %p", si->dynamic);
1681        }
1682    }
1683
1684#if defined(__arm__)
1685    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1686                                    &si->ARM_exidx, &si->ARM_exidx_count);
1687#endif
1688
1689    // Extract useful information from dynamic section.
1690    uint32_t needed_count = 0;
1691    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1692        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1693              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1694        switch (d->d_tag) {
1695        case DT_HASH:
1696            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1697            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1698            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1699            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1700            break;
1701        case DT_STRTAB:
1702            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1703            break;
1704        case DT_SYMTAB:
1705            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1706            break;
1707#if !defined(__LP64__)
1708        case DT_PLTREL:
1709            if (d->d_un.d_val != DT_REL) {
1710                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1711                return false;
1712            }
1713            break;
1714#endif
1715        case DT_JMPREL:
1716#if defined(USE_RELA)
1717            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1718#else
1719            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1720#endif
1721            break;
1722        case DT_PLTRELSZ:
1723#if defined(USE_RELA)
1724            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1725#else
1726            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1727#endif
1728            break;
1729#if defined(__mips__)
1730        case DT_PLTGOT:
1731            // Used by mips and mips64.
1732            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1733            break;
1734#endif
1735        case DT_DEBUG:
1736            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1737            // if the dynamic table is writable
1738// FIXME: not working currently for N64
1739// The flags for the LOAD and DYNAMIC program headers do not agree.
1740// The LOAD section containng the dynamic table has been mapped as
1741// read-only, but the DYNAMIC header claims it is writable.
1742#if !(defined(__mips__) && defined(__LP64__))
1743            if ((dynamic_flags & PF_W) != 0) {
1744                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1745            }
1746            break;
1747#endif
1748#if defined(USE_RELA)
1749         case DT_RELA:
1750            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1751            break;
1752         case DT_RELASZ:
1753            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1754            break;
1755        case DT_REL:
1756            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1757            return false;
1758        case DT_RELSZ:
1759            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1760            return false;
1761#else
1762        case DT_REL:
1763            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1764            break;
1765        case DT_RELSZ:
1766            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1767            break;
1768         case DT_RELA:
1769            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1770            return false;
1771#endif
1772        case DT_INIT:
1773            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1774            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1775            break;
1776        case DT_FINI:
1777            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1778            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1779            break;
1780        case DT_INIT_ARRAY:
1781            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1782            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1783            break;
1784        case DT_INIT_ARRAYSZ:
1785            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1786            break;
1787        case DT_FINI_ARRAY:
1788            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1789            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1790            break;
1791        case DT_FINI_ARRAYSZ:
1792            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1793            break;
1794        case DT_PREINIT_ARRAY:
1795            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1796            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1797            break;
1798        case DT_PREINIT_ARRAYSZ:
1799            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1800            break;
1801        case DT_TEXTREL:
1802#if defined(__LP64__)
1803            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1804            return false;
1805#else
1806            si->has_text_relocations = true;
1807            break;
1808#endif
1809        case DT_SYMBOLIC:
1810            si->has_DT_SYMBOLIC = true;
1811            break;
1812        case DT_NEEDED:
1813            ++needed_count;
1814            break;
1815        case DT_FLAGS:
1816            if (d->d_un.d_val & DF_TEXTREL) {
1817#if defined(__LP64__)
1818                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1819                return false;
1820#else
1821                si->has_text_relocations = true;
1822#endif
1823            }
1824            if (d->d_un.d_val & DF_SYMBOLIC) {
1825                si->has_DT_SYMBOLIC = true;
1826            }
1827            break;
1828#if defined(__mips__)
1829        case DT_STRSZ:
1830        case DT_SYMENT:
1831        case DT_RELENT:
1832             break;
1833        case DT_MIPS_RLD_MAP:
1834            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1835            {
1836              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1837              *dp = &_r_debug;
1838            }
1839            break;
1840        case DT_MIPS_RLD_VERSION:
1841        case DT_MIPS_FLAGS:
1842        case DT_MIPS_BASE_ADDRESS:
1843        case DT_MIPS_UNREFEXTNO:
1844            break;
1845
1846        case DT_MIPS_SYMTABNO:
1847            si->mips_symtabno = d->d_un.d_val;
1848            break;
1849
1850        case DT_MIPS_LOCAL_GOTNO:
1851            si->mips_local_gotno = d->d_un.d_val;
1852            break;
1853
1854        case DT_MIPS_GOTSYM:
1855            si->mips_gotsym = d->d_un.d_val;
1856            break;
1857#endif
1858
1859        default:
1860            DEBUG("Unused DT entry: type %p arg %p",
1861                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1862            break;
1863        }
1864    }
1865
1866    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1867          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1868
1869    // Sanity checks.
1870    if (relocating_linker && needed_count != 0) {
1871        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1872        return false;
1873    }
1874    if (si->nbucket == 0) {
1875        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1876        return false;
1877    }
1878    if (si->strtab == 0) {
1879        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1880        return false;
1881    }
1882    if (si->symtab == 0) {
1883        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1884        return false;
1885    }
1886
1887    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1888    if (si->flags & FLAG_EXE) {
1889        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1890        size_t preload_count = 0;
1891        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1892            soinfo* lsi = find_library(g_ld_preload_names[i], 0, NULL);
1893            if (lsi != NULL) {
1894                g_ld_preloads[preload_count++] = lsi;
1895            } else {
1896                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1897                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1898                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1899            }
1900        }
1901    }
1902
1903    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1904    soinfo** pneeded = needed;
1905
1906    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1907        if (d->d_tag == DT_NEEDED) {
1908            const char* library_name = si->strtab + d->d_un.d_val;
1909            DEBUG("%s needs %s", si->name, library_name);
1910            soinfo* lsi = find_library(library_name, 0, NULL);
1911            if (lsi == NULL) {
1912                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1913                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1914                       library_name, si->name, tmp_err_buf);
1915                return false;
1916            }
1917
1918            si->add_child(lsi);
1919            *pneeded++ = lsi;
1920        }
1921    }
1922    *pneeded = NULL;
1923
1924#if !defined(__LP64__)
1925    if (si->has_text_relocations) {
1926        // Make segments writable to allow text relocations to work properly. We will later call
1927        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1928        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1929                "security hardening. Please fix.", si->name);
1930        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1931            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1932                   si->name, strerror(errno));
1933            return false;
1934        }
1935    }
1936#endif
1937
1938#if defined(USE_RELA)
1939    if (si->plt_rela != NULL) {
1940        DEBUG("[ relocating %s plt ]\n", si->name);
1941        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1942            return false;
1943        }
1944    }
1945    if (si->rela != NULL) {
1946        DEBUG("[ relocating %s ]\n", si->name);
1947        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1948            return false;
1949        }
1950    }
1951#else
1952    if (si->plt_rel != NULL) {
1953        DEBUG("[ relocating %s plt ]", si->name);
1954        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1955            return false;
1956        }
1957    }
1958    if (si->rel != NULL) {
1959        DEBUG("[ relocating %s ]", si->name);
1960        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1961            return false;
1962        }
1963    }
1964#endif
1965
1966#if defined(__mips__)
1967    if (!mips_relocate_got(si, needed)) {
1968        return false;
1969    }
1970#endif
1971
1972    si->flags |= FLAG_LINKED;
1973    DEBUG("[ finished linking %s ]", si->name);
1974
1975#if !defined(__LP64__)
1976    if (si->has_text_relocations) {
1977        // All relocations are done, we can protect our segments back to read-only.
1978        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1979            DL_ERR("can't protect segments for \"%s\": %s",
1980                   si->name, strerror(errno));
1981            return false;
1982        }
1983    }
1984#endif
1985
1986    /* We can also turn on GNU RELRO protection */
1987    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1988        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1989               si->name, strerror(errno));
1990        return false;
1991    }
1992
1993    /* Handle serializing/sharing the RELRO segment */
1994    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1995      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
1996                                         extinfo->relro_fd) < 0) {
1997        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
1998               si->name, strerror(errno));
1999        return false;
2000      }
2001    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
2002      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
2003                                   extinfo->relro_fd) < 0) {
2004        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
2005               si->name, strerror(errno));
2006        return false;
2007      }
2008    }
2009
2010    notify_gdb_of_load(si);
2011    return true;
2012}
2013
2014/*
2015 * This function add vdso to internal dso list.
2016 * It helps to stack unwinding through signal handlers.
2017 * Also, it makes bionic more like glibc.
2018 */
2019static void add_vdso(KernelArgumentBlock& args __unused) {
2020#if defined(AT_SYSINFO_EHDR)
2021  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
2022  if (ehdr_vdso == NULL) {
2023    return;
2024  }
2025
2026  soinfo* si = soinfo_alloc("[vdso]", NULL);
2027
2028  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
2029  si->phnum = ehdr_vdso->e_phnum;
2030  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
2031  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2032  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
2033
2034  soinfo_link_image(si, NULL);
2035#endif
2036}
2037
2038/*
2039 * This is linker soinfo for GDB. See details below.
2040 */
2041static soinfo linker_soinfo_for_gdb;
2042
2043/* gdb expects the linker to be in the debug shared object list.
2044 * Without this, gdb has trouble locating the linker's ".text"
2045 * and ".plt" sections. Gdb could also potentially use this to
2046 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2047 * Don't use soinfo_alloc(), because the linker shouldn't
2048 * be on the soinfo list.
2049 */
2050static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
2051#if defined(__LP64__)
2052  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
2053#else
2054  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
2055#endif
2056  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
2057  linker_soinfo_for_gdb.base = linker_base;
2058
2059  /*
2060   * Set the dynamic field in the link map otherwise gdb will complain with
2061   * the following:
2062   *   warning: .dynamic section for "/system/bin/linker" is not at the
2063   *   expected address (wrong library or version mismatch?)
2064   */
2065  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2066  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2067  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2068                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
2069  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2070}
2071
2072/*
2073 * This code is called after the linker has linked itself and
2074 * fixed it's own GOT. It is safe to make references to externs
2075 * and other non-local data at this point.
2076 */
2077static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2078    /* NOTE: we store the args pointer on a special location
2079     *       of the temporary TLS area in order to pass it to
2080     *       the C Library's runtime initializer.
2081     *
2082     *       The initializer must clear the slot and reset the TLS
2083     *       to point to a different location to ensure that no other
2084     *       shared library constructor can access it.
2085     */
2086  __libc_init_tls(args);
2087
2088#if TIMING
2089    struct timeval t0, t1;
2090    gettimeofday(&t0, 0);
2091#endif
2092
2093    // Initialize environment functions, and get to the ELF aux vectors table.
2094    linker_env_init(args);
2095
2096    // If this is a setuid/setgid program, close the security hole described in
2097    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2098    if (get_AT_SECURE()) {
2099        nullify_closed_stdio();
2100    }
2101
2102    debuggerd_init();
2103
2104    // Get a few environment variables.
2105    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2106    if (LD_DEBUG != NULL) {
2107      g_ld_debug_verbosity = atoi(LD_DEBUG);
2108    }
2109
2110    // Normally, these are cleaned by linker_env_init, but the test
2111    // doesn't cost us anything.
2112    const char* ldpath_env = NULL;
2113    const char* ldpreload_env = NULL;
2114    if (!get_AT_SECURE()) {
2115      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2116      ldpreload_env = linker_env_get("LD_PRELOAD");
2117    }
2118
2119    INFO("[ android linker & debugger ]");
2120
2121    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2122    if (si == NULL) {
2123        exit(EXIT_FAILURE);
2124    }
2125
2126    /* bootstrap the link map, the main exe always needs to be first */
2127    si->flags |= FLAG_EXE;
2128    link_map* map = &(si->link_map_head);
2129
2130    map->l_addr = 0;
2131    map->l_name = args.argv[0];
2132    map->l_prev = NULL;
2133    map->l_next = NULL;
2134
2135    _r_debug.r_map = map;
2136    r_debug_tail = map;
2137
2138    init_linker_info_for_gdb(linker_base);
2139
2140    // Extract information passed from the kernel.
2141    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2142    si->phnum = args.getauxval(AT_PHNUM);
2143    si->entry = args.getauxval(AT_ENTRY);
2144
2145    /* Compute the value of si->base. We can't rely on the fact that
2146     * the first entry is the PHDR because this will not be true
2147     * for certain executables (e.g. some in the NDK unit test suite)
2148     */
2149    si->base = 0;
2150    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2151    si->load_bias = 0;
2152    for (size_t i = 0; i < si->phnum; ++i) {
2153      if (si->phdr[i].p_type == PT_PHDR) {
2154        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2155        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2156        break;
2157      }
2158    }
2159    si->dynamic = NULL;
2160    si->ref_count = 1;
2161
2162    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2163    if (elf_hdr->e_type != ET_DYN) {
2164        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2165        exit(EXIT_FAILURE);
2166    }
2167
2168    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2169    parse_LD_LIBRARY_PATH(ldpath_env);
2170    parse_LD_PRELOAD(ldpreload_env);
2171
2172    somain = si;
2173
2174    if (!soinfo_link_image(si, NULL)) {
2175        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2176        exit(EXIT_FAILURE);
2177    }
2178
2179    add_vdso(args);
2180
2181    si->CallPreInitConstructors();
2182
2183    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2184        g_ld_preloads[i]->CallConstructors();
2185    }
2186
2187    /* After the link_image, the si->load_bias is initialized.
2188     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2189     * We need to update this value for so exe here. So Unwind_Backtrace
2190     * for some arch like x86 could work correctly within so exe.
2191     */
2192    map->l_addr = si->load_bias;
2193    si->CallConstructors();
2194
2195#if TIMING
2196    gettimeofday(&t1, NULL);
2197    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2198               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2199               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2200#endif
2201#if STATS
2202    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2203           linker_stats.count[kRelocAbsolute],
2204           linker_stats.count[kRelocRelative],
2205           linker_stats.count[kRelocCopy],
2206           linker_stats.count[kRelocSymbol]);
2207#endif
2208#if COUNT_PAGES
2209    {
2210        unsigned n;
2211        unsigned i;
2212        unsigned count = 0;
2213        for (n = 0; n < 4096; n++) {
2214            if (bitmask[n]) {
2215                unsigned x = bitmask[n];
2216#if defined(__LP64__)
2217                for (i = 0; i < 32; i++) {
2218#else
2219                for (i = 0; i < 8; i++) {
2220#endif
2221                    if (x & 1) {
2222                        count++;
2223                    }
2224                    x >>= 1;
2225                }
2226            }
2227        }
2228        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2229    }
2230#endif
2231
2232#if TIMING || STATS || COUNT_PAGES
2233    fflush(stdout);
2234#endif
2235
2236    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2237    return si->entry;
2238}
2239
2240/* Compute the load-bias of an existing executable. This shall only
2241 * be used to compute the load bias of an executable or shared library
2242 * that was loaded by the kernel itself.
2243 *
2244 * Input:
2245 *    elf    -> address of ELF header, assumed to be at the start of the file.
2246 * Return:
2247 *    load bias, i.e. add the value of any p_vaddr in the file to get
2248 *    the corresponding address in memory.
2249 */
2250static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2251  ElfW(Addr) offset = elf->e_phoff;
2252  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2253  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2254
2255  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2256    if (phdr->p_type == PT_LOAD) {
2257      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2258    }
2259  }
2260  return 0;
2261}
2262
2263extern "C" void _start();
2264
2265/*
2266 * This is the entry point for the linker, called from begin.S. This
2267 * method is responsible for fixing the linker's own relocations, and
2268 * then calling __linker_init_post_relocation().
2269 *
2270 * Because this method is called before the linker has fixed it's own
2271 * relocations, any attempt to reference an extern variable, extern
2272 * function, or other GOT reference will generate a segfault.
2273 */
2274extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2275  // Initialize static variables.
2276  solist = get_libdl_info();
2277  sonext = get_libdl_info();
2278
2279  KernelArgumentBlock args(raw_args);
2280
2281  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2282  ElfW(Addr) entry_point = args.getauxval(AT_ENTRY);
2283  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2284  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2285
2286  soinfo linker_so;
2287  memset(&linker_so, 0, sizeof(soinfo));
2288
2289  // If the linker is not acting as PT_INTERP entry_point is equal to
2290  // _start. Which means that the linker is running as an executable and
2291  // already linked by PT_INTERP.
2292  //
2293  // This happens when user tries to run 'adb shell /system/bin/linker'
2294  // see also https://code.google.com/p/android/issues/detail?id=63174
2295  if (reinterpret_cast<ElfW(Addr)>(&_start) == entry_point) {
2296    __libc_fatal("This is %s, the helper program for shared library executables.\n", args.argv[0]);
2297  }
2298
2299  strcpy(linker_so.name, "[dynamic linker]");
2300  linker_so.base = linker_addr;
2301  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2302  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2303  linker_so.dynamic = NULL;
2304  linker_so.phdr = phdr;
2305  linker_so.phnum = elf_hdr->e_phnum;
2306  linker_so.flags |= FLAG_LINKER;
2307
2308  if (!soinfo_link_image(&linker_so, NULL)) {
2309    // It would be nice to print an error message, but if the linker
2310    // can't link itself, there's no guarantee that we'll be able to
2311    // call write() (because it involves a GOT reference). We may as
2312    // well try though...
2313    const char* msg = "CANNOT LINK EXECUTABLE: ";
2314    write(2, msg, strlen(msg));
2315    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2316    write(2, "\n", 1);
2317    _exit(EXIT_FAILURE);
2318  }
2319
2320  // Initialize the linker's own global variables
2321  linker_so.CallConstructors();
2322
2323  // We have successfully fixed our own relocations. It's safe to run
2324  // the main part of the linker now.
2325  args.abort_message_ptr = &g_abort_message;
2326  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2327
2328  protect_data(PROT_READ);
2329
2330  // Return the address that the calling assembly stub should jump to.
2331  return start_address;
2332}
2333