linker.cpp revision d59e50063ad708509f3ad83350be33f5612c4f54
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/atomics.h>
38#include <sys/mman.h>
39#include <sys/stat.h>
40#include <unistd.h>
41
42// Private C library headers.
43#include "private/bionic_tls.h"
44#include "private/KernelArgumentBlock.h"
45#include "private/ScopedPthreadMutexLocker.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
68static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
69static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
70
71static LinkerAllocator<soinfo> g_soinfo_allocator;
72static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
73
74static soinfo* solist;
75static soinfo* sonext;
76static soinfo* somain; /* main process, always the one after libdl_info */
77
78static const char* const kDefaultLdPaths[] = {
79#if defined(__LP64__)
80  "/vendor/lib64",
81  "/system/lib64",
82#else
83  "/vendor/lib",
84  "/system/lib",
85#endif
86  NULL
87};
88
89#define LDPATH_BUFSIZE (LDPATH_MAX*64)
90#define LDPATH_MAX 8
91
92#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
93#define LDPRELOAD_MAX 8
94
95static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
96static const char* g_ld_library_paths[LDPATH_MAX + 1];
97
98static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
99static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
100
101static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
102
103__LIBC_HIDDEN__ int g_ld_debug_verbosity;
104
105__LIBC_HIDDEN__ abort_msg_t* g_abort_message = NULL; // For debuggerd.
106
107enum RelocationKind {
108    kRelocAbsolute = 0,
109    kRelocRelative,
110    kRelocCopy,
111    kRelocSymbol,
112    kRelocMax
113};
114
115#if STATS
116struct linker_stats_t {
117    int count[kRelocMax];
118};
119
120static linker_stats_t linker_stats;
121
122static void count_relocation(RelocationKind kind) {
123    ++linker_stats.count[kind];
124}
125#else
126static void count_relocation(RelocationKind) {
127}
128#endif
129
130#if COUNT_PAGES
131static unsigned bitmask[4096];
132#if defined(__LP64__)
133#define MARK(offset) \
134    do { \
135        if ((((offset) >> 12) >> 5) < 4096) \
136            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
137    } while (0)
138#else
139#define MARK(offset) \
140    do { \
141        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
142    } while (0)
143#endif
144#else
145#define MARK(x) do {} while (0)
146#endif
147
148// You shouldn't try to call memory-allocating functions in the dynamic linker.
149// Guard against the most obvious ones.
150#define DISALLOW_ALLOCATION(return_type, name, ...) \
151    return_type name __VA_ARGS__ \
152    { \
153        const char* msg = "ERROR: " #name " called from the dynamic linker!\n"; \
154        __libc_format_log(ANDROID_LOG_FATAL, "linker", "%s", msg); \
155        write(2, msg, strlen(msg)); \
156        abort(); \
157    }
158DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
159DISALLOW_ALLOCATION(void, free, (void* u __unused));
160DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
161DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
162
163static char tmp_err_buf[768];
164static char __linker_dl_err_buf[768];
165
166char* linker_get_error_buffer() {
167  return &__linker_dl_err_buf[0];
168}
169
170size_t linker_get_error_buffer_size() {
171  return sizeof(__linker_dl_err_buf);
172}
173
174/*
175 * This function is an empty stub where GDB locates a breakpoint to get notified
176 * about linker activity.
177 */
178extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
179
180static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
181static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
182static link_map* r_debug_tail = 0;
183
184static void insert_soinfo_into_debug_map(soinfo* info) {
185    // Copy the necessary fields into the debug structure.
186    link_map* map = &(info->link_map_head);
187    map->l_addr = info->load_bias;
188    map->l_name = reinterpret_cast<char*>(info->name);
189    map->l_ld = info->dynamic;
190
191    /* Stick the new library at the end of the list.
192     * gdb tends to care more about libc than it does
193     * about leaf libraries, and ordering it this way
194     * reduces the back-and-forth over the wire.
195     */
196    if (r_debug_tail) {
197        r_debug_tail->l_next = map;
198        map->l_prev = r_debug_tail;
199        map->l_next = 0;
200    } else {
201        _r_debug.r_map = map;
202        map->l_prev = 0;
203        map->l_next = 0;
204    }
205    r_debug_tail = map;
206}
207
208static void remove_soinfo_from_debug_map(soinfo* info) {
209    link_map* map = &(info->link_map_head);
210
211    if (r_debug_tail == map) {
212        r_debug_tail = map->l_prev;
213    }
214
215    if (map->l_prev) {
216        map->l_prev->l_next = map->l_next;
217    }
218    if (map->l_next) {
219        map->l_next->l_prev = map->l_prev;
220    }
221}
222
223static void notify_gdb_of_load(soinfo* info) {
224    if (info->flags & FLAG_EXE) {
225        // GDB already knows about the main executable
226        return;
227    }
228
229    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
230
231    _r_debug.r_state = r_debug::RT_ADD;
232    rtld_db_dlactivity();
233
234    insert_soinfo_into_debug_map(info);
235
236    _r_debug.r_state = r_debug::RT_CONSISTENT;
237    rtld_db_dlactivity();
238}
239
240static void notify_gdb_of_unload(soinfo* info) {
241    if (info->flags & FLAG_EXE) {
242        // GDB already knows about the main executable
243        return;
244    }
245
246    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
247
248    _r_debug.r_state = r_debug::RT_DELETE;
249    rtld_db_dlactivity();
250
251    remove_soinfo_from_debug_map(info);
252
253    _r_debug.r_state = r_debug::RT_CONSISTENT;
254    rtld_db_dlactivity();
255}
256
257void notify_gdb_of_libraries() {
258  _r_debug.r_state = r_debug::RT_ADD;
259  rtld_db_dlactivity();
260  _r_debug.r_state = r_debug::RT_CONSISTENT;
261  rtld_db_dlactivity();
262}
263
264LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
265  return g_soinfo_links_allocator.alloc();
266}
267
268void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
269  g_soinfo_links_allocator.free(entry);
270}
271
272static void protect_data(int protection) {
273  g_soinfo_allocator.protect_all(protection);
274  g_soinfo_links_allocator.protect_all(protection);
275}
276
277static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
278  if (strlen(name) >= SOINFO_NAME_LEN) {
279    DL_ERR("library name \"%s\" too long", name);
280    return NULL;
281  }
282
283  soinfo* si = g_soinfo_allocator.alloc();
284
285  // Initialize the new element.
286  memset(si, 0, sizeof(soinfo));
287  strlcpy(si->name, name, sizeof(si->name));
288  si->flags = FLAG_NEW_SOINFO;
289
290  if (file_stat != NULL) {
291    si->set_st_dev(file_stat->st_dev);
292    si->set_st_ino(file_stat->st_ino);
293  }
294
295  sonext->next = si;
296  sonext = si;
297
298  TRACE("name %s: allocated soinfo @ %p", name, si);
299  return si;
300}
301
302static void soinfo_free(soinfo* si) {
303    if (si == NULL) {
304        return;
305    }
306
307    if (si->base != 0 && si->size != 0) {
308      munmap(reinterpret_cast<void*>(si->base), si->size);
309    }
310
311    soinfo *prev = NULL, *trav;
312
313    TRACE("name %s: freeing soinfo @ %p", si->name, si);
314
315    for (trav = solist; trav != NULL; trav = trav->next) {
316        if (trav == si)
317            break;
318        prev = trav;
319    }
320    if (trav == NULL) {
321        /* si was not in solist */
322        DL_ERR("name \"%s\" is not in solist!", si->name);
323        return;
324    }
325
326    // clear links to/from si
327    si->remove_all_links();
328
329    /* prev will never be NULL, because the first entry in solist is
330       always the static libdl_info.
331    */
332    prev->next = si->next;
333    if (si == sonext) {
334        sonext = prev;
335    }
336
337    g_soinfo_allocator.free(si);
338}
339
340
341static void parse_path(const char* path, const char* delimiters,
342                       const char** array, char* buf, size_t buf_size, size_t max_count) {
343  if (path == NULL) {
344    return;
345  }
346
347  size_t len = strlcpy(buf, path, buf_size);
348
349  size_t i = 0;
350  char* buf_p = buf;
351  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
352    if (*array[i] != '\0') {
353      ++i;
354    }
355  }
356
357  // Forget the last path if we had to truncate; this occurs if the 2nd to
358  // last char isn't '\0' (i.e. wasn't originally a delimiter).
359  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
360    array[i - 1] = NULL;
361  } else {
362    array[i] = NULL;
363  }
364}
365
366static void parse_LD_LIBRARY_PATH(const char* path) {
367  parse_path(path, ":", g_ld_library_paths,
368             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
369}
370
371static void parse_LD_PRELOAD(const char* path) {
372  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
373  parse_path(path, " :", g_ld_preload_names,
374             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
375}
376
377#if defined(__arm__)
378
379/* For a given PC, find the .so that it belongs to.
380 * Returns the base address of the .ARM.exidx section
381 * for that .so, and the number of 8-byte entries
382 * in that section (via *pcount).
383 *
384 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
385 *
386 * This function is exposed via dlfcn.cpp and libdl.so.
387 */
388_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
389    unsigned addr = (unsigned)pc;
390
391    for (soinfo* si = solist; si != 0; si = si->next) {
392        if ((addr >= si->base) && (addr < (si->base + si->size))) {
393            *pcount = si->ARM_exidx_count;
394            return (_Unwind_Ptr)si->ARM_exidx;
395        }
396    }
397    *pcount = 0;
398    return NULL;
399}
400
401#endif
402
403/* Here, we only have to provide a callback to iterate across all the
404 * loaded libraries. gcc_eh does the rest. */
405int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
406    int rv = 0;
407    for (soinfo* si = solist; si != NULL; si = si->next) {
408        dl_phdr_info dl_info;
409        dl_info.dlpi_addr = si->link_map_head.l_addr;
410        dl_info.dlpi_name = si->link_map_head.l_name;
411        dl_info.dlpi_phdr = si->phdr;
412        dl_info.dlpi_phnum = si->phnum;
413        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
414        if (rv != 0) {
415            break;
416        }
417    }
418    return rv;
419}
420
421static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
422  ElfW(Sym)* symtab = si->symtab;
423  const char* strtab = si->strtab;
424
425  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
426             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
427
428  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
429    ElfW(Sym)* s = symtab + n;
430    if (strcmp(strtab + s->st_name, name)) continue;
431
432    /* only concern ourselves with global and weak symbol definitions */
433    switch (ELF_ST_BIND(s->st_info)) {
434      case STB_GLOBAL:
435      case STB_WEAK:
436        if (s->st_shndx == SHN_UNDEF) {
437        continue;
438      }
439
440      TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
441                 name, si->name, reinterpret_cast<void*>(s->st_value),
442                 static_cast<size_t>(s->st_size));
443      return s;
444    }
445  }
446
447  return NULL;
448}
449
450static unsigned elfhash(const char* _name) {
451    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
452    unsigned h = 0, g;
453
454    while (*name) {
455        h = (h << 4) + *name++;
456        g = h & 0xf0000000;
457        h ^= g;
458        h ^= g >> 24;
459    }
460    return h;
461}
462
463static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
464    unsigned elf_hash = elfhash(name);
465    ElfW(Sym)* s = NULL;
466
467    if (si != NULL && somain != NULL) {
468        /*
469         * Local scope is executable scope. Just start looking into it right away
470         * for the shortcut.
471         */
472
473        if (si == somain) {
474            s = soinfo_elf_lookup(si, elf_hash, name);
475            if (s != NULL) {
476                *lsi = si;
477                goto done;
478            }
479        } else {
480            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
481
482            /*
483             * If this object was built with symbolic relocations disabled, the
484             * first place to look to resolve external references is the main
485             * executable.
486             */
487
488            if (!si->has_DT_SYMBOLIC) {
489                DEBUG("%s: looking up %s in executable %s",
490                      si->name, name, somain->name);
491                s = soinfo_elf_lookup(somain, elf_hash, name);
492                if (s != NULL) {
493                    *lsi = somain;
494                    goto done;
495                }
496            }
497
498            /* Look for symbols in the local scope (the object who is
499             * searching). This happens with C++ templates on x86 for some
500             * reason.
501             *
502             * Notes on weak symbols:
503             * The ELF specs are ambiguous about treatment of weak definitions in
504             * dynamic linking.  Some systems return the first definition found
505             * and some the first non-weak definition.   This is system dependent.
506             * Here we return the first definition found for simplicity.  */
507
508            s = soinfo_elf_lookup(si, elf_hash, name);
509            if (s != NULL) {
510                *lsi = si;
511                goto done;
512            }
513
514            /*
515             * If this object was built with -Bsymbolic and symbol is not found
516             * in the local scope, try to find the symbol in the main executable.
517             */
518
519            if (si->has_DT_SYMBOLIC) {
520                DEBUG("%s: looking up %s in executable %s after local scope",
521                      si->name, name, somain->name);
522                s = soinfo_elf_lookup(somain, elf_hash, name);
523                if (s != NULL) {
524                    *lsi = somain;
525                    goto done;
526                }
527            }
528        }
529    }
530
531    /* Next, look for it in the preloads list */
532    for (int i = 0; g_ld_preloads[i] != NULL; i++) {
533        s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name);
534        if (s != NULL) {
535            *lsi = g_ld_preloads[i];
536            goto done;
537        }
538    }
539
540    for (int i = 0; needed[i] != NULL; i++) {
541        DEBUG("%s: looking up %s in %s",
542              si->name, name, needed[i]->name);
543        s = soinfo_elf_lookup(needed[i], elf_hash, name);
544        if (s != NULL) {
545            *lsi = needed[i];
546            goto done;
547        }
548    }
549
550done:
551    if (s != NULL) {
552        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
553                   "found in %s, base = %p, load bias = %p",
554                   si->name, name, reinterpret_cast<void*>(s->st_value),
555                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
556                   reinterpret_cast<void*>((*lsi)->load_bias));
557        return s;
558    }
559
560    return NULL;
561}
562
563/* This is used by dlsym(3).  It performs symbol lookup only within the
564   specified soinfo object and not in any of its dependencies.
565
566   TODO: Only looking in the specified soinfo seems wrong. dlsym(3) says
567   that it should do a breadth first search through the dependency
568   tree. This agrees with the ELF spec (aka System V Application
569   Binary Interface) where in Chapter 5 it discuss resolving "Shared
570   Object Dependencies" in breadth first search order.
571 */
572ElfW(Sym)* dlsym_handle_lookup(soinfo* si, const char* name) {
573    return soinfo_elf_lookup(si, elfhash(name), name);
574}
575
576/* This is used by dlsym(3) to performs a global symbol lookup. If the
577   start value is null (for RTLD_DEFAULT), the search starts at the
578   beginning of the global solist. Otherwise the search starts at the
579   specified soinfo (for RTLD_NEXT).
580 */
581ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
582  unsigned elf_hash = elfhash(name);
583
584  if (start == NULL) {
585    start = solist;
586  }
587
588  ElfW(Sym)* s = NULL;
589  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
590    s = soinfo_elf_lookup(si, elf_hash, name);
591    if (s != NULL) {
592      *found = si;
593      break;
594    }
595  }
596
597  if (s != NULL) {
598    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
599               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
600  }
601
602  return s;
603}
604
605soinfo* find_containing_library(const void* p) {
606  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
607  for (soinfo* si = solist; si != NULL; si = si->next) {
608    if (address >= si->base && address - si->base < si->size) {
609      return si;
610    }
611  }
612  return NULL;
613}
614
615ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
616  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
617
618  // Search the library's symbol table for any defined symbol which
619  // contains this address.
620  for (size_t i = 0; i < si->nchain; ++i) {
621    ElfW(Sym)* sym = &si->symtab[i];
622    if (sym->st_shndx != SHN_UNDEF &&
623        soaddr >= sym->st_value &&
624        soaddr < sym->st_value + sym->st_size) {
625      return sym;
626    }
627  }
628
629  return NULL;
630}
631
632static int open_library_on_path(const char* name, const char* const paths[]) {
633  char buf[512];
634  for (size_t i = 0; paths[i] != NULL; ++i) {
635    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
636    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
637      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
638      continue;
639    }
640    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
641    if (fd != -1) {
642      return fd;
643    }
644  }
645  return -1;
646}
647
648static int open_library(const char* name) {
649  TRACE("[ opening %s ]", name);
650
651  // If the name contains a slash, we should attempt to open it directly and not search the paths.
652  if (strchr(name, '/') != NULL) {
653    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
654    if (fd != -1) {
655      return fd;
656    }
657    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
658#if defined(__LP64__)
659    return -1;
660#endif
661  }
662
663  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
664  int fd = open_library_on_path(name, g_ld_library_paths);
665  if (fd == -1) {
666    fd = open_library_on_path(name, kDefaultLdPaths);
667  }
668  return fd;
669}
670
671static soinfo* load_library(const char* name, const android_dlextinfo* extinfo) {
672    // Open the file.
673    int fd = open_library(name);
674    if (fd == -1) {
675        DL_ERR("library \"%s\" not found", name);
676        return NULL;
677    }
678
679    ElfReader elf_reader(name, fd);
680
681    struct stat file_stat;
682    if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
683      DL_ERR("unable to stat file for the library %s: %s", name, strerror(errno));
684      return NULL;
685    }
686
687    // Check for symlink and other situations where
688    // file can have different names.
689    for (soinfo* si = solist; si != NULL; si = si->next) {
690      if (si->get_st_dev() != 0 &&
691          si->get_st_ino() != 0 &&
692          si->get_st_dev() == file_stat.st_dev &&
693          si->get_st_ino() == file_stat.st_ino) {
694        TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
695        return si;
696      }
697    }
698
699    // Read the ELF header and load the segments.
700    if (!elf_reader.Load(extinfo)) {
701        return NULL;
702    }
703
704    const char* bname = strrchr(name, '/');
705    soinfo* si = soinfo_alloc(bname ? bname + 1 : name, &file_stat);
706    if (si == NULL) {
707        return NULL;
708    }
709    si->base = elf_reader.load_start();
710    si->size = elf_reader.load_size();
711    si->load_bias = elf_reader.load_bias();
712    si->phnum = elf_reader.phdr_count();
713    si->phdr = elf_reader.loaded_phdr();
714
715    // At this point we know that whatever is loaded @ base is a valid ELF
716    // shared library whose segments are properly mapped in.
717    TRACE("[ find_library_internal base=%p size=%zu name='%s' ]",
718          reinterpret_cast<void*>(si->base), si->size, si->name);
719
720    if (!soinfo_link_image(si, extinfo)) {
721      soinfo_free(si);
722      return NULL;
723    }
724
725    return si;
726}
727
728static soinfo *find_loaded_library(const char* name) {
729    // TODO: don't use basename only for determining libraries
730    // http://code.google.com/p/android/issues/detail?id=6670
731
732    const char* bname = strrchr(name, '/');
733    bname = bname ? bname + 1 : name;
734
735    for (soinfo* si = solist; si != NULL; si = si->next) {
736        if (!strcmp(bname, si->name)) {
737            return si;
738        }
739    }
740    return NULL;
741}
742
743static soinfo* find_library_internal(const char* name, const android_dlextinfo* extinfo) {
744  if (name == NULL) {
745    return somain;
746  }
747
748  soinfo* si = find_loaded_library(name);
749  if (si != NULL) {
750    if (si->flags & FLAG_LINKED) {
751      return si;
752    }
753    DL_ERR("OOPS: recursive link to \"%s\"", si->name);
754    return NULL;
755  }
756
757  TRACE("[ '%s' has not been loaded yet.  Locating...]", name);
758  return load_library(name, extinfo);
759}
760
761static soinfo* find_library(const char* name, const android_dlextinfo* extinfo) {
762  soinfo* si = find_library_internal(name, extinfo);
763  if (si != NULL) {
764    si->ref_count++;
765  }
766  return si;
767}
768
769static int soinfo_unload(soinfo* si) {
770  if (si->ref_count == 1) {
771    TRACE("unloading '%s'", si->name);
772    si->CallDestructors();
773
774    if ((si->flags | FLAG_NEW_SOINFO) != 0) {
775      si->get_children().for_each([&] (soinfo* child) {
776        TRACE("%s needs to unload %s", si->name, child->name);
777        soinfo_unload(child);
778      });
779    } else {
780      for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
781        if (d->d_tag == DT_NEEDED) {
782          const char* library_name = si->strtab + d->d_un.d_val;
783          TRACE("%s needs to unload %s", si->name, library_name);
784          soinfo_unload(find_loaded_library(library_name));
785        }
786      }
787    }
788
789    notify_gdb_of_unload(si);
790    si->ref_count = 0;
791    soinfo_free(si);
792  } else {
793    si->ref_count--;
794    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
795  }
796  return 0;
797}
798
799void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
800  snprintf(buffer, buffer_size, "%s:%s", kDefaultLdPaths[0], kDefaultLdPaths[1]);
801}
802
803void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
804  if (!get_AT_SECURE()) {
805    parse_LD_LIBRARY_PATH(ld_library_path);
806  }
807}
808
809soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
810  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL)) != 0) {
811    DL_ERR("invalid flags to dlopen: %x", flags);
812    return NULL;
813  }
814  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
815    DL_ERR("invalid extended flags to android_dlopen_ext: %x", extinfo->flags);
816    return NULL;
817  }
818  protect_data(PROT_READ | PROT_WRITE);
819  soinfo* si = find_library(name, extinfo);
820  if (si != NULL) {
821    si->CallConstructors();
822    somain->add_child(si);
823  }
824  protect_data(PROT_READ);
825  return si;
826}
827
828int do_dlclose(soinfo* si) {
829  protect_data(PROT_READ | PROT_WRITE);
830  int result = soinfo_unload(si);
831  protect_data(PROT_READ);
832  return result;
833}
834
835#if defined(USE_RELA)
836static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
837  ElfW(Sym)* s;
838  soinfo* lsi;
839
840  for (size_t idx = 0; idx < count; ++idx, ++rela) {
841    unsigned type = ELFW(R_TYPE)(rela->r_info);
842    unsigned sym = ELFW(R_SYM)(rela->r_info);
843    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
844    ElfW(Addr) sym_addr = 0;
845    const char* sym_name = NULL;
846
847    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
848    if (type == 0) { // R_*_NONE
849      continue;
850    }
851    if (sym != 0) {
852      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
853      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
854      if (s == NULL) {
855        // We only allow an undefined symbol if this is a weak reference...
856        s = &si->symtab[sym];
857        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
858          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
859          return -1;
860        }
861
862        /* IHI0044C AAELF 4.5.1.1:
863
864           Libraries are not searched to resolve weak references.
865           It is not an error for a weak reference to remain unsatisfied.
866
867           During linking, the value of an undefined weak reference is:
868           - Zero if the relocation type is absolute
869           - The address of the place if the relocation is pc-relative
870           - The address of nominal base address if the relocation
871             type is base-relative.
872         */
873
874        switch (type) {
875#if defined(__aarch64__)
876        case R_AARCH64_JUMP_SLOT:
877        case R_AARCH64_GLOB_DAT:
878        case R_AARCH64_ABS64:
879        case R_AARCH64_ABS32:
880        case R_AARCH64_ABS16:
881        case R_AARCH64_RELATIVE:
882          /*
883           * The sym_addr was initialized to be zero above, or the relocation
884           * code below does not care about value of sym_addr.
885           * No need to do anything.
886           */
887          break;
888#elif defined(__x86_64__)
889        case R_X86_64_JUMP_SLOT:
890        case R_X86_64_GLOB_DAT:
891        case R_X86_64_32:
892        case R_X86_64_RELATIVE:
893          // No need to do anything.
894          break;
895        case R_X86_64_PC32:
896          sym_addr = reloc;
897          break;
898#endif
899        default:
900          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
901          return -1;
902        }
903      } else {
904        // We got a definition.
905        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
906      }
907      count_relocation(kRelocSymbol);
908    } else {
909      s = NULL;
910    }
911
912    switch (type) {
913#if defined(__aarch64__)
914    case R_AARCH64_JUMP_SLOT:
915        count_relocation(kRelocAbsolute);
916        MARK(rela->r_offset);
917        TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
918                   reloc, (sym_addr + rela->r_addend), sym_name);
919        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
920        break;
921    case R_AARCH64_GLOB_DAT:
922        count_relocation(kRelocAbsolute);
923        MARK(rela->r_offset);
924        TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
925                   reloc, (sym_addr + rela->r_addend), sym_name);
926        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
927        break;
928    case R_AARCH64_ABS64:
929        count_relocation(kRelocAbsolute);
930        MARK(rela->r_offset);
931        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
932                   reloc, (sym_addr + rela->r_addend), sym_name);
933        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
934        break;
935    case R_AARCH64_ABS32:
936        count_relocation(kRelocAbsolute);
937        MARK(rela->r_offset);
938        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
939                   reloc, (sym_addr + rela->r_addend), sym_name);
940        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
941            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
942            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
943        } else {
944            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
945                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
946                   static_cast<ElfW(Addr)>(INT32_MIN),
947                   static_cast<ElfW(Addr)>(UINT32_MAX));
948            return -1;
949        }
950        break;
951    case R_AARCH64_ABS16:
952        count_relocation(kRelocAbsolute);
953        MARK(rela->r_offset);
954        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
955                   reloc, (sym_addr + rela->r_addend), sym_name);
956        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
957            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
958            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
959        } else {
960            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
961                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
962                   static_cast<ElfW(Addr)>(INT16_MIN),
963                   static_cast<ElfW(Addr)>(UINT16_MAX));
964            return -1;
965        }
966        break;
967    case R_AARCH64_PREL64:
968        count_relocation(kRelocRelative);
969        MARK(rela->r_offset);
970        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
971                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
972        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
973        break;
974    case R_AARCH64_PREL32:
975        count_relocation(kRelocRelative);
976        MARK(rela->r_offset);
977        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
978                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
979        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
980            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
981            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
982        } else {
983            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
984                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
985                   static_cast<ElfW(Addr)>(INT32_MIN),
986                   static_cast<ElfW(Addr)>(UINT32_MAX));
987            return -1;
988        }
989        break;
990    case R_AARCH64_PREL16:
991        count_relocation(kRelocRelative);
992        MARK(rela->r_offset);
993        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
994                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
995        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
996            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
997            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
998        } else {
999            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1000                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1001                   static_cast<ElfW(Addr)>(INT16_MIN),
1002                   static_cast<ElfW(Addr)>(UINT16_MAX));
1003            return -1;
1004        }
1005        break;
1006
1007    case R_AARCH64_RELATIVE:
1008        count_relocation(kRelocRelative);
1009        MARK(rela->r_offset);
1010        if (sym) {
1011            DL_ERR("odd RELATIVE form...");
1012            return -1;
1013        }
1014        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1015                   reloc, (si->base + rela->r_addend));
1016        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1017        break;
1018
1019    case R_AARCH64_COPY:
1020        /*
1021         * ET_EXEC is not supported so this should not happen.
1022         *
1023         * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1024         *
1025         * Section 4.7.1.10 "Dynamic relocations"
1026         * R_AARCH64_COPY may only appear in executable objects where e_type is
1027         * set to ET_EXEC.
1028         */
1029        DL_ERR("%s R_AARCH64_COPY relocations are not supported", si->name);
1030        return -1;
1031    case R_AARCH64_TLS_TPREL64:
1032        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1033                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1034        break;
1035    case R_AARCH64_TLS_DTPREL32:
1036        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1037                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1038        break;
1039#elif defined(__x86_64__)
1040    case R_X86_64_JUMP_SLOT:
1041      count_relocation(kRelocAbsolute);
1042      MARK(rela->r_offset);
1043      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1044                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1045      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1046      break;
1047    case R_X86_64_GLOB_DAT:
1048      count_relocation(kRelocAbsolute);
1049      MARK(rela->r_offset);
1050      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1051                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1052      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1053      break;
1054    case R_X86_64_RELATIVE:
1055      count_relocation(kRelocRelative);
1056      MARK(rela->r_offset);
1057      if (sym) {
1058        DL_ERR("odd RELATIVE form...");
1059        return -1;
1060      }
1061      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1062                 static_cast<size_t>(si->base));
1063      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1064      break;
1065    case R_X86_64_32:
1066      count_relocation(kRelocRelative);
1067      MARK(rela->r_offset);
1068      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1069                 static_cast<size_t>(sym_addr), sym_name);
1070      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1071      break;
1072    case R_X86_64_64:
1073      count_relocation(kRelocRelative);
1074      MARK(rela->r_offset);
1075      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1076                 static_cast<size_t>(sym_addr), sym_name);
1077      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1078      break;
1079    case R_X86_64_PC32:
1080      count_relocation(kRelocRelative);
1081      MARK(rela->r_offset);
1082      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1083                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1084                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1085      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1086      break;
1087#endif
1088
1089    default:
1090      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1091      return -1;
1092    }
1093  }
1094  return 0;
1095}
1096
1097#else // REL, not RELA.
1098
1099static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1100    ElfW(Sym)* s;
1101    soinfo* lsi;
1102
1103    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1104        unsigned type = ELFW(R_TYPE)(rel->r_info);
1105        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1106        unsigned sym = ELFW(R_SYM)(rel->r_info);
1107        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1108        ElfW(Addr) sym_addr = 0;
1109        const char* sym_name = NULL;
1110
1111        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1112        if (type == 0) { // R_*_NONE
1113            continue;
1114        }
1115        if (sym != 0) {
1116            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1117            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1118            if (s == NULL) {
1119                // We only allow an undefined symbol if this is a weak reference...
1120                s = &si->symtab[sym];
1121                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1122                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1123                    return -1;
1124                }
1125
1126                /* IHI0044C AAELF 4.5.1.1:
1127
1128                   Libraries are not searched to resolve weak references.
1129                   It is not an error for a weak reference to remain
1130                   unsatisfied.
1131
1132                   During linking, the value of an undefined weak reference is:
1133                   - Zero if the relocation type is absolute
1134                   - The address of the place if the relocation is pc-relative
1135                   - The address of nominal base address if the relocation
1136                     type is base-relative.
1137                  */
1138
1139                switch (type) {
1140#if defined(__arm__)
1141                case R_ARM_JUMP_SLOT:
1142                case R_ARM_GLOB_DAT:
1143                case R_ARM_ABS32:
1144                case R_ARM_RELATIVE:    /* Don't care. */
1145                    // sym_addr was initialized to be zero above or relocation
1146                    // code below does not care about value of sym_addr.
1147                    // No need to do anything.
1148                    break;
1149#elif defined(__i386__)
1150                case R_386_JMP_SLOT:
1151                case R_386_GLOB_DAT:
1152                case R_386_32:
1153                case R_386_RELATIVE:    /* Don't care. */
1154                    // sym_addr was initialized to be zero above or relocation
1155                    // code below does not care about value of sym_addr.
1156                    // No need to do anything.
1157                    break;
1158                case R_386_PC32:
1159                    sym_addr = reloc;
1160                    break;
1161#endif
1162
1163#if defined(__arm__)
1164                case R_ARM_COPY:
1165                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1166#endif
1167                default:
1168                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1169                    return -1;
1170                }
1171            } else {
1172                // We got a definition.
1173                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1174            }
1175            count_relocation(kRelocSymbol);
1176        } else {
1177            s = NULL;
1178        }
1179
1180        switch (type) {
1181#if defined(__arm__)
1182        case R_ARM_JUMP_SLOT:
1183            count_relocation(kRelocAbsolute);
1184            MARK(rel->r_offset);
1185            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1186            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1187            break;
1188        case R_ARM_GLOB_DAT:
1189            count_relocation(kRelocAbsolute);
1190            MARK(rel->r_offset);
1191            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1192            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1193            break;
1194        case R_ARM_ABS32:
1195            count_relocation(kRelocAbsolute);
1196            MARK(rel->r_offset);
1197            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1198            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1199            break;
1200        case R_ARM_REL32:
1201            count_relocation(kRelocRelative);
1202            MARK(rel->r_offset);
1203            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1204                       reloc, sym_addr, rel->r_offset, sym_name);
1205            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1206            break;
1207        case R_ARM_COPY:
1208            /*
1209             * ET_EXEC is not supported so this should not happen.
1210             *
1211             * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1212             *
1213             * Section 4.7.1.10 "Dynamic relocations"
1214             * R_ARM_COPY may only appear in executable objects where e_type is
1215             * set to ET_EXEC.
1216             */
1217            DL_ERR("%s R_ARM_COPY relocations are not supported", si->name);
1218            return -1;
1219#elif defined(__i386__)
1220        case R_386_JMP_SLOT:
1221            count_relocation(kRelocAbsolute);
1222            MARK(rel->r_offset);
1223            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1224            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1225            break;
1226        case R_386_GLOB_DAT:
1227            count_relocation(kRelocAbsolute);
1228            MARK(rel->r_offset);
1229            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1230            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1231            break;
1232        case R_386_32:
1233            count_relocation(kRelocRelative);
1234            MARK(rel->r_offset);
1235            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1236            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1237            break;
1238        case R_386_PC32:
1239            count_relocation(kRelocRelative);
1240            MARK(rel->r_offset);
1241            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1242                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1243            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1244            break;
1245#elif defined(__mips__)
1246        case R_MIPS_REL32:
1247#if defined(__LP64__)
1248            // MIPS Elf64_Rel entries contain compound relocations
1249            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1250            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1251                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1252                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1253                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1254                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1255                return -1;
1256            }
1257#endif
1258            count_relocation(kRelocAbsolute);
1259            MARK(rel->r_offset);
1260            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1261                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1262            if (s) {
1263                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1264            } else {
1265                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1266            }
1267            break;
1268#endif
1269
1270#if defined(__arm__)
1271        case R_ARM_RELATIVE:
1272#elif defined(__i386__)
1273        case R_386_RELATIVE:
1274#endif
1275            count_relocation(kRelocRelative);
1276            MARK(rel->r_offset);
1277            if (sym) {
1278                DL_ERR("odd RELATIVE form...");
1279                return -1;
1280            }
1281            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1282                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1283            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1284            break;
1285
1286        default:
1287            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1288            return -1;
1289        }
1290    }
1291    return 0;
1292}
1293#endif
1294
1295#if defined(__mips__)
1296static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1297    ElfW(Addr)** got = si->plt_got;
1298    if (got == NULL) {
1299        return true;
1300    }
1301    unsigned local_gotno = si->mips_local_gotno;
1302    unsigned gotsym = si->mips_gotsym;
1303    unsigned symtabno = si->mips_symtabno;
1304    ElfW(Sym)* symtab = si->symtab;
1305
1306    // got[0] is the address of the lazy resolver function.
1307    // got[1] may be used for a GNU extension.
1308    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1309    // FIXME: maybe this should be in a separate routine?
1310    if ((si->flags & FLAG_LINKER) == 0) {
1311        size_t g = 0;
1312        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1313        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1314            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1315        }
1316        // Relocate the local GOT entries.
1317        for (; g < local_gotno; g++) {
1318            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1319        }
1320    }
1321
1322    // Now for the global GOT entries...
1323    ElfW(Sym)* sym = symtab + gotsym;
1324    got = si->plt_got + local_gotno;
1325    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1326        // This is an undefined reference... try to locate it.
1327        const char* sym_name = si->strtab + sym->st_name;
1328        soinfo* lsi;
1329        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1330        if (s == NULL) {
1331            // We only allow an undefined symbol if this is a weak reference.
1332            s = &symtab[g];
1333            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1334                DL_ERR("cannot locate \"%s\"...", sym_name);
1335                return false;
1336            }
1337            *got = 0;
1338        } else {
1339            // FIXME: is this sufficient?
1340            // For reference see NetBSD link loader
1341            // 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
1342            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1343        }
1344    }
1345    return true;
1346}
1347#endif
1348
1349void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1350  if (functions == NULL) {
1351    return;
1352  }
1353
1354  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1355
1356  int begin = reverse ? (count - 1) : 0;
1357  int end = reverse ? -1 : count;
1358  int step = reverse ? -1 : 1;
1359
1360  for (int i = begin; i != end; i += step) {
1361    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1362    CallFunction("function", functions[i]);
1363  }
1364
1365  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1366}
1367
1368void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1369  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1370    return;
1371  }
1372
1373  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1374  function();
1375  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1376
1377  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1378  // are still writable. This happens with our debug malloc (see http://b/7941716).
1379  protect_data(PROT_READ | PROT_WRITE);
1380}
1381
1382void soinfo::CallPreInitConstructors() {
1383  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1384  // but ignored in a shared library.
1385  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1386}
1387
1388void soinfo::CallConstructors() {
1389  if (constructors_called) {
1390    return;
1391  }
1392
1393  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1394  // protect against recursive constructor calls. One simple example of constructor recursion
1395  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1396  // 1. The program depends on libc, so libc's constructor is called here.
1397  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1398  // 3. dlopen() calls the constructors on the newly created
1399  //    soinfo for libc_malloc_debug_leak.so.
1400  // 4. The debug .so depends on libc, so CallConstructors is
1401  //    called again with the libc soinfo. If it doesn't trigger the early-
1402  //    out above, the libc constructor will be called again (recursively!).
1403  constructors_called = true;
1404
1405  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1406    // The GNU dynamic linker silently ignores these, but we warn the developer.
1407    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1408          name, preinit_array_count);
1409  }
1410
1411  get_children().for_each([] (soinfo* si) {
1412    si->CallConstructors();
1413  });
1414
1415  TRACE("\"%s\": calling constructors", name);
1416
1417  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1418  CallFunction("DT_INIT", init_func);
1419  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1420}
1421
1422void soinfo::CallDestructors() {
1423  TRACE("\"%s\": calling destructors", name);
1424
1425  // DT_FINI_ARRAY must be parsed in reverse order.
1426  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1427
1428  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1429  CallFunction("DT_FINI", fini_func);
1430}
1431
1432void soinfo::add_child(soinfo* child) {
1433  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1434    return;
1435  }
1436
1437  this->children.push_front(child);
1438  child->parents.push_front(this);
1439}
1440
1441void soinfo::remove_all_links() {
1442  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1443    return;
1444  }
1445
1446  // 1. Untie connected soinfos from 'this'.
1447  children.for_each([&] (soinfo* child) {
1448    child->parents.remove_if([&] (const soinfo* parent) {
1449      return parent == this;
1450    });
1451  });
1452
1453  parents.for_each([&] (soinfo* parent) {
1454    parent->children.for_each([&] (const soinfo* child) {
1455      return child == this;
1456    });
1457  });
1458
1459  // 2. Once everything untied - clear local lists.
1460  parents.clear();
1461  children.clear();
1462}
1463
1464void soinfo::set_st_dev(dev_t dev) {
1465  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1466    return;
1467  }
1468
1469  st_dev = dev;
1470}
1471
1472void soinfo::set_st_ino(ino_t ino) {
1473  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1474    return;
1475  }
1476
1477  st_ino = ino;
1478}
1479
1480dev_t soinfo::get_st_dev() {
1481  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1482    return 0;
1483  }
1484
1485  return st_dev;
1486};
1487
1488ino_t soinfo::get_st_ino() {
1489  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1490    return 0;
1491  }
1492
1493  return st_ino;
1494}
1495
1496// This is a return on get_children() in case
1497// 'this->flags' does not have FLAG_NEW_SOINFO set.
1498static soinfo::soinfo_list_t g_empty_list;
1499
1500soinfo::soinfo_list_t& soinfo::get_children() {
1501  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1502    return g_empty_list;
1503  }
1504
1505  return this->children;
1506}
1507
1508/* Force any of the closed stdin, stdout and stderr to be associated with
1509   /dev/null. */
1510static int nullify_closed_stdio() {
1511    int dev_null, i, status;
1512    int return_value = 0;
1513
1514    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1515    if (dev_null < 0) {
1516        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1517        return -1;
1518    }
1519    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1520
1521    /* If any of the stdio file descriptors is valid and not associated
1522       with /dev/null, dup /dev/null to it.  */
1523    for (i = 0; i < 3; i++) {
1524        /* If it is /dev/null already, we are done. */
1525        if (i == dev_null) {
1526            continue;
1527        }
1528
1529        TRACE("[ Nullifying stdio file descriptor %d]", i);
1530        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1531
1532        /* If file is opened, we are good. */
1533        if (status != -1) {
1534            continue;
1535        }
1536
1537        /* The only error we allow is that the file descriptor does not
1538           exist, in which case we dup /dev/null to it. */
1539        if (errno != EBADF) {
1540            DL_ERR("fcntl failed: %s", strerror(errno));
1541            return_value = -1;
1542            continue;
1543        }
1544
1545        /* Try dupping /dev/null to this stdio file descriptor and
1546           repeat if there is a signal.  Note that any errors in closing
1547           the stdio descriptor are lost.  */
1548        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1549        if (status < 0) {
1550            DL_ERR("dup2 failed: %s", strerror(errno));
1551            return_value = -1;
1552            continue;
1553        }
1554    }
1555
1556    /* If /dev/null is not one of the stdio file descriptors, close it. */
1557    if (dev_null > 2) {
1558        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1559        status = TEMP_FAILURE_RETRY(close(dev_null));
1560        if (status == -1) {
1561            DL_ERR("close failed: %s", strerror(errno));
1562            return_value = -1;
1563        }
1564    }
1565
1566    return return_value;
1567}
1568
1569static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1570    /* "base" might wrap around UINT32_MAX. */
1571    ElfW(Addr) base = si->load_bias;
1572    const ElfW(Phdr)* phdr = si->phdr;
1573    int phnum = si->phnum;
1574    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1575
1576    /* We can't debug anything until the linker is relocated */
1577    if (!relocating_linker) {
1578        INFO("[ linking %s ]", si->name);
1579        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1580    }
1581
1582    /* Extract dynamic section */
1583    size_t dynamic_count;
1584    ElfW(Word) dynamic_flags;
1585    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1586                                   &dynamic_count, &dynamic_flags);
1587    if (si->dynamic == NULL) {
1588        if (!relocating_linker) {
1589            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1590        }
1591        return false;
1592    } else {
1593        if (!relocating_linker) {
1594            DEBUG("dynamic = %p", si->dynamic);
1595        }
1596    }
1597
1598#if defined(__arm__)
1599    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1600                                    &si->ARM_exidx, &si->ARM_exidx_count);
1601#endif
1602
1603    // Extract useful information from dynamic section.
1604    uint32_t needed_count = 0;
1605    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1606        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1607              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1608        switch (d->d_tag) {
1609        case DT_HASH:
1610            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1611            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1612            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1613            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1614            break;
1615        case DT_STRTAB:
1616            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1617            break;
1618        case DT_SYMTAB:
1619            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1620            break;
1621#if !defined(__LP64__)
1622        case DT_PLTREL:
1623            if (d->d_un.d_val != DT_REL) {
1624                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1625                return false;
1626            }
1627            break;
1628#endif
1629        case DT_JMPREL:
1630#if defined(USE_RELA)
1631            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1632#else
1633            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1634#endif
1635            break;
1636        case DT_PLTRELSZ:
1637#if defined(USE_RELA)
1638            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1639#else
1640            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1641#endif
1642            break;
1643#if defined(__mips__)
1644        case DT_PLTGOT:
1645            // Used by mips and mips64.
1646            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1647            break;
1648#endif
1649        case DT_DEBUG:
1650            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1651            // if the dynamic table is writable
1652// FIXME: not working currently for N64
1653// The flags for the LOAD and DYNAMIC program headers do not agree.
1654// The LOAD section containng the dynamic table has been mapped as
1655// read-only, but the DYNAMIC header claims it is writable.
1656#if !(defined(__mips__) && defined(__LP64__))
1657            if ((dynamic_flags & PF_W) != 0) {
1658                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1659            }
1660            break;
1661#endif
1662#if defined(USE_RELA)
1663         case DT_RELA:
1664            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1665            break;
1666         case DT_RELASZ:
1667            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1668            break;
1669        case DT_REL:
1670            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1671            return false;
1672        case DT_RELSZ:
1673            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1674            return false;
1675#else
1676        case DT_REL:
1677            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1678            break;
1679        case DT_RELSZ:
1680            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1681            break;
1682         case DT_RELA:
1683            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1684            return false;
1685#endif
1686        case DT_INIT:
1687            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1688            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1689            break;
1690        case DT_FINI:
1691            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1692            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1693            break;
1694        case DT_INIT_ARRAY:
1695            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1696            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1697            break;
1698        case DT_INIT_ARRAYSZ:
1699            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1700            break;
1701        case DT_FINI_ARRAY:
1702            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1703            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1704            break;
1705        case DT_FINI_ARRAYSZ:
1706            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1707            break;
1708        case DT_PREINIT_ARRAY:
1709            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1710            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1711            break;
1712        case DT_PREINIT_ARRAYSZ:
1713            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1714            break;
1715        case DT_TEXTREL:
1716#if defined(__LP64__)
1717            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1718            return false;
1719#else
1720            si->has_text_relocations = true;
1721            break;
1722#endif
1723        case DT_SYMBOLIC:
1724            si->has_DT_SYMBOLIC = true;
1725            break;
1726        case DT_NEEDED:
1727            ++needed_count;
1728            break;
1729        case DT_FLAGS:
1730            if (d->d_un.d_val & DF_TEXTREL) {
1731#if defined(__LP64__)
1732                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1733                return false;
1734#else
1735                si->has_text_relocations = true;
1736#endif
1737            }
1738            if (d->d_un.d_val & DF_SYMBOLIC) {
1739                si->has_DT_SYMBOLIC = true;
1740            }
1741            break;
1742#if defined(__mips__)
1743        case DT_STRSZ:
1744        case DT_SYMENT:
1745        case DT_RELENT:
1746             break;
1747        case DT_MIPS_RLD_MAP:
1748            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1749            {
1750              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1751              *dp = &_r_debug;
1752            }
1753            break;
1754        case DT_MIPS_RLD_VERSION:
1755        case DT_MIPS_FLAGS:
1756        case DT_MIPS_BASE_ADDRESS:
1757        case DT_MIPS_UNREFEXTNO:
1758            break;
1759
1760        case DT_MIPS_SYMTABNO:
1761            si->mips_symtabno = d->d_un.d_val;
1762            break;
1763
1764        case DT_MIPS_LOCAL_GOTNO:
1765            si->mips_local_gotno = d->d_un.d_val;
1766            break;
1767
1768        case DT_MIPS_GOTSYM:
1769            si->mips_gotsym = d->d_un.d_val;
1770            break;
1771#endif
1772
1773        default:
1774            DEBUG("Unused DT entry: type %p arg %p",
1775                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1776            break;
1777        }
1778    }
1779
1780    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1781          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1782
1783    // Sanity checks.
1784    if (relocating_linker && needed_count != 0) {
1785        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1786        return false;
1787    }
1788    if (si->nbucket == 0) {
1789        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1790        return false;
1791    }
1792    if (si->strtab == 0) {
1793        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1794        return false;
1795    }
1796    if (si->symtab == 0) {
1797        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1798        return false;
1799    }
1800
1801    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1802    if (si->flags & FLAG_EXE) {
1803        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1804        size_t preload_count = 0;
1805        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1806            soinfo* lsi = find_library(g_ld_preload_names[i], NULL);
1807            if (lsi != NULL) {
1808                g_ld_preloads[preload_count++] = lsi;
1809            } else {
1810                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1811                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1812                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1813            }
1814        }
1815    }
1816
1817    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1818    soinfo** pneeded = needed;
1819
1820    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1821        if (d->d_tag == DT_NEEDED) {
1822            const char* library_name = si->strtab + d->d_un.d_val;
1823            DEBUG("%s needs %s", si->name, library_name);
1824            soinfo* lsi = find_library(library_name, NULL);
1825            if (lsi == NULL) {
1826                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1827                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1828                       library_name, si->name, tmp_err_buf);
1829                return false;
1830            }
1831
1832            si->add_child(lsi);
1833            *pneeded++ = lsi;
1834        }
1835    }
1836    *pneeded = NULL;
1837
1838#if !defined(__LP64__)
1839    if (si->has_text_relocations) {
1840        // Make segments writable to allow text relocations to work properly. We will later call
1841        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1842#if !defined(__i386__) // The platform itself has too many text relocations on x86.
1843        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1844                "security hardening. Please fix.", si->name);
1845#endif
1846        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1847            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1848                   si->name, strerror(errno));
1849            return false;
1850        }
1851    }
1852#endif
1853
1854#if defined(USE_RELA)
1855    if (si->plt_rela != NULL) {
1856        DEBUG("[ relocating %s plt ]\n", si->name);
1857        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1858            return false;
1859        }
1860    }
1861    if (si->rela != NULL) {
1862        DEBUG("[ relocating %s ]\n", si->name);
1863        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1864            return false;
1865        }
1866    }
1867#else
1868    if (si->plt_rel != NULL) {
1869        DEBUG("[ relocating %s plt ]", si->name);
1870        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1871            return false;
1872        }
1873    }
1874    if (si->rel != NULL) {
1875        DEBUG("[ relocating %s ]", si->name);
1876        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1877            return false;
1878        }
1879    }
1880#endif
1881
1882#if defined(__mips__)
1883    if (!mips_relocate_got(si, needed)) {
1884        return false;
1885    }
1886#endif
1887
1888    si->flags |= FLAG_LINKED;
1889    DEBUG("[ finished linking %s ]", si->name);
1890
1891#if !defined(__LP64__)
1892    if (si->has_text_relocations) {
1893        // All relocations are done, we can protect our segments back to read-only.
1894        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1895            DL_ERR("can't protect segments for \"%s\": %s",
1896                   si->name, strerror(errno));
1897            return false;
1898        }
1899    }
1900#endif
1901
1902    /* We can also turn on GNU RELRO protection */
1903    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1904        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1905               si->name, strerror(errno));
1906        return false;
1907    }
1908
1909    /* Handle serializing/sharing the RELRO segment */
1910    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1911      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
1912                                         extinfo->relro_fd) < 0) {
1913        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
1914               si->name, strerror(errno));
1915        return false;
1916      }
1917    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
1918      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
1919                                   extinfo->relro_fd) < 0) {
1920        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
1921               si->name, strerror(errno));
1922        return false;
1923      }
1924    }
1925
1926    notify_gdb_of_load(si);
1927    return true;
1928}
1929
1930/*
1931 * This function add vdso to internal dso list.
1932 * It helps to stack unwinding through signal handlers.
1933 * Also, it makes bionic more like glibc.
1934 */
1935static void add_vdso(KernelArgumentBlock& args __unused) {
1936#if defined(AT_SYSINFO_EHDR)
1937  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
1938  if (ehdr_vdso == NULL) {
1939    return;
1940  }
1941
1942  soinfo* si = soinfo_alloc("[vdso]", NULL);
1943
1944  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
1945  si->phnum = ehdr_vdso->e_phnum;
1946  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
1947  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
1948  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
1949
1950  soinfo_link_image(si, NULL);
1951#endif
1952}
1953
1954/*
1955 * This is linker soinfo for GDB. See details below.
1956 */
1957static soinfo linker_soinfo_for_gdb;
1958
1959/* gdb expects the linker to be in the debug shared object list.
1960 * Without this, gdb has trouble locating the linker's ".text"
1961 * and ".plt" sections. Gdb could also potentially use this to
1962 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
1963 * Don't use soinfo_alloc(), because the linker shouldn't
1964 * be on the soinfo list.
1965 */
1966static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
1967#if defined(__LP64__)
1968  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
1969#else
1970  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
1971#endif
1972  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
1973  linker_soinfo_for_gdb.base = linker_base;
1974
1975  /*
1976   * Set the dynamic field in the link map otherwise gdb will complain with
1977   * the following:
1978   *   warning: .dynamic section for "/system/bin/linker" is not at the
1979   *   expected address (wrong library or version mismatch?)
1980   */
1981  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
1982  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
1983  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
1984                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
1985  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
1986}
1987
1988/*
1989 * This code is called after the linker has linked itself and
1990 * fixed it's own GOT. It is safe to make references to externs
1991 * and other non-local data at this point.
1992 */
1993static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
1994    /* NOTE: we store the args pointer on a special location
1995     *       of the temporary TLS area in order to pass it to
1996     *       the C Library's runtime initializer.
1997     *
1998     *       The initializer must clear the slot and reset the TLS
1999     *       to point to a different location to ensure that no other
2000     *       shared library constructor can access it.
2001     */
2002  __libc_init_tls(args);
2003
2004#if TIMING
2005    struct timeval t0, t1;
2006    gettimeofday(&t0, 0);
2007#endif
2008
2009    // Initialize environment functions, and get to the ELF aux vectors table.
2010    linker_env_init(args);
2011
2012    // If this is a setuid/setgid program, close the security hole described in
2013    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2014    if (get_AT_SECURE()) {
2015        nullify_closed_stdio();
2016    }
2017
2018    debuggerd_init();
2019
2020    // Get a few environment variables.
2021    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2022    if (LD_DEBUG != NULL) {
2023      g_ld_debug_verbosity = atoi(LD_DEBUG);
2024    }
2025
2026    // Normally, these are cleaned by linker_env_init, but the test
2027    // doesn't cost us anything.
2028    const char* ldpath_env = NULL;
2029    const char* ldpreload_env = NULL;
2030    if (!get_AT_SECURE()) {
2031      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2032      ldpreload_env = linker_env_get("LD_PRELOAD");
2033    }
2034
2035    // Linker does not call constructors for its own
2036    // global variables so we need to initialize
2037    // the allocators explicitly.
2038    g_soinfo_allocator.init();
2039    g_soinfo_links_allocator.init();
2040
2041    INFO("[ android linker & debugger ]");
2042
2043    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2044    if (si == NULL) {
2045        exit(EXIT_FAILURE);
2046    }
2047
2048    /* bootstrap the link map, the main exe always needs to be first */
2049    si->flags |= FLAG_EXE;
2050    link_map* map = &(si->link_map_head);
2051
2052    map->l_addr = 0;
2053    map->l_name = args.argv[0];
2054    map->l_prev = NULL;
2055    map->l_next = NULL;
2056
2057    _r_debug.r_map = map;
2058    r_debug_tail = map;
2059
2060    init_linker_info_for_gdb(linker_base);
2061
2062    // Extract information passed from the kernel.
2063    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2064    si->phnum = args.getauxval(AT_PHNUM);
2065    si->entry = args.getauxval(AT_ENTRY);
2066
2067    /* Compute the value of si->base. We can't rely on the fact that
2068     * the first entry is the PHDR because this will not be true
2069     * for certain executables (e.g. some in the NDK unit test suite)
2070     */
2071    si->base = 0;
2072    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2073    si->load_bias = 0;
2074    for (size_t i = 0; i < si->phnum; ++i) {
2075      if (si->phdr[i].p_type == PT_PHDR) {
2076        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2077        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2078        break;
2079      }
2080    }
2081    si->dynamic = NULL;
2082    si->ref_count = 1;
2083
2084    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2085    if (elf_hdr->e_type != ET_DYN) {
2086        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2087        exit(EXIT_FAILURE);
2088    }
2089
2090    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2091    parse_LD_LIBRARY_PATH(ldpath_env);
2092    parse_LD_PRELOAD(ldpreload_env);
2093
2094    somain = si;
2095
2096    if (!soinfo_link_image(si, NULL)) {
2097        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2098        exit(EXIT_FAILURE);
2099    }
2100
2101    add_vdso(args);
2102
2103    si->CallPreInitConstructors();
2104
2105    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2106        g_ld_preloads[i]->CallConstructors();
2107    }
2108
2109    /* After the link_image, the si->load_bias is initialized.
2110     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2111     * We need to update this value for so exe here. So Unwind_Backtrace
2112     * for some arch like x86 could work correctly within so exe.
2113     */
2114    map->l_addr = si->load_bias;
2115    si->CallConstructors();
2116
2117#if TIMING
2118    gettimeofday(&t1, NULL);
2119    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2120               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2121               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2122#endif
2123#if STATS
2124    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2125           linker_stats.count[kRelocAbsolute],
2126           linker_stats.count[kRelocRelative],
2127           linker_stats.count[kRelocCopy],
2128           linker_stats.count[kRelocSymbol]);
2129#endif
2130#if COUNT_PAGES
2131    {
2132        unsigned n;
2133        unsigned i;
2134        unsigned count = 0;
2135        for (n = 0; n < 4096; n++) {
2136            if (bitmask[n]) {
2137                unsigned x = bitmask[n];
2138#if defined(__LP64__)
2139                for (i = 0; i < 32; i++) {
2140#else
2141                for (i = 0; i < 8; i++) {
2142#endif
2143                    if (x & 1) {
2144                        count++;
2145                    }
2146                    x >>= 1;
2147                }
2148            }
2149        }
2150        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2151    }
2152#endif
2153
2154#if TIMING || STATS || COUNT_PAGES
2155    fflush(stdout);
2156#endif
2157
2158    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2159    return si->entry;
2160}
2161
2162/* Compute the load-bias of an existing executable. This shall only
2163 * be used to compute the load bias of an executable or shared library
2164 * that was loaded by the kernel itself.
2165 *
2166 * Input:
2167 *    elf    -> address of ELF header, assumed to be at the start of the file.
2168 * Return:
2169 *    load bias, i.e. add the value of any p_vaddr in the file to get
2170 *    the corresponding address in memory.
2171 */
2172static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2173  ElfW(Addr) offset = elf->e_phoff;
2174  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2175  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2176
2177  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2178    if (phdr->p_type == PT_LOAD) {
2179      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2180    }
2181  }
2182  return 0;
2183}
2184
2185/*
2186 * This is the entry point for the linker, called from begin.S. This
2187 * method is responsible for fixing the linker's own relocations, and
2188 * then calling __linker_init_post_relocation().
2189 *
2190 * Because this method is called before the linker has fixed it's own
2191 * relocations, any attempt to reference an extern variable, extern
2192 * function, or other GOT reference will generate a segfault.
2193 */
2194extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2195  // Initialize static variables.
2196  solist = get_libdl_info();
2197  sonext = get_libdl_info();
2198
2199  KernelArgumentBlock args(raw_args);
2200
2201  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2202  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2203  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2204
2205  soinfo linker_so;
2206  memset(&linker_so, 0, sizeof(soinfo));
2207
2208  strcpy(linker_so.name, "[dynamic linker]");
2209  linker_so.base = linker_addr;
2210  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2211  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2212  linker_so.dynamic = NULL;
2213  linker_so.phdr = phdr;
2214  linker_so.phnum = elf_hdr->e_phnum;
2215  linker_so.flags |= FLAG_LINKER;
2216
2217  if (!soinfo_link_image(&linker_so, NULL)) {
2218    // It would be nice to print an error message, but if the linker
2219    // can't link itself, there's no guarantee that we'll be able to
2220    // call write() (because it involves a GOT reference). We may as
2221    // well try though...
2222    const char* msg = "CANNOT LINK EXECUTABLE: ";
2223    write(2, msg, strlen(msg));
2224    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2225    write(2, "\n", 1);
2226    _exit(EXIT_FAILURE);
2227  }
2228
2229  // We have successfully fixed our own relocations. It's safe to run
2230  // the main part of the linker now.
2231  args.abort_message_ptr = &g_abort_message;
2232  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2233
2234  protect_data(PROT_READ);
2235
2236  // Return the address that the calling assembly stub should jump to.
2237  return start_address;
2238}
2239