linker.cpp revision 2aebf5429bb1241a3298b5b642d38f73124c2026
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
52/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
53 *
54 * Do NOT use malloc() and friends or pthread_*() code here.
55 * Don't use printf() either; it's caused mysterious memory
56 * corruption in the past.
57 * The linker runs before we bring up libc and it's easiest
58 * to make sure it does not depend on any complex libc features
59 *
60 * open issues / todo:
61 *
62 * - are we doing everything we should for ARM_COPY relocations?
63 * - cleaner error reporting
64 * - after linking, set as much stuff as possible to READONLY
65 *   and NOEXEC
66 */
67
68static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
69static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
70
71// We can't use malloc(3) in the dynamic linker. We use a linked list of anonymous
72// maps, each a single page in size. The pages are broken up into as many struct soinfo
73// objects as will fit, and they're all threaded together on a free list.
74#define SOINFO_PER_POOL ((PAGE_SIZE - sizeof(soinfo_pool_t*)) / sizeof(soinfo))
75struct soinfo_pool_t {
76  soinfo_pool_t* next;
77  soinfo info[SOINFO_PER_POOL];
78};
79static struct soinfo_pool_t* gSoInfoPools = NULL;
80static soinfo* gSoInfoFreeList = NULL;
81
82static soinfo* solist = &libdl_info;
83static soinfo* sonext = &libdl_info;
84static soinfo* somain; /* main process, always the one after libdl_info */
85
86static const char* const gDefaultLdPaths[] = {
87#if defined(__LP64__)
88  "/vendor/lib64",
89  "/system/lib64",
90#else
91  "/vendor/lib",
92  "/system/lib",
93#endif
94  NULL
95};
96
97#define LDPATH_BUFSIZE (LDPATH_MAX*64)
98#define LDPATH_MAX 8
99
100#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
101#define LDPRELOAD_MAX 8
102
103static char gLdPathsBuffer[LDPATH_BUFSIZE];
104static const char* gLdPaths[LDPATH_MAX + 1];
105
106static char gLdPreloadsBuffer[LDPRELOAD_BUFSIZE];
107static const char* gLdPreloadNames[LDPRELOAD_MAX + 1];
108
109static soinfo* gLdPreloads[LDPRELOAD_MAX + 1];
110
111__LIBC_HIDDEN__ int gLdDebugVerbosity;
112
113__LIBC_HIDDEN__ abort_msg_t* gAbortMessage = NULL; // For debuggerd.
114
115enum RelocationKind {
116    kRelocAbsolute = 0,
117    kRelocRelative,
118    kRelocCopy,
119    kRelocSymbol,
120    kRelocMax
121};
122
123#if STATS
124struct linker_stats_t {
125    int count[kRelocMax];
126};
127
128static linker_stats_t linker_stats;
129
130static void count_relocation(RelocationKind kind) {
131    ++linker_stats.count[kind];
132}
133#else
134static void count_relocation(RelocationKind) {
135}
136#endif
137
138#if COUNT_PAGES
139static unsigned bitmask[4096];
140#if defined(__LP64__)
141#define MARK(offset) \
142    do { \
143        if ((((offset) >> 12) >> 5) < 4096) \
144            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
145    } while (0)
146#else
147#define MARK(offset) \
148    do { \
149        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
150    } while (0)
151#endif
152#else
153#define MARK(x) do {} while (0)
154#endif
155
156// You shouldn't try to call memory-allocating functions in the dynamic linker.
157// Guard against the most obvious ones.
158#define DISALLOW_ALLOCATION(return_type, name, ...) \
159    return_type name __VA_ARGS__ \
160    { \
161        const char* msg = "ERROR: " #name " called from the dynamic linker!\n"; \
162        __libc_format_log(ANDROID_LOG_FATAL, "linker", "%s", msg); \
163        write(2, msg, strlen(msg)); \
164        abort(); \
165    }
166DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
167DISALLOW_ALLOCATION(void, free, (void* u __unused));
168DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
169DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
170
171static char tmp_err_buf[768];
172static char __linker_dl_err_buf[768];
173
174char* linker_get_error_buffer() {
175  return &__linker_dl_err_buf[0];
176}
177
178size_t linker_get_error_buffer_size() {
179  return sizeof(__linker_dl_err_buf);
180}
181
182/*
183 * This function is an empty stub where GDB locates a breakpoint to get notified
184 * about linker activity.
185 */
186extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
187
188static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
189static link_map* r_debug_tail = 0;
190
191static pthread_mutex_t gDebugMutex = PTHREAD_MUTEX_INITIALIZER;
192
193static void insert_soinfo_into_debug_map(soinfo* info) {
194    // Copy the necessary fields into the debug structure.
195    link_map* map = &(info->link_map_head);
196    map->l_addr = info->load_bias;
197    map->l_name = reinterpret_cast<char*>(info->name);
198    map->l_ld = info->dynamic;
199
200    /* Stick the new library at the end of the list.
201     * gdb tends to care more about libc than it does
202     * about leaf libraries, and ordering it this way
203     * reduces the back-and-forth over the wire.
204     */
205    if (r_debug_tail) {
206        r_debug_tail->l_next = map;
207        map->l_prev = r_debug_tail;
208        map->l_next = 0;
209    } else {
210        _r_debug.r_map = map;
211        map->l_prev = 0;
212        map->l_next = 0;
213    }
214    r_debug_tail = map;
215}
216
217static void remove_soinfo_from_debug_map(soinfo* info) {
218    link_map* map = &(info->link_map_head);
219
220    if (r_debug_tail == map) {
221        r_debug_tail = map->l_prev;
222    }
223
224    if (map->l_prev) {
225        map->l_prev->l_next = map->l_next;
226    }
227    if (map->l_next) {
228        map->l_next->l_prev = map->l_prev;
229    }
230}
231
232static void notify_gdb_of_load(soinfo* info) {
233    if (info->flags & FLAG_EXE) {
234        // GDB already knows about the main executable
235        return;
236    }
237
238    ScopedPthreadMutexLocker locker(&gDebugMutex);
239
240    _r_debug.r_state = r_debug::RT_ADD;
241    rtld_db_dlactivity();
242
243    insert_soinfo_into_debug_map(info);
244
245    _r_debug.r_state = r_debug::RT_CONSISTENT;
246    rtld_db_dlactivity();
247}
248
249static void notify_gdb_of_unload(soinfo* info) {
250    if (info->flags & FLAG_EXE) {
251        // GDB already knows about the main executable
252        return;
253    }
254
255    ScopedPthreadMutexLocker locker(&gDebugMutex);
256
257    _r_debug.r_state = r_debug::RT_DELETE;
258    rtld_db_dlactivity();
259
260    remove_soinfo_from_debug_map(info);
261
262    _r_debug.r_state = r_debug::RT_CONSISTENT;
263    rtld_db_dlactivity();
264}
265
266void notify_gdb_of_libraries() {
267  _r_debug.r_state = r_debug::RT_ADD;
268  rtld_db_dlactivity();
269  _r_debug.r_state = r_debug::RT_CONSISTENT;
270  rtld_db_dlactivity();
271}
272
273static bool ensure_free_list_non_empty() {
274  if (gSoInfoFreeList != NULL) {
275    return true;
276  }
277
278  // Allocate a new pool.
279  soinfo_pool_t* pool = reinterpret_cast<soinfo_pool_t*>(mmap(NULL, sizeof(*pool),
280                                                              PROT_READ|PROT_WRITE,
281                                                              MAP_PRIVATE|MAP_ANONYMOUS, 0, 0));
282  if (pool == MAP_FAILED) {
283    return false;
284  }
285
286  // Add the pool to our list of pools.
287  pool->next = gSoInfoPools;
288  gSoInfoPools = pool;
289
290  // Chain the entries in the new pool onto the free list.
291  gSoInfoFreeList = &pool->info[0];
292  soinfo* next = NULL;
293  for (int i = SOINFO_PER_POOL - 1; i >= 0; --i) {
294    pool->info[i].next = next;
295    next = &pool->info[i];
296  }
297
298  return true;
299}
300
301static void set_soinfo_pool_protection(int protection) {
302  for (soinfo_pool_t* p = gSoInfoPools; p != NULL; p = p->next) {
303    if (mprotect(p, sizeof(*p), protection) == -1) {
304      abort(); // Can't happen.
305    }
306  }
307}
308
309static soinfo* soinfo_alloc(const char* name) {
310  if (strlen(name) >= SOINFO_NAME_LEN) {
311    DL_ERR("library name \"%s\" too long", name);
312    return NULL;
313  }
314
315  if (!ensure_free_list_non_empty()) {
316    DL_ERR("out of memory when loading \"%s\"", name);
317    return NULL;
318  }
319
320  // Take the head element off the free list.
321  soinfo* si = gSoInfoFreeList;
322  gSoInfoFreeList = gSoInfoFreeList->next;
323
324  // Initialize the new element.
325  memset(si, 0, sizeof(soinfo));
326  strlcpy(si->name, name, sizeof(si->name));
327  sonext->next = si;
328  sonext = si;
329
330  TRACE("name %s: allocated soinfo @ %p", name, si);
331  return si;
332}
333
334static void soinfo_free(soinfo* si) {
335    if (si == NULL) {
336        return;
337    }
338
339    soinfo *prev = NULL, *trav;
340
341    TRACE("name %s: freeing soinfo @ %p", si->name, si);
342
343    for (trav = solist; trav != NULL; trav = trav->next) {
344        if (trav == si)
345            break;
346        prev = trav;
347    }
348    if (trav == NULL) {
349        /* si was not in solist */
350        DL_ERR("name \"%s\" is not in solist!", si->name);
351        return;
352    }
353
354    /* prev will never be NULL, because the first entry in solist is
355       always the static libdl_info.
356    */
357    prev->next = si->next;
358    if (si == sonext) {
359        sonext = prev;
360    }
361    si->next = gSoInfoFreeList;
362    gSoInfoFreeList = si;
363}
364
365
366static void parse_path(const char* path, const char* delimiters,
367                       const char** array, char* buf, size_t buf_size, size_t max_count) {
368  if (path == NULL) {
369    return;
370  }
371
372  size_t len = strlcpy(buf, path, buf_size);
373
374  size_t i = 0;
375  char* buf_p = buf;
376  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
377    if (*array[i] != '\0') {
378      ++i;
379    }
380  }
381
382  // Forget the last path if we had to truncate; this occurs if the 2nd to
383  // last char isn't '\0' (i.e. wasn't originally a delimiter).
384  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
385    array[i - 1] = NULL;
386  } else {
387    array[i] = NULL;
388  }
389}
390
391static void parse_LD_LIBRARY_PATH(const char* path) {
392  parse_path(path, ":", gLdPaths,
393             gLdPathsBuffer, sizeof(gLdPathsBuffer), LDPATH_MAX);
394}
395
396static void parse_LD_PRELOAD(const char* path) {
397  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
398  parse_path(path, " :", gLdPreloadNames,
399             gLdPreloadsBuffer, sizeof(gLdPreloadsBuffer), LDPRELOAD_MAX);
400}
401
402#if defined(__arm__)
403
404/* For a given PC, find the .so that it belongs to.
405 * Returns the base address of the .ARM.exidx section
406 * for that .so, and the number of 8-byte entries
407 * in that section (via *pcount).
408 *
409 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
410 *
411 * This function is exposed via dlfcn.cpp and libdl.so.
412 */
413_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
414    unsigned addr = (unsigned)pc;
415
416    for (soinfo* si = solist; si != 0; si = si->next) {
417        if ((addr >= si->base) && (addr < (si->base + si->size))) {
418            *pcount = si->ARM_exidx_count;
419            return (_Unwind_Ptr)si->ARM_exidx;
420        }
421    }
422    *pcount = 0;
423    return NULL;
424}
425
426#endif
427
428/* Here, we only have to provide a callback to iterate across all the
429 * loaded libraries. gcc_eh does the rest. */
430int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
431    int rv = 0;
432    for (soinfo* si = solist; si != NULL; si = si->next) {
433        dl_phdr_info dl_info;
434        dl_info.dlpi_addr = si->link_map_head.l_addr;
435        dl_info.dlpi_name = si->link_map_head.l_name;
436        dl_info.dlpi_phdr = si->phdr;
437        dl_info.dlpi_phnum = si->phnum;
438        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
439        if (rv != 0) {
440            break;
441        }
442    }
443    return rv;
444}
445
446static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
447  ElfW(Sym)* symtab = si->symtab;
448  const char* strtab = si->strtab;
449
450  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
451             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
452
453  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
454    ElfW(Sym)* s = symtab + n;
455    if (strcmp(strtab + s->st_name, name)) continue;
456
457    /* only concern ourselves with global and weak symbol definitions */
458    switch (ELF_ST_BIND(s->st_info)) {
459      case STB_GLOBAL:
460      case STB_WEAK:
461        if (s->st_shndx == SHN_UNDEF) {
462        continue;
463      }
464
465      TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
466                 name, si->name, reinterpret_cast<void*>(s->st_value),
467                 static_cast<size_t>(s->st_size));
468      return s;
469    }
470  }
471
472  return NULL;
473}
474
475static unsigned elfhash(const char* _name) {
476    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
477    unsigned h = 0, g;
478
479    while (*name) {
480        h = (h << 4) + *name++;
481        g = h & 0xf0000000;
482        h ^= g;
483        h ^= g >> 24;
484    }
485    return h;
486}
487
488static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
489    unsigned elf_hash = elfhash(name);
490    ElfW(Sym)* s = NULL;
491
492    if (si != NULL && somain != NULL) {
493        /*
494         * Local scope is executable scope. Just start looking into it right away
495         * for the shortcut.
496         */
497
498        if (si == somain) {
499            s = soinfo_elf_lookup(si, elf_hash, name);
500            if (s != NULL) {
501                *lsi = si;
502                goto done;
503            }
504        } else {
505            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
506
507            /*
508             * If this object was built with symbolic relocations disabled, the
509             * first place to look to resolve external references is the main
510             * executable.
511             */
512
513            if (!si->has_DT_SYMBOLIC) {
514                DEBUG("%s: looking up %s in executable %s",
515                      si->name, name, somain->name);
516                s = soinfo_elf_lookup(somain, elf_hash, name);
517                if (s != NULL) {
518                    *lsi = somain;
519                    goto done;
520                }
521            }
522
523            /* Look for symbols in the local scope (the object who is
524             * searching). This happens with C++ templates on x86 for some
525             * reason.
526             *
527             * Notes on weak symbols:
528             * The ELF specs are ambiguous about treatment of weak definitions in
529             * dynamic linking.  Some systems return the first definition found
530             * and some the first non-weak definition.   This is system dependent.
531             * Here we return the first definition found for simplicity.  */
532
533            s = soinfo_elf_lookup(si, elf_hash, name);
534            if (s != NULL) {
535                *lsi = si;
536                goto done;
537            }
538
539            /*
540             * If this object was built with -Bsymbolic and symbol is not found
541             * in the local scope, try to find the symbol in the main executable.
542             */
543
544            if (si->has_DT_SYMBOLIC) {
545                DEBUG("%s: looking up %s in executable %s after local scope",
546                      si->name, name, somain->name);
547                s = soinfo_elf_lookup(somain, elf_hash, name);
548                if (s != NULL) {
549                    *lsi = somain;
550                    goto done;
551                }
552            }
553        }
554    }
555
556    /* Next, look for it in the preloads list */
557    for (int i = 0; gLdPreloads[i] != NULL; i++) {
558        s = soinfo_elf_lookup(gLdPreloads[i], elf_hash, name);
559        if (s != NULL) {
560            *lsi = gLdPreloads[i];
561            goto done;
562        }
563    }
564
565    for (int i = 0; needed[i] != NULL; i++) {
566        DEBUG("%s: looking up %s in %s",
567              si->name, name, needed[i]->name);
568        s = soinfo_elf_lookup(needed[i], elf_hash, name);
569        if (s != NULL) {
570            *lsi = needed[i];
571            goto done;
572        }
573    }
574
575done:
576    if (s != NULL) {
577        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
578                   "found in %s, base = %p, load bias = %p",
579                   si->name, name, reinterpret_cast<void*>(s->st_value),
580                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
581                   reinterpret_cast<void*>((*lsi)->load_bias));
582        return s;
583    }
584
585    return NULL;
586}
587
588/* This is used by dlsym(3).  It performs symbol lookup only within the
589   specified soinfo object and not in any of its dependencies.
590
591   TODO: Only looking in the specified soinfo seems wrong. dlsym(3) says
592   that it should do a breadth first search through the dependency
593   tree. This agrees with the ELF spec (aka System V Application
594   Binary Interface) where in Chapter 5 it discuss resolving "Shared
595   Object Dependencies" in breadth first search order.
596 */
597ElfW(Sym)* dlsym_handle_lookup(soinfo* si, const char* name) {
598    return soinfo_elf_lookup(si, elfhash(name), name);
599}
600
601/* This is used by dlsym(3) to performs a global symbol lookup. If the
602   start value is null (for RTLD_DEFAULT), the search starts at the
603   beginning of the global solist. Otherwise the search starts at the
604   specified soinfo (for RTLD_NEXT).
605 */
606ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start) {
607  unsigned elf_hash = elfhash(name);
608
609  if (start == NULL) {
610    start = solist;
611  }
612
613  ElfW(Sym)* s = NULL;
614  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
615    s = soinfo_elf_lookup(si, elf_hash, name);
616    if (s != NULL) {
617      *found = si;
618      break;
619    }
620  }
621
622  if (s != NULL) {
623    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
624               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
625  }
626
627  return s;
628}
629
630soinfo* find_containing_library(const void* p) {
631  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
632  for (soinfo* si = solist; si != NULL; si = si->next) {
633    if (address >= si->base && address - si->base < si->size) {
634      return si;
635    }
636  }
637  return NULL;
638}
639
640ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
641  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
642
643  // Search the library's symbol table for any defined symbol which
644  // contains this address.
645  for (size_t i = 0; i < si->nchain; ++i) {
646    ElfW(Sym)* sym = &si->symtab[i];
647    if (sym->st_shndx != SHN_UNDEF &&
648        soaddr >= sym->st_value &&
649        soaddr < sym->st_value + sym->st_size) {
650      return sym;
651    }
652  }
653
654  return NULL;
655}
656
657static int open_library_on_path(const char* name, const char* const paths[]) {
658  char buf[512];
659  for (size_t i = 0; paths[i] != NULL; ++i) {
660    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
661    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
662      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
663      continue;
664    }
665    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
666    if (fd != -1) {
667      return fd;
668    }
669  }
670  return -1;
671}
672
673static int open_library(const char* name) {
674  TRACE("[ opening %s ]", name);
675
676  // If the name contains a slash, we should attempt to open it directly and not search the paths.
677  if (strchr(name, '/') != NULL) {
678    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
679    if (fd != -1) {
680      return fd;
681    }
682    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
683#if defined(__LP64__)
684    return -1;
685#endif
686  }
687
688  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
689  int fd = open_library_on_path(name, gLdPaths);
690  if (fd == -1) {
691    fd = open_library_on_path(name, gDefaultLdPaths);
692  }
693  return fd;
694}
695
696static soinfo* load_library(const char* name, const android_dlextinfo* extinfo) {
697    // Open the file.
698    int fd = open_library(name);
699    if (fd == -1) {
700        DL_ERR("library \"%s\" not found", name);
701        return NULL;
702    }
703
704    // Read the ELF header and load the segments.
705    ElfReader elf_reader(name, fd);
706    if (!elf_reader.Load(extinfo)) {
707        return NULL;
708    }
709
710    const char* bname = strrchr(name, '/');
711    soinfo* si = soinfo_alloc(bname ? bname + 1 : name);
712    if (si == NULL) {
713        return NULL;
714    }
715    si->base = elf_reader.load_start();
716    si->size = elf_reader.load_size();
717    si->load_bias = elf_reader.load_bias();
718    si->flags = 0;
719    si->entry = 0;
720    si->dynamic = NULL;
721    si->phnum = elf_reader.phdr_count();
722    si->phdr = elf_reader.loaded_phdr();
723    return si;
724}
725
726static soinfo *find_loaded_library(const char* name) {
727    // TODO: don't use basename only for determining libraries
728    // http://code.google.com/p/android/issues/detail?id=6670
729
730    const char* bname = strrchr(name, '/');
731    bname = bname ? bname + 1 : name;
732
733    for (soinfo* si = solist; si != NULL; si = si->next) {
734        if (!strcmp(bname, si->name)) {
735            return si;
736        }
737    }
738    return NULL;
739}
740
741static soinfo* find_library_internal(const char* name, const android_dlextinfo* extinfo) {
742  if (name == NULL) {
743    return somain;
744  }
745
746  soinfo* si = find_loaded_library(name);
747  if (si != NULL) {
748    if (si->flags & FLAG_LINKED) {
749      return si;
750    }
751    DL_ERR("OOPS: recursive link to \"%s\"", si->name);
752    return NULL;
753  }
754
755  TRACE("[ '%s' has not been loaded yet.  Locating...]", name);
756  si = load_library(name, extinfo);
757  if (si == NULL) {
758    return NULL;
759  }
760
761  // At this point we know that whatever is loaded @ base is a valid ELF
762  // shared library whose segments are properly mapped in.
763  TRACE("[ find_library_internal base=%p size=%zu name='%s' ]",
764        reinterpret_cast<void*>(si->base), si->size, si->name);
765
766  if (!soinfo_link_image(si, extinfo)) {
767    munmap(reinterpret_cast<void*>(si->base), si->size);
768    soinfo_free(si);
769    return NULL;
770  }
771
772  return si;
773}
774
775static soinfo* find_library(const char* name, const android_dlextinfo* extinfo) {
776  soinfo* si = find_library_internal(name, extinfo);
777  if (si != NULL) {
778    si->ref_count++;
779  }
780  return si;
781}
782
783static int soinfo_unload(soinfo* si) {
784  if (si->ref_count == 1) {
785    TRACE("unloading '%s'", si->name);
786    si->CallDestructors();
787
788    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
789      if (d->d_tag == DT_NEEDED) {
790        const char* library_name = si->strtab + d->d_un.d_val;
791        TRACE("%s needs to unload %s", si->name, library_name);
792        soinfo_unload(find_loaded_library(library_name));
793      }
794    }
795
796    munmap(reinterpret_cast<void*>(si->base), si->size);
797    notify_gdb_of_unload(si);
798    soinfo_free(si);
799    si->ref_count = 0;
800  } else {
801    si->ref_count--;
802    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
803  }
804  return 0;
805}
806
807void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
808  snprintf(buffer, buffer_size, "%s:%s", gDefaultLdPaths[0], gDefaultLdPaths[1]);
809}
810
811void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
812  if (!get_AT_SECURE()) {
813    parse_LD_LIBRARY_PATH(ld_library_path);
814  }
815}
816
817soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
818  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL)) != 0) {
819    DL_ERR("invalid flags to dlopen: %x", flags);
820    return NULL;
821  }
822  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
823    DL_ERR("invalid extended flags to android_dlopen_ext: %x", extinfo->flags);
824    return NULL;
825  }
826  set_soinfo_pool_protection(PROT_READ | PROT_WRITE);
827  soinfo* si = find_library(name, extinfo);
828  if (si != NULL) {
829    si->CallConstructors();
830  }
831  set_soinfo_pool_protection(PROT_READ);
832  return si;
833}
834
835int do_dlclose(soinfo* si) {
836  set_soinfo_pool_protection(PROT_READ | PROT_WRITE);
837  int result = soinfo_unload(si);
838  set_soinfo_pool_protection(PROT_READ);
839  return result;
840}
841
842#if defined(USE_RELA)
843static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
844  ElfW(Sym)* s;
845  soinfo* lsi;
846
847  for (size_t idx = 0; idx < count; ++idx, ++rela) {
848    unsigned type = ELFW(R_TYPE)(rela->r_info);
849    unsigned sym = ELFW(R_SYM)(rela->r_info);
850    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
851    ElfW(Addr) sym_addr = 0;
852    const char* sym_name = NULL;
853
854    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
855    if (type == 0) { // R_*_NONE
856      continue;
857    }
858    if (sym != 0) {
859      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
860      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
861      if (s == NULL) {
862        // We only allow an undefined symbol if this is a weak reference...
863        s = &si->symtab[sym];
864        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
865          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
866          return -1;
867        }
868
869        /* IHI0044C AAELF 4.5.1.1:
870
871           Libraries are not searched to resolve weak references.
872           It is not an error for a weak reference to remain unsatisfied.
873
874           During linking, the value of an undefined weak reference is:
875           - Zero if the relocation type is absolute
876           - The address of the place if the relocation is pc-relative
877           - The address of nominal base address if the relocation
878             type is base-relative.
879         */
880
881        switch (type) {
882#if defined(__aarch64__)
883        case R_AARCH64_JUMP_SLOT:
884        case R_AARCH64_GLOB_DAT:
885        case R_AARCH64_ABS64:
886        case R_AARCH64_ABS32:
887        case R_AARCH64_ABS16:
888        case R_AARCH64_RELATIVE:
889          /*
890           * The sym_addr was initialized to be zero above, or the relocation
891           * code below does not care about value of sym_addr.
892           * No need to do anything.
893           */
894          break;
895#elif defined(__x86_64__)
896        case R_X86_64_JUMP_SLOT:
897        case R_X86_64_GLOB_DAT:
898        case R_X86_64_32:
899        case R_X86_64_RELATIVE:
900          // No need to do anything.
901          break;
902        case R_X86_64_PC32:
903          sym_addr = reloc;
904          break;
905#endif
906        default:
907          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
908          return -1;
909        }
910      } else {
911        // We got a definition.
912        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
913      }
914      count_relocation(kRelocSymbol);
915    } else {
916      s = NULL;
917    }
918
919    switch (type) {
920#if defined(__aarch64__)
921    case R_AARCH64_JUMP_SLOT:
922        count_relocation(kRelocAbsolute);
923        MARK(rela->r_offset);
924        TRACE_TYPE(RELO, "RELO JMP_SLOT %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_GLOB_DAT:
929        count_relocation(kRelocAbsolute);
930        MARK(rela->r_offset);
931        TRACE_TYPE(RELO, "RELO GLOB_DAT %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_ABS64:
936        count_relocation(kRelocAbsolute);
937        MARK(rela->r_offset);
938        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
939                   reloc, (sym_addr + rela->r_addend), sym_name);
940        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
941        break;
942    case R_AARCH64_ABS32:
943        count_relocation(kRelocAbsolute);
944        MARK(rela->r_offset);
945        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
946                   reloc, (sym_addr + rela->r_addend), sym_name);
947        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
948            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
949            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
950        } else {
951            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
952                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
953                   static_cast<ElfW(Addr)>(INT32_MIN),
954                   static_cast<ElfW(Addr)>(UINT32_MAX));
955            return -1;
956        }
957        break;
958    case R_AARCH64_ABS16:
959        count_relocation(kRelocAbsolute);
960        MARK(rela->r_offset);
961        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
962                   reloc, (sym_addr + rela->r_addend), sym_name);
963        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
964            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
965            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
966        } else {
967            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
968                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
969                   static_cast<ElfW(Addr)>(INT16_MIN),
970                   static_cast<ElfW(Addr)>(UINT16_MAX));
971            return -1;
972        }
973        break;
974    case R_AARCH64_PREL64:
975        count_relocation(kRelocRelative);
976        MARK(rela->r_offset);
977        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
978                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
979        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
980        break;
981    case R_AARCH64_PREL32:
982        count_relocation(kRelocRelative);
983        MARK(rela->r_offset);
984        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
985                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
986        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
987            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
988            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
989        } else {
990            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
991                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
992                   static_cast<ElfW(Addr)>(INT32_MIN),
993                   static_cast<ElfW(Addr)>(UINT32_MAX));
994            return -1;
995        }
996        break;
997    case R_AARCH64_PREL16:
998        count_relocation(kRelocRelative);
999        MARK(rela->r_offset);
1000        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1001                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1002        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1003            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1004            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1005        } else {
1006            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1007                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1008                   static_cast<ElfW(Addr)>(INT16_MIN),
1009                   static_cast<ElfW(Addr)>(UINT16_MAX));
1010            return -1;
1011        }
1012        break;
1013
1014    case R_AARCH64_RELATIVE:
1015        count_relocation(kRelocRelative);
1016        MARK(rela->r_offset);
1017        if (sym) {
1018            DL_ERR("odd RELATIVE form...");
1019            return -1;
1020        }
1021        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1022                   reloc, (si->base + rela->r_addend));
1023        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1024        break;
1025
1026    case R_AARCH64_COPY:
1027        if ((si->flags & FLAG_EXE) == 0) {
1028            /*
1029              * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1030              *
1031              * Section 4.7.1.10 "Dynamic relocations"
1032              * R_AARCH64_COPY may only appear in executable objects where e_type is
1033              * set to ET_EXEC.
1034              *
1035              * FLAG_EXE is set for both ET_DYN and ET_EXEC executables.
1036              * We should explicitly disallow ET_DYN executables from having
1037              * R_AARCH64_COPY relocations.
1038              */
1039            DL_ERR("%s R_AARCH64_COPY relocations only supported for ET_EXEC", si->name);
1040            return -1;
1041        }
1042        count_relocation(kRelocCopy);
1043        MARK(rela->r_offset);
1044        TRACE_TYPE(RELO, "RELO COPY %16llx <- %lld @ %16llx %s\n",
1045                   reloc,
1046                   s->st_size,
1047                   (sym_addr + rela->r_addend),
1048                   sym_name);
1049        if (reloc == (sym_addr + rela->r_addend)) {
1050            ElfW(Sym)* src = soinfo_do_lookup(NULL, sym_name, &lsi, needed);
1051
1052            if (src == NULL) {
1053                DL_ERR("%s R_AARCH64_COPY relocation source cannot be resolved", si->name);
1054                return -1;
1055            }
1056            if (lsi->has_DT_SYMBOLIC) {
1057                DL_ERR("%s invalid R_AARCH64_COPY relocation against DT_SYMBOLIC shared "
1058                       "library %s (built with -Bsymbolic?)", si->name, lsi->name);
1059                return -1;
1060            }
1061            if (s->st_size < src->st_size) {
1062                DL_ERR("%s R_AARCH64_COPY relocation size mismatch (%lld < %lld)",
1063                       si->name, s->st_size, src->st_size);
1064                return -1;
1065            }
1066            memcpy(reinterpret_cast<void*>(reloc),
1067                   reinterpret_cast<void*>(src->st_value + lsi->load_bias), src->st_size);
1068        } else {
1069            DL_ERR("%s R_AARCH64_COPY relocation target cannot be resolved", si->name);
1070            return -1;
1071        }
1072        break;
1073    case R_AARCH64_TLS_TPREL64:
1074        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1075                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1076        break;
1077    case R_AARCH64_TLS_DTPREL32:
1078        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1079                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1080        break;
1081#elif defined(__x86_64__)
1082    case R_X86_64_JUMP_SLOT:
1083      count_relocation(kRelocAbsolute);
1084      MARK(rela->r_offset);
1085      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1086                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1087      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1088      break;
1089    case R_X86_64_GLOB_DAT:
1090      count_relocation(kRelocAbsolute);
1091      MARK(rela->r_offset);
1092      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1093                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1094      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1095      break;
1096    case R_X86_64_RELATIVE:
1097      count_relocation(kRelocRelative);
1098      MARK(rela->r_offset);
1099      if (sym) {
1100        DL_ERR("odd RELATIVE form...");
1101        return -1;
1102      }
1103      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1104                 static_cast<size_t>(si->base));
1105      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1106      break;
1107    case R_X86_64_32:
1108      count_relocation(kRelocRelative);
1109      MARK(rela->r_offset);
1110      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1111                 static_cast<size_t>(sym_addr), sym_name);
1112      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1113      break;
1114    case R_X86_64_64:
1115      count_relocation(kRelocRelative);
1116      MARK(rela->r_offset);
1117      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1118                 static_cast<size_t>(sym_addr), sym_name);
1119      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1120      break;
1121    case R_X86_64_PC32:
1122      count_relocation(kRelocRelative);
1123      MARK(rela->r_offset);
1124      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1125                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1126                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1127      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1128      break;
1129#endif
1130
1131    default:
1132      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1133      return -1;
1134    }
1135  }
1136  return 0;
1137}
1138
1139#else // REL, not RELA.
1140
1141static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1142    ElfW(Sym)* s;
1143    soinfo* lsi;
1144
1145    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1146        unsigned type = ELFW(R_TYPE)(rel->r_info);
1147        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1148        unsigned sym = ELFW(R_SYM)(rel->r_info);
1149        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1150        ElfW(Addr) sym_addr = 0;
1151        const char* sym_name = NULL;
1152
1153        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1154        if (type == 0) { // R_*_NONE
1155            continue;
1156        }
1157        if (sym != 0) {
1158            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1159            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1160            if (s == NULL) {
1161                // We only allow an undefined symbol if this is a weak reference...
1162                s = &si->symtab[sym];
1163                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1164                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1165                    return -1;
1166                }
1167
1168                /* IHI0044C AAELF 4.5.1.1:
1169
1170                   Libraries are not searched to resolve weak references.
1171                   It is not an error for a weak reference to remain
1172                   unsatisfied.
1173
1174                   During linking, the value of an undefined weak reference is:
1175                   - Zero if the relocation type is absolute
1176                   - The address of the place if the relocation is pc-relative
1177                   - The address of nominal base address if the relocation
1178                     type is base-relative.
1179                  */
1180
1181                switch (type) {
1182#if defined(__arm__)
1183                case R_ARM_JUMP_SLOT:
1184                case R_ARM_GLOB_DAT:
1185                case R_ARM_ABS32:
1186                case R_ARM_RELATIVE:    /* Don't care. */
1187                    // sym_addr was initialized to be zero above or relocation
1188                    // code below does not care about value of sym_addr.
1189                    // No need to do anything.
1190                    break;
1191#elif defined(__i386__)
1192                case R_386_JMP_SLOT:
1193                case R_386_GLOB_DAT:
1194                case R_386_32:
1195                case R_386_RELATIVE:    /* Don't care. */
1196                    // sym_addr was initialized to be zero above or relocation
1197                    // code below does not care about value of sym_addr.
1198                    // No need to do anything.
1199                    break;
1200                case R_386_PC32:
1201                    sym_addr = reloc;
1202                    break;
1203#endif
1204
1205#if defined(__arm__)
1206                case R_ARM_COPY:
1207                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1208#endif
1209                default:
1210                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1211                    return -1;
1212                }
1213            } else {
1214                // We got a definition.
1215                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1216            }
1217            count_relocation(kRelocSymbol);
1218        } else {
1219            s = NULL;
1220        }
1221
1222        switch (type) {
1223#if defined(__arm__)
1224        case R_ARM_JUMP_SLOT:
1225            count_relocation(kRelocAbsolute);
1226            MARK(rel->r_offset);
1227            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1228            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1229            break;
1230        case R_ARM_GLOB_DAT:
1231            count_relocation(kRelocAbsolute);
1232            MARK(rel->r_offset);
1233            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1234            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1235            break;
1236        case R_ARM_ABS32:
1237            count_relocation(kRelocAbsolute);
1238            MARK(rel->r_offset);
1239            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1240            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1241            break;
1242        case R_ARM_REL32:
1243            count_relocation(kRelocRelative);
1244            MARK(rel->r_offset);
1245            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1246                       reloc, sym_addr, rel->r_offset, sym_name);
1247            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1248            break;
1249        case R_ARM_COPY:
1250            if ((si->flags & FLAG_EXE) == 0) {
1251                /*
1252                 * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1253                 *
1254                 * Section 4.7.1.10 "Dynamic relocations"
1255                 * R_ARM_COPY may only appear in executable objects where e_type is
1256                 * set to ET_EXEC.
1257                 *
1258                 * TODO: FLAG_EXE is set for both ET_DYN and ET_EXEC executables.
1259                 * We should explicitly disallow ET_DYN executables from having
1260                 * R_ARM_COPY relocations.
1261                 */
1262                DL_ERR("%s R_ARM_COPY relocations only supported for ET_EXEC", si->name);
1263                return -1;
1264            }
1265            count_relocation(kRelocCopy);
1266            MARK(rel->r_offset);
1267            TRACE_TYPE(RELO, "RELO %08x <- %d @ %08x %s", reloc, s->st_size, sym_addr, sym_name);
1268            if (reloc == sym_addr) {
1269                ElfW(Sym)* src = soinfo_do_lookup(NULL, sym_name, &lsi, needed);
1270
1271                if (src == NULL) {
1272                    DL_ERR("%s R_ARM_COPY relocation source cannot be resolved", si->name);
1273                    return -1;
1274                }
1275                if (lsi->has_DT_SYMBOLIC) {
1276                    DL_ERR("%s invalid R_ARM_COPY relocation against DT_SYMBOLIC shared "
1277                           "library %s (built with -Bsymbolic?)", si->name, lsi->name);
1278                    return -1;
1279                }
1280                if (s->st_size < src->st_size) {
1281                    DL_ERR("%s R_ARM_COPY relocation size mismatch (%d < %d)",
1282                           si->name, s->st_size, src->st_size);
1283                    return -1;
1284                }
1285                memcpy(reinterpret_cast<void*>(reloc),
1286                       reinterpret_cast<void*>(src->st_value + lsi->load_bias), src->st_size);
1287            } else {
1288                DL_ERR("%s R_ARM_COPY relocation target cannot be resolved", si->name);
1289                return -1;
1290            }
1291            break;
1292#elif defined(__i386__)
1293        case R_386_JMP_SLOT:
1294            count_relocation(kRelocAbsolute);
1295            MARK(rel->r_offset);
1296            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1297            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1298            break;
1299        case R_386_GLOB_DAT:
1300            count_relocation(kRelocAbsolute);
1301            MARK(rel->r_offset);
1302            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1303            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1304            break;
1305        case R_386_32:
1306            count_relocation(kRelocRelative);
1307            MARK(rel->r_offset);
1308            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1309            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1310            break;
1311        case R_386_PC32:
1312            count_relocation(kRelocRelative);
1313            MARK(rel->r_offset);
1314            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1315                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1316            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1317            break;
1318#elif defined(__mips__)
1319        case R_MIPS_REL32:
1320#if defined(__LP64__)
1321            // MIPS Elf64_Rel entries contain compound relocations
1322            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1323            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1324                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1325                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1326                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1327                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1328                return -1;
1329            }
1330#endif
1331            count_relocation(kRelocAbsolute);
1332            MARK(rel->r_offset);
1333            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1334                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1335            if (s) {
1336                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1337            } else {
1338                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1339            }
1340            break;
1341#endif
1342
1343#if defined(__arm__)
1344        case R_ARM_RELATIVE:
1345#elif defined(__i386__)
1346        case R_386_RELATIVE:
1347#endif
1348            count_relocation(kRelocRelative);
1349            MARK(rel->r_offset);
1350            if (sym) {
1351                DL_ERR("odd RELATIVE form...");
1352                return -1;
1353            }
1354            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1355                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1356            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1357            break;
1358
1359        default:
1360            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1361            return -1;
1362        }
1363    }
1364    return 0;
1365}
1366#endif
1367
1368#if defined(__mips__)
1369static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1370    ElfW(Addr)** got = si->plt_got;
1371    if (got == NULL) {
1372        return true;
1373    }
1374    unsigned local_gotno = si->mips_local_gotno;
1375    unsigned gotsym = si->mips_gotsym;
1376    unsigned symtabno = si->mips_symtabno;
1377    ElfW(Sym)* symtab = si->symtab;
1378
1379    // got[0] is the address of the lazy resolver function.
1380    // got[1] may be used for a GNU extension.
1381    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1382    // FIXME: maybe this should be in a separate routine?
1383    if ((si->flags & FLAG_LINKER) == 0) {
1384        size_t g = 0;
1385        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1386        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1387            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1388        }
1389        // Relocate the local GOT entries.
1390        for (; g < local_gotno; g++) {
1391            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1392        }
1393    }
1394
1395    // Now for the global GOT entries...
1396    ElfW(Sym)* sym = symtab + gotsym;
1397    got = si->plt_got + local_gotno;
1398    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1399        // This is an undefined reference... try to locate it.
1400        const char* sym_name = si->strtab + sym->st_name;
1401        soinfo* lsi;
1402        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1403        if (s == NULL) {
1404            // We only allow an undefined symbol if this is a weak reference.
1405            s = &symtab[g];
1406            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1407                DL_ERR("cannot locate \"%s\"...", sym_name);
1408                return false;
1409            }
1410            *got = 0;
1411        } else {
1412            // FIXME: is this sufficient?
1413            // For reference see NetBSD link loader
1414            // 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
1415            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1416        }
1417    }
1418    return true;
1419}
1420#endif
1421
1422void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1423  if (functions == NULL) {
1424    return;
1425  }
1426
1427  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1428
1429  int begin = reverse ? (count - 1) : 0;
1430  int end = reverse ? -1 : count;
1431  int step = reverse ? -1 : 1;
1432
1433  for (int i = begin; i != end; i += step) {
1434    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1435    CallFunction("function", functions[i]);
1436  }
1437
1438  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1439}
1440
1441void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1442  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1443    return;
1444  }
1445
1446  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1447  function();
1448  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1449
1450  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1451  // are still writable. This happens with our debug malloc (see http://b/7941716).
1452  set_soinfo_pool_protection(PROT_READ | PROT_WRITE);
1453}
1454
1455void soinfo::CallPreInitConstructors() {
1456  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1457  // but ignored in a shared library.
1458  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1459}
1460
1461void soinfo::CallConstructors() {
1462  if (constructors_called) {
1463    return;
1464  }
1465
1466  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1467  // protect against recursive constructor calls. One simple example of constructor recursion
1468  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1469  // 1. The program depends on libc, so libc's constructor is called here.
1470  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1471  // 3. dlopen() calls the constructors on the newly created
1472  //    soinfo for libc_malloc_debug_leak.so.
1473  // 4. The debug .so depends on libc, so CallConstructors is
1474  //    called again with the libc soinfo. If it doesn't trigger the early-
1475  //    out above, the libc constructor will be called again (recursively!).
1476  constructors_called = true;
1477
1478  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1479    // The GNU dynamic linker silently ignores these, but we warn the developer.
1480    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1481          name, preinit_array_count);
1482  }
1483
1484  if (dynamic != NULL) {
1485    for (ElfW(Dyn)* d = dynamic; d->d_tag != DT_NULL; ++d) {
1486      if (d->d_tag == DT_NEEDED) {
1487        const char* library_name = strtab + d->d_un.d_val;
1488        TRACE("\"%s\": calling constructors in DT_NEEDED \"%s\"", name, library_name);
1489        find_loaded_library(library_name)->CallConstructors();
1490      }
1491    }
1492  }
1493
1494  TRACE("\"%s\": calling constructors", name);
1495
1496  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1497  CallFunction("DT_INIT", init_func);
1498  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1499}
1500
1501void soinfo::CallDestructors() {
1502  TRACE("\"%s\": calling destructors", name);
1503
1504  // DT_FINI_ARRAY must be parsed in reverse order.
1505  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1506
1507  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1508  CallFunction("DT_FINI", fini_func);
1509}
1510
1511/* Force any of the closed stdin, stdout and stderr to be associated with
1512   /dev/null. */
1513static int nullify_closed_stdio() {
1514    int dev_null, i, status;
1515    int return_value = 0;
1516
1517    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1518    if (dev_null < 0) {
1519        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1520        return -1;
1521    }
1522    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1523
1524    /* If any of the stdio file descriptors is valid and not associated
1525       with /dev/null, dup /dev/null to it.  */
1526    for (i = 0; i < 3; i++) {
1527        /* If it is /dev/null already, we are done. */
1528        if (i == dev_null) {
1529            continue;
1530        }
1531
1532        TRACE("[ Nullifying stdio file descriptor %d]", i);
1533        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1534
1535        /* If file is opened, we are good. */
1536        if (status != -1) {
1537            continue;
1538        }
1539
1540        /* The only error we allow is that the file descriptor does not
1541           exist, in which case we dup /dev/null to it. */
1542        if (errno != EBADF) {
1543            DL_ERR("fcntl failed: %s", strerror(errno));
1544            return_value = -1;
1545            continue;
1546        }
1547
1548        /* Try dupping /dev/null to this stdio file descriptor and
1549           repeat if there is a signal.  Note that any errors in closing
1550           the stdio descriptor are lost.  */
1551        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1552        if (status < 0) {
1553            DL_ERR("dup2 failed: %s", strerror(errno));
1554            return_value = -1;
1555            continue;
1556        }
1557    }
1558
1559    /* If /dev/null is not one of the stdio file descriptors, close it. */
1560    if (dev_null > 2) {
1561        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1562        status = TEMP_FAILURE_RETRY(close(dev_null));
1563        if (status == -1) {
1564            DL_ERR("close failed: %s", strerror(errno));
1565            return_value = -1;
1566        }
1567    }
1568
1569    return return_value;
1570}
1571
1572static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1573    /* "base" might wrap around UINT32_MAX. */
1574    ElfW(Addr) base = si->load_bias;
1575    const ElfW(Phdr)* phdr = si->phdr;
1576    int phnum = si->phnum;
1577    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1578
1579    /* We can't debug anything until the linker is relocated */
1580    if (!relocating_linker) {
1581        INFO("[ linking %s ]", si->name);
1582        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1583    }
1584
1585    /* Extract dynamic section */
1586    size_t dynamic_count;
1587    ElfW(Word) dynamic_flags;
1588    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1589                                   &dynamic_count, &dynamic_flags);
1590    if (si->dynamic == NULL) {
1591        if (!relocating_linker) {
1592            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1593        }
1594        return false;
1595    } else {
1596        if (!relocating_linker) {
1597            DEBUG("dynamic = %p", si->dynamic);
1598        }
1599    }
1600
1601#if defined(__arm__)
1602    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1603                                    &si->ARM_exidx, &si->ARM_exidx_count);
1604#endif
1605
1606    // Extract useful information from dynamic section.
1607    uint32_t needed_count = 0;
1608    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1609        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1610              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1611        switch (d->d_tag) {
1612        case DT_HASH:
1613            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1614            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1615            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1616            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1617            break;
1618        case DT_STRTAB:
1619            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1620            break;
1621        case DT_SYMTAB:
1622            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1623            break;
1624#if !defined(__LP64__)
1625        case DT_PLTREL:
1626            if (d->d_un.d_val != DT_REL) {
1627                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1628                return false;
1629            }
1630            break;
1631#endif
1632        case DT_JMPREL:
1633#if defined(USE_RELA)
1634            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1635#else
1636            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1637#endif
1638            break;
1639        case DT_PLTRELSZ:
1640#if defined(USE_RELA)
1641            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1642#else
1643            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1644#endif
1645            break;
1646#if defined(__mips__)
1647        case DT_PLTGOT:
1648            // Used by mips and mips64.
1649            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1650            break;
1651#endif
1652        case DT_DEBUG:
1653            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1654            // if the dynamic table is writable
1655// FIXME: not working currently for N64
1656// The flags for the LOAD and DYNAMIC program headers do not agree.
1657// The LOAD section containng the dynamic table has been mapped as
1658// read-only, but the DYNAMIC header claims it is writable.
1659#if !(defined(__mips__) && defined(__LP64__))
1660            if ((dynamic_flags & PF_W) != 0) {
1661                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1662            }
1663            break;
1664#endif
1665#if defined(USE_RELA)
1666         case DT_RELA:
1667            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1668            break;
1669         case DT_RELASZ:
1670            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1671            break;
1672        case DT_REL:
1673            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1674            return false;
1675        case DT_RELSZ:
1676            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1677            return false;
1678#else
1679        case DT_REL:
1680            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1681            break;
1682        case DT_RELSZ:
1683            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1684            break;
1685         case DT_RELA:
1686            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1687            return false;
1688#endif
1689        case DT_INIT:
1690            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1691            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1692            break;
1693        case DT_FINI:
1694            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1695            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1696            break;
1697        case DT_INIT_ARRAY:
1698            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1699            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1700            break;
1701        case DT_INIT_ARRAYSZ:
1702            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1703            break;
1704        case DT_FINI_ARRAY:
1705            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1706            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1707            break;
1708        case DT_FINI_ARRAYSZ:
1709            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1710            break;
1711        case DT_PREINIT_ARRAY:
1712            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1713            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1714            break;
1715        case DT_PREINIT_ARRAYSZ:
1716            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1717            break;
1718        case DT_TEXTREL:
1719#if defined(__LP64__)
1720            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1721            return false;
1722#else
1723            si->has_text_relocations = true;
1724            break;
1725#endif
1726        case DT_SYMBOLIC:
1727            si->has_DT_SYMBOLIC = true;
1728            break;
1729        case DT_NEEDED:
1730            ++needed_count;
1731            break;
1732        case DT_FLAGS:
1733            if (d->d_un.d_val & DF_TEXTREL) {
1734#if defined(__LP64__)
1735                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1736                return false;
1737#else
1738                si->has_text_relocations = true;
1739#endif
1740            }
1741            if (d->d_un.d_val & DF_SYMBOLIC) {
1742                si->has_DT_SYMBOLIC = true;
1743            }
1744            break;
1745#if defined(__mips__)
1746        case DT_STRSZ:
1747        case DT_SYMENT:
1748        case DT_RELENT:
1749             break;
1750        case DT_MIPS_RLD_MAP:
1751            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1752            {
1753              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1754              *dp = &_r_debug;
1755            }
1756            break;
1757        case DT_MIPS_RLD_VERSION:
1758        case DT_MIPS_FLAGS:
1759        case DT_MIPS_BASE_ADDRESS:
1760        case DT_MIPS_UNREFEXTNO:
1761            break;
1762
1763        case DT_MIPS_SYMTABNO:
1764            si->mips_symtabno = d->d_un.d_val;
1765            break;
1766
1767        case DT_MIPS_LOCAL_GOTNO:
1768            si->mips_local_gotno = d->d_un.d_val;
1769            break;
1770
1771        case DT_MIPS_GOTSYM:
1772            si->mips_gotsym = d->d_un.d_val;
1773            break;
1774#endif
1775
1776        default:
1777            DEBUG("Unused DT entry: type %p arg %p",
1778                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1779            break;
1780        }
1781    }
1782
1783    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1784          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1785
1786    // Sanity checks.
1787    if (relocating_linker && needed_count != 0) {
1788        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1789        return false;
1790    }
1791    if (si->nbucket == 0) {
1792        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1793        return false;
1794    }
1795    if (si->strtab == 0) {
1796        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1797        return false;
1798    }
1799    if (si->symtab == 0) {
1800        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1801        return false;
1802    }
1803
1804    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1805    if (si->flags & FLAG_EXE) {
1806        memset(gLdPreloads, 0, sizeof(gLdPreloads));
1807        size_t preload_count = 0;
1808        for (size_t i = 0; gLdPreloadNames[i] != NULL; i++) {
1809            soinfo* lsi = find_library(gLdPreloadNames[i], NULL);
1810            if (lsi != NULL) {
1811                gLdPreloads[preload_count++] = lsi;
1812            } else {
1813                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1814                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1815                        gLdPreloadNames[i], si->name, linker_get_error_buffer());
1816            }
1817        }
1818    }
1819
1820    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1821    soinfo** pneeded = needed;
1822
1823    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1824        if (d->d_tag == DT_NEEDED) {
1825            const char* library_name = si->strtab + d->d_un.d_val;
1826            DEBUG("%s needs %s", si->name, library_name);
1827            soinfo* lsi = find_library(library_name, NULL);
1828            if (lsi == NULL) {
1829                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1830                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1831                       library_name, si->name, tmp_err_buf);
1832                return false;
1833            }
1834            *pneeded++ = lsi;
1835        }
1836    }
1837    *pneeded = NULL;
1838
1839#if !defined(__LP64__)
1840    if (si->has_text_relocations) {
1841        // Make segments writable to allow text relocations to work properly. We will later call
1842        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1843#if !defined(__i386__) // The platform itself has too many text relocations on x86.
1844        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1845                "security hardening. Please fix.", si->name);
1846#endif
1847        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1848            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1849                   si->name, strerror(errno));
1850            return false;
1851        }
1852    }
1853#endif
1854
1855#if defined(USE_RELA)
1856    if (si->plt_rela != NULL) {
1857        DEBUG("[ relocating %s plt ]\n", si->name);
1858        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1859            return false;
1860        }
1861    }
1862    if (si->rela != NULL) {
1863        DEBUG("[ relocating %s ]\n", si->name);
1864        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1865            return false;
1866        }
1867    }
1868#else
1869    if (si->plt_rel != NULL) {
1870        DEBUG("[ relocating %s plt ]", si->name);
1871        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1872            return false;
1873        }
1874    }
1875    if (si->rel != NULL) {
1876        DEBUG("[ relocating %s ]", si->name);
1877        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1878            return false;
1879        }
1880    }
1881#endif
1882
1883#if defined(__mips__)
1884    if (!mips_relocate_got(si, needed)) {
1885        return false;
1886    }
1887#endif
1888
1889    si->flags |= FLAG_LINKED;
1890    DEBUG("[ finished linking %s ]", si->name);
1891
1892#if !defined(__LP64__)
1893    if (si->has_text_relocations) {
1894        // All relocations are done, we can protect our segments back to read-only.
1895        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1896            DL_ERR("can't protect segments for \"%s\": %s",
1897                   si->name, strerror(errno));
1898            return false;
1899        }
1900    }
1901#endif
1902
1903    /* We can also turn on GNU RELRO protection */
1904    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1905        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1906               si->name, strerror(errno));
1907        return false;
1908    }
1909
1910    /* Handle serializing/sharing the RELRO segment */
1911    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1912      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
1913                                         extinfo->relro_fd) < 0) {
1914        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
1915               si->name, strerror(errno));
1916        return false;
1917      }
1918    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
1919      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
1920                                   extinfo->relro_fd) < 0) {
1921        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
1922               si->name, strerror(errno));
1923        return false;
1924      }
1925    }
1926
1927    notify_gdb_of_load(si);
1928    return true;
1929}
1930
1931/*
1932 * This function add vdso to internal dso list.
1933 * It helps to stack unwinding through signal handlers.
1934 * Also, it makes bionic more like glibc.
1935 */
1936static void add_vdso(KernelArgumentBlock& args __unused) {
1937#if defined(AT_SYSINFO_EHDR)
1938  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
1939  if (ehdr_vdso == NULL) {
1940    return;
1941  }
1942
1943  soinfo* si = soinfo_alloc("[vdso]");
1944
1945  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
1946  si->phnum = ehdr_vdso->e_phnum;
1947  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
1948  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
1949  si->flags = 0;
1950  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
1951
1952  soinfo_link_image(si, NULL);
1953#endif
1954}
1955
1956/*
1957 * This code is called after the linker has linked itself and
1958 * fixed it's own GOT. It is safe to make references to externs
1959 * and other non-local data at this point.
1960 */
1961static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
1962    /* NOTE: we store the args pointer on a special location
1963     *       of the temporary TLS area in order to pass it to
1964     *       the C Library's runtime initializer.
1965     *
1966     *       The initializer must clear the slot and reset the TLS
1967     *       to point to a different location to ensure that no other
1968     *       shared library constructor can access it.
1969     */
1970  __libc_init_tls(args);
1971
1972#if TIMING
1973    struct timeval t0, t1;
1974    gettimeofday(&t0, 0);
1975#endif
1976
1977    // Initialize environment functions, and get to the ELF aux vectors table.
1978    linker_env_init(args);
1979
1980    // If this is a setuid/setgid program, close the security hole described in
1981    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
1982    if (get_AT_SECURE()) {
1983        nullify_closed_stdio();
1984    }
1985
1986    debuggerd_init();
1987
1988    // Get a few environment variables.
1989    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
1990    if (LD_DEBUG != NULL) {
1991      gLdDebugVerbosity = atoi(LD_DEBUG);
1992    }
1993
1994    // Normally, these are cleaned by linker_env_init, but the test
1995    // doesn't cost us anything.
1996    const char* ldpath_env = NULL;
1997    const char* ldpreload_env = NULL;
1998    if (!get_AT_SECURE()) {
1999      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2000      ldpreload_env = linker_env_get("LD_PRELOAD");
2001    }
2002
2003    INFO("[ android linker & debugger ]");
2004
2005    soinfo* si = soinfo_alloc(args.argv[0]);
2006    if (si == NULL) {
2007        exit(EXIT_FAILURE);
2008    }
2009
2010    /* bootstrap the link map, the main exe always needs to be first */
2011    si->flags |= FLAG_EXE;
2012    link_map* map = &(si->link_map_head);
2013
2014    map->l_addr = 0;
2015    map->l_name = args.argv[0];
2016    map->l_prev = NULL;
2017    map->l_next = NULL;
2018
2019    _r_debug.r_map = map;
2020    r_debug_tail = map;
2021
2022    /* gdb expects the linker to be in the debug shared object list.
2023     * Without this, gdb has trouble locating the linker's ".text"
2024     * and ".plt" sections. Gdb could also potentially use this to
2025     * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2026     * Don't use soinfo_alloc(), because the linker shouldn't
2027     * be on the soinfo list.
2028     */
2029    {
2030        static soinfo linker_soinfo;
2031#if defined(__LP64__)
2032        strlcpy(linker_soinfo.name, "/system/bin/linker64", sizeof(linker_soinfo.name));
2033#else
2034        strlcpy(linker_soinfo.name, "/system/bin/linker", sizeof(linker_soinfo.name));
2035#endif
2036        linker_soinfo.flags = 0;
2037        linker_soinfo.base = linker_base;
2038
2039        /*
2040         * Set the dynamic field in the link map otherwise gdb will complain with
2041         * the following:
2042         *   warning: .dynamic section for "/system/bin/linker" is not at the
2043         *   expected address (wrong library or version mismatch?)
2044         */
2045        ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2046        ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2047        phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2048                                       &linker_soinfo.dynamic, NULL, NULL);
2049        insert_soinfo_into_debug_map(&linker_soinfo);
2050    }
2051
2052    // Extract information passed from the kernel.
2053    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2054    si->phnum = args.getauxval(AT_PHNUM);
2055    si->entry = args.getauxval(AT_ENTRY);
2056
2057    /* Compute the value of si->base. We can't rely on the fact that
2058     * the first entry is the PHDR because this will not be true
2059     * for certain executables (e.g. some in the NDK unit test suite)
2060     */
2061    si->base = 0;
2062    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2063    si->load_bias = 0;
2064    for (size_t i = 0; i < si->phnum; ++i) {
2065      if (si->phdr[i].p_type == PT_PHDR) {
2066        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2067        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2068        break;
2069      }
2070    }
2071    si->dynamic = NULL;
2072    si->ref_count = 1;
2073
2074    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2075    if (elf_hdr->e_type != ET_DYN) {
2076        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2077        exit(EXIT_FAILURE);
2078    }
2079
2080    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2081    parse_LD_LIBRARY_PATH(ldpath_env);
2082    parse_LD_PRELOAD(ldpreload_env);
2083
2084    somain = si;
2085
2086    if (!soinfo_link_image(si, NULL)) {
2087        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2088        exit(EXIT_FAILURE);
2089    }
2090
2091    add_vdso(args);
2092
2093    si->CallPreInitConstructors();
2094
2095    for (size_t i = 0; gLdPreloads[i] != NULL; ++i) {
2096        gLdPreloads[i]->CallConstructors();
2097    }
2098
2099    /* After the link_image, the si->load_bias is initialized.
2100     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2101     * We need to update this value for so exe here. So Unwind_Backtrace
2102     * for some arch like x86 could work correctly within so exe.
2103     */
2104    map->l_addr = si->load_bias;
2105    si->CallConstructors();
2106
2107#if TIMING
2108    gettimeofday(&t1, NULL);
2109    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2110               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2111               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2112#endif
2113#if STATS
2114    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2115           linker_stats.count[kRelocAbsolute],
2116           linker_stats.count[kRelocRelative],
2117           linker_stats.count[kRelocCopy],
2118           linker_stats.count[kRelocSymbol]);
2119#endif
2120#if COUNT_PAGES
2121    {
2122        unsigned n;
2123        unsigned i;
2124        unsigned count = 0;
2125        for (n = 0; n < 4096; n++) {
2126            if (bitmask[n]) {
2127                unsigned x = bitmask[n];
2128#if defined(__LP64__)
2129                for (i = 0; i < 32; i++) {
2130#else
2131                for (i = 0; i < 8; i++) {
2132#endif
2133                    if (x & 1) {
2134                        count++;
2135                    }
2136                    x >>= 1;
2137                }
2138            }
2139        }
2140        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2141    }
2142#endif
2143
2144#if TIMING || STATS || COUNT_PAGES
2145    fflush(stdout);
2146#endif
2147
2148    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2149    return si->entry;
2150}
2151
2152/* Compute the load-bias of an existing executable. This shall only
2153 * be used to compute the load bias of an executable or shared library
2154 * that was loaded by the kernel itself.
2155 *
2156 * Input:
2157 *    elf    -> address of ELF header, assumed to be at the start of the file.
2158 * Return:
2159 *    load bias, i.e. add the value of any p_vaddr in the file to get
2160 *    the corresponding address in memory.
2161 */
2162static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2163  ElfW(Addr) offset = elf->e_phoff;
2164  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2165  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2166
2167  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2168    if (phdr->p_type == PT_LOAD) {
2169      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2170    }
2171  }
2172  return 0;
2173}
2174
2175/*
2176 * This is the entry point for the linker, called from begin.S. This
2177 * method is responsible for fixing the linker's own relocations, and
2178 * then calling __linker_init_post_relocation().
2179 *
2180 * Because this method is called before the linker has fixed it's own
2181 * relocations, any attempt to reference an extern variable, extern
2182 * function, or other GOT reference will generate a segfault.
2183 */
2184extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2185  KernelArgumentBlock args(raw_args);
2186
2187  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2188  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2189  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2190
2191  soinfo linker_so;
2192  memset(&linker_so, 0, sizeof(soinfo));
2193
2194  strcpy(linker_so.name, "[dynamic linker]");
2195  linker_so.base = linker_addr;
2196  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2197  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2198  linker_so.dynamic = NULL;
2199  linker_so.phdr = phdr;
2200  linker_so.phnum = elf_hdr->e_phnum;
2201  linker_so.flags |= FLAG_LINKER;
2202
2203  if (!soinfo_link_image(&linker_so, NULL)) {
2204    // It would be nice to print an error message, but if the linker
2205    // can't link itself, there's no guarantee that we'll be able to
2206    // call write() (because it involves a GOT reference). We may as
2207    // well try though...
2208    const char* msg = "CANNOT LINK EXECUTABLE: ";
2209    write(2, msg, strlen(msg));
2210    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2211    write(2, "\n", 1);
2212    _exit(EXIT_FAILURE);
2213  }
2214
2215  // We have successfully fixed our own relocations. It's safe to run
2216  // the main part of the linker now.
2217  args.abort_message_ptr = &gAbortMessage;
2218  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2219
2220  set_soinfo_pool_protection(PROT_READ);
2221
2222  // Return the address that the calling assembly stub should jump to.
2223  return start_address;
2224}
2225