linker.cpp revision 422106a24d620af4be58e8d92a2e9b7b6167b72d
1/*
2 * Copyright (C) 2008, 2009 The Android Open Source Project
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *  * Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 *  * Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in
12 *    the documentation and/or other materials provided with the
13 *    distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
18 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
19 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
21 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
22 * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
23 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
24 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
25 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <dlfcn.h>
30#include <errno.h>
31#include <fcntl.h>
32#include <inttypes.h>
33#include <pthread.h>
34#include <stdio.h>
35#include <stdlib.h>
36#include <string.h>
37#include <sys/mman.h>
38#include <sys/stat.h>
39#include <unistd.h>
40
41// Private C library headers.
42#include "private/bionic_tls.h"
43#include "private/KernelArgumentBlock.h"
44#include "private/ScopedPthreadMutexLocker.h"
45#include "private/ScopedFd.h"
46
47#include "linker.h"
48#include "linker_debug.h"
49#include "linker_environ.h"
50#include "linker_phdr.h"
51#include "linker_allocator.h"
52
53/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
54 *
55 * Do NOT use malloc() and friends or pthread_*() code here.
56 * Don't use printf() either; it's caused mysterious memory
57 * corruption in the past.
58 * The linker runs before we bring up libc and it's easiest
59 * to make sure it does not depend on any complex libc features
60 *
61 * open issues / todo:
62 *
63 * - cleaner error reporting
64 * - after linking, set as much stuff as possible to READONLY
65 *   and NOEXEC
66 */
67
68#if defined(__LP64__)
69#define SEARCH_NAME(x) x
70#else
71// Nvidia drivers are relying on the bug:
72// http://code.google.com/p/android/issues/detail?id=6670
73// so we continue to use base-name lookup for lp32
74static const char* get_base_name(const char* name) {
75  const char* bname = strrchr(name, '/');
76  return bname ? bname + 1 : name;
77}
78#define SEARCH_NAME(x) get_base_name(x)
79#endif
80
81static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo);
82static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf);
83
84static LinkerAllocator<soinfo> g_soinfo_allocator;
85static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_links_allocator;
86
87static soinfo* solist;
88static soinfo* sonext;
89static soinfo* somain; /* main process, always the one after libdl_info */
90
91static const char* const kDefaultLdPaths[] = {
92#if defined(__LP64__)
93  "/vendor/lib64",
94  "/system/lib64",
95#else
96  "/vendor/lib",
97  "/system/lib",
98#endif
99  NULL
100};
101
102#define LDPATH_BUFSIZE (LDPATH_MAX*64)
103#define LDPATH_MAX 8
104
105#define LDPRELOAD_BUFSIZE (LDPRELOAD_MAX*64)
106#define LDPRELOAD_MAX 8
107
108static char g_ld_library_paths_buffer[LDPATH_BUFSIZE];
109static const char* g_ld_library_paths[LDPATH_MAX + 1];
110
111static char g_ld_preloads_buffer[LDPRELOAD_BUFSIZE];
112static const char* g_ld_preload_names[LDPRELOAD_MAX + 1];
113
114static soinfo* g_ld_preloads[LDPRELOAD_MAX + 1];
115
116__LIBC_HIDDEN__ int g_ld_debug_verbosity;
117
118__LIBC_HIDDEN__ abort_msg_t* g_abort_message = NULL; // For debuggerd.
119
120enum RelocationKind {
121    kRelocAbsolute = 0,
122    kRelocRelative,
123    kRelocCopy,
124    kRelocSymbol,
125    kRelocMax
126};
127
128enum class SymbolLookupScope {
129  kAllowLocal,
130  kExcludeLocal,
131};
132
133#if STATS
134struct linker_stats_t {
135    int count[kRelocMax];
136};
137
138static linker_stats_t linker_stats;
139
140static void count_relocation(RelocationKind kind) {
141    ++linker_stats.count[kind];
142}
143#else
144static void count_relocation(RelocationKind) {
145}
146#endif
147
148#if COUNT_PAGES
149static unsigned bitmask[4096];
150#if defined(__LP64__)
151#define MARK(offset) \
152    do { \
153        if ((((offset) >> 12) >> 5) < 4096) \
154            bitmask[((offset) >> 12) >> 5] |= (1 << (((offset) >> 12) & 31)); \
155    } while (0)
156#else
157#define MARK(offset) \
158    do { \
159        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
160    } while (0)
161#endif
162#else
163#define MARK(x) do {} while (0)
164#endif
165
166// You shouldn't try to call memory-allocating functions in the dynamic linker.
167// Guard against the most obvious ones.
168#define DISALLOW_ALLOCATION(return_type, name, ...) \
169    return_type name __VA_ARGS__ \
170    { \
171        __libc_fatal("ERROR: " #name " called from the dynamic linker!\n"); \
172    }
173DISALLOW_ALLOCATION(void*, malloc, (size_t u __unused));
174DISALLOW_ALLOCATION(void, free, (void* u __unused));
175DISALLOW_ALLOCATION(void*, realloc, (void* u1 __unused, size_t u2 __unused));
176DISALLOW_ALLOCATION(void*, calloc, (size_t u1 __unused, size_t u2 __unused));
177
178static char tmp_err_buf[768];
179static char __linker_dl_err_buf[768];
180
181char* linker_get_error_buffer() {
182  return &__linker_dl_err_buf[0];
183}
184
185size_t linker_get_error_buffer_size() {
186  return sizeof(__linker_dl_err_buf);
187}
188
189/*
190 * This function is an empty stub where GDB locates a breakpoint to get notified
191 * about linker activity.
192 */
193extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
194
195static pthread_mutex_t g__r_debug_mutex = PTHREAD_MUTEX_INITIALIZER;
196static r_debug _r_debug = {1, NULL, reinterpret_cast<uintptr_t>(&rtld_db_dlactivity), r_debug::RT_CONSISTENT, 0};
197static link_map* r_debug_tail = 0;
198
199static void insert_soinfo_into_debug_map(soinfo* info) {
200    // Copy the necessary fields into the debug structure.
201    link_map* map = &(info->link_map_head);
202    map->l_addr = info->load_bias;
203    map->l_name = reinterpret_cast<char*>(info->name);
204    map->l_ld = info->dynamic;
205
206    /* Stick the new library at the end of the list.
207     * gdb tends to care more about libc than it does
208     * about leaf libraries, and ordering it this way
209     * reduces the back-and-forth over the wire.
210     */
211    if (r_debug_tail) {
212        r_debug_tail->l_next = map;
213        map->l_prev = r_debug_tail;
214        map->l_next = 0;
215    } else {
216        _r_debug.r_map = map;
217        map->l_prev = 0;
218        map->l_next = 0;
219    }
220    r_debug_tail = map;
221}
222
223static void remove_soinfo_from_debug_map(soinfo* info) {
224    link_map* map = &(info->link_map_head);
225
226    if (r_debug_tail == map) {
227        r_debug_tail = map->l_prev;
228    }
229
230    if (map->l_prev) {
231        map->l_prev->l_next = map->l_next;
232    }
233    if (map->l_next) {
234        map->l_next->l_prev = map->l_prev;
235    }
236}
237
238static void notify_gdb_of_load(soinfo* info) {
239    if (info->flags & FLAG_EXE) {
240        // GDB already knows about the main executable
241        return;
242    }
243
244    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
245
246    _r_debug.r_state = r_debug::RT_ADD;
247    rtld_db_dlactivity();
248
249    insert_soinfo_into_debug_map(info);
250
251    _r_debug.r_state = r_debug::RT_CONSISTENT;
252    rtld_db_dlactivity();
253}
254
255static void notify_gdb_of_unload(soinfo* info) {
256    if (info->flags & FLAG_EXE) {
257        // GDB already knows about the main executable
258        return;
259    }
260
261    ScopedPthreadMutexLocker locker(&g__r_debug_mutex);
262
263    _r_debug.r_state = r_debug::RT_DELETE;
264    rtld_db_dlactivity();
265
266    remove_soinfo_from_debug_map(info);
267
268    _r_debug.r_state = r_debug::RT_CONSISTENT;
269    rtld_db_dlactivity();
270}
271
272void notify_gdb_of_libraries() {
273  _r_debug.r_state = r_debug::RT_ADD;
274  rtld_db_dlactivity();
275  _r_debug.r_state = r_debug::RT_CONSISTENT;
276  rtld_db_dlactivity();
277}
278
279LinkedListEntry<soinfo>* SoinfoListAllocator::alloc() {
280  return g_soinfo_links_allocator.alloc();
281}
282
283void SoinfoListAllocator::free(LinkedListEntry<soinfo>* entry) {
284  g_soinfo_links_allocator.free(entry);
285}
286
287static void protect_data(int protection) {
288  g_soinfo_allocator.protect_all(protection);
289  g_soinfo_links_allocator.protect_all(protection);
290}
291
292static soinfo* soinfo_alloc(const char* name, struct stat* file_stat) {
293  if (strlen(name) >= SOINFO_NAME_LEN) {
294    DL_ERR("library name \"%s\" too long", name);
295    return NULL;
296  }
297
298  soinfo* si = g_soinfo_allocator.alloc();
299
300  // Initialize the new element.
301  memset(si, 0, sizeof(soinfo));
302  strlcpy(si->name, name, sizeof(si->name));
303  si->flags = FLAG_NEW_SOINFO;
304
305  if (file_stat != NULL) {
306    si->set_st_dev(file_stat->st_dev);
307    si->set_st_ino(file_stat->st_ino);
308  }
309
310  sonext->next = si;
311  sonext = si;
312
313  TRACE("name %s: allocated soinfo @ %p", name, si);
314  return si;
315}
316
317static void soinfo_free(soinfo* si) {
318    if (si == NULL) {
319        return;
320    }
321
322    if (si->base != 0 && si->size != 0) {
323      munmap(reinterpret_cast<void*>(si->base), si->size);
324    }
325
326    soinfo *prev = NULL, *trav;
327
328    TRACE("name %s: freeing soinfo @ %p", si->name, si);
329
330    for (trav = solist; trav != NULL; trav = trav->next) {
331        if (trav == si)
332            break;
333        prev = trav;
334    }
335    if (trav == NULL) {
336        /* si was not in solist */
337        DL_ERR("name \"%s\" is not in solist!", si->name);
338        return;
339    }
340
341    // clear links to/from si
342    si->remove_all_links();
343
344    /* prev will never be NULL, because the first entry in solist is
345       always the static libdl_info.
346    */
347    prev->next = si->next;
348    if (si == sonext) {
349        sonext = prev;
350    }
351
352    g_soinfo_allocator.free(si);
353}
354
355
356static void parse_path(const char* path, const char* delimiters,
357                       const char** array, char* buf, size_t buf_size, size_t max_count) {
358  if (path == NULL) {
359    return;
360  }
361
362  size_t len = strlcpy(buf, path, buf_size);
363
364  size_t i = 0;
365  char* buf_p = buf;
366  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
367    if (*array[i] != '\0') {
368      ++i;
369    }
370  }
371
372  // Forget the last path if we had to truncate; this occurs if the 2nd to
373  // last char isn't '\0' (i.e. wasn't originally a delimiter).
374  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
375    array[i - 1] = NULL;
376  } else {
377    array[i] = NULL;
378  }
379}
380
381static void parse_LD_LIBRARY_PATH(const char* path) {
382  parse_path(path, ":", g_ld_library_paths,
383             g_ld_library_paths_buffer, sizeof(g_ld_library_paths_buffer), LDPATH_MAX);
384}
385
386static void parse_LD_PRELOAD(const char* path) {
387  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
388  parse_path(path, " :", g_ld_preload_names,
389             g_ld_preloads_buffer, sizeof(g_ld_preloads_buffer), LDPRELOAD_MAX);
390}
391
392#if defined(__arm__)
393
394/* For a given PC, find the .so that it belongs to.
395 * Returns the base address of the .ARM.exidx section
396 * for that .so, and the number of 8-byte entries
397 * in that section (via *pcount).
398 *
399 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
400 *
401 * This function is exposed via dlfcn.cpp and libdl.so.
402 */
403_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int* pcount) {
404    unsigned addr = (unsigned)pc;
405
406    for (soinfo* si = solist; si != 0; si = si->next) {
407        if ((addr >= si->base) && (addr < (si->base + si->size))) {
408            *pcount = si->ARM_exidx_count;
409            return (_Unwind_Ptr)si->ARM_exidx;
410        }
411    }
412    *pcount = 0;
413    return NULL;
414}
415
416#endif
417
418/* Here, we only have to provide a callback to iterate across all the
419 * loaded libraries. gcc_eh does the rest. */
420int dl_iterate_phdr(int (*cb)(dl_phdr_info* info, size_t size, void* data), void* data) {
421    int rv = 0;
422    for (soinfo* si = solist; si != NULL; si = si->next) {
423        dl_phdr_info dl_info;
424        dl_info.dlpi_addr = si->link_map_head.l_addr;
425        dl_info.dlpi_name = si->link_map_head.l_name;
426        dl_info.dlpi_phdr = si->phdr;
427        dl_info.dlpi_phnum = si->phnum;
428        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
429        if (rv != 0) {
430            break;
431        }
432    }
433    return rv;
434}
435
436static ElfW(Sym)* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name, const SymbolLookupScope& lookup_scope) {
437  ElfW(Sym)* symtab = si->symtab;
438  const char* strtab = si->strtab;
439
440  TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
441             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
442
443  for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
444    ElfW(Sym)* s = symtab + n;
445    if (strcmp(strtab + s->st_name, name)) continue;
446
447    switch (ELF_ST_BIND(s->st_info)) {
448      case STB_GLOBAL:
449      case STB_WEAK:
450        if (s->st_shndx == SHN_UNDEF) {
451          continue;
452        }
453
454        TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
455                 name, si->name, reinterpret_cast<void*>(s->st_value),
456                 static_cast<size_t>(s->st_size));
457        return s;
458      case STB_LOCAL:
459        if (lookup_scope != SymbolLookupScope::kAllowLocal) {
460          continue;
461        }
462        TRACE_TYPE(LOOKUP, "FOUND LOCAL %s in %s (%p) %zd",
463                name, si->name, reinterpret_cast<void*>(s->st_value),
464                static_cast<size_t>(s->st_size));
465        return s;
466      default:
467        __libc_fatal("ERROR: Unexpected ST_BIND value: %d for '%s' in '%s'",
468            ELF_ST_BIND(s->st_info), name, si->name);
469    }
470  }
471
472  TRACE_TYPE(LOOKUP, "NOT FOUND %s in %s@%p %x %zd",
473             name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
474
475
476  return NULL;
477}
478
479static unsigned elfhash(const char* _name) {
480    const unsigned char* name = reinterpret_cast<const unsigned char*>(_name);
481    unsigned h = 0, g;
482
483    while (*name) {
484        h = (h << 4) + *name++;
485        g = h & 0xf0000000;
486        h ^= g;
487        h ^= g >> 24;
488    }
489    return h;
490}
491
492static ElfW(Sym)* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
493    unsigned elf_hash = elfhash(name);
494    ElfW(Sym)* s = NULL;
495
496    if (si != NULL && somain != NULL) {
497        /*
498         * Local scope is executable scope. Just start looking into it right away
499         * for the shortcut.
500         */
501
502        if (si == somain) {
503            s = soinfo_elf_lookup(si, elf_hash, name, SymbolLookupScope::kAllowLocal);
504            if (s != NULL) {
505                *lsi = si;
506                goto done;
507            }
508        } else {
509            /* Order of symbol lookup is controlled by DT_SYMBOLIC flag */
510
511            /*
512             * If this object was built with symbolic relocations disabled, the
513             * first place to look to resolve external references is the main
514             * executable.
515             */
516
517            if (!si->has_DT_SYMBOLIC) {
518                DEBUG("%s: looking up %s in executable %s",
519                      si->name, name, somain->name);
520                s = soinfo_elf_lookup(somain, elf_hash, name, SymbolLookupScope::kExcludeLocal);
521                if (s != NULL) {
522                    *lsi = somain;
523                    goto done;
524                }
525            }
526
527            /* Look for symbols in the local scope (the object who is
528             * searching). This happens with C++ templates on x86 for some
529             * reason.
530             *
531             * Notes on weak symbols:
532             * The ELF specs are ambiguous about treatment of weak definitions in
533             * dynamic linking.  Some systems return the first definition found
534             * and some the first non-weak definition.   This is system dependent.
535             * Here we return the first definition found for simplicity.  */
536
537            s = soinfo_elf_lookup(si, elf_hash, name, SymbolLookupScope::kAllowLocal);
538            if (s != NULL) {
539                *lsi = si;
540                goto done;
541            }
542
543            /*
544             * If this object was built with -Bsymbolic and symbol is not found
545             * in the local scope, try to find the symbol in the main executable.
546             */
547
548            if (si->has_DT_SYMBOLIC) {
549                DEBUG("%s: looking up %s in executable %s after local scope",
550                      si->name, name, somain->name);
551                s = soinfo_elf_lookup(somain, elf_hash, name, SymbolLookupScope::kExcludeLocal);
552                if (s != NULL) {
553                    *lsi = somain;
554                    goto done;
555                }
556            }
557        }
558    }
559
560    /* Next, look for it in the preloads list */
561    for (int i = 0; g_ld_preloads[i] != NULL; i++) {
562        s = soinfo_elf_lookup(g_ld_preloads[i], elf_hash, name, SymbolLookupScope::kExcludeLocal);
563        if (s != NULL) {
564            *lsi = g_ld_preloads[i];
565            goto done;
566        }
567    }
568
569    for (int i = 0; needed[i] != NULL; i++) {
570        DEBUG("%s: looking up %s in %s",
571              si->name, name, needed[i]->name);
572        s = soinfo_elf_lookup(needed[i], elf_hash, name, SymbolLookupScope::kExcludeLocal);
573        if (s != NULL) {
574            *lsi = needed[i];
575            goto done;
576        }
577    }
578
579done:
580    if (s != NULL) {
581        TRACE_TYPE(LOOKUP, "si %s sym %s s->st_value = %p, "
582                   "found in %s, base = %p, load bias = %p",
583                   si->name, name, reinterpret_cast<void*>(s->st_value),
584                   (*lsi)->name, reinterpret_cast<void*>((*lsi)->base),
585                   reinterpret_cast<void*>((*lsi)->load_bias));
586        return s;
587    }
588
589    return NULL;
590}
591
592// Another soinfo list allocator to use in dlsym. We don't reuse
593// SoinfoListAllocator because it is write-protected most of the time.
594static LinkerAllocator<LinkedListEntry<soinfo>> g_soinfo_list_allocator_rw;
595class SoinfoListAllocatorRW {
596 public:
597  static LinkedListEntry<soinfo>* alloc() {
598    return g_soinfo_list_allocator_rw.alloc();
599  }
600
601  static void free(LinkedListEntry<soinfo>* ptr) {
602    g_soinfo_list_allocator_rw.free(ptr);
603  }
604};
605
606// This is used by dlsym(3).  It performs symbol lookup only within the
607// specified soinfo object and its dependencies in breadth first order.
608ElfW(Sym)* dlsym_handle_lookup(soinfo* si, soinfo** found, const char* name, soinfo* caller) {
609  LinkedList<soinfo, SoinfoListAllocatorRW> visit_list;
610  visit_list.push_back(si);
611  soinfo* current_soinfo;
612  while ((current_soinfo = visit_list.pop_front()) != nullptr) {
613    ElfW(Sym)* result = soinfo_elf_lookup(current_soinfo, elfhash(name), name,
614        caller == current_soinfo ? SymbolLookupScope::kAllowLocal : SymbolLookupScope::kExcludeLocal);
615
616    if (result != nullptr) {
617      *found = current_soinfo;
618      visit_list.clear();
619      return result;
620    }
621
622    current_soinfo->get_children().for_each([&](soinfo* child) {
623      visit_list.push_back(child);
624    });
625  }
626
627  visit_list.clear();
628  return nullptr;
629}
630
631/* This is used by dlsym(3) to performs a global symbol lookup. If the
632   start value is null (for RTLD_DEFAULT), the search starts at the
633   beginning of the global solist. Otherwise the search starts at the
634   specified soinfo (for RTLD_NEXT).
635 */
636ElfW(Sym)* dlsym_linear_lookup(const char* name, soinfo** found, soinfo* start, soinfo* caller) {
637  unsigned elf_hash = elfhash(name);
638
639  if (start == NULL) {
640    start = solist;
641  }
642
643  ElfW(Sym)* s = NULL;
644  for (soinfo* si = start; (s == NULL) && (si != NULL); si = si->next) {
645    s = soinfo_elf_lookup(si, elf_hash, name,
646        caller == si ? SymbolLookupScope::kAllowLocal : SymbolLookupScope::kExcludeLocal);
647    if (s != NULL) {
648      *found = si;
649      break;
650    }
651  }
652
653  if (s != NULL) {
654    TRACE_TYPE(LOOKUP, "%s s->st_value = %p, found->base = %p",
655               name, reinterpret_cast<void*>(s->st_value), reinterpret_cast<void*>((*found)->base));
656  }
657
658  return s;
659}
660
661soinfo* find_containing_library(const void* p) {
662  ElfW(Addr) address = reinterpret_cast<ElfW(Addr)>(p);
663  for (soinfo* si = solist; si != NULL; si = si->next) {
664    if (address >= si->base && address - si->base < si->size) {
665      return si;
666    }
667  }
668  return NULL;
669}
670
671ElfW(Sym)* dladdr_find_symbol(soinfo* si, const void* addr) {
672  ElfW(Addr) soaddr = reinterpret_cast<ElfW(Addr)>(addr) - si->base;
673
674  // Search the library's symbol table for any defined symbol which
675  // contains this address.
676  for (size_t i = 0; i < si->nchain; ++i) {
677    ElfW(Sym)* sym = &si->symtab[i];
678    if (sym->st_shndx != SHN_UNDEF &&
679        soaddr >= sym->st_value &&
680        soaddr < sym->st_value + sym->st_size) {
681      return sym;
682    }
683  }
684
685  return NULL;
686}
687
688static int open_library_on_path(const char* name, const char* const paths[]) {
689  char buf[512];
690  for (size_t i = 0; paths[i] != NULL; ++i) {
691    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
692    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
693      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
694      continue;
695    }
696    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
697    if (fd != -1) {
698      return fd;
699    }
700  }
701  return -1;
702}
703
704static int open_library(const char* name) {
705  TRACE("[ opening %s ]", name);
706
707  // If the name contains a slash, we should attempt to open it directly and not search the paths.
708  if (strchr(name, '/') != NULL) {
709    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
710    if (fd != -1) {
711      return fd;
712    }
713    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
714#if defined(__LP64__)
715    return -1;
716#endif
717  }
718
719  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
720  int fd = open_library_on_path(name, g_ld_library_paths);
721  if (fd == -1) {
722    fd = open_library_on_path(name, kDefaultLdPaths);
723  }
724  return fd;
725}
726
727static soinfo* load_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
728    int fd = -1;
729    ScopedFd file_guard(-1);
730
731    if (extinfo != NULL && (extinfo->flags & ANDROID_DLEXT_USE_LIBRARY_FD) != 0) {
732      fd = extinfo->library_fd;
733    } else {
734      // Open the file.
735      fd = open_library(name);
736      if (fd == -1) {
737        DL_ERR("library \"%s\" not found", name);
738        return NULL;
739      }
740
741      file_guard.reset(fd);
742    }
743
744    ElfReader elf_reader(name, fd);
745
746    struct stat file_stat;
747    if (TEMP_FAILURE_RETRY(fstat(fd, &file_stat)) != 0) {
748      DL_ERR("unable to stat file for the library %s: %s", name, strerror(errno));
749      return NULL;
750    }
751
752    // Check for symlink and other situations where
753    // file can have different names.
754    for (soinfo* si = solist; si != NULL; si = si->next) {
755      if (si->get_st_dev() != 0 &&
756          si->get_st_ino() != 0 &&
757          si->get_st_dev() == file_stat.st_dev &&
758          si->get_st_ino() == file_stat.st_ino) {
759        TRACE("library \"%s\" is already loaded under different name/path \"%s\" - will return existing soinfo", name, si->name);
760        return si;
761      }
762    }
763
764    if ((dlflags & RTLD_NOLOAD) != 0) {
765      return NULL;
766    }
767
768    // Read the ELF header and load the segments.
769    if (!elf_reader.Load(extinfo)) {
770        return NULL;
771    }
772
773    soinfo* si = soinfo_alloc(SEARCH_NAME(name), &file_stat);
774    if (si == NULL) {
775        return NULL;
776    }
777    si->base = elf_reader.load_start();
778    si->size = elf_reader.load_size();
779    si->load_bias = elf_reader.load_bias();
780    si->phnum = elf_reader.phdr_count();
781    si->phdr = elf_reader.loaded_phdr();
782
783    // At this point we know that whatever is loaded @ base is a valid ELF
784    // shared library whose segments are properly mapped in.
785    TRACE("[ load_library base=%p size=%zu name='%s' ]",
786          reinterpret_cast<void*>(si->base), si->size, si->name);
787
788    if (!soinfo_link_image(si, extinfo)) {
789      soinfo_free(si);
790      return NULL;
791    }
792
793    return si;
794}
795
796static soinfo *find_loaded_library_by_name(const char* name) {
797  const char* search_name = SEARCH_NAME(name);
798  for (soinfo* si = solist; si != NULL; si = si->next) {
799    if (!strcmp(search_name, si->name)) {
800      return si;
801    }
802  }
803  return NULL;
804}
805
806static soinfo* find_library_internal(const char* name, int dlflags, const android_dlextinfo* extinfo) {
807  if (name == NULL) {
808    return somain;
809  }
810
811  soinfo* si = find_loaded_library_by_name(name);
812
813  // Library might still be loaded, the accurate detection
814  // of this fact is done by load_library
815  if (si == NULL) {
816    TRACE("[ '%s' has not been found by name.  Trying harder...]", name);
817    si = load_library(name, dlflags, extinfo);
818  }
819
820  if (si != NULL && (si->flags & FLAG_LINKED) == 0) {
821    DL_ERR("recursive link to \"%s\"", si->name);
822    return NULL;
823  }
824
825  return si;
826}
827
828static soinfo* find_library(const char* name, int dlflags, const android_dlextinfo* extinfo) {
829  soinfo* si = find_library_internal(name, dlflags, extinfo);
830  if (si != NULL) {
831    si->ref_count++;
832  }
833  return si;
834}
835
836static void soinfo_unload(soinfo* si) {
837  if (si->ref_count == 1) {
838    TRACE("unloading '%s'", si->name);
839    si->CallDestructors();
840
841    if ((si->flags | FLAG_NEW_SOINFO) != 0) {
842      si->get_children().for_each([&] (soinfo* child) {
843        TRACE("%s needs to unload %s", si->name, child->name);
844        soinfo_unload(child);
845      });
846    } else {
847      for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
848        if (d->d_tag == DT_NEEDED) {
849          const char* library_name = si->strtab + d->d_un.d_val;
850          TRACE("%s needs to unload %s", si->name, library_name);
851          soinfo* needed = find_library(library_name, RTLD_NOLOAD, NULL);
852          if (needed != NULL) {
853            soinfo_unload(needed);
854          } else {
855            // Not found: for example if symlink was deleted between dlopen and dlclose
856            // Since we cannot really handle errors at this point - print and continue.
857            PRINT("warning: couldn't find %s needed by %s on unload.", library_name, si->name);
858          }
859        }
860      }
861    }
862
863    notify_gdb_of_unload(si);
864    si->ref_count = 0;
865    soinfo_free(si);
866  } else {
867    si->ref_count--;
868    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
869  }
870}
871
872void do_android_get_LD_LIBRARY_PATH(char* buffer, size_t buffer_size) {
873  snprintf(buffer, buffer_size, "%s:%s", kDefaultLdPaths[0], kDefaultLdPaths[1]);
874}
875
876void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
877  if (!get_AT_SECURE()) {
878    parse_LD_LIBRARY_PATH(ld_library_path);
879  }
880}
881
882soinfo* do_dlopen(const char* name, int flags, const android_dlextinfo* extinfo) {
883  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL|RTLD_NOLOAD)) != 0) {
884    DL_ERR("invalid flags to dlopen: %x", flags);
885    return NULL;
886  }
887  if (extinfo != NULL && ((extinfo->flags & ~(ANDROID_DLEXT_VALID_FLAG_BITS)) != 0)) {
888    DL_ERR("invalid extended flags to android_dlopen_ext: %" PRIx64, extinfo->flags);
889    return NULL;
890  }
891  protect_data(PROT_READ | PROT_WRITE);
892  soinfo* si = find_library(name, flags, extinfo);
893  if (si != NULL) {
894    si->CallConstructors();
895  }
896  protect_data(PROT_READ);
897  return si;
898}
899
900void do_dlclose(soinfo* si) {
901  protect_data(PROT_READ | PROT_WRITE);
902  soinfo_unload(si);
903  protect_data(PROT_READ);
904}
905
906#if defined(USE_RELA)
907static int soinfo_relocate(soinfo* si, ElfW(Rela)* rela, unsigned count, soinfo* needed[]) {
908  ElfW(Sym)* s;
909  soinfo* lsi;
910
911  for (size_t idx = 0; idx < count; ++idx, ++rela) {
912    unsigned type = ELFW(R_TYPE)(rela->r_info);
913    unsigned sym = ELFW(R_SYM)(rela->r_info);
914    ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rela->r_offset + si->load_bias);
915    ElfW(Addr) sym_addr = 0;
916    const char* sym_name = NULL;
917
918    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
919    if (type == 0) { // R_*_NONE
920      continue;
921    }
922    if (sym != 0) {
923      sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
924      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
925      if (s == NULL) {
926        // We only allow an undefined symbol if this is a weak reference...
927        s = &si->symtab[sym];
928        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
929          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
930          return -1;
931        }
932
933        /* IHI0044C AAELF 4.5.1.1:
934
935           Libraries are not searched to resolve weak references.
936           It is not an error for a weak reference to remain unsatisfied.
937
938           During linking, the value of an undefined weak reference is:
939           - Zero if the relocation type is absolute
940           - The address of the place if the relocation is pc-relative
941           - The address of nominal base address if the relocation
942             type is base-relative.
943         */
944
945        switch (type) {
946#if defined(__aarch64__)
947        case R_AARCH64_JUMP_SLOT:
948        case R_AARCH64_GLOB_DAT:
949        case R_AARCH64_ABS64:
950        case R_AARCH64_ABS32:
951        case R_AARCH64_ABS16:
952        case R_AARCH64_RELATIVE:
953          /*
954           * The sym_addr was initialized to be zero above, or the relocation
955           * code below does not care about value of sym_addr.
956           * No need to do anything.
957           */
958          break;
959#elif defined(__x86_64__)
960        case R_X86_64_JUMP_SLOT:
961        case R_X86_64_GLOB_DAT:
962        case R_X86_64_32:
963        case R_X86_64_64:
964        case R_X86_64_RELATIVE:
965          // No need to do anything.
966          break;
967        case R_X86_64_PC32:
968          sym_addr = reloc;
969          break;
970#endif
971        default:
972          DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rela, idx);
973          return -1;
974        }
975      } else {
976        // We got a definition.
977        sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
978      }
979      count_relocation(kRelocSymbol);
980    } else {
981      s = NULL;
982    }
983
984    switch (type) {
985#if defined(__aarch64__)
986    case R_AARCH64_JUMP_SLOT:
987        count_relocation(kRelocAbsolute);
988        MARK(rela->r_offset);
989        TRACE_TYPE(RELO, "RELO JMP_SLOT %16llx <- %16llx %s\n",
990                   reloc, (sym_addr + rela->r_addend), sym_name);
991        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
992        break;
993    case R_AARCH64_GLOB_DAT:
994        count_relocation(kRelocAbsolute);
995        MARK(rela->r_offset);
996        TRACE_TYPE(RELO, "RELO GLOB_DAT %16llx <- %16llx %s\n",
997                   reloc, (sym_addr + rela->r_addend), sym_name);
998        *reinterpret_cast<ElfW(Addr)*>(reloc) = (sym_addr + rela->r_addend);
999        break;
1000    case R_AARCH64_ABS64:
1001        count_relocation(kRelocAbsolute);
1002        MARK(rela->r_offset);
1003        TRACE_TYPE(RELO, "RELO ABS64 %16llx <- %16llx %s\n",
1004                   reloc, (sym_addr + rela->r_addend), sym_name);
1005        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1006        break;
1007    case R_AARCH64_ABS32:
1008        count_relocation(kRelocAbsolute);
1009        MARK(rela->r_offset);
1010        TRACE_TYPE(RELO, "RELO ABS32 %16llx <- %16llx %s\n",
1011                   reloc, (sym_addr + rela->r_addend), sym_name);
1012        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1013            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1014            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1015        } else {
1016            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1017                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
1018                   static_cast<ElfW(Addr)>(INT32_MIN),
1019                   static_cast<ElfW(Addr)>(UINT32_MAX));
1020            return -1;
1021        }
1022        break;
1023    case R_AARCH64_ABS16:
1024        count_relocation(kRelocAbsolute);
1025        MARK(rela->r_offset);
1026        TRACE_TYPE(RELO, "RELO ABS16 %16llx <- %16llx %s\n",
1027                   reloc, (sym_addr + rela->r_addend), sym_name);
1028        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend))) &&
1029            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1030            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend);
1031        } else {
1032            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1033                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + (sym_addr + rela->r_addend)),
1034                   static_cast<ElfW(Addr)>(INT16_MIN),
1035                   static_cast<ElfW(Addr)>(UINT16_MAX));
1036            return -1;
1037        }
1038        break;
1039    case R_AARCH64_PREL64:
1040        count_relocation(kRelocRelative);
1041        MARK(rela->r_offset);
1042        TRACE_TYPE(RELO, "RELO REL64 %16llx <- %16llx - %16llx %s\n",
1043                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1044        *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr + rela->r_addend) - rela->r_offset;
1045        break;
1046    case R_AARCH64_PREL32:
1047        count_relocation(kRelocRelative);
1048        MARK(rela->r_offset);
1049        TRACE_TYPE(RELO, "RELO REL32 %16llx <- %16llx - %16llx %s\n",
1050                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1051        if ((static_cast<ElfW(Addr)>(INT32_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1052            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT32_MAX))) {
1053            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1054        } else {
1055            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1056                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1057                   static_cast<ElfW(Addr)>(INT32_MIN),
1058                   static_cast<ElfW(Addr)>(UINT32_MAX));
1059            return -1;
1060        }
1061        break;
1062    case R_AARCH64_PREL16:
1063        count_relocation(kRelocRelative);
1064        MARK(rela->r_offset);
1065        TRACE_TYPE(RELO, "RELO REL16 %16llx <- %16llx - %16llx %s\n",
1066                   reloc, (sym_addr + rela->r_addend), rela->r_offset, sym_name);
1067        if ((static_cast<ElfW(Addr)>(INT16_MIN) <= (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset))) &&
1068            ((*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)) <= static_cast<ElfW(Addr)>(UINT16_MAX))) {
1069            *reinterpret_cast<ElfW(Addr)*>(reloc) += ((sym_addr + rela->r_addend) - rela->r_offset);
1070        } else {
1071            DL_ERR("0x%016llx out of range 0x%016llx to 0x%016llx",
1072                   (*reinterpret_cast<ElfW(Addr)*>(reloc) + ((sym_addr + rela->r_addend) - rela->r_offset)),
1073                   static_cast<ElfW(Addr)>(INT16_MIN),
1074                   static_cast<ElfW(Addr)>(UINT16_MAX));
1075            return -1;
1076        }
1077        break;
1078
1079    case R_AARCH64_RELATIVE:
1080        count_relocation(kRelocRelative);
1081        MARK(rela->r_offset);
1082        if (sym) {
1083            DL_ERR("odd RELATIVE form...");
1084            return -1;
1085        }
1086        TRACE_TYPE(RELO, "RELO RELATIVE %16llx <- %16llx\n",
1087                   reloc, (si->base + rela->r_addend));
1088        *reinterpret_cast<ElfW(Addr)*>(reloc) = (si->base + rela->r_addend);
1089        break;
1090
1091    case R_AARCH64_COPY:
1092        /*
1093         * ET_EXEC is not supported so this should not happen.
1094         *
1095         * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1096         *
1097         * Section 4.7.1.10 "Dynamic relocations"
1098         * R_AARCH64_COPY may only appear in executable objects where e_type is
1099         * set to ET_EXEC.
1100         */
1101        DL_ERR("%s R_AARCH64_COPY relocations are not supported", si->name);
1102        return -1;
1103    case R_AARCH64_TLS_TPREL64:
1104        TRACE_TYPE(RELO, "RELO TLS_TPREL64 *** %16llx <- %16llx - %16llx\n",
1105                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1106        break;
1107    case R_AARCH64_TLS_DTPREL32:
1108        TRACE_TYPE(RELO, "RELO TLS_DTPREL32 *** %16llx <- %16llx - %16llx\n",
1109                   reloc, (sym_addr + rela->r_addend), rela->r_offset);
1110        break;
1111#elif defined(__x86_64__)
1112    case R_X86_64_JUMP_SLOT:
1113      count_relocation(kRelocAbsolute);
1114      MARK(rela->r_offset);
1115      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1116                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1117      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1118      break;
1119    case R_X86_64_GLOB_DAT:
1120      count_relocation(kRelocAbsolute);
1121      MARK(rela->r_offset);
1122      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
1123                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
1124      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1125      break;
1126    case R_X86_64_RELATIVE:
1127      count_relocation(kRelocRelative);
1128      MARK(rela->r_offset);
1129      if (sym) {
1130        DL_ERR("odd RELATIVE form...");
1131        return -1;
1132      }
1133      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
1134                 static_cast<size_t>(si->base));
1135      *reinterpret_cast<ElfW(Addr)*>(reloc) = si->base + rela->r_addend;
1136      break;
1137    case R_X86_64_32:
1138      count_relocation(kRelocRelative);
1139      MARK(rela->r_offset);
1140      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1141                 static_cast<size_t>(sym_addr), sym_name);
1142      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1143      break;
1144    case R_X86_64_64:
1145      count_relocation(kRelocRelative);
1146      MARK(rela->r_offset);
1147      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
1148                 static_cast<size_t>(sym_addr), sym_name);
1149      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend;
1150      break;
1151    case R_X86_64_PC32:
1152      count_relocation(kRelocRelative);
1153      MARK(rela->r_offset);
1154      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
1155                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
1156                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
1157      *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr + rela->r_addend - reloc;
1158      break;
1159#endif
1160
1161    default:
1162      DL_ERR("unknown reloc type %d @ %p (%zu)", type, rela, idx);
1163      return -1;
1164    }
1165  }
1166  return 0;
1167}
1168
1169#else // REL, not RELA.
1170
1171static int soinfo_relocate(soinfo* si, ElfW(Rel)* rel, unsigned count, soinfo* needed[]) {
1172    ElfW(Sym)* s;
1173    soinfo* lsi;
1174
1175    for (size_t idx = 0; idx < count; ++idx, ++rel) {
1176        unsigned type = ELFW(R_TYPE)(rel->r_info);
1177        // TODO: don't use unsigned for 'sym'. Use uint32_t or ElfW(Addr) instead.
1178        unsigned sym = ELFW(R_SYM)(rel->r_info);
1179        ElfW(Addr) reloc = static_cast<ElfW(Addr)>(rel->r_offset + si->load_bias);
1180        ElfW(Addr) sym_addr = 0;
1181        const char* sym_name = NULL;
1182
1183        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
1184        if (type == 0) { // R_*_NONE
1185            continue;
1186        }
1187        if (sym != 0) {
1188            sym_name = reinterpret_cast<const char*>(si->strtab + si->symtab[sym].st_name);
1189            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1190            if (s == NULL) {
1191                // We only allow an undefined symbol if this is a weak reference...
1192                s = &si->symtab[sym];
1193                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1194                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1195                    return -1;
1196                }
1197
1198                /* IHI0044C AAELF 4.5.1.1:
1199
1200                   Libraries are not searched to resolve weak references.
1201                   It is not an error for a weak reference to remain
1202                   unsatisfied.
1203
1204                   During linking, the value of an undefined weak reference is:
1205                   - Zero if the relocation type is absolute
1206                   - The address of the place if the relocation is pc-relative
1207                   - The address of nominal base address if the relocation
1208                     type is base-relative.
1209                  */
1210
1211                switch (type) {
1212#if defined(__arm__)
1213                case R_ARM_JUMP_SLOT:
1214                case R_ARM_GLOB_DAT:
1215                case R_ARM_ABS32:
1216                case R_ARM_RELATIVE:    /* Don't care. */
1217                    // sym_addr was initialized to be zero above or relocation
1218                    // code below does not care about value of sym_addr.
1219                    // No need to do anything.
1220                    break;
1221#elif defined(__i386__)
1222                case R_386_JMP_SLOT:
1223                case R_386_GLOB_DAT:
1224                case R_386_32:
1225                case R_386_RELATIVE:    /* Don't care. */
1226                    // sym_addr was initialized to be zero above or relocation
1227                    // code below does not care about value of sym_addr.
1228                    // No need to do anything.
1229                    break;
1230                case R_386_PC32:
1231                    sym_addr = reloc;
1232                    break;
1233#endif
1234
1235#if defined(__arm__)
1236                case R_ARM_COPY:
1237                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1238#endif
1239                default:
1240                    DL_ERR("unknown weak reloc type %d @ %p (%zu)", type, rel, idx);
1241                    return -1;
1242                }
1243            } else {
1244                // We got a definition.
1245                sym_addr = static_cast<ElfW(Addr)>(s->st_value + lsi->load_bias);
1246            }
1247            count_relocation(kRelocSymbol);
1248        } else {
1249            s = NULL;
1250        }
1251
1252        switch (type) {
1253#if defined(__arm__)
1254        case R_ARM_JUMP_SLOT:
1255            count_relocation(kRelocAbsolute);
1256            MARK(rel->r_offset);
1257            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1258            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1259            break;
1260        case R_ARM_GLOB_DAT:
1261            count_relocation(kRelocAbsolute);
1262            MARK(rel->r_offset);
1263            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1264            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1265            break;
1266        case R_ARM_ABS32:
1267            count_relocation(kRelocAbsolute);
1268            MARK(rel->r_offset);
1269            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1270            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1271            break;
1272        case R_ARM_REL32:
1273            count_relocation(kRelocRelative);
1274            MARK(rel->r_offset);
1275            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1276                       reloc, sym_addr, rel->r_offset, sym_name);
1277            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr - rel->r_offset;
1278            break;
1279        case R_ARM_COPY:
1280            /*
1281             * ET_EXEC is not supported so this should not happen.
1282             *
1283             * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1284             *
1285             * Section 4.7.1.10 "Dynamic relocations"
1286             * R_ARM_COPY may only appear in executable objects where e_type is
1287             * set to ET_EXEC.
1288             */
1289            DL_ERR("%s R_ARM_COPY relocations are not supported", si->name);
1290            return -1;
1291#elif defined(__i386__)
1292        case R_386_JMP_SLOT:
1293            count_relocation(kRelocAbsolute);
1294            MARK(rel->r_offset);
1295            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1296            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1297            break;
1298        case R_386_GLOB_DAT:
1299            count_relocation(kRelocAbsolute);
1300            MARK(rel->r_offset);
1301            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1302            *reinterpret_cast<ElfW(Addr)*>(reloc) = sym_addr;
1303            break;
1304        case R_386_32:
1305            count_relocation(kRelocRelative);
1306            MARK(rel->r_offset);
1307            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1308            *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1309            break;
1310        case R_386_PC32:
1311            count_relocation(kRelocRelative);
1312            MARK(rel->r_offset);
1313            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1314                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1315            *reinterpret_cast<ElfW(Addr)*>(reloc) += (sym_addr - reloc);
1316            break;
1317#elif defined(__mips__)
1318        case R_MIPS_REL32:
1319#if defined(__LP64__)
1320            // MIPS Elf64_Rel entries contain compound relocations
1321            // We only handle the R_MIPS_NONE|R_MIPS_64|R_MIPS_REL32 case
1322            if (ELF64_R_TYPE2(rel->r_info) != R_MIPS_64 ||
1323                ELF64_R_TYPE3(rel->r_info) != R_MIPS_NONE) {
1324                DL_ERR("Unexpected compound relocation type:%d type2:%d type3:%d @ %p (%zu)",
1325                       type, (unsigned)ELF64_R_TYPE2(rel->r_info),
1326                       (unsigned)ELF64_R_TYPE3(rel->r_info), rel, idx);
1327                return -1;
1328            }
1329#endif
1330            count_relocation(kRelocAbsolute);
1331            MARK(rel->r_offset);
1332            TRACE_TYPE(RELO, "RELO REL32 %08zx <- %08zx %s", static_cast<size_t>(reloc),
1333                       static_cast<size_t>(sym_addr), sym_name ? sym_name : "*SECTIONHDR*");
1334            if (s) {
1335                *reinterpret_cast<ElfW(Addr)*>(reloc) += sym_addr;
1336            } else {
1337                *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1338            }
1339            break;
1340#endif
1341
1342#if defined(__arm__)
1343        case R_ARM_RELATIVE:
1344#elif defined(__i386__)
1345        case R_386_RELATIVE:
1346#endif
1347            count_relocation(kRelocRelative);
1348            MARK(rel->r_offset);
1349            if (sym) {
1350                DL_ERR("odd RELATIVE form...");
1351                return -1;
1352            }
1353            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1354                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1355            *reinterpret_cast<ElfW(Addr)*>(reloc) += si->base;
1356            break;
1357
1358        default:
1359            DL_ERR("unknown reloc type %d @ %p (%zu)", type, rel, idx);
1360            return -1;
1361        }
1362    }
1363    return 0;
1364}
1365#endif
1366
1367#if defined(__mips__)
1368static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1369    ElfW(Addr)** got = si->plt_got;
1370    if (got == NULL) {
1371        return true;
1372    }
1373    unsigned local_gotno = si->mips_local_gotno;
1374    unsigned gotsym = si->mips_gotsym;
1375    unsigned symtabno = si->mips_symtabno;
1376    ElfW(Sym)* symtab = si->symtab;
1377
1378    // got[0] is the address of the lazy resolver function.
1379    // got[1] may be used for a GNU extension.
1380    // Set it to a recognizable address in case someone calls it (should be _rtld_bind_start).
1381    // FIXME: maybe this should be in a separate routine?
1382    if ((si->flags & FLAG_LINKER) == 0) {
1383        size_t g = 0;
1384        got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadbeef);
1385        if (reinterpret_cast<intptr_t>(got[g]) < 0) {
1386            got[g++] = reinterpret_cast<ElfW(Addr)*>(0xdeadfeed);
1387        }
1388        // Relocate the local GOT entries.
1389        for (; g < local_gotno; g++) {
1390            got[g] = reinterpret_cast<ElfW(Addr)*>(reinterpret_cast<uintptr_t>(got[g]) + si->load_bias);
1391        }
1392    }
1393
1394    // Now for the global GOT entries...
1395    ElfW(Sym)* sym = symtab + gotsym;
1396    got = si->plt_got + local_gotno;
1397    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1398        // This is an undefined reference... try to locate it.
1399        const char* sym_name = si->strtab + sym->st_name;
1400        soinfo* lsi;
1401        ElfW(Sym)* s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1402        if (s == NULL) {
1403            // We only allow an undefined symbol if this is a weak reference.
1404            s = &symtab[g];
1405            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1406                DL_ERR("cannot locate \"%s\"...", sym_name);
1407                return false;
1408            }
1409            *got = 0;
1410        } else {
1411            // FIXME: is this sufficient?
1412            // For reference see NetBSD link loader
1413            // 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
1414            *got = reinterpret_cast<ElfW(Addr)*>(lsi->load_bias + s->st_value);
1415        }
1416    }
1417    return true;
1418}
1419#endif
1420
1421void soinfo::CallArray(const char* array_name __unused, linker_function_t* functions, size_t count, bool reverse) {
1422  if (functions == NULL) {
1423    return;
1424  }
1425
1426  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1427
1428  int begin = reverse ? (count - 1) : 0;
1429  int end = reverse ? -1 : count;
1430  int step = reverse ? -1 : 1;
1431
1432  for (int i = begin; i != end; i += step) {
1433    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1434    CallFunction("function", functions[i]);
1435  }
1436
1437  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1438}
1439
1440void soinfo::CallFunction(const char* function_name __unused, linker_function_t function) {
1441  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1442    return;
1443  }
1444
1445  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1446  function();
1447  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1448
1449  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1450  // are still writable. This happens with our debug malloc (see http://b/7941716).
1451  protect_data(PROT_READ | PROT_WRITE);
1452}
1453
1454void soinfo::CallPreInitConstructors() {
1455  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1456  // but ignored in a shared library.
1457  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1458}
1459
1460void soinfo::CallConstructors() {
1461  if (constructors_called) {
1462    return;
1463  }
1464
1465  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1466  // protect against recursive constructor calls. One simple example of constructor recursion
1467  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1468  // 1. The program depends on libc, so libc's constructor is called here.
1469  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1470  // 3. dlopen() calls the constructors on the newly created
1471  //    soinfo for libc_malloc_debug_leak.so.
1472  // 4. The debug .so depends on libc, so CallConstructors is
1473  //    called again with the libc soinfo. If it doesn't trigger the early-
1474  //    out above, the libc constructor will be called again (recursively!).
1475  constructors_called = true;
1476
1477  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1478    // The GNU dynamic linker silently ignores these, but we warn the developer.
1479    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1480          name, preinit_array_count);
1481  }
1482
1483  get_children().for_each([] (soinfo* si) {
1484    si->CallConstructors();
1485  });
1486
1487  TRACE("\"%s\": calling constructors", name);
1488
1489  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1490  CallFunction("DT_INIT", init_func);
1491  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1492}
1493
1494void soinfo::CallDestructors() {
1495  TRACE("\"%s\": calling destructors", name);
1496
1497  // DT_FINI_ARRAY must be parsed in reverse order.
1498  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1499
1500  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1501  CallFunction("DT_FINI", fini_func);
1502
1503  // This is needed on second call to dlopen
1504  // after library has been unloaded with RTLD_NODELETE
1505  constructors_called = false;
1506}
1507
1508void soinfo::add_child(soinfo* child) {
1509  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1510    return;
1511  }
1512
1513  this->children.push_front(child);
1514  child->parents.push_front(this);
1515}
1516
1517void soinfo::remove_all_links() {
1518  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1519    return;
1520  }
1521
1522  // 1. Untie connected soinfos from 'this'.
1523  children.for_each([&] (soinfo* child) {
1524    child->parents.remove_if([&] (const soinfo* parent) {
1525      return parent == this;
1526    });
1527  });
1528
1529  parents.for_each([&] (soinfo* parent) {
1530    parent->children.for_each([&] (const soinfo* child) {
1531      return child == this;
1532    });
1533  });
1534
1535  // 2. Once everything untied - clear local lists.
1536  parents.clear();
1537  children.clear();
1538}
1539
1540void soinfo::set_st_dev(dev_t dev) {
1541  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1542    return;
1543  }
1544
1545  st_dev = dev;
1546}
1547
1548void soinfo::set_st_ino(ino_t ino) {
1549  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1550    return;
1551  }
1552
1553  st_ino = ino;
1554}
1555
1556dev_t soinfo::get_st_dev() {
1557  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1558    return 0;
1559  }
1560
1561  return st_dev;
1562};
1563
1564ino_t soinfo::get_st_ino() {
1565  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1566    return 0;
1567  }
1568
1569  return st_ino;
1570}
1571
1572// This is a return on get_children() in case
1573// 'this->flags' does not have FLAG_NEW_SOINFO set.
1574static soinfo::soinfo_list_t g_empty_list;
1575
1576soinfo::soinfo_list_t& soinfo::get_children() {
1577  if ((this->flags & FLAG_NEW_SOINFO) == 0) {
1578    return g_empty_list;
1579  }
1580
1581  return this->children;
1582}
1583
1584/* Force any of the closed stdin, stdout and stderr to be associated with
1585   /dev/null. */
1586static int nullify_closed_stdio() {
1587    int dev_null, i, status;
1588    int return_value = 0;
1589
1590    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1591    if (dev_null < 0) {
1592        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1593        return -1;
1594    }
1595    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1596
1597    /* If any of the stdio file descriptors is valid and not associated
1598       with /dev/null, dup /dev/null to it.  */
1599    for (i = 0; i < 3; i++) {
1600        /* If it is /dev/null already, we are done. */
1601        if (i == dev_null) {
1602            continue;
1603        }
1604
1605        TRACE("[ Nullifying stdio file descriptor %d]", i);
1606        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1607
1608        /* If file is opened, we are good. */
1609        if (status != -1) {
1610            continue;
1611        }
1612
1613        /* The only error we allow is that the file descriptor does not
1614           exist, in which case we dup /dev/null to it. */
1615        if (errno != EBADF) {
1616            DL_ERR("fcntl failed: %s", strerror(errno));
1617            return_value = -1;
1618            continue;
1619        }
1620
1621        /* Try dupping /dev/null to this stdio file descriptor and
1622           repeat if there is a signal.  Note that any errors in closing
1623           the stdio descriptor are lost.  */
1624        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1625        if (status < 0) {
1626            DL_ERR("dup2 failed: %s", strerror(errno));
1627            return_value = -1;
1628            continue;
1629        }
1630    }
1631
1632    /* If /dev/null is not one of the stdio file descriptors, close it. */
1633    if (dev_null > 2) {
1634        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1635        status = TEMP_FAILURE_RETRY(close(dev_null));
1636        if (status == -1) {
1637            DL_ERR("close failed: %s", strerror(errno));
1638            return_value = -1;
1639        }
1640    }
1641
1642    return return_value;
1643}
1644
1645static bool soinfo_link_image(soinfo* si, const android_dlextinfo* extinfo) {
1646    /* "base" might wrap around UINT32_MAX. */
1647    ElfW(Addr) base = si->load_bias;
1648    const ElfW(Phdr)* phdr = si->phdr;
1649    int phnum = si->phnum;
1650    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1651
1652    /* We can't debug anything until the linker is relocated */
1653    if (!relocating_linker) {
1654        INFO("[ linking %s ]", si->name);
1655        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1656    }
1657
1658    /* Extract dynamic section */
1659    size_t dynamic_count;
1660    ElfW(Word) dynamic_flags;
1661    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1662                                   &dynamic_count, &dynamic_flags);
1663    if (si->dynamic == NULL) {
1664        if (!relocating_linker) {
1665            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1666        }
1667        return false;
1668    } else {
1669        if (!relocating_linker) {
1670            DEBUG("dynamic = %p", si->dynamic);
1671        }
1672    }
1673
1674#if defined(__arm__)
1675    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1676                                    &si->ARM_exidx, &si->ARM_exidx_count);
1677#endif
1678
1679    // Extract useful information from dynamic section.
1680    uint32_t needed_count = 0;
1681    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1682        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1683              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1684        switch (d->d_tag) {
1685        case DT_HASH:
1686            si->nbucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[0];
1687            si->nchain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr)[1];
1688            si->bucket = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8);
1689            si->chain = reinterpret_cast<uint32_t*>(base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1690            break;
1691        case DT_STRTAB:
1692            si->strtab = reinterpret_cast<const char*>(base + d->d_un.d_ptr);
1693            break;
1694        case DT_SYMTAB:
1695            si->symtab = reinterpret_cast<ElfW(Sym)*>(base + d->d_un.d_ptr);
1696            break;
1697#if !defined(__LP64__)
1698        case DT_PLTREL:
1699            if (d->d_un.d_val != DT_REL) {
1700                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1701                return false;
1702            }
1703            break;
1704#endif
1705        case DT_JMPREL:
1706#if defined(USE_RELA)
1707            si->plt_rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1708#else
1709            si->plt_rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1710#endif
1711            break;
1712        case DT_PLTRELSZ:
1713#if defined(USE_RELA)
1714            si->plt_rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1715#else
1716            si->plt_rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1717#endif
1718            break;
1719#if defined(__mips__)
1720        case DT_PLTGOT:
1721            // Used by mips and mips64.
1722            si->plt_got = reinterpret_cast<ElfW(Addr)**>(base + d->d_un.d_ptr);
1723            break;
1724#endif
1725        case DT_DEBUG:
1726            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1727            // if the dynamic table is writable
1728// FIXME: not working currently for N64
1729// The flags for the LOAD and DYNAMIC program headers do not agree.
1730// The LOAD section containng the dynamic table has been mapped as
1731// read-only, but the DYNAMIC header claims it is writable.
1732#if !(defined(__mips__) && defined(__LP64__))
1733            if ((dynamic_flags & PF_W) != 0) {
1734                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1735            }
1736            break;
1737#endif
1738#if defined(USE_RELA)
1739         case DT_RELA:
1740            si->rela = reinterpret_cast<ElfW(Rela)*>(base + d->d_un.d_ptr);
1741            break;
1742         case DT_RELASZ:
1743            si->rela_count = d->d_un.d_val / sizeof(ElfW(Rela));
1744            break;
1745        case DT_REL:
1746            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1747            return false;
1748        case DT_RELSZ:
1749            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1750            return false;
1751#else
1752        case DT_REL:
1753            si->rel = reinterpret_cast<ElfW(Rel)*>(base + d->d_un.d_ptr);
1754            break;
1755        case DT_RELSZ:
1756            si->rel_count = d->d_un.d_val / sizeof(ElfW(Rel));
1757            break;
1758         case DT_RELA:
1759            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1760            return false;
1761#endif
1762        case DT_INIT:
1763            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1764            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1765            break;
1766        case DT_FINI:
1767            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1768            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1769            break;
1770        case DT_INIT_ARRAY:
1771            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1772            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1773            break;
1774        case DT_INIT_ARRAYSZ:
1775            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1776            break;
1777        case DT_FINI_ARRAY:
1778            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1779            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1780            break;
1781        case DT_FINI_ARRAYSZ:
1782            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1783            break;
1784        case DT_PREINIT_ARRAY:
1785            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1786            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1787            break;
1788        case DT_PREINIT_ARRAYSZ:
1789            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(ElfW(Addr));
1790            break;
1791        case DT_TEXTREL:
1792#if defined(__LP64__)
1793            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1794            return false;
1795#else
1796            si->has_text_relocations = true;
1797            break;
1798#endif
1799        case DT_SYMBOLIC:
1800            si->has_DT_SYMBOLIC = true;
1801            break;
1802        case DT_NEEDED:
1803            ++needed_count;
1804            break;
1805        case DT_FLAGS:
1806            if (d->d_un.d_val & DF_TEXTREL) {
1807#if defined(__LP64__)
1808                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1809                return false;
1810#else
1811                si->has_text_relocations = true;
1812#endif
1813            }
1814            if (d->d_un.d_val & DF_SYMBOLIC) {
1815                si->has_DT_SYMBOLIC = true;
1816            }
1817            break;
1818#if defined(__mips__)
1819        case DT_STRSZ:
1820        case DT_SYMENT:
1821        case DT_RELENT:
1822             break;
1823        case DT_MIPS_RLD_MAP:
1824            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1825            {
1826              r_debug** dp = reinterpret_cast<r_debug**>(base + d->d_un.d_ptr);
1827              *dp = &_r_debug;
1828            }
1829            break;
1830        case DT_MIPS_RLD_VERSION:
1831        case DT_MIPS_FLAGS:
1832        case DT_MIPS_BASE_ADDRESS:
1833        case DT_MIPS_UNREFEXTNO:
1834            break;
1835
1836        case DT_MIPS_SYMTABNO:
1837            si->mips_symtabno = d->d_un.d_val;
1838            break;
1839
1840        case DT_MIPS_LOCAL_GOTNO:
1841            si->mips_local_gotno = d->d_un.d_val;
1842            break;
1843
1844        case DT_MIPS_GOTSYM:
1845            si->mips_gotsym = d->d_un.d_val;
1846            break;
1847#endif
1848
1849        default:
1850            DEBUG("Unused DT entry: type %p arg %p",
1851                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1852            break;
1853        }
1854    }
1855
1856    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1857          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1858
1859    // Sanity checks.
1860    if (relocating_linker && needed_count != 0) {
1861        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1862        return false;
1863    }
1864    if (si->nbucket == 0) {
1865        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1866        return false;
1867    }
1868    if (si->strtab == 0) {
1869        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1870        return false;
1871    }
1872    if (si->symtab == 0) {
1873        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1874        return false;
1875    }
1876
1877    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1878    if (si->flags & FLAG_EXE) {
1879        memset(g_ld_preloads, 0, sizeof(g_ld_preloads));
1880        size_t preload_count = 0;
1881        for (size_t i = 0; g_ld_preload_names[i] != NULL; i++) {
1882            soinfo* lsi = find_library(g_ld_preload_names[i], 0, NULL);
1883            if (lsi != NULL) {
1884                g_ld_preloads[preload_count++] = lsi;
1885            } else {
1886                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1887                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1888                        g_ld_preload_names[i], si->name, linker_get_error_buffer());
1889            }
1890        }
1891    }
1892
1893    soinfo** needed = reinterpret_cast<soinfo**>(alloca((1 + needed_count) * sizeof(soinfo*)));
1894    soinfo** pneeded = needed;
1895
1896    for (ElfW(Dyn)* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1897        if (d->d_tag == DT_NEEDED) {
1898            const char* library_name = si->strtab + d->d_un.d_val;
1899            DEBUG("%s needs %s", si->name, library_name);
1900            soinfo* lsi = find_library(library_name, 0, NULL);
1901            if (lsi == NULL) {
1902                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1903                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1904                       library_name, si->name, tmp_err_buf);
1905                return false;
1906            }
1907
1908            si->add_child(lsi);
1909            *pneeded++ = lsi;
1910        }
1911    }
1912    *pneeded = NULL;
1913
1914#if !defined(__LP64__)
1915    if (si->has_text_relocations) {
1916        // Make segments writable to allow text relocations to work properly. We will later call
1917        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1918        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1919                "security hardening. Please fix.", si->name);
1920        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1921            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1922                   si->name, strerror(errno));
1923            return false;
1924        }
1925    }
1926#endif
1927
1928#if defined(USE_RELA)
1929    if (si->plt_rela != NULL) {
1930        DEBUG("[ relocating %s plt ]\n", si->name);
1931        if (soinfo_relocate(si, si->plt_rela, si->plt_rela_count, needed)) {
1932            return false;
1933        }
1934    }
1935    if (si->rela != NULL) {
1936        DEBUG("[ relocating %s ]\n", si->name);
1937        if (soinfo_relocate(si, si->rela, si->rela_count, needed)) {
1938            return false;
1939        }
1940    }
1941#else
1942    if (si->plt_rel != NULL) {
1943        DEBUG("[ relocating %s plt ]", si->name);
1944        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1945            return false;
1946        }
1947    }
1948    if (si->rel != NULL) {
1949        DEBUG("[ relocating %s ]", si->name);
1950        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1951            return false;
1952        }
1953    }
1954#endif
1955
1956#if defined(__mips__)
1957    if (!mips_relocate_got(si, needed)) {
1958        return false;
1959    }
1960#endif
1961
1962    si->flags |= FLAG_LINKED;
1963    DEBUG("[ finished linking %s ]", si->name);
1964
1965#if !defined(__LP64__)
1966    if (si->has_text_relocations) {
1967        // All relocations are done, we can protect our segments back to read-only.
1968        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1969            DL_ERR("can't protect segments for \"%s\": %s",
1970                   si->name, strerror(errno));
1971            return false;
1972        }
1973    }
1974#endif
1975
1976    /* We can also turn on GNU RELRO protection */
1977    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1978        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1979               si->name, strerror(errno));
1980        return false;
1981    }
1982
1983    /* Handle serializing/sharing the RELRO segment */
1984    if (extinfo && (extinfo->flags & ANDROID_DLEXT_WRITE_RELRO)) {
1985      if (phdr_table_serialize_gnu_relro(si->phdr, si->phnum, si->load_bias,
1986                                         extinfo->relro_fd) < 0) {
1987        DL_ERR("failed serializing GNU RELRO section for \"%s\": %s",
1988               si->name, strerror(errno));
1989        return false;
1990      }
1991    } else if (extinfo && (extinfo->flags & ANDROID_DLEXT_USE_RELRO)) {
1992      if (phdr_table_map_gnu_relro(si->phdr, si->phnum, si->load_bias,
1993                                   extinfo->relro_fd) < 0) {
1994        DL_ERR("failed mapping GNU RELRO section for \"%s\": %s",
1995               si->name, strerror(errno));
1996        return false;
1997      }
1998    }
1999
2000    notify_gdb_of_load(si);
2001    return true;
2002}
2003
2004/*
2005 * This function add vdso to internal dso list.
2006 * It helps to stack unwinding through signal handlers.
2007 * Also, it makes bionic more like glibc.
2008 */
2009static void add_vdso(KernelArgumentBlock& args __unused) {
2010#if defined(AT_SYSINFO_EHDR)
2011  ElfW(Ehdr)* ehdr_vdso = reinterpret_cast<ElfW(Ehdr)*>(args.getauxval(AT_SYSINFO_EHDR));
2012  if (ehdr_vdso == NULL) {
2013    return;
2014  }
2015
2016  soinfo* si = soinfo_alloc("[vdso]", NULL);
2017
2018  si->phdr = reinterpret_cast<ElfW(Phdr)*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
2019  si->phnum = ehdr_vdso->e_phnum;
2020  si->base = reinterpret_cast<ElfW(Addr)>(ehdr_vdso);
2021  si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2022  si->load_bias = get_elf_exec_load_bias(ehdr_vdso);
2023
2024  soinfo_link_image(si, NULL);
2025#endif
2026}
2027
2028/*
2029 * This is linker soinfo for GDB. See details below.
2030 */
2031static soinfo linker_soinfo_for_gdb;
2032
2033/* gdb expects the linker to be in the debug shared object list.
2034 * Without this, gdb has trouble locating the linker's ".text"
2035 * and ".plt" sections. Gdb could also potentially use this to
2036 * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
2037 * Don't use soinfo_alloc(), because the linker shouldn't
2038 * be on the soinfo list.
2039 */
2040static void init_linker_info_for_gdb(ElfW(Addr) linker_base) {
2041#if defined(__LP64__)
2042  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker64", sizeof(linker_soinfo_for_gdb.name));
2043#else
2044  strlcpy(linker_soinfo_for_gdb.name, "/system/bin/linker", sizeof(linker_soinfo_for_gdb.name));
2045#endif
2046  linker_soinfo_for_gdb.flags = FLAG_NEW_SOINFO;
2047  linker_soinfo_for_gdb.base = linker_base;
2048
2049  /*
2050   * Set the dynamic field in the link map otherwise gdb will complain with
2051   * the following:
2052   *   warning: .dynamic section for "/system/bin/linker" is not at the
2053   *   expected address (wrong library or version mismatch?)
2054   */
2055  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_base);
2056  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_base + elf_hdr->e_phoff);
2057  phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
2058                                 &linker_soinfo_for_gdb.dynamic, NULL, NULL);
2059  insert_soinfo_into_debug_map(&linker_soinfo_for_gdb);
2060}
2061
2062/*
2063 * This code is called after the linker has linked itself and
2064 * fixed it's own GOT. It is safe to make references to externs
2065 * and other non-local data at this point.
2066 */
2067static ElfW(Addr) __linker_init_post_relocation(KernelArgumentBlock& args, ElfW(Addr) linker_base) {
2068    /* NOTE: we store the args pointer on a special location
2069     *       of the temporary TLS area in order to pass it to
2070     *       the C Library's runtime initializer.
2071     *
2072     *       The initializer must clear the slot and reset the TLS
2073     *       to point to a different location to ensure that no other
2074     *       shared library constructor can access it.
2075     */
2076  __libc_init_tls(args);
2077
2078#if TIMING
2079    struct timeval t0, t1;
2080    gettimeofday(&t0, 0);
2081#endif
2082
2083    // Initialize environment functions, and get to the ELF aux vectors table.
2084    linker_env_init(args);
2085
2086    // If this is a setuid/setgid program, close the security hole described in
2087    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
2088    if (get_AT_SECURE()) {
2089        nullify_closed_stdio();
2090    }
2091
2092    debuggerd_init();
2093
2094    // Get a few environment variables.
2095    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
2096    if (LD_DEBUG != NULL) {
2097      g_ld_debug_verbosity = atoi(LD_DEBUG);
2098    }
2099
2100    // Normally, these are cleaned by linker_env_init, but the test
2101    // doesn't cost us anything.
2102    const char* ldpath_env = NULL;
2103    const char* ldpreload_env = NULL;
2104    if (!get_AT_SECURE()) {
2105      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
2106      ldpreload_env = linker_env_get("LD_PRELOAD");
2107    }
2108
2109    INFO("[ android linker & debugger ]");
2110
2111    soinfo* si = soinfo_alloc(args.argv[0], NULL);
2112    if (si == NULL) {
2113        exit(EXIT_FAILURE);
2114    }
2115
2116    /* bootstrap the link map, the main exe always needs to be first */
2117    si->flags |= FLAG_EXE;
2118    link_map* map = &(si->link_map_head);
2119
2120    map->l_addr = 0;
2121    map->l_name = args.argv[0];
2122    map->l_prev = NULL;
2123    map->l_next = NULL;
2124
2125    _r_debug.r_map = map;
2126    r_debug_tail = map;
2127
2128    init_linker_info_for_gdb(linker_base);
2129
2130    // Extract information passed from the kernel.
2131    si->phdr = reinterpret_cast<ElfW(Phdr)*>(args.getauxval(AT_PHDR));
2132    si->phnum = args.getauxval(AT_PHNUM);
2133    si->entry = args.getauxval(AT_ENTRY);
2134
2135    /* Compute the value of si->base. We can't rely on the fact that
2136     * the first entry is the PHDR because this will not be true
2137     * for certain executables (e.g. some in the NDK unit test suite)
2138     */
2139    si->base = 0;
2140    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
2141    si->load_bias = 0;
2142    for (size_t i = 0; i < si->phnum; ++i) {
2143      if (si->phdr[i].p_type == PT_PHDR) {
2144        si->load_bias = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_vaddr;
2145        si->base = reinterpret_cast<ElfW(Addr)>(si->phdr) - si->phdr[i].p_offset;
2146        break;
2147      }
2148    }
2149    si->dynamic = NULL;
2150    si->ref_count = 1;
2151
2152    ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(si->base);
2153    if (elf_hdr->e_type != ET_DYN) {
2154        __libc_format_fd(2, "error: only position independent executables (PIE) are supported.\n");
2155        exit(EXIT_FAILURE);
2156    }
2157
2158    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
2159    parse_LD_LIBRARY_PATH(ldpath_env);
2160    parse_LD_PRELOAD(ldpreload_env);
2161
2162    somain = si;
2163
2164    if (!soinfo_link_image(si, NULL)) {
2165        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
2166        exit(EXIT_FAILURE);
2167    }
2168
2169    add_vdso(args);
2170
2171    si->CallPreInitConstructors();
2172
2173    for (size_t i = 0; g_ld_preloads[i] != NULL; ++i) {
2174        g_ld_preloads[i]->CallConstructors();
2175    }
2176
2177    /* After the link_image, the si->load_bias is initialized.
2178     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
2179     * We need to update this value for so exe here. So Unwind_Backtrace
2180     * for some arch like x86 could work correctly within so exe.
2181     */
2182    map->l_addr = si->load_bias;
2183    si->CallConstructors();
2184
2185#if TIMING
2186    gettimeofday(&t1, NULL);
2187    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
2188               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
2189               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)));
2190#endif
2191#if STATS
2192    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
2193           linker_stats.count[kRelocAbsolute],
2194           linker_stats.count[kRelocRelative],
2195           linker_stats.count[kRelocCopy],
2196           linker_stats.count[kRelocSymbol]);
2197#endif
2198#if COUNT_PAGES
2199    {
2200        unsigned n;
2201        unsigned i;
2202        unsigned count = 0;
2203        for (n = 0; n < 4096; n++) {
2204            if (bitmask[n]) {
2205                unsigned x = bitmask[n];
2206#if defined(__LP64__)
2207                for (i = 0; i < 32; i++) {
2208#else
2209                for (i = 0; i < 8; i++) {
2210#endif
2211                    if (x & 1) {
2212                        count++;
2213                    }
2214                    x >>= 1;
2215                }
2216            }
2217        }
2218        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
2219    }
2220#endif
2221
2222#if TIMING || STATS || COUNT_PAGES
2223    fflush(stdout);
2224#endif
2225
2226    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
2227    return si->entry;
2228}
2229
2230/* Compute the load-bias of an existing executable. This shall only
2231 * be used to compute the load bias of an executable or shared library
2232 * that was loaded by the kernel itself.
2233 *
2234 * Input:
2235 *    elf    -> address of ELF header, assumed to be at the start of the file.
2236 * Return:
2237 *    load bias, i.e. add the value of any p_vaddr in the file to get
2238 *    the corresponding address in memory.
2239 */
2240static ElfW(Addr) get_elf_exec_load_bias(const ElfW(Ehdr)* elf) {
2241  ElfW(Addr) offset = elf->e_phoff;
2242  const ElfW(Phdr)* phdr_table = reinterpret_cast<const ElfW(Phdr)*>(reinterpret_cast<uintptr_t>(elf) + offset);
2243  const ElfW(Phdr)* phdr_end = phdr_table + elf->e_phnum;
2244
2245  for (const ElfW(Phdr)* phdr = phdr_table; phdr < phdr_end; phdr++) {
2246    if (phdr->p_type == PT_LOAD) {
2247      return reinterpret_cast<ElfW(Addr)>(elf) + phdr->p_offset - phdr->p_vaddr;
2248    }
2249  }
2250  return 0;
2251}
2252
2253extern "C" void _start();
2254
2255/*
2256 * This is the entry point for the linker, called from begin.S. This
2257 * method is responsible for fixing the linker's own relocations, and
2258 * then calling __linker_init_post_relocation().
2259 *
2260 * Because this method is called before the linker has fixed it's own
2261 * relocations, any attempt to reference an extern variable, extern
2262 * function, or other GOT reference will generate a segfault.
2263 */
2264extern "C" ElfW(Addr) __linker_init(void* raw_args) {
2265  // Initialize static variables.
2266  solist = get_libdl_info();
2267  sonext = get_libdl_info();
2268
2269  KernelArgumentBlock args(raw_args);
2270
2271  ElfW(Addr) linker_addr = args.getauxval(AT_BASE);
2272  ElfW(Addr) entry_point = args.getauxval(AT_ENTRY);
2273  ElfW(Ehdr)* elf_hdr = reinterpret_cast<ElfW(Ehdr)*>(linker_addr);
2274  ElfW(Phdr)* phdr = reinterpret_cast<ElfW(Phdr)*>(linker_addr + elf_hdr->e_phoff);
2275
2276  soinfo linker_so;
2277  memset(&linker_so, 0, sizeof(soinfo));
2278
2279  // If the linker is not acting as PT_INTERP entry_point is equal to
2280  // _start. Which means that the linker is running as an executable and
2281  // already linked by PT_INTERP.
2282  //
2283  // This happens when user tries to run 'adb shell /system/bin/linker'
2284  // see also https://code.google.com/p/android/issues/detail?id=63174
2285  if (reinterpret_cast<ElfW(Addr)>(&_start) == entry_point) {
2286    __libc_fatal("This is %s, the helper program for shared library executables.\n", args.argv[0]);
2287  }
2288
2289  strcpy(linker_so.name, "[dynamic linker]");
2290  linker_so.base = linker_addr;
2291  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
2292  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
2293  linker_so.dynamic = NULL;
2294  linker_so.phdr = phdr;
2295  linker_so.phnum = elf_hdr->e_phnum;
2296  linker_so.flags |= FLAG_LINKER;
2297
2298  if (!soinfo_link_image(&linker_so, NULL)) {
2299    // It would be nice to print an error message, but if the linker
2300    // can't link itself, there's no guarantee that we'll be able to
2301    // call write() (because it involves a GOT reference). We may as
2302    // well try though...
2303    const char* msg = "CANNOT LINK EXECUTABLE: ";
2304    write(2, msg, strlen(msg));
2305    write(2, __linker_dl_err_buf, strlen(__linker_dl_err_buf));
2306    write(2, "\n", 1);
2307    _exit(EXIT_FAILURE);
2308  }
2309
2310  // Initialize the linker's own global variables
2311  linker_so.CallConstructors();
2312
2313  // We have successfully fixed our own relocations. It's safe to run
2314  // the main part of the linker now.
2315  args.abort_message_ptr = &g_abort_message;
2316  ElfW(Addr) start_address = __linker_init_post_relocation(args, linker_addr);
2317
2318  protect_data(PROT_READ);
2319
2320  // Return the address that the calling assembly stub should jump to.
2321  return start_address;
2322}
2323