linker.cpp revision e4d792adb8d6f9228b9ac9dc1ad7f43b271f085f
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 <linux/auxvec.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/* Assume average path length of 64 and max 8 paths */
53#define LDPATH_BUFSIZE 512
54#define LDPATH_MAX 8
55
56#define LDPRELOAD_BUFSIZE 512
57#define LDPRELOAD_MAX 8
58
59/* >>> IMPORTANT NOTE - READ ME BEFORE MODIFYING <<<
60 *
61 * Do NOT use malloc() and friends or pthread_*() code here.
62 * Don't use printf() either; it's caused mysterious memory
63 * corruption in the past.
64 * The linker runs before we bring up libc and it's easiest
65 * to make sure it does not depend on any complex libc features
66 *
67 * open issues / todo:
68 *
69 * - are we doing everything we should for ARM_COPY relocations?
70 * - cleaner error reporting
71 * - after linking, set as much stuff as possible to READONLY
72 *   and NOEXEC
73 */
74
75static bool soinfo_link_image(soinfo* si);
76
77// We can't use malloc(3) in the dynamic linker. We use a linked list of anonymous
78// maps, each a single page in size. The pages are broken up into as many struct soinfo
79// objects as will fit, and they're all threaded together on a free list.
80#define SOINFO_PER_POOL ((PAGE_SIZE - sizeof(soinfo_pool_t*)) / sizeof(soinfo))
81struct soinfo_pool_t {
82  soinfo_pool_t* next;
83  soinfo info[SOINFO_PER_POOL];
84};
85static struct soinfo_pool_t* gSoInfoPools = NULL;
86static soinfo* gSoInfoFreeList = NULL;
87
88static soinfo* solist = &libdl_info;
89static soinfo* sonext = &libdl_info;
90static soinfo* somain; /* main process, always the one after libdl_info */
91
92static const char* const gSoPaths[] = {
93#if defined(__LP64__)
94  "/vendor/lib64",
95  "/system/lib64",
96#else
97  "/vendor/lib",
98  "/system/lib",
99#endif
100  NULL
101};
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#define MARK(offset) \
141    do { \
142        bitmask[((offset) >> 12) >> 3] |= (1 << (((offset) >> 12) & 7)); \
143    } while(0)
144#else
145#define MARK(x) do {} while (0)
146#endif
147
148// You shouldn't try to call memory-allocating functions in the dynamic linker.
149// Guard against the most obvious ones.
150#define DISALLOW_ALLOCATION(return_type, name, ...) \
151    return_type name __VA_ARGS__ \
152    { \
153        const char* msg = "ERROR: " #name " called from the dynamic linker!\n"; \
154        __libc_format_log(ANDROID_LOG_FATAL, "linker", "%s", msg); \
155        write(2, msg, strlen(msg)); \
156        abort(); \
157    }
158#define UNUSED __attribute__((unused))
159DISALLOW_ALLOCATION(void*, malloc, (size_t u UNUSED));
160DISALLOW_ALLOCATION(void, free, (void* u UNUSED));
161DISALLOW_ALLOCATION(void*, realloc, (void* u1 UNUSED, size_t u2 UNUSED));
162DISALLOW_ALLOCATION(void*, calloc, (size_t u1 UNUSED, size_t u2 UNUSED));
163
164static char tmp_err_buf[768];
165static char __linker_dl_err_buf[768];
166
167char* linker_get_error_buffer() {
168  return &__linker_dl_err_buf[0];
169}
170
171size_t linker_get_error_buffer_size() {
172  return sizeof(__linker_dl_err_buf);
173}
174
175/*
176 * This function is an empty stub where GDB locates a breakpoint to get notified
177 * about linker activity.
178 */
179extern "C" void __attribute__((noinline)) __attribute__((visibility("default"))) rtld_db_dlactivity();
180
181static r_debug _r_debug = {1, NULL, &rtld_db_dlactivity, RT_CONSISTENT, 0};
182static link_map_t* r_debug_tail = 0;
183
184static pthread_mutex_t gDebugMutex = PTHREAD_MUTEX_INITIALIZER;
185
186static void insert_soinfo_into_debug_map(soinfo * info) {
187    // Copy the necessary fields into the debug structure.
188    link_map_t* map = &(info->link_map);
189    map->l_addr = info->base;
190    map->l_name = (char*) info->name;
191    map->l_ld = (uintptr_t)info->dynamic;
192
193    /* Stick the new library at the end of the list.
194     * gdb tends to care more about libc than it does
195     * about leaf libraries, and ordering it this way
196     * reduces the back-and-forth over the wire.
197     */
198    if (r_debug_tail) {
199        r_debug_tail->l_next = map;
200        map->l_prev = r_debug_tail;
201        map->l_next = 0;
202    } else {
203        _r_debug.r_map = map;
204        map->l_prev = 0;
205        map->l_next = 0;
206    }
207    r_debug_tail = map;
208}
209
210static void remove_soinfo_from_debug_map(soinfo* info) {
211    link_map_t* map = &(info->link_map);
212
213    if (r_debug_tail == map) {
214        r_debug_tail = map->l_prev;
215    }
216
217    if (map->l_prev) {
218        map->l_prev->l_next = map->l_next;
219    }
220    if (map->l_next) {
221        map->l_next->l_prev = map->l_prev;
222    }
223}
224
225static void notify_gdb_of_load(soinfo* info) {
226    if (info->flags & FLAG_EXE) {
227        // GDB already knows about the main executable
228        return;
229    }
230
231    ScopedPthreadMutexLocker locker(&gDebugMutex);
232
233    _r_debug.r_state = RT_ADD;
234    rtld_db_dlactivity();
235
236    insert_soinfo_into_debug_map(info);
237
238    _r_debug.r_state = RT_CONSISTENT;
239    rtld_db_dlactivity();
240}
241
242static void notify_gdb_of_unload(soinfo* info) {
243    if (info->flags & FLAG_EXE) {
244        // GDB already knows about the main executable
245        return;
246    }
247
248    ScopedPthreadMutexLocker locker(&gDebugMutex);
249
250    _r_debug.r_state = RT_DELETE;
251    rtld_db_dlactivity();
252
253    remove_soinfo_from_debug_map(info);
254
255    _r_debug.r_state = RT_CONSISTENT;
256    rtld_db_dlactivity();
257}
258
259void notify_gdb_of_libraries() {
260    _r_debug.r_state = RT_ADD;
261    rtld_db_dlactivity();
262    _r_debug.r_state = RT_CONSISTENT;
263    rtld_db_dlactivity();
264}
265
266static bool ensure_free_list_non_empty() {
267  if (gSoInfoFreeList != NULL) {
268    return true;
269  }
270
271  // Allocate a new pool.
272  soinfo_pool_t* pool = reinterpret_cast<soinfo_pool_t*>(mmap(NULL, sizeof(*pool),
273                                                              PROT_READ|PROT_WRITE,
274                                                              MAP_PRIVATE|MAP_ANONYMOUS, 0, 0));
275  if (pool == MAP_FAILED) {
276    return false;
277  }
278
279  // Add the pool to our list of pools.
280  pool->next = gSoInfoPools;
281  gSoInfoPools = pool;
282
283  // Chain the entries in the new pool onto the free list.
284  gSoInfoFreeList = &pool->info[0];
285  soinfo* next = NULL;
286  for (int i = SOINFO_PER_POOL - 1; i >= 0; --i) {
287    pool->info[i].next = next;
288    next = &pool->info[i];
289  }
290
291  return true;
292}
293
294static void set_soinfo_pool_protection(int protection) {
295  for (soinfo_pool_t* p = gSoInfoPools; p != NULL; p = p->next) {
296    if (mprotect(p, sizeof(*p), protection) == -1) {
297      abort(); // Can't happen.
298    }
299  }
300}
301
302static soinfo* soinfo_alloc(const char* name) {
303  if (strlen(name) >= SOINFO_NAME_LEN) {
304    DL_ERR("library name \"%s\" too long", name);
305    return NULL;
306  }
307
308  if (!ensure_free_list_non_empty()) {
309    DL_ERR("out of memory when loading \"%s\"", name);
310    return NULL;
311  }
312
313  // Take the head element off the free list.
314  soinfo* si = gSoInfoFreeList;
315  gSoInfoFreeList = gSoInfoFreeList->next;
316
317  // Initialize the new element.
318  memset(si, 0, sizeof(soinfo));
319  strlcpy(si->name, name, sizeof(si->name));
320  sonext->next = si;
321  sonext = si;
322
323  TRACE("name %s: allocated soinfo @ %p", name, si);
324  return si;
325}
326
327static void soinfo_free(soinfo* si)
328{
329    if (si == NULL) {
330        return;
331    }
332
333    soinfo *prev = NULL, *trav;
334
335    TRACE("name %s: freeing soinfo @ %p", si->name, si);
336
337    for (trav = solist; trav != NULL; trav = trav->next) {
338        if (trav == si)
339            break;
340        prev = trav;
341    }
342    if (trav == NULL) {
343        /* si was not in solist */
344        DL_ERR("name \"%s\" is not in solist!", si->name);
345        return;
346    }
347
348    /* prev will never be NULL, because the first entry in solist is
349       always the static libdl_info.
350    */
351    prev->next = si->next;
352    if (si == sonext) {
353        sonext = prev;
354    }
355    si->next = gSoInfoFreeList;
356    gSoInfoFreeList = si;
357}
358
359
360static void parse_path(const char* path, const char* delimiters,
361                       const char** array, char* buf, size_t buf_size, size_t max_count) {
362  if (path == NULL) {
363    return;
364  }
365
366  size_t len = strlcpy(buf, path, buf_size);
367
368  size_t i = 0;
369  char* buf_p = buf;
370  while (i < max_count && (array[i] = strsep(&buf_p, delimiters))) {
371    if (*array[i] != '\0') {
372      ++i;
373    }
374  }
375
376  // Forget the last path if we had to truncate; this occurs if the 2nd to
377  // last char isn't '\0' (i.e. wasn't originally a delimiter).
378  if (i > 0 && len >= buf_size && buf[buf_size - 2] != '\0') {
379    array[i - 1] = NULL;
380  } else {
381    array[i] = NULL;
382  }
383}
384
385static void parse_LD_LIBRARY_PATH(const char* path) {
386  parse_path(path, ":", gLdPaths,
387             gLdPathsBuffer, sizeof(gLdPathsBuffer), LDPATH_MAX);
388}
389
390static void parse_LD_PRELOAD(const char* path) {
391  // We have historically supported ':' as well as ' ' in LD_PRELOAD.
392  parse_path(path, " :", gLdPreloadNames,
393             gLdPreloadsBuffer, sizeof(gLdPreloadsBuffer), LDPRELOAD_MAX);
394}
395
396#if defined(__arm__)
397
398/* For a given PC, find the .so that it belongs to.
399 * Returns the base address of the .ARM.exidx section
400 * for that .so, and the number of 8-byte entries
401 * in that section (via *pcount).
402 *
403 * Intended to be called by libc's __gnu_Unwind_Find_exidx().
404 *
405 * This function is exposed via dlfcn.cpp and libdl.so.
406 */
407_Unwind_Ptr dl_unwind_find_exidx(_Unwind_Ptr pc, int *pcount)
408{
409    soinfo *si;
410    unsigned addr = (unsigned)pc;
411
412    for (si = solist; si != 0; si = si->next) {
413        if ((addr >= si->base) && (addr < (si->base + si->size))) {
414            *pcount = si->ARM_exidx_count;
415            return (_Unwind_Ptr)si->ARM_exidx;
416        }
417    }
418   *pcount = 0;
419    return NULL;
420}
421
422#endif
423
424/* Here, we only have to provide a callback to iterate across all the
425 * loaded libraries. gcc_eh does the rest. */
426int
427dl_iterate_phdr(int (*cb)(dl_phdr_info *info, size_t size, void *data),
428                void *data)
429{
430    int rv = 0;
431    for (soinfo* si = solist; si != NULL; si = si->next) {
432        dl_phdr_info dl_info;
433        dl_info.dlpi_addr = si->link_map.l_addr;
434        dl_info.dlpi_name = si->link_map.l_name;
435        dl_info.dlpi_phdr = si->phdr;
436        dl_info.dlpi_phnum = si->phnum;
437        rv = cb(&dl_info, sizeof(dl_phdr_info), data);
438        if (rv != 0) {
439            break;
440        }
441    }
442    return rv;
443}
444
445static Elf_Sym* soinfo_elf_lookup(soinfo* si, unsigned hash, const char* name) {
446    Elf_Sym* symtab = si->symtab;
447    const char* strtab = si->strtab;
448
449    TRACE_TYPE(LOOKUP, "SEARCH %s in %s@%p %x %zd",
450               name, si->name, reinterpret_cast<void*>(si->base), hash, hash % si->nbucket);
451
452    for (unsigned n = si->bucket[hash % si->nbucket]; n != 0; n = si->chain[n]) {
453        Elf_Sym* s = symtab + n;
454        if (strcmp(strtab + s->st_name, name)) continue;
455
456            /* only concern ourselves with global and weak symbol definitions */
457        switch (ELF_ST_BIND(s->st_info)) {
458        case STB_GLOBAL:
459        case STB_WEAK:
460            if (s->st_shndx == SHN_UNDEF) {
461                continue;
462            }
463
464            TRACE_TYPE(LOOKUP, "FOUND %s in %s (%p) %zd",
465                       name, si->name, reinterpret_cast<void*>(s->st_value),
466                       static_cast<size_t>(s->st_size));
467            return s;
468        }
469    }
470
471    return NULL;
472}
473
474static unsigned elfhash(const char* _name) {
475    const unsigned char* name = (const unsigned char*) _name;
476    unsigned h = 0, g;
477
478    while(*name) {
479        h = (h << 4) + *name++;
480        g = h & 0xf0000000;
481        h ^= g;
482        h ^= g >> 24;
483    }
484    return h;
485}
486
487static Elf_Sym* soinfo_do_lookup(soinfo* si, const char* name, soinfo** lsi, soinfo* needed[]) {
488    unsigned elf_hash = elfhash(name);
489    Elf_Sym* s = NULL;
490
491    if (si != NULL && somain != NULL) {
492
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 */
597Elf_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 */
606Elf_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  Elf_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  Elf_Addr address = reinterpret_cast<Elf_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
640Elf_Sym* dladdr_find_symbol(soinfo* si, const void* addr) {
641  Elf_Addr soaddr = reinterpret_cast<Elf_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    Elf_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
657#if 0
658static void dump(soinfo* si)
659{
660    Elf_Sym* s = si->symtab;
661    for (unsigned n = 0; n < si->nchain; n++) {
662        TRACE("%04d> %08x: %02x %04x %08x %08x %s", n, s,
663               s->st_info, s->st_shndx, s->st_value, s->st_size,
664               si->strtab + s->st_name);
665        s++;
666    }
667}
668#endif
669
670static int open_library_on_path(const char* name, const char* const paths[]) {
671  char buf[512];
672  for (size_t i = 0; paths[i] != NULL; ++i) {
673    int n = __libc_format_buffer(buf, sizeof(buf), "%s/%s", paths[i], name);
674    if (n < 0 || n >= static_cast<int>(sizeof(buf))) {
675      PRINT("Warning: ignoring very long library path: %s/%s", paths[i], name);
676      continue;
677    }
678    int fd = TEMP_FAILURE_RETRY(open(buf, O_RDONLY | O_CLOEXEC));
679    if (fd != -1) {
680      return fd;
681    }
682  }
683  return -1;
684}
685
686static int open_library(const char* name) {
687  TRACE("[ opening %s ]", name);
688
689  // If the name contains a slash, we should attempt to open it directly and not search the paths.
690  if (strchr(name, '/') != NULL) {
691    int fd = TEMP_FAILURE_RETRY(open(name, O_RDONLY | O_CLOEXEC));
692    if (fd != -1) {
693      return fd;
694    }
695    // ...but nvidia binary blobs (at least) rely on this behavior, so fall through for now.
696  }
697
698  // Otherwise we try LD_LIBRARY_PATH first, and fall back to the built-in well known paths.
699  int fd = open_library_on_path(name, gLdPaths);
700  if (fd == -1) {
701    fd = open_library_on_path(name, gSoPaths);
702  }
703  return fd;
704}
705
706static soinfo* load_library(const char* name) {
707    // Open the file.
708    int fd = open_library(name);
709    if (fd == -1) {
710        DL_ERR("library \"%s\" not found", name);
711        return NULL;
712    }
713
714    // Read the ELF header and load the segments.
715    ElfReader elf_reader(name, fd);
716    if (!elf_reader.Load()) {
717        return NULL;
718    }
719
720    const char* bname = strrchr(name, '/');
721    soinfo* si = soinfo_alloc(bname ? bname + 1 : name);
722    if (si == NULL) {
723        return NULL;
724    }
725    si->base = elf_reader.load_start();
726    si->size = elf_reader.load_size();
727    si->load_bias = elf_reader.load_bias();
728    si->flags = 0;
729    si->entry = 0;
730    si->dynamic = NULL;
731    si->phnum = elf_reader.phdr_count();
732    si->phdr = elf_reader.loaded_phdr();
733    return si;
734}
735
736static soinfo *find_loaded_library(const char *name)
737{
738    soinfo *si;
739    const char *bname;
740
741    // TODO: don't use basename only for determining libraries
742    // http://code.google.com/p/android/issues/detail?id=6670
743
744    bname = strrchr(name, '/');
745    bname = bname ? bname + 1 : name;
746
747    for (si = solist; si != NULL; si = si->next) {
748        if (!strcmp(bname, si->name)) {
749            return si;
750        }
751    }
752    return NULL;
753}
754
755static soinfo* find_library_internal(const char* name) {
756  if (name == NULL) {
757    return somain;
758  }
759
760  soinfo* si = find_loaded_library(name);
761  if (si != NULL) {
762    if (si->flags & FLAG_LINKED) {
763      return si;
764    }
765    DL_ERR("OOPS: recursive link to \"%s\"", si->name);
766    return NULL;
767  }
768
769  TRACE("[ '%s' has not been loaded yet.  Locating...]", name);
770  si = load_library(name);
771  if (si == NULL) {
772    return NULL;
773  }
774
775  // At this point we know that whatever is loaded @ base is a valid ELF
776  // shared library whose segments are properly mapped in.
777  TRACE("[ init_library base=%p sz=0x%08x name='%s' ]",
778        reinterpret_cast<void*>(si->base), si->size, si->name);
779
780  if (!soinfo_link_image(si)) {
781    munmap(reinterpret_cast<void*>(si->base), si->size);
782    soinfo_free(si);
783    return NULL;
784  }
785
786  return si;
787}
788
789static soinfo* find_library(const char* name) {
790  soinfo* si = find_library_internal(name);
791  if (si != NULL) {
792    si->ref_count++;
793  }
794  return si;
795}
796
797static int soinfo_unload(soinfo* si) {
798  if (si->ref_count == 1) {
799    TRACE("unloading '%s'", si->name);
800    si->CallDestructors();
801
802    for (Elf_Dyn* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
803      if (d->d_tag == DT_NEEDED) {
804        const char* library_name = si->strtab + d->d_un.d_val;
805        TRACE("%s needs to unload %s", si->name, library_name);
806        soinfo_unload(find_loaded_library(library_name));
807      }
808    }
809
810    munmap(reinterpret_cast<void*>(si->base), si->size);
811    notify_gdb_of_unload(si);
812    soinfo_free(si);
813    si->ref_count = 0;
814  } else {
815    si->ref_count--;
816    TRACE("not unloading '%s', decrementing ref_count to %zd", si->name, si->ref_count);
817  }
818  return 0;
819}
820
821void do_android_update_LD_LIBRARY_PATH(const char* ld_library_path) {
822  if (!get_AT_SECURE()) {
823    parse_LD_LIBRARY_PATH(ld_library_path);
824  }
825}
826
827soinfo* do_dlopen(const char* name, int flags) {
828  if ((flags & ~(RTLD_NOW|RTLD_LAZY|RTLD_LOCAL|RTLD_GLOBAL)) != 0) {
829    DL_ERR("invalid flags to dlopen: %x", flags);
830    return NULL;
831  }
832  set_soinfo_pool_protection(PROT_READ | PROT_WRITE);
833  soinfo* si = find_library(name);
834  if (si != NULL) {
835    si->CallConstructors();
836  }
837  set_soinfo_pool_protection(PROT_READ);
838  return si;
839}
840
841int do_dlclose(soinfo* si) {
842  set_soinfo_pool_protection(PROT_READ | PROT_WRITE);
843  int result = soinfo_unload(si);
844  set_soinfo_pool_protection(PROT_READ);
845  return result;
846}
847
848#if defined(USE_RELA)
849static int soinfo_relocate_a(soinfo* si, Elf_Rela* rela, unsigned count, soinfo* needed[]) {
850  Elf_Sym* symtab = si->symtab;
851  const char* strtab = si->strtab;
852  Elf_Sym* s;
853  Elf_Rela* start = rela;
854  soinfo* lsi;
855
856  for (size_t idx = 0; idx < count; ++idx, ++rela) {
857    unsigned type = ELF_R_TYPE(rela->r_info);
858    unsigned sym = ELF_R_SYM(rela->r_info);
859    Elf_Addr reloc = static_cast<Elf_Addr>(rela->r_offset + si->load_bias);
860    Elf_Addr sym_addr = 0;
861    char* sym_name = NULL;
862
863    DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
864    if (type == 0) { // R_*_NONE
865      continue;
866    }
867    if (sym != 0) {
868      sym_name = (char *)(strtab + symtab[sym].st_name);
869      s = soinfo_do_lookup(si, sym_name, &lsi, needed);
870      if (s == NULL) {
871        // We only allow an undefined symbol if this is a weak reference...
872        s = &symtab[sym];
873        if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
874          DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
875          return -1;
876        }
877
878        /* IHI0044C AAELF 4.5.1.1:
879
880           Libraries are not searched to resolve weak references.
881           It is not an error for a weak reference to remain unsatisfied.
882
883           During linking, the value of an undefined weak reference is:
884           - Zero if the relocation type is absolute
885           - The address of the place if the relocation is pc-relative
886           - The address of nominal base address if the relocation
887             type is base-relative.
888         */
889
890        switch (type) {
891#if defined(__x86_64__)
892        case R_X86_64_JUMP_SLOT:
893        case R_X86_64_GLOB_DAT:
894        case R_X86_64_32:
895        case R_X86_64_RELATIVE:
896          // No need to do anything.
897          break;
898        case R_X86_64_PC32:
899          sym_addr = reloc;
900          break;
901#endif
902
903        default:
904          DL_ERR("unknown weak reloc type %d @ %p (%d)", type, rela, (int) (rela - start));
905          return -1;
906        }
907      } else {
908        // We got a definition.
909        sym_addr = static_cast<Elf_Addr>(s->st_value + lsi->load_bias);
910      }
911      count_relocation(kRelocSymbol);
912    } else {
913      s = NULL;
914    }
915
916    switch (type) {
917#if defined(__x86_64__)
918    case R_X86_64_JUMP_SLOT:
919      count_relocation(kRelocAbsolute);
920      MARK(rela->r_offset);
921      TRACE_TYPE(RELO, "RELO JMP_SLOT %08zx <- %08zx %s", static_cast<size_t>(reloc),
922                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
923      *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr + rela->r_addend;
924      break;
925    case R_X86_64_GLOB_DAT:
926      count_relocation(kRelocAbsolute);
927      MARK(rela->r_offset);
928      TRACE_TYPE(RELO, "RELO GLOB_DAT %08zx <- %08zx %s", static_cast<size_t>(reloc),
929                 static_cast<size_t>(sym_addr + rela->r_addend), sym_name);
930      *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr + rela->r_addend;
931      break;
932    case R_X86_64_RELATIVE:
933      count_relocation(kRelocRelative);
934      MARK(rela->r_offset);
935      if (sym) {
936        DL_ERR("odd RELATIVE form...");
937        return -1;
938      }
939      TRACE_TYPE(RELO, "RELO RELATIVE %08zx <- +%08zx", static_cast<size_t>(reloc),
940                 static_cast<size_t>(si->base));
941      *reinterpret_cast<Elf_Addr*>(reloc) = si->base + rela->r_addend;
942      break;
943    case R_X86_64_32:
944      count_relocation(kRelocRelative);
945      MARK(rela->r_offset);
946      TRACE_TYPE(RELO, "RELO R_X86_64_32 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
947                 static_cast<size_t>(sym_addr), sym_name);
948      *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr + rela->r_addend;
949      break;
950    case R_X86_64_64:
951      count_relocation(kRelocRelative);
952      MARK(rela->r_offset);
953      TRACE_TYPE(RELO, "RELO R_X86_64_64 %08zx <- +%08zx %s", static_cast<size_t>(reloc),
954                 static_cast<size_t>(sym_addr), sym_name);
955      *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr + rela->r_addend;
956      break;
957    case R_X86_64_PC32:
958      count_relocation(kRelocRelative);
959      MARK(rela->r_offset);
960      TRACE_TYPE(RELO, "RELO R_X86_64_PC32 %08zx <- +%08zx (%08zx - %08zx) %s",
961                 static_cast<size_t>(reloc), static_cast<size_t>(sym_addr - reloc),
962                 static_cast<size_t>(sym_addr), static_cast<size_t>(reloc), sym_name);
963      *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr + rela->r_addend - reloc;
964      break;
965#endif
966    default:
967      DL_ERR("unknown reloc type %d @ %p (%d)", type, rela, (int) (rela - start));
968      return -1;
969    }
970  }
971  return 0;
972}
973#else
974static int soinfo_relocate(soinfo* si, Elf_Rel* rel, unsigned count,
975                           soinfo* needed[])
976{
977    Elf_Sym* symtab = si->symtab;
978    const char* strtab = si->strtab;
979    Elf_Sym* s;
980    Elf_Rel* start = rel;
981    soinfo* lsi;
982
983    for (size_t idx = 0; idx < count; ++idx, ++rel) {
984        unsigned type = ELF_R_TYPE(rel->r_info);
985        // TODO: don't use unsigned for 'sym'. Use uint32_t or Elf_Addr instead.
986        unsigned sym = ELF_R_SYM(rel->r_info);
987        Elf_Addr reloc = static_cast<Elf_Addr>(rel->r_offset + si->load_bias);
988        Elf_Addr sym_addr = 0;
989        char* sym_name = NULL;
990
991        DEBUG("Processing '%s' relocation at index %zd", si->name, idx);
992        if (type == 0) { // R_*_NONE
993            continue;
994        }
995        if (sym != 0) {
996            sym_name = (char *)(strtab + symtab[sym].st_name);
997            s = soinfo_do_lookup(si, sym_name, &lsi, needed);
998            if (s == NULL) {
999                // We only allow an undefined symbol if this is a weak reference...
1000                s = &symtab[sym];
1001                if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1002                    DL_ERR("cannot locate symbol \"%s\" referenced by \"%s\"...", sym_name, si->name);
1003                    return -1;
1004                }
1005
1006                /* IHI0044C AAELF 4.5.1.1:
1007
1008                   Libraries are not searched to resolve weak references.
1009                   It is not an error for a weak reference to remain
1010                   unsatisfied.
1011
1012                   During linking, the value of an undefined weak reference is:
1013                   - Zero if the relocation type is absolute
1014                   - The address of the place if the relocation is pc-relative
1015                   - The address of nominal base address if the relocation
1016                     type is base-relative.
1017                  */
1018
1019                switch (type) {
1020#if defined(__arm__)
1021                case R_ARM_JUMP_SLOT:
1022                case R_ARM_GLOB_DAT:
1023                case R_ARM_ABS32:
1024                case R_ARM_RELATIVE:    /* Don't care. */
1025                    // sym_addr was initialized to be zero above or relocation
1026                    // code below does not care about value of sym_addr.
1027                    // No need to do anything.
1028                    break;
1029#elif defined(__i386__)
1030                case R_386_JMP_SLOT:
1031                case R_386_GLOB_DAT:
1032                case R_386_32:
1033                case R_386_RELATIVE:    /* Don't care. */
1034                    // sym_addr was initialized to be zero above or relocation
1035                    // code below does not care about value of sym_addr.
1036                    // No need to do anything.
1037                    break;
1038                case R_386_PC32:
1039                    sym_addr = reloc;
1040                    break;
1041#endif
1042
1043#if defined(__arm__)
1044                case R_ARM_COPY:
1045                    // Fall through. Can't really copy if weak symbol is not found at run-time.
1046#endif
1047                default:
1048                    DL_ERR("unknown weak reloc type %d @ %p (%d)", type, rel, (int) (rel - start));
1049                    return -1;
1050                }
1051            } else {
1052                // We got a definition.
1053                sym_addr = static_cast<Elf_Addr>(s->st_value + lsi->load_bias);
1054            }
1055            count_relocation(kRelocSymbol);
1056        } else {
1057            s = NULL;
1058        }
1059
1060        switch (type) {
1061#if defined(__arm__)
1062        case R_ARM_JUMP_SLOT:
1063            count_relocation(kRelocAbsolute);
1064            MARK(rel->r_offset);
1065            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1066            *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr;
1067            break;
1068        case R_ARM_GLOB_DAT:
1069            count_relocation(kRelocAbsolute);
1070            MARK(rel->r_offset);
1071            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1072            *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr;
1073            break;
1074        case R_ARM_ABS32:
1075            count_relocation(kRelocAbsolute);
1076            MARK(rel->r_offset);
1077            TRACE_TYPE(RELO, "RELO ABS %08x <- %08x %s", reloc, sym_addr, sym_name);
1078            *reinterpret_cast<Elf_Addr*>(reloc) += sym_addr;
1079            break;
1080        case R_ARM_REL32:
1081            count_relocation(kRelocRelative);
1082            MARK(rel->r_offset);
1083            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x - %08x %s",
1084                       reloc, sym_addr, rel->r_offset, sym_name);
1085            *reinterpret_cast<Elf_Addr*>(reloc) += sym_addr - rel->r_offset;
1086            break;
1087        case R_ARM_COPY:
1088            if ((si->flags & FLAG_EXE) == 0) {
1089                /*
1090                 * http://infocenter.arm.com/help/topic/com.arm.doc.ihi0044d/IHI0044D_aaelf.pdf
1091                 *
1092                 * Section 4.7.1.10 "Dynamic relocations"
1093                 * R_ARM_COPY may only appear in executable objects where e_type is
1094                 * set to ET_EXEC.
1095                 *
1096                 * TODO: FLAG_EXE is set for both ET_DYN and ET_EXEC executables.
1097                 * We should explicitly disallow ET_DYN executables from having
1098                 * R_ARM_COPY relocations.
1099                 */
1100                DL_ERR("%s R_ARM_COPY relocations only supported for ET_EXEC", si->name);
1101                return -1;
1102            }
1103            count_relocation(kRelocCopy);
1104            MARK(rel->r_offset);
1105            TRACE_TYPE(RELO, "RELO %08x <- %d @ %08x %s", reloc, s->st_size, sym_addr, sym_name);
1106            if (reloc == sym_addr) {
1107                Elf_Sym *src = soinfo_do_lookup(NULL, sym_name, &lsi, needed);
1108
1109                if (src == NULL) {
1110                    DL_ERR("%s R_ARM_COPY relocation source cannot be resolved", si->name);
1111                    return -1;
1112                }
1113                if (lsi->has_DT_SYMBOLIC) {
1114                    DL_ERR("%s invalid R_ARM_COPY relocation against DT_SYMBOLIC shared "
1115                           "library %s (built with -Bsymbolic?)", si->name, lsi->name);
1116                    return -1;
1117                }
1118                if (s->st_size < src->st_size) {
1119                    DL_ERR("%s R_ARM_COPY relocation size mismatch (%d < %d)",
1120                           si->name, s->st_size, src->st_size);
1121                    return -1;
1122                }
1123                memcpy((void*)reloc, (void*)(src->st_value + lsi->load_bias), src->st_size);
1124            } else {
1125                DL_ERR("%s R_ARM_COPY relocation target cannot be resolved", si->name);
1126                return -1;
1127            }
1128            break;
1129#elif defined(__i386__)
1130        case R_386_JMP_SLOT:
1131            count_relocation(kRelocAbsolute);
1132            MARK(rel->r_offset);
1133            TRACE_TYPE(RELO, "RELO JMP_SLOT %08x <- %08x %s", reloc, sym_addr, sym_name);
1134            *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr;
1135            break;
1136        case R_386_GLOB_DAT:
1137            count_relocation(kRelocAbsolute);
1138            MARK(rel->r_offset);
1139            TRACE_TYPE(RELO, "RELO GLOB_DAT %08x <- %08x %s", reloc, sym_addr, sym_name);
1140            *reinterpret_cast<Elf_Addr*>(reloc) = sym_addr;
1141            break;
1142        case R_386_32:
1143            count_relocation(kRelocRelative);
1144            MARK(rel->r_offset);
1145            TRACE_TYPE(RELO, "RELO R_386_32 %08x <- +%08x %s", reloc, sym_addr, sym_name);
1146            *reinterpret_cast<Elf_Addr*>(reloc) += sym_addr;
1147            break;
1148        case R_386_PC32:
1149            count_relocation(kRelocRelative);
1150            MARK(rel->r_offset);
1151            TRACE_TYPE(RELO, "RELO R_386_PC32 %08x <- +%08x (%08x - %08x) %s",
1152                       reloc, (sym_addr - reloc), sym_addr, reloc, sym_name);
1153            *reinterpret_cast<Elf_Addr*>(reloc) += (sym_addr - reloc);
1154            break;
1155#elif defined(__mips__)
1156        case R_MIPS_REL32:
1157            count_relocation(kRelocAbsolute);
1158            MARK(rel->r_offset);
1159            TRACE_TYPE(RELO, "RELO REL32 %08x <- %08x %s",
1160                       reloc, sym_addr, (sym_name) ? sym_name : "*SECTIONHDR*");
1161            if (s) {
1162                *reinterpret_cast<Elf_Addr*>(reloc) += sym_addr;
1163            } else {
1164                *reinterpret_cast<Elf_Addr*>(reloc) += si->base;
1165            }
1166            break;
1167#endif
1168
1169#if defined(__arm__)
1170        case R_ARM_RELATIVE:
1171#elif defined(__i386__)
1172        case R_386_RELATIVE:
1173#endif
1174            count_relocation(kRelocRelative);
1175            MARK(rel->r_offset);
1176            if (sym) {
1177                DL_ERR("odd RELATIVE form...");
1178                return -1;
1179            }
1180            TRACE_TYPE(RELO, "RELO RELATIVE %p <- +%p",
1181                       reinterpret_cast<void*>(reloc), reinterpret_cast<void*>(si->base));
1182            *reinterpret_cast<Elf_Addr*>(reloc) += si->base;
1183            break;
1184
1185        default:
1186            DL_ERR("unknown reloc type %d @ %p (%d)", type, rel, (int) (rel - start));
1187            return -1;
1188        }
1189    }
1190    return 0;
1191}
1192#endif
1193
1194#if defined(__mips__)
1195static bool mips_relocate_got(soinfo* si, soinfo* needed[]) {
1196    unsigned* got = si->plt_got;
1197    if (got == NULL) {
1198        return true;
1199    }
1200    unsigned local_gotno = si->mips_local_gotno;
1201    unsigned gotsym = si->mips_gotsym;
1202    unsigned symtabno = si->mips_symtabno;
1203    Elf_Sym* symtab = si->symtab;
1204
1205    /*
1206     * got[0] is address of lazy resolver function
1207     * got[1] may be used for a GNU extension
1208     * set it to a recognizable address in case someone calls it
1209     * (should be _rtld_bind_start)
1210     * FIXME: maybe this should be in a separate routine
1211     */
1212
1213    if ((si->flags & FLAG_LINKER) == 0) {
1214        size_t g = 0;
1215        got[g++] = 0xdeadbeef;
1216        if (got[g] & 0x80000000) {
1217            got[g++] = 0xdeadfeed;
1218        }
1219        /*
1220         * Relocate the local GOT entries need to be relocated
1221         */
1222        for (; g < local_gotno; g++) {
1223            got[g] += si->load_bias;
1224        }
1225    }
1226
1227    /* Now for the global GOT entries */
1228    Elf_Sym* sym = symtab + gotsym;
1229    got = si->plt_got + local_gotno;
1230    for (size_t g = gotsym; g < symtabno; g++, sym++, got++) {
1231        const char* sym_name;
1232        Elf_Sym* s;
1233        soinfo* lsi;
1234
1235        /* This is an undefined reference... try to locate it */
1236        sym_name = si->strtab + sym->st_name;
1237        s = soinfo_do_lookup(si, sym_name, &lsi, needed);
1238        if (s == NULL) {
1239            /* We only allow an undefined symbol if this is a weak
1240               reference..   */
1241            s = &symtab[g];
1242            if (ELF_ST_BIND(s->st_info) != STB_WEAK) {
1243                DL_ERR("cannot locate \"%s\"...", sym_name);
1244                return false;
1245            }
1246            *got = 0;
1247        }
1248        else {
1249            /* FIXME: is this sufficient?
1250             * For reference see NetBSD link loader
1251             * 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
1252             */
1253             *got = lsi->load_bias + s->st_value;
1254        }
1255    }
1256    return true;
1257}
1258#endif
1259
1260void soinfo::CallArray(const char* array_name UNUSED, linker_function_t* functions, size_t count, bool reverse) {
1261  if (functions == NULL) {
1262    return;
1263  }
1264
1265  TRACE("[ Calling %s (size %zd) @ %p for '%s' ]", array_name, count, functions, name);
1266
1267  int begin = reverse ? (count - 1) : 0;
1268  int end = reverse ? -1 : count;
1269  int step = reverse ? -1 : 1;
1270
1271  for (int i = begin; i != end; i += step) {
1272    TRACE("[ %s[%d] == %p ]", array_name, i, functions[i]);
1273    CallFunction("function", functions[i]);
1274  }
1275
1276  TRACE("[ Done calling %s for '%s' ]", array_name, name);
1277}
1278
1279void soinfo::CallFunction(const char* function_name UNUSED, linker_function_t function) {
1280  if (function == NULL || reinterpret_cast<uintptr_t>(function) == static_cast<uintptr_t>(-1)) {
1281    return;
1282  }
1283
1284  TRACE("[ Calling %s @ %p for '%s' ]", function_name, function, name);
1285  function();
1286  TRACE("[ Done calling %s @ %p for '%s' ]", function_name, function, name);
1287
1288  // The function may have called dlopen(3) or dlclose(3), so we need to ensure our data structures
1289  // are still writable. This happens with our debug malloc (see http://b/7941716).
1290  set_soinfo_pool_protection(PROT_READ | PROT_WRITE);
1291}
1292
1293void soinfo::CallPreInitConstructors() {
1294  // DT_PREINIT_ARRAY functions are called before any other constructors for executables,
1295  // but ignored in a shared library.
1296  CallArray("DT_PREINIT_ARRAY", preinit_array, preinit_array_count, false);
1297}
1298
1299void soinfo::CallConstructors() {
1300  if (constructors_called) {
1301    return;
1302  }
1303
1304  // We set constructors_called before actually calling the constructors, otherwise it doesn't
1305  // protect against recursive constructor calls. One simple example of constructor recursion
1306  // is the libc debug malloc, which is implemented in libc_malloc_debug_leak.so:
1307  // 1. The program depends on libc, so libc's constructor is called here.
1308  // 2. The libc constructor calls dlopen() to load libc_malloc_debug_leak.so.
1309  // 3. dlopen() calls the constructors on the newly created
1310  //    soinfo for libc_malloc_debug_leak.so.
1311  // 4. The debug .so depends on libc, so CallConstructors is
1312  //    called again with the libc soinfo. If it doesn't trigger the early-
1313  //    out above, the libc constructor will be called again (recursively!).
1314  constructors_called = true;
1315
1316  if ((flags & FLAG_EXE) == 0 && preinit_array != NULL) {
1317    // The GNU dynamic linker silently ignores these, but we warn the developer.
1318    PRINT("\"%s\": ignoring %zd-entry DT_PREINIT_ARRAY in shared library!",
1319          name, preinit_array_count);
1320  }
1321
1322  if (dynamic != NULL) {
1323    for (Elf_Dyn* d = dynamic; d->d_tag != DT_NULL; ++d) {
1324      if (d->d_tag == DT_NEEDED) {
1325        const char* library_name = strtab + d->d_un.d_val;
1326        TRACE("\"%s\": calling constructors in DT_NEEDED \"%s\"", name, library_name);
1327        find_loaded_library(library_name)->CallConstructors();
1328      }
1329    }
1330  }
1331
1332  TRACE("\"%s\": calling constructors", name);
1333
1334  // DT_INIT should be called before DT_INIT_ARRAY if both are present.
1335  CallFunction("DT_INIT", init_func);
1336  CallArray("DT_INIT_ARRAY", init_array, init_array_count, false);
1337}
1338
1339void soinfo::CallDestructors() {
1340  TRACE("\"%s\": calling destructors", name);
1341
1342  // DT_FINI_ARRAY must be parsed in reverse order.
1343  CallArray("DT_FINI_ARRAY", fini_array, fini_array_count, true);
1344
1345  // DT_FINI should be called after DT_FINI_ARRAY if both are present.
1346  CallFunction("DT_FINI", fini_func);
1347}
1348
1349/* Force any of the closed stdin, stdout and stderr to be associated with
1350   /dev/null. */
1351static int nullify_closed_stdio() {
1352    int dev_null, i, status;
1353    int return_value = 0;
1354
1355    dev_null = TEMP_FAILURE_RETRY(open("/dev/null", O_RDWR));
1356    if (dev_null < 0) {
1357        DL_ERR("cannot open /dev/null: %s", strerror(errno));
1358        return -1;
1359    }
1360    TRACE("[ Opened /dev/null file-descriptor=%d]", dev_null);
1361
1362    /* If any of the stdio file descriptors is valid and not associated
1363       with /dev/null, dup /dev/null to it.  */
1364    for (i = 0; i < 3; i++) {
1365        /* If it is /dev/null already, we are done. */
1366        if (i == dev_null) {
1367            continue;
1368        }
1369
1370        TRACE("[ Nullifying stdio file descriptor %d]", i);
1371        status = TEMP_FAILURE_RETRY(fcntl(i, F_GETFL));
1372
1373        /* If file is opened, we are good. */
1374        if (status != -1) {
1375            continue;
1376        }
1377
1378        /* The only error we allow is that the file descriptor does not
1379           exist, in which case we dup /dev/null to it. */
1380        if (errno != EBADF) {
1381            DL_ERR("fcntl failed: %s", strerror(errno));
1382            return_value = -1;
1383            continue;
1384        }
1385
1386        /* Try dupping /dev/null to this stdio file descriptor and
1387           repeat if there is a signal.  Note that any errors in closing
1388           the stdio descriptor are lost.  */
1389        status = TEMP_FAILURE_RETRY(dup2(dev_null, i));
1390        if (status < 0) {
1391            DL_ERR("dup2 failed: %s", strerror(errno));
1392            return_value = -1;
1393            continue;
1394        }
1395    }
1396
1397    /* If /dev/null is not one of the stdio file descriptors, close it. */
1398    if (dev_null > 2) {
1399        TRACE("[ Closing /dev/null file-descriptor=%d]", dev_null);
1400        status = TEMP_FAILURE_RETRY(close(dev_null));
1401        if (status == -1) {
1402            DL_ERR("close failed: %s", strerror(errno));
1403            return_value = -1;
1404        }
1405    }
1406
1407    return return_value;
1408}
1409
1410static bool soinfo_link_image(soinfo* si) {
1411    /* "base" might wrap around UINT32_MAX. */
1412    Elf_Addr base = si->load_bias;
1413    const Elf_Phdr *phdr = si->phdr;
1414    int phnum = si->phnum;
1415    bool relocating_linker = (si->flags & FLAG_LINKER) != 0;
1416
1417    /* We can't debug anything until the linker is relocated */
1418    if (!relocating_linker) {
1419        INFO("[ linking %s ]", si->name);
1420        DEBUG("si->base = %p si->flags = 0x%08x", reinterpret_cast<void*>(si->base), si->flags);
1421    }
1422
1423    /* Extract dynamic section */
1424    size_t dynamic_count;
1425    Elf_Word dynamic_flags;
1426    phdr_table_get_dynamic_section(phdr, phnum, base, &si->dynamic,
1427                                   &dynamic_count, &dynamic_flags);
1428    if (si->dynamic == NULL) {
1429        if (!relocating_linker) {
1430            DL_ERR("missing PT_DYNAMIC in \"%s\"", si->name);
1431        }
1432        return false;
1433    } else {
1434        if (!relocating_linker) {
1435            DEBUG("dynamic = %p", si->dynamic);
1436        }
1437    }
1438
1439#if defined(__arm__)
1440    (void) phdr_table_get_arm_exidx(phdr, phnum, base,
1441                                    &si->ARM_exidx, &si->ARM_exidx_count);
1442#endif
1443
1444    // Extract useful information from dynamic section.
1445    uint32_t needed_count = 0;
1446    for (Elf_Dyn* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1447        DEBUG("d = %p, d[0](tag) = %p d[1](val) = %p",
1448              d, reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1449        switch (d->d_tag) {
1450        case DT_HASH:
1451            si->nbucket = ((unsigned *) (base + d->d_un.d_ptr))[0];
1452            si->nchain = ((unsigned *) (base + d->d_un.d_ptr))[1];
1453            si->bucket = (unsigned *) (base + d->d_un.d_ptr + 8);
1454            si->chain = (unsigned *) (base + d->d_un.d_ptr + 8 + si->nbucket * 4);
1455            break;
1456        case DT_STRTAB:
1457            si->strtab = (const char *) (base + d->d_un.d_ptr);
1458            break;
1459        case DT_SYMTAB:
1460            si->symtab = (Elf_Sym *) (base + d->d_un.d_ptr);
1461            break;
1462#if !defined(__LP64__)
1463        case DT_PLTREL:
1464            if (d->d_un.d_val != DT_REL) {
1465                DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1466                return false;
1467            }
1468            break;
1469#endif
1470        case DT_JMPREL:
1471#if defined(USE_RELA)
1472            si->plt_rela = (Elf_Rela*) (base + d->d_un.d_ptr);
1473#else
1474            si->plt_rel = (Elf_Rel*) (base + d->d_un.d_ptr);
1475#endif
1476            break;
1477        case DT_PLTRELSZ:
1478#if defined(USE_RELA)
1479            si->plt_rela_count = d->d_un.d_val / sizeof(Elf_Rela);
1480#else
1481            si->plt_rel_count = d->d_un.d_val / sizeof(Elf_Rel);
1482#endif
1483            break;
1484#if !defined(__LP64__)
1485        case DT_PLTGOT:
1486            // Used by 32-bit MIPS.
1487            si->plt_got = (unsigned *)(base + d->d_un.d_ptr);
1488            break;
1489#endif
1490        case DT_DEBUG:
1491            // Set the DT_DEBUG entry to the address of _r_debug for GDB
1492            // if the dynamic table is writable
1493            if ((dynamic_flags & PF_W) != 0) {
1494                d->d_un.d_val = reinterpret_cast<uintptr_t>(&_r_debug);
1495            }
1496            break;
1497#if defined(USE_RELA)
1498         case DT_RELA:
1499            si->rela = (Elf_Rela*) (base + d->d_un.d_ptr);
1500            break;
1501         case DT_RELASZ:
1502            si->rela_count = d->d_un.d_val / sizeof(Elf_Rela);
1503            break;
1504        case DT_REL:
1505            DL_ERR("unsupported DT_REL in \"%s\"", si->name);
1506            return false;
1507        case DT_RELSZ:
1508            DL_ERR("unsupported DT_RELSZ in \"%s\"", si->name);
1509            return false;
1510#else
1511        case DT_REL:
1512            si->rel = (Elf_Rel*) (base + d->d_un.d_ptr);
1513            break;
1514        case DT_RELSZ:
1515            si->rel_count = d->d_un.d_val / sizeof(Elf_Rel);
1516            break;
1517         case DT_RELA:
1518            DL_ERR("unsupported DT_RELA in \"%s\"", si->name);
1519            return false;
1520#endif
1521        case DT_INIT:
1522            si->init_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1523            DEBUG("%s constructors (DT_INIT) found at %p", si->name, si->init_func);
1524            break;
1525        case DT_FINI:
1526            si->fini_func = reinterpret_cast<linker_function_t>(base + d->d_un.d_ptr);
1527            DEBUG("%s destructors (DT_FINI) found at %p", si->name, si->fini_func);
1528            break;
1529        case DT_INIT_ARRAY:
1530            si->init_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1531            DEBUG("%s constructors (DT_INIT_ARRAY) found at %p", si->name, si->init_array);
1532            break;
1533        case DT_INIT_ARRAYSZ:
1534            si->init_array_count = ((unsigned)d->d_un.d_val) / sizeof(Elf_Addr);
1535            break;
1536        case DT_FINI_ARRAY:
1537            si->fini_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1538            DEBUG("%s destructors (DT_FINI_ARRAY) found at %p", si->name, si->fini_array);
1539            break;
1540        case DT_FINI_ARRAYSZ:
1541            si->fini_array_count = ((unsigned)d->d_un.d_val) / sizeof(Elf_Addr);
1542            break;
1543        case DT_PREINIT_ARRAY:
1544            si->preinit_array = reinterpret_cast<linker_function_t*>(base + d->d_un.d_ptr);
1545            DEBUG("%s constructors (DT_PREINIT_ARRAY) found at %p", si->name, si->preinit_array);
1546            break;
1547        case DT_PREINIT_ARRAYSZ:
1548            si->preinit_array_count = ((unsigned)d->d_un.d_val) / sizeof(Elf_Addr);
1549            break;
1550        case DT_TEXTREL:
1551#if defined(__LP64__)
1552            DL_ERR("text relocations (DT_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1553            return false;
1554#else
1555            si->has_text_relocations = true;
1556            break;
1557#endif
1558        case DT_SYMBOLIC:
1559            si->has_DT_SYMBOLIC = true;
1560            break;
1561        case DT_NEEDED:
1562            ++needed_count;
1563            break;
1564        case DT_FLAGS:
1565            if (d->d_un.d_val & DF_TEXTREL) {
1566#if defined(__LP64__)
1567                DL_ERR("text relocations (DF_TEXTREL) found in 64-bit ELF file \"%s\"", si->name);
1568                return false;
1569#else
1570                si->has_text_relocations = true;
1571#endif
1572            }
1573            if (d->d_un.d_val & DF_SYMBOLIC) {
1574                si->has_DT_SYMBOLIC = true;
1575            }
1576            break;
1577#if defined(__mips__)
1578        case DT_STRSZ:
1579        case DT_SYMENT:
1580        case DT_RELENT:
1581             break;
1582        case DT_MIPS_RLD_MAP:
1583            // Set the DT_MIPS_RLD_MAP entry to the address of _r_debug for GDB.
1584            {
1585              r_debug** dp = (r_debug**) d->d_un.d_ptr;
1586              *dp = &_r_debug;
1587            }
1588            break;
1589        case DT_MIPS_RLD_VERSION:
1590        case DT_MIPS_FLAGS:
1591        case DT_MIPS_BASE_ADDRESS:
1592        case DT_MIPS_UNREFEXTNO:
1593            break;
1594
1595        case DT_MIPS_SYMTABNO:
1596            si->mips_symtabno = d->d_un.d_val;
1597            break;
1598
1599        case DT_MIPS_LOCAL_GOTNO:
1600            si->mips_local_gotno = d->d_un.d_val;
1601            break;
1602
1603        case DT_MIPS_GOTSYM:
1604            si->mips_gotsym = d->d_un.d_val;
1605            break;
1606#endif
1607
1608        default:
1609            DEBUG("Unused DT entry: type %p arg %p",
1610                  reinterpret_cast<void*>(d->d_tag), reinterpret_cast<void*>(d->d_un.d_val));
1611            break;
1612        }
1613    }
1614
1615    DEBUG("si->base = %p, si->strtab = %p, si->symtab = %p",
1616          reinterpret_cast<void*>(si->base), si->strtab, si->symtab);
1617
1618    // Sanity checks.
1619    if (relocating_linker && needed_count != 0) {
1620        DL_ERR("linker cannot have DT_NEEDED dependencies on other libraries");
1621        return false;
1622    }
1623    if (si->nbucket == 0) {
1624        DL_ERR("empty/missing DT_HASH in \"%s\" (built with --hash-style=gnu?)", si->name);
1625        return false;
1626    }
1627    if (si->strtab == 0) {
1628        DL_ERR("empty/missing DT_STRTAB in \"%s\"", si->name);
1629        return false;
1630    }
1631    if (si->symtab == 0) {
1632        DL_ERR("empty/missing DT_SYMTAB in \"%s\"", si->name);
1633        return false;
1634    }
1635
1636    // If this is the main executable, then load all of the libraries from LD_PRELOAD now.
1637    if (si->flags & FLAG_EXE) {
1638        memset(gLdPreloads, 0, sizeof(gLdPreloads));
1639        size_t preload_count = 0;
1640        for (size_t i = 0; gLdPreloadNames[i] != NULL; i++) {
1641            soinfo* lsi = find_library(gLdPreloadNames[i]);
1642            if (lsi != NULL) {
1643                gLdPreloads[preload_count++] = lsi;
1644            } else {
1645                // As with glibc, failure to load an LD_PRELOAD library is just a warning.
1646                DL_WARN("could not load library \"%s\" from LD_PRELOAD for \"%s\"; caused by %s",
1647                        gLdPreloadNames[i], si->name, linker_get_error_buffer());
1648            }
1649        }
1650    }
1651
1652    soinfo** needed = (soinfo**) alloca((1 + needed_count) * sizeof(soinfo*));
1653    soinfo** pneeded = needed;
1654
1655    for (Elf_Dyn* d = si->dynamic; d->d_tag != DT_NULL; ++d) {
1656        if (d->d_tag == DT_NEEDED) {
1657            const char* library_name = si->strtab + d->d_un.d_val;
1658            DEBUG("%s needs %s", si->name, library_name);
1659            soinfo* lsi = find_library(library_name);
1660            if (lsi == NULL) {
1661                strlcpy(tmp_err_buf, linker_get_error_buffer(), sizeof(tmp_err_buf));
1662                DL_ERR("could not load library \"%s\" needed by \"%s\"; caused by %s",
1663                       library_name, si->name, tmp_err_buf);
1664                return false;
1665            }
1666            *pneeded++ = lsi;
1667        }
1668    }
1669    *pneeded = NULL;
1670
1671#if !defined(__LP64__)
1672    if (si->has_text_relocations) {
1673        // Make segments writable to allow text relocations to work properly. We will later call
1674        // phdr_table_protect_segments() after all of them are applied and all constructors are run.
1675        DL_WARN("%s has text relocations. This is wasting memory and prevents "
1676                "security hardening. Please fix.", si->name);
1677        if (phdr_table_unprotect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1678            DL_ERR("can't unprotect loadable segments for \"%s\": %s",
1679                   si->name, strerror(errno));
1680            return false;
1681        }
1682    }
1683#endif
1684
1685#if defined(USE_RELA)
1686    if (si->plt_rela != NULL) {
1687        DEBUG("[ relocating %s plt ]\n", si->name );
1688        if (soinfo_relocate_a(si, si->plt_rela, si->plt_rela_count, needed)) {
1689            return false;
1690        }
1691    }
1692    if (si->rela != NULL) {
1693        DEBUG("[ relocating %s ]\n", si->name );
1694        if (soinfo_relocate_a(si, si->rela, si->rela_count, needed)) {
1695            return false;
1696        }
1697    }
1698#else
1699    if (si->plt_rel != NULL) {
1700        DEBUG("[ relocating %s plt ]", si->name );
1701        if (soinfo_relocate(si, si->plt_rel, si->plt_rel_count, needed)) {
1702            return false;
1703        }
1704    }
1705    if (si->rel != NULL) {
1706        DEBUG("[ relocating %s ]", si->name );
1707        if (soinfo_relocate(si, si->rel, si->rel_count, needed)) {
1708            return false;
1709        }
1710    }
1711#endif
1712
1713#if defined(__mips__)
1714    if (!mips_relocate_got(si, needed)) {
1715        return false;
1716    }
1717#endif
1718
1719    si->flags |= FLAG_LINKED;
1720    DEBUG("[ finished linking %s ]", si->name);
1721
1722#if !defined(__LP64__)
1723    if (si->has_text_relocations) {
1724        // All relocations are done, we can protect our segments back to read-only.
1725        if (phdr_table_protect_segments(si->phdr, si->phnum, si->load_bias) < 0) {
1726            DL_ERR("can't protect segments for \"%s\": %s",
1727                   si->name, strerror(errno));
1728            return false;
1729        }
1730    }
1731#endif
1732
1733    /* We can also turn on GNU RELRO protection */
1734    if (phdr_table_protect_gnu_relro(si->phdr, si->phnum, si->load_bias) < 0) {
1735        DL_ERR("can't enable GNU RELRO protection for \"%s\": %s",
1736               si->name, strerror(errno));
1737        return false;
1738    }
1739
1740    notify_gdb_of_load(si);
1741    return true;
1742}
1743
1744/*
1745 * This function add vdso to internal dso list.
1746 * It helps to stack unwinding through signal handlers.
1747 * Also, it makes bionic more like glibc.
1748 */
1749static void add_vdso(KernelArgumentBlock& args UNUSED) {
1750#if defined(AT_SYSINFO_EHDR)
1751    Elf_Ehdr* ehdr_vdso = reinterpret_cast<Elf_Ehdr*>(args.getauxval(AT_SYSINFO_EHDR));
1752
1753    soinfo* si = soinfo_alloc("[vdso]");
1754    si->phdr = reinterpret_cast<Elf_Phdr*>(reinterpret_cast<char*>(ehdr_vdso) + ehdr_vdso->e_phoff);
1755    si->phnum = ehdr_vdso->e_phnum;
1756    si->link_map.l_name = si->name;
1757    for (size_t i = 0; i < si->phnum; ++i) {
1758        if (si->phdr[i].p_type == PT_LOAD) {
1759            si->link_map.l_addr = reinterpret_cast<Elf_Addr>(ehdr_vdso) - si->phdr[i].p_vaddr;
1760            break;
1761        }
1762    }
1763#endif
1764}
1765
1766/*
1767 * This code is called after the linker has linked itself and
1768 * fixed it's own GOT. It is safe to make references to externs
1769 * and other non-local data at this point.
1770 */
1771static Elf_Addr __linker_init_post_relocation(KernelArgumentBlock& args, Elf_Addr linker_base) {
1772    /* NOTE: we store the args pointer on a special location
1773     *       of the temporary TLS area in order to pass it to
1774     *       the C Library's runtime initializer.
1775     *
1776     *       The initializer must clear the slot and reset the TLS
1777     *       to point to a different location to ensure that no other
1778     *       shared library constructor can access it.
1779     */
1780  __libc_init_tls(args);
1781
1782#if TIMING
1783    struct timeval t0, t1;
1784    gettimeofday(&t0, 0);
1785#endif
1786
1787    // Initialize environment functions, and get to the ELF aux vectors table.
1788    linker_env_init(args);
1789
1790    // If this is a setuid/setgid program, close the security hole described in
1791    // ftp://ftp.freebsd.org/pub/FreeBSD/CERT/advisories/FreeBSD-SA-02:23.stdio.asc
1792    if (get_AT_SECURE()) {
1793        nullify_closed_stdio();
1794    }
1795
1796    debuggerd_init();
1797
1798    // Get a few environment variables.
1799    const char* LD_DEBUG = linker_env_get("LD_DEBUG");
1800    if (LD_DEBUG != NULL) {
1801      gLdDebugVerbosity = atoi(LD_DEBUG);
1802    }
1803
1804    // Normally, these are cleaned by linker_env_init, but the test
1805    // doesn't cost us anything.
1806    const char* ldpath_env = NULL;
1807    const char* ldpreload_env = NULL;
1808    if (!get_AT_SECURE()) {
1809      ldpath_env = linker_env_get("LD_LIBRARY_PATH");
1810      ldpreload_env = linker_env_get("LD_PRELOAD");
1811    }
1812
1813    INFO("[ android linker & debugger ]");
1814
1815    soinfo* si = soinfo_alloc(args.argv[0]);
1816    if (si == NULL) {
1817        exit(EXIT_FAILURE);
1818    }
1819
1820    /* bootstrap the link map, the main exe always needs to be first */
1821    si->flags |= FLAG_EXE;
1822    link_map_t* map = &(si->link_map);
1823
1824    map->l_addr = 0;
1825    map->l_name = args.argv[0];
1826    map->l_prev = NULL;
1827    map->l_next = NULL;
1828
1829    _r_debug.r_map = map;
1830    r_debug_tail = map;
1831
1832    /* gdb expects the linker to be in the debug shared object list.
1833     * Without this, gdb has trouble locating the linker's ".text"
1834     * and ".plt" sections. Gdb could also potentially use this to
1835     * relocate the offset of our exported 'rtld_db_dlactivity' symbol.
1836     * Don't use soinfo_alloc(), because the linker shouldn't
1837     * be on the soinfo list.
1838     */
1839    {
1840        static soinfo linker_soinfo;
1841#if defined(__LP64__)
1842        strlcpy(linker_soinfo.name, "/system/bin/linker64", sizeof(linker_soinfo.name));
1843#else
1844        strlcpy(linker_soinfo.name, "/system/bin/linker", sizeof(linker_soinfo.name));
1845#endif
1846        linker_soinfo.flags = 0;
1847        linker_soinfo.base = linker_base;
1848
1849        /*
1850         * Set the dynamic field in the link map otherwise gdb will complain with
1851         * the following:
1852         *   warning: .dynamic section for "/system/bin/linker" is not at the
1853         *   expected address (wrong library or version mismatch?)
1854         */
1855        Elf_Ehdr *elf_hdr = (Elf_Ehdr *) linker_base;
1856        Elf_Phdr *phdr = (Elf_Phdr*)((unsigned char*) linker_base + elf_hdr->e_phoff);
1857        phdr_table_get_dynamic_section(phdr, elf_hdr->e_phnum, linker_base,
1858                                       &linker_soinfo.dynamic, NULL, NULL);
1859        insert_soinfo_into_debug_map(&linker_soinfo);
1860    }
1861
1862    // Extract information passed from the kernel.
1863    si->phdr = reinterpret_cast<Elf_Phdr*>(args.getauxval(AT_PHDR));
1864    si->phnum = args.getauxval(AT_PHNUM);
1865    si->entry = args.getauxval(AT_ENTRY);
1866
1867    /* Compute the value of si->base. We can't rely on the fact that
1868     * the first entry is the PHDR because this will not be true
1869     * for certain executables (e.g. some in the NDK unit test suite)
1870     */
1871    si->base = 0;
1872    si->size = phdr_table_get_load_size(si->phdr, si->phnum);
1873    si->load_bias = 0;
1874    for (size_t i = 0; i < si->phnum; ++i) {
1875      if (si->phdr[i].p_type == PT_PHDR) {
1876        si->load_bias = reinterpret_cast<Elf_Addr>(si->phdr) - si->phdr[i].p_vaddr;
1877        si->base = reinterpret_cast<Elf_Addr>(si->phdr) - si->phdr[i].p_offset;
1878        break;
1879      }
1880    }
1881    si->dynamic = NULL;
1882    si->ref_count = 1;
1883
1884    // Use LD_LIBRARY_PATH and LD_PRELOAD (but only if we aren't setuid/setgid).
1885    parse_LD_LIBRARY_PATH(ldpath_env);
1886    parse_LD_PRELOAD(ldpreload_env);
1887
1888    somain = si;
1889
1890    if (!soinfo_link_image(si)) {
1891        __libc_format_fd(2, "CANNOT LINK EXECUTABLE: %s\n", linker_get_error_buffer());
1892        exit(EXIT_FAILURE);
1893    }
1894
1895    add_vdso(args);
1896
1897    si->CallPreInitConstructors();
1898
1899    for (size_t i = 0; gLdPreloads[i] != NULL; ++i) {
1900        gLdPreloads[i]->CallConstructors();
1901    }
1902
1903    /* After the link_image, the si->load_bias is initialized.
1904     * For so lib, the map->l_addr will be updated in notify_gdb_of_load.
1905     * We need to update this value for so exe here. So Unwind_Backtrace
1906     * for some arch like x86 could work correctly within so exe.
1907     */
1908    map->l_addr = si->load_bias;
1909    si->CallConstructors();
1910
1911#if TIMING
1912    gettimeofday(&t1,NULL);
1913    PRINT("LINKER TIME: %s: %d microseconds", args.argv[0], (int) (
1914               (((long long)t1.tv_sec * 1000000LL) + (long long)t1.tv_usec) -
1915               (((long long)t0.tv_sec * 1000000LL) + (long long)t0.tv_usec)
1916               ));
1917#endif
1918#if STATS
1919    PRINT("RELO STATS: %s: %d abs, %d rel, %d copy, %d symbol", args.argv[0],
1920           linker_stats.count[kRelocAbsolute],
1921           linker_stats.count[kRelocRelative],
1922           linker_stats.count[kRelocCopy],
1923           linker_stats.count[kRelocSymbol]);
1924#endif
1925#if COUNT_PAGES
1926    {
1927        unsigned n;
1928        unsigned i;
1929        unsigned count = 0;
1930        for (n = 0; n < 4096; n++) {
1931            if (bitmask[n]) {
1932                unsigned x = bitmask[n];
1933                for (i = 0; i < 8; i++) {
1934                    if (x & 1) {
1935                        count++;
1936                    }
1937                    x >>= 1;
1938                }
1939            }
1940        }
1941        PRINT("PAGES MODIFIED: %s: %d (%dKB)", args.argv[0], count, count * 4);
1942    }
1943#endif
1944
1945#if TIMING || STATS || COUNT_PAGES
1946    fflush(stdout);
1947#endif
1948
1949    TRACE("[ Ready to execute '%s' @ %p ]", si->name, reinterpret_cast<void*>(si->entry));
1950    return si->entry;
1951}
1952
1953/* Compute the load-bias of an existing executable. This shall only
1954 * be used to compute the load bias of an executable or shared library
1955 * that was loaded by the kernel itself.
1956 *
1957 * Input:
1958 *    elf    -> address of ELF header, assumed to be at the start of the file.
1959 * Return:
1960 *    load bias, i.e. add the value of any p_vaddr in the file to get
1961 *    the corresponding address in memory.
1962 */
1963static Elf_Addr get_elf_exec_load_bias(const Elf_Ehdr* elf) {
1964  Elf_Addr offset = elf->e_phoff;
1965  const Elf_Phdr* phdr_table = (const Elf_Phdr*)((char*)elf + offset);
1966  const Elf_Phdr* phdr_end = phdr_table + elf->e_phnum;
1967
1968  for (const Elf_Phdr* phdr = phdr_table; phdr < phdr_end; phdr++) {
1969    if (phdr->p_type == PT_LOAD) {
1970      return reinterpret_cast<Elf_Addr>(elf) + phdr->p_offset - phdr->p_vaddr;
1971    }
1972  }
1973  return 0;
1974}
1975
1976/*
1977 * This is the entry point for the linker, called from begin.S. This
1978 * method is responsible for fixing the linker's own relocations, and
1979 * then calling __linker_init_post_relocation().
1980 *
1981 * Because this method is called before the linker has fixed it's own
1982 * relocations, any attempt to reference an extern variable, extern
1983 * function, or other GOT reference will generate a segfault.
1984 */
1985extern "C" Elf_Addr __linker_init(void* raw_args) {
1986  KernelArgumentBlock args(raw_args);
1987
1988  Elf_Addr linker_addr = args.getauxval(AT_BASE);
1989
1990  Elf_Ehdr* elf_hdr = reinterpret_cast<Elf_Ehdr*>(linker_addr);
1991  Elf_Phdr* phdr = (Elf_Phdr*)((unsigned char*) linker_addr + elf_hdr->e_phoff);
1992
1993  soinfo linker_so;
1994  memset(&linker_so, 0, sizeof(soinfo));
1995
1996  linker_so.base = linker_addr;
1997  linker_so.size = phdr_table_get_load_size(phdr, elf_hdr->e_phnum);
1998  linker_so.load_bias = get_elf_exec_load_bias(elf_hdr);
1999  linker_so.dynamic = NULL;
2000  linker_so.phdr = phdr;
2001  linker_so.phnum = elf_hdr->e_phnum;
2002  linker_so.flags |= FLAG_LINKER;
2003
2004  if (!soinfo_link_image(&linker_so)) {
2005    // It would be nice to print an error message, but if the linker
2006    // can't link itself, there's no guarantee that we'll be able to
2007    // call write() (because it involves a GOT reference).
2008    //
2009    // This situation should never occur unless the linker itself
2010    // is corrupt.
2011    exit(EXIT_FAILURE);
2012  }
2013
2014  // We have successfully fixed our own relocations. It's safe to run
2015  // the main part of the linker now.
2016  args.abort_message_ptr = &gAbortMessage;
2017  Elf_Addr start_address = __linker_init_post_relocation(args, linker_addr);
2018
2019  set_soinfo_pool_protection(PROT_READ);
2020
2021  // Return the address that the calling assembly stub should jump to.
2022  return start_address;
2023}
2024