Set up the data structures for vDSO in libc.a
[glibc.git] / elf / dl-sysdep.c
blobfb0849a54399a03e9aaee60c79f205af03ba72c2
1 /* Operating system support for run-time dynamic linker. Generic Unix version.
2 Copyright (C) 1995-2012 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, see
17 <http://www.gnu.org/licenses/>. */
19 /* We conditionalize the whole of this file rather than simply eliding it
20 from the static build, because other sysdeps/ versions of this file
21 might define things needed by a static build. */
23 #ifdef SHARED
25 #include <assert.h>
26 #include <elf.h>
27 #include <errno.h>
28 #include <fcntl.h>
29 #include <libintl.h>
30 #include <stdlib.h>
31 #include <string.h>
32 #include <unistd.h>
33 #include <sys/types.h>
34 #include <sys/stat.h>
35 #include <sys/mman.h>
36 #include <ldsodefs.h>
37 #include <_itoa.h>
38 #include <fpu_control.h>
40 #include <entry.h>
41 #include <dl-machine.h>
42 #include <dl-procinfo.h>
43 #include <dl-osinfo.h>
44 #include <hp-timing.h>
45 #include <tls.h>
47 #ifdef _DL_FIRST_PLATFORM
48 # define _DL_FIRST_EXTRA (_DL_FIRST_PLATFORM + _DL_PLATFORMS_COUNT)
49 #else
50 # define _DL_FIRST_EXTRA _DL_HWCAP_COUNT
51 #endif
53 extern char **_environ attribute_hidden;
54 extern char _end[] attribute_hidden;
56 /* Protect SUID program against misuse of file descriptors. */
57 extern void __libc_check_standard_fds (void);
59 #ifdef NEED_DL_BASE_ADDR
60 ElfW(Addr) _dl_base_addr;
61 #endif
62 int __libc_enable_secure attribute_relro = 0;
63 INTVARDEF(__libc_enable_secure)
64 int __libc_multiple_libcs = 0; /* Defining this here avoids the inclusion
65 of init-first. */
66 /* This variable contains the lowest stack address ever used. */
67 void *__libc_stack_end attribute_relro = NULL;
68 rtld_hidden_data_def(__libc_stack_end)
69 void *_dl_random attribute_relro = NULL;
71 #ifndef DL_FIND_ARG_COMPONENTS
72 # define DL_FIND_ARG_COMPONENTS(cookie, argc, argv, envp, auxp) \
73 do { \
74 void **_tmp; \
75 (argc) = *(long int *) cookie; \
76 (argv) = (char **) ((long int *) cookie + 1); \
77 (envp) = (argv) + (argc) + 1; \
78 for (_tmp = (void **) (envp); *_tmp; ++_tmp) \
79 continue; \
80 (auxp) = (void *) ++_tmp; \
81 } while (0)
82 #endif
84 #ifndef DL_STACK_END
85 # define DL_STACK_END(cookie) ((void *) (cookie))
86 #endif
88 ElfW(Addr)
89 _dl_sysdep_start (void **start_argptr,
90 void (*dl_main) (const ElfW(Phdr) *phdr, ElfW(Word) phnum,
91 ElfW(Addr) *user_entry, ElfW(auxv_t) *auxv))
93 const ElfW(Phdr) *phdr = NULL;
94 ElfW(Word) phnum = 0;
95 ElfW(Addr) user_entry;
96 ElfW(auxv_t) *av;
97 #ifdef HAVE_AUX_SECURE
98 # define set_seen(tag) (tag) /* Evaluate for the side effects. */
99 # define set_seen_secure() ((void) 0)
100 #else
101 uid_t uid = 0;
102 gid_t gid = 0;
103 unsigned int seen = 0;
104 # define set_seen_secure() (seen = -1)
105 # ifdef HAVE_AUX_XID
106 # define set_seen(tag) (tag) /* Evaluate for the side effects. */
107 # else
108 # define M(type) (1 << (type))
109 # define set_seen(tag) seen |= M ((tag)->a_type)
110 # endif
111 #endif
112 #ifdef NEED_DL_SYSINFO
113 uintptr_t new_sysinfo = 0;
114 #endif
116 __libc_stack_end = DL_STACK_END (start_argptr);
117 DL_FIND_ARG_COMPONENTS (start_argptr, _dl_argc, INTUSE(_dl_argv), _environ,
118 GLRO(dl_auxv));
120 user_entry = (ElfW(Addr)) ENTRY_POINT;
121 GLRO(dl_platform) = NULL; /* Default to nothing known about the platform. */
123 for (av = GLRO(dl_auxv); av->a_type != AT_NULL; set_seen (av++))
124 switch (av->a_type)
126 case AT_PHDR:
127 phdr = (void *) av->a_un.a_val;
128 break;
129 case AT_PHNUM:
130 phnum = av->a_un.a_val;
131 break;
132 case AT_PAGESZ:
133 GLRO(dl_pagesize) = av->a_un.a_val;
134 break;
135 case AT_ENTRY:
136 user_entry = av->a_un.a_val;
137 break;
138 #ifdef NEED_DL_BASE_ADDR
139 case AT_BASE:
140 _dl_base_addr = av->a_un.a_val;
141 break;
142 #endif
143 #ifndef HAVE_AUX_SECURE
144 case AT_UID:
145 case AT_EUID:
146 uid ^= av->a_un.a_val;
147 break;
148 case AT_GID:
149 case AT_EGID:
150 gid ^= av->a_un.a_val;
151 break;
152 #endif
153 case AT_SECURE:
154 #ifndef HAVE_AUX_SECURE
155 seen = -1;
156 #endif
157 INTUSE(__libc_enable_secure) = av->a_un.a_val;
158 break;
159 case AT_PLATFORM:
160 GLRO(dl_platform) = (void *) av->a_un.a_val;
161 break;
162 case AT_HWCAP:
163 GLRO(dl_hwcap) = (unsigned long int) av->a_un.a_val;
164 break;
165 case AT_CLKTCK:
166 GLRO(dl_clktck) = av->a_un.a_val;
167 break;
168 case AT_FPUCW:
169 GLRO(dl_fpu_control) = av->a_un.a_val;
170 break;
171 #ifdef NEED_DL_SYSINFO
172 case AT_SYSINFO:
173 new_sysinfo = av->a_un.a_val;
174 break;
175 #endif
176 #if defined NEED_DL_SYSINFO || defined NEED_DL_SYSINFO_DSO
177 case AT_SYSINFO_EHDR:
178 GLRO(dl_sysinfo_dso) = (void *) av->a_un.a_val;
179 break;
180 #endif
181 case AT_RANDOM:
182 _dl_random = (void *) av->a_un.a_val;
183 break;
184 #ifdef DL_PLATFORM_AUXV
185 DL_PLATFORM_AUXV
186 #endif
189 #ifndef HAVE_AUX_SECURE
190 if (seen != -1)
192 /* Fill in the values we have not gotten from the kernel through the
193 auxiliary vector. */
194 # ifndef HAVE_AUX_XID
195 # define SEE(UID, var, uid) \
196 if ((seen & M (AT_##UID)) == 0) var ^= __get##uid ()
197 SEE (UID, uid, uid);
198 SEE (EUID, uid, euid);
199 SEE (GID, gid, gid);
200 SEE (EGID, gid, egid);
201 # endif
203 /* If one of the two pairs of IDs does not match this is a setuid
204 or setgid run. */
205 INTUSE(__libc_enable_secure) = uid | gid;
207 #endif
209 #ifndef HAVE_AUX_PAGESIZE
210 if (GLRO(dl_pagesize) == 0)
211 GLRO(dl_pagesize) = __getpagesize ();
212 #endif
214 #if defined NEED_DL_SYSINFO
215 /* Only set the sysinfo value if we also have the vsyscall DSO. */
216 if (GLRO(dl_sysinfo_dso) != 0 && new_sysinfo)
217 GLRO(dl_sysinfo) = new_sysinfo;
218 #endif
220 #ifdef DL_SYSDEP_INIT
221 DL_SYSDEP_INIT;
222 #endif
224 #ifdef DL_PLATFORM_INIT
225 DL_PLATFORM_INIT;
226 #endif
228 /* Determine the length of the platform name. */
229 if (GLRO(dl_platform) != NULL)
230 GLRO(dl_platformlen) = strlen (GLRO(dl_platform));
232 if (__sbrk (0) == _end)
233 /* The dynamic linker was run as a program, and so the initial break
234 starts just after our bss, at &_end. The malloc in dl-minimal.c
235 will consume the rest of this page, so tell the kernel to move the
236 break up that far. When the user program examines its break, it
237 will see this new value and not clobber our data. */
238 __sbrk (GLRO(dl_pagesize)
239 - ((_end - (char *) 0) & (GLRO(dl_pagesize) - 1)));
241 /* If this is a SUID program we make sure that FDs 0, 1, and 2 are
242 allocated. If necessary we are doing it ourself. If it is not
243 possible we stop the program. */
244 if (__builtin_expect (INTUSE(__libc_enable_secure), 0))
245 __libc_check_standard_fds ();
247 (*dl_main) (phdr, phnum, &user_entry, GLRO(dl_auxv));
248 return user_entry;
251 void
252 internal_function
253 _dl_sysdep_start_cleanup (void)
257 void
258 internal_function
259 _dl_show_auxv (void)
261 char buf[64];
262 ElfW(auxv_t) *av;
264 /* Terminate string. */
265 buf[63] = '\0';
267 /* The following code assumes that the AT_* values are encoded
268 starting from 0 with AT_NULL, 1 for AT_IGNORE, and all other values
269 close by (otherwise the array will be too large). In case we have
270 to support a platform where these requirements are not fulfilled
271 some alternative implementation has to be used. */
272 for (av = GLRO(dl_auxv); av->a_type != AT_NULL; ++av)
274 static const struct
276 const char label[17];
277 enum { unknown = 0, dec, hex, str, ignore } form : 8;
278 } auxvars[] =
280 [AT_EXECFD - 2] = { "EXECFD: ", dec },
281 [AT_EXECFN - 2] = { "EXECFN: ", str },
282 [AT_PHDR - 2] = { "PHDR: 0x", hex },
283 [AT_PHENT - 2] = { "PHENT: ", dec },
284 [AT_PHNUM - 2] = { "PHNUM: ", dec },
285 [AT_PAGESZ - 2] = { "PAGESZ: ", dec },
286 [AT_BASE - 2] = { "BASE: 0x", hex },
287 [AT_FLAGS - 2] = { "FLAGS: 0x", hex },
288 [AT_ENTRY - 2] = { "ENTRY: 0x", hex },
289 [AT_NOTELF - 2] = { "NOTELF: ", hex },
290 [AT_UID - 2] = { "UID: ", dec },
291 [AT_EUID - 2] = { "EUID: ", dec },
292 [AT_GID - 2] = { "GID: ", dec },
293 [AT_EGID - 2] = { "EGID: ", dec },
294 [AT_PLATFORM - 2] = { "PLATFORM: ", str },
295 [AT_HWCAP - 2] = { "HWCAP: ", hex },
296 [AT_CLKTCK - 2] = { "CLKTCK: ", dec },
297 [AT_FPUCW - 2] = { "FPUCW: ", hex },
298 [AT_DCACHEBSIZE - 2] = { "DCACHEBSIZE: 0x", hex },
299 [AT_ICACHEBSIZE - 2] = { "ICACHEBSIZE: 0x", hex },
300 [AT_UCACHEBSIZE - 2] = { "UCACHEBSIZE: 0x", hex },
301 [AT_IGNOREPPC - 2] = { "IGNOREPPC", ignore },
302 [AT_SECURE - 2] = { "SECURE: ", dec },
303 [AT_BASE_PLATFORM - 2] = { "BASE_PLATFORM:", str },
304 [AT_SYSINFO - 2] = { "SYSINFO: 0x", hex },
305 [AT_SYSINFO_EHDR - 2] = { "SYSINFO_EHDR: 0x", hex },
306 [AT_RANDOM - 2] = { "RANDOM: 0x", hex },
308 unsigned int idx = (unsigned int) (av->a_type - 2);
310 if ((unsigned int) av->a_type < 2u
311 || (idx < sizeof (auxvars) / sizeof (auxvars[0])
312 && auxvars[idx].form == ignore))
313 continue;
315 assert (AT_NULL == 0);
316 assert (AT_IGNORE == 1);
318 if (av->a_type == AT_HWCAP)
320 /* This is handled special. */
321 if (_dl_procinfo (av->a_un.a_val) == 0)
322 continue;
325 if (idx < sizeof (auxvars) / sizeof (auxvars[0])
326 && auxvars[idx].form != unknown)
328 const char *val = (char *) av->a_un.a_val;
330 if (__builtin_expect (auxvars[idx].form, dec) == dec)
331 val = _itoa ((unsigned long int) av->a_un.a_val,
332 buf + sizeof buf - 1, 10, 0);
333 else if (__builtin_expect (auxvars[idx].form, hex) == hex)
334 val = _itoa ((unsigned long int) av->a_un.a_val,
335 buf + sizeof buf - 1, 16, 0);
337 _dl_printf ("AT_%s%s\n", auxvars[idx].label, val);
339 continue;
342 /* Unknown value: print a generic line. */
343 char buf2[17];
344 buf2[sizeof (buf2) - 1] = '\0';
345 const char *val2 = _itoa ((unsigned long int) av->a_un.a_val,
346 buf2 + sizeof buf2 - 1, 16, 0);
347 const char *val = _itoa ((unsigned long int) av->a_type,
348 buf + sizeof buf - 1, 16, 0);
349 _dl_printf ("AT_??? (0x%s): 0x%s\n", val, val2);
354 /* Return an array of useful/necessary hardware capability names. */
355 const struct r_strlenpair *
356 internal_function
357 _dl_important_hwcaps (const char *platform, size_t platform_len, size_t *sz,
358 size_t *max_capstrlen)
360 /* Determine how many important bits are set. */
361 uint64_t masked = GLRO(dl_hwcap) & GLRO(dl_hwcap_mask);
362 size_t cnt = platform != NULL;
363 size_t n, m;
364 size_t total;
365 struct r_strlenpair *temp;
366 struct r_strlenpair *result;
367 struct r_strlenpair *rp;
368 char *cp;
370 /* Count the number of bits set in the masked value. */
371 for (n = 0; (~((1ULL << n) - 1) & masked) != 0; ++n)
372 if ((masked & (1ULL << n)) != 0)
373 ++cnt;
375 #if (defined NEED_DL_SYSINFO || defined NEED_DL_SYSINFO_DSO) && defined SHARED
376 /* The system-supplied DSO can contain a note of type 2, vendor "GNU".
377 This gives us a list of names to treat as fake hwcap bits. */
379 const char *dsocaps = NULL;
380 size_t dsocapslen = 0;
381 if (GLRO(dl_sysinfo_map) != NULL)
383 const ElfW(Phdr) *const phdr = GLRO(dl_sysinfo_map)->l_phdr;
384 const ElfW(Word) phnum = GLRO(dl_sysinfo_map)->l_phnum;
385 for (uint_fast16_t i = 0; i < phnum; ++i)
386 if (phdr[i].p_type == PT_NOTE)
388 const ElfW(Addr) start = (phdr[i].p_vaddr
389 + GLRO(dl_sysinfo_map)->l_addr);
390 const struct
392 ElfW(Word) vendorlen;
393 ElfW(Word) datalen;
394 ElfW(Word) type;
395 } *note = (const void *) start;
396 while ((ElfW(Addr)) (note + 1) - start < phdr[i].p_memsz)
398 #define ROUND(len) (((len) + sizeof (ElfW(Word)) - 1) & -sizeof (ElfW(Word)))
399 if (note->type == NT_GNU_HWCAP
400 && note->vendorlen == sizeof "GNU"
401 && !memcmp ((note + 1), "GNU", sizeof "GNU")
402 && note->datalen > 2 * sizeof (ElfW(Word)) + 2)
404 const ElfW(Word) *p = ((const void *) (note + 1)
405 + ROUND (sizeof "GNU"));
406 cnt += *p++;
407 ++p; /* Skip mask word. */
408 dsocaps = (const char *) p;
409 dsocapslen = note->datalen - sizeof *p * 2;
410 break;
412 note = ((const void *) (note + 1)
413 + ROUND (note->vendorlen) + ROUND (note->datalen));
414 #undef ROUND
416 if (dsocaps != NULL)
417 break;
420 #endif
422 /* For TLS enabled builds always add 'tls'. */
423 ++cnt;
425 /* Create temporary data structure to generate result table. */
426 temp = (struct r_strlenpair *) alloca (cnt * sizeof (*temp));
427 m = 0;
428 #if (defined NEED_DL_SYSINFO || defined NEED_DL_SYSINFO_DSO) && defined SHARED
429 if (dsocaps != NULL)
431 const ElfW(Word) mask = ((const ElfW(Word) *) dsocaps)[-1];
432 GLRO(dl_hwcap) |= (uint64_t) mask << _DL_FIRST_EXTRA;
433 /* Note that we add the dsocaps to the set already chosen by the
434 LD_HWCAP_MASK environment variable (or default HWCAP_IMPORTANT).
435 So there is no way to request ignoring an OS-supplied dsocap
436 string and bit like you can ignore an OS-supplied HWCAP bit. */
437 GLRO(dl_hwcap_mask) |= (uint64_t) mask << _DL_FIRST_EXTRA;
438 size_t len;
439 for (const char *p = dsocaps; p < dsocaps + dsocapslen; p += len + 1)
441 uint_fast8_t bit = *p++;
442 len = strlen (p);
444 /* Skip entries that are not enabled in the mask word. */
445 if (__builtin_expect (mask & ((ElfW(Word)) 1 << bit), 1))
447 temp[m].str = p;
448 temp[m].len = len;
449 ++m;
451 else
452 --cnt;
455 #endif
456 for (n = 0; masked != 0; ++n)
457 if ((masked & (1ULL << n)) != 0)
459 temp[m].str = _dl_hwcap_string (n);
460 temp[m].len = strlen (temp[m].str);
461 masked ^= 1ULL << n;
462 ++m;
464 if (platform != NULL)
466 temp[m].str = platform;
467 temp[m].len = platform_len;
468 ++m;
471 temp[m].str = "tls";
472 temp[m].len = 3;
473 ++m;
475 assert (m == cnt);
477 /* Determine the total size of all strings together. */
478 if (cnt == 1)
479 total = temp[0].len + 1;
480 else
482 total = temp[0].len + temp[cnt - 1].len + 2;
483 if (cnt > 2)
485 total <<= 1;
486 for (n = 1; n + 1 < cnt; ++n)
487 total += temp[n].len + 1;
488 if (cnt > 3
489 && (cnt >= sizeof (size_t) * 8
490 || total + (sizeof (*result) << 3)
491 >= (1UL << (sizeof (size_t) * 8 - cnt + 3))))
492 _dl_signal_error (ENOMEM, NULL, NULL,
493 N_("cannot create capability list"));
495 total <<= cnt - 3;
499 /* The result structure: we use a very compressed way to store the
500 various combinations of capability names. */
501 *sz = 1 << cnt;
502 result = (struct r_strlenpair *) malloc (*sz * sizeof (*result) + total);
503 if (result == NULL)
504 _dl_signal_error (ENOMEM, NULL, NULL,
505 N_("cannot create capability list"));
507 if (cnt == 1)
509 result[0].str = (char *) (result + *sz);
510 result[0].len = temp[0].len + 1;
511 result[1].str = (char *) (result + *sz);
512 result[1].len = 0;
513 cp = __mempcpy ((char *) (result + *sz), temp[0].str, temp[0].len);
514 *cp = '/';
515 *sz = 2;
516 *max_capstrlen = result[0].len;
518 return result;
521 /* Fill in the information. This follows the following scheme
522 (indeces from TEMP for four strings):
523 entry #0: 0, 1, 2, 3 binary: 1111
524 #1: 0, 1, 3 1101
525 #2: 0, 2, 3 1011
526 #3: 0, 3 1001
527 This allows the representation of all possible combinations of
528 capability names in the string. First generate the strings. */
529 result[1].str = result[0].str = cp = (char *) (result + *sz);
530 #define add(idx) \
531 cp = __mempcpy (__mempcpy (cp, temp[idx].str, temp[idx].len), "/", 1);
532 if (cnt == 2)
534 add (1);
535 add (0);
537 else
539 n = 1 << (cnt - 1);
542 n -= 2;
544 /* We always add the last string. */
545 add (cnt - 1);
547 /* Add the strings which have the bit set in N. */
548 for (m = cnt - 2; m > 0; --m)
549 if ((n & (1 << m)) != 0)
550 add (m);
552 /* Always add the first string. */
553 add (0);
555 while (n != 0);
557 #undef add
559 /* Now we are ready to install the string pointers and length. */
560 for (n = 0; n < (1UL << cnt); ++n)
561 result[n].len = 0;
562 n = cnt;
565 size_t mask = 1 << --n;
567 rp = result;
568 for (m = 1 << cnt; m > 0; ++rp)
569 if ((--m & mask) != 0)
570 rp->len += temp[n].len + 1;
572 while (n != 0);
574 /* The first half of the strings all include the first string. */
575 n = (1 << cnt) - 2;
576 rp = &result[2];
577 while (n != (1UL << (cnt - 1)))
579 if ((--n & 1) != 0)
580 rp[0].str = rp[-2].str + rp[-2].len;
581 else
582 rp[0].str = rp[-1].str;
583 ++rp;
586 /* The second half starts right after the first part of the string of
587 the corresponding entry in the first half. */
590 rp[0].str = rp[-(1 << (cnt - 1))].str + temp[cnt - 1].len + 1;
591 ++rp;
593 while (--n != 0);
595 /* The maximum string length. */
596 *max_capstrlen = result[0].len;
598 return result;
601 #endif