1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2004 Doug Rabson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 /*
30 * Define stubs for TLS internals so that programs and libraries can
31 * link. These functions will be replaced by functional versions at
32 * runtime from ld-elf.so.1.
33 */
34
35 #include <sys/param.h>
36 #include <stdlib.h>
37 #include <string.h>
38 #include <elf.h>
39 #include <unistd.h>
40
41 #include "rtld.h"
42 #include "libc_private.h"
43
44 #define tls_assert(cond) ((cond) ? (void) 0 : \
45 (tls_msg(#cond ": assert failed: " __FILE__ ":" \
46 __XSTRING(__LINE__) "\n"), abort()))
47 #define tls_msg(s) write(STDOUT_FILENO, s, strlen(s))
48
49 /* Provided by jemalloc to avoid bootstrapping issues. */
50 void *__je_bootstrap_malloc(size_t size);
51 void *__je_bootstrap_calloc(size_t num, size_t size);
52 void __je_bootstrap_free(void *ptr);
53
54 __weak_reference(__libc_allocate_tls, _rtld_allocate_tls);
55 __weak_reference(__libc_free_tls, _rtld_free_tls);
56
57 #ifdef __i386__
58
59 __weak_reference(___libc_tls_get_addr, ___tls_get_addr);
60 __attribute__((__regparm__(1))) void * ___libc_tls_get_addr(void *);
61
62 #endif
63
64 void * __libc_tls_get_addr(void *);
65 __weak_reference(__libc_tls_get_addr, __tls_get_addr);
66
67 void *_rtld_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign);
68 void _rtld_free_tls(void *tls, size_t tcbsize, size_t tcbalign);
69 void *__libc_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign);
70 void __libc_free_tls(void *tls, size_t tcbsize, size_t tcbalign);
71
72 #ifndef PIC
73
74 static size_t libc_tls_static_space;
75 static size_t libc_tls_init_size;
76 static size_t libc_tls_init_align;
77 static void *libc_tls_init;
78 #endif
79
80 void *
__libc_tls_get_addr(void * vti)81 __libc_tls_get_addr(void *vti)
82 {
83 uintptr_t *dtv;
84 tls_index *ti;
85
86 dtv = _tcb_get()->tcb_dtv;
87 ti = vti;
88 return ((char *)(dtv[ti->ti_module + 1] + ti->ti_offset) +
89 TLS_DTV_OFFSET);
90 }
91
92 #ifdef __i386__
93
94 /* GNU ABI */
95
96 __attribute__((__regparm__(1)))
97 void *
___libc_tls_get_addr(void * vti)98 ___libc_tls_get_addr(void *vti)
99 {
100 return (__libc_tls_get_addr(vti));
101 }
102
103 #endif
104
105 #ifndef PIC
106
107 static void *
libc_malloc_aligned(size_t size,size_t align)108 libc_malloc_aligned(size_t size, size_t align)
109 {
110 void *mem, *res;
111
112 if (align < sizeof(void *))
113 align = sizeof(void *);
114
115 mem = __je_bootstrap_malloc(size + sizeof(void *) + align - 1);
116 res = (void *)roundup2((uintptr_t)mem + sizeof(void *), align);
117 *(void **)((uintptr_t)res - sizeof(void *)) = mem;
118 return (res);
119 }
120
121 static void
libc_free_aligned(void * ptr)122 libc_free_aligned(void *ptr)
123 {
124 void *mem;
125 uintptr_t x;
126
127 if (ptr == NULL)
128 return;
129
130 x = (uintptr_t)ptr;
131 x -= sizeof(void *);
132 mem = *(void **)x;
133 __je_bootstrap_free(mem);
134 }
135
136 #ifdef TLS_VARIANT_I
137
138 /*
139 * There are two versions of variant I of TLS
140 *
141 * - ARM and aarch64 uses original variant I as is described in [1] and [2],
142 * where TP points to start of TCB followed by aligned TLS segment.
143 * Both TCB and TLS must be aligned to alignment of TLS section. The TCB[0]
144 * points to DTV vector and DTV values are real addresses (without bias).
145 * Note: for Local Exec TLS Model, the offsets from TP (TCB in this case) to
146 * TLS variables are computed by linker, so we cannot overalign TLS section.
147 *
148 * - PowerPC and RISC-V use modified version of variant I, described in [3]
149 * where TP points (with bias) to TLS and TCB immediately precedes TLS without
150 * any alignment gap[4]. Only TLS should be aligned. The TCB[0] points to DTV
151 * vector and DTV values are biased by constant value (TLS_DTV_OFFSET) from
152 * real addresses[5].
153 *
154 * [1] Ulrich Drepper: ELF Handling for Thread-Local Storage
155 * www.akkadia.org/drepper/tls.pdf
156 *
157 * [2] ARM IHI 0045E: Addenda to, and Errata in, the ABI for the ARM(r)
158 * Architecture
159 * infocenter.arm.com/help/topic/com.arm.doc.ihi0045e/IHI0045E_ABI_addenda.pdf
160 *
161 * [3] OpenPOWER: Power Architecture 64-Bit ELF V2 ABI Specification
162 * https://members.openpowerfoundation.org/document/dl/576
163 *
164 * [4] Its unclear if "without any alignment gap" is hard ABI requirement,
165 * but we must follow this rule due to suboptimal _tcb_set()
166 * (aka <ARCH>_SET_TP) implementation. This function doesn't expect TP but
167 * TCB as argument.
168 *
169 * [5] I'm not able to validate "values are biased" assertions.
170 */
171
172 /*
173 * Return pointer to allocated TLS block
174 */
175 static void *
get_tls_block_ptr(void * tcb,size_t tcbsize)176 get_tls_block_ptr(void *tcb, size_t tcbsize)
177 {
178 size_t extra_size, post_size, pre_size, tls_block_size;
179
180 /* Compute fragments sizes. */
181 extra_size = tcbsize - TLS_TCB_SIZE;
182 #if defined(__aarch64__) || defined(__arm__)
183 post_size = roundup2(TLS_TCB_SIZE, libc_tls_init_align) - TLS_TCB_SIZE;
184 #else
185 post_size = 0;
186 #endif
187 tls_block_size = tcbsize + post_size;
188 pre_size = roundup2(tls_block_size, libc_tls_init_align) -
189 tls_block_size;
190
191 return ((char *)tcb - pre_size - extra_size);
192 }
193
194 /*
195 * Free Static TLS using the Variant I method. The tcbsize
196 * and tcbalign parameters must be the same as those used to allocate
197 * the block.
198 */
199 void
__libc_free_tls(void * tcb,size_t tcbsize,size_t tcbalign __unused)200 __libc_free_tls(void *tcb, size_t tcbsize, size_t tcbalign __unused)
201 {
202 Elf_Addr *dtv;
203 Elf_Addr **tls;
204
205 tls = (Elf_Addr **)tcb;
206 dtv = tls[0];
207 __je_bootstrap_free(dtv);
208 libc_free_aligned(get_tls_block_ptr(tcb, tcbsize));
209 }
210
211 /*
212 * Allocate Static TLS using the Variant I method.
213 *
214 * To handle all above requirements, we setup the following layout for
215 * TLS block:
216 * (whole memory block is aligned with MAX(TLS_TCB_ALIGN, tls_init_align))
217 *
218 * +----------+--------------+--------------+-----------+------------------+
219 * | pre gap | extended TCB | TCB | post gap | TLS segment |
220 * | pre_size | extra_size | TLS_TCB_SIZE | post_size | tls_static_space |
221 * +----------+--------------+--------------+-----------+------------------+
222 *
223 * where:
224 * extra_size is tcbsize - TLS_TCB_SIZE
225 * post_size is used to adjust TCB to TLS alignment for first version of TLS
226 * layout and is always 0 for second version.
227 * pre_size is used to adjust TCB alignment for first version and to adjust
228 * TLS alignment for second version.
229 *
230 */
231 void *
__libc_allocate_tls(void * oldtcb,size_t tcbsize,size_t tcbalign)232 __libc_allocate_tls(void *oldtcb, size_t tcbsize, size_t tcbalign)
233 {
234 Elf_Addr *dtv, **tcb;
235 char *tls_block, *tls;
236 size_t extra_size, maxalign, post_size, pre_size, tls_block_size;
237
238 if (oldtcb != NULL && tcbsize == TLS_TCB_SIZE)
239 return (oldtcb);
240
241 tls_assert(tcbalign >= TLS_TCB_ALIGN);
242 maxalign = MAX(tcbalign, libc_tls_init_align);
243
244 /* Compute fragmets sizes. */
245 extra_size = tcbsize - TLS_TCB_SIZE;
246 #if defined(__aarch64__) || defined(__arm__)
247 post_size = roundup2(TLS_TCB_SIZE, libc_tls_init_align) - TLS_TCB_SIZE;
248 #else
249 post_size = 0;
250 #endif
251 tls_block_size = tcbsize + post_size;
252 pre_size = roundup2(tls_block_size, libc_tls_init_align) -
253 tls_block_size;
254 tls_block_size += pre_size + libc_tls_static_space;
255
256 /* Allocate whole TLS block */
257 tls_block = libc_malloc_aligned(tls_block_size, maxalign);
258 if (tls_block == NULL) {
259 tls_msg("__libc_allocate_tls: Out of memory.\n");
260 abort();
261 }
262 memset(tls_block, 0, tls_block_size);
263 tcb = (Elf_Addr **)(tls_block + pre_size + extra_size);
264 tls = (char *)tcb + TLS_TCB_SIZE + post_size;
265
266 if (oldtcb != NULL) {
267 memcpy(tls_block, get_tls_block_ptr(oldtcb, tcbsize),
268 tls_block_size);
269 libc_free_aligned(oldtcb);
270
271 /* Adjust the DTV. */
272 dtv = tcb[0];
273 dtv[2] = (Elf_Addr)(tls + TLS_DTV_OFFSET);
274 } else {
275 dtv = __je_bootstrap_malloc(3 * sizeof(Elf_Addr));
276 if (dtv == NULL) {
277 tls_msg("__libc_allocate_tls: Out of memory.\n");
278 abort();
279 }
280 /* Build the DTV. */
281 tcb[0] = dtv;
282 dtv[0] = 1; /* Generation. */
283 dtv[1] = 1; /* Segments count. */
284 dtv[2] = (Elf_Addr)(tls + TLS_DTV_OFFSET);
285
286 if (libc_tls_init_size > 0)
287 memcpy(tls, libc_tls_init, libc_tls_init_size);
288 }
289
290 return (tcb);
291 }
292
293 #endif
294
295 #ifdef TLS_VARIANT_II
296
297 /*
298 * Free Static TLS using the Variant II method.
299 */
300 void
__libc_free_tls(void * tcb,size_t tcbsize __unused,size_t tcbalign)301 __libc_free_tls(void *tcb, size_t tcbsize __unused, size_t tcbalign)
302 {
303 size_t size;
304 Elf_Addr* dtv;
305 Elf_Addr tlsstart, tlsend;
306
307 /*
308 * Figure out the size of the initial TLS block so that we can
309 * find stuff which ___tls_get_addr() allocated dynamically.
310 */
311 tcbalign = MAX(tcbalign, libc_tls_init_align);
312 size = roundup2(libc_tls_static_space, tcbalign);
313
314 dtv = ((Elf_Addr**)tcb)[1];
315 tlsend = (Elf_Addr) tcb;
316 tlsstart = tlsend - size;
317 libc_free_aligned((void*)tlsstart);
318 __je_bootstrap_free(dtv);
319 }
320
321 /*
322 * Allocate Static TLS using the Variant II method.
323 */
324 void *
__libc_allocate_tls(void * oldtls,size_t tcbsize,size_t tcbalign)325 __libc_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign)
326 {
327 size_t size;
328 char *tls;
329 Elf_Addr *dtv;
330 Elf_Addr segbase, oldsegbase;
331
332 tcbalign = MAX(tcbalign, libc_tls_init_align);
333 size = roundup2(libc_tls_static_space, tcbalign);
334
335 if (tcbsize < 2 * sizeof(Elf_Addr))
336 tcbsize = 2 * sizeof(Elf_Addr);
337 tls = libc_malloc_aligned(size + tcbsize, tcbalign);
338 if (tls == NULL) {
339 tls_msg("__libc_allocate_tls: Out of memory.\n");
340 abort();
341 }
342 memset(tls, 0, size + tcbsize);
343 dtv = __je_bootstrap_malloc(3 * sizeof(Elf_Addr));
344 if (dtv == NULL) {
345 tls_msg("__libc_allocate_tls: Out of memory.\n");
346 abort();
347 }
348
349 segbase = (Elf_Addr)(tls + size);
350 ((Elf_Addr*)segbase)[0] = segbase;
351 ((Elf_Addr*)segbase)[1] = (Elf_Addr) dtv;
352
353 dtv[0] = 1;
354 dtv[1] = 1;
355 dtv[2] = segbase - libc_tls_static_space;
356
357 if (oldtls) {
358 /*
359 * Copy the static TLS block over whole.
360 */
361 oldsegbase = (Elf_Addr) oldtls;
362 memcpy((void *)(segbase - libc_tls_static_space),
363 (const void *)(oldsegbase - libc_tls_static_space),
364 libc_tls_static_space);
365
366 /*
367 * We assume that this block was the one we created with
368 * allocate_initial_tls().
369 */
370 _rtld_free_tls(oldtls, 2*sizeof(Elf_Addr), sizeof(Elf_Addr));
371 } else {
372 memcpy((void *)(segbase - libc_tls_static_space),
373 libc_tls_init, libc_tls_init_size);
374 memset((void *)(segbase - libc_tls_static_space +
375 libc_tls_init_size), 0,
376 libc_tls_static_space - libc_tls_init_size);
377 }
378
379 return (void*) segbase;
380 }
381
382 #endif /* TLS_VARIANT_II */
383
384 #else
385
386 void *
__libc_allocate_tls(void * oldtls __unused,size_t tcbsize __unused,size_t tcbalign __unused)387 __libc_allocate_tls(void *oldtls __unused, size_t tcbsize __unused,
388 size_t tcbalign __unused)
389 {
390 return (0);
391 }
392
393 void
__libc_free_tls(void * tcb __unused,size_t tcbsize __unused,size_t tcbalign __unused)394 __libc_free_tls(void *tcb __unused, size_t tcbsize __unused,
395 size_t tcbalign __unused)
396 {
397 }
398
399 #endif /* PIC */
400
401 void
_init_tls(void)402 _init_tls(void)
403 {
404 #ifndef PIC
405 Elf_Addr *sp;
406 Elf_Auxinfo *aux, *auxp;
407 Elf_Phdr *phdr;
408 size_t phent, phnum;
409 int i;
410 void *tls;
411
412 sp = (Elf_Addr *) environ;
413 while (*sp++ != 0)
414 ;
415 aux = (Elf_Auxinfo *) sp;
416 phdr = NULL;
417 phent = phnum = 0;
418 for (auxp = aux; auxp->a_type != AT_NULL; auxp++) {
419 switch (auxp->a_type) {
420 case AT_PHDR:
421 phdr = auxp->a_un.a_ptr;
422 break;
423
424 case AT_PHENT:
425 phent = auxp->a_un.a_val;
426 break;
427
428 case AT_PHNUM:
429 phnum = auxp->a_un.a_val;
430 break;
431 }
432 }
433 if (phdr == NULL || phent != sizeof(Elf_Phdr) || phnum == 0)
434 return;
435
436 for (i = 0; (unsigned) i < phnum; i++) {
437 if (phdr[i].p_type == PT_TLS) {
438 libc_tls_static_space = roundup2(phdr[i].p_memsz,
439 phdr[i].p_align);
440 libc_tls_init_size = phdr[i].p_filesz;
441 libc_tls_init_align = phdr[i].p_align;
442 libc_tls_init = (void *)phdr[i].p_vaddr;
443 break;
444 }
445 }
446 tls = _rtld_allocate_tls(NULL, TLS_TCB_SIZE, TLS_TCB_ALIGN);
447
448 _tcb_set(tls);
449 #endif
450 }
451