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 struct dtv *dtv;
84 tls_index *ti;
85
86 dtv = _tcb_get()->tcb_dtv;
87 ti = vti;
88 return (dtv->dtv_slots[ti->ti_module - 1].dtvs_tls +
89 (ti->ti_offset + 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 if (mem == NULL)
117 return (NULL);
118 res = (void *)roundup2((uintptr_t)mem + sizeof(void *), align);
119 *(void **)((uintptr_t)res - sizeof(void *)) = mem;
120 return (res);
121 }
122
123 static void
libc_free_aligned(void * ptr)124 libc_free_aligned(void *ptr)
125 {
126 void *mem;
127 uintptr_t x;
128
129 if (ptr == NULL)
130 return;
131
132 x = (uintptr_t)ptr;
133 x -= sizeof(void *);
134 mem = *(void **)x;
135 __je_bootstrap_free(mem);
136 }
137
138 #ifdef TLS_VARIANT_I
139
140 /*
141 * There are two versions of variant I of TLS
142 *
143 * - ARM and aarch64 uses original variant I as is described in [1] and [2],
144 * where TP points to start of TCB followed by aligned TLS segment.
145 * Both TCB and TLS must be aligned to alignment of TLS section. The TCB[0]
146 * points to DTV vector and DTV values are real addresses (without bias).
147 * Note: for Local Exec TLS Model, the offsets from TP (TCB in this case) to
148 * TLS variables are computed by linker, so we cannot overalign TLS section.
149 *
150 * - PowerPC and RISC-V use modified version of variant I, described in [3]
151 * where TP points (with bias) to TLS and TCB immediately precedes TLS without
152 * any alignment gap[4]. Only TLS should be aligned. The TCB[0] points to DTV
153 * vector and DTV values are biased by constant value (TLS_DTV_OFFSET) from
154 * real addresses. However, like RTLD, we don't actually bias the DTV values,
155 * instead we compensate in __tls_get_addr for ti_offset's bias.
156 *
157 * [1] Ulrich Drepper: ELF Handling for Thread-Local Storage
158 * www.akkadia.org/drepper/tls.pdf
159 *
160 * [2] ARM IHI 0045E: Addenda to, and Errata in, the ABI for the ARM(r)
161 * Architecture
162 * infocenter.arm.com/help/topic/com.arm.doc.ihi0045e/IHI0045E_ABI_addenda.pdf
163 *
164 * [3] OpenPOWER: Power Architecture 64-Bit ELF V2 ABI Specification
165 * https://members.openpowerfoundation.org/document/dl/576
166 *
167 * [4] Its unclear if "without any alignment gap" is hard ABI requirement,
168 * but we must follow this rule due to suboptimal _tcb_set()
169 * (aka <ARCH>_SET_TP) implementation. This function doesn't expect TP but
170 * TCB as argument.
171 */
172
173 /*
174 * Return pointer to allocated TLS block
175 */
176 static void *
get_tls_block_ptr(void * tcb,size_t tcbsize)177 get_tls_block_ptr(void *tcb, size_t tcbsize)
178 {
179 size_t extra_size, post_size, pre_size, tls_block_size;
180
181 /* Compute fragments sizes. */
182 extra_size = tcbsize - TLS_TCB_SIZE;
183 #if defined(__aarch64__) || defined(__arm__)
184 post_size = roundup2(TLS_TCB_SIZE, libc_tls_init_align) - TLS_TCB_SIZE;
185 #else
186 post_size = 0;
187 #endif
188 tls_block_size = tcbsize + post_size;
189 pre_size = roundup2(tls_block_size, libc_tls_init_align) -
190 tls_block_size;
191
192 return ((char *)tcb - pre_size - extra_size);
193 }
194
195 /*
196 * Free Static TLS using the Variant I method. The tcbsize
197 * and tcbalign parameters must be the same as those used to allocate
198 * the block.
199 */
200 void
__libc_free_tls(void * tcb,size_t tcbsize,size_t tcbalign __unused)201 __libc_free_tls(void *tcb, size_t tcbsize, size_t tcbalign __unused)
202 {
203 struct dtv *dtv;
204
205 dtv = ((struct tcb *)tcb)->tcb_dtv;
206 __je_bootstrap_free(dtv);
207 libc_free_aligned(get_tls_block_ptr(tcb, tcbsize));
208 }
209
210 /*
211 * Allocate Static TLS using the Variant I method.
212 *
213 * To handle all above requirements, we setup the following layout for
214 * TLS block:
215 * (whole memory block is aligned with MAX(TLS_TCB_ALIGN, tls_init_align))
216 *
217 * +----------+--------------+--------------+-----------+------------------+
218 * | pre gap | extended TCB | TCB | post gap | TLS segment |
219 * | pre_size | extra_size | TLS_TCB_SIZE | post_size | tls_static_space |
220 * +----------+--------------+--------------+-----------+------------------+
221 *
222 * where:
223 * extra_size is tcbsize - TLS_TCB_SIZE
224 * post_size is used to adjust TCB to TLS alignment for first version of TLS
225 * layout and is always 0 for second version.
226 * pre_size is used to adjust TCB alignment for first version and to adjust
227 * TLS alignment for second version.
228 *
229 */
230 void *
__libc_allocate_tls(void * oldtcb,size_t tcbsize,size_t tcbalign)231 __libc_allocate_tls(void *oldtcb, size_t tcbsize, size_t tcbalign)
232 {
233 struct dtv *dtv;
234 struct tcb *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 = (struct tcb *)(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->tcb_dtv;
273 dtv->dtv_slots[0].dtvs_tls = tls;
274 } else {
275 dtv = __je_bootstrap_malloc(sizeof(struct dtv) +
276 sizeof(struct dtv_slot));
277 if (dtv == NULL) {
278 tls_msg("__libc_allocate_tls: Out of memory.\n");
279 abort();
280 }
281 /* Build the DTV. */
282 tcb->tcb_dtv = dtv;
283 dtv->dtv_gen = 1; /* Generation. */
284 dtv->dtv_size = 1; /* Segments count. */
285 dtv->dtv_slots[0].dtvs_tls = tls;
286
287 if (libc_tls_init_size > 0)
288 memcpy(tls, libc_tls_init, libc_tls_init_size);
289 }
290
291 return (tcb);
292 }
293
294 #endif
295
296 #ifdef TLS_VARIANT_II
297
298 /*
299 * Free Static TLS using the Variant II method.
300 */
301 void
__libc_free_tls(void * tcb,size_t tcbsize __unused,size_t tcbalign)302 __libc_free_tls(void *tcb, size_t tcbsize __unused, size_t tcbalign)
303 {
304 size_t size;
305 struct dtv *dtv;
306 uintptr_t tlsstart, tlsend;
307
308 /*
309 * Figure out the size of the initial TLS block so that we can
310 * find stuff which ___tls_get_addr() allocated dynamically.
311 */
312 tcbalign = MAX(tcbalign, libc_tls_init_align);
313 size = roundup2(libc_tls_static_space, tcbalign);
314
315 dtv = ((struct tcb *)tcb)->tcb_dtv;
316 tlsend = (uintptr_t)tcb;
317 tlsstart = tlsend - size;
318 libc_free_aligned((void*)tlsstart);
319 __je_bootstrap_free(dtv);
320 }
321
322 /*
323 * Allocate Static TLS using the Variant II method.
324 */
325 void *
__libc_allocate_tls(void * oldtcb,size_t tcbsize,size_t tcbalign)326 __libc_allocate_tls(void *oldtcb, size_t tcbsize, size_t tcbalign)
327 {
328 size_t size;
329 char *tls_block, *tls;
330 struct dtv *dtv;
331 struct tcb *tcb;
332
333 tcbalign = MAX(tcbalign, libc_tls_init_align);
334 size = roundup2(libc_tls_static_space, tcbalign);
335
336 if (tcbsize < 2 * sizeof(uintptr_t))
337 tcbsize = 2 * sizeof(uintptr_t);
338 tls_block = libc_malloc_aligned(size + tcbsize, tcbalign);
339 if (tls_block == NULL) {
340 tls_msg("__libc_allocate_tls: Out of memory.\n");
341 abort();
342 }
343 memset(tls_block, 0, size + tcbsize);
344 dtv = __je_bootstrap_malloc(sizeof(struct dtv) +
345 sizeof(struct dtv_slot));
346 if (dtv == NULL) {
347 tls_msg("__libc_allocate_tls: Out of memory.\n");
348 abort();
349 }
350
351 tcb = (struct tcb *)(tls_block + size);
352 tls = (char *)tcb - libc_tls_static_space;
353 tcb->tcb_self = tcb;
354 tcb->tcb_dtv = dtv;
355
356 dtv->dtv_gen = 1;
357 dtv->dtv_size = 1;
358 dtv->dtv_slots[0].dtvs_tls = tls;
359
360 if (oldtcb != NULL) {
361 /*
362 * Copy the static TLS block over whole.
363 */
364 memcpy(tls, (const char *)oldtcb - libc_tls_static_space,
365 libc_tls_static_space);
366
367 /*
368 * We assume that this block was the one we created with
369 * allocate_initial_tls().
370 */
371 _rtld_free_tls(oldtcb, 2 * sizeof(uintptr_t),
372 sizeof(uintptr_t));
373 } else {
374 memcpy(tls, libc_tls_init, libc_tls_init_size);
375 memset(tls + libc_tls_init_size, 0,
376 libc_tls_static_space - libc_tls_init_size);
377 }
378
379 return (tcb);
380 }
381
382 #endif /* TLS_VARIANT_II */
383
384 #else
385
386 void *
__libc_allocate_tls(void * oldtcb __unused,size_t tcbsize __unused,size_t tcbalign __unused)387 __libc_allocate_tls(void *oldtcb __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