1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 * $FreeBSD$ 29 */ 30 31 /* 32 * Define stubs for TLS internals so that programs and libraries can 33 * link. These functions will be replaced by functional versions at 34 * runtime from ld-elf.so.1. 35 */ 36 37 #include <sys/cdefs.h> 38 #include <sys/param.h> 39 #include <stdlib.h> 40 #include <string.h> 41 #include <elf.h> 42 #include <unistd.h> 43 44 #include "libc_private.h" 45 46 #define tls_assert(cond) ((cond) ? (void) 0 : \ 47 (tls_msg(#cond ": assert failed: " __FILE__ ":" \ 48 __XSTRING(__LINE__) "\n"), abort())) 49 #define tls_msg(s) write(STDOUT_FILENO, s, strlen(s)) 50 51 /* Provided by jemalloc to avoid bootstrapping issues. */ 52 void *__je_bootstrap_malloc(size_t size); 53 void *__je_bootstrap_calloc(size_t num, size_t size); 54 void __je_bootstrap_free(void *ptr); 55 56 __weak_reference(__libc_allocate_tls, _rtld_allocate_tls); 57 __weak_reference(__libc_free_tls, _rtld_free_tls); 58 59 #ifdef __i386__ 60 61 __weak_reference(___libc_tls_get_addr, ___tls_get_addr); 62 __attribute__((__regparm__(1))) void * ___libc_tls_get_addr(void *); 63 64 #endif 65 66 void * __libc_tls_get_addr(void *); 67 __weak_reference(__libc_tls_get_addr, __tls_get_addr); 68 69 void *_rtld_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign); 70 void _rtld_free_tls(void *tls, size_t tcbsize, size_t tcbalign); 71 void *__libc_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign); 72 void __libc_free_tls(void *tls, size_t tcbsize, size_t tcbalign); 73 74 #if defined(__amd64__) 75 #define TLS_TCB_ALIGN 16 76 #elif defined(__aarch64__) || defined(__arm__) || defined(__i386__) || \ 77 defined(__mips__) || defined(__powerpc__) || defined(__riscv) 78 #define TLS_TCB_ALIGN sizeof(void *) 79 #else 80 #error TLS_TCB_ALIGN undefined for target architecture 81 #endif 82 83 #if defined(__aarch64__) || defined(__arm__) || defined(__mips__) || \ 84 defined(__powerpc__) || defined(__riscv) 85 #define TLS_VARIANT_I 86 #endif 87 #if defined(__i386__) || defined(__amd64__) 88 #define TLS_VARIANT_II 89 #endif 90 91 #if defined(__mips__) || defined(__powerpc__) || defined(__riscv) 92 #define DTV_OFFSET 0x8000 93 #else 94 #define DTV_OFFSET 0 95 #endif 96 97 #ifndef PIC 98 99 static size_t tls_static_space; 100 static size_t tls_init_size; 101 static size_t tls_init_align; 102 static void *tls_init; 103 #endif 104 105 #ifdef __i386__ 106 107 /* GNU ABI */ 108 109 __attribute__((__regparm__(1))) 110 void * 111 ___libc_tls_get_addr(void *ti __unused) 112 { 113 return (0); 114 } 115 116 #endif 117 118 void * 119 __libc_tls_get_addr(void *ti __unused) 120 { 121 return (0); 122 } 123 124 #ifndef PIC 125 126 static void * 127 malloc_aligned(size_t size, size_t align) 128 { 129 void *mem, *res; 130 131 if (align < sizeof(void *)) 132 align = sizeof(void *); 133 134 mem = __je_bootstrap_malloc(size + sizeof(void *) + align - 1); 135 res = (void *)roundup2((uintptr_t)mem + sizeof(void *), align); 136 *(void **)((uintptr_t)res - sizeof(void *)) = mem; 137 return (res); 138 } 139 140 static void 141 free_aligned(void *ptr) 142 { 143 void *mem; 144 uintptr_t x; 145 146 if (ptr == NULL) 147 return; 148 149 x = (uintptr_t)ptr; 150 x -= sizeof(void *); 151 mem = *(void **)x; 152 __je_bootstrap_free(mem); 153 } 154 155 #ifdef TLS_VARIANT_I 156 157 /* 158 * There are two versions of variant I of TLS 159 * 160 * - ARM and aarch64 uses original variant I as is described in [1] and [2], 161 * where TP points to start of TCB followed by aligned TLS segment. 162 * Both TCB and TLS must be aligned to alignment of TLS section. The TCB[0] 163 * points to DTV vector and DTV values are real addresses (without bias). 164 * Note: for Local Exec TLS Model, the offsets from TP (TCB in this case) to 165 * TLS variables are computed by linker, so we cannot overalign TLS section. 166 * 167 * - MIPS, PowerPC and RISC-V use modified version of variant I, 168 * described in [3] where TP points (with bias) to TLS and TCB immediately 169 * precedes TLS without any alignment gap[4]. Only TLS should be aligned. 170 * The TCB[0] points to DTV vector and DTV values are biased by constant 171 * value (0x8000) from real addresses[5]. 172 * 173 * [1] Ulrich Drepper: ELF Handling for Thread-Local Storage 174 * www.akkadia.org/drepper/tls.pdf 175 * 176 * [2] ARM IHI 0045E: Addenda to, and Errata in, the ABI for the ARM(r) 177 * Architecture 178 * infocenter.arm.com/help/topic/com.arm.doc.ihi0045e/IHI0045E_ABI_addenda.pdf 179 * 180 * [3] OpenPOWER: Power Architecture 64-Bit ELF V2 ABI Specification 181 * https://members.openpowerfoundation.org/document/dl/576 182 * 183 * [4] Its unclear if "without any alignment gap" is hard ABI requirement, 184 * but we must follow this rule due to suboptimal _set_tp() 185 * (aka <ARCH>_SET_TP) implementation. This function doesn't expect TP but 186 * TCB as argument. 187 * 188 * [5] I'm not able to validate "values are biased" assertions. 189 */ 190 191 #define TLS_TCB_SIZE (2 * sizeof(void *)) 192 193 /* 194 * Return pointer to allocated TLS block 195 */ 196 static void * 197 get_tls_block_ptr(void *tcb, size_t tcbsize) 198 { 199 size_t extra_size, post_size, pre_size, tls_block_size; 200 201 /* Compute fragments sizes. */ 202 extra_size = tcbsize - TLS_TCB_SIZE; 203 #if defined(__aarch64__) || defined(__arm__) 204 post_size = roundup2(TLS_TCB_SIZE, tls_init_align) - TLS_TCB_SIZE; 205 #else 206 post_size = 0; 207 #endif 208 tls_block_size = tcbsize + post_size; 209 pre_size = roundup2(tls_block_size, tls_init_align) - tls_block_size; 210 211 return ((char *)tcb - pre_size - extra_size); 212 } 213 214 /* 215 * Free Static TLS using the Variant I method. The tcbsize 216 * and tcbalign parameters must be the same as those used to allocate 217 * the block. 218 */ 219 void 220 __libc_free_tls(void *tcb, size_t tcbsize, size_t tcbalign __unused) 221 { 222 Elf_Addr *dtv; 223 Elf_Addr **tls; 224 225 tls = (Elf_Addr **)tcb; 226 dtv = tls[0]; 227 __je_bootstrap_free(dtv); 228 free_aligned(get_tls_block_ptr(tcb, tcbsize)); 229 } 230 231 /* 232 * Allocate Static TLS using the Variant I method. 233 * 234 * To handle all above requirements, we setup the following layout for 235 * TLS block: 236 * (whole memory block is aligned with MAX(TLS_TCB_ALIGN, tls_init_align)) 237 * 238 * +----------+--------------+--------------+-----------+------------------+ 239 * | pre gap | extended TCB | TCB | post gap | TLS segment | 240 * | pre_size | extra_size | TLS_TCB_SIZE | post_size | tls_static_space | 241 * +----------+--------------+--------------+-----------+------------------+ 242 * 243 * where: 244 * extra_size is tcbsize - TLS_TCB_SIZE 245 * post_size is used to adjust TCB to TLS aligment for first version of TLS 246 * layout and is always 0 for second version. 247 * pre_size is used to adjust TCB aligment for first version and to adjust 248 * TLS alignment for second version. 249 * 250 */ 251 void * 252 __libc_allocate_tls(void *oldtcb, size_t tcbsize, size_t tcbalign) 253 { 254 Elf_Addr *dtv, **tcb; 255 char *tls_block, *tls; 256 size_t extra_size, maxalign, post_size, pre_size, tls_block_size; 257 258 if (oldtcb != NULL && tcbsize == TLS_TCB_SIZE) 259 return (oldtcb); 260 261 tls_assert(tcbalign >= TLS_TCB_ALIGN); 262 maxalign = MAX(tcbalign, tls_init_align); 263 264 /* Compute fragmets sizes. */ 265 extra_size = tcbsize - TLS_TCB_SIZE; 266 #if defined(__aarch64__) || defined(__arm__) 267 post_size = roundup2(TLS_TCB_SIZE, tls_init_align) - TLS_TCB_SIZE; 268 #else 269 post_size = 0; 270 #endif 271 tls_block_size = tcbsize + post_size; 272 pre_size = roundup2(tls_block_size, tls_init_align) - tls_block_size; 273 tls_block_size += pre_size + tls_static_space; 274 275 /* Allocate whole TLS block */ 276 tls_block = malloc_aligned(tls_block_size, maxalign); 277 if (tls_block == NULL) { 278 tls_msg("__libc_allocate_tls: Out of memory.\n"); 279 abort(); 280 } 281 memset(tls_block, 0, tls_block_size); 282 tcb = (Elf_Addr **)(tls_block + pre_size + extra_size); 283 tls = (char *)tcb + TLS_TCB_SIZE + post_size; 284 285 if (oldtcb != NULL) { 286 memcpy(tls_block, get_tls_block_ptr(oldtcb, tcbsize), 287 tls_block_size); 288 free_aligned(oldtcb); 289 290 /* Adjust the DTV. */ 291 dtv = tcb[0]; 292 dtv[2] = (Elf_Addr)(tls + DTV_OFFSET); 293 } else { 294 dtv = __je_bootstrap_malloc(3 * sizeof(Elf_Addr)); 295 if (dtv == NULL) { 296 tls_msg("__libc_allocate_tls: Out of memory.\n"); 297 abort(); 298 } 299 /* Build the DTV. */ 300 tcb[0] = dtv; 301 dtv[0] = 1; /* Generation. */ 302 dtv[1] = 1; /* Segments count. */ 303 dtv[2] = (Elf_Addr)(tls + DTV_OFFSET); 304 305 if (tls_init_size > 0) 306 memcpy(tls, tls_init, tls_init_size); 307 } 308 309 return (tcb); 310 } 311 312 #endif 313 314 #ifdef TLS_VARIANT_II 315 316 #define TLS_TCB_SIZE (3 * sizeof(Elf_Addr)) 317 318 /* 319 * Free Static TLS using the Variant II method. 320 */ 321 void 322 __libc_free_tls(void *tcb, size_t tcbsize __unused, size_t tcbalign) 323 { 324 size_t size; 325 Elf_Addr* dtv; 326 Elf_Addr tlsstart, tlsend; 327 328 /* 329 * Figure out the size of the initial TLS block so that we can 330 * find stuff which ___tls_get_addr() allocated dynamically. 331 */ 332 tcbalign = MAX(tcbalign, tls_init_align); 333 size = roundup2(tls_static_space, tcbalign); 334 335 dtv = ((Elf_Addr**)tcb)[1]; 336 tlsend = (Elf_Addr) tcb; 337 tlsstart = tlsend - size; 338 free_aligned((void*)tlsstart); 339 __je_bootstrap_free(dtv); 340 } 341 342 /* 343 * Allocate Static TLS using the Variant II method. 344 */ 345 void * 346 __libc_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign) 347 { 348 size_t size; 349 char *tls; 350 Elf_Addr *dtv; 351 Elf_Addr segbase, oldsegbase; 352 353 tcbalign = MAX(tcbalign, tls_init_align); 354 size = roundup2(tls_static_space, tcbalign); 355 356 if (tcbsize < 2 * sizeof(Elf_Addr)) 357 tcbsize = 2 * sizeof(Elf_Addr); 358 tls = malloc_aligned(size + tcbsize, tcbalign); 359 if (tls == NULL) { 360 tls_msg("__libc_allocate_tls: Out of memory.\n"); 361 abort(); 362 } 363 memset(tls, 0, size + tcbsize); 364 dtv = __je_bootstrap_malloc(3 * sizeof(Elf_Addr)); 365 if (dtv == NULL) { 366 tls_msg("__libc_allocate_tls: Out of memory.\n"); 367 abort(); 368 } 369 370 segbase = (Elf_Addr)(tls + size); 371 ((Elf_Addr*)segbase)[0] = segbase; 372 ((Elf_Addr*)segbase)[1] = (Elf_Addr) dtv; 373 374 dtv[0] = 1; 375 dtv[1] = 1; 376 dtv[2] = segbase - tls_static_space; 377 378 if (oldtls) { 379 /* 380 * Copy the static TLS block over whole. 381 */ 382 oldsegbase = (Elf_Addr) oldtls; 383 memcpy((void *)(segbase - tls_static_space), 384 (const void *)(oldsegbase - tls_static_space), 385 tls_static_space); 386 387 /* 388 * We assume that this block was the one we created with 389 * allocate_initial_tls(). 390 */ 391 _rtld_free_tls(oldtls, 2*sizeof(Elf_Addr), sizeof(Elf_Addr)); 392 } else { 393 memcpy((void *)(segbase - tls_static_space), 394 tls_init, tls_init_size); 395 memset((void *)(segbase - tls_static_space + tls_init_size), 396 0, tls_static_space - tls_init_size); 397 } 398 399 return (void*) segbase; 400 } 401 402 #endif /* TLS_VARIANT_II */ 403 404 #else 405 406 void * 407 __libc_allocate_tls(void *oldtls __unused, size_t tcbsize __unused, 408 size_t tcbalign __unused) 409 { 410 return (0); 411 } 412 413 void 414 __libc_free_tls(void *tcb __unused, size_t tcbsize __unused, 415 size_t tcbalign __unused) 416 { 417 } 418 419 #endif /* PIC */ 420 421 extern char **environ; 422 423 void 424 _init_tls(void) 425 { 426 #ifndef PIC 427 Elf_Addr *sp; 428 Elf_Auxinfo *aux, *auxp; 429 Elf_Phdr *phdr; 430 size_t phent, phnum; 431 int i; 432 void *tls; 433 434 sp = (Elf_Addr *) environ; 435 while (*sp++ != 0) 436 ; 437 aux = (Elf_Auxinfo *) sp; 438 phdr = NULL; 439 phent = phnum = 0; 440 for (auxp = aux; auxp->a_type != AT_NULL; auxp++) { 441 switch (auxp->a_type) { 442 case AT_PHDR: 443 phdr = auxp->a_un.a_ptr; 444 break; 445 446 case AT_PHENT: 447 phent = auxp->a_un.a_val; 448 break; 449 450 case AT_PHNUM: 451 phnum = auxp->a_un.a_val; 452 break; 453 } 454 } 455 if (phdr == NULL || phent != sizeof(Elf_Phdr) || phnum == 0) 456 return; 457 458 for (i = 0; (unsigned) i < phnum; i++) { 459 if (phdr[i].p_type == PT_TLS) { 460 tls_static_space = roundup2(phdr[i].p_memsz, 461 phdr[i].p_align); 462 tls_init_size = phdr[i].p_filesz; 463 tls_init_align = phdr[i].p_align; 464 tls_init = (void*) phdr[i].p_vaddr; 465 break; 466 } 467 } 468 tls = _rtld_allocate_tls(NULL, TLS_TCB_SIZE, TLS_TCB_ALIGN); 469 470 _set_tp(tls); 471 #endif 472 } 473