1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright 1999, 2000 John D. Polstra. 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 ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 20 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 21 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 22 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 23 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 24 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 25 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 * 27 * from: FreeBSD: src/libexec/rtld-elf/sparc64/lockdflt.c,v 1.3 2002/10/09 28 */ 29 30 /* 31 * Thread locking implementation for the dynamic linker. 32 * 33 * We use the "simple, non-scalable reader-preference lock" from: 34 * 35 * J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer 36 * Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on 37 * Principles and Practice of Parallel Programming, April 1991. 38 * 39 * In this algorithm the lock is a single word. Its low-order bit is 40 * set when a writer holds the lock. The remaining high-order bits 41 * contain a count of readers desiring the lock. The algorithm requires 42 * atomic "compare_and_store" and "add" operations, which we take 43 * from machine/atomic.h. 44 */ 45 46 #include <sys/param.h> 47 #include <sys/signalvar.h> 48 #include <signal.h> 49 #include <stdlib.h> 50 #include <time.h> 51 52 #include "debug.h" 53 #include "rtld.h" 54 #include "rtld_machdep.h" 55 #include "rtld_libc.h" 56 57 void _rtld_thread_init(struct RtldLockInfo *) __exported; 58 void _rtld_atfork_pre(int *) __exported; 59 void _rtld_atfork_post(int *) __exported; 60 61 static char def_dlerror_msg[512]; 62 static int def_dlerror_seen_val = 1; 63 64 static char * 65 def_dlerror_loc(void) 66 { 67 return (def_dlerror_msg); 68 } 69 70 static int * 71 def_dlerror_seen(void) 72 { 73 return (&def_dlerror_seen_val); 74 } 75 76 #define WAFLAG 0x1 /* A writer holds the lock */ 77 #define RC_INCR 0x2 /* Adjusts count of readers desiring lock */ 78 79 typedef struct Struct_Lock { 80 volatile u_int lock; 81 void *base; 82 } Lock; 83 84 static sigset_t fullsigmask, oldsigmask; 85 static int thread_flag, wnested; 86 static uint32_t fsigblock; 87 88 static void * 89 def_lock_create(void) 90 { 91 void *base; 92 char *p; 93 uintptr_t r; 94 Lock *l; 95 96 /* 97 * Arrange for the lock to occupy its own cache line. First, we 98 * optimistically allocate just a cache line, hoping that malloc 99 * will give us a well-aligned block of memory. If that doesn't 100 * work, we allocate a larger block and take a well-aligned cache 101 * line from it. 102 */ 103 base = xmalloc(CACHE_LINE_SIZE); 104 p = base; 105 if ((uintptr_t)p % CACHE_LINE_SIZE != 0) { 106 free(base); 107 base = xmalloc(2 * CACHE_LINE_SIZE); 108 p = base; 109 if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0) 110 p += CACHE_LINE_SIZE - r; 111 } 112 l = (Lock *)p; 113 l->base = base; 114 l->lock = 0; 115 return (l); 116 } 117 118 static void 119 def_lock_destroy(void *lock) 120 { 121 Lock *l = lock; 122 123 free(l->base); 124 } 125 126 static void 127 sig_fastunblock(void) 128 { 129 uint32_t oldval; 130 131 assert((fsigblock & ~SIGFASTBLOCK_FLAGS) >= SIGFASTBLOCK_INC); 132 oldval = atomic_fetchadd_32(&fsigblock, -SIGFASTBLOCK_INC); 133 if (oldval == (SIGFASTBLOCK_PEND | SIGFASTBLOCK_INC)) 134 __sys_sigfastblock(SIGFASTBLOCK_UNBLOCK, NULL); 135 } 136 137 static bool 138 def_lock_acquire_set(Lock *l, bool wlock) 139 { 140 if (wlock) { 141 if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG)) 142 return (true); 143 } else { 144 atomic_add_acq_int(&l->lock, RC_INCR); 145 if ((l->lock & WAFLAG) == 0) 146 return (true); 147 atomic_add_int(&l->lock, -RC_INCR); 148 } 149 return (false); 150 } 151 152 static void 153 def_lock_acquire(Lock *l, bool wlock) 154 { 155 sigset_t tmp_oldsigmask; 156 157 if (ld_fast_sigblock) { 158 for (;;) { 159 atomic_add_32(&fsigblock, SIGFASTBLOCK_INC); 160 if (def_lock_acquire_set(l, wlock)) 161 break; 162 sig_fastunblock(); 163 } 164 } else { 165 for (;;) { 166 sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask); 167 if (def_lock_acquire_set(l, wlock)) 168 break; 169 sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL); 170 } 171 if (atomic_fetchadd_int(&wnested, 1) == 0) 172 oldsigmask = tmp_oldsigmask; 173 } 174 } 175 176 static void 177 def_rlock_acquire(void *lock) 178 { 179 def_lock_acquire(lock, false); 180 } 181 182 static void 183 def_wlock_acquire(void *lock) 184 { 185 def_lock_acquire(lock, true); 186 } 187 188 static void 189 def_lock_release(void *lock) 190 { 191 Lock *l = lock; 192 193 atomic_add_rel_int(&l->lock, -((l->lock & WAFLAG) == 0 ? 194 RC_INCR : WAFLAG)); 195 if (ld_fast_sigblock) 196 sig_fastunblock(); 197 else if (atomic_fetchadd_int(&wnested, -1) == 1) 198 sigprocmask(SIG_SETMASK, &oldsigmask, NULL); 199 } 200 201 static int 202 def_thread_set_flag(int mask) 203 { 204 int old_val = thread_flag; 205 206 thread_flag |= mask; 207 return (old_val); 208 } 209 210 static int 211 def_thread_clr_flag(int mask) 212 { 213 int old_val = thread_flag; 214 215 thread_flag &= ~mask; 216 return (old_val); 217 } 218 219 /* 220 * Public interface exposed to the rest of the dynamic linker. 221 */ 222 struct RtldLockInfo lockinfo; 223 static struct RtldLockInfo deflockinfo; 224 225 static __inline int 226 thread_mask_set(int mask) 227 { 228 return (lockinfo.thread_set_flag(mask)); 229 } 230 231 static __inline void 232 thread_mask_clear(int mask) 233 { 234 lockinfo.thread_clr_flag(mask); 235 } 236 237 #define RTLD_LOCK_CNT 3 238 static struct rtld_lock { 239 void *handle; 240 int mask; 241 } rtld_locks[RTLD_LOCK_CNT]; 242 243 rtld_lock_t rtld_bind_lock = &rtld_locks[0]; 244 rtld_lock_t rtld_libc_lock = &rtld_locks[1]; 245 rtld_lock_t rtld_phdr_lock = &rtld_locks[2]; 246 247 void 248 rlock_acquire(rtld_lock_t lock, RtldLockState *lockstate) 249 { 250 251 if (lockstate == NULL) 252 return; 253 254 if ((thread_mask_set(lock->mask) & lock->mask) != 0) { 255 dbg("rlock_acquire: recursed"); 256 lockstate->lockstate = RTLD_LOCK_UNLOCKED; 257 return; 258 } 259 lockinfo.rlock_acquire(lock->handle); 260 lockstate->lockstate = RTLD_LOCK_RLOCKED; 261 } 262 263 void 264 wlock_acquire(rtld_lock_t lock, RtldLockState *lockstate) 265 { 266 267 if (lockstate == NULL) 268 return; 269 270 if ((thread_mask_set(lock->mask) & lock->mask) != 0) { 271 dbg("wlock_acquire: recursed"); 272 lockstate->lockstate = RTLD_LOCK_UNLOCKED; 273 return; 274 } 275 lockinfo.wlock_acquire(lock->handle); 276 lockstate->lockstate = RTLD_LOCK_WLOCKED; 277 } 278 279 void 280 lock_release(rtld_lock_t lock, RtldLockState *lockstate) 281 { 282 283 if (lockstate == NULL) 284 return; 285 286 switch (lockstate->lockstate) { 287 case RTLD_LOCK_UNLOCKED: 288 break; 289 case RTLD_LOCK_RLOCKED: 290 case RTLD_LOCK_WLOCKED: 291 thread_mask_clear(lock->mask); 292 lockinfo.lock_release(lock->handle); 293 break; 294 default: 295 assert(0); 296 } 297 } 298 299 void 300 lock_upgrade(rtld_lock_t lock, RtldLockState *lockstate) 301 { 302 303 if (lockstate == NULL) 304 return; 305 306 lock_release(lock, lockstate); 307 wlock_acquire(lock, lockstate); 308 } 309 310 void 311 lock_restart_for_upgrade(RtldLockState *lockstate) 312 { 313 314 if (lockstate == NULL) 315 return; 316 317 switch (lockstate->lockstate) { 318 case RTLD_LOCK_UNLOCKED: 319 case RTLD_LOCK_WLOCKED: 320 break; 321 case RTLD_LOCK_RLOCKED: 322 siglongjmp(lockstate->env, 1); 323 break; 324 default: 325 assert(0); 326 } 327 } 328 329 bool 330 lockstate_wlocked(const RtldLockState *lockstate) 331 { 332 return (lockstate->lockstate == RTLD_LOCK_WLOCKED); 333 } 334 335 void 336 dlerror_dflt_init(void) 337 { 338 lockinfo.dlerror_loc = def_dlerror_loc; 339 lockinfo.dlerror_loc_sz = sizeof(def_dlerror_msg); 340 lockinfo.dlerror_seen = def_dlerror_seen; 341 } 342 343 void 344 lockdflt_init(void) 345 { 346 int i; 347 348 deflockinfo.rtli_version = RTLI_VERSION; 349 deflockinfo.lock_create = def_lock_create; 350 deflockinfo.lock_destroy = def_lock_destroy; 351 deflockinfo.rlock_acquire = def_rlock_acquire; 352 deflockinfo.wlock_acquire = def_wlock_acquire; 353 deflockinfo.lock_release = def_lock_release; 354 deflockinfo.thread_set_flag = def_thread_set_flag; 355 deflockinfo.thread_clr_flag = def_thread_clr_flag; 356 deflockinfo.at_fork = NULL; 357 deflockinfo.dlerror_loc = def_dlerror_loc; 358 deflockinfo.dlerror_loc_sz = sizeof(def_dlerror_msg); 359 deflockinfo.dlerror_seen = def_dlerror_seen; 360 361 for (i = 0; i < RTLD_LOCK_CNT; i++) { 362 rtld_locks[i].mask = (1 << i); 363 rtld_locks[i].handle = NULL; 364 } 365 366 memcpy(&lockinfo, &deflockinfo, sizeof(lockinfo)); 367 _rtld_thread_init(NULL); 368 if (ld_fast_sigblock) { 369 __sys_sigfastblock(SIGFASTBLOCK_SETPTR, &fsigblock); 370 } else { 371 /* 372 * Construct a mask to block all signals. Note that 373 * blocked traps mean that the process is terminated 374 * if trap occurs while we are in locked section, with 375 * the default settings for kern.forcesigexit. 376 */ 377 sigfillset(&fullsigmask); 378 } 379 } 380 381 /* 382 * Callback function to allow threads implementation to 383 * register their own locking primitives if the default 384 * one is not suitable. 385 * The current context should be the only context 386 * executing at the invocation time. 387 */ 388 void 389 _rtld_thread_init(struct RtldLockInfo *pli) 390 { 391 const Obj_Entry *obj; 392 SymLook req; 393 void *locks[RTLD_LOCK_CNT]; 394 int flags, i, res; 395 396 if (pli == NULL) { 397 lockinfo.rtli_version = RTLI_VERSION; 398 } else { 399 lockinfo.rtli_version = RTLI_VERSION_ONE; 400 obj = obj_from_addr(pli->lock_create); 401 if (obj != NULL) { 402 symlook_init(&req, "_pli_rtli_version"); 403 res = symlook_obj(&req, obj); 404 if (res == 0) 405 lockinfo.rtli_version = pli->rtli_version; 406 } 407 } 408 409 /* disable all locking while this function is running */ 410 flags = thread_mask_set(~0); 411 412 if (pli == NULL) 413 pli = &deflockinfo; 414 else if (ld_fast_sigblock) { 415 fsigblock = 0; 416 __sys_sigfastblock(SIGFASTBLOCK_UNSETPTR, NULL); 417 } 418 419 for (i = 0; i < RTLD_LOCK_CNT; i++) 420 if ((locks[i] = pli->lock_create()) == NULL) 421 break; 422 423 if (i < RTLD_LOCK_CNT) { 424 while (--i >= 0) 425 pli->lock_destroy(locks[i]); 426 abort(); 427 } 428 429 for (i = 0; i < RTLD_LOCK_CNT; i++) { 430 if (rtld_locks[i].handle == NULL) 431 continue; 432 if (flags & rtld_locks[i].mask) 433 lockinfo.lock_release(rtld_locks[i].handle); 434 lockinfo.lock_destroy(rtld_locks[i].handle); 435 } 436 437 for (i = 0; i < RTLD_LOCK_CNT; i++) { 438 rtld_locks[i].handle = locks[i]; 439 if (flags & rtld_locks[i].mask) 440 pli->wlock_acquire(rtld_locks[i].handle); 441 } 442 443 lockinfo.lock_create = pli->lock_create; 444 lockinfo.lock_destroy = pli->lock_destroy; 445 lockinfo.rlock_acquire = pli->rlock_acquire; 446 lockinfo.wlock_acquire = pli->wlock_acquire; 447 lockinfo.lock_release = pli->lock_release; 448 lockinfo.thread_set_flag = pli->thread_set_flag; 449 lockinfo.thread_clr_flag = pli->thread_clr_flag; 450 lockinfo.at_fork = pli->at_fork; 451 if (lockinfo.rtli_version > RTLI_VERSION_ONE && pli != NULL) { 452 strlcpy(pli->dlerror_loc(), lockinfo.dlerror_loc(), 453 lockinfo.dlerror_loc_sz); 454 lockinfo.dlerror_loc = pli->dlerror_loc; 455 lockinfo.dlerror_loc_sz = pli->dlerror_loc_sz; 456 lockinfo.dlerror_seen = pli->dlerror_seen; 457 } 458 459 /* restore thread locking state, this time with new locks */ 460 thread_mask_clear(~0); 461 thread_mask_set(flags); 462 dbg("_rtld_thread_init: done"); 463 } 464 465 void 466 _rtld_atfork_pre(int *locks) 467 { 468 RtldLockState ls[2]; 469 470 if (locks == NULL) 471 return; 472 bzero(ls, sizeof(ls)); 473 474 /* 475 * Warning: this did not worked well with the rtld compat 476 * locks above, when the thread signal mask was corrupted (set 477 * to all signals blocked) if two locks were taken 478 * simultaneously in the write mode. The caller of the 479 * _rtld_atfork_pre() must provide the working implementation 480 * of the locks anyway, and libthr locks are fine. 481 */ 482 if (ld_get_env_var(LD_NO_DL_ITERATE_PHDR_AFTER_FORK) == NULL) 483 wlock_acquire(rtld_phdr_lock, &ls[0]); 484 wlock_acquire(rtld_bind_lock, &ls[1]); 485 486 /* XXXKIB: I am really sorry for this. */ 487 locks[0] = ls[1].lockstate; 488 locks[2] = ls[0].lockstate; 489 } 490 491 void 492 _rtld_atfork_post(int *locks) 493 { 494 RtldLockState ls[2]; 495 496 if (locks == NULL) 497 return; 498 499 bzero(ls, sizeof(ls)); 500 ls[0].lockstate = locks[2]; 501 ls[1].lockstate = locks[0]; 502 lock_release(rtld_bind_lock, &ls[1]); 503 if (ld_get_env_var(LD_NO_DL_ITERATE_PHDR_AFTER_FORK) == NULL) 504 lock_release(rtld_phdr_lock, &ls[0]); 505 } 506