1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 * $FreeBSD$ 29 */ 30 31 /* 32 * Thread locking implementation for the dynamic linker. 33 * 34 * We use the "simple, non-scalable reader-preference lock" from: 35 * 36 * J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer 37 * Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on 38 * Principles and Practice of Parallel Programming, April 1991. 39 * 40 * In this algorithm the lock is a single word. Its low-order bit is 41 * set when a writer holds the lock. The remaining high-order bits 42 * contain a count of readers desiring the lock. The algorithm requires 43 * atomic "compare_and_store" and "add" operations, which we take 44 * from machine/atomic.h. 45 */ 46 47 #include <sys/param.h> 48 #include <sys/signalvar.h> 49 #include <signal.h> 50 #include <stdlib.h> 51 #include <time.h> 52 53 #include "debug.h" 54 #include "rtld.h" 55 #include "rtld_machdep.h" 56 #include "rtld_libc.h" 57 58 void _rtld_thread_init(struct RtldLockInfo *) __exported; 59 void _rtld_atfork_pre(int *) __exported; 60 void _rtld_atfork_post(int *) __exported; 61 62 #define WAFLAG 0x1 /* A writer holds the lock */ 63 #define RC_INCR 0x2 /* Adjusts count of readers desiring lock */ 64 65 typedef struct Struct_Lock { 66 volatile u_int lock; 67 void *base; 68 } Lock; 69 70 static sigset_t fullsigmask, oldsigmask; 71 static int thread_flag, wnested; 72 static uint32_t fsigblock; 73 74 static void * 75 def_lock_create(void) 76 { 77 void *base; 78 char *p; 79 uintptr_t r; 80 Lock *l; 81 82 /* 83 * Arrange for the lock to occupy its own cache line. First, we 84 * optimistically allocate just a cache line, hoping that malloc 85 * will give us a well-aligned block of memory. If that doesn't 86 * work, we allocate a larger block and take a well-aligned cache 87 * line from it. 88 */ 89 base = xmalloc(CACHE_LINE_SIZE); 90 p = (char *)base; 91 if ((uintptr_t)p % CACHE_LINE_SIZE != 0) { 92 free(base); 93 base = xmalloc(2 * CACHE_LINE_SIZE); 94 p = (char *)base; 95 if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0) 96 p += CACHE_LINE_SIZE - r; 97 } 98 l = (Lock *)p; 99 l->base = base; 100 l->lock = 0; 101 return l; 102 } 103 104 static void 105 def_lock_destroy(void *lock) 106 { 107 Lock *l = (Lock *)lock; 108 109 free(l->base); 110 } 111 112 static void 113 def_rlock_acquire(void *lock) 114 { 115 Lock *l = (Lock *)lock; 116 117 atomic_add_acq_int(&l->lock, RC_INCR); 118 while (l->lock & WAFLAG) 119 ; /* Spin */ 120 } 121 122 static void 123 sig_fastunblock(void) 124 { 125 uint32_t oldval; 126 127 assert((fsigblock & ~SIGFASTBLOCK_FLAGS) >= SIGFASTBLOCK_INC); 128 oldval = atomic_fetchadd_32(&fsigblock, -SIGFASTBLOCK_INC); 129 if (oldval == (SIGFASTBLOCK_PEND | SIGFASTBLOCK_INC)) 130 __sys_sigfastblock(SIGFASTBLOCK_UNBLOCK, NULL); 131 } 132 133 static void 134 def_wlock_acquire(void *lock) 135 { 136 Lock *l; 137 sigset_t tmp_oldsigmask; 138 139 l = (Lock *)lock; 140 if (ld_fast_sigblock) { 141 for (;;) { 142 atomic_add_32(&fsigblock, SIGFASTBLOCK_INC); 143 if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG)) 144 break; 145 sig_fastunblock(); 146 } 147 } else { 148 for (;;) { 149 sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask); 150 if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG)) 151 break; 152 sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL); 153 } 154 if (atomic_fetchadd_int(&wnested, 1) == 0) 155 oldsigmask = tmp_oldsigmask; 156 } 157 } 158 159 static void 160 def_lock_release(void *lock) 161 { 162 Lock *l; 163 164 l = (Lock *)lock; 165 if ((l->lock & WAFLAG) == 0) 166 atomic_add_rel_int(&l->lock, -RC_INCR); 167 else { 168 atomic_add_rel_int(&l->lock, -WAFLAG); 169 if (ld_fast_sigblock) 170 sig_fastunblock(); 171 else if (atomic_fetchadd_int(&wnested, -1) == 1) 172 sigprocmask(SIG_SETMASK, &oldsigmask, NULL); 173 } 174 } 175 176 static int 177 def_thread_set_flag(int mask) 178 { 179 int old_val = thread_flag; 180 thread_flag |= mask; 181 return (old_val); 182 } 183 184 static int 185 def_thread_clr_flag(int mask) 186 { 187 int old_val = thread_flag; 188 thread_flag &= ~mask; 189 return (old_val); 190 } 191 192 /* 193 * Public interface exposed to the rest of the dynamic linker. 194 */ 195 static struct RtldLockInfo lockinfo; 196 static struct RtldLockInfo deflockinfo; 197 198 static __inline int 199 thread_mask_set(int mask) 200 { 201 return lockinfo.thread_set_flag(mask); 202 } 203 204 static __inline void 205 thread_mask_clear(int mask) 206 { 207 lockinfo.thread_clr_flag(mask); 208 } 209 210 #define RTLD_LOCK_CNT 3 211 static struct rtld_lock { 212 void *handle; 213 int mask; 214 } rtld_locks[RTLD_LOCK_CNT]; 215 216 rtld_lock_t rtld_bind_lock = &rtld_locks[0]; 217 rtld_lock_t rtld_libc_lock = &rtld_locks[1]; 218 rtld_lock_t rtld_phdr_lock = &rtld_locks[2]; 219 220 void 221 rlock_acquire(rtld_lock_t lock, RtldLockState *lockstate) 222 { 223 224 if (lockstate == NULL) 225 return; 226 227 if (thread_mask_set(lock->mask) & lock->mask) { 228 dbg("rlock_acquire: recursed"); 229 lockstate->lockstate = RTLD_LOCK_UNLOCKED; 230 return; 231 } 232 lockinfo.rlock_acquire(lock->handle); 233 lockstate->lockstate = RTLD_LOCK_RLOCKED; 234 } 235 236 void 237 wlock_acquire(rtld_lock_t lock, RtldLockState *lockstate) 238 { 239 240 if (lockstate == NULL) 241 return; 242 243 if (thread_mask_set(lock->mask) & lock->mask) { 244 dbg("wlock_acquire: recursed"); 245 lockstate->lockstate = RTLD_LOCK_UNLOCKED; 246 return; 247 } 248 lockinfo.wlock_acquire(lock->handle); 249 lockstate->lockstate = RTLD_LOCK_WLOCKED; 250 } 251 252 void 253 lock_release(rtld_lock_t lock, RtldLockState *lockstate) 254 { 255 256 if (lockstate == NULL) 257 return; 258 259 switch (lockstate->lockstate) { 260 case RTLD_LOCK_UNLOCKED: 261 break; 262 case RTLD_LOCK_RLOCKED: 263 case RTLD_LOCK_WLOCKED: 264 thread_mask_clear(lock->mask); 265 lockinfo.lock_release(lock->handle); 266 break; 267 default: 268 assert(0); 269 } 270 } 271 272 void 273 lock_upgrade(rtld_lock_t lock, RtldLockState *lockstate) 274 { 275 276 if (lockstate == NULL) 277 return; 278 279 lock_release(lock, lockstate); 280 wlock_acquire(lock, lockstate); 281 } 282 283 void 284 lock_restart_for_upgrade(RtldLockState *lockstate) 285 { 286 287 if (lockstate == NULL) 288 return; 289 290 switch (lockstate->lockstate) { 291 case RTLD_LOCK_UNLOCKED: 292 case RTLD_LOCK_WLOCKED: 293 break; 294 case RTLD_LOCK_RLOCKED: 295 siglongjmp(lockstate->env, 1); 296 break; 297 default: 298 assert(0); 299 } 300 } 301 302 void 303 lockdflt_init(void) 304 { 305 int i; 306 307 deflockinfo.rtli_version = RTLI_VERSION; 308 deflockinfo.lock_create = def_lock_create; 309 deflockinfo.lock_destroy = def_lock_destroy; 310 deflockinfo.rlock_acquire = def_rlock_acquire; 311 deflockinfo.wlock_acquire = def_wlock_acquire; 312 deflockinfo.lock_release = def_lock_release; 313 deflockinfo.thread_set_flag = def_thread_set_flag; 314 deflockinfo.thread_clr_flag = def_thread_clr_flag; 315 deflockinfo.at_fork = NULL; 316 317 for (i = 0; i < RTLD_LOCK_CNT; i++) { 318 rtld_locks[i].mask = (1 << i); 319 rtld_locks[i].handle = NULL; 320 } 321 322 memcpy(&lockinfo, &deflockinfo, sizeof(lockinfo)); 323 _rtld_thread_init(NULL); 324 if (ld_fast_sigblock) { 325 __sys_sigfastblock(SIGFASTBLOCK_SETPTR, &fsigblock); 326 } else { 327 /* 328 * Construct a mask to block all signals. Note that 329 * blocked traps mean that the process is terminated 330 * if trap occurs while we are in locked section, with 331 * the default settings for kern.forcesigexit. 332 */ 333 sigfillset(&fullsigmask); 334 } 335 } 336 337 /* 338 * Callback function to allow threads implementation to 339 * register their own locking primitives if the default 340 * one is not suitable. 341 * The current context should be the only context 342 * executing at the invocation time. 343 */ 344 void 345 _rtld_thread_init(struct RtldLockInfo *pli) 346 { 347 int flags, i; 348 void *locks[RTLD_LOCK_CNT]; 349 350 /* disable all locking while this function is running */ 351 flags = thread_mask_set(~0); 352 353 if (pli == NULL) 354 pli = &deflockinfo; 355 else if (ld_fast_sigblock) { 356 fsigblock = 0; 357 __sys_sigfastblock(SIGFASTBLOCK_UNSETPTR, NULL); 358 } 359 360 for (i = 0; i < RTLD_LOCK_CNT; i++) 361 if ((locks[i] = pli->lock_create()) == NULL) 362 break; 363 364 if (i < RTLD_LOCK_CNT) { 365 while (--i >= 0) 366 pli->lock_destroy(locks[i]); 367 abort(); 368 } 369 370 for (i = 0; i < RTLD_LOCK_CNT; i++) { 371 if (rtld_locks[i].handle == NULL) 372 continue; 373 if (flags & rtld_locks[i].mask) 374 lockinfo.lock_release(rtld_locks[i].handle); 375 lockinfo.lock_destroy(rtld_locks[i].handle); 376 } 377 378 for (i = 0; i < RTLD_LOCK_CNT; i++) { 379 rtld_locks[i].handle = locks[i]; 380 if (flags & rtld_locks[i].mask) 381 pli->wlock_acquire(rtld_locks[i].handle); 382 } 383 384 lockinfo.lock_create = pli->lock_create; 385 lockinfo.lock_destroy = pli->lock_destroy; 386 lockinfo.rlock_acquire = pli->rlock_acquire; 387 lockinfo.wlock_acquire = pli->wlock_acquire; 388 lockinfo.lock_release = pli->lock_release; 389 lockinfo.thread_set_flag = pli->thread_set_flag; 390 lockinfo.thread_clr_flag = pli->thread_clr_flag; 391 lockinfo.at_fork = pli->at_fork; 392 393 /* restore thread locking state, this time with new locks */ 394 thread_mask_clear(~0); 395 thread_mask_set(flags); 396 dbg("_rtld_thread_init: done"); 397 } 398 399 void 400 _rtld_atfork_pre(int *locks) 401 { 402 RtldLockState ls[2]; 403 404 if (locks == NULL) 405 return; 406 407 /* 408 * Warning: this did not worked well with the rtld compat 409 * locks above, when the thread signal mask was corrupted (set 410 * to all signals blocked) if two locks were taken 411 * simultaneously in the write mode. The caller of the 412 * _rtld_atfork_pre() must provide the working implementation 413 * of the locks anyway, and libthr locks are fine. 414 */ 415 wlock_acquire(rtld_phdr_lock, &ls[0]); 416 wlock_acquire(rtld_bind_lock, &ls[1]); 417 418 /* XXXKIB: I am really sorry for this. */ 419 locks[0] = ls[1].lockstate; 420 locks[2] = ls[0].lockstate; 421 } 422 423 void 424 _rtld_atfork_post(int *locks) 425 { 426 RtldLockState ls[2]; 427 428 if (locks == NULL) 429 return; 430 431 bzero(ls, sizeof(ls)); 432 ls[0].lockstate = locks[2]; 433 ls[1].lockstate = locks[0]; 434 lock_release(rtld_bind_lock, &ls[1]); 435 lock_release(rtld_phdr_lock, &ls[0]); 436 } 437