1 /*- 2 * Copyright 1999, 2000 John D. Polstra. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 17 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 18 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 19 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 20 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 21 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 22 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 23 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 24 * 25 * from: FreeBSD: src/libexec/rtld-elf/sparc64/lockdflt.c,v 1.3 2002/10/09 26 * $FreeBSD$ 27 */ 28 29 /* 30 * Thread locking implementation for the dynamic linker. 31 * 32 * We use the "simple, non-scalable reader-preference lock" from: 33 * 34 * J. M. Mellor-Crummey and M. L. Scott. "Scalable Reader-Writer 35 * Synchronization for Shared-Memory Multiprocessors." 3rd ACM Symp. on 36 * Principles and Practice of Parallel Programming, April 1991. 37 * 38 * In this algorithm the lock is a single word. Its low-order bit is 39 * set when a writer holds the lock. The remaining high-order bits 40 * contain a count of readers desiring the lock. The algorithm requires 41 * atomic "compare_and_store" and "add" operations, which we implement 42 * using assembly language sequences in "rtld_start.S". 43 */ 44 45 #include <sys/param.h> 46 #include <signal.h> 47 #include <stdlib.h> 48 #include <time.h> 49 50 #include "debug.h" 51 #include "rtld.h" 52 #include "rtld_machdep.h" 53 54 #define WAFLAG 0x1 /* A writer holds the lock */ 55 #define RC_INCR 0x2 /* Adjusts count of readers desiring lock */ 56 57 typedef struct Struct_Lock { 58 volatile u_int lock; 59 void *base; 60 } Lock; 61 62 static sigset_t fullsigmask, oldsigmask; 63 static int thread_flag; 64 65 static void * 66 def_lock_create() 67 { 68 void *base; 69 char *p; 70 uintptr_t r; 71 Lock *l; 72 73 /* 74 * Arrange for the lock to occupy its own cache line. First, we 75 * optimistically allocate just a cache line, hoping that malloc 76 * will give us a well-aligned block of memory. If that doesn't 77 * work, we allocate a larger block and take a well-aligned cache 78 * line from it. 79 */ 80 base = xmalloc(CACHE_LINE_SIZE); 81 p = (char *)base; 82 if ((uintptr_t)p % CACHE_LINE_SIZE != 0) { 83 free(base); 84 base = xmalloc(2 * CACHE_LINE_SIZE); 85 p = (char *)base; 86 if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0) 87 p += CACHE_LINE_SIZE - r; 88 } 89 l = (Lock *)p; 90 l->base = base; 91 l->lock = 0; 92 return l; 93 } 94 95 static void 96 def_lock_destroy(void *lock) 97 { 98 Lock *l = (Lock *)lock; 99 100 free(l->base); 101 } 102 103 static void 104 def_rlock_acquire(void *lock) 105 { 106 Lock *l = (Lock *)lock; 107 108 atomic_add_acq_int(&l->lock, RC_INCR); 109 while (l->lock & WAFLAG) 110 ; /* Spin */ 111 } 112 113 static void 114 def_wlock_acquire(void *lock) 115 { 116 Lock *l = (Lock *)lock; 117 sigset_t tmp_oldsigmask; 118 119 for ( ; ; ) { 120 sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask); 121 if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG)) 122 break; 123 sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL); 124 } 125 oldsigmask = tmp_oldsigmask; 126 } 127 128 static void 129 def_lock_release(void *lock) 130 { 131 Lock *l = (Lock *)lock; 132 133 if ((l->lock & WAFLAG) == 0) 134 atomic_add_rel_int(&l->lock, -RC_INCR); 135 else { 136 atomic_add_rel_int(&l->lock, -WAFLAG); 137 sigprocmask(SIG_SETMASK, &oldsigmask, NULL); 138 } 139 } 140 141 static int 142 def_thread_set_flag(int mask) 143 { 144 int old_val = thread_flag; 145 thread_flag |= mask; 146 return (old_val); 147 } 148 149 static int 150 def_thread_clr_flag(int mask) 151 { 152 int old_val = thread_flag; 153 thread_flag &= ~mask; 154 return (old_val); 155 } 156 157 /* 158 * Public interface exposed to the rest of the dynamic linker. 159 */ 160 static struct RtldLockInfo lockinfo; 161 static struct RtldLockInfo deflockinfo; 162 163 static __inline int 164 thread_mask_set(int mask) 165 { 166 return lockinfo.thread_set_flag(mask); 167 } 168 169 static __inline void 170 thread_mask_clear(int mask) 171 { 172 lockinfo.thread_clr_flag(mask); 173 } 174 175 #define RTLD_LOCK_CNT 3 176 struct rtld_lock { 177 void *handle; 178 int mask; 179 } rtld_locks[RTLD_LOCK_CNT]; 180 181 rtld_lock_t rtld_bind_lock = &rtld_locks[0]; 182 rtld_lock_t rtld_libc_lock = &rtld_locks[1]; 183 rtld_lock_t rtld_phdr_lock = &rtld_locks[2]; 184 185 #define print_ebp(str) do {register long ebp asm("ebp"); printf("%s 0x%0lx\n", str, ebp);} while (0) 186 187 void 188 rlock_acquire(rtld_lock_t lock, RtldLockState *lockstate) 189 { 190 191 if (lockstate == NULL) 192 return; 193 194 if (thread_mask_set(lock->mask) & lock->mask) { 195 dbg("rlock_acquire: recursed"); 196 lockstate->lockstate = RTLD_LOCK_UNLOCKED; 197 return; 198 } 199 lockinfo.rlock_acquire(lock->handle); 200 lockstate->lockstate = RTLD_LOCK_RLOCKED; 201 } 202 203 void 204 wlock_acquire(rtld_lock_t lock, RtldLockState *lockstate) 205 { 206 207 if (lockstate == NULL) 208 return; 209 210 if (thread_mask_set(lock->mask) & lock->mask) { 211 dbg("wlock_acquire: recursed"); 212 lockstate->lockstate = RTLD_LOCK_UNLOCKED; 213 return; 214 } 215 lockinfo.wlock_acquire(lock->handle); 216 lockstate->lockstate = RTLD_LOCK_WLOCKED; 217 } 218 219 void 220 lock_release(rtld_lock_t lock, RtldLockState *lockstate) 221 { 222 223 if (lockstate == NULL) 224 return; 225 226 switch (lockstate->lockstate) { 227 case RTLD_LOCK_UNLOCKED: 228 break; 229 case RTLD_LOCK_RLOCKED: 230 case RTLD_LOCK_WLOCKED: 231 thread_mask_clear(lock->mask); 232 lockinfo.lock_release(lock->handle); 233 break; 234 default: 235 assert(0); 236 } 237 } 238 239 void 240 lock_upgrade(rtld_lock_t lock, RtldLockState *lockstate) 241 { 242 243 if (lockstate == NULL) 244 return; 245 246 lock_release(lock, lockstate); 247 wlock_acquire(lock, lockstate); 248 } 249 250 void 251 lock_restart_for_upgrade(RtldLockState *lockstate) 252 { 253 254 if (lockstate == NULL) 255 return; 256 257 switch (lockstate->lockstate) { 258 case RTLD_LOCK_UNLOCKED: 259 case RTLD_LOCK_WLOCKED: 260 break; 261 case RTLD_LOCK_RLOCKED: 262 siglongjmp(lockstate->env, 1); 263 break; 264 default: 265 assert(0); 266 } 267 } 268 269 void 270 lockdflt_init() 271 { 272 int i; 273 274 deflockinfo.rtli_version = RTLI_VERSION; 275 deflockinfo.lock_create = def_lock_create; 276 deflockinfo.lock_destroy = def_lock_destroy; 277 deflockinfo.rlock_acquire = def_rlock_acquire; 278 deflockinfo.wlock_acquire = def_wlock_acquire; 279 deflockinfo.lock_release = def_lock_release; 280 deflockinfo.thread_set_flag = def_thread_set_flag; 281 deflockinfo.thread_clr_flag = def_thread_clr_flag; 282 deflockinfo.at_fork = NULL; 283 284 for (i = 0; i < RTLD_LOCK_CNT; i++) { 285 rtld_locks[i].mask = (1 << i); 286 rtld_locks[i].handle = NULL; 287 } 288 289 memcpy(&lockinfo, &deflockinfo, sizeof(lockinfo)); 290 _rtld_thread_init(NULL); 291 /* 292 * Construct a mask to block all signals except traps which might 293 * conceivably be generated within the dynamic linker itself. 294 */ 295 sigfillset(&fullsigmask); 296 sigdelset(&fullsigmask, SIGILL); 297 sigdelset(&fullsigmask, SIGTRAP); 298 sigdelset(&fullsigmask, SIGABRT); 299 sigdelset(&fullsigmask, SIGEMT); 300 sigdelset(&fullsigmask, SIGFPE); 301 sigdelset(&fullsigmask, SIGBUS); 302 sigdelset(&fullsigmask, SIGSEGV); 303 sigdelset(&fullsigmask, SIGSYS); 304 } 305 306 /* 307 * Callback function to allow threads implementation to 308 * register their own locking primitives if the default 309 * one is not suitable. 310 * The current context should be the only context 311 * executing at the invocation time. 312 */ 313 void 314 _rtld_thread_init(struct RtldLockInfo *pli) 315 { 316 int flags, i; 317 void *locks[RTLD_LOCK_CNT]; 318 319 /* disable all locking while this function is running */ 320 flags = thread_mask_set(~0); 321 322 if (pli == NULL) 323 pli = &deflockinfo; 324 325 326 for (i = 0; i < RTLD_LOCK_CNT; i++) 327 if ((locks[i] = pli->lock_create()) == NULL) 328 break; 329 330 if (i < RTLD_LOCK_CNT) { 331 while (--i >= 0) 332 pli->lock_destroy(locks[i]); 333 abort(); 334 } 335 336 for (i = 0; i < RTLD_LOCK_CNT; i++) { 337 if (rtld_locks[i].handle == NULL) 338 continue; 339 if (flags & rtld_locks[i].mask) 340 lockinfo.lock_release(rtld_locks[i].handle); 341 lockinfo.lock_destroy(rtld_locks[i].handle); 342 } 343 344 for (i = 0; i < RTLD_LOCK_CNT; i++) { 345 rtld_locks[i].handle = locks[i]; 346 if (flags & rtld_locks[i].mask) 347 pli->wlock_acquire(rtld_locks[i].handle); 348 } 349 350 lockinfo.lock_create = pli->lock_create; 351 lockinfo.lock_destroy = pli->lock_destroy; 352 lockinfo.rlock_acquire = pli->rlock_acquire; 353 lockinfo.wlock_acquire = pli->wlock_acquire; 354 lockinfo.lock_release = pli->lock_release; 355 lockinfo.thread_set_flag = pli->thread_set_flag; 356 lockinfo.thread_clr_flag = pli->thread_clr_flag; 357 lockinfo.at_fork = pli->at_fork; 358 359 /* restore thread locking state, this time with new locks */ 360 thread_mask_clear(~0); 361 thread_mask_set(flags); 362 dbg("_rtld_thread_init: done"); 363 } 364 365 void 366 _rtld_atfork_pre(int *locks) 367 { 368 RtldLockState ls[2]; 369 370 wlock_acquire(rtld_phdr_lock, &ls[0]); 371 rlock_acquire(rtld_bind_lock, &ls[1]); 372 373 /* XXXKIB: I am really sorry for this. */ 374 locks[0] = ls[1].lockstate; 375 locks[2] = ls[0].lockstate; 376 } 377 378 void 379 _rtld_atfork_post(int *locks) 380 { 381 RtldLockState ls[2]; 382 383 bzero(ls, sizeof(ls)); 384 ls[0].lockstate = locks[2]; 385 ls[1].lockstate = locks[0]; 386 lock_release(rtld_bind_lock, &ls[1]); 387 lock_release(rtld_phdr_lock, &ls[0]); 388 } 389