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 <signal.h> 46 #include <stdlib.h> 47 #include <time.h> 48 49 #include "debug.h" 50 #include "rtld.h" 51 #include "rtld_machdep.h" 52 53 #define WAFLAG 0x1 /* A writer holds the lock */ 54 #define RC_INCR 0x2 /* Adjusts count of readers desiring lock */ 55 56 typedef struct Struct_Lock { 57 volatile u_int lock; 58 void *base; 59 } Lock; 60 61 static sigset_t fullsigmask, oldsigmask; 62 static int thread_flag; 63 64 static void * 65 def_lock_create() 66 { 67 void *base; 68 char *p; 69 uintptr_t r; 70 Lock *l; 71 72 /* 73 * Arrange for the lock to occupy its own cache line. First, we 74 * optimistically allocate just a cache line, hoping that malloc 75 * will give us a well-aligned block of memory. If that doesn't 76 * work, we allocate a larger block and take a well-aligned cache 77 * line from it. 78 */ 79 base = xmalloc(CACHE_LINE_SIZE); 80 p = (char *)base; 81 if ((uintptr_t)p % CACHE_LINE_SIZE != 0) { 82 free(base); 83 base = xmalloc(2 * CACHE_LINE_SIZE); 84 p = (char *)base; 85 if ((r = (uintptr_t)p % CACHE_LINE_SIZE) != 0) 86 p += CACHE_LINE_SIZE - r; 87 } 88 l = (Lock *)p; 89 l->base = base; 90 l->lock = 0; 91 return l; 92 } 93 94 static void 95 def_lock_destroy(void *lock) 96 { 97 Lock *l = (Lock *)lock; 98 99 free(l->base); 100 } 101 102 static void 103 def_rlock_acquire(void *lock) 104 { 105 Lock *l = (Lock *)lock; 106 107 atomic_add_acq_int(&l->lock, RC_INCR); 108 while (l->lock & WAFLAG) 109 ; /* Spin */ 110 } 111 112 static void 113 def_wlock_acquire(void *lock) 114 { 115 Lock *l = (Lock *)lock; 116 sigset_t tmp_oldsigmask; 117 118 for ( ; ; ) { 119 sigprocmask(SIG_BLOCK, &fullsigmask, &tmp_oldsigmask); 120 if (atomic_cmpset_acq_int(&l->lock, 0, WAFLAG)) 121 break; 122 sigprocmask(SIG_SETMASK, &tmp_oldsigmask, NULL); 123 } 124 oldsigmask = tmp_oldsigmask; 125 } 126 127 static void 128 def_lock_release(void *lock) 129 { 130 Lock *l = (Lock *)lock; 131 132 if ((l->lock & WAFLAG) == 0) 133 atomic_add_rel_int(&l->lock, -RC_INCR); 134 else { 135 atomic_add_rel_int(&l->lock, -WAFLAG); 136 sigprocmask(SIG_SETMASK, &oldsigmask, NULL); 137 } 138 } 139 140 static int 141 def_thread_set_flag(int mask) 142 { 143 int old_val = thread_flag; 144 thread_flag |= mask; 145 return (old_val); 146 } 147 148 static int 149 def_thread_clr_flag(int mask) 150 { 151 int old_val = thread_flag; 152 thread_flag &= ~mask; 153 return (old_val); 154 } 155 156 /* 157 * Public interface exposed to the rest of the dynamic linker. 158 */ 159 static struct RtldLockInfo lockinfo; 160 static struct RtldLockInfo deflockinfo; 161 162 static __inline int 163 thread_mask_set(int mask) 164 { 165 return lockinfo.thread_set_flag(mask); 166 } 167 168 static __inline void 169 thread_mask_clear(int mask) 170 { 171 lockinfo.thread_clr_flag(mask); 172 } 173 174 #define RTLD_LOCK_CNT 3 175 struct rtld_lock { 176 void *handle; 177 int mask; 178 } rtld_locks[RTLD_LOCK_CNT]; 179 180 rtld_lock_t rtld_bind_lock = &rtld_locks[0]; 181 rtld_lock_t rtld_libc_lock = &rtld_locks[1]; 182 rtld_lock_t rtld_phdr_lock = &rtld_locks[2]; 183 184 int 185 rlock_acquire(rtld_lock_t lock) 186 { 187 if (thread_mask_set(lock->mask) & lock->mask) { 188 dbg("rlock_acquire: recursed"); 189 return (0); 190 } 191 lockinfo.rlock_acquire(lock->handle); 192 return (1); 193 } 194 195 int 196 wlock_acquire(rtld_lock_t lock) 197 { 198 if (thread_mask_set(lock->mask) & lock->mask) { 199 dbg("wlock_acquire: recursed"); 200 return (0); 201 } 202 lockinfo.wlock_acquire(lock->handle); 203 return (1); 204 } 205 206 void 207 rlock_release(rtld_lock_t lock, int locked) 208 { 209 if (locked == 0) 210 return; 211 thread_mask_clear(lock->mask); 212 lockinfo.lock_release(lock->handle); 213 } 214 215 void 216 wlock_release(rtld_lock_t lock, int locked) 217 { 218 if (locked == 0) 219 return; 220 thread_mask_clear(lock->mask); 221 lockinfo.lock_release(lock->handle); 222 } 223 224 void 225 lockdflt_init() 226 { 227 int i; 228 229 deflockinfo.rtli_version = RTLI_VERSION; 230 deflockinfo.lock_create = def_lock_create; 231 deflockinfo.lock_destroy = def_lock_destroy; 232 deflockinfo.rlock_acquire = def_rlock_acquire; 233 deflockinfo.wlock_acquire = def_wlock_acquire; 234 deflockinfo.lock_release = def_lock_release; 235 deflockinfo.thread_set_flag = def_thread_set_flag; 236 deflockinfo.thread_clr_flag = def_thread_clr_flag; 237 deflockinfo.at_fork = NULL; 238 239 for (i = 0; i < RTLD_LOCK_CNT; i++) { 240 rtld_locks[i].mask = (1 << i); 241 rtld_locks[i].handle = NULL; 242 } 243 244 memcpy(&lockinfo, &deflockinfo, sizeof(lockinfo)); 245 _rtld_thread_init(NULL); 246 /* 247 * Construct a mask to block all signals except traps which might 248 * conceivably be generated within the dynamic linker itself. 249 */ 250 sigfillset(&fullsigmask); 251 sigdelset(&fullsigmask, SIGILL); 252 sigdelset(&fullsigmask, SIGTRAP); 253 sigdelset(&fullsigmask, SIGABRT); 254 sigdelset(&fullsigmask, SIGEMT); 255 sigdelset(&fullsigmask, SIGFPE); 256 sigdelset(&fullsigmask, SIGBUS); 257 sigdelset(&fullsigmask, SIGSEGV); 258 sigdelset(&fullsigmask, SIGSYS); 259 } 260 261 /* 262 * Callback function to allow threads implementation to 263 * register their own locking primitives if the default 264 * one is not suitable. 265 * The current context should be the only context 266 * executing at the invocation time. 267 */ 268 void 269 _rtld_thread_init(struct RtldLockInfo *pli) 270 { 271 int flags, i; 272 void *locks[RTLD_LOCK_CNT]; 273 274 /* disable all locking while this function is running */ 275 flags = thread_mask_set(~0); 276 277 if (pli == NULL) 278 pli = &deflockinfo; 279 280 281 for (i = 0; i < RTLD_LOCK_CNT; i++) 282 if ((locks[i] = pli->lock_create()) == NULL) 283 break; 284 285 if (i < RTLD_LOCK_CNT) { 286 while (--i >= 0) 287 pli->lock_destroy(locks[i]); 288 abort(); 289 } 290 291 for (i = 0; i < RTLD_LOCK_CNT; i++) { 292 if (rtld_locks[i].handle == NULL) 293 continue; 294 if (flags & rtld_locks[i].mask) 295 lockinfo.lock_release(rtld_locks[i].handle); 296 lockinfo.lock_destroy(rtld_locks[i].handle); 297 } 298 299 for (i = 0; i < RTLD_LOCK_CNT; i++) { 300 rtld_locks[i].handle = locks[i]; 301 if (flags & rtld_locks[i].mask) 302 pli->wlock_acquire(rtld_locks[i].handle); 303 } 304 305 lockinfo.lock_create = pli->lock_create; 306 lockinfo.lock_destroy = pli->lock_destroy; 307 lockinfo.rlock_acquire = pli->rlock_acquire; 308 lockinfo.wlock_acquire = pli->wlock_acquire; 309 lockinfo.lock_release = pli->lock_release; 310 lockinfo.thread_set_flag = pli->thread_set_flag; 311 lockinfo.thread_clr_flag = pli->thread_clr_flag; 312 lockinfo.at_fork = pli->at_fork; 313 314 /* restore thread locking state, this time with new locks */ 315 thread_mask_clear(~0); 316 thread_mask_set(flags); 317 dbg("_rtld_thread_init: done"); 318 } 319 320 void 321 _rtld_atfork_pre(int *locks) 322 { 323 324 locks[2] = wlock_acquire(rtld_phdr_lock); 325 locks[0] = rlock_acquire(rtld_bind_lock); 326 } 327 328 void 329 _rtld_atfork_post(int *locks) 330 { 331 332 rlock_release(rtld_bind_lock, locks[0]); 333 wlock_release(rtld_phdr_lock, locks[2]); 334 } 335