| 1 | /* |
| 2 | * Copyright (C) 2009 Pierre-Marc Fournier |
| 3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 4 | * |
| 5 | * This library is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU Lesser General Public |
| 7 | * License as published by the Free Software Foundation; either |
| 8 | * version 2.1 of the License, or (at your option) any later version. |
| 9 | * |
| 10 | * This library is distributed in the hope that it will be useful, |
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | * Lesser General Public License for more details. |
| 14 | * |
| 15 | * You should have received a copy of the GNU Lesser General Public |
| 16 | * License along with this library; if not, write to the Free Software |
| 17 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 18 | */ |
| 19 | |
| 20 | #define _GNU_SOURCE |
| 21 | /* |
| 22 | * Do _not_ define _LGPL_SOURCE because we don't want to create a |
| 23 | * circular dependency loop between this malloc wrapper, liburcu and |
| 24 | * libc. |
| 25 | */ |
| 26 | #include <lttng/ust-dlfcn.h> |
| 27 | #include <sys/types.h> |
| 28 | #include <stdio.h> |
| 29 | #include <assert.h> |
| 30 | #include <urcu/system.h> |
| 31 | #include <urcu/uatomic.h> |
| 32 | #include <urcu/compiler.h> |
| 33 | #include <urcu/tls-compat.h> |
| 34 | #include <urcu/arch.h> |
| 35 | #include <lttng/align.h> |
| 36 | #include <helper.h> |
| 37 | |
| 38 | #define TRACEPOINT_DEFINE |
| 39 | #define TRACEPOINT_CREATE_PROBES |
| 40 | #define TP_IP_PARAM ip |
| 41 | #include "ust_libc.h" |
| 42 | |
| 43 | #define STATIC_CALLOC_LEN 4096 |
| 44 | static char static_calloc_buf[STATIC_CALLOC_LEN]; |
| 45 | static unsigned long static_calloc_buf_offset; |
| 46 | |
| 47 | struct alloc_functions { |
| 48 | void *(*calloc)(size_t nmemb, size_t size); |
| 49 | void *(*malloc)(size_t size); |
| 50 | void (*free)(void *ptr); |
| 51 | void *(*realloc)(void *ptr, size_t size); |
| 52 | void *(*memalign)(size_t alignment, size_t size); |
| 53 | int (*posix_memalign)(void **memptr, size_t alignment, size_t size); |
| 54 | }; |
| 55 | |
| 56 | static |
| 57 | struct alloc_functions cur_alloc; |
| 58 | |
| 59 | /* |
| 60 | * Make sure our own use of the LTS compat layer will not cause infinite |
| 61 | * recursion by calling calloc. |
| 62 | */ |
| 63 | |
| 64 | static |
| 65 | void *static_calloc(size_t nmemb, size_t size); |
| 66 | |
| 67 | /* |
| 68 | * pthread mutex replacement for URCU tls compat layer. |
| 69 | */ |
| 70 | static int ust_malloc_lock; |
| 71 | |
| 72 | static __attribute__((unused)) |
| 73 | void ust_malloc_spin_lock(pthread_mutex_t *lock) |
| 74 | { |
| 75 | /* |
| 76 | * The memory barrier within cmpxchg takes care of ordering |
| 77 | * memory accesses with respect to the start of the critical |
| 78 | * section. |
| 79 | */ |
| 80 | while (uatomic_cmpxchg(&ust_malloc_lock, 0, 1) != 0) |
| 81 | caa_cpu_relax(); |
| 82 | } |
| 83 | |
| 84 | static __attribute__((unused)) |
| 85 | void ust_malloc_spin_unlock(pthread_mutex_t *lock) |
| 86 | { |
| 87 | /* |
| 88 | * Ensure memory accesses within the critical section do not |
| 89 | * leak outside. |
| 90 | */ |
| 91 | cmm_smp_mb(); |
| 92 | uatomic_set(&ust_malloc_lock, 0); |
| 93 | } |
| 94 | |
| 95 | #define calloc static_calloc |
| 96 | #define pthread_mutex_lock ust_malloc_spin_lock |
| 97 | #define pthread_mutex_unlock ust_malloc_spin_unlock |
| 98 | static DEFINE_URCU_TLS(int, malloc_nesting); |
| 99 | #undef ust_malloc_spin_unlock |
| 100 | #undef ust_malloc_spin_lock |
| 101 | #undef calloc |
| 102 | |
| 103 | /* |
| 104 | * Static allocator to use when initially executing dlsym(). It keeps a |
| 105 | * size_t value of each object size prior to the object. |
| 106 | */ |
| 107 | static |
| 108 | void *static_calloc_aligned(size_t nmemb, size_t size, size_t alignment) |
| 109 | { |
| 110 | size_t prev_offset, new_offset, res_offset, aligned_offset; |
| 111 | |
| 112 | if (nmemb * size == 0) { |
| 113 | return NULL; |
| 114 | } |
| 115 | |
| 116 | /* |
| 117 | * Protect static_calloc_buf_offset from concurrent updates |
| 118 | * using a cmpxchg loop rather than a mutex to remove a |
| 119 | * dependency on pthread. This will minimize the risk of bad |
| 120 | * interaction between mutex and malloc instrumentation. |
| 121 | */ |
| 122 | res_offset = CMM_LOAD_SHARED(static_calloc_buf_offset); |
| 123 | do { |
| 124 | prev_offset = res_offset; |
| 125 | aligned_offset = ALIGN(prev_offset + sizeof(size_t), alignment); |
| 126 | new_offset = aligned_offset + nmemb * size; |
| 127 | if (new_offset > sizeof(static_calloc_buf)) { |
| 128 | abort(); |
| 129 | } |
| 130 | } while ((res_offset = uatomic_cmpxchg(&static_calloc_buf_offset, |
| 131 | prev_offset, new_offset)) != prev_offset); |
| 132 | *(size_t *) &static_calloc_buf[aligned_offset - sizeof(size_t)] = size; |
| 133 | return &static_calloc_buf[aligned_offset]; |
| 134 | } |
| 135 | |
| 136 | static |
| 137 | void *static_calloc(size_t nmemb, size_t size) |
| 138 | { |
| 139 | void *retval; |
| 140 | |
| 141 | retval = static_calloc_aligned(nmemb, size, 1); |
| 142 | return retval; |
| 143 | } |
| 144 | |
| 145 | static |
| 146 | void *static_malloc(size_t size) |
| 147 | { |
| 148 | void *retval; |
| 149 | |
| 150 | retval = static_calloc_aligned(1, size, 1); |
| 151 | return retval; |
| 152 | } |
| 153 | |
| 154 | static |
| 155 | void static_free(void *ptr) |
| 156 | { |
| 157 | /* no-op. */ |
| 158 | } |
| 159 | |
| 160 | static |
| 161 | void *static_realloc(void *ptr, size_t size) |
| 162 | { |
| 163 | size_t *old_size = NULL; |
| 164 | void *retval; |
| 165 | |
| 166 | if (size == 0) { |
| 167 | retval = NULL; |
| 168 | goto end; |
| 169 | } |
| 170 | |
| 171 | if (ptr) { |
| 172 | old_size = (size_t *) ptr - 1; |
| 173 | if (size <= *old_size) { |
| 174 | /* We can re-use the old entry. */ |
| 175 | *old_size = size; |
| 176 | retval = ptr; |
| 177 | goto end; |
| 178 | } |
| 179 | } |
| 180 | /* We need to expand. Don't free previous memory location. */ |
| 181 | retval = static_calloc_aligned(1, size, 1); |
| 182 | assert(retval); |
| 183 | if (ptr) |
| 184 | memcpy(retval, ptr, *old_size); |
| 185 | end: |
| 186 | return retval; |
| 187 | } |
| 188 | |
| 189 | static |
| 190 | void *static_memalign(size_t alignment, size_t size) |
| 191 | { |
| 192 | void *retval; |
| 193 | |
| 194 | retval = static_calloc_aligned(1, size, alignment); |
| 195 | return retval; |
| 196 | } |
| 197 | |
| 198 | static |
| 199 | int static_posix_memalign(void **memptr, size_t alignment, size_t size) |
| 200 | { |
| 201 | void *ptr; |
| 202 | |
| 203 | /* Check for power of 2, larger than void *. */ |
| 204 | if (alignment & (alignment - 1) |
| 205 | || alignment < sizeof(void *) |
| 206 | || alignment == 0) { |
| 207 | goto end; |
| 208 | } |
| 209 | ptr = static_calloc_aligned(1, size, alignment); |
| 210 | *memptr = ptr; |
| 211 | end: |
| 212 | return 0; |
| 213 | } |
| 214 | |
| 215 | static |
| 216 | void setup_static_allocator(void) |
| 217 | { |
| 218 | assert(cur_alloc.calloc == NULL); |
| 219 | cur_alloc.calloc = static_calloc; |
| 220 | assert(cur_alloc.malloc == NULL); |
| 221 | cur_alloc.malloc = static_malloc; |
| 222 | assert(cur_alloc.free == NULL); |
| 223 | cur_alloc.free = static_free; |
| 224 | assert(cur_alloc.realloc == NULL); |
| 225 | cur_alloc.realloc = static_realloc; |
| 226 | assert(cur_alloc.memalign == NULL); |
| 227 | cur_alloc.memalign = static_memalign; |
| 228 | assert(cur_alloc.posix_memalign == NULL); |
| 229 | cur_alloc.posix_memalign = static_posix_memalign; |
| 230 | } |
| 231 | |
| 232 | static |
| 233 | void lookup_all_symbols(void) |
| 234 | { |
| 235 | struct alloc_functions af; |
| 236 | |
| 237 | /* |
| 238 | * Temporarily redirect allocation functions to |
| 239 | * static_calloc_aligned, and free function to static_free |
| 240 | * (no-op), until the dlsym lookup has completed. |
| 241 | */ |
| 242 | setup_static_allocator(); |
| 243 | |
| 244 | /* Perform the actual lookups */ |
| 245 | af.calloc = dlsym(RTLD_NEXT, "calloc"); |
| 246 | af.malloc = dlsym(RTLD_NEXT, "malloc"); |
| 247 | af.free = dlsym(RTLD_NEXT, "free"); |
| 248 | af.realloc = dlsym(RTLD_NEXT, "realloc"); |
| 249 | af.memalign = dlsym(RTLD_NEXT, "memalign"); |
| 250 | af.posix_memalign = dlsym(RTLD_NEXT, "posix_memalign"); |
| 251 | |
| 252 | /* Populate the new allocator functions */ |
| 253 | memcpy(&cur_alloc, &af, sizeof(cur_alloc)); |
| 254 | } |
| 255 | |
| 256 | void *malloc(size_t size) |
| 257 | { |
| 258 | void *retval; |
| 259 | |
| 260 | URCU_TLS(malloc_nesting)++; |
| 261 | if (cur_alloc.malloc == NULL) { |
| 262 | lookup_all_symbols(); |
| 263 | if (cur_alloc.malloc == NULL) { |
| 264 | fprintf(stderr, "mallocwrap: unable to find malloc\n"); |
| 265 | abort(); |
| 266 | } |
| 267 | } |
| 268 | retval = cur_alloc.malloc(size); |
| 269 | if (URCU_TLS(malloc_nesting) == 1) { |
| 270 | tracepoint(lttng_ust_libc, malloc, |
| 271 | size, retval, LTTNG_UST_CALLER_IP()); |
| 272 | } |
| 273 | URCU_TLS(malloc_nesting)--; |
| 274 | return retval; |
| 275 | } |
| 276 | |
| 277 | void free(void *ptr) |
| 278 | { |
| 279 | URCU_TLS(malloc_nesting)++; |
| 280 | /* |
| 281 | * Check whether the memory was allocated with |
| 282 | * static_calloc_align, in which case there is nothing to free. |
| 283 | */ |
| 284 | if (caa_unlikely((char *)ptr >= static_calloc_buf && |
| 285 | (char *)ptr < static_calloc_buf + STATIC_CALLOC_LEN)) { |
| 286 | goto end; |
| 287 | } |
| 288 | |
| 289 | if (URCU_TLS(malloc_nesting) == 1) { |
| 290 | tracepoint(lttng_ust_libc, free, |
| 291 | ptr, LTTNG_UST_CALLER_IP()); |
| 292 | } |
| 293 | |
| 294 | if (cur_alloc.free == NULL) { |
| 295 | lookup_all_symbols(); |
| 296 | if (cur_alloc.free == NULL) { |
| 297 | fprintf(stderr, "mallocwrap: unable to find free\n"); |
| 298 | abort(); |
| 299 | } |
| 300 | } |
| 301 | cur_alloc.free(ptr); |
| 302 | end: |
| 303 | URCU_TLS(malloc_nesting)--; |
| 304 | } |
| 305 | |
| 306 | void *calloc(size_t nmemb, size_t size) |
| 307 | { |
| 308 | void *retval; |
| 309 | |
| 310 | URCU_TLS(malloc_nesting)++; |
| 311 | if (cur_alloc.calloc == NULL) { |
| 312 | lookup_all_symbols(); |
| 313 | if (cur_alloc.calloc == NULL) { |
| 314 | fprintf(stderr, "callocwrap: unable to find calloc\n"); |
| 315 | abort(); |
| 316 | } |
| 317 | } |
| 318 | retval = cur_alloc.calloc(nmemb, size); |
| 319 | if (URCU_TLS(malloc_nesting) == 1) { |
| 320 | tracepoint(lttng_ust_libc, calloc, |
| 321 | nmemb, size, retval, LTTNG_UST_CALLER_IP()); |
| 322 | } |
| 323 | URCU_TLS(malloc_nesting)--; |
| 324 | return retval; |
| 325 | } |
| 326 | |
| 327 | void *realloc(void *ptr, size_t size) |
| 328 | { |
| 329 | void *retval; |
| 330 | |
| 331 | URCU_TLS(malloc_nesting)++; |
| 332 | /* |
| 333 | * Check whether the memory was allocated with |
| 334 | * static_calloc_align, in which case there is nothing |
| 335 | * to free, and we need to copy the old data. |
| 336 | */ |
| 337 | if (caa_unlikely((char *)ptr >= static_calloc_buf && |
| 338 | (char *)ptr < static_calloc_buf + STATIC_CALLOC_LEN)) { |
| 339 | size_t *old_size; |
| 340 | |
| 341 | old_size = (size_t *) ptr - 1; |
| 342 | if (cur_alloc.calloc == NULL) { |
| 343 | lookup_all_symbols(); |
| 344 | if (cur_alloc.calloc == NULL) { |
| 345 | fprintf(stderr, "reallocwrap: unable to find calloc\n"); |
| 346 | abort(); |
| 347 | } |
| 348 | } |
| 349 | retval = cur_alloc.calloc(1, size); |
| 350 | if (retval) { |
| 351 | memcpy(retval, ptr, *old_size); |
| 352 | } |
| 353 | /* |
| 354 | * Mimick that a NULL pointer has been received, so |
| 355 | * memory allocation analysis based on the trace don't |
| 356 | * get confused by the address from the static |
| 357 | * allocator. |
| 358 | */ |
| 359 | ptr = NULL; |
| 360 | goto end; |
| 361 | } |
| 362 | |
| 363 | if (cur_alloc.realloc == NULL) { |
| 364 | lookup_all_symbols(); |
| 365 | if (cur_alloc.realloc == NULL) { |
| 366 | fprintf(stderr, "reallocwrap: unable to find realloc\n"); |
| 367 | abort(); |
| 368 | } |
| 369 | } |
| 370 | retval = cur_alloc.realloc(ptr, size); |
| 371 | end: |
| 372 | if (URCU_TLS(malloc_nesting) == 1) { |
| 373 | tracepoint(lttng_ust_libc, realloc, |
| 374 | ptr, size, retval, LTTNG_UST_CALLER_IP()); |
| 375 | } |
| 376 | URCU_TLS(malloc_nesting)--; |
| 377 | return retval; |
| 378 | } |
| 379 | |
| 380 | void *memalign(size_t alignment, size_t size) |
| 381 | { |
| 382 | void *retval; |
| 383 | |
| 384 | URCU_TLS(malloc_nesting)++; |
| 385 | if (cur_alloc.memalign == NULL) { |
| 386 | lookup_all_symbols(); |
| 387 | if (cur_alloc.memalign == NULL) { |
| 388 | fprintf(stderr, "memalignwrap: unable to find memalign\n"); |
| 389 | abort(); |
| 390 | } |
| 391 | } |
| 392 | retval = cur_alloc.memalign(alignment, size); |
| 393 | if (URCU_TLS(malloc_nesting) == 1) { |
| 394 | tracepoint(lttng_ust_libc, memalign, |
| 395 | alignment, size, retval, |
| 396 | LTTNG_UST_CALLER_IP()); |
| 397 | } |
| 398 | URCU_TLS(malloc_nesting)--; |
| 399 | return retval; |
| 400 | } |
| 401 | |
| 402 | int posix_memalign(void **memptr, size_t alignment, size_t size) |
| 403 | { |
| 404 | int retval; |
| 405 | |
| 406 | URCU_TLS(malloc_nesting)++; |
| 407 | if (cur_alloc.posix_memalign == NULL) { |
| 408 | lookup_all_symbols(); |
| 409 | if (cur_alloc.posix_memalign == NULL) { |
| 410 | fprintf(stderr, "posix_memalignwrap: unable to find posix_memalign\n"); |
| 411 | abort(); |
| 412 | } |
| 413 | } |
| 414 | retval = cur_alloc.posix_memalign(memptr, alignment, size); |
| 415 | if (URCU_TLS(malloc_nesting) == 1) { |
| 416 | tracepoint(lttng_ust_libc, posix_memalign, |
| 417 | *memptr, alignment, size, |
| 418 | retval, LTTNG_UST_CALLER_IP()); |
| 419 | } |
| 420 | URCU_TLS(malloc_nesting)--; |
| 421 | return retval; |
| 422 | } |
| 423 | |
| 424 | static |
| 425 | void lttng_ust_fixup_malloc_nesting_tls(void) |
| 426 | { |
| 427 | asm volatile ("" : : "m" (URCU_TLS(malloc_nesting))); |
| 428 | } |
| 429 | |
| 430 | __attribute__((constructor)) |
| 431 | void lttng_ust_malloc_wrapper_init(void) |
| 432 | { |
| 433 | /* Initialization already done */ |
| 434 | if (cur_alloc.calloc) { |
| 435 | return; |
| 436 | } |
| 437 | lttng_ust_fixup_malloc_nesting_tls(); |
| 438 | /* |
| 439 | * Ensure the allocator is in place before the process becomes |
| 440 | * multithreaded. |
| 441 | */ |
| 442 | lookup_all_symbols(); |
| 443 | } |