Commit | Line | Data |
---|---|---|
b27f8e75 MD |
1 | /* |
2 | * Copyright (C) 2009 Pierre-Marc Fournier | |
1622ba22 | 3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
c39c72ee PMF |
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 | |
6d4658aa | 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
c39c72ee PMF |
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 | ||
e541a28d | 20 | #define _GNU_SOURCE |
f02baefb | 21 | #include <lttng/ust-dlfcn.h> |
e541a28d PMF |
22 | #include <sys/types.h> |
23 | #include <stdio.h> | |
2594a5b4 | 24 | #include <assert.h> |
4c3536e0 MD |
25 | #include <urcu/system.h> |
26 | #include <urcu/uatomic.h> | |
2594a5b4 | 27 | #include <urcu/compiler.h> |
8c06ba6f | 28 | #include <urcu/tls-compat.h> |
20ef5166 | 29 | #include <urcu/arch.h> |
2594a5b4 | 30 | #include <lttng/align.h> |
1de7abff | 31 | #include <helper.h> |
1622ba22 MD |
32 | |
33 | #define TRACEPOINT_DEFINE | |
34 | #define TRACEPOINT_CREATE_PROBES | |
52c95399 | 35 | #define TP_IP_PARAM ip |
1622ba22 | 36 | #include "ust_libc.h" |
fbd8191b | 37 | |
f95b2888 SS |
38 | #define STATIC_CALLOC_LEN 4096 |
39 | static char static_calloc_buf[STATIC_CALLOC_LEN]; | |
4c3536e0 | 40 | static unsigned long static_calloc_buf_offset; |
f95b2888 | 41 | |
2594a5b4 MD |
42 | struct alloc_functions { |
43 | void *(*calloc)(size_t nmemb, size_t size); | |
44 | void *(*malloc)(size_t size); | |
45 | void (*free)(void *ptr); | |
46 | void *(*realloc)(void *ptr, size_t size); | |
47 | void *(*memalign)(size_t alignment, size_t size); | |
48 | int (*posix_memalign)(void **memptr, size_t alignment, size_t size); | |
49 | }; | |
50 | ||
51 | static | |
52 | struct alloc_functions cur_alloc; | |
53 | ||
8c06ba6f MD |
54 | /* |
55 | * Make sure our own use of the LTS compat layer will not cause infinite | |
56 | * recursion by calling calloc. | |
57 | */ | |
58 | ||
59 | static | |
60 | void *static_calloc(size_t nmemb, size_t size); | |
61 | ||
20ef5166 MD |
62 | /* |
63 | * pthread mutex replacement for URCU tls compat layer. | |
64 | */ | |
65 | static int ust_malloc_lock; | |
66 | ||
67 | static __attribute__((unused)) | |
68 | void ust_malloc_spin_lock(pthread_mutex_t *lock) | |
69 | { | |
70 | /* | |
71 | * The memory barrier within cmpxchg takes care of ordering | |
72 | * memory accesses with respect to the start of the critical | |
73 | * section. | |
74 | */ | |
75 | while (uatomic_cmpxchg(&ust_malloc_lock, 0, 1) != 0) | |
76 | caa_cpu_relax(); | |
77 | } | |
78 | ||
79 | static __attribute__((unused)) | |
80 | void ust_malloc_spin_unlock(pthread_mutex_t *lock) | |
81 | { | |
82 | /* | |
83 | * Ensure memory accesses within the critical section do not | |
84 | * leak outside. | |
85 | */ | |
86 | cmm_smp_mb(); | |
87 | uatomic_set(&ust_malloc_lock, 0); | |
88 | } | |
89 | ||
8c06ba6f | 90 | #define calloc static_calloc |
20ef5166 MD |
91 | #define pthread_mutex_lock ust_malloc_spin_lock |
92 | #define pthread_mutex_unlock ust_malloc_spin_unlock | |
8c06ba6f | 93 | static DEFINE_URCU_TLS(int, malloc_nesting); |
20ef5166 MD |
94 | #undef ust_malloc_spin_unlock |
95 | #undef ust_malloc_spin_lock | |
8c06ba6f MD |
96 | #undef calloc |
97 | ||
2594a5b4 MD |
98 | /* |
99 | * Static allocator to use when initially executing dlsym(). It keeps a | |
100 | * size_t value of each object size prior to the object. | |
101 | */ | |
102 | static | |
103 | void *static_calloc_aligned(size_t nmemb, size_t size, size_t alignment) | |
f95b2888 | 104 | { |
2594a5b4 MD |
105 | size_t prev_offset, new_offset, res_offset, aligned_offset; |
106 | ||
107 | if (nmemb * size == 0) { | |
108 | return NULL; | |
109 | } | |
f95b2888 | 110 | |
4c3536e0 MD |
111 | /* |
112 | * Protect static_calloc_buf_offset from concurrent updates | |
113 | * using a cmpxchg loop rather than a mutex to remove a | |
114 | * dependency on pthread. This will minimize the risk of bad | |
115 | * interaction between mutex and malloc instrumentation. | |
116 | */ | |
117 | res_offset = CMM_LOAD_SHARED(static_calloc_buf_offset); | |
118 | do { | |
119 | prev_offset = res_offset; | |
2594a5b4 MD |
120 | aligned_offset = ALIGN(prev_offset + sizeof(size_t), alignment); |
121 | new_offset = aligned_offset + nmemb * size; | |
122 | if (new_offset > sizeof(static_calloc_buf)) { | |
123 | abort(); | |
4c3536e0 | 124 | } |
4c3536e0 MD |
125 | } while ((res_offset = uatomic_cmpxchg(&static_calloc_buf_offset, |
126 | prev_offset, new_offset)) != prev_offset); | |
2594a5b4 MD |
127 | *(size_t *) &static_calloc_buf[aligned_offset - sizeof(size_t)] = size; |
128 | return &static_calloc_buf[aligned_offset]; | |
129 | } | |
130 | ||
131 | static | |
132 | void *static_calloc(size_t nmemb, size_t size) | |
133 | { | |
134 | void *retval; | |
135 | ||
136 | retval = static_calloc_aligned(nmemb, size, 1); | |
2594a5b4 MD |
137 | return retval; |
138 | } | |
139 | ||
140 | static | |
141 | void *static_malloc(size_t size) | |
142 | { | |
143 | void *retval; | |
144 | ||
145 | retval = static_calloc_aligned(1, size, 1); | |
2594a5b4 MD |
146 | return retval; |
147 | } | |
148 | ||
149 | static | |
150 | void static_free(void *ptr) | |
151 | { | |
152 | /* no-op. */ | |
2594a5b4 MD |
153 | } |
154 | ||
155 | static | |
156 | void *static_realloc(void *ptr, size_t size) | |
157 | { | |
158 | size_t *old_size = NULL; | |
159 | void *retval; | |
160 | ||
161 | if (size == 0) { | |
162 | retval = NULL; | |
163 | goto end; | |
164 | } | |
165 | ||
166 | if (ptr) { | |
167 | old_size = (size_t *) ptr - 1; | |
168 | if (size <= *old_size) { | |
169 | /* We can re-use the old entry. */ | |
170 | *old_size = size; | |
171 | retval = ptr; | |
172 | goto end; | |
173 | } | |
174 | } | |
175 | /* We need to expand. Don't free previous memory location. */ | |
176 | retval = static_calloc_aligned(1, size, 1); | |
177 | assert(retval); | |
178 | if (ptr) | |
179 | memcpy(retval, ptr, *old_size); | |
180 | end: | |
2594a5b4 MD |
181 | return retval; |
182 | } | |
183 | ||
184 | static | |
185 | void *static_memalign(size_t alignment, size_t size) | |
186 | { | |
187 | void *retval; | |
188 | ||
189 | retval = static_calloc_aligned(1, size, alignment); | |
2594a5b4 MD |
190 | return retval; |
191 | } | |
192 | ||
193 | static | |
194 | int static_posix_memalign(void **memptr, size_t alignment, size_t size) | |
195 | { | |
2594a5b4 MD |
196 | void *ptr; |
197 | ||
198 | /* Check for power of 2, larger than void *. */ | |
199 | if (alignment & (alignment - 1) | |
200 | || alignment < sizeof(void *) | |
201 | || alignment == 0) { | |
2594a5b4 MD |
202 | goto end; |
203 | } | |
204 | ptr = static_calloc_aligned(1, size, alignment); | |
205 | *memptr = ptr; | |
2594a5b4 | 206 | end: |
2594a5b4 MD |
207 | return 0; |
208 | } | |
209 | ||
210 | static | |
211 | void setup_static_allocator(void) | |
212 | { | |
213 | assert(cur_alloc.calloc == NULL); | |
214 | cur_alloc.calloc = static_calloc; | |
215 | assert(cur_alloc.malloc == NULL); | |
216 | cur_alloc.malloc = static_malloc; | |
217 | assert(cur_alloc.free == NULL); | |
218 | cur_alloc.free = static_free; | |
219 | assert(cur_alloc.realloc == NULL); | |
220 | cur_alloc.realloc = static_realloc; | |
221 | assert(cur_alloc.memalign == NULL); | |
222 | cur_alloc.memalign = static_memalign; | |
223 | assert(cur_alloc.posix_memalign == NULL); | |
224 | cur_alloc.posix_memalign = static_posix_memalign; | |
225 | } | |
226 | ||
227 | static | |
228 | void lookup_all_symbols(void) | |
229 | { | |
230 | struct alloc_functions af; | |
231 | ||
232 | /* | |
233 | * Temporarily redirect allocation functions to | |
234 | * static_calloc_aligned, and free function to static_free | |
235 | * (no-op), until the dlsym lookup has completed. | |
236 | */ | |
237 | setup_static_allocator(); | |
238 | ||
239 | /* Perform the actual lookups */ | |
240 | af.calloc = dlsym(RTLD_NEXT, "calloc"); | |
241 | af.malloc = dlsym(RTLD_NEXT, "malloc"); | |
242 | af.free = dlsym(RTLD_NEXT, "free"); | |
243 | af.realloc = dlsym(RTLD_NEXT, "realloc"); | |
244 | af.memalign = dlsym(RTLD_NEXT, "memalign"); | |
245 | af.posix_memalign = dlsym(RTLD_NEXT, "posix_memalign"); | |
246 | ||
247 | /* Populate the new allocator functions */ | |
248 | memcpy(&cur_alloc, &af, sizeof(cur_alloc)); | |
f95b2888 SS |
249 | } |
250 | ||
e541a28d PMF |
251 | void *malloc(size_t size) |
252 | { | |
1c184644 PMF |
253 | void *retval; |
254 | ||
8c06ba6f | 255 | URCU_TLS(malloc_nesting)++; |
2594a5b4 MD |
256 | if (cur_alloc.malloc == NULL) { |
257 | lookup_all_symbols(); | |
258 | if (cur_alloc.malloc == NULL) { | |
e541a28d | 259 | fprintf(stderr, "mallocwrap: unable to find malloc\n"); |
2594a5b4 | 260 | abort(); |
e541a28d PMF |
261 | } |
262 | } | |
2594a5b4 | 263 | retval = cur_alloc.malloc(size); |
8c06ba6f | 264 | if (URCU_TLS(malloc_nesting) == 1) { |
6d4658aa | 265 | tracepoint(lttng_ust_libc, malloc, |
1de7abff | 266 | size, retval, LTTNG_UST_CALLER_IP()); |
8c06ba6f MD |
267 | } |
268 | URCU_TLS(malloc_nesting)--; | |
1c184644 PMF |
269 | return retval; |
270 | } | |
271 | ||
272 | void free(void *ptr) | |
273 | { | |
8c06ba6f | 274 | URCU_TLS(malloc_nesting)++; |
2594a5b4 MD |
275 | /* |
276 | * Check whether the memory was allocated with | |
277 | * static_calloc_align, in which case there is nothing to free. | |
f95b2888 | 278 | */ |
2594a5b4 MD |
279 | if (caa_unlikely((char *)ptr >= static_calloc_buf && |
280 | (char *)ptr < static_calloc_buf + STATIC_CALLOC_LEN)) { | |
8c06ba6f MD |
281 | goto end; |
282 | } | |
283 | ||
284 | if (URCU_TLS(malloc_nesting) == 1) { | |
6d4658aa | 285 | tracepoint(lttng_ust_libc, free, |
1de7abff | 286 | ptr, LTTNG_UST_CALLER_IP()); |
f95b2888 | 287 | } |
1c184644 | 288 | |
2594a5b4 MD |
289 | if (cur_alloc.free == NULL) { |
290 | lookup_all_symbols(); | |
291 | if (cur_alloc.free == NULL) { | |
1c184644 | 292 | fprintf(stderr, "mallocwrap: unable to find free\n"); |
2594a5b4 | 293 | abort(); |
1c184644 PMF |
294 | } |
295 | } | |
2594a5b4 | 296 | cur_alloc.free(ptr); |
8c06ba6f MD |
297 | end: |
298 | URCU_TLS(malloc_nesting)--; | |
e541a28d | 299 | } |
f95b2888 SS |
300 | |
301 | void *calloc(size_t nmemb, size_t size) | |
302 | { | |
f95b2888 SS |
303 | void *retval; |
304 | ||
8c06ba6f | 305 | URCU_TLS(malloc_nesting)++; |
2594a5b4 MD |
306 | if (cur_alloc.calloc == NULL) { |
307 | lookup_all_symbols(); | |
308 | if (cur_alloc.calloc == NULL) { | |
f95b2888 | 309 | fprintf(stderr, "callocwrap: unable to find calloc\n"); |
2594a5b4 | 310 | abort(); |
f95b2888 SS |
311 | } |
312 | } | |
2594a5b4 | 313 | retval = cur_alloc.calloc(nmemb, size); |
8c06ba6f | 314 | if (URCU_TLS(malloc_nesting) == 1) { |
6d4658aa | 315 | tracepoint(lttng_ust_libc, calloc, |
1de7abff | 316 | nmemb, size, retval, LTTNG_UST_CALLER_IP()); |
8c06ba6f MD |
317 | } |
318 | URCU_TLS(malloc_nesting)--; | |
f95b2888 SS |
319 | return retval; |
320 | } | |
321 | ||
322 | void *realloc(void *ptr, size_t size) | |
323 | { | |
f95b2888 SS |
324 | void *retval; |
325 | ||
8c06ba6f MD |
326 | URCU_TLS(malloc_nesting)++; |
327 | /* | |
328 | * Check whether the memory was allocated with | |
2594a5b4 MD |
329 | * static_calloc_align, in which case there is nothing |
330 | * to free, and we need to copy the old data. | |
331 | */ | |
332 | if (caa_unlikely((char *)ptr >= static_calloc_buf && | |
333 | (char *)ptr < static_calloc_buf + STATIC_CALLOC_LEN)) { | |
334 | size_t *old_size; | |
335 | ||
336 | old_size = (size_t *) ptr - 1; | |
337 | if (cur_alloc.calloc == NULL) { | |
338 | lookup_all_symbols(); | |
339 | if (cur_alloc.calloc == NULL) { | |
340 | fprintf(stderr, "reallocwrap: unable to find calloc\n"); | |
341 | abort(); | |
342 | } | |
343 | } | |
344 | retval = cur_alloc.calloc(1, size); | |
345 | if (retval) { | |
346 | memcpy(retval, ptr, *old_size); | |
347 | } | |
8c06ba6f MD |
348 | /* |
349 | * Mimick that a NULL pointer has been received, so | |
350 | * memory allocation analysis based on the trace don't | |
351 | * get confused by the address from the static | |
352 | * allocator. | |
353 | */ | |
354 | ptr = NULL; | |
2594a5b4 MD |
355 | goto end; |
356 | } | |
357 | ||
358 | if (cur_alloc.realloc == NULL) { | |
359 | lookup_all_symbols(); | |
360 | if (cur_alloc.realloc == NULL) { | |
f95b2888 | 361 | fprintf(stderr, "reallocwrap: unable to find realloc\n"); |
2594a5b4 | 362 | abort(); |
f95b2888 SS |
363 | } |
364 | } | |
2594a5b4 MD |
365 | retval = cur_alloc.realloc(ptr, size); |
366 | end: | |
8c06ba6f | 367 | if (URCU_TLS(malloc_nesting) == 1) { |
6d4658aa | 368 | tracepoint(lttng_ust_libc, realloc, |
1de7abff | 369 | ptr, size, retval, LTTNG_UST_CALLER_IP()); |
8c06ba6f MD |
370 | } |
371 | URCU_TLS(malloc_nesting)--; | |
f95b2888 SS |
372 | return retval; |
373 | } | |
9d34b226 SS |
374 | |
375 | void *memalign(size_t alignment, size_t size) | |
376 | { | |
9d34b226 SS |
377 | void *retval; |
378 | ||
8c06ba6f | 379 | URCU_TLS(malloc_nesting)++; |
2594a5b4 MD |
380 | if (cur_alloc.memalign == NULL) { |
381 | lookup_all_symbols(); | |
382 | if (cur_alloc.memalign == NULL) { | |
9d34b226 | 383 | fprintf(stderr, "memalignwrap: unable to find memalign\n"); |
2594a5b4 | 384 | abort(); |
9d34b226 SS |
385 | } |
386 | } | |
2594a5b4 | 387 | retval = cur_alloc.memalign(alignment, size); |
8c06ba6f | 388 | if (URCU_TLS(malloc_nesting) == 1) { |
6d4658aa AB |
389 | tracepoint(lttng_ust_libc, memalign, |
390 | alignment, size, retval, | |
1de7abff | 391 | LTTNG_UST_CALLER_IP()); |
8c06ba6f MD |
392 | } |
393 | URCU_TLS(malloc_nesting)--; | |
9d34b226 SS |
394 | return retval; |
395 | } | |
396 | ||
397 | int posix_memalign(void **memptr, size_t alignment, size_t size) | |
398 | { | |
9d34b226 SS |
399 | int retval; |
400 | ||
8c06ba6f | 401 | URCU_TLS(malloc_nesting)++; |
2594a5b4 MD |
402 | if (cur_alloc.posix_memalign == NULL) { |
403 | lookup_all_symbols(); | |
404 | if (cur_alloc.posix_memalign == NULL) { | |
9d34b226 | 405 | fprintf(stderr, "posix_memalignwrap: unable to find posix_memalign\n"); |
2594a5b4 | 406 | abort(); |
9d34b226 SS |
407 | } |
408 | } | |
2594a5b4 | 409 | retval = cur_alloc.posix_memalign(memptr, alignment, size); |
8c06ba6f | 410 | if (URCU_TLS(malloc_nesting) == 1) { |
6d4658aa AB |
411 | tracepoint(lttng_ust_libc, posix_memalign, |
412 | *memptr, alignment, size, | |
1de7abff | 413 | retval, LTTNG_UST_CALLER_IP()); |
8c06ba6f MD |
414 | } |
415 | URCU_TLS(malloc_nesting)--; | |
9d34b226 SS |
416 | return retval; |
417 | } | |
2594a5b4 | 418 | |
85b2b5d8 MD |
419 | static |
420 | void lttng_ust_fixup_malloc_nesting_tls(void) | |
421 | { | |
422 | asm volatile ("" : : "m" (URCU_TLS(malloc_nesting))); | |
423 | } | |
424 | ||
2594a5b4 MD |
425 | __attribute__((constructor)) |
426 | void lttng_ust_malloc_wrapper_init(void) | |
427 | { | |
428 | /* Initialization already done */ | |
429 | if (cur_alloc.calloc) { | |
430 | return; | |
431 | } | |
85b2b5d8 | 432 | lttng_ust_fixup_malloc_nesting_tls(); |
2594a5b4 MD |
433 | /* |
434 | * Ensure the allocator is in place before the process becomes | |
435 | * multithreaded. | |
436 | */ | |
437 | lookup_all_symbols(); | |
438 | } |