1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
|
//===-- tsan_fd.cpp -------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file is a part of ThreadSanitizer (TSan), a race detector.
//
//===----------------------------------------------------------------------===//
#include "tsan_fd.h"
#include <sanitizer_common/sanitizer_atomic.h>
#include "tsan_interceptors.h"
#include "tsan_rtl.h"
namespace __tsan {
const int kTableSizeL1 = 1024;
const int kTableSizeL2 = 1024;
const int kTableSize = kTableSizeL1 * kTableSizeL2;
struct FdSync {
atomic_uint64_t rc;
};
struct FdDesc {
FdSync *sync;
Tid creation_tid;
StackID creation_stack;
};
struct FdContext {
atomic_uintptr_t tab[kTableSizeL1];
// Addresses used for synchronization.
FdSync globsync;
FdSync filesync;
FdSync socksync;
u64 connectsync;
};
static FdContext fdctx;
static bool bogusfd(int fd) {
// Apparently a bogus fd value.
return fd < 0 || fd >= kTableSize;
}
static FdSync *allocsync(ThreadState *thr, uptr pc) {
FdSync *s = (FdSync*)user_alloc_internal(thr, pc, sizeof(FdSync),
kDefaultAlignment, false);
atomic_store(&s->rc, 1, memory_order_relaxed);
return s;
}
static FdSync *ref(FdSync *s) {
if (s && atomic_load(&s->rc, memory_order_relaxed) != (u64)-1)
atomic_fetch_add(&s->rc, 1, memory_order_relaxed);
return s;
}
static void unref(ThreadState *thr, uptr pc, FdSync *s) {
if (s && atomic_load(&s->rc, memory_order_relaxed) != (u64)-1) {
if (atomic_fetch_sub(&s->rc, 1, memory_order_acq_rel) == 1) {
CHECK_NE(s, &fdctx.globsync);
CHECK_NE(s, &fdctx.filesync);
CHECK_NE(s, &fdctx.socksync);
user_free(thr, pc, s, false);
}
}
}
static FdDesc *fddesc(ThreadState *thr, uptr pc, int fd) {
CHECK_GE(fd, 0);
CHECK_LT(fd, kTableSize);
atomic_uintptr_t *pl1 = &fdctx.tab[fd / kTableSizeL2];
uptr l1 = atomic_load(pl1, memory_order_consume);
if (l1 == 0) {
uptr size = kTableSizeL2 * sizeof(FdDesc);
// We need this to reside in user memory to properly catch races on it.
void *p = user_alloc_internal(thr, pc, size, kDefaultAlignment, false);
internal_memset(p, 0, size);
MemoryResetRange(thr, (uptr)&fddesc, (uptr)p, size);
if (atomic_compare_exchange_strong(pl1, &l1, (uptr)p, memory_order_acq_rel))
l1 = (uptr)p;
else
user_free(thr, pc, p, false);
}
FdDesc *fds = reinterpret_cast<FdDesc *>(l1);
return &fds[fd % kTableSizeL2];
}
// pd must be already ref'ed.
static void init(ThreadState *thr, uptr pc, int fd, FdSync *s,
bool write = true) {
FdDesc *d = fddesc(thr, pc, fd);
// As a matter of fact, we don't intercept all close calls.
// See e.g. libc __res_iclose().
if (d->sync) {
unref(thr, pc, d->sync);
d->sync = 0;
}
if (flags()->io_sync == 0) {
unref(thr, pc, s);
} else if (flags()->io_sync == 1) {
d->sync = s;
} else if (flags()->io_sync == 2) {
unref(thr, pc, s);
d->sync = &fdctx.globsync;
}
d->creation_tid = thr->tid;
d->creation_stack = CurrentStackId(thr, pc);
if (write) {
// To catch races between fd usage and open.
MemoryRangeImitateWrite(thr, pc, (uptr)d, 8);
} else {
// See the dup-related comment in FdClose.
MemoryAccess(thr, pc, (uptr)d, 8, kAccessRead);
}
}
void FdInit() {
atomic_store(&fdctx.globsync.rc, (u64)-1, memory_order_relaxed);
atomic_store(&fdctx.filesync.rc, (u64)-1, memory_order_relaxed);
atomic_store(&fdctx.socksync.rc, (u64)-1, memory_order_relaxed);
}
void FdOnFork(ThreadState *thr, uptr pc) {
// On fork() we need to reset all fd's, because the child is going
// close all them, and that will cause races between previous read/write
// and the close.
for (int l1 = 0; l1 < kTableSizeL1; l1++) {
FdDesc *tab = (FdDesc*)atomic_load(&fdctx.tab[l1], memory_order_relaxed);
if (tab == 0)
break;
for (int l2 = 0; l2 < kTableSizeL2; l2++) {
FdDesc *d = &tab[l2];
MemoryResetRange(thr, pc, (uptr)d, 8);
}
}
}
bool FdLocation(uptr addr, int *fd, Tid *tid, StackID *stack) {
for (int l1 = 0; l1 < kTableSizeL1; l1++) {
FdDesc *tab = (FdDesc*)atomic_load(&fdctx.tab[l1], memory_order_relaxed);
if (tab == 0)
break;
if (addr >= (uptr)tab && addr < (uptr)(tab + kTableSizeL2)) {
int l2 = (addr - (uptr)tab) / sizeof(FdDesc);
FdDesc *d = &tab[l2];
*fd = l1 * kTableSizeL1 + l2;
*tid = d->creation_tid;
*stack = d->creation_stack;
return true;
}
}
return false;
}
void FdAcquire(ThreadState *thr, uptr pc, int fd) {
if (bogusfd(fd))
return;
FdDesc *d = fddesc(thr, pc, fd);
FdSync *s = d->sync;
DPrintf("#%d: FdAcquire(%d) -> %p\n", thr->tid, fd, s);
MemoryAccess(thr, pc, (uptr)d, 8, kAccessRead);
if (s)
Acquire(thr, pc, (uptr)s);
}
void FdRelease(ThreadState *thr, uptr pc, int fd) {
if (bogusfd(fd))
return;
FdDesc *d = fddesc(thr, pc, fd);
FdSync *s = d->sync;
DPrintf("#%d: FdRelease(%d) -> %p\n", thr->tid, fd, s);
MemoryAccess(thr, pc, (uptr)d, 8, kAccessRead);
if (s)
Release(thr, pc, (uptr)s);
}
void FdAccess(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdAccess(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
FdDesc *d = fddesc(thr, pc, fd);
MemoryAccess(thr, pc, (uptr)d, 8, kAccessRead);
}
void FdClose(ThreadState *thr, uptr pc, int fd, bool write) {
DPrintf("#%d: FdClose(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
FdDesc *d = fddesc(thr, pc, fd);
if (!MustIgnoreInterceptor(thr)) {
if (write) {
// To catch races between fd usage and close.
MemoryAccess(thr, pc, (uptr)d, 8, kAccessWrite);
} else {
// This path is used only by dup2/dup3 calls.
// We do read instead of write because there is a number of legitimate
// cases where write would lead to false positives:
// 1. Some software dups a closed pipe in place of a socket before closing
// the socket (to prevent races actually).
// 2. Some daemons dup /dev/null in place of stdin/stdout.
// On the other hand we have not seen cases when write here catches real
// bugs.
MemoryAccess(thr, pc, (uptr)d, 8, kAccessRead);
}
}
// We need to clear it, because if we do not intercept any call out there
// that creates fd, we will hit false postives.
MemoryResetRange(thr, pc, (uptr)d, 8);
unref(thr, pc, d->sync);
d->sync = 0;
d->creation_tid = kInvalidTid;
d->creation_stack = kInvalidStackID;
}
void FdFileCreate(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdFileCreate(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
init(thr, pc, fd, &fdctx.filesync);
}
void FdDup(ThreadState *thr, uptr pc, int oldfd, int newfd, bool write) {
DPrintf("#%d: FdDup(%d, %d)\n", thr->tid, oldfd, newfd);
if (bogusfd(oldfd) || bogusfd(newfd))
return;
// Ignore the case when user dups not yet connected socket.
FdDesc *od = fddesc(thr, pc, oldfd);
MemoryAccess(thr, pc, (uptr)od, 8, kAccessRead);
FdClose(thr, pc, newfd, write);
init(thr, pc, newfd, ref(od->sync), write);
}
void FdPipeCreate(ThreadState *thr, uptr pc, int rfd, int wfd) {
DPrintf("#%d: FdCreatePipe(%d, %d)\n", thr->tid, rfd, wfd);
FdSync *s = allocsync(thr, pc);
init(thr, pc, rfd, ref(s));
init(thr, pc, wfd, ref(s));
unref(thr, pc, s);
}
void FdEventCreate(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdEventCreate(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
init(thr, pc, fd, allocsync(thr, pc));
}
void FdSignalCreate(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdSignalCreate(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
init(thr, pc, fd, 0);
}
void FdInotifyCreate(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdInotifyCreate(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
init(thr, pc, fd, 0);
}
void FdPollCreate(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdPollCreate(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
init(thr, pc, fd, allocsync(thr, pc));
}
void FdSocketCreate(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdSocketCreate(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
// It can be a UDP socket.
init(thr, pc, fd, &fdctx.socksync);
}
void FdSocketAccept(ThreadState *thr, uptr pc, int fd, int newfd) {
DPrintf("#%d: FdSocketAccept(%d, %d)\n", thr->tid, fd, newfd);
if (bogusfd(fd))
return;
// Synchronize connect->accept.
Acquire(thr, pc, (uptr)&fdctx.connectsync);
init(thr, pc, newfd, &fdctx.socksync);
}
void FdSocketConnecting(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdSocketConnecting(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
// Synchronize connect->accept.
Release(thr, pc, (uptr)&fdctx.connectsync);
}
void FdSocketConnect(ThreadState *thr, uptr pc, int fd) {
DPrintf("#%d: FdSocketConnect(%d)\n", thr->tid, fd);
if (bogusfd(fd))
return;
init(thr, pc, fd, &fdctx.socksync);
}
uptr File2addr(const char *path) {
(void)path;
static u64 addr;
return (uptr)&addr;
}
uptr Dir2addr(const char *path) {
(void)path;
static u64 addr;
return (uptr)&addr;
}
} // namespace __tsan
|