about summary refs log tree commit diff
path: root/src/thread/pthread_create.c
blob: 7494a486b2b3a4d4b4fee7c05d64538cb51e9eed (plain) (blame)
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
#include "pthread_impl.h"

void __pthread_unwind_next(struct __ptcb *cb)
{
	int i, j, not_finished;
	pthread_t self;

	if (cb->__next) longjmp((void *)cb->__next->__jb, 1);

	self = pthread_self();
	if (self->cancel) self->result = PTHREAD_CANCELLED;

	LOCK(&self->exitlock);

	not_finished = self->tsd_used;
	for (j=0; not_finished && j<PTHREAD_DESTRUCTOR_ITERATIONS; j++) {
		not_finished = 0;
		for (i=0; i<PTHREAD_KEYS_MAX; i++) {
			if (self->tsd[i] && libc.tsd_keys[i]) {
				void *tmp = self->tsd[i];
				self->tsd[i] = 0;
				libc.tsd_keys[i](tmp);
				not_finished = 1;
			}
		}
	}

	/* Mark this thread dead before decrementing count */
	self->dead = 1;

	if (!a_fetch_add(&libc.threads_minus_1, -1))
		exit(0);

	if (self->detached && self->map_base) {
		syscall4(__NR_rt_sigprocmask, SIG_BLOCK, (long)(uint64_t[1]){-1},0,8);
		__unmapself(self->map_base, self->map_size);
	}

	__syscall_exit(0);
}

static void docancel(struct pthread *self)
{
	struct __ptcb cb = { .__next = self->cancelbuf };
	__pthread_unwind_next(&cb);
}

static void cancel_handler(int sig, siginfo_t *si, void *ctx)
{
	struct pthread *self = __pthread_self();
	if (si->si_code > 0 || si->si_pid != self->pid) return;
	self->cancel = 1;
	if (self->canceldisable || (!self->cancelasync && !self->cancelpoint))
		return;
	docancel(self);
}

static void cancelpt(int x)
{
	struct pthread *self = __pthread_self();
	if (self->canceldisable) return;
	self->cancelpoint = x;
	if (self->cancel) docancel(self);
}

/* "rsyscall" is a mechanism by which a thread can synchronously force all
 * other threads to perform an arbitrary syscall. It is necessary to work
 * around the non-conformant implementation of setuid() et al on Linux,
 * which affect only the calling thread and not the whole process. This
 * implementation performs some tricks with signal delivery to work around
 * the fact that it does not keep any list of threads in userspace. */

static struct {
	volatile int lock, hold, blocks, cnt;
	unsigned long arg[6];
	int nr;
	int err;
} rs;

static void rsyscall_handler(int sig, siginfo_t *si, void *ctx)
{
	struct pthread *self = __pthread_self();

	if (si->si_code > 0 || si->si_pid != self->pid ||
		rs.cnt == libc.threads_minus_1) return;

	/* Threads which have already decremented themselves from the
	 * thread count must not increment rs.cnt or otherwise act. */
	if (self->dead) {
		__wait(&rs.hold, 0, 1, 1);
		return;
	}

	if (syscall6(rs.nr, rs.arg[0], rs.arg[1], rs.arg[2],
		rs.arg[3], rs.arg[4], rs.arg[5]) < 0 && !rs.err) rs.err=errno;

	a_inc(&rs.cnt);
	__wake(&rs.cnt, 1, 1);
	while(rs.hold)
		__wait(&rs.hold, 0, 1, 1);
	a_dec(&rs.cnt);
	if (!rs.cnt) __wake(&rs.cnt, 1, 1);
}

static int rsyscall(int nr, long a, long b, long c, long d, long e, long f)
{
	int i, ret;
	sigset_t set = { 0 };
	struct pthread *self = __pthread_self();
	sigaddset(&set, SIGSYSCALL);

	LOCK(&rs.lock);
	while ((i=rs.blocks))
		__wait(&rs.blocks, 0, i, 1);

	__libc_sigprocmask(SIG_BLOCK, &set, 0);

	rs.nr = nr;
	rs.arg[0] = a; rs.arg[1] = b;
	rs.arg[2] = c; rs.arg[3] = d;
	rs.arg[4] = d; rs.arg[5] = f;
	rs.hold = 1;
	rs.err = 0;
	rs.cnt = 0;

	/* Dispatch signals until all threads respond */
	for (i=libc.threads_minus_1; i; i--)
		sigqueue(self->pid, SIGSYSCALL, (union sigval){0});
	while ((i=rs.cnt) < libc.threads_minus_1) {
		sigqueue(self->pid, SIGSYSCALL, (union sigval){0});
		__wait(&rs.cnt, 0, i, 1);
	}

	/* Handle any lingering signals with no-op */
	__libc_sigprocmask(SIG_UNBLOCK, &set, 0);

	/* Resume other threads' signal handlers and wait for them */
	rs.hold = 0;
	__wake(&rs.hold, -1, 0);
	while((i=rs.cnt)) __wait(&rs.cnt, 0, i, 1);

	if (rs.err) errno = rs.err, ret = -1;
	else ret = syscall6(nr, a, b, c, d, e, f);

	UNLOCK(&rs.lock);
	return ret;
}

static void init_threads()
{
	struct sigaction sa = { .sa_flags = SA_SIGINFO | SA_RESTART };
	libc.lock = __lock;
	libc.lockfile = __lockfile;
	libc.cancelpt = cancelpt;
	libc.rsyscall = rsyscall;
	sa.sa_sigaction = cancel_handler;
	__libc_sigaction(SIGCANCEL, &sa, 0);
	sigaddset(&sa.sa_mask, SIGSYSCALL);
	sigaddset(&sa.sa_mask, SIGCANCEL);
	sa.sa_sigaction = rsyscall_handler;
	__libc_sigaction(SIGSYSCALL, &sa, 0);
	sigprocmask(SIG_UNBLOCK, &sa.sa_mask, 0);
}

static int start(void *p)
{
	struct pthread *self = p;
	pthread_exit(self->start(self->start_arg));
	return 0;
}

int __uniclone(void *, int (*)(), void *);

#define ROUND(x) (((x)+PAGE_SIZE-1)&-PAGE_SIZE)

/* pthread_key_create.c overrides this */
static const size_t dummy = 0;
weak_alias(dummy, __pthread_tsd_size);

int pthread_create(pthread_t *res, const pthread_attr_t *attr, void *(*entry)(void *), void *arg)
{
	static int init;
	int ret;
	size_t size, guard;
	struct pthread *self = pthread_self(), *new;
	unsigned char *map, *stack, *tsd;
	static const pthread_attr_t default_attr;

	if (!self) return errno = ENOSYS;
	if (!init && ++init) init_threads();

	if (!attr) attr = &default_attr;
	guard = ROUND(attr->_a_guardsize + DEFAULT_GUARD_SIZE);
	size = guard + ROUND(attr->_a_stacksize + DEFAULT_STACK_SIZE);
	size += __pthread_tsd_size;
	map = mmap(0, size, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE|MAP_ANON, -1, 0);
	if (!map) return EAGAIN;
	if (guard) mprotect(map, guard, PROT_NONE);

	tsd = map + size - __pthread_tsd_size;
	new = (void *)(tsd - sizeof *new - PAGE_SIZE%sizeof *new);
	new->map_base = map;
	new->map_size = size;
	new->pid = self->pid;
	new->errno_ptr = &new->errno_val;
	new->start = entry;
	new->start_arg = arg;
	new->self = new;
	new->tsd = (void *)tsd;
	new->detached = attr->_a_detach;
	new->attr = *attr;
	memcpy(new->tlsdesc, self->tlsdesc, sizeof new->tlsdesc);
	new->tlsdesc[1] = (uintptr_t)new;
	stack = (void *)((uintptr_t)new-1 & ~(uintptr_t)15);

	/* We must synchronize new thread creation with rsyscall
	 * delivery. This looks to be the least expensive way: */
	a_inc(&rs.blocks);
	while (rs.lock) __wait(&rs.lock, 0, 1, 1);

	a_inc(&libc.threads_minus_1);
	ret = __uniclone(stack, start, new);

	a_dec(&rs.blocks);
	if (rs.lock) __wake(&rs.blocks, 1, 1);

	if (ret < 0) {
		a_dec(&libc.threads_minus_1);
		munmap(map, size);
		return EAGAIN;
	}
	*res = new;
	return 0;
}

void pthread_exit(void *result)
{
	struct pthread *self = pthread_self();
	self->result = result;
	docancel(self);
}