summaryrefslogtreecommitdiff
path: root/src/process/posix_spawn.c
blob: c55907d31b176b05cbbe89a0fa52682b81ece2c8 (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
#define _GNU_SOURCE
#include <spawn.h>
#include <sched.h>
#include <unistd.h>
#include <signal.h>
#include <fcntl.h>
#include <sys/wait.h>
#include "syscall.h"
#include "pthread_impl.h"
#include "fdop.h"
#include "libc.h"

static void dummy_0()
{
}
weak_alias(dummy_0, __acquire_ptc);
weak_alias(dummy_0, __release_ptc);

struct args {
	int p[2];
	sigset_t oldmask;
	const char *path;
	int (*exec)(const char *, char *const *, char *const *);
	const posix_spawn_file_actions_t *fa;
	const posix_spawnattr_t *restrict attr;
	char *const *argv, *const *envp;
};

static int child(void *args_vp)
{
	int i, ret;
	struct sigaction sa;
	struct args *args = args_vp;
	int p = args->p[1];
	const posix_spawn_file_actions_t *fa = args->fa;
	const posix_spawnattr_t *restrict attr = args->attr;

	close(args->p[0]);

	/* All signal dispositions must be either SIG_DFL or SIG_IGN
	 * before signals are unblocked. Otherwise a signal handler
	 * from the parent might get run in the child while sharing
	 * memory, with unpredictable and dangerous results. */
	for (i=1; i<_NSIG; i++) {
		__libc_sigaction(i, 0, &sa);
		if (sa.sa_handler!=SIG_DFL && (sa.sa_handler!=SIG_IGN ||
		    ((attr->__flags & POSIX_SPAWN_SETSIGDEF)
		     && sigismember(&attr->__def, i) ))) {
			sa.sa_handler = SIG_DFL;
			__libc_sigaction(i, &sa, 0);
		}
	}

	if (attr->__flags & POSIX_SPAWN_SETPGROUP)
		if ((ret=__syscall(SYS_setpgid, 0, attr->__pgrp)))
			goto fail;

	/* Use syscalls directly because pthread state because the
	 * library functions attempt to do a multi-threaded synchronized
	 * id-change, which would trash the parent's state. */
	if (attr->__flags & POSIX_SPAWN_RESETIDS)
		if ((ret=__syscall(SYS_setgid, __syscall(SYS_getgid))) ||
		    (ret=__syscall(SYS_setuid, __syscall(SYS_getuid))) )
			goto fail;

	if (fa && fa->__actions) {
		struct fdop *op;
		int fd;
		for (op = fa->__actions; op->next; op = op->next);
		for (; op; op = op->prev) {
			/* It's possible that a file operation would clobber
			 * the pipe fd used for synchronizing with the
			 * parent. To avoid that, we dup the pipe onto
			 * an unoccupied fd. */
			if (op->fd == p) {
				ret = __syscall(SYS_dup, p);
				if (ret < 0) goto fail;
				__syscall(SYS_close, p);
				p = ret;
			}
			switch(op->cmd) {
			case FDOP_CLOSE:
				if ((ret=__syscall(SYS_close, op->fd)))
					goto fail;
				break;
			case FDOP_DUP2:
				if ((ret=__syscall(SYS_dup2, op->srcfd, op->fd))<0)
					goto fail;
				break;
			case FDOP_OPEN:
				fd = __syscall(SYS_open, op->path,
					op->oflag | O_LARGEFILE, op->mode);
				if ((ret=fd) < 0) goto fail;
				if (fd != op->fd) {
					if ((ret=__syscall(SYS_dup2, fd, op->fd))<0)
						goto fail;
					__syscall(SYS_close, fd);
				}
				break;
			}
		}
	}

	/* Close-on-exec flag may have been lost if we moved the pipe
	 * to a different fd. We don't use F_DUPFD_CLOEXEC above because
	 * it would fail on older kernels and atomicity is not needed --
	 * in this process there are no threads or signal handlers. */
	__syscall(SYS_fcntl, p, F_SETFD, FD_CLOEXEC);

	pthread_sigmask(SIG_SETMASK, (attr->__flags & POSIX_SPAWN_SETSIGMASK)
		? &attr->__mask : &args->oldmask, 0);

	args->exec(args->path, args->argv, args->envp);

fail:
	/* Since sizeof errno < PIPE_BUF, the write is atomic. */
	ret = -ret;
	if (ret) while (write(p, &ret, sizeof ret) < 0);
	_exit(127);
}


int __posix_spawnx(pid_t *restrict res, const char *restrict path,
	int (*exec)(const char *, char *const *, char *const *),
	const posix_spawn_file_actions_t *fa,
	const posix_spawnattr_t *restrict attr,
	char *const argv[restrict], char *const envp[restrict])
{
	pid_t pid;
	char stack[1024];
	int ec=0, cs;
	struct args args;

	if (pipe2(args.p, O_CLOEXEC))
		return errno;

	pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cs);

	args.path = path;
	args.exec = exec;
	args.fa = fa;
	args.attr = attr ? attr : &(const posix_spawnattr_t){0};
	args.argv = argv;
	args.envp = envp;
	pthread_sigmask(SIG_BLOCK, SIGALL_SET, &args.oldmask);

	/* This lock prevents setuid/setgid operations while the parent
	 * is sharing memory with the child. Situations where processes
	 * with different permissions share VM are fundamentally unsafe. */
	__acquire_ptc();
	pid = __clone(child, stack+sizeof stack, CLONE_VM|SIGCHLD, &args);
	close(args.p[1]);

	if (pid > 0) {
		if (read(args.p[0], &ec, sizeof ec) < sizeof ec) ec = 0;
		else waitpid(pid, &(int){0}, 0);
	} else {
		ec = -pid;
	}

	/* At this point, the child has either exited or successfully
	 * performed exec, so the lock may be released. */
	__release_ptc();
	close(args.p[0]);

	if (!ec) *res = pid;

	pthread_sigmask(SIG_SETMASK, &args.oldmask, 0);
	pthread_setcancelstate(cs, 0);

	return ec;
}

int posix_spawn(pid_t *restrict res, const char *restrict path,
	const posix_spawn_file_actions_t *fa,
	const posix_spawnattr_t *restrict attr,
	char *const argv[restrict], char *const envp[restrict])
{
	return __posix_spawnx(res, path, execve, fa, attr, argv, envp);
}