procutil: avoid using os.fork() to implement runbgcommand
We ran into the following deadlock:
- some command creates an ssh peer, then raises without explicitly
closing the peer (hg id + extension in our case)
- dispatch catches the exception, calls ui.log('commandfinish', ..)
(the sshpeer is still not closed), which calls logtoprocess, which
calls procutil.runbgcommand.
- in the child of runbgcommand's fork(), between the fork and the
exec, the opening of file descriptors triggers a gc which runs the
destructor for sshpeer, which waits on ssh's stderr being closed,
which never happens since ssh's stderr is held open by the parent of
the fork where said destructor hasn't run
Remotefilelog appears to have a hack around this deadlock as well.
I don't know if there's more subtlety to it, because even though the
problem is determistic, it is very fragile, so I didn't manage to
reduce it.
I can imagine three ways of tackling this problem:
1. don't run any python between fork and exec in runbgcommand
2. make the finalizer harmless after the fork
3. close the peer without relying on gc behavior
This commit goes with 1, as forking without exec'ing is tricky in
general in a language with gc finalizers. And maybe it's better in the
presence of rust threads. A future commit will try 2 or 3.
Performance wise: at low memory usage, it's an improvement. At higher
memory usage, it's about 2x faster than before when ensurestart=True,
but 2x slower when ensurestart=False. Not sure if that matters. The
reason for that last bit is that the subprocess.Popen always waits for
the execve to finish, and at high memory usage, execve is slow because
it deallocates the large page table. Numbers and script:
before after
mem=1.0GB, ensurestart=True 52.1ms 26.0ms
mem=1.0GB, ensurestart=False 14.7ms 26.0ms
mem=0.5GB, ensurestart=True 23.2ms 11.2ms
mem=0.5GB, ensurestart=False 6.2ms 11.3ms
mem=0.2GB, ensurestart=True 15.7ms 7.4ms
mem=0.2GB, ensurestart=False 4.3ms 8.1ms
mem=0.0GB, ensurestart=True 2.3ms 0.7ms
mem=0.0GB, ensurestart=False 0.8ms 0.8ms
import time
for memsize in [1_000_000_000, 500_000_000, 250_000_000, 0]:
mem = 'a' * memsize
for ensurestart in [True, False]:
now = time.time()
n = 100
for i in range(n):
procutil.runbgcommand([b'true'], {}, ensurestart=ensurestart)
after = time.time()
ms = (after - now) / float(n) * 1000
print(f'mem={memsize / 1e9:.1f}GB, ensurestart={ensurestart} -> {ms:.1f}ms')
Differential Revision: https://phab.mercurial-scm.org/D9019
/*
* A fast client for Mercurial command server
*
* Copyright (c) 2011 Yuya Nishihara <yuya@tcha.org>
*
* This software may be used and distributed according to the terms of the
* GNU General Public License version 2 or any later version.
*/
#include <assert.h>
#include <dirent.h>
#include <errno.h>
#include <fcntl.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/file.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/un.h>
#include <sys/wait.h>
#include <time.h>
#include <unistd.h>
#include "hgclient.h"
#include "procutil.h"
#include "util.h"
#ifndef PATH_MAX
#define PATH_MAX 4096
#endif
struct cmdserveropts {
char sockname[PATH_MAX];
char initsockname[PATH_MAX];
char redirectsockname[PATH_MAX];
size_t argsize;
const char **args;
};
static void initcmdserveropts(struct cmdserveropts *opts)
{
memset(opts, 0, sizeof(struct cmdserveropts));
}
static void freecmdserveropts(struct cmdserveropts *opts)
{
free(opts->args);
opts->args = NULL;
opts->argsize = 0;
}
/*
* Test if an argument is a sensitive flag that should be passed to the server.
* Return 0 if not, otherwise the number of arguments starting from the current
* one that should be passed to the server.
*/
static size_t testsensitiveflag(const char *arg)
{
static const struct {
const char *name;
size_t narg;
} flags[] = {
{"--config", 1}, {"--cwd", 1}, {"--repo", 1},
{"--repository", 1}, {"--traceback", 0}, {"-R", 1},
};
size_t i;
for (i = 0; i < sizeof(flags) / sizeof(flags[0]); ++i) {
size_t len = strlen(flags[i].name);
size_t narg = flags[i].narg;
if (memcmp(arg, flags[i].name, len) == 0) {
if (arg[len] == '\0') {
/* --flag (value) */
return narg + 1;
} else if (arg[len] == '=' && narg > 0) {
/* --flag=value */
return 1;
} else if (flags[i].name[1] != '-') {
/* short flag */
return 1;
}
}
}
return 0;
}
/*
* Parse argv[] and put sensitive flags to opts->args
*/
static void setcmdserverargs(struct cmdserveropts *opts, int argc,
const char *argv[])
{
size_t i, step;
opts->argsize = 0;
for (i = 0, step = 1; i < (size_t)argc; i += step, step = 1) {
if (!argv[i])
continue; /* pass clang-analyse */
if (strcmp(argv[i], "--") == 0)
break;
size_t n = testsensitiveflag(argv[i]);
if (n == 0 || i + n > (size_t)argc)
continue;
opts->args =
reallocx(opts->args, (n + opts->argsize) * sizeof(char *));
memcpy(opts->args + opts->argsize, argv + i,
sizeof(char *) * n);
opts->argsize += n;
step = n;
}
}
static void preparesockdir(const char *sockdir)
{
int r;
r = mkdir(sockdir, 0700);
if (r < 0 && errno != EEXIST)
abortmsgerrno("cannot create sockdir %s", sockdir);
struct stat st;
r = lstat(sockdir, &st);
if (r < 0)
abortmsgerrno("cannot stat %s", sockdir);
if (!S_ISDIR(st.st_mode))
abortmsg("cannot create sockdir %s (file exists)", sockdir);
if (st.st_uid != geteuid() || st.st_mode & 0077)
abortmsg("insecure sockdir %s", sockdir);
}
/*
* Check if a socket directory exists and is only owned by the current user.
* Return 1 if so, 0 if not. This is used to check if XDG_RUNTIME_DIR can be
* used or not. According to the specification [1], XDG_RUNTIME_DIR should be
* ignored if the directory is not owned by the user with mode 0700.
* [1]: https://standards.freedesktop.org/basedir-spec/basedir-spec-latest.html
*/
static int checkruntimedir(const char *sockdir)
{
struct stat st;
int r = lstat(sockdir, &st);
if (r < 0) /* ex. does not exist */
return 0;
if (!S_ISDIR(st.st_mode)) /* ex. is a file, not a directory */
return 0;
return st.st_uid == geteuid() && (st.st_mode & 0777) == 0700;
}
static void getdefaultsockdir(char sockdir[], size_t size)
{
/* by default, put socket file in secure directory
* (${XDG_RUNTIME_DIR}/chg, or /${TMPDIR:-tmp}/chg$UID)
* (permission of socket file may be ignored on some Unices) */
const char *runtimedir = getenv("XDG_RUNTIME_DIR");
int r;
if (runtimedir && checkruntimedir(runtimedir)) {
r = snprintf(sockdir, size, "%s/chg", runtimedir);
} else {
const char *tmpdir = getenv("TMPDIR");
if (!tmpdir)
tmpdir = "/tmp";
r = snprintf(sockdir, size, "%s/chg%d", tmpdir, geteuid());
}
if (r < 0 || (size_t)r >= size)
abortmsg("too long TMPDIR (r = %d)", r);
}
static void setcmdserveropts(struct cmdserveropts *opts)
{
int r;
char sockdir[PATH_MAX];
const char *envsockname = getenv("CHGSOCKNAME");
if (!envsockname) {
getdefaultsockdir(sockdir, sizeof(sockdir));
preparesockdir(sockdir);
}
const char *basename = (envsockname) ? envsockname : sockdir;
const char *sockfmt = (envsockname) ? "%s" : "%s/server";
r = snprintf(opts->sockname, sizeof(opts->sockname), sockfmt, basename);
if (r < 0 || (size_t)r >= sizeof(opts->sockname))
abortmsg("too long TMPDIR or CHGSOCKNAME (r = %d)", r);
r = snprintf(opts->initsockname, sizeof(opts->initsockname), "%s.%u",
opts->sockname, (unsigned)getpid());
if (r < 0 || (size_t)r >= sizeof(opts->initsockname))
abortmsg("too long TMPDIR or CHGSOCKNAME (r = %d)", r);
}
/* If the current program is, say, /a/b/c/chg, returns /a/b/c/hg. */
static char *getrelhgcmd(void)
{
ssize_t n;
char *res, *slash;
int maxsize = 4096;
res = malloc(maxsize);
if (res == NULL)
goto cleanup;
n = readlink("/proc/self/exe", res, maxsize);
if (n < 0 || n >= maxsize)
goto cleanup;
res[n] = '\0';
slash = strrchr(res, '/');
if (slash == NULL)
goto cleanup;
/* 4 is strlen("/hg") + nul byte */
if (slash + 4 >= res + maxsize)
goto cleanup;
memcpy(slash, "/hg", 4);
return res;
cleanup:
free(res);
return NULL;
}
static const char *gethgcmd(void)
{
static const char *hgcmd = NULL;
#ifdef HGPATHREL
int tryrelhgcmd = 1;
#else
int tryrelhgcmd = 0;
#endif
if (!hgcmd) {
hgcmd = getenv("CHGHG");
if (!hgcmd || hgcmd[0] == '\0')
hgcmd = getenv("HG");
if (tryrelhgcmd && (!hgcmd || hgcmd[0] == '\0'))
hgcmd = getrelhgcmd();
if (!hgcmd || hgcmd[0] == '\0')
#ifdef HGPATH
hgcmd = (HGPATH);
#else
hgcmd = "hg";
#endif
}
return hgcmd;
}
static void execcmdserver(const struct cmdserveropts *opts)
{
const char *hgcmd = gethgcmd();
const char *baseargv[] = {
hgcmd,
"serve",
"--cmdserver",
"chgunix",
"--address",
opts->initsockname,
"--daemon-postexec",
"chdir:/",
};
size_t baseargvsize = sizeof(baseargv) / sizeof(baseargv[0]);
size_t argsize = baseargvsize + opts->argsize + 1;
const char **argv = mallocx(sizeof(char *) * argsize);
memcpy(argv, baseargv, sizeof(baseargv));
if (opts->args) {
size_t size = sizeof(char *) * opts->argsize;
memcpy(argv + baseargvsize, opts->args, size);
}
argv[argsize - 1] = NULL;
const char *lc_ctype_env = getenv("LC_CTYPE");
if (lc_ctype_env == NULL) {
if (putenv("CHG_CLEAR_LC_CTYPE=") != 0)
abortmsgerrno("failed to putenv CHG_CLEAR_LC_CTYPE");
} else {
if (setenv("CHGORIG_LC_CTYPE", lc_ctype_env, 1) != 0) {
abortmsgerrno("failed to setenv CHGORIG_LC_CTYPE");
}
}
/* close any open files to avoid hanging locks */
DIR *dp = opendir("/proc/self/fd");
if (dp != NULL) {
debugmsg("closing files based on /proc contents");
struct dirent *de;
while ((de = readdir(dp))) {
errno = 0;
char *end;
long fd_value = strtol(de->d_name, &end, 10);
if (end == de->d_name) {
/* unable to convert to int (. or ..) */
continue;
}
if (errno == ERANGE) {
debugmsg("tried to parse %s, but range error "
"occurred",
de->d_name);
continue;
}
if (fd_value > STDERR_FILENO && fd_value != dirfd(dp)) {
debugmsg("closing fd %ld", fd_value);
int res = close(fd_value);
if (res) {
debugmsg("tried to close fd %ld: %d "
"(errno: %d)",
fd_value, res, errno);
}
}
}
closedir(dp);
}
if (putenv("CHGINTERNALMARK=") != 0)
abortmsgerrno("failed to putenv");
if (execvp(hgcmd, (char **)argv) < 0)
abortmsgerrno("failed to exec cmdserver");
free(argv);
}
/* Retry until we can connect to the server. Give up after some time. */
static hgclient_t *retryconnectcmdserver(struct cmdserveropts *opts, pid_t pid)
{
static const struct timespec sleepreq = {0, 10 * 1000000};
int pst = 0;
debugmsg("try connect to %s repeatedly", opts->initsockname);
unsigned int timeoutsec = 60; /* default: 60 seconds */
const char *timeoutenv = getenv("CHGTIMEOUT");
if (timeoutenv)
sscanf(timeoutenv, "%u", &timeoutsec);
for (unsigned int i = 0; !timeoutsec || i < timeoutsec * 100; i++) {
hgclient_t *hgc = hgc_open(opts->initsockname);
if (hgc) {
debugmsg("rename %s to %s", opts->initsockname,
opts->sockname);
int r = rename(opts->initsockname, opts->sockname);
if (r != 0)
abortmsgerrno("cannot rename");
return hgc;
}
if (pid > 0) {
/* collect zombie if child process fails to start */
int r = waitpid(pid, &pst, WNOHANG);
if (r != 0)
goto cleanup;
}
nanosleep(&sleepreq, NULL);
}
abortmsg("timed out waiting for cmdserver %s", opts->initsockname);
return NULL;
cleanup:
if (WIFEXITED(pst)) {
if (WEXITSTATUS(pst) == 0)
abortmsg("could not connect to cmdserver "
"(exited with status 0)");
debugmsg("cmdserver exited with status %d", WEXITSTATUS(pst));
exit(WEXITSTATUS(pst));
} else if (WIFSIGNALED(pst)) {
abortmsg("cmdserver killed by signal %d", WTERMSIG(pst));
} else {
abortmsg("error while waiting for cmdserver");
}
return NULL;
}
/* Connect to a cmdserver. Will start a new server on demand. */
static hgclient_t *connectcmdserver(struct cmdserveropts *opts)
{
const char *sockname =
opts->redirectsockname[0] ? opts->redirectsockname : opts->sockname;
debugmsg("try connect to %s", sockname);
hgclient_t *hgc = hgc_open(sockname);
if (hgc)
return hgc;
/* prevent us from being connected to an outdated server: we were
* told by a server to redirect to opts->redirectsockname and that
* address does not work. we do not want to connect to the server
* again because it will probably tell us the same thing. */
if (sockname == opts->redirectsockname)
unlink(opts->sockname);
debugmsg("start cmdserver at %s", opts->initsockname);
pid_t pid = fork();
if (pid < 0)
abortmsg("failed to fork cmdserver process");
if (pid == 0) {
execcmdserver(opts);
} else {
hgc = retryconnectcmdserver(opts, pid);
}
return hgc;
}
static void killcmdserver(const struct cmdserveropts *opts)
{
/* resolve config hash */
char *resolvedpath = realpath(opts->sockname, NULL);
if (resolvedpath) {
unlink(resolvedpath);
free(resolvedpath);
}
}
/* Run instructions sent from the server like unlink and set redirect path
* Return 1 if reconnect is needed, otherwise 0 */
static int runinstructions(struct cmdserveropts *opts, const char **insts)
{
int needreconnect = 0;
if (!insts)
return needreconnect;
assert(insts);
opts->redirectsockname[0] = '\0';
const char **pinst;
for (pinst = insts; *pinst; pinst++) {
debugmsg("instruction: %s", *pinst);
if (strncmp(*pinst, "unlink ", 7) == 0) {
unlink(*pinst + 7);
} else if (strncmp(*pinst, "redirect ", 9) == 0) {
int r = snprintf(opts->redirectsockname,
sizeof(opts->redirectsockname), "%s",
*pinst + 9);
if (r < 0 || r >= (int)sizeof(opts->redirectsockname))
abortmsg("redirect path is too long (%d)", r);
needreconnect = 1;
} else if (strncmp(*pinst, "exit ", 5) == 0) {
int n = 0;
if (sscanf(*pinst + 5, "%d", &n) != 1)
abortmsg("cannot read the exit code");
exit(n);
} else if (strcmp(*pinst, "reconnect") == 0) {
needreconnect = 1;
} else {
abortmsg("unknown instruction: %s", *pinst);
}
}
return needreconnect;
}
/*
* Test whether the command and the environment is unsupported or not.
*
* If any of the stdio file descriptors are not present (rare, but some tools
* might spawn new processes without stdio instead of redirecting them to the
* null device), then mark it as not supported because attachio won't work
* correctly.
*
* The command list is not designed to cover all cases. But it's fast, and does
* not depend on the server.
*/
static int isunsupported(int argc, const char *argv[])
{
enum {
SERVE = 1,
DAEMON = 2,
SERVEDAEMON = SERVE | DAEMON,
};
unsigned int state = 0;
int i;
/* use fcntl to test missing stdio fds */
if (fcntl(STDIN_FILENO, F_GETFD) == -1 ||
fcntl(STDOUT_FILENO, F_GETFD) == -1 ||
fcntl(STDERR_FILENO, F_GETFD) == -1) {
debugmsg("stdio fds are missing");
return 1;
}
for (i = 0; i < argc; ++i) {
if (strcmp(argv[i], "--") == 0)
break;
/*
* there can be false positives but no false negative
* we cannot assume `serve` will always be first argument
* because global options can be passed before the command name
*/
if (strcmp("serve", argv[i]) == 0)
state |= SERVE;
else if (strcmp("-d", argv[i]) == 0 ||
strcmp("--daemon", argv[i]) == 0)
state |= DAEMON;
}
return (state & SERVEDAEMON) == SERVEDAEMON;
}
static void execoriginalhg(const char *argv[])
{
debugmsg("execute original hg");
if (execvp(gethgcmd(), (char **)argv) < 0)
abortmsgerrno("failed to exec original hg");
}
int main(int argc, const char *argv[], const char *envp[])
{
if (getenv("CHGDEBUG"))
enabledebugmsg();
if (!getenv("HGPLAIN") && isatty(fileno(stderr)))
enablecolor();
if (getenv("CHGINTERNALMARK"))
abortmsg("chg started by chg detected.\n"
"Please make sure ${HG:-hg} is not a symlink or "
"wrapper to chg. Alternatively, set $CHGHG to the "
"path of real hg.");
if (isunsupported(argc - 1, argv + 1))
execoriginalhg(argv);
struct cmdserveropts opts;
initcmdserveropts(&opts);
setcmdserveropts(&opts);
setcmdserverargs(&opts, argc, argv);
if (argc == 2) {
if (strcmp(argv[1], "--kill-chg-daemon") == 0) {
killcmdserver(&opts);
return 0;
}
}
hgclient_t *hgc;
size_t retry = 0;
while (1) {
hgc = connectcmdserver(&opts);
if (!hgc)
abortmsg("cannot open hg client");
hgc_setenv(hgc, envp);
const char **insts = hgc_validate(hgc, argv + 1, argc - 1);
int needreconnect = runinstructions(&opts, insts);
free(insts);
if (!needreconnect)
break;
hgc_close(hgc);
if (++retry > 10)
abortmsg("too many redirections.\n"
"Please make sure %s is not a wrapper which "
"changes sensitive environment variables "
"before executing hg. If you have to use a "
"wrapper, wrap chg instead of hg.",
gethgcmd());
}
setupsignalhandler(hgc_peerpid(hgc), hgc_peerpgid(hgc));
atexit(waitpager);
int exitcode = hgc_runcommand(hgc, argv + 1, argc - 1);
restoresignalhandler();
hgc_close(hgc);
freecmdserveropts(&opts);
return exitcode;
}