2020-06-15 14:18:57 +00:00
|
|
|
/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:2;tab-width:8;coding:utf-8 -*-│
|
2023-12-08 03:11:56 +00:00
|
|
|
│ vi: set et ft=c ts=2 sts=2 sw=2 fenc=utf-8 :vi │
|
2020-06-15 14:18:57 +00:00
|
|
|
╞══════════════════════════════════════════════════════════════════════════════╡
|
2021-03-25 09:21:13 +00:00
|
|
|
│ Copyright 2021 Justine Alexandra Roberts Tunney │
|
2020-06-15 14:18:57 +00:00
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
│ Permission to use, copy, modify, and/or distribute this software for │
|
|
|
|
│ any purpose with or without fee is hereby granted, provided that the │
|
|
|
|
│ above copyright notice and this permission notice appear in all copies. │
|
2020-06-15 14:18:57 +00:00
|
|
|
│ │
|
2020-12-28 01:18:44 +00:00
|
|
|
│ THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL │
|
|
|
|
│ WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED │
|
|
|
|
│ WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE │
|
|
|
|
│ AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL │
|
|
|
|
│ DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR │
|
|
|
|
│ PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER │
|
|
|
|
│ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR │
|
|
|
|
│ PERFORMANCE OF THIS SOFTWARE. │
|
2020-06-15 14:18:57 +00:00
|
|
|
╚─────────────────────────────────────────────────────────────────────────────*/
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/assert.h"
|
|
|
|
#include "libc/atomic.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/calls/calls.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/calls/struct/rusage.h"
|
2022-09-13 06:10:38 +00:00
|
|
|
#include "libc/calls/struct/sigaction.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/calls/struct/sigset.h"
|
2022-06-09 03:01:28 +00:00
|
|
|
#include "libc/calls/struct/stat.h"
|
2023-08-08 03:22:49 +00:00
|
|
|
#include "libc/calls/struct/timespec.h"
|
2022-05-28 12:50:01 +00:00
|
|
|
#include "libc/calls/struct/timeval.h"
|
2022-04-20 16:56:53 +00:00
|
|
|
#include "libc/dce.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/errno.h"
|
2020-12-09 23:04:54 +00:00
|
|
|
#include "libc/fmt/conv.h"
|
2023-09-12 15:58:57 +00:00
|
|
|
#include "libc/fmt/itoa.h"
|
2023-06-18 12:39:31 +00:00
|
|
|
#include "libc/fmt/libgen.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/intrin/kprintf.h"
|
|
|
|
#include "libc/log/appendresourcereport.internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/log/check.h"
|
2022-04-20 16:56:53 +00:00
|
|
|
#include "libc/macros.internal.h"
|
2022-09-13 06:10:38 +00:00
|
|
|
#include "libc/mem/gc.h"
|
2024-06-23 17:08:48 +00:00
|
|
|
#include "libc/mem/leaks.h"
|
2023-08-20 09:13:02 +00:00
|
|
|
#include "libc/mem/mem.h"
|
2021-08-26 04:35:58 +00:00
|
|
|
#include "libc/nexgen32e/crc32.h"
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
#include "libc/proc/posix_spawn.h"
|
2023-09-21 14:30:39 +00:00
|
|
|
#include "libc/runtime/internal.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/runtime/runtime.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/runtime/syslib.internal.h"
|
2024-03-03 00:57:56 +00:00
|
|
|
#include "libc/serialize.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/sock/sock.h"
|
2022-08-06 10:51:50 +00:00
|
|
|
#include "libc/sock/struct/pollfd.h"
|
|
|
|
#include "libc/sock/struct/sockaddr.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/stdio/append.h"
|
|
|
|
#include "libc/stdio/rand.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/stdio/stdio.h"
|
|
|
|
#include "libc/str/str.h"
|
|
|
|
#include "libc/sysv/consts/af.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/sysv/consts/at.h"
|
|
|
|
#include "libc/sysv/consts/clock.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/sysv/consts/ex.h"
|
|
|
|
#include "libc/sysv/consts/exit.h"
|
|
|
|
#include "libc/sysv/consts/f.h"
|
|
|
|
#include "libc/sysv/consts/fd.h"
|
|
|
|
#include "libc/sysv/consts/inaddr.h"
|
|
|
|
#include "libc/sysv/consts/ipproto.h"
|
2021-02-05 20:19:43 +00:00
|
|
|
#include "libc/sysv/consts/itimer.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/sysv/consts/o.h"
|
|
|
|
#include "libc/sysv/consts/poll.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/sysv/consts/posix.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/sysv/consts/sa.h"
|
2022-05-22 15:13:13 +00:00
|
|
|
#include "libc/sysv/consts/sig.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/sysv/consts/so.h"
|
|
|
|
#include "libc/sysv/consts/sock.h"
|
|
|
|
#include "libc/sysv/consts/sol.h"
|
2021-01-25 21:08:05 +00:00
|
|
|
#include "libc/sysv/consts/w.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/temp.h"
|
|
|
|
#include "libc/thread/thread.h"
|
|
|
|
#include "libc/thread/thread2.h"
|
2024-05-05 06:05:36 +00:00
|
|
|
#include "libc/time.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "libc/x/x.h"
|
2023-09-10 15:12:43 +00:00
|
|
|
#include "libc/x/xsigaction.h"
|
|
|
|
#include "net/http/escape.h"
|
2021-08-14 13:17:56 +00:00
|
|
|
#include "net/https/https.h"
|
2023-07-03 02:57:43 +00:00
|
|
|
#include "third_party/getopt/getopt.internal.h"
|
2023-09-12 15:58:57 +00:00
|
|
|
#include "third_party/mbedtls/debug.h"
|
2021-08-07 20:22:35 +00:00
|
|
|
#include "third_party/mbedtls/ssl.h"
|
2022-05-15 06:17:22 +00:00
|
|
|
#include "third_party/zlib/zlib.h"
|
2021-08-07 20:22:35 +00:00
|
|
|
#include "tool/build/lib/eztls.h"
|
2021-08-14 13:17:56 +00:00
|
|
|
#include "tool/build/lib/psk.h"
|
2020-06-15 14:18:57 +00:00
|
|
|
#include "tool/build/runit.h"
|
|
|
|
|
|
|
|
/**
|
|
|
|
* @fileoverview Remote test runner daemon.
|
|
|
|
* Delivers 10x latency improvement over SSH (100x if Debian defaults)
|
|
|
|
*
|
|
|
|
* Here's how it handles connections:
|
|
|
|
*
|
|
|
|
* 1. Receives atomically-written request header, comprised of:
|
|
|
|
*
|
|
|
|
* - 4 byte nbo magic = 0xFEEDABEEu
|
|
|
|
* - 1 byte command = kRunitExecute
|
|
|
|
* - 4 byte nbo name length in bytes, e.g. "test1"
|
|
|
|
* - 4 byte nbo executable file length in bytes
|
|
|
|
* - <name bytes> (no NUL terminator)
|
|
|
|
* - <file bytes> (it's binary data)
|
|
|
|
*
|
|
|
|
* 2. Runs program, after verifying it came from the IP that spawned
|
|
|
|
* this program via SSH. Be sure to only run this over a trusted
|
|
|
|
* physically-wired network. To use this software on untrustworthy
|
|
|
|
* networks, wrap it with stunnel and use your own CA.
|
|
|
|
*
|
|
|
|
* 3. Sends stdout/stderr fragments, potentially multiple times:
|
|
|
|
*
|
|
|
|
* - 4 byte nbo magic = 0xFEEDABEEu
|
|
|
|
* - 1 byte command = kRunitStdout/Stderr
|
|
|
|
* - 4 byte nbo byte length
|
|
|
|
* - <chunk bytes>
|
|
|
|
*
|
|
|
|
* 4. Sends process exit code:
|
|
|
|
*
|
|
|
|
* - 4 byte nbo magic = 0xFEEDABEEu
|
|
|
|
* - 1 byte command = kRunitExit
|
|
|
|
* - 1 byte exit status
|
|
|
|
*/
|
|
|
|
|
2023-08-08 03:22:49 +00:00
|
|
|
#define DEATH_CLOCK_SECONDS 300
|
2021-06-24 19:31:26 +00:00
|
|
|
|
2020-06-15 14:18:57 +00:00
|
|
|
#define kLogFile "o/runitd.log"
|
|
|
|
#define kLogMaxBytes (2 * 1000 * 1000)
|
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
#define LOG_LEVEL_WARN 0
|
|
|
|
#define LOG_LEVEL_INFO 1
|
2023-09-12 15:58:57 +00:00
|
|
|
#define LOG_LEVEL_VERB 2
|
2023-09-10 15:12:43 +00:00
|
|
|
#define LOG_LEVEL_DEBU 3
|
|
|
|
|
|
|
|
#define DEBUF(FMT, ...) LOGF(DEBU, FMT, ##__VA_ARGS__)
|
|
|
|
#define VERBF(FMT, ...) LOGF(VERB, FMT, ##__VA_ARGS__)
|
|
|
|
#define INFOF(FMT, ...) LOGF(INFO, FMT, ##__VA_ARGS__)
|
|
|
|
#define WARNF(FMT, ...) LOGF(WARN, FMT, ##__VA_ARGS__)
|
|
|
|
|
|
|
|
#define LOGF(LVL, FMT, ...) \
|
|
|
|
do { \
|
|
|
|
if (g_log_level >= LOG_LEVEL_##LVL) { \
|
|
|
|
kprintf("%r" #LVL " %6P %'18T %s:%d " FMT "\n", __FILE__, __LINE__, \
|
|
|
|
##__VA_ARGS__); \
|
|
|
|
} \
|
|
|
|
} while (0)
|
|
|
|
|
|
|
|
struct Client {
|
|
|
|
int fd;
|
|
|
|
int pid;
|
|
|
|
int pipe[2];
|
|
|
|
pthread_t th;
|
|
|
|
uint32_t addrsize;
|
|
|
|
struct sockaddr_in addr;
|
|
|
|
bool once;
|
|
|
|
int zstatus;
|
|
|
|
z_stream zs;
|
|
|
|
struct {
|
|
|
|
size_t off;
|
|
|
|
size_t len;
|
|
|
|
size_t cap;
|
|
|
|
char *data;
|
|
|
|
} rbuf;
|
|
|
|
char *output;
|
2023-09-12 15:58:57 +00:00
|
|
|
char tmpexepath[128];
|
2023-09-10 15:12:43 +00:00
|
|
|
char buf[32768];
|
|
|
|
};
|
|
|
|
|
|
|
|
char *g_psk;
|
|
|
|
int g_log_level;
|
2022-04-15 06:39:48 +00:00
|
|
|
bool use_ftrace;
|
|
|
|
bool use_strace;
|
2023-09-10 15:12:43 +00:00
|
|
|
char g_hostname[256];
|
|
|
|
int g_bogusfd, g_servfd;
|
|
|
|
atomic_bool g_interrupted;
|
2020-06-15 14:18:57 +00:00
|
|
|
struct sockaddr_in g_servaddr;
|
2022-04-20 16:56:53 +00:00
|
|
|
bool g_daemonize, g_sendready;
|
2020-06-15 14:18:57 +00:00
|
|
|
|
2021-01-28 23:49:15 +00:00
|
|
|
void OnInterrupt(int sig) {
|
|
|
|
g_interrupted = true;
|
|
|
|
}
|
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
void Close(int *fd) {
|
|
|
|
if (*fd > 0) {
|
|
|
|
close(*fd);
|
|
|
|
*fd = -1; // poll ignores -1
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2020-12-05 20:20:41 +00:00
|
|
|
wontreturn void ShowUsage(FILE *f, int rc) {
|
2020-06-15 14:18:57 +00:00
|
|
|
fprintf(f, "%s: %s %s\n", "Usage", program_invocation_name,
|
|
|
|
"[-d] [-r] [-l LISTENIP] [-p PORT] [-t TIMEOUTMS]");
|
|
|
|
exit(rc);
|
|
|
|
}
|
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
char *DescribeAddress(struct sockaddr_in *addr) {
|
|
|
|
static _Thread_local char res[64];
|
|
|
|
char ip4buf[64];
|
|
|
|
sprintf(res, "%s:%hu",
|
|
|
|
inet_ntop(addr->sin_family, &addr->sin_addr.s_addr, ip4buf,
|
|
|
|
sizeof(ip4buf)),
|
|
|
|
ntohs(addr->sin_port));
|
|
|
|
return res;
|
|
|
|
}
|
|
|
|
|
2020-06-15 14:18:57 +00:00
|
|
|
void GetOpts(int argc, char *argv[]) {
|
|
|
|
int opt;
|
|
|
|
g_servaddr.sin_family = AF_INET;
|
|
|
|
g_servaddr.sin_port = htons(RUNITD_PORT);
|
|
|
|
g_servaddr.sin_addr.s_addr = INADDR_ANY;
|
2023-09-12 15:58:57 +00:00
|
|
|
while ((opt = getopt(argc, argv, "fqhvVsdrl:p:t:w:")) != -1) {
|
2020-06-15 14:18:57 +00:00
|
|
|
switch (opt) {
|
2022-04-15 06:39:48 +00:00
|
|
|
case 'f':
|
|
|
|
use_ftrace = true;
|
|
|
|
break;
|
Improve ZIP filesystem and change its prefix
The ZIP filesystem has a breaking change. You now need to use /zip/ to
open() / opendir() / etc. assets within the ZIP structure of your APE
binary, instead of the previous convention of using zip: or zip! URIs.
This is needed because Python likes to use absolute paths, and having
ZIP paths encoded like URIs simply broke too many things.
Many more system calls have been updated to be able to operate on ZIP
files and file descriptors. In particular fcntl() and ioctl() since
Python would do things like ask if a ZIP file is a terminal and get
confused when the old implementation mistakenly said yes, because the
fastest way to guarantee native file descriptors is to dup(2). This
change also improves the async signal safety of zipos and ensures it
doesn't maintain any open file descriptors beyond that which the user
has opened.
This change makes a lot of progress towards adding magic numbers that
are specific to platforms other than Linux. The philosophy here is that,
if you use an operating system like FreeBSD, then you should be able to
take advantage of FreeBSD exclusive features, even if we don't polyfill
them on other platforms. For example, you can now open() a file with the
O_VERIFY flag. If your program runs on other platforms, then Cosmo will
automatically set O_VERIFY to zero. This lets you safely use it without
the need for #ifdef or ifstatements which detract from readability.
One of the blindspots of the ASAN memory hardening we use to offer Rust
like assurances has always been that memory passed to the kernel via
system calls (e.g. writev) can't be checked automatically since the
kernel wasn't built with MODE=asan. This change makes more progress
ensuring that each system call will verify the soundness of memory
before it's passed to the kernel. The code for doing these checks is
fast, particularly for buffers, where it can verify 64 bytes a cycle.
- Correct O_LOOP definition on NT
- Introduce program_executable_name
- Add ASAN guards to more system calls
- Improve termios compatibility with BSDs
- Fix bug in Windows auxiliary value encoding
- Add BSD and XNU specific errnos and open flags
- Add check to ensure build doesn't talk to internet
2021-08-22 08:04:18 +00:00
|
|
|
case 's':
|
2022-04-15 06:39:48 +00:00
|
|
|
use_strace = true;
|
|
|
|
break;
|
|
|
|
case 'q':
|
2023-09-10 15:12:43 +00:00
|
|
|
--g_log_level;
|
Improve ZIP filesystem and change its prefix
The ZIP filesystem has a breaking change. You now need to use /zip/ to
open() / opendir() / etc. assets within the ZIP structure of your APE
binary, instead of the previous convention of using zip: or zip! URIs.
This is needed because Python likes to use absolute paths, and having
ZIP paths encoded like URIs simply broke too many things.
Many more system calls have been updated to be able to operate on ZIP
files and file descriptors. In particular fcntl() and ioctl() since
Python would do things like ask if a ZIP file is a terminal and get
confused when the old implementation mistakenly said yes, because the
fastest way to guarantee native file descriptors is to dup(2). This
change also improves the async signal safety of zipos and ensures it
doesn't maintain any open file descriptors beyond that which the user
has opened.
This change makes a lot of progress towards adding magic numbers that
are specific to platforms other than Linux. The philosophy here is that,
if you use an operating system like FreeBSD, then you should be able to
take advantage of FreeBSD exclusive features, even if we don't polyfill
them on other platforms. For example, you can now open() a file with the
O_VERIFY flag. If your program runs on other platforms, then Cosmo will
automatically set O_VERIFY to zero. This lets you safely use it without
the need for #ifdef or ifstatements which detract from readability.
One of the blindspots of the ASAN memory hardening we use to offer Rust
like assurances has always been that memory passed to the kernel via
system calls (e.g. writev) can't be checked automatically since the
kernel wasn't built with MODE=asan. This change makes more progress
ensuring that each system call will verify the soundness of memory
before it's passed to the kernel. The code for doing these checks is
fast, particularly for buffers, where it can verify 64 bytes a cycle.
- Correct O_LOOP definition on NT
- Introduce program_executable_name
- Add ASAN guards to more system calls
- Improve termios compatibility with BSDs
- Fix bug in Windows auxiliary value encoding
- Add BSD and XNU specific errnos and open flags
- Add check to ensure build doesn't talk to internet
2021-08-22 08:04:18 +00:00
|
|
|
break;
|
|
|
|
case 'v':
|
2023-09-10 15:12:43 +00:00
|
|
|
++g_log_level;
|
Improve ZIP filesystem and change its prefix
The ZIP filesystem has a breaking change. You now need to use /zip/ to
open() / opendir() / etc. assets within the ZIP structure of your APE
binary, instead of the previous convention of using zip: or zip! URIs.
This is needed because Python likes to use absolute paths, and having
ZIP paths encoded like URIs simply broke too many things.
Many more system calls have been updated to be able to operate on ZIP
files and file descriptors. In particular fcntl() and ioctl() since
Python would do things like ask if a ZIP file is a terminal and get
confused when the old implementation mistakenly said yes, because the
fastest way to guarantee native file descriptors is to dup(2). This
change also improves the async signal safety of zipos and ensures it
doesn't maintain any open file descriptors beyond that which the user
has opened.
This change makes a lot of progress towards adding magic numbers that
are specific to platforms other than Linux. The philosophy here is that,
if you use an operating system like FreeBSD, then you should be able to
take advantage of FreeBSD exclusive features, even if we don't polyfill
them on other platforms. For example, you can now open() a file with the
O_VERIFY flag. If your program runs on other platforms, then Cosmo will
automatically set O_VERIFY to zero. This lets you safely use it without
the need for #ifdef or ifstatements which detract from readability.
One of the blindspots of the ASAN memory hardening we use to offer Rust
like assurances has always been that memory passed to the kernel via
system calls (e.g. writev) can't be checked automatically since the
kernel wasn't built with MODE=asan. This change makes more progress
ensuring that each system call will verify the soundness of memory
before it's passed to the kernel. The code for doing these checks is
fast, particularly for buffers, where it can verify 64 bytes a cycle.
- Correct O_LOOP definition on NT
- Introduce program_executable_name
- Add ASAN guards to more system calls
- Improve termios compatibility with BSDs
- Fix bug in Windows auxiliary value encoding
- Add BSD and XNU specific errnos and open flags
- Add check to ensure build doesn't talk to internet
2021-08-22 08:04:18 +00:00
|
|
|
break;
|
2023-09-12 15:58:57 +00:00
|
|
|
case 'V':
|
|
|
|
++mbedtls_debug_threshold;
|
|
|
|
break;
|
2020-06-15 14:18:57 +00:00
|
|
|
case 'd':
|
|
|
|
g_daemonize = true;
|
|
|
|
break;
|
|
|
|
case 'r':
|
|
|
|
g_sendready = true;
|
|
|
|
break;
|
|
|
|
case 't':
|
|
|
|
break;
|
|
|
|
case 'p':
|
Make improvements
- We now serialize the file descriptor table when spawning / executing
processes on Windows. This means you can now inherit more stuff than
just standard i/o. It's needed by bash, which duplicates the console
to file descriptor #255. We also now do a better job serializing the
environment variables, so you're less likely to encounter E2BIG when
using your bash shell. We also no longer coerce environ to uppercase
- execve() on Windows now remotely controls its parent process to make
them spawn a replacement for itself. Then it'll be able to terminate
immediately once the spawn succeeds, without having to linger around
for the lifetime as a shell process for proxying the exit code. When
process worker thread running in the parent sees the child die, it's
given a handle to the new child, to replace it in the process table.
- execve() and posix_spawn() on Windows will now provide CreateProcess
an explicit handle list. This allows us to remove handle locks which
enables better fork/spawn concurrency, with seriously correct thread
safety. Other codebases like Go use the same technique. On the other
hand fork() still favors the conventional WIN32 inheritence approach
which can be a little bit messy, but is *controlled* by guaranteeing
perfectly clean slates at both the spawning and execution boundaries
- sigset_t is now 64 bits. Having it be 128 bits was a mistake because
there's no reason to use that and it's only supported by FreeBSD. By
using the system word size, signal mask manipulation on Windows goes
very fast. Furthermore @asyncsignalsafe funcs have been rewritten on
Windows to take advantage of signal masking, now that it's much more
pleasant to use.
- All the overlapped i/o code on Windows has been rewritten for pretty
good signal and cancelation safety. We're now able to ensure overlap
data structures are cleaned up so long as you don't longjmp() out of
out of a signal handler that interrupted an i/o operation. Latencies
are also improved thanks to the removal of lots of "busy wait" code.
Waits should be optimal for everything except poll(), which shall be
the last and final demon we slay in the win32 i/o horror show.
- getrusage() on Windows is now able to report RUSAGE_CHILDREN as well
as RUSAGE_SELF, thanks to aggregation in the process manager thread.
2023-10-08 12:36:18 +00:00
|
|
|
g_servaddr.sin_port = htons(atoi(optarg));
|
2020-06-15 14:18:57 +00:00
|
|
|
break;
|
|
|
|
case 'l':
|
2023-09-10 15:12:43 +00:00
|
|
|
inet_pton(AF_INET, optarg, &g_servaddr.sin_addr);
|
2020-06-15 14:18:57 +00:00
|
|
|
break;
|
|
|
|
case 'h':
|
|
|
|
ShowUsage(stdout, EXIT_SUCCESS);
|
|
|
|
default:
|
|
|
|
ShowUsage(stderr, EX_USAGE);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void StartTcpServer(void) {
|
|
|
|
int yes = true;
|
|
|
|
uint32_t asize;
|
2023-09-10 15:12:43 +00:00
|
|
|
g_servfd = socket(AF_INET, SOCK_STREAM | SOCK_CLOEXEC, IPPROTO_TCP);
|
|
|
|
if (g_servfd == -1) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
fprintf(stderr, "%s: socket failed: %s\n", program_invocation_short_name,
|
|
|
|
strerror(errno));
|
2023-09-10 15:12:43 +00:00
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
struct timeval timeo = {DEATH_CLOCK_SECONDS / 10};
|
2022-05-28 12:50:01 +00:00
|
|
|
setsockopt(g_servfd, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo));
|
|
|
|
setsockopt(g_servfd, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo));
|
2023-09-10 15:12:43 +00:00
|
|
|
setsockopt(g_servfd, SOL_SOCKET, SO_REUSEADDR, &yes, sizeof(yes));
|
2022-10-11 00:52:41 +00:00
|
|
|
if (bind(g_servfd, (struct sockaddr *)&g_servaddr, sizeof(g_servaddr)) ==
|
|
|
|
-1) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
fprintf(stderr, "%s: bind failed: %s\n", program_invocation_short_name,
|
|
|
|
strerror(errno));
|
2023-09-10 15:12:43 +00:00
|
|
|
exit(1);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
unassert(!listen(g_servfd, 10));
|
2020-06-15 14:18:57 +00:00
|
|
|
asize = sizeof(g_servaddr);
|
2023-09-10 15:12:43 +00:00
|
|
|
unassert(!getsockname(g_servfd, (struct sockaddr *)&g_servaddr, &asize));
|
|
|
|
INFOF("listening on tcp:%s", DescribeAddress(&g_servaddr));
|
2020-06-15 14:18:57 +00:00
|
|
|
if (g_sendready) {
|
|
|
|
printf("ready %hu\n", ntohs(g_servaddr.sin_port));
|
2023-09-12 04:34:53 +00:00
|
|
|
close(1);
|
|
|
|
dup2(g_bogusfd, 1);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-08-07 20:22:35 +00:00
|
|
|
void SendExitMessage(int rc) {
|
2023-09-10 15:12:43 +00:00
|
|
|
EzSanity();
|
|
|
|
int res;
|
2020-06-15 14:18:57 +00:00
|
|
|
unsigned char msg[4 + 1 + 1];
|
2023-09-10 15:12:43 +00:00
|
|
|
DEBUF("SendExitMessage");
|
2021-09-28 05:58:51 +00:00
|
|
|
msg[0 + 0] = (RUNITD_MAGIC & 0xff000000) >> 030;
|
|
|
|
msg[0 + 1] = (RUNITD_MAGIC & 0x00ff0000) >> 020;
|
|
|
|
msg[0 + 2] = (RUNITD_MAGIC & 0x0000ff00) >> 010;
|
|
|
|
msg[0 + 3] = (RUNITD_MAGIC & 0x000000ff) >> 000;
|
2020-06-15 14:18:57 +00:00
|
|
|
msg[4] = kRunitExit;
|
2021-09-28 05:58:51 +00:00
|
|
|
msg[5] = rc;
|
2023-09-10 15:12:43 +00:00
|
|
|
DEBUF("mbedtls_ssl_write");
|
|
|
|
if (sizeof(msg) != (res = mbedtls_ssl_write(&ezssl, msg, sizeof(msg)))) {
|
|
|
|
EzTlsDie("SendExitMessage mbedtls_ssl_write failed", res);
|
|
|
|
}
|
|
|
|
if ((res = EzTlsFlush(&ezbio, 0, 0))) {
|
|
|
|
EzTlsDie("SendExitMessage EzTlsFlush failed", res);
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
void SendOutputFragmentMessage(enum RunitCommand kind, char *buf, size_t size) {
|
|
|
|
EzSanity();
|
2020-06-15 14:18:57 +00:00
|
|
|
ssize_t rc;
|
|
|
|
unsigned char msg[4 + 1 + 4];
|
2021-09-28 05:58:51 +00:00
|
|
|
msg[0 + 0] = (RUNITD_MAGIC & 0xff000000) >> 030;
|
|
|
|
msg[0 + 1] = (RUNITD_MAGIC & 0x00ff0000) >> 020;
|
|
|
|
msg[0 + 2] = (RUNITD_MAGIC & 0x0000ff00) >> 010;
|
|
|
|
msg[0 + 3] = (RUNITD_MAGIC & 0x000000ff) >> 000;
|
2020-06-15 14:18:57 +00:00
|
|
|
msg[4 + 0] = kind;
|
2021-09-28 05:58:51 +00:00
|
|
|
msg[5 + 0] = (size & 0xff000000) >> 030;
|
|
|
|
msg[5 + 1] = (size & 0x00ff0000) >> 020;
|
|
|
|
msg[5 + 2] = (size & 0x0000ff00) >> 010;
|
|
|
|
msg[5 + 3] = (size & 0x000000ff) >> 000;
|
2023-09-10 15:12:43 +00:00
|
|
|
DEBUF("mbedtls_ssl_write");
|
|
|
|
if (sizeof(msg) != (rc = mbedtls_ssl_write(&ezssl, msg, sizeof(msg)))) {
|
|
|
|
EzTlsDie("SendOutputFragmentMessage mbedtls_ssl_write failed", rc);
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
while (size) {
|
2023-09-10 15:12:43 +00:00
|
|
|
if ((rc = mbedtls_ssl_write(&ezssl, buf, size)) <= 0) {
|
|
|
|
EzTlsDie("SendOutputFragmentMessage mbedtls_ssl_write #2 failed", rc);
|
|
|
|
}
|
|
|
|
size -= rc;
|
|
|
|
buf += rc;
|
|
|
|
}
|
|
|
|
if ((rc = EzTlsFlush(&ezbio, 0, 0))) {
|
|
|
|
EzTlsDie("SendOutputFragmentMessage EzTlsFlush failed", rc);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
void Recv(struct Client *client, void *output, size_t outputsize) {
|
|
|
|
EzSanity();
|
2023-09-02 03:49:13 +00:00
|
|
|
ssize_t chunk, received, totalgot;
|
2023-09-10 15:12:43 +00:00
|
|
|
if (!client->once) {
|
|
|
|
unassert(Z_OK == inflateInit(&client->zs));
|
|
|
|
client->once = true;
|
2022-05-15 06:17:22 +00:00
|
|
|
}
|
2022-06-13 18:02:13 +00:00
|
|
|
totalgot = 0;
|
2022-05-15 06:17:22 +00:00
|
|
|
for (;;) {
|
2023-09-10 15:12:43 +00:00
|
|
|
if (client->rbuf.len >= outputsize) {
|
|
|
|
memcpy(output, client->rbuf.data + client->rbuf.off, outputsize);
|
|
|
|
client->rbuf.len -= outputsize;
|
|
|
|
client->rbuf.off += outputsize;
|
2022-05-15 06:17:22 +00:00
|
|
|
// trim dymanic buffer once it empties
|
2023-09-10 15:12:43 +00:00
|
|
|
if (!client->rbuf.len) {
|
|
|
|
client->rbuf.off = 0;
|
|
|
|
client->rbuf.cap = 4096;
|
|
|
|
client->rbuf.data = realloc(client->rbuf.data, client->rbuf.cap);
|
2022-05-15 06:17:22 +00:00
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
if (client->zstatus == Z_STREAM_END) {
|
|
|
|
WARNF("recv zlib unexpected eof");
|
|
|
|
pthread_exit(0);
|
2022-05-15 06:17:22 +00:00
|
|
|
}
|
|
|
|
// get another fixed-size data packet from network
|
|
|
|
// pass along error conditions to caller
|
|
|
|
// pass along eof condition to zlib
|
2023-09-10 15:12:43 +00:00
|
|
|
received = mbedtls_ssl_read(&ezssl, client->buf, sizeof(client->buf));
|
2022-06-09 04:52:47 +00:00
|
|
|
if (!received) {
|
2023-09-10 15:12:43 +00:00
|
|
|
EzTlsDie("got unexpected eof", received);
|
2022-06-09 04:52:47 +00:00
|
|
|
}
|
|
|
|
if (received < 0) {
|
2023-09-10 15:12:43 +00:00
|
|
|
EzTlsDie("read failed", received);
|
2022-06-09 04:52:47 +00:00
|
|
|
}
|
2022-06-13 18:02:13 +00:00
|
|
|
totalgot += received;
|
2022-05-15 06:17:22 +00:00
|
|
|
// decompress packet completely
|
|
|
|
// into a dynamical size buffer
|
2023-09-10 15:12:43 +00:00
|
|
|
client->zs.avail_in = received;
|
|
|
|
client->zs.next_in = (unsigned char *)client->buf;
|
|
|
|
unassert(Z_OK == client->zstatus);
|
2021-08-26 04:35:58 +00:00
|
|
|
do {
|
2022-05-15 06:17:22 +00:00
|
|
|
// make sure we have a reasonable capacity for zlib output
|
2023-09-10 15:12:43 +00:00
|
|
|
if (client->rbuf.cap - (client->rbuf.off + client->rbuf.len) <
|
|
|
|
sizeof(client->buf)) {
|
|
|
|
client->rbuf.cap += sizeof(client->buf);
|
|
|
|
client->rbuf.data = realloc(client->rbuf.data, client->rbuf.cap);
|
2022-05-15 06:17:22 +00:00
|
|
|
}
|
|
|
|
// inflate packet, which naturally can be much larger
|
|
|
|
// permit zlib no delay flushes that come from sender
|
2023-09-10 15:12:43 +00:00
|
|
|
client->zs.next_out = (unsigned char *)client->rbuf.data +
|
|
|
|
(client->rbuf.off + client->rbuf.len);
|
|
|
|
client->zs.avail_out = chunk =
|
|
|
|
client->rbuf.cap - (client->rbuf.off + client->rbuf.len);
|
|
|
|
client->zstatus = inflate(&client->zs, Z_SYNC_FLUSH);
|
|
|
|
unassert(Z_STREAM_ERROR != client->zstatus);
|
|
|
|
switch (client->zstatus) {
|
2022-05-15 06:17:22 +00:00
|
|
|
case Z_NEED_DICT:
|
2022-06-13 18:02:13 +00:00
|
|
|
WARNF("tls recv Z_NEED_DICT %ld total %ld", received, totalgot);
|
2023-09-10 15:12:43 +00:00
|
|
|
pthread_exit(0);
|
2022-05-15 06:17:22 +00:00
|
|
|
case Z_DATA_ERROR:
|
2022-06-13 18:02:13 +00:00
|
|
|
WARNF("tls recv Z_DATA_ERROR %ld total %ld", received, totalgot);
|
2023-09-10 15:12:43 +00:00
|
|
|
pthread_exit(0);
|
2022-05-15 06:17:22 +00:00
|
|
|
case Z_MEM_ERROR:
|
2022-06-13 18:02:13 +00:00
|
|
|
WARNF("tls recv Z_MEM_ERROR %ld total %ld", received, totalgot);
|
2023-09-10 15:12:43 +00:00
|
|
|
pthread_exit(0);
|
2022-05-15 06:17:22 +00:00
|
|
|
case Z_BUF_ERROR:
|
2023-09-10 15:12:43 +00:00
|
|
|
client->zstatus = Z_OK; // harmless? nothing for inflate to do
|
|
|
|
break; // it probably just our wraparound eof
|
2022-05-15 06:17:22 +00:00
|
|
|
default:
|
2023-09-10 15:12:43 +00:00
|
|
|
client->rbuf.len += chunk - client->zs.avail_out;
|
2022-05-15 06:17:22 +00:00
|
|
|
break;
|
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
} while (!client->zs.avail_out);
|
2021-08-26 04:35:58 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2023-09-11 20:51:37 +00:00
|
|
|
void SendProgramOutput(struct Client *client) {
|
2023-09-10 15:12:43 +00:00
|
|
|
if (client->output) {
|
|
|
|
SendOutputFragmentMessage(kRunitStderr, client->output,
|
|
|
|
appendz(client->output).i);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void PrintProgramOutput(struct Client *client) {
|
|
|
|
if (client->output) {
|
|
|
|
char *p = client->output;
|
|
|
|
size_t z = appendz(p).i;
|
|
|
|
if ((p = IndentLines(p, z, &z, 2))) {
|
|
|
|
fwrite(p, 1, z, stderr);
|
|
|
|
free(p);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void FreeClient(struct Client *client) {
|
|
|
|
DEBUF("FreeClient");
|
2023-09-10 18:52:03 +00:00
|
|
|
Close(&client->pipe[1]);
|
|
|
|
Close(&client->pipe[0]);
|
2023-09-10 15:12:43 +00:00
|
|
|
if (client->pid) {
|
|
|
|
kill(client->pid, SIGHUP);
|
|
|
|
waitpid(client->pid, 0, 0);
|
|
|
|
}
|
|
|
|
Close(&client->fd);
|
2023-09-12 15:58:57 +00:00
|
|
|
if (*client->tmpexepath) {
|
|
|
|
unlink(client->tmpexepath);
|
2023-09-10 15:12:43 +00:00
|
|
|
}
|
|
|
|
if (client->once) {
|
|
|
|
inflateEnd(&client->zs);
|
|
|
|
}
|
|
|
|
EzDestroy();
|
|
|
|
free(client->rbuf.data);
|
|
|
|
free(client->output);
|
|
|
|
free(client);
|
|
|
|
VERBF("---------------");
|
|
|
|
}
|
|
|
|
|
|
|
|
void *ClientWorker(void *arg) {
|
2021-08-26 04:35:58 +00:00
|
|
|
uint32_t crc;
|
2023-08-08 03:22:49 +00:00
|
|
|
sigset_t sigmask;
|
2023-09-10 15:12:43 +00:00
|
|
|
int events, wstatus;
|
|
|
|
struct Client *client = arg;
|
|
|
|
uint32_t namesize, filesize;
|
2023-09-12 15:58:57 +00:00
|
|
|
char *addrstr, *origname;
|
2021-08-26 04:35:58 +00:00
|
|
|
unsigned char msg[4 + 1 + 4 + 4 + 4];
|
2023-09-10 15:12:43 +00:00
|
|
|
|
|
|
|
SetupPresharedKeySsl(MBEDTLS_SSL_IS_SERVER, g_psk);
|
|
|
|
defer(FreeClient, client);
|
|
|
|
|
|
|
|
// read request to run program
|
|
|
|
EzFd(client->fd);
|
|
|
|
DEBUF("EzHandshake");
|
2021-08-09 21:45:52 +00:00
|
|
|
EzHandshake();
|
2023-09-10 15:12:43 +00:00
|
|
|
addrstr = DescribeAddress(&client->addr);
|
|
|
|
DEBUF("%s %s %s", DescribeAddress(&g_servaddr), "accepted", addrstr);
|
2022-05-15 06:17:22 +00:00
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
// get the executable
|
|
|
|
Recv(client, msg, sizeof(msg));
|
|
|
|
if (READ32BE(msg) != RUNITD_MAGIC) {
|
|
|
|
WARNF("%s magic mismatch!", addrstr);
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
if (msg[4] != kRunitExecute) {
|
|
|
|
WARNF("%s unknown command!", addrstr);
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
2021-08-26 04:35:58 +00:00
|
|
|
namesize = READ32BE(msg + 5);
|
|
|
|
filesize = READ32BE(msg + 9);
|
|
|
|
crc = READ32BE(msg + 13);
|
2023-09-12 15:58:57 +00:00
|
|
|
origname = gc(calloc(1, namesize + 1));
|
|
|
|
Recv(client, origname, namesize);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
VERBF("%s sent %#s (%'u bytes @ %#s)", addrstr, origname, filesize,
|
2023-09-12 15:58:57 +00:00
|
|
|
client->tmpexepath);
|
|
|
|
char *exedata = gc(malloc(filesize));
|
|
|
|
Recv(client, exedata, filesize);
|
|
|
|
if (crc32_z(0, exedata, filesize) != crc) {
|
|
|
|
WARNF("%s crc mismatch! %#s", addrstr, origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
// create the executable file
|
|
|
|
// if another thread vforks while we're writing it then a race
|
|
|
|
// condition can happen, where etxtbsy is raised by our execve
|
|
|
|
// we're using o_cloexec so it's guaranteed to fix itself fast
|
|
|
|
// thus we use an optimistic approach to avoid expensive locks
|
2024-03-03 00:57:56 +00:00
|
|
|
sprintf(client->tmpexepath, "o/%s.XXXXXX",
|
2023-09-21 14:30:39 +00:00
|
|
|
basename(stripext(gc(strdup(origname)))));
|
2024-03-31 02:03:30 +00:00
|
|
|
int exefd = openatemp(AT_FDCWD, client->tmpexepath, 0, O_CLOEXEC, 0700);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
if (exefd == -1) {
|
|
|
|
WARNF("%s failed to open temporary file %#s due to %m", addrstr,
|
|
|
|
client->tmpexepath);
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
if (ftruncate(exefd, filesize)) {
|
|
|
|
WARNF("%s failed to write %#s due to %m", addrstr, origname);
|
|
|
|
close(exefd);
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
2023-09-12 15:58:57 +00:00
|
|
|
if (write(exefd, exedata, filesize) != filesize) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
WARNF("%s failed to write %#s due to %m", addrstr, origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
close(exefd);
|
|
|
|
pthread_exit(0);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
if (close(exefd)) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
WARNF("%s failed to close %#s due to %m", addrstr, origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
|
|
|
|
// do the args
|
|
|
|
int i = 0;
|
|
|
|
char *args[8] = {0};
|
2023-09-12 15:58:57 +00:00
|
|
|
args[i++] = client->tmpexepath;
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
if (use_strace)
|
|
|
|
args[i++] = "--strace";
|
|
|
|
if (use_ftrace)
|
|
|
|
args[i++] = "--ftrace";
|
2020-06-15 14:18:57 +00:00
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
// run program, tee'ing stderr to both log and client
|
2023-09-12 15:58:57 +00:00
|
|
|
DEBUF("spawning %s", client->tmpexepath);
|
2023-08-08 03:22:49 +00:00
|
|
|
sigemptyset(&sigmask);
|
|
|
|
sigaddset(&sigmask, SIGINT);
|
|
|
|
sigaddset(&sigmask, SIGCHLD);
|
|
|
|
sigprocmask(SIG_BLOCK, &sigmask, 0);
|
2023-09-10 15:12:43 +00:00
|
|
|
|
|
|
|
// spawn the program
|
|
|
|
int etxtbsy_tries = 0;
|
|
|
|
RetryOnEtxtbsyRaceCondition:
|
2023-09-10 18:52:03 +00:00
|
|
|
Close(&client->pipe[1]);
|
|
|
|
Close(&client->pipe[0]);
|
2023-09-10 15:12:43 +00:00
|
|
|
if (etxtbsy_tries++) {
|
|
|
|
if (etxtbsy_tries == 24) { // ~30 seconds
|
|
|
|
WARNF("%s failed to spawn on %s due because either (1) the ETXTBSY race "
|
|
|
|
"condition kept happening or (2) the program in question actually "
|
|
|
|
"is crashing with SIGVTALRM, without printing anything to out/err!",
|
2023-09-12 15:58:57 +00:00
|
|
|
origname, g_hostname);
|
2023-09-10 15:12:43 +00:00
|
|
|
pthread_exit(0);
|
2023-06-04 15:19:45 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
if (usleep(1u << etxtbsy_tries)) {
|
|
|
|
INFOF("interrupted exponential spawn backoff");
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
errno_t err;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
struct timespec started;
|
2023-09-10 15:12:43 +00:00
|
|
|
posix_spawnattr_t spawnattr;
|
|
|
|
posix_spawn_file_actions_t spawnfila;
|
|
|
|
sigemptyset(&sigmask);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
started = timespec_real();
|
2023-09-10 15:12:43 +00:00
|
|
|
pipe2(client->pipe, O_CLOEXEC);
|
|
|
|
posix_spawnattr_init(&spawnattr);
|
2023-11-07 00:38:44 +00:00
|
|
|
posix_spawnattr_setflags(&spawnattr,
|
|
|
|
POSIX_SPAWN_SETSIGMASK | POSIX_SPAWN_SETPGROUP);
|
2023-09-10 15:12:43 +00:00
|
|
|
posix_spawnattr_setsigmask(&spawnattr, &sigmask);
|
|
|
|
posix_spawn_file_actions_init(&spawnfila);
|
|
|
|
posix_spawn_file_actions_adddup2(&spawnfila, g_bogusfd, 0);
|
|
|
|
posix_spawn_file_actions_adddup2(&spawnfila, client->pipe[1], 1);
|
|
|
|
posix_spawn_file_actions_adddup2(&spawnfila, client->pipe[1], 2);
|
2023-09-12 15:58:57 +00:00
|
|
|
err = posix_spawn(&client->pid, client->tmpexepath, &spawnfila, &spawnattr,
|
|
|
|
args, environ);
|
2023-09-10 15:12:43 +00:00
|
|
|
if (err) {
|
|
|
|
if (err == ETXTBSY) {
|
|
|
|
goto RetryOnEtxtbsyRaceCondition;
|
|
|
|
}
|
2023-09-12 15:58:57 +00:00
|
|
|
WARNF("%s failed to spawn on %s due to %s", client->tmpexepath, g_hostname,
|
2023-09-10 15:12:43 +00:00
|
|
|
strerror(err));
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
posix_spawn_file_actions_destroy(&spawnfila);
|
|
|
|
posix_spawnattr_destroy(&spawnattr);
|
|
|
|
Close(&client->pipe[1]);
|
|
|
|
|
2023-09-12 15:58:57 +00:00
|
|
|
DEBUF("communicating %s[%d]", origname, client->pid);
|
2023-09-10 15:12:43 +00:00
|
|
|
struct timespec deadline =
|
2023-08-08 03:22:49 +00:00
|
|
|
timespec_add(timespec_real(), timespec_fromseconds(DEATH_CLOCK_SECONDS));
|
2021-06-24 19:31:26 +00:00
|
|
|
for (;;) {
|
2023-09-10 15:12:43 +00:00
|
|
|
if (g_interrupted) {
|
|
|
|
WARNF("killing %d %s and hanging up %d due to interrupt", client->fd,
|
2023-09-12 15:58:57 +00:00
|
|
|
origname, client->pid);
|
2023-09-10 15:12:43 +00:00
|
|
|
HangupClientAndTerminateJob:
|
2023-09-11 20:51:37 +00:00
|
|
|
SendProgramOutput(client);
|
2023-09-10 15:12:43 +00:00
|
|
|
mbedtls_ssl_close_notify(&ezssl);
|
2023-08-08 03:22:49 +00:00
|
|
|
TerminateJob:
|
2023-09-10 15:12:43 +00:00
|
|
|
PrintProgramOutput(client);
|
|
|
|
pthread_exit(0);
|
|
|
|
}
|
|
|
|
struct timespec now = timespec_real();
|
|
|
|
if (timespec_cmp(now, deadline) >= 0) {
|
2023-09-12 15:58:57 +00:00
|
|
|
WARNF("killing %s (pid %d) which timed out after %d seconds", origname,
|
2023-09-10 15:12:43 +00:00
|
|
|
client->pid, DEATH_CLOCK_SECONDS);
|
|
|
|
goto HangupClientAndTerminateJob;
|
2022-04-20 16:56:53 +00:00
|
|
|
}
|
|
|
|
struct pollfd fds[2];
|
2023-09-10 15:12:43 +00:00
|
|
|
fds[0].fd = client->fd;
|
2022-04-20 16:56:53 +00:00
|
|
|
fds[0].events = POLLIN;
|
2023-09-10 15:12:43 +00:00
|
|
|
fds[1].fd = client->pipe[0];
|
2022-04-20 16:56:53 +00:00
|
|
|
fds[1].events = POLLIN;
|
2023-09-10 15:12:43 +00:00
|
|
|
events = poll(fds, ARRAYLEN(fds),
|
|
|
|
timespec_tomillis(timespec_sub(deadline, now)));
|
|
|
|
if (events == -1) {
|
|
|
|
if (errno == EINTR) {
|
|
|
|
INFOF("poll interrupted");
|
|
|
|
continue;
|
|
|
|
} else {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
WARNF("killing %d %s and hanging up %d because poll failed with %m",
|
|
|
|
client->fd, origname, client->pid);
|
2023-09-10 15:12:43 +00:00
|
|
|
goto HangupClientAndTerminateJob;
|
|
|
|
}
|
|
|
|
}
|
2023-08-08 03:22:49 +00:00
|
|
|
if (events) {
|
|
|
|
if (fds[0].revents) {
|
2023-08-08 11:00:29 +00:00
|
|
|
int received;
|
|
|
|
char buf[512];
|
|
|
|
received = mbedtls_ssl_read(&ezssl, buf, sizeof(buf));
|
|
|
|
if (!received) {
|
2023-09-12 15:58:57 +00:00
|
|
|
WARNF("%s client disconnected so killing worker %d", origname,
|
2023-09-10 15:12:43 +00:00
|
|
|
client->pid);
|
2023-08-08 11:00:29 +00:00
|
|
|
goto TerminateJob;
|
2023-08-08 03:22:49 +00:00
|
|
|
}
|
2023-08-08 11:00:29 +00:00
|
|
|
if (received > 0) {
|
2023-10-14 08:06:00 +00:00
|
|
|
WARNF("%s client sent %d unexpected bytes", origname, received);
|
|
|
|
continue;
|
2023-08-08 11:00:29 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
if (received == MBEDTLS_ERR_SSL_WANT_READ) { // EAGAIN SO_RCVTIMEO
|
2023-09-12 15:58:57 +00:00
|
|
|
WARNF("%s (pid %d) is taking a really long time", origname,
|
2023-09-10 15:12:43 +00:00
|
|
|
client->pid);
|
|
|
|
continue;
|
2023-08-08 11:00:29 +00:00
|
|
|
}
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
if (received == MBEDTLS_ERR_SSL_CANCELED) { // EAGAIN SO_RCVTIMEO
|
|
|
|
WARNF("%s (pid %d) is is canceling job", origname, client->pid);
|
|
|
|
goto HangupClientAndTerminateJob;
|
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
WARNF("client ssl read failed with -0x%04x (%s) so killing %s",
|
2023-09-12 15:58:57 +00:00
|
|
|
-received, GetTlsError(received), origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
goto TerminateJob;
|
2023-08-08 03:22:49 +00:00
|
|
|
}
|
|
|
|
if (fds[1].revents) {
|
2023-09-10 15:12:43 +00:00
|
|
|
char buf[512];
|
|
|
|
ssize_t got = read(client->pipe[0], buf, sizeof(buf));
|
|
|
|
if (got == -1) {
|
2023-09-12 15:58:57 +00:00
|
|
|
WARNF("got %s reading %s output", strerror(errno), origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
goto HangupClientAndTerminateJob;
|
|
|
|
}
|
2023-08-08 03:22:49 +00:00
|
|
|
if (!got) {
|
2023-09-12 15:58:57 +00:00
|
|
|
VERBF("got eof reading %s output", origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
Close(&client->pipe[0]);
|
2023-08-08 03:22:49 +00:00
|
|
|
break;
|
|
|
|
}
|
2023-09-12 15:58:57 +00:00
|
|
|
DEBUF("got %ld bytes reading %s output", got, origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
appendd(&client->output, buf, got);
|
2022-04-20 16:56:53 +00:00
|
|
|
}
|
2021-02-05 20:19:43 +00:00
|
|
|
}
|
2021-01-25 21:08:05 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
WaitAgain:
|
|
|
|
DEBUF("waitpid");
|
|
|
|
struct rusage rusage;
|
|
|
|
int wrc = wait4(client->pid, &wstatus, 0, &rusage);
|
|
|
|
if (wrc == -1) {
|
|
|
|
if (errno == EINTR) {
|
2023-09-12 15:58:57 +00:00
|
|
|
WARNF("waitpid interrupted; killing %s pid %d", origname, client->pid);
|
2023-09-10 15:12:43 +00:00
|
|
|
kill(client->pid, SIGINT);
|
|
|
|
goto WaitAgain;
|
|
|
|
}
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
if (errno == ECANCELED) {
|
|
|
|
WARNF("thread is canceled; killing %s pid %d", origname, client->pid);
|
|
|
|
kill(client->pid, SIGKILL);
|
|
|
|
goto WaitAgain;
|
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
WARNF("waitpid failed %m");
|
|
|
|
client->pid = 0;
|
|
|
|
goto HangupClientAndTerminateJob;
|
|
|
|
}
|
|
|
|
client->pid = 0;
|
|
|
|
int exitcode;
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
struct timespec ended = timespec_real();
|
|
|
|
int64_t micros = timespec_tomicros(timespec_sub(ended, started));
|
2021-01-28 03:34:02 +00:00
|
|
|
if (WIFEXITED(wstatus)) {
|
2021-08-26 04:35:58 +00:00
|
|
|
if (WEXITSTATUS(wstatus)) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
WARNF("%s on %s exited with $?=%d after %'ldµs", origname, g_hostname,
|
|
|
|
WEXITSTATUS(wstatus), micros);
|
|
|
|
appendf(&client->output, "------ %s %s $?=%d (0x%08x) %,ldµs ------\n",
|
|
|
|
g_hostname, origname, WEXITSTATUS(wstatus), wstatus, micros);
|
2021-08-26 04:35:58 +00:00
|
|
|
} else {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
INFOF("%s on %s exited with $?=%d after %'ldµs", origname, g_hostname,
|
|
|
|
WEXITSTATUS(wstatus), micros);
|
2021-08-26 04:35:58 +00:00
|
|
|
}
|
2021-01-28 03:34:02 +00:00
|
|
|
exitcode = WEXITSTATUS(wstatus);
|
2023-09-10 15:12:43 +00:00
|
|
|
} else if (WIFSIGNALED(wstatus)) {
|
|
|
|
if (WTERMSIG(wstatus) == SIGVTALRM && !client->output) {
|
|
|
|
free(client->output);
|
|
|
|
client->output = 0;
|
|
|
|
goto RetryOnEtxtbsyRaceCondition;
|
|
|
|
}
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
char sigbuf[21];
|
|
|
|
WARNF("%s on %s terminated after %'ldµs with %s", origname, g_hostname,
|
|
|
|
micros, strsignal_r(WTERMSIG(wstatus), sigbuf));
|
2021-01-28 03:34:02 +00:00
|
|
|
exitcode = 128 + WTERMSIG(wstatus);
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
appendf(&client->output, "------ %s %s $?=%s (0x%08x) %,ldµs ------\n",
|
|
|
|
g_hostname, origname, strsignal(WTERMSIG(wstatus)), wstatus,
|
|
|
|
micros);
|
2023-09-10 15:12:43 +00:00
|
|
|
} else {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
WARNF("%s on %s died after %'ldµs with wait status 0x%08x", origname,
|
|
|
|
g_hostname, micros, wstatus);
|
2023-09-10 15:12:43 +00:00
|
|
|
exitcode = 127;
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
if (wstatus) {
|
|
|
|
AppendResourceReport(&client->output, &rusage, "\n");
|
|
|
|
PrintProgramOutput(client);
|
|
|
|
}
|
2023-09-11 20:51:37 +00:00
|
|
|
SendProgramOutput(client);
|
2021-08-07 20:22:35 +00:00
|
|
|
SendExitMessage(exitcode);
|
|
|
|
mbedtls_ssl_close_notify(&ezssl);
|
2023-09-12 15:58:57 +00:00
|
|
|
if (etxtbsy_tries > 1) {
|
|
|
|
WARNF("encountered %d ETXTBSY race conditions spawning %s",
|
|
|
|
etxtbsy_tries - 1, origname);
|
2023-09-10 15:12:43 +00:00
|
|
|
}
|
|
|
|
pthread_exit(0);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
|
2023-09-10 15:12:43 +00:00
|
|
|
void HandleClient(void) {
|
2023-09-12 04:34:53 +00:00
|
|
|
struct stat st;
|
2023-09-10 15:12:43 +00:00
|
|
|
struct Client *client;
|
|
|
|
client = calloc(1, sizeof(struct Client));
|
|
|
|
client->addrsize = sizeof(client->addr);
|
|
|
|
for (;;) {
|
|
|
|
if (g_interrupted) {
|
Make improvements
- Every unit test now passes on Apple Silicon. The final piece of this
puzzle was porting our POSIX threads cancelation support, since that
works differently on ARM64 XNU vs. AMD64. Our semaphore support on
Apple Silicon is also superior now compared to AMD64, thanks to the
grand central dispatch library which lets *NSYNC locks go faster.
- The Cosmopolitan runtime is now more stable, particularly on Windows.
To do this, thread local storage is mandatory at all runtime levels,
and the innermost packages of the C library is no longer being built
using ASAN. TLS is being bootstrapped with a 128-byte TIB during the
process startup phase, and then later on the runtime re-allocates it
either statically or dynamically to support code using _Thread_local.
fork() and execve() now do a better job cooperating with threads. We
can now check how much stack memory is left in the process or thread
when functions like kprintf() / execve() etc. call alloca(), so that
ENOMEM can be raised, reduce a buffer size, or just print a warning.
- POSIX signal emulation is now implemented the same way kernels do it
with pthread_kill() and raise(). Any thread can interrupt any other
thread, regardless of what it's doing. If it's blocked on read/write
then the killer thread will cancel its i/o operation so that EINTR can
be returned in the mark thread immediately. If it's doing a tight CPU
bound operation, then that's also interrupted by the signal delivery.
Signal delivery works now by suspending a thread and pushing context
data structures onto its stack, and redirecting its execution to a
trampoline function, which calls SetThreadContext(GetCurrentThread())
when it's done.
- We're now doing a better job managing locks and handles. On NetBSD we
now close semaphore file descriptors in forked children. Semaphores on
Windows can now be canceled immediately, which means mutexes/condition
variables will now go faster. Apple Silicon semaphores can be canceled
too. We're now using Apple's pthread_yield() funciton. Apple _nocancel
syscalls are now used on XNU when appropriate to ensure pthread_cancel
requests aren't lost. The MbedTLS library has been updated to support
POSIX thread cancelations. See tool/build/runitd.c for an example of
how it can be used for production multi-threaded tls servers. Handles
on Windows now leak less often across processes. All i/o operations on
Windows are now overlapped, which means file pointers can no longer be
inherited across dup() and fork() for the time being.
- We now spawn a thread on Windows to deliver SIGCHLD and wakeup wait4()
which means, for example, that posix_spawn() now goes 3x faster. POSIX
spawn is also now more correct. Like Musl, it's now able to report the
failure code of execve() via a pipe although our approach favors using
shared memory to do that on systems that have a true vfork() function.
- We now spawn a thread to deliver SIGALRM to threads when setitimer()
is used. This enables the most precise wakeups the OS makes possible.
- The Cosmopolitan runtime now uses less memory. On NetBSD for example,
it turned out the kernel would actually commit the PT_GNU_STACK size
which caused RSS to be 6mb for every process. Now it's down to ~4kb.
On Apple Silicon, we reduce the mandatory upstream thread size to the
smallest possible size to reduce the memory overhead of Cosmo threads.
The examples directory has a program called greenbean which can spawn
a web server on Linux with 10,000 worker threads and have the memory
usage of the process be ~77mb. The 1024 byte overhead of POSIX-style
thread-local storage is now optional; it won't be allocated until the
pthread_setspecific/getspecific functions are called. On Windows, the
threads that get spawned which are internal to the libc implementation
use reserve rather than commit memory, which shaves a few hundred kb.
- sigaltstack() is now supported on Windows, however it's currently not
able to be used to handle stack overflows, since crash signals are
still generated by WIN32. However the crash handler will still switch
to the alt stack, which is helpful in environments with tiny threads.
- Test binaries are now smaller. Many of the mandatory dependencies of
the test runner have been removed. This ensures many programs can do a
better job only linking the the thing they're testing. This caused the
test binaries for LIBC_FMT for example, to decrease from 200kb to 50kb
- long double is no longer used in the implementation details of libc,
except in the APIs that define it. The old code that used long double
for time (instead of struct timespec) has now been thoroughly removed.
- ShowCrashReports() is now much tinier in MODE=tiny. Instead of doing
backtraces itself, it'll just print a command you can run on the shell
using our new `cosmoaddr2line` program to view the backtrace.
- Crash report signal handling now works in a much better way. Instead
of terminating the process, it now relies on SA_RESETHAND so that the
default SIG_IGN behavior can terminate the process if necessary.
- Our pledge() functionality has now been fully ported to AARCH64 Linux.
2023-09-19 03:44:45 +00:00
|
|
|
pthread_cancel(0);
|
2023-09-10 15:12:43 +00:00
|
|
|
free(client);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
// poll() because we use SA_RESTART and accept() is @restartable
|
|
|
|
if (poll(&(struct pollfd){g_servfd, POLLIN}, 1, -1) > 0) {
|
|
|
|
client->fd = accept4(g_servfd, (struct sockaddr *)&client->addr,
|
|
|
|
&client->addrsize, SOCK_CLOEXEC);
|
|
|
|
if (client->fd != -1) {
|
|
|
|
VERBF("accepted client fd %d", client->fd);
|
|
|
|
break;
|
|
|
|
} else if (errno != EINTR && errno != EAGAIN) {
|
|
|
|
WARNF("accept4 failed %m");
|
|
|
|
}
|
|
|
|
} else if (errno != EINTR && errno != EAGAIN) {
|
|
|
|
WARNF("poll failed %m");
|
|
|
|
}
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
2023-09-12 04:34:53 +00:00
|
|
|
if (fstat(2, &st) != -1 && st.st_size > kLogMaxBytes) {
|
|
|
|
ftruncate(2, 0); // auto rotate log
|
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
sigset_t mask;
|
|
|
|
pthread_attr_t attr;
|
|
|
|
sigfillset(&mask);
|
|
|
|
pthread_attr_init(&attr);
|
|
|
|
pthread_attr_setsigmask_np(&attr, &mask);
|
|
|
|
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
|
|
|
|
pthread_create(&client->th, &attr, ClientWorker, client);
|
|
|
|
pthread_attr_destroy(&attr);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int Serve(void) {
|
2023-09-10 15:12:43 +00:00
|
|
|
sigset_t mask;
|
2020-06-15 14:18:57 +00:00
|
|
|
StartTcpServer();
|
2023-09-10 15:12:43 +00:00
|
|
|
sigemptyset(&mask);
|
|
|
|
sigaddset(&mask, SIGCHLD);
|
|
|
|
signal(SIGINT, OnInterrupt);
|
|
|
|
sigprocmask(SIG_BLOCK, &mask, 0);
|
|
|
|
while (!g_interrupted) {
|
|
|
|
HandleClient();
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
if (g_interrupted) {
|
|
|
|
WARNF("got ctrl-c, shutting down...");
|
2021-01-29 09:27:09 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
WARNF("server exiting");
|
|
|
|
close(g_servfd);
|
2021-01-28 03:34:02 +00:00
|
|
|
return 0;
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void Daemonize(void) {
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
if (fork() > 0)
|
|
|
|
_exit(0);
|
2020-06-15 14:18:57 +00:00
|
|
|
setsid();
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
if (fork() > 0)
|
|
|
|
_exit(0);
|
2023-09-12 04:34:53 +00:00
|
|
|
dup2(g_bogusfd, 0);
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
if (!g_sendready)
|
|
|
|
dup2(g_bogusfd, 1);
|
2023-09-12 04:34:53 +00:00
|
|
|
close(2);
|
|
|
|
open(kLogFile, O_CREAT | O_WRONLY | O_APPEND | O_CLOEXEC, 0644);
|
|
|
|
extern long __klog_handle;
|
|
|
|
if (__klog_handle > 0) {
|
|
|
|
close(__klog_handle);
|
|
|
|
}
|
|
|
|
__klog_handle = 2;
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int main(int argc, char *argv[]) {
|
|
|
|
GetOpts(argc, argv);
|
2023-09-10 15:12:43 +00:00
|
|
|
g_psk = GetRunitPsk();
|
|
|
|
signal(SIGPIPE, SIG_IGN);
|
|
|
|
setenv("TZ", "PST", true);
|
|
|
|
gethostname(g_hostname, sizeof(g_hostname));
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
for (int i = 3; i < 16; ++i)
|
|
|
|
close(i);
|
2022-04-29 13:06:23 +00:00
|
|
|
errno = 0;
|
2022-04-16 17:40:23 +00:00
|
|
|
// poll()'ing /dev/null stdin file descriptor on xnu returns POLLNVAL?!
|
|
|
|
if (IsWindows()) {
|
2023-09-10 15:12:43 +00:00
|
|
|
g_bogusfd = open("/dev/null", O_RDONLY | O_CLOEXEC);
|
2022-04-16 17:40:23 +00:00
|
|
|
} else {
|
2023-09-10 15:12:43 +00:00
|
|
|
g_bogusfd = open("/dev/zero", O_RDONLY | O_CLOEXEC);
|
2022-04-16 17:40:23 +00:00
|
|
|
}
|
2023-09-10 15:12:43 +00:00
|
|
|
mkdir("o", 0700);
|
Apply clang-format update to repo (#1154)
Commit bc6c183 introduced a bunch of discrepancies between what files
look like in the repo and what clang-format says they should look like.
However, there were already a few discrepancies prior to that. Most of
these discrepancies seemed to be unintentional, but a few of them were
load-bearing (e.g., a #include that violated header ordering needing
something to have been #defined by a 'later' #include.)
I opted to take what I hope is a relatively smooth-brained approach: I
reverted the .clang-format change, ran clang-format on the whole repo,
reapplied the .clang-format change, reran clang-format again, and then
reverted the commit that contained the first run. Thus the full effect
of this PR should only be to apply the changed formatting rules to the
repo, and from skimming the results, this seems to be the case.
My work can be checked by applying the short, manual commits, and then
rerunning the command listed in the autogenerated commits (those whose
messages I have prefixed auto:) and seeing if your results agree.
It might be that the other diffs should be fixed at some point but I'm
leaving that aside for now.
fd '\.c(c|pp)?$' --print0| xargs -0 clang-format -i
2024-04-25 17:38:00 +00:00
|
|
|
if (g_daemonize)
|
|
|
|
Daemonize();
|
2023-09-10 15:12:43 +00:00
|
|
|
Serve();
|
|
|
|
free(g_psk);
|
2023-11-19 00:56:11 +00:00
|
|
|
#ifdef MODE_DBG
|
2023-09-10 15:12:43 +00:00
|
|
|
CheckForMemoryLeaks();
|
2023-09-10 18:52:03 +00:00
|
|
|
CheckForFileLeaks();
|
2023-09-10 15:12:43 +00:00
|
|
|
#endif
|
|
|
|
pthread_exit(0);
|
2020-06-15 14:18:57 +00:00
|
|
|
}
|