os2: process API for Darwin and most of it for BSDs

This commit is contained in:
Laytan Laats
2024-08-22 23:07:24 +02:00
parent b9043db434
commit a66520ba57
6 changed files with 587 additions and 121 deletions
+7 -4
View File
@@ -576,10 +576,13 @@ _process_start :: proc(desc: Process_Desc) -> (process: Process, err: Error) {
success_byte: [1]u8
linux.write(child_pipe_fds[WRITE], success_byte[:])
if errno = linux.execveat(exe_fd, "", &cargs[0], env, {.AT_EMPTY_PATH}); errno != .NONE {
write_errno_to_parent_and_abort(child_pipe_fds[WRITE], errno)
}
unreachable()
errno = linux.execveat(exe_fd, "", &cargs[0], env, {.AT_EMPTY_PATH})
// NOTE: we can't tell the parent about this failure because we already wrote the success byte.
// So if this happens the user will just see the process failed when they call process_wait.
assert(errno != nil)
intrinsics.trap()
}
process.pid = int(pid)
+243 -97
View File
@@ -3,9 +3,13 @@
package os2
import "base:runtime"
import "core:time"
import "core:sys/posix"
import "core:time"
import "core:strings"
import "core:path/filepath"
import kq "core:sys/kqueue"
import "core:sys/posix"
_exit :: proc "contextless" (code: int) -> ! {
posix.exit(i32(code))
@@ -51,141 +55,286 @@ _process_start :: proc(desc: Process_Desc) -> (process: Process, err: Error) {
return
}
TEMP_ALLOCATOR_GUARD()
// search PATH if just a plain name is provided.
exe_builder := strings.builder_make(temp_allocator())
exe_name := desc.command[0]
if strings.index_byte(exe_name, '/') < 0 {
path_env := get_env("PATH", temp_allocator())
path_dirs := filepath.split_list(path_env, temp_allocator())
found: bool
for dir in path_dirs {
strings.builder_reset(&exe_builder)
strings.write_string(&exe_builder, dir)
strings.write_byte(&exe_builder, '/')
strings.write_string(&exe_builder, exe_name)
if exe_fd := posix.open(strings.to_cstring(&exe_builder), {.CLOEXEC, .EXEC}); exe_fd == -1 {
continue
} else {
posix.close(exe_fd)
found = true
break
}
}
if !found {
// check in cwd to match windows behavior
strings.builder_reset(&exe_builder)
strings.write_string(&exe_builder, desc.working_dir)
if len(desc.working_dir) > 0 && desc.working_dir[len(desc.working_dir)-1] != '/' {
strings.write_byte(&exe_builder, '/')
}
strings.write_string(&exe_builder, "./")
strings.write_string(&exe_builder, exe_name)
// "hello/./world" is fine right?
if exe_fd := posix.open(strings.to_cstring(&exe_builder), {.CLOEXEC, .EXEC}); exe_fd == -1 {
err = .Not_Exist
return
} else {
posix.close(exe_fd)
}
}
} else {
strings.builder_reset(&exe_builder)
strings.write_string(&exe_builder, exe_name)
if exe_fd := posix.open(strings.to_cstring(&exe_builder), {.CLOEXEC, .EXEC}); exe_fd == -1 {
err = .Not_Exist
return
} else {
posix.close(exe_fd)
}
}
cwd: cstring; if desc.working_dir != "" {
cwd = temp_cstring(desc.working_dir)
}
cmd := make([]cstring, len(desc.command)+1, temp_allocator())
cmd := make([]cstring, len(desc.command) + 1, temp_allocator())
for part, i in desc.command {
cmd[i] = temp_cstring(part)
}
env: [^]cstring
if desc.env == nil {
// take this process's current environment
env = posix.environ
} else {
cenv := make([]cstring, len(desc.env) + 1, temp_allocator())
for env, i in desc.env {
cenv[i] = temp_cstring(env)
}
env = raw_data(cenv)
}
READ :: 0
WRITE :: 1
pipe: [2]posix.FD
if posix.pipe(&pipe) != .OK {
err = _get_platform_error()
return
}
defer posix.close(pipe[WRITE])
defer posix.close(pipe[READ])
if posix.fcntl(pipe[READ], .SETFD, i32(posix.FD_CLOEXEC)) == -1 {
err = _get_platform_error()
return
}
if posix.fcntl(pipe[WRITE], .SETFD, i32(posix.FD_CLOEXEC)) == -1 {
err = _get_platform_error()
return
}
switch pid := posix.fork(); pid {
case -1:
err = _get_platform_error()
return
case 0:
// NOTE(laytan): would need to use execvp and look up the command in the PATH.
assert(len(desc.env) == 0, "unimplemented: process_start with env")
abort :: proc(parent_fd: posix.FD) -> ! {
#assert(len(posix.Errno) < max(u8))
errno := u8(posix.errno())
posix.write(parent_fd, &errno, 1)
runtime.trap()
}
null := posix.open("/dev/null", { .RDWR, .CLOEXEC })
assert(null != -1) // TODO: Does this happen/need to be handled?
null := posix.open("/dev/null", {.RDWR})
if null == -1 { abort(pipe[WRITE]) }
stderr := (^File_Impl)(desc.stderr.impl).fd if desc.stderr != nil else null
stdout := (^File_Impl)(desc.stdout.impl).fd if desc.stdout != nil else null
stdin := (^File_Impl)(desc.stdin.impl).fd if desc.stdin != nil else null
posix.dup2(stderr, posix.STDERR_FILENO)
posix.dup2(stdout, posix.STDOUT_FILENO)
posix.dup2(stdin, posix.STDIN_FILENO )
// NOTE(laytan): is this how we should handle these?
// Maybe we can try to `stat` the cwd in the parent before forking?
// Does that mean no other errors could happen in chdir?
// How about execvp?
if posix.dup2(stderr, posix.STDERR_FILENO) == -1 { abort(pipe[WRITE]) }
if posix.dup2(stdout, posix.STDOUT_FILENO) == -1 { abort(pipe[WRITE]) }
if posix.dup2(stdin, posix.STDIN_FILENO ) == -1 { abort(pipe[WRITE]) }
if cwd != nil {
if posix.chdir(cwd) != .OK {
posix.exit(i32(posix.errno())) // TODO: handle, or is it fine this way?
}
if posix.chdir(cwd) != .OK { abort(pipe[WRITE]) }
}
posix.execvp(cmd[0], raw_data(cmd))
posix.exit(i32(posix.errno())) // TODO: handle, or is it fine this way?
ok := u8(0)
posix.write(pipe[WRITE], &ok, 1)
res := posix.execve(strings.to_cstring(&exe_builder), raw_data(cmd), env)
// NOTE: we can't tell the parent about this failure because we already wrote the success byte.
// So if this happens the user will just see the process failed when they call process_wait.
assert(res == -1)
runtime.trap()
case:
fmt.println("returning")
process, _ = _process_open(int(pid), {})
errno: posix.Errno
for {
errno_byte: u8
switch posix.read(pipe[READ], &errno_byte, 1) {
case 1:
errno = posix.Errno(errno_byte)
case:
errno = posix.errno()
if errno == .EINTR {
continue
} else {
// If the read failed, something weird happened. Do not return the read
// error so the user knows to wait on it.
errno = nil
}
}
break
}
if errno != nil {
// We can assume it trapped here.
for {
info: posix.siginfo_t
wpid := posix.waitid(.P_PID, posix.id_t(process.pid), &info, {.EXITED})
if wpid == -1 && posix.errno() == .EINTR {
continue
}
break
}
err = errno
return
}
process.pid = int(pid)
process, _ = _process_open(int(pid), {})
return
}
}
import "core:fmt"
import "core:nbio/kqueue"
_process_wait :: proc(process: Process, timeout: time.Duration) -> (process_state: Process_State, err: Error) {
process_state.pid = process.pid
if !process_posix_handle_still_valid(process) {
err = Platform_Error(posix.Errno.ESRCH)
return
}
_process_handle_still_valid(process) or_return
// prev := posix.signal(.SIGALRM, proc "c" (_: posix.Signal) {
// context = runtime.default_context()
// fmt.println("alarm")
// })
// defer posix.signal(.SIGALRM, prev)
// timeout > 0 = use kqueue to wait (with a timeout) on process exit
// timeout == 0 = use waitid with WNOHANG so it returns immediately
// timeout > 0 = use waitid without WNOHANG so it waits indefinitely
//
// posix.alarm(u32(time.duration_seconds(timeout)))
// defer posix.alarm(0)
// at the end use waitid to actually reap the process and get it's status
// TODO: if there's no timeout, don't set up a kqueue.
if timeout > 0 {
timeout := timeout
// TODO: if timeout is 0, don't set up a kqueue and use NO_HANG.
queue := kq.kqueue() or_return
defer posix.close(queue)
kq, qerr := kqueue.kqueue()
if qerr != nil {
err = Platform_Error(qerr)
return
changelist, eventlist: [1]kq.KEvent
changelist[0] = {
ident = uintptr(process.pid),
filter = .Proc,
flags = { .Add },
fflags = {
fproc = { .Exit },
},
}
for {
start := time.tick_now()
n, kerr := kq.kevent(queue, changelist[:], eventlist[:], &{
tv_sec = posix.time_t(timeout / time.Second),
tv_nsec = i64(timeout % time.Second),
})
if kerr == .EINTR {
timeout -= time.tick_since(start)
continue
} else if kerr != nil {
err = kerr
return
} else if n == 0 {
err = .Timeout
_process_state_update_times(process, &process_state)
return
} else {
_process_state_update_times(process, &process_state)
break
}
}
} else {
flags := posix.Wait_Flags{.EXITED, .NOWAIT}
if timeout == 0 {
flags += {.NOHANG}
}
info: posix.siginfo_t
for {
wpid := posix.waitid(.P_PID, posix.id_t(process.pid), &info, flags)
if wpid == -1 {
if errno := posix.errno(); errno == .EINTR {
continue
} else {
err = _get_platform_error()
return
}
}
break
}
_process_state_update_times(process, &process_state)
if info.si_signo == nil {
assert(timeout == 0)
err = .Timeout
return
}
}
changelist, eventlist: [1]kqueue.KEvent
changelist[0] = {
ident = uintptr(process.pid),
filter = .Proc,
flags = { .Add },
fflags = {
fproc = 0x80000000,
},
info: posix.siginfo_t
for {
wpid := posix.waitid(.P_PID, posix.id_t(process.pid), &info, {.EXITED})
if wpid == -1 {
if errno := posix.errno(); errno == .EINTR {
continue
} else {
err = _get_platform_error()
return
}
}
break
}
// NOTE: could this be interrupted which means it should be looped and subtracting the timeout on EINTR.
n, eerr := kqueue.kevent(kq, changelist[:], eventlist[:], &{
seconds = i64(timeout / time.Second),
nanoseconds = i64(timeout % time.Second),
})
if eerr != nil {
err = Platform_Error(eerr)
return
}
if n == 0 {
err = .Timeout
// TODO: populate the time fields.
return
}
// NOTE(laytan): should this be looped untill WIFEXITED/WIFSIGNALED?
status: i32
wpid := posix.waitpid(posix.pid_t(process.pid), &status, {})
if wpid == -1 {
err = _get_platform_error()
return
}
process_state.exited = true
// TODO: populate times
switch {
case posix.WIFEXITED(status):
fmt.printfln("child exited, status=%v", posix.WEXITSTATUS(status))
process_state.exit_code = int(posix.WEXITSTATUS(status))
process_state.success = true
case posix.WIFSIGNALED(status):
fmt.printfln("child killed (signal %v)", posix.WTERMSIG(status))
process_state.exit_code = int(posix.WTERMSIG(status))
switch info.si_code.chld {
case: unreachable()
case .CONTINUED, .STOPPED: unreachable()
case .EXITED:
process_state.exited = true
process_state.exit_code = int(info.si_status)
process_state.success = process_state.exit_code == 0
case .KILLED, .DUMPED, .TRAPPED:
process_state.exited = true
process_state.exit_code = int(info.si_status)
process_state.success = false
case:
fmt.panicf("unexpected status (%x)", status)
}
return
@@ -196,10 +345,7 @@ _process_close :: proc(process: Process) -> Error {
}
_process_kill :: proc(process: Process) -> (err: Error) {
if !process_posix_handle_still_valid(process) {
err = Platform_Error(posix.Errno.ESRCH)
return
}
_process_handle_still_valid(process) or_return
if posix.kill(posix.pid_t(process.pid), .SIGKILL) != .OK {
err = _get_platform_error()
+44 -20
View File
@@ -8,6 +8,7 @@ import "core:bytes"
import "core:sys/darwin"
import "core:sys/posix"
import "core:sys/unix"
import "core:time"
foreign import lib "system:System.framework"
@@ -19,8 +20,6 @@ foreign lib {
) -> posix.result ---
}
import "core:fmt"
_process_info_by_pid :: proc(pid: int, selection: Process_Info_Fields, allocator: runtime.Allocator) -> (info: Process_Info, err: Error) {
get_pidinfo :: proc(pid: int, selection: Process_Info_Fields) -> (ppid: u32, prio: Maybe(i32), uid: posix.uid_t, ok: bool) {
// Short info is enough and requires less permissions if the priority isn't requested.
@@ -258,31 +257,56 @@ _process_list :: proc(allocator: runtime.Allocator) -> (list: []int, err: Error)
}
_process_open :: proc(pid: int, flags: Process_Open_Flags) -> (process: Process, err: Error) {
// NOTE(laytan): pids can get reused, and afaik posix/macos doesn't have a unique identifier
// for a specific process execution, next best thing to me is checking the time the process
// started as some extra "uniqueness". We could also hash a bunch of the fields in this info.
// This incidentally also checks if the pid is actually valid so that's nice.
pinfo: darwin.proc_bsdinfo
ret := darwin.proc_pidinfo(posix.pid_t(pid), .BSDINFO, 0, &pinfo, size_of(pinfo))
if ret <= 0 {
rusage: darwin.rusage_info_v0
if ret := darwin.proc_pid_rusage(posix.pid_t(pid), .V0, &rusage); ret != 0 {
err = _get_platform_error()
return
}
assert(ret == size_of(pinfo))
process = { int(pid), uintptr(pinfo.pbi_start_tvusec) }
// XOR fold the UUID so it fits the handle, I think this is enough to verify pid uniqueness.
#assert(size_of(uintptr) == size_of(u64))
a := intrinsics.unaligned_load((^u64)(&rusage.ri_uuid))
b := intrinsics.unaligned_load((^u64)(&rusage.ri_uuid[8]))
process.handle = uintptr(a ~ b)
process.pid = int(pid)
return
}
process_posix_handle_still_valid :: proc(p: Process) -> bool {
pinfo: darwin.proc_bsdinfo
ret := darwin.proc_pidinfo(posix.pid_t(p.pid), .BSDINFO, 0, &pinfo, size_of(pinfo))
if ret <= 0 {
return false
_process_handle_still_valid :: proc(p: Process) -> Error {
rusage: darwin.rusage_info_v0
if ret := darwin.proc_pid_rusage(posix.pid_t(p.pid), .V0, &rusage); ret != 0 {
return _get_platform_error()
}
return uintptr(pinfo.pbi_start_tvusec) == p.handle
// XOR fold the UUID so it fits the handle, I think this is enough to verify pid uniqueness.
#assert(size_of(uintptr) == size_of(u64))
a := intrinsics.unaligned_load((^u64)(&rusage.ri_uuid))
b := intrinsics.unaligned_load((^u64)(&rusage.ri_uuid[8]))
handle := uintptr(a ~ b)
if p.handle != handle {
return posix.Errno.ESRCH
}
return nil
}
_process_state_update_times :: proc(p: Process, state: ^Process_State) {
rusage: darwin.rusage_info_v0
if ret := darwin.proc_pid_rusage(posix.pid_t(p.pid), .V0, &rusage); ret != 0 {
return
}
// NOTE(laytan): I have no clue if this is correct, the output seems correct comparing it with `time`'s output.
HZ :: 20000000
state.user_time = (
(time.Duration(rusage.ri_user_time) / HZ * time.Second) +
time.Duration(rusage.ri_user_time % HZ))
state.system_time = (
(time.Duration(rusage.ri_system_time) / HZ * time.Second) +
time.Duration(rusage.ri_system_time % HZ))
return
}
+13
View File
@@ -13,3 +13,16 @@ _process_list :: proc(allocator: runtime.Allocator) -> (list: []int, err: Error)
err = .Unsupported
return
}
_process_open :: proc(pid: int, flags: Process_Open_Flags) -> (process: Process, err: Error) {
err = .Unsupported
return
}
_process_handle_still_valid :: proc(p: Process) -> Error {
return nil
}
_process_state_update_times :: proc(p: Process, state: ^Process_State) {
return
}