Threaded version 0.7

component-source/src/atom_worker.c

@@ -0,0 +1,115 @@
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include "atom_bridge.h"
+#include "atom_mb.h"
+#include <pthread.h>
+#include <sched.h>
+#include <errno.h>
+#include <time.h>
+#include <string.h>
+
+/* All engine access is serialized. The worker never holds an IEC pointer.
+ * Network I/O is nonblocking; no mutex is held across sleep or thread join.
+ * Serial open/ioctl latency is driver dependent: this is not a hard RT API. */
+uint32_t ab_dispatch(uint32_t,uint32_t,uint32_t,uint16_t,uint8_t,uint8_t,uint16_t,uint16_t,
+ uint16_t *,uint32_t,uint32_t *,uint32_t,uint64_t,uint32_t,uint32_t,uint32_t);
+void ab_pump(uint64_t,uint32_t);
+void ab_close_all(void);
+static pthread_mutex_t aw_lifecycle=PTHREAD_MUTEX_INITIALIZER;
+static pthread_mutex_t aw_engine;
+static pthread_once_t aw_once=PTHREAD_ONCE_INIT;
+static int aw_init_error;
+static void aw_init_mutex(void){
+ pthread_mutexattr_t a;
+ int e=pthread_mutexattr_init(&a);
+ if(e){aw_init_error=e;return;}
+ /* Priority inheritance limits inversion if the PLC task has RT priority. */
+ e=pthread_mutexattr_setprotocol(&a,PTHREAD_PRIO_INHERIT);
+ if(e){pthread_mutexattr_destroy(&a);aw_init_error=e;return;}
+ aw_init_error=pthread_mutex_init(&aw_engine,&a);
+ pthread_mutexattr_destroy(&a);
+}
+static pthread_t aw_thread;
+static int aw_running,aw_stop,aw_error,aw_cpu=-1;
+static uint32_t aw_interval=1,aw_watchdog=2000;
+static uint64_t aw_cycles,aw_heartbeat;
+static cpu_set_t aw_original_mask;
+static uint64_t aw_now(void){
+ struct timespec t;clock_gettime(CLOCK_MONOTONIC,&t);
+ return (uint64_t)t.tv_sec*1000u+(uint64_t)t.tv_nsec/1000000u;
+}
+static void *aw_main(void *unused){
+ (void)unused;
+ for(;;){
+  pthread_mutex_lock(&aw_engine);
+  if(aw_stop){pthread_mutex_unlock(&aw_engine);break;}
+  aw_heartbeat=aw_now();ab_pump(aw_heartbeat,aw_watchdog);aw_cycles++;
+  uint32_t delay=aw_interval;
+  pthread_mutex_unlock(&aw_engine);
+  struct timespec t={delay/1000u,(long)(delay%1000u)*1000000L};
+  while(nanosleep(&t,&t)<0 && errno==EINTR){}
+ }
+ return NULL;
+}
+static int aw_start(void){
+ if(aw_running)return 0;
+ int e=pthread_getaffinity_np(pthread_self(),sizeof(aw_original_mask),&aw_original_mask);
+ if(e)return e;
+ pthread_attr_t attr;struct sched_param priority={0};e=pthread_attr_init(&attr);
+ if(e)return e;
+ /* Do not inherit a high-priority real-time PLC task's scheduling policy. */
+ e=pthread_attr_setinheritsched(&attr,PTHREAD_EXPLICIT_SCHED);
+ if(!e)e=pthread_attr_setschedpolicy(&attr,SCHED_OTHER);
+ if(!e)e=pthread_attr_setschedparam(&attr,&priority);
+ aw_stop=0;
+ if(!e)e=pthread_create(&aw_thread,&attr,aw_main,NULL);
+ pthread_attr_destroy(&attr);
+ if(!e){aw_running=1;aw_cpu=-1;}
+ return e;
+}
+static void aw_snapshot(uint32_t *o){
+ memset(o,0,24*sizeof(*o));o[0]=0x00040000u;o[1]=(uint32_t)aw_running;
+ o[2]=(uint32_t)aw_error;o[3]=(uint32_t)(aw_cpu+1);o[4]=aw_interval;o[5]=aw_watchdog;
+ o[6]=(uint32_t)aw_cycles;o[7]=(uint32_t)(aw_cycles>>32);
+ o[8]=(uint32_t)aw_heartbeat;o[9]=(uint32_t)(aw_heartbeat>>32);
+}
+uint32_t ab_call(uint32_t cmd,uint32_t h,uint32_t ip,uint16_t port,uint8_t unit,uint8_t f,
+ uint16_t addr,uint16_t q,uint16_t *data,uint32_t cap,uint32_t *status,uint32_t sc,
+ uint64_t ignored_now,uint32_t ct,uint32_t rt,uint32_t retry){
+ (void)ignored_now;
+ if(!status || sc<AM_STATUS_WORDS)return AB_BAD_CAPACITY;
+ pthread_once(&aw_once,aw_init_mutex);
+ if(aw_init_error){memset(status,0,24*sizeof(*status));status[2]=status[7]=(uint32_t)aw_init_error;return AB_WORKER_ERROR;}
+ pthread_mutex_lock(&aw_lifecycle);pthread_mutex_lock(&aw_engine);
+ uint32_t result=0;
+ if(cmd==1 || cmd==20 || cmd==30 || cmd==40){
+  aw_error=aw_start();
+  if(aw_error){memset(status,0,24*sizeof(*status));status[2]=status[7]=(uint32_t)aw_error;result=AB_WORKER_ERROR;goto done;}
+ }
+ if(cmd==40){
+  /* addr=CPU+1 (0 restores inherited mask), ct=pump interval, rt=scan watchdog. */
+  if(addr>CPU_SETSIZE || ct<1 || ct>20 || rt<100 || rt>60000){result=AM_ARGUMENT;goto worker_status;}
+  cpu_set_t mask=aw_original_mask;
+  if(addr){CPU_ZERO(&mask);CPU_SET(addr-1,&mask);}
+  aw_error=pthread_setaffinity_np(aw_thread,sizeof(mask),&mask);
+  if(aw_error){result=AB_WORKER_ERROR;goto worker_status;}
+  aw_cpu=(int)addr-1;aw_interval=ct;aw_watchdog=rt;
+worker_status:
+  aw_snapshot(status);
+ }else if(cmd==41){aw_snapshot(status);}
+ else result=ab_dispatch(cmd,h,ip,port,unit,f,addr,q,data,cap,status,sc,aw_now(),ct,rt,retry);
+done:
+ pthread_mutex_unlock(&aw_engine);pthread_mutex_unlock(&aw_lifecycle);
+ return result;
+}
+void ab_shutdown(void){
+ pthread_once(&aw_once,aw_init_mutex);
+ if(aw_init_error)return;
+ pthread_mutex_lock(&aw_lifecycle);pthread_mutex_lock(&aw_engine);
+ int join=aw_running;aw_stop=1;pthread_mutex_unlock(&aw_engine);
+ if(join)pthread_join(aw_thread,NULL);
+ pthread_mutex_lock(&aw_engine);ab_close_all();aw_running=0;aw_cpu=-1;
+ aw_error=0;aw_interval=1;aw_watchdog=2000;aw_cycles=0;aw_heartbeat=0;
+ pthread_mutex_unlock(&aw_engine);pthread_mutex_unlock(&aw_lifecycle);
+}

component-source/tests/test_transport.py

@@ -0,0 +1,132 @@
+#!/usr/bin/env python3
+"""Real PTY/socket integration tests for the transport RTU and TCP server bridge."""
+import ctypes as C,os,pty,select,socket,struct,subprocess,time,tempfile
+from pathlib import Path
+root=Path(__file__).resolve().parents[1]
+with tempfile.TemporaryDirectory() as tmp:
+ so=Path(tmp)/'test-transport.so'
+ subprocess.run(['cc','-shared','-fPIC','-std=c11','-Wall','-Wextra','-Werror','-O2','-pthread','-I'+str(root/'src'),*[str(root/'src'/f) for f in ['atom_mb.c','atom_rtu.c','atom_server.c','atom_bridge.c','atom_worker.c']],'-o',str(so)],check=True)
+ lib=C.CDLL(str(so));fn=lib.ab_call
+ U=C.c_uint32;W=C.c_uint16;B=C.c_uint8
+ fn.argtypes=[U,U,U,W,B,B,W,W,C.POINTER(W),U,C.POINTER(U),U,C.c_uint64,U,U,U];fn.restype=U
+ data=(W*2000)();status=(U*24)()
+ def now():return int(time.monotonic()*1000)
+ def call(cmd,h=0,ip=0,port=0,unit=1,f=3,addr=0,q=1,ct=100,rt=100,retry=10,when=None):
+  return fn(cmd,h,ip,port,unit,f,addr,q,data,2000,status,24,now() if when is None else when,ct,rt,retry)
+ def crc(frame):
+  v=0xffff
+  for byte in frame:
+   v^=byte
+   for _ in range(8):v=(v>>1)^(0xa001 if v&1 else 0)
+  return v
+ def adu(payload):return payload+struct.pack('<H',crc(payload))
+ def ready(h):
+  end=time.monotonic()+1
+  while time.monotonic()<end:
+   assert call(5,h)==0
+   if status[1]==3:return
+   time.sleep(.001)
+  raise AssertionError(('not ready',list(status)))
+ def rtu_scenario(mode,f=3):
+  master,slave=pty.openpty();path=os.ttyname(slave);os.close(slave);os.set_blocking(master,False)
+  for i,b in enumerate(path.encode()):data[i]=b
+  assert call(20,ip=9600,unit=0,f=8,addr=1,q=len(path),ct=0)==0
+  h=status[0];assert call(9,h)==0;assert call(3,h)==0;ready(h)
+  for attempt in range(2 if mode in ['crc','timeout','echo'] else 1):
+   if f==16:data[0]=123;data[1]=456
+   if f==15:data[0]=1;data[1]=0
+   assert call(6,h,unit=1,f=f,addr=10,q=2)==0
+   received=b'';sent=False;fragments=[];deadline=time.monotonic()+1
+   while time.monotonic()<deadline:
+    assert call(5,h)==0
+    if select.select([master],[],[],0)[0]:
+     try:received+=os.read(master,256)
+     except OSError:pass
+    need=(9+received[6]) if len(received)>=7 and f>=15 else 8
+    if len(received)>=need and not sent:
+     assert crc(received)==0;sent=True
+     payload=(bytes([1,f])+received[2:6]) if f>=15 else bytes([1,f,4,0,42,0,43])
+     if mode=='exception':payload=bytes([1,f|0x80,2])
+     if mode=='echo' and attempt==0:payload=payload[:3]+bytes([payload[3]^1])+payload[4:]
+     response=adu(payload)
+     if mode=='crc' and attempt==0:response=response[:-1]+bytes([response[-1]^1])
+     if mode!='timeout' or attempt>0:
+      fragments=[response[:2],response[2:4],response[4:]] if mode=='fragment' else [response]
+    if fragments:os.write(master,fragments.pop(0))
+    if status[3] or status[4]:break
+    time.sleep(.002)
+   assert sent and (status[3] or status[4]),list(status)
+   if mode in ['crc','echo'] and attempt==0:assert status[5]==8
+   elif mode=='timeout' and attempt==0:assert status[5]==6
+   elif mode=='exception':assert status[5]==9 and status[8]==2 and status[1]==3
+   else:
+    assert status[3]==1,list(status)
+    if f<=4:assert call(7,h)==0 and list(data[:2])==[42,43]
+   assert call(10,h)==0
+   if attempt==0 and mode in ['crc','timeout','echo']:ready(h)
+  assert call(6,h,unit=0,f=3,addr=0,q=1)==1 # Broadcast unsupported, no fake completion.
+  assert call(2,h)==0;assert call(5,h)==256;os.close(master)
+ for mode in ['normal','fragment','crc','timeout','exception']:rtu_scenario(mode);print('PASS RTU',mode)
+ rtu_scenario('normal',16);print('PASS RTU register write')
+ rtu_scenario('echo',16);print('PASS RTU invalid echo and reopen recovery')
+ rtu_scenario('normal',15);print('PASS RTU coil write')
+ # Transport TCP server, four independent tables mapped at nonzero Modbus address.
+ probe=socket.socket();probe.bind(('127.0.0.1',0));port=probe.getsockname()[1];probe.close()
+ assert call(30,ip=0x7f000001,port=port)==0;h=status[0]
+ for table in range(4):
+  for i in range(16):data[i]=i%2 if table<2 else 100+i
+  assert call(35,h,f=table,addr=10,q=16)==0
+ assert call(32,h)==0
+ deadline=time.monotonic()+1
+ while time.monotonic()<deadline:
+  assert call(34,h)==0
+  if status[1]:break
+  time.sleep(.002)
+ assert status[1]==1
+ peer=socket.create_connection(('127.0.0.1',port));peer.setblocking(False)
+ tid=0
+ def exchange(pdu):
+  global tid
+  tid+=1;packet=struct.pack('>HHHB',tid,0,len(pdu)+1,1)+pdu
+  peer.send(packet[:3]);assert call(34,h)==0;peer.send(packet[3:])
+  reply=b'';end=time.monotonic()+1
+  while time.monotonic()<end:
+   assert call(34,h)==0
+   try:reply+=peer.recv(260)
+   except BlockingIOError:pass
+   if len(reply)>=7 and len(reply)==6+struct.unpack('>H',reply[4:6])[0]:return reply[7:]
+   time.sleep(.001)
+  raise AssertionError(('server timeout',pdu,reply))
+ for f in [1,2,3,4]:
+  rep=exchange(struct.pack('>BHH',f,10,2));assert rep==(bytes([f,1,2]) if f<=2 else bytes([f,4,0,100,0,101]));print('PASS slave read FC',f)
+ for f in [5,6]:
+  value=0xff00 if f==5 else 456;pdu=struct.pack('>BHH',f,10,value);assert exchange(pdu)==pdu
+  assert call(36,h,f=0 if f==5 else 2,addr=10,q=16)==0 and data[0]==(1 if f==5 else 456);print('PASS slave write FC',f)
+ for f in [15,16]:
+  payload=b'\x01' if f==15 else struct.pack('>HH',321,654)
+  pdu=struct.pack('>BHHB',f,10,2,len(payload))+payload
+  assert exchange(pdu)==struct.pack('>BHH',f,10,2)
+  assert call(36,h,f=0 if f==15 else 2,addr=10,q=16)==0 and list(data[:2])==([1,0] if f==15 else [321,654]);print('PASS slave write FC',f)
+ assert exchange(struct.pack('>BHH',3,0,1))==bytes([0x83,2])
+ assert exchange(struct.pack('>BHH',16,10,0))==bytes([0x90,3])
+ assert exchange(struct.pack('>BHH',99,10,1))==bytes([0xe3,1])
+ print('PASS slave exceptions and nonzero table bases')
+ assert call(31,h)==0;assert call(34,h)==256;peer.close()
+ # Listener bind error recovers after the conflicting listener disappears.
+ blocker=socket.socket();blocker.setsockopt(socket.SOL_SOCKET,socket.SO_REUSEADDR,1);blocker.bind(('127.0.0.1',port));blocker.listen()
+ assert call(30,ip=0x7f000001,port=port)==0;h=status[0];call(32,h)
+ end=time.monotonic()+1
+ while time.monotonic()<end:
+  call(34,h)
+  if status[2]:break
+  time.sleep(.002)
+ assert status[2]==1
+ blocker.close()
+ end=time.monotonic()+1
+ while time.monotonic()<end:
+  call(34,h)
+  if status[1] and not status[2]:break
+  time.sleep(.002)
+ assert status[1]==1 and status[2]==0
+ call(31,h);print('PASS slave listener recovery and stale handles')
+ lib.ab_shutdown()

component-source/tests/test_worker.py

@@ -0,0 +1,117 @@
+#!/usr/bin/env python3
+"""Exercise production worker with real sockets/PTYS and deliberately sparse PLC calls."""
+import ctypes as C
+import os,pty,select,socket,struct,subprocess,tempfile,time,threading
+from pathlib import Path
+root=Path(__file__).resolve().parents[1]
+with tempfile.TemporaryDirectory() as td:
+ so=Path(td)/'worker.so'
+ subprocess.run(['cc','-shared','-fPIC','-std=c11','-Wall','-Wextra','-Werror','-O2','-pthread',
+  *[str(root/'src'/f) for f in ['atom_mb.c','atom_rtu.c','atom_server.c','atom_bridge.c','atom_worker.c']],'-o',str(so)],check=True)
+ lib=C.CDLL(str(so));U=C.c_uint32;W=C.c_uint16;B=C.c_uint8
+ lib.ab_call.argtypes=[U,U,U,W,B,B,W,W,C.POINTER(W),U,C.POINTER(U),U,C.c_uint64,U,U,U];lib.ab_call.restype=U
+ def call(cmd,h=0,ip=0,port=0,f=3,addr=0,q=1,values=(),ct=100,rt=100,retry=10,unit=1):
+  data=(W*2000)(*values);status=(U*24)()
+  result=lib.ab_call(cmd,h,ip,port,unit,f,addr,q,data,2000,status,24,0,ct,rt,retry)
+  return result,list(status),list(data)
+ def until(fn,seconds=1):
+  end=time.monotonic()+seconds
+  while time.monotonic()<end:
+   v=fn()
+   if v:return v
+   time.sleep(.002)
+  raise AssertionError('condition timed out')
+ def configure(watchdog=2000,cpu=-1):
+  r,s,_=call(40,addr=cpu+1,ct=1,rt=watchdog);assert r==0,(r,s)
+ def client(port):
+  r,s,_=call(1,ip=0x7f000001,port=port);assert not r;h=s[0]
+  assert not call(9,h,rt=100)[0];assert not call(3,h)[0]
+  until(lambda:call(5,h)[1][1]==3)
+  return h
+ def recv_exact(peer,count):
+  result=b''
+  while len(result)<count:
+   v=peer.recv(count-len(result));assert v;result+=v
+  return result
+ def server():
+  r,s,_=call(30,ip=0x7f000001,port=port);assert not r;h=s[0]
+  for t in range(4):assert not call(35,h,f=t,q=8,values=[0]*8)[0]
+  assert not call(32,h)[0]
+  until(lambda:call(34,h)[1][1])
+  return h
+ probe=socket.socket();probe.bind(('127.0.0.1',0));port=probe.getsockname()[1];probe.close()
+ try:
+  configure();sh=server();ch=client(port)
+  assert not call(6,ch,q=2)[0]
+  # The reply deadline is 100 ms; the PLC deliberately does nothing for 250 ms.
+  time.sleep(.25)
+  r,s,d=call(7,ch);assert not r and s[3] and not s[4] and d[:2]==[0,0],s
+  assert call(6,ch)[0]==2 # Completion cannot be overwritten before acknowledgment.
+  r,m,_=call(12,ch);assert not r and m[21]<100 and m[22]>=100,m
+  assert call(10,ch)[0]==0
+  print('PASS worker completes TCP before timeout with PLC absent for 250 ms; result retained and timed')
+  # Remote write must survive a stale PLC push before the PLC pulls its result.
+  peer=socket.create_connection(('127.0.0.1',port));peer.settimeout(1)
+  pdu=struct.pack('>BHH',6,0,4321);peer.sendall(struct.pack('>HHHB',1,0,len(pdu)+1,1)+pdu)
+  assert recv_exact(peer,12)[7:]==pdu
+  assert call(35,sh,f=2,q=8,values=[0]*8)[0]==0
+  assert call(36,sh,f=2,q=8)[2][0]==4321
+  # After acknowledgment a deliberate application override is permitted.
+  call(35,sh,f=2,q=8,values=[42]*8);assert call(36,sh,f=2,q=8)[2][0]==42
+  assert call(35,sh,f=2,addr=1,q=8,values=[0]*8)[0]==1 # explicit recreate for remap
+  peer.close();call(2,ch);call(31,sh)
+  print('PASS autonomous TCP slave and lossless remote-write handoff')
+  # Real serial PTY, again no PLC ticks after submission.
+  master,slave=pty.openpty();path=os.ttyname(slave);os.close(slave)
+  r,s,_=call(20,ip=9600,unit=0,f=8,addr=1,q=len(path),ct=0,values=list(path.encode()));assert not r;h=s[0]
+  call(9,h,rt=100);call(3,h);until(lambda:call(5,h)[1][1]==3)
+  call(6,h,unit=1,f=3,q=1)
+  assert select.select([master],[],[],1)[0];request=os.read(master,256);assert len(request)==8
+  def crc(p):
+   c=65535
+   for b in p:
+    c^=b
+    for _ in range(8):c=(c>>1)^(0xa001 if c&1 else 0)
+   return c
+  p=bytes([1,3,2,0,77]);os.write(master,p+struct.pack('<H',crc(p)))
+  time.sleep(.25);r,s,d=call(7,h);assert not r and s[3] and d[0]==77,s
+  call(2,h);os.close(master)
+  print('PASS RTU completion while PLC does not scan for 250 ms')
+  # Watchdog cancels an in-flight write without replaying it on resume.
+  configure(100)
+  listener=socket.socket();listener.bind(('127.0.0.1',0));listener.listen();listener.settimeout(1)
+  h=client(listener.getsockname()[1]);peer,_=listener.accept();peer.settimeout(1)
+  call(11,h,rt=1000);assert not call(6,h,f=16,q=1,values=[9])[0]
+  packet=recv_exact(peer,15);assert packet[7]==16
+  time.sleep(.16)
+  # command 12 is read-only and does not renew the scan lease.
+  r,s,_=call(12,h);assert not r and s[1]==0 and s[5]==10 and s[9]==1,s
+  assert peer.recv(1)==b''
+  call(5,h);peer2,_=listener.accept();peer2.settimeout(.03)
+  try:assert not peer2.recv(1), 'write replayed'
+  except socket.timeout:pass
+  call(2,h);peer.close();peer2.close();listener.close()
+  print('PASS scan watchdog closes transport, preserves uncertain write, resumes without replay')
+  # Server closes its listener when scanning ceases and recovers on refresh.
+  configure(100);sh=server();time.sleep(.16)
+  probe=socket.socket();probe.settimeout(.1);assert probe.connect_ex(('127.0.0.1',port))!=0;probe.close()
+  call(34,sh);until(lambda:call(34,sh)[1][1]);call(31,sh)
+  print('PASS server watchdog and restart')
+  # CPU selection, failure reporting, and restoring inherited mask.
+  configure(cpu=min(os.sched_getaffinity(0)))
+  r,s,_=call(41);assert not r and s[3]==min(os.sched_getaffinity(0))+1
+  assert call(40,addr=65535,ct=1,rt=2000)[0]==1
+  configure()
+  failures=[]
+  def reader():
+   for _ in range(200):
+    if call(41)[0]:failures.append('snapshot failure')
+  threads=[threading.Thread(target=reader) for _ in range(4)]
+  for t in threads:t.start()
+  for t in threads:t.join()
+  assert not failures
+  print('PASS configurable affinity, invalid arguments, concurrent snapshot access')
+  for _ in range(20):
+   lib.ab_shutdown();assert call(41)[1][1]==0;configure();assert call(41)[1][1]==1
+  print('PASS repeated joined shutdown/restart')
+ finally:lib.ab_shutdown()