-
Notifications
You must be signed in to change notification settings - Fork 49
/
iocp.zig
2356 lines (2010 loc) · 78.4 KB
/
iocp.zig
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
//! Backend to use win32 IOCP.
const std = @import("std");
const assert = std.debug.assert;
const windows = @import("../windows.zig");
const queue = @import("../queue.zig");
const heap = @import("../heap.zig");
const xev = @import("../main.zig").IOCP;
const posix = std.posix;
const log = std.log.scoped(.libxev_iocp);
pub const Loop = struct {
const TimerHeap = heap.Intrusive(Timer, void, Timer.less);
/// The handle to the IO completion port.
iocp_handle: windows.HANDLE = windows.INVALID_HANDLE_VALUE,
/// The number of active completions. This DOES NOT include completions that are queued in the
/// submissions queue.
active: usize = 0,
/// Our queue of submissions that we want to enqueue on the next tick.
/// These are NOT started.
submissions: queue.Intrusive(Completion) = .{},
/// The queue of cancellation requests. These will point to the completion that we need to
/// cancel. We don't enqueue the exact completion to cancel because it may be in another queue.
cancellations: queue.Intrusive(Completion) = .{},
/// Our queue of completed completions where the callback hasn't been called yet, but the
/// "result" field should be set on every completion. This is used to delay completion callbacks
/// until the next tick.
completions: queue.Intrusive(Completion) = .{},
/// Our queue of waiting completions
asyncs: queue.Intrusive(Completion) = .{},
/// Heap of timers.
timers: TimerHeap = .{ .context = {} },
/// Cached time
cached_now: u64,
/// Duration of a tick of Windows QueryPerformanceCounter.
qpc_duration: u64,
/// Some internal fields we can pack for better space.
flags: packed struct {
/// Whether we're in a run of not (to prevent nested runs).
in_run: bool = false,
/// Whether our loop is in a stopped state or not.
stopped: bool = false,
} = .{},
/// Initialize a new IOCP-backed event loop. See the Options docs
/// for what options matter for IOCP.
pub fn init(options: xev.Options) !Loop {
_ = options;
// Get the duration of the QueryPerformanceCounter.
// We should check if the division is lossless, but it returns 10_000_000 on my machine so
// we'll handle that later.
const qpc_duration = 1_000_000_000 / windows.QueryPerformanceFrequency();
// This creates a new Completion Port
const handle = try windows.CreateIoCompletionPort(windows.INVALID_HANDLE_VALUE, null, 0, 1);
var res: Loop = .{
.iocp_handle = handle,
.qpc_duration = qpc_duration,
.cached_now = undefined,
};
res.update_now();
return res;
}
/// Deinitialize the loop, this closes the handle to the Completion Port. Any events that were
/// unprocessed are lost -- their callbacks will never be called.
pub fn deinit(self: *Loop) void {
windows.CloseHandle(self.iocp_handle);
}
/// Stop the loop. This can only be called from the main thread.
/// This will stop the loop forever. Future ticks will do nothing.
///
/// This does NOT stop any completions associated to operations that are in-flight.
pub fn stop(self: *Loop) void {
self.flags.stopped = true;
}
/// Add a completion to the loop. The completion is not started until the loop is run (`run`) or
/// an explicit submission request is made (`submit`).
pub fn add(self: *Loop, completion: *Completion) void {
// If the completion is a cancel operation, we start it immediately as it will be put in the
// cancellations queue.
if (completion.op == .cancel) {
self.start_completion(completion);
return;
}
switch (completion.flags.state) {
// The completion is in an adding state already, nothing needs to be done.
.adding => return,
// The completion is dead, probably because it was canceled.
.dead => {},
// If we reach this point, we have a problem...
.active => unreachable,
}
// We just add the completion to the queue. Failures can happen
// at submission or tick time.
completion.flags.state = .adding;
self.submissions.push(completion);
}
/// Submit any enqueued completions. This does not fire any callbacks for completed events
/// (success or error). Callbacks are only fired on the next tick.
pub fn submit(self: *Loop) !void {
// Submit all the submissions. We copy the submission queue so that any resubmits don't
// cause an infinite loop.
var queued = self.submissions;
self.submissions = .{};
// On error, we have to restore the queue because we may be batching.
errdefer self.submissions = queued;
while (queued.pop()) |c| {
switch (c.flags.state) {
.adding => self.start_completion(c),
.dead => self.stop_completion(c, null),
.active => std.log.err(
"invalid state in submission queue state={}",
.{c.flags.state},
),
}
}
}
/// Process the cancellations queue. This doesn't call any callbacks but can potentially make
/// system call to cancel active IO.
fn process_cancellations(self: *Loop) void {
while (self.cancellations.pop()) |c| {
const target = c.op.cancel.c;
var cancel_result: CancelError!void = {};
switch (target.flags.state) {
// If the target is dead already we do nothing.
.dead => {},
// If it is in the submission queue, mark them as dead so they will never be
// submitted.
.adding => target.flags.state = .dead,
// If it is active we need to schedule the deletion.
.active => self.stop_completion(target, &cancel_result),
}
// We completed the cancellation.
c.result = .{ .cancel = cancel_result };
self.completions.push(c);
}
}
/// Run the event loop. See RunMode documentation for details on modes.
/// Once the loop is run, the pointer MUST remain stable.
pub fn run(self: *Loop, mode: xev.RunMode) !void {
switch (mode) {
.no_wait => try self.tick(0),
.once => try self.tick(1),
.until_done => while (!self.done()) try self.tick(1),
}
}
/// Tick through the event loop once, waiting for at least "wait" completions to be processed by
/// the loop itself.
pub fn tick(self: *Loop, wait: u32) !void {
// If we're stopped then the loop is fully over.
if (self.flags.stopped) return;
// We can't nest runs.
if (self.flags.in_run) return error.NestedRunsNotAllowed;
self.flags.in_run = true;
defer self.flags.in_run = false;
// The list of entry that will be filled with a call to GetQueuedCompletionStatusEx.
var entries: [128]windows.OVERLAPPED_ENTRY = undefined;
var wait_rem = @as(usize, @intCast(wait));
// Handle all of our cancellations first because we may be able to stop submissions from
// even happening if its still queued. Plus, cancellations sometimes add more to the
// submission queue.
self.process_cancellations();
// Submit pending completions.
try self.submit();
// Loop condition is inspired from the kqueue backend. See its documentation for details.
while (true) {
// If we're stopped then the loop is fully over.
if (self.flags.stopped) return;
// We must update our time no matter what.
self.update_now();
const should_continue = (self.active > 0 and (wait == 0 or wait_rem > 0)) or !self.completions.empty();
if (!should_continue) break;
// Run our expired timers.
const now_timer: Timer = .{ .next = self.cached_now };
while (self.timers.peek()) |t| {
if (!Timer.less({}, t, &now_timer)) break;
// Remove the timer
assert(self.timers.deleteMin().? == t);
// Mark completion as done
const c = t.c;
c.flags.state = .dead;
// We mark it as inactive here because if we rearm below the start() function will
// reincrement this.
self.active -= 1;
// Lower our remaining count since we have processed something.
wait_rem -|= 1;
// Invoke
const action = c.callback(c.userdata, self, c, .{ .timer = .expiration });
switch (action) {
.disarm => {},
.rearm => self.start_completion(c),
}
}
// Process the completions we already have completed.
while (self.completions.pop()) |c| {
// We store whether this completion was active so we can decrement the active count
// later.
const c_active = c.flags.state == .active;
c.flags.state = .dead;
// Decrease our waiters because we are definitely processing one.
wait_rem -|= 1;
// Completion queue items MUST have a result set.
const action = c.callback(c.userdata, self, c, c.result.?);
switch (action) {
.disarm => {
// If we were active, decrement the number of active completions.
if (c_active) self.active -= 1;
},
// Only resubmit if we aren't already active
.rearm => if (!c_active) self.submissions.push(c),
}
}
// Process asyncs
if (!self.asyncs.empty()) {
var asyncs = self.asyncs;
self.asyncs = .{};
while (asyncs.pop()) |c| {
const c_wakeup = c.op.async_wait.wakeup.swap(false, .seq_cst);
// If we aren't waking this one up, requeue
if (!c_wakeup) {
self.asyncs.push(c);
continue;
}
// We are waking up, mark this as dead and call it.
c.flags.state = .dead;
self.active -= 1;
// Lower our waiters
wait_rem -|= 1;
const action = c.callback(c.userdata, self, c, .{ .async_wait = {} });
switch (action) {
.disarm => {},
.rearm => self.start_completion(c),
}
}
}
// If we have processed enough event, we break out of the loop.
if (wait_rem == 0) break;
// Determine our next timeout based on the timers.
const timeout: ?windows.DWORD = timeout: {
// If we have a timer, we want to set the timeout to our next timer value. If we
// have no timer, we wait forever.
const t = self.timers.peek() orelse break :timeout null;
// Determin the time in milliseconds. If the cast fails, we fallback to the maximum
// acceptable value.
const ms_now = self.cached_now / std.time.ns_per_ms;
const ms_next = t.next / std.time.ns_per_ms;
const ms = ms_next -| ms_now;
break :timeout std.math.cast(windows.DWORD, ms) orelse windows.INFINITE - 1;
};
// Wait for changes IO completions.
const count: u32 = windows.GetQueuedCompletionStatusEx(self.iocp_handle, &entries, timeout, false) catch |err| switch (err) {
// A timeout means that nothing was completed.
error.Timeout => 0,
else => return err,
};
// Go through the entries and perform completions callbacks.
for (entries[0..count]) |entry| {
const completion: *Completion = if (entry.lpCompletionKey == 0) completion: {
// We retrieve the Completion from the OVERLAPPED pointer as we know it's a part of
// the Completion struct.
const overlapped_ptr: ?*windows.OVERLAPPED = @as(?*windows.OVERLAPPED, @ptrCast(entry.lpOverlapped));
if (overlapped_ptr == null) {
// Probably an async wakeup
continue;
}
break :completion @alignCast(@fieldParentPtr("overlapped", overlapped_ptr.?));
} else completion: {
// JobObjects are a special case where the OVERLAPPED_ENTRY fields are interpreted differently.
// When JOBOBJECT_ASSOCIATE_COMPLETION_PORT is used, lpOverlapped actually contains the message
// value, and not the address of the overlapped structure. The Completion pointer is passed
// as the completion key instead.
const completion: *Completion = @ptrFromInt(entry.lpCompletionKey);
completion.result = .{ .job_object = .{
.message = .{
.type = @enumFromInt(entry.dwNumberOfBytesTransferred),
.value = @intFromPtr(entry.lpOverlapped),
},
} };
break :completion completion;
};
wait_rem -|= 1;
self.active -= 1;
completion.flags.state = .dead;
const result = completion.perform();
const action = completion.callback(completion.userdata, self, completion, result);
switch (action) {
.disarm => {},
.rearm => {
completion.reset();
self.start_completion(completion);
},
}
}
// If we ran through the loop once we break if we don't care.
if (wait == 0) break;
}
}
/// Returns the "loop" time in milliseconds. The loop time is updated once per loop tick, before
/// IO polling occurs. It remains constant throughout callback execution.
///
/// You can force an update of the "now" value by calling update_now() at any time from the main
/// thread.
///
/// QueryPerformanceCounter is used to get the current timestamp.
pub fn now(self: *Loop) i64 {
return @as(i64, @intCast(self.cached_now));
}
/// Update the cached time.
pub fn update_now(self: *Loop) void {
// Compute the current timestamp in ms by multiplying the QueryPerfomanceCounter value in
// ticks by the duration of a tick.
self.cached_now = windows.QueryPerformanceCounter() * self.qpc_duration;
}
/// Add a timer to the loop. The timer will execute in "next_ms". This is oneshot: the timer
/// will not repeat. To repeat a timer, either schedule another in your callback or return rearm
/// from the callback.
pub fn timer(
self: *Loop,
c: *Completion,
next_ms: u64,
userdata: ?*anyopaque,
comptime cb: xev.Callback,
) void {
c.* = .{
.op = .{
.timer = .{
.next = self.timer_next(next_ms),
},
},
.userdata = userdata,
.callback = cb,
};
self.add(c);
}
/// see io_uring.timer_reset for docs.
pub fn timer_reset(
self: *Loop,
c: *Completion,
c_cancel: *Completion,
next_ms: u64,
userdata: ?*anyopaque,
comptime cb: xev.Callback,
) void {
switch (c.flags.state) {
.dead => {
self.timer(c, next_ms, userdata, cb);
return;
},
// Adding state we can just modify the metadata and return since the timer isn't in the
// heap yet.
.adding => {
c.op.timer.next = self.timer_next(next_ms);
c.userdata = userdata;
c.callback = cb;
},
.active => {
// Update the reset time for the timer to the desired time along with all the
// callbacks.
c.op.timer.reset = self.timer_next(next_ms);
c.userdata = userdata;
c.callback = cb;
// If the cancellation is active, we assume its for this timer.
if (c_cancel.state() == .active) return;
assert(c_cancel.state() == .dead and c.state() == .active);
c_cancel.* = .{ .op = .{ .cancel = .{ .c = c } } };
self.add(c_cancel);
},
}
}
// Get the absolute timestamp corresponding to the given "next_ms".
pub fn timer_next(self: *Loop, next_ms: u64) u64 {
return self.cached_now + next_ms * std.time.ns_per_ms;
}
pub fn done(self: *Loop) bool {
return self.flags.stopped or (self.active == 0 and
self.submissions.empty() and
self.completions.empty());
}
// Start the completion.
fn start_completion(self: *Loop, completion: *Completion) void {
const StartAction = union(enum) {
// We successfully submitted the operation.
submitted: void,
// We are a timer.
timer: void,
// We are a cancellation.
cancel: void,
// We are an async wait
async_wait: void,
// We have a result code from making a system call now.
result: Result,
};
const action: StartAction = switch (completion.op) {
.noop => {
completion.flags.state = .dead;
return;
},
.accept => |*v| action: {
if (v.internal_accept_socket == null) {
var addr: posix.sockaddr.storage = undefined;
var addr_len: i32 = @sizeOf(posix.sockaddr.storage);
std.debug.assert(windows.ws2_32.getsockname(asSocket(v.socket), @as(*posix.sockaddr, @ptrCast(&addr)), &addr_len) == 0);
var socket_type: i32 = 0;
const socket_type_bytes = std.mem.asBytes(&socket_type);
var opt_len: i32 = @as(i32, @intCast(socket_type_bytes.len));
std.debug.assert(windows.ws2_32.getsockopt(asSocket(v.socket), posix.SOL.SOCKET, posix.SO.TYPE, socket_type_bytes, &opt_len) == 0);
v.internal_accept_socket = windows.WSASocketW(addr.family, socket_type, 0, null, 0, windows.ws2_32.WSA_FLAG_OVERLAPPED) catch |err| {
break :action .{ .result = .{ .accept = err } };
};
}
self.associate_fd(completion.handle().?) catch unreachable;
var discard: u32 = undefined;
const result = windows.ws2_32.AcceptEx(
asSocket(v.socket),
asSocket(v.internal_accept_socket.?),
&v.storage,
0,
0,
@as(u32, @intCast(@sizeOf(posix.sockaddr.storage))),
&discard,
&completion.overlapped,
);
if (result != windows.TRUE) {
const err = windows.ws2_32.WSAGetLastError();
switch (err) {
windows.ws2_32.WinsockError.WSA_IO_PENDING => break :action .{ .submitted = {} },
else => {
windows.CloseHandle(v.internal_accept_socket.?);
break :action .{ .result = .{ .accept = windows.unexpectedWSAError(err) } };
},
}
}
break :action .{ .submitted = {} };
},
.close => |v| .{ .result = .{ .close = windows.CloseHandle(v.fd) } },
.connect => |*v| action: {
const result = windows.ws2_32.connect(asSocket(v.socket), &v.addr.any, @as(i32, @intCast(v.addr.getOsSockLen())));
if (result != 0) {
const err = windows.ws2_32.WSAGetLastError();
break :action switch (err) {
else => .{ .result = .{ .connect = windows.unexpectedWSAError(err) } },
};
}
break :action .{ .result = .{ .connect = {} } };
},
.read => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []u8 = if (v.buffer == .slice) v.buffer.slice else &v.buffer.array;
break :action if (windows.exp.ReadFile(v.fd, buffer, &completion.overlapped)) |_|
.{
.submitted = {},
}
else |err|
.{
.result = .{ .read = err },
};
},
.pread => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []u8 = if (v.buffer == .slice) v.buffer.slice else &v.buffer.array;
completion.overlapped.DUMMYUNIONNAME.DUMMYSTRUCTNAME.Offset = @intCast(v.offset & 0xFFFF_FFFF_FFFF_FFFF);
completion.overlapped.DUMMYUNIONNAME.DUMMYSTRUCTNAME.OffsetHigh = @intCast(v.offset >> 32);
break :action if (windows.exp.ReadFile(v.fd, buffer, &completion.overlapped)) |_|
.{
.submitted = {},
}
else |err|
.{
.result = .{ .pread = err },
};
},
.shutdown => |*v| .{ .result = .{ .shutdown = posix.shutdown(asSocket(v.socket), v.how) } },
.write => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []const u8 = if (v.buffer == .slice) v.buffer.slice else v.buffer.array.array[0..v.buffer.array.len];
break :action if (windows.exp.WriteFile(v.fd, buffer, &completion.overlapped)) |_|
.{
.submitted = {},
}
else |err|
.{
.result = .{ .write = err },
};
},
.pwrite => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []const u8 = if (v.buffer == .slice) v.buffer.slice else v.buffer.array.array[0..v.buffer.array.len];
completion.overlapped.DUMMYUNIONNAME.DUMMYSTRUCTNAME.Offset = @intCast(v.offset & 0xFFFF_FFFF_FFFF_FFFF);
completion.overlapped.DUMMYUNIONNAME.DUMMYSTRUCTNAME.OffsetHigh = @intCast(v.offset >> 32);
break :action if (windows.exp.WriteFile(v.fd, buffer, &completion.overlapped)) |_|
.{
.submitted = {},
}
else |err|
.{
.result = .{ .pwrite = err },
};
},
.send => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []const u8 = if (v.buffer == .slice) v.buffer.slice else v.buffer.array.array[0..v.buffer.array.len];
v.wsa_buffer = .{ .buf = @constCast(buffer.ptr), .len = @as(u32, @intCast(buffer.len)) };
const result = windows.ws2_32.WSASend(
asSocket(v.fd),
@as([*]windows.ws2_32.WSABUF, @ptrCast(&v.wsa_buffer)),
1,
null,
0,
&completion.overlapped,
null,
);
if (result != 0) {
const err = windows.ws2_32.WSAGetLastError();
break :action switch (err) {
windows.ws2_32.WinsockError.WSA_IO_PENDING => .{ .submitted = {} },
.WSA_OPERATION_ABORTED, .WSAECONNABORTED => .{ .result = .{ .send = error.Canceled } },
.WSAECONNRESET, .WSAENETRESET => .{ .result = .{ .send = error.ConnectionReset } },
else => .{ .result = .{ .send = windows.unexpectedWSAError(err) } },
};
}
break :action .{ .submitted = {} };
},
.recv => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []u8 = if (v.buffer == .slice) v.buffer.slice else &v.buffer.array;
v.wsa_buffer = .{ .buf = buffer.ptr, .len = @as(u32, @intCast(buffer.len)) };
var flags: u32 = 0;
const result = windows.ws2_32.WSARecv(
asSocket(v.fd),
@as([*]windows.ws2_32.WSABUF, @ptrCast(&v.wsa_buffer)),
1,
null,
&flags,
&completion.overlapped,
null,
);
if (result != 0) {
const err = windows.ws2_32.WSAGetLastError();
break :action switch (err) {
windows.ws2_32.WinsockError.WSA_IO_PENDING => .{ .submitted = {} },
.WSA_OPERATION_ABORTED, .WSAECONNABORTED => .{ .result = .{ .recv = error.Canceled } },
.WSAECONNRESET, .WSAENETRESET => .{ .result = .{ .recv = error.ConnectionReset } },
else => .{ .result = .{ .recv = windows.unexpectedWSAError(err) } },
};
}
break :action .{ .submitted = {} };
},
.sendto => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []const u8 = if (v.buffer == .slice) v.buffer.slice else v.buffer.array.array[0..v.buffer.array.len];
v.wsa_buffer = .{ .buf = @constCast(buffer.ptr), .len = @as(u32, @intCast(buffer.len)) };
const result = windows.ws2_32.WSASendTo(
asSocket(v.fd),
@as([*]windows.ws2_32.WSABUF, @ptrCast(&v.wsa_buffer)),
1,
null,
0,
&v.addr.any,
@as(i32, @intCast(v.addr.getOsSockLen())),
&completion.overlapped,
null,
);
if (result != 0) {
const err = windows.ws2_32.WSAGetLastError();
break :action switch (err) {
windows.ws2_32.WinsockError.WSA_IO_PENDING => .{ .submitted = {} },
.WSA_OPERATION_ABORTED, .WSAECONNABORTED => .{ .result = .{ .sendto = error.Canceled } },
.WSAECONNRESET, .WSAENETRESET => .{ .result = .{ .sendto = error.ConnectionReset } },
else => .{ .result = .{ .sendto = windows.unexpectedWSAError(err) } },
};
}
break :action .{ .submitted = {} };
},
.recvfrom => |*v| action: {
self.associate_fd(completion.handle().?) catch unreachable;
const buffer: []u8 = if (v.buffer == .slice) v.buffer.slice else &v.buffer.array;
v.wsa_buffer = .{ .buf = buffer.ptr, .len = @as(u32, @intCast(buffer.len)) };
var flags: u32 = 0;
const result = windows.ws2_32.WSARecvFrom(
asSocket(v.fd),
@as([*]windows.ws2_32.WSABUF, @ptrCast(&v.wsa_buffer)),
1,
null,
&flags,
&v.addr,
@as(*i32, @ptrCast(&v.addr_size)),
&completion.overlapped,
null,
);
if (result != 0) {
const err = windows.ws2_32.WSAGetLastError();
break :action switch (err) {
windows.ws2_32.WinsockError.WSA_IO_PENDING => .{ .submitted = {} },
.WSA_OPERATION_ABORTED, .WSAECONNABORTED => .{ .result = .{ .recvfrom = error.Canceled } },
.WSAECONNRESET, .WSAENETRESET => .{ .result = .{ .recvfrom = error.ConnectionReset } },
else => .{ .result = .{ .recvfrom = windows.unexpectedWSAError(err) } },
};
}
break :action .{ .submitted = {} };
},
.timer => |*v| action: {
v.c = completion;
self.timers.insert(v);
break :action .{ .timer = {} };
},
.cancel => action: {
self.cancellations.push(completion);
break :action .{ .cancel = {} };
},
.async_wait => action: {
self.asyncs.push(completion);
break :action .{ .async_wait = {} };
},
.job_object => |*v| action: {
if (!v.associated) {
var port = windows.exp.JOBOBJECT_ASSOCIATE_COMPLETION_PORT{
.CompletionKey = @intFromPtr(completion),
.CompletionPort = self.iocp_handle,
};
windows.exp.SetInformationJobObject(
v.job,
.JobObjectAssociateCompletionPortInformation,
&port,
@sizeOf(windows.exp.JOBOBJECT_ASSOCIATE_COMPLETION_PORT),
) catch |err| break :action .{ .result = .{ .job_object = err } };
v.associated = true;
const action = completion.callback(completion.userdata, self, completion, .{ .job_object = .{ .associated = {} } });
switch (action) {
.disarm => {
completion.flags.state = .dead;
return;
},
.rearm => break :action .{ .submitted = {} },
}
}
break :action .{ .submitted = {} };
},
};
switch (action) {
.timer, .submitted, .cancel => {
// Increase our active count so we now wait for this. We assume it'll successfully
// queue. If it doesn't we handle that later (see submit).
self.active += 1;
completion.flags.state = .active;
},
.async_wait => {
// We are considered an active completion.
self.active += 1;
completion.flags.state = .active;
},
// A result is immediately available. Queue the completion to be invoked.
.result => |r| {
completion.result = r;
self.completions.push(completion);
},
}
}
/// Stop the completion. Fill `cancel_result` if it is non-null.
fn stop_completion(self: *Loop, completion: *Completion, cancel_result: ?*CancelError!void) void {
if (completion.flags.state == .active and completion.result != null) return;
// Inspect other operations. WARNING: the state can be anything here so per op be sure to
// check the state flag.
switch (completion.op) {
.timer => |*v| {
if (completion.flags.state == .active) {
// Remove from the heap so it never fires...
self.timers.remove(v);
// If we have reset AND we got cancellation result, that means that we were
// canceled so that we can update our expiration time.
if (v.reset) |r| {
v.next = r;
v.reset = null;
completion.flags.state = .dead;
self.active -= 1;
self.add(completion);
return;
}
}
// Add to our completion so we trigger the callback.
completion.result = .{ .timer = .cancel };
self.completions.push(completion);
// Note the timers state purposely remains ACTIVE so that
// when we process the completion we decrement the
// active count.
},
.accept => |*v| {
if (completion.flags.state == .active) {
const result = windows.kernel32.CancelIoEx(asSocket(v.socket), &completion.overlapped);
cancel_result.?.* = if (result == windows.FALSE)
windows.unexpectedError(windows.kernel32.GetLastError())
else {};
}
},
inline .read, .pread, .write, .pwrite, .recv, .send, .sendto, .recvfrom => |*v| {
if (completion.flags.state == .active) {
const result = windows.kernel32.CancelIoEx(asSocket(v.fd), &completion.overlapped);
cancel_result.?.* = if (result == windows.FALSE)
windows.unexpectedError(windows.kernel32.GetLastError())
else {};
}
},
else => @panic("Not implemented"),
}
}
// Sens an empty Completion token so that the loop wakes up if it is waiting for a completion
// event.
pub fn async_notify(self: *Loop, completion: *Completion) void {
// The completion must be in a waiting state.
assert(completion.op == .async_wait);
// The completion has been wakeup, this is used to see which completion in the async queue
// needs to be removed.
completion.op.async_wait.wakeup.store(true, .seq_cst);
const result = windows.kernel32.PostQueuedCompletionStatus(
self.iocp_handle,
0,
0,
null,
);
// NOTE(Corendos): if something goes wrong, ignore it for the moment.
if (result == windows.FALSE) {
const err = windows.kernel32.GetLastError();
windows.unexpectedError(err) catch {};
}
}
/// Associate a handler to the internal completion port.
/// This has to be done only once per handle so we delegate the responsibility to the caller.
pub fn associate_fd(self: Loop, fd: windows.HANDLE) !void {
if (fd == windows.INVALID_HANDLE_VALUE or self.iocp_handle == windows.INVALID_HANDLE_VALUE) return error.InvalidParameter;
// We ignore the error here because multiple call to CreateIoCompletionPort with a HANDLE
// already registered triggers a INVALID_PARAMETER error and we have no way to see the cause
// of it.
_ = windows.kernel32.CreateIoCompletionPort(fd, self.iocp_handle, 0, 0);
}
};
/// Convenience to convert from windows.HANDLE to windows.ws2_32.SOCKET (which are the same thing).
inline fn asSocket(h: windows.HANDLE) windows.ws2_32.SOCKET {
return @as(windows.ws2_32.SOCKET, @ptrCast(h));
}
/// A completion is a request to perform some work with the loop.
pub const Completion = struct {
/// Operation to execute.
op: Operation = .{ .noop = {} },
/// Userdata and callback for when the completion is finished.
userdata: ?*anyopaque = null,
callback: xev.Callback = xev.noopCallback,
//---------------------------------------------------------------
// Internal fields
/// Intrusive queue field.
next: ?*Completion = null,
/// Result code of the syscall. Only used internally in certain scenarios, should not be relied
/// upon by program authors.
result: ?Result = null,
flags: packed struct {
/// Watch state of this completion. We use this to determine whether we're active, adding or
/// dead. This lets us add and abd delete multiple times before a loop tick and handle the
/// state properly.
state: State = .dead,
} = .{},
/// Win32 OVERLAPPED struct used for asynchronous IO. Only used internally in certain scenarios.
/// It needs to be there as we rely on @fieldParentPtr to get the completion using a pointer to
/// that field.
overlapped: windows.OVERLAPPED = .{
.Internal = 0,
.InternalHigh = 0,
.DUMMYUNIONNAME = .{ .Pointer = null },
.hEvent = null,
},
/// Loop associated with this completion. HANDLE are required to be associated with an I/O
/// Completion Port to work properly.
loop: ?*const xev.Loop = null,
const State = enum(u2) {
/// completion is not part of any loop
dead = 0,
/// completion is in the submission queue
adding = 1,
/// completion is submitted successfully
active = 2,
};
/// Returns the state of this completion. There are some things to be cautious about when
/// calling this function.
///
/// First, this is only safe to call from the main thread. This cannot be called from any other
/// thread.
///
/// Second, if you are using default "undefined" completions, this will NOT return a valid value
/// if you access it. You must zero your completion using ".{}". You only need to zero the
/// completion once. Once the completion is in use, it will always be valid.
///
/// Third, if you stop the loop (loop.stop()), the completions registered with the loop will NOT
/// be reset to a dead state.
pub fn state(self: Completion) xev.CompletionState {
return switch (self.flags.state) {
.dead => .dead,
.adding, .active => .active,
};
}
/// Returns a handle for the current operation if it makes sense.
fn handle(self: Completion) ?windows.HANDLE {
return switch (self.op) {
inline .accept => |*v| v.socket,
inline .read, .pread, .write, .pwrite, .recv, .send, .recvfrom, .sendto => |*v| v.fd,
else => null,
};
}
/// Perform the operation associated with this completion. This will perform the full blocking
/// operation for the completion.
pub fn perform(self: *Completion) Result {
return switch (self.op) {
.noop, .close, .connect, .shutdown, .timer, .cancel => {
std.log.warn("perform op={s}", .{@tagName(self.op)});
unreachable;
},
.accept => |*v| r: {
var bytes_transferred: u32 = 0;
var flags: u32 = 0;
const result = windows.ws2_32.WSAGetOverlappedResult(asSocket(v.socket), &self.overlapped, &bytes_transferred, windows.FALSE, &flags);
if (result != windows.TRUE) {
const err = windows.ws2_32.WSAGetLastError();
const r = .{
.accept = switch (err) {
windows.ws2_32.WinsockError.WSA_OPERATION_ABORTED => error.Canceled,
else => windows.unexpectedWSAError(err),
},
};
windows.CloseHandle(v.internal_accept_socket.?);
break :r r;
}
break :r .{ .accept = self.op.accept.internal_accept_socket.? };
},
.read => |*v| r: {
var bytes_transferred: windows.DWORD = 0;
const result = windows.kernel32.GetOverlappedResult(v.fd, &self.overlapped, &bytes_transferred, windows.FALSE);
if (result == windows.FALSE) {
const err = windows.kernel32.GetLastError();
break :r .{ .read = switch (err) {
windows.Win32Error.OPERATION_ABORTED => error.Canceled,
else => error.Unexpected,
} };
}
break :r .{ .read = @as(usize, @intCast(bytes_transferred)) };
},
.pread => |*v| r: {
var bytes_transferred: windows.DWORD = 0;
const result = windows.kernel32.GetOverlappedResult(v.fd, &self.overlapped, &bytes_transferred, windows.FALSE);
if (result == windows.FALSE) {
const err = windows.kernel32.GetLastError();
break :r .{ .read = switch (err) {
windows.Win32Error.OPERATION_ABORTED => error.Canceled,
else => error.Unexpected,
} };
}
break :r .{ .pread = @as(usize, @intCast(bytes_transferred)) };
},
.write => |*v| r: {