Commit | Line | Data |
---|---|---|
ecc314ba | 1 | /* |
03244c19 BC |
2 | * windowsaio engine |
3 | * | |
4 | * IO engine using Windows IO Completion Ports. | |
ecc314ba BC |
5 | */ |
6 | ||
ecc314ba BC |
7 | #include <stdio.h> |
8 | #include <stdlib.h> | |
9 | #include <unistd.h> | |
10 | #include <signal.h> | |
11 | #include <errno.h> | |
ecc314ba BC |
12 | |
13 | #include "../fio.h" | |
14 | ||
ea4500d8 BC |
15 | typedef BOOL (WINAPI *CANCELIOEX)(HANDLE hFile, LPOVERLAPPED lpOverlapped); |
16 | ||
66c098b8 BC |
17 | int geterrno_from_win_error (DWORD code, int deferrno); |
18 | ||
67897036 BC |
19 | struct fio_overlapped { |
20 | OVERLAPPED o; | |
21 | struct io_u *io_u; | |
22 | BOOL io_complete; | |
67897036 | 23 | }; |
ecc314ba | 24 | |
9b836561 | 25 | struct windowsaio_data { |
9b836561 | 26 | struct io_u **aio_events; |
5a90bb5f | 27 | HANDLE iocp; |
67897036 | 28 | HANDLE iothread; |
9b836561 | 29 | HANDLE iocomplete_event; |
67897036 | 30 | BOOL iothread_running; |
9b836561 BC |
31 | }; |
32 | ||
ecc314ba | 33 | struct thread_ctx { |
9b836561 | 34 | HANDLE iocp; |
ecc314ba BC |
35 | struct windowsaio_data *wd; |
36 | }; | |
37 | ||
67897036 | 38 | static BOOL timeout_expired(DWORD start_count, DWORD end_count); |
ecc314ba | 39 | static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min, |
1f440ece | 40 | unsigned int max, const struct timespec *t); |
ecc314ba BC |
41 | static struct io_u *fio_windowsaio_event(struct thread_data *td, int event); |
42 | static int fio_windowsaio_queue(struct thread_data *td, | |
03244c19 | 43 | struct io_u *io_u); |
ecc314ba BC |
44 | static void fio_windowsaio_cleanup(struct thread_data *td); |
45 | static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter); | |
46 | static int fio_windowsaio_init(struct thread_data *td); | |
47 | static int fio_windowsaio_open_file(struct thread_data *td, struct fio_file *f); | |
48 | static int fio_windowsaio_close_file(struct thread_data fio_unused *td, struct fio_file *f); | |
49 | ||
ecc314ba BC |
50 | static int fio_windowsaio_init(struct thread_data *td) |
51 | { | |
ecc314ba | 52 | struct windowsaio_data *wd; |
9b836561 | 53 | int rc = 0; |
ecc314ba | 54 | |
03244c19 BC |
55 | wd = calloc(1, sizeof(struct windowsaio_data)); |
56 | if (wd == NULL) { | |
57 | log_err("windowsaio: failed to allocate memory for engine data\n"); | |
9b836561 | 58 | rc = 1; |
03244c19 | 59 | } |
ecc314ba | 60 | |
9b836561 BC |
61 | if (!rc) { |
62 | wd->aio_events = malloc(td->o.iodepth * sizeof(struct io_u*)); | |
03244c19 BC |
63 | if (wd->aio_events == NULL) { |
64 | log_err("windowsaio: failed to allocate memory for aio events list\n"); | |
9b836561 | 65 | rc = 1; |
03244c19 | 66 | } |
e4db9fec BC |
67 | } |
68 | ||
9b836561 BC |
69 | if (!rc) { |
70 | /* Create an auto-reset event */ | |
71 | wd->iocomplete_event = CreateEvent(NULL, FALSE, FALSE, NULL); | |
03244c19 BC |
72 | if (wd->iocomplete_event == NULL) { |
73 | log_err("windowsaio: failed to create io complete event handle\n"); | |
9b836561 | 74 | rc = 1; |
03244c19 | 75 | } |
9b836561 BC |
76 | } |
77 | ||
9b836561 | 78 | if (rc) { |
9b836561 | 79 | if (wd != NULL) { |
9b836561 BC |
80 | if (wd->aio_events != NULL) |
81 | free(wd->aio_events); | |
82 | ||
83 | free(wd); | |
84 | } | |
85 | } | |
ecc314ba BC |
86 | |
87 | td->io_ops->data = wd; | |
93bcfd20 | 88 | |
5a90bb5f BC |
89 | if (!rc) { |
90 | struct thread_ctx *ctx; | |
91 | struct windowsaio_data *wd; | |
92 | HANDLE hFile; | |
93 | ||
94 | hFile = CreateIoCompletionPort(INVALID_HANDLE_VALUE, NULL, 0, 0); | |
03244c19 BC |
95 | if (hFile == INVALID_HANDLE_VALUE) { |
96 | log_err("windowsaio: failed to create io completion port\n"); | |
5a90bb5f | 97 | rc = 1; |
03244c19 | 98 | } |
5a90bb5f BC |
99 | |
100 | wd = td->io_ops->data; | |
101 | wd->iothread_running = TRUE; | |
102 | wd->iocp = hFile; | |
103 | ||
104 | if (!rc) | |
105 | ctx = malloc(sizeof(struct thread_ctx)); | |
106 | ||
107 | if (!rc && ctx == NULL) | |
108 | { | |
03244c19 | 109 | log_err("windowsaio: failed to allocate memory for thread context structure\n"); |
5a90bb5f BC |
110 | CloseHandle(hFile); |
111 | rc = 1; | |
112 | } | |
113 | ||
114 | if (!rc) | |
115 | { | |
116 | ctx->iocp = hFile; | |
117 | ctx->wd = wd; | |
118 | wd->iothread = CreateThread(NULL, 0, IoCompletionRoutine, ctx, 0, NULL); | |
03244c19 BC |
119 | if (wd->iothread == NULL) |
120 | log_err("windowsaio: failed to create io completion thread\n"); | |
5a90bb5f BC |
121 | } |
122 | ||
123 | if (rc || wd->iothread == NULL) | |
124 | rc = 1; | |
125 | } | |
126 | ||
4e79098f | 127 | return rc; |
ecc314ba BC |
128 | } |
129 | ||
67897036 BC |
130 | static void fio_windowsaio_cleanup(struct thread_data *td) |
131 | { | |
67897036 BC |
132 | struct windowsaio_data *wd; |
133 | ||
134 | wd = td->io_ops->data; | |
135 | ||
136 | if (wd != NULL) { | |
40c5db35 JA |
137 | wd->iothread_running = FALSE; |
138 | WaitForSingleObject(wd->iothread, INFINITE); | |
67897036 BC |
139 | |
140 | CloseHandle(wd->iothread); | |
141 | CloseHandle(wd->iocomplete_event); | |
142 | ||
67897036 | 143 | free(wd->aio_events); |
67897036 BC |
144 | free(wd); |
145 | ||
146 | td->io_ops->data = NULL; | |
147 | } | |
148 | } | |
149 | ||
150 | ||
ecc314ba BC |
151 | static int fio_windowsaio_open_file(struct thread_data *td, struct fio_file *f) |
152 | { | |
153 | int rc = 0; | |
4e79098f | 154 | DWORD flags = FILE_FLAG_POSIX_SEMANTICS | FILE_FLAG_OVERLAPPED; |
ecc314ba BC |
155 | DWORD sharemode = FILE_SHARE_READ | FILE_SHARE_WRITE; |
156 | DWORD openmode = OPEN_ALWAYS; | |
157 | DWORD access; | |
158 | ||
159 | dprint(FD_FILE, "fd open %s\n", f->file_name); | |
160 | ||
ecc314ba | 161 | if (f->filetype == FIO_TYPE_PIPE) { |
03244c19 | 162 | log_err("windowsaio: pipes are not supported\n"); |
ecc314ba BC |
163 | return 1; |
164 | } | |
165 | ||
166 | if (!strcmp(f->file_name, "-")) { | |
03244c19 | 167 | log_err("windowsaio: can't read/write to stdin/out\n"); |
ecc314ba BC |
168 | return 1; |
169 | } | |
170 | ||
171 | if (td->o.odirect) | |
172 | flags |= FILE_FLAG_NO_BUFFERING; | |
173 | if (td->o.sync_io) | |
174 | flags |= FILE_FLAG_WRITE_THROUGH; | |
175 | ||
93bcfd20 BC |
176 | /* |
177 | * Inform Windows whether we're going to be doing sequential or | |
178 | * random io so it can tune the Cache Manager | |
179 | */ | |
ecc314ba | 180 | if (td->o.td_ddir == TD_DDIR_READ || |
ea4500d8 | 181 | td->o.td_ddir == TD_DDIR_WRITE) |
ecc314ba | 182 | flags |= FILE_FLAG_SEQUENTIAL_SCAN; |
ecc314ba | 183 | else |
ecc314ba | 184 | flags |= FILE_FLAG_RANDOM_ACCESS; |
ecc314ba | 185 | |
ea4500d8 | 186 | if (!td_write(td) || read_only) |
ecc314ba BC |
187 | access = GENERIC_READ; |
188 | else | |
189 | access = (GENERIC_READ | GENERIC_WRITE); | |
190 | ||
93bcfd20 | 191 | if (td->o.create_on_open) |
ecc314ba BC |
192 | openmode = OPEN_ALWAYS; |
193 | else | |
194 | openmode = OPEN_EXISTING; | |
195 | ||
196 | f->hFile = CreateFile(f->file_name, access, sharemode, | |
197 | NULL, openmode, flags, NULL); | |
198 | ||
03244c19 BC |
199 | if (f->hFile == INVALID_HANDLE_VALUE) { |
200 | log_err("windowsaio: failed to open file \"%s\"\n", f->file_name); | |
ecc314ba | 201 | rc = 1; |
03244c19 | 202 | } |
ecc314ba | 203 | |
93bcfd20 | 204 | /* Only set up the completion port and thread if we're not just |
ecc314ba | 205 | * querying the device size */ |
40c5db35 | 206 | if (!rc && td->io_ops->data != NULL) { |
40c5db35 | 207 | struct windowsaio_data *wd; |
ecc314ba | 208 | |
40c5db35 | 209 | wd = td->io_ops->data; |
ecc314ba | 210 | |
03244c19 BC |
211 | if (CreateIoCompletionPort(f->hFile, wd->iocp, 0, 0) == NULL) { |
212 | log_err("windowsaio: failed to create io completion port\n"); | |
ecc314ba | 213 | rc = 1; |
03244c19 | 214 | } |
ecc314ba BC |
215 | } |
216 | ||
ecc314ba BC |
217 | return rc; |
218 | } | |
219 | ||
220 | static int fio_windowsaio_close_file(struct thread_data fio_unused *td, struct fio_file *f) | |
221 | { | |
4e79098f | 222 | int rc = 0; |
1e42cc3d | 223 | |
9b836561 BC |
224 | dprint(FD_FILE, "fd close %s\n", f->file_name); |
225 | ||
ecc314ba | 226 | if (f->hFile != INVALID_HANDLE_VALUE) { |
03244c19 BC |
227 | if (!CloseHandle(f->hFile)) { |
228 | log_info("windowsaio: failed to close file handle for \"%s\"\n", f->file_name); | |
4e79098f | 229 | rc = 1; |
03244c19 | 230 | } |
ecc314ba BC |
231 | } |
232 | ||
233 | f->hFile = INVALID_HANDLE_VALUE; | |
4e79098f | 234 | return rc; |
ecc314ba BC |
235 | } |
236 | ||
67897036 BC |
237 | static BOOL timeout_expired(DWORD start_count, DWORD end_count) |
238 | { | |
239 | BOOL expired = FALSE; | |
240 | DWORD current_time; | |
241 | ||
242 | current_time = GetTickCount(); | |
243 | ||
244 | if ((end_count > start_count) && current_time >= end_count) | |
245 | expired = TRUE; | |
246 | else if (current_time < start_count && current_time > end_count) | |
247 | expired = TRUE; | |
248 | ||
249 | return expired; | |
250 | } | |
251 | ||
252 | static struct io_u* fio_windowsaio_event(struct thread_data *td, int event) | |
253 | { | |
254 | struct windowsaio_data *wd = td->io_ops->data; | |
255 | return wd->aio_events[event]; | |
256 | } | |
257 | ||
258 | static int fio_windowsaio_getevents(struct thread_data *td, unsigned int min, | |
1f440ece JA |
259 | unsigned int max, |
260 | const struct timespec *t) | |
67897036 BC |
261 | { |
262 | struct windowsaio_data *wd = td->io_ops->data; | |
67897036 BC |
263 | unsigned int dequeued = 0; |
264 | struct io_u *io_u; | |
7b2dfab1 | 265 | int i; |
67897036 BC |
266 | struct fio_overlapped *fov; |
267 | DWORD start_count = 0; | |
268 | DWORD end_count = 0; | |
269 | DWORD status; | |
270 | DWORD mswait = 250; | |
271 | ||
272 | if (t != NULL) { | |
273 | mswait = (t->tv_sec * 1000) + (t->tv_nsec / 1000000); | |
274 | start_count = GetTickCount(); | |
275 | end_count = start_count + (t->tv_sec * 1000) + (t->tv_nsec / 1000000); | |
276 | } | |
277 | ||
278 | do { | |
7b2dfab1 BC |
279 | io_u_qiter(&td->io_u_all, io_u, i) { |
280 | if (!(io_u->flags & IO_U_F_FLIGHT)) | |
281 | continue; | |
282 | ||
67897036 BC |
283 | fov = (struct fio_overlapped*)io_u->engine_data; |
284 | ||
285 | if (fov->io_complete) { | |
40c5db35 | 286 | fov->io_complete = FALSE; |
f9a58c2a | 287 | ResetEvent(fov->o.hEvent); |
67897036 BC |
288 | wd->aio_events[dequeued] = io_u; |
289 | dequeued++; | |
290 | } | |
291 | ||
292 | if (dequeued >= min) | |
293 | break; | |
294 | } | |
295 | ||
40c5db35 | 296 | if (dequeued < min) { |
67897036 | 297 | status = WaitForSingleObject(wd->iocomplete_event, mswait); |
f9a58c2a | 298 | if (status != WAIT_OBJECT_0 && dequeued >= min) |
03244c19 | 299 | break; |
67897036 BC |
300 | } |
301 | ||
302 | if (dequeued >= min || (t != NULL && timeout_expired(start_count, end_count))) | |
303 | break; | |
304 | } while (1); | |
305 | ||
306 | return dequeued; | |
307 | } | |
308 | ||
40c5db35 | 309 | static int fio_windowsaio_queue(struct thread_data *td, struct io_u *io_u) |
67897036 | 310 | { |
c73ed246 JA |
311 | struct fio_overlapped *o = io_u->engine_data; |
312 | LPOVERLAPPED lpOvl = &o->o; | |
67897036 | 313 | DWORD iobytes; |
93bcfd20 | 314 | BOOL success = FALSE; |
67897036 BC |
315 | int rc = FIO_Q_COMPLETED; |
316 | ||
317 | fio_ro_check(td, io_u); | |
318 | ||
4e79098f BC |
319 | lpOvl->Internal = STATUS_PENDING; |
320 | lpOvl->InternalHigh = 0; | |
321 | lpOvl->Offset = io_u->offset & 0xFFFFFFFF; | |
322 | lpOvl->OffsetHigh = io_u->offset >> 32; | |
67897036 BC |
323 | |
324 | switch (io_u->ddir) { | |
40c5db35 | 325 | case DDIR_WRITE: |
67897036 BC |
326 | success = WriteFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, lpOvl); |
327 | break; | |
328 | case DDIR_READ: | |
329 | success = ReadFile(io_u->file->hFile, io_u->xfer_buf, io_u->xfer_buflen, &iobytes, lpOvl); | |
330 | break; | |
331 | case DDIR_SYNC: | |
332 | case DDIR_DATASYNC: | |
333 | case DDIR_SYNC_FILE_RANGE: | |
334 | success = FlushFileBuffers(io_u->file->hFile); | |
03244c19 BC |
335 | if (!success) { |
336 | log_err("windowsaio: failed to flush file buffers\n"); | |
337 | io_u->error = win_to_posix_error(GetLastError()); | |
338 | } | |
67897036 BC |
339 | |
340 | return FIO_Q_COMPLETED; | |
341 | break; | |
342 | case DDIR_TRIM: | |
03244c19 | 343 | log_err("windowsaio: manual TRIM isn't supported on Windows\n"); |
67897036 BC |
344 | io_u->error = 1; |
345 | io_u->resid = io_u->xfer_buflen; | |
346 | return FIO_Q_COMPLETED; | |
347 | break; | |
348 | default: | |
349 | assert(0); | |
93bcfd20 | 350 | break; |
67897036 BC |
351 | } |
352 | ||
40c5db35 | 353 | if (success || GetLastError() == ERROR_IO_PENDING) |
67897036 | 354 | rc = FIO_Q_QUEUED; |
40c5db35 | 355 | else { |
2277d5d5 | 356 | io_u->error = win_to_posix_error(GetLastError()); |
67897036 BC |
357 | io_u->resid = io_u->xfer_buflen; |
358 | } | |
359 | ||
360 | return rc; | |
361 | } | |
362 | ||
363 | /* Runs as a thread and waits for queued IO to complete */ | |
364 | static DWORD WINAPI IoCompletionRoutine(LPVOID lpParameter) | |
365 | { | |
366 | OVERLAPPED *ovl; | |
367 | struct fio_overlapped *fov; | |
368 | struct io_u *io_u; | |
369 | struct windowsaio_data *wd; | |
370 | struct thread_ctx *ctx; | |
371 | ULONG_PTR ulKey = 0; | |
372 | DWORD bytes; | |
373 | ||
374 | ctx = (struct thread_ctx*)lpParameter; | |
375 | wd = ctx->wd; | |
376 | ||
377 | do { | |
66c098b8 | 378 | if (!GetQueuedCompletionStatus(ctx->iocp, &bytes, &ulKey, &ovl, 250) && ovl == NULL) |
67897036 BC |
379 | continue; |
380 | ||
381 | fov = CONTAINING_RECORD(ovl, struct fio_overlapped, o); | |
382 | io_u = fov->io_u; | |
383 | ||
384 | if (ovl->Internal == ERROR_SUCCESS) { | |
385 | io_u->resid = io_u->xfer_buflen - ovl->InternalHigh; | |
386 | io_u->error = 0; | |
387 | } else { | |
388 | io_u->resid = io_u->xfer_buflen; | |
2277d5d5 | 389 | io_u->error = win_to_posix_error(GetLastError()); |
67897036 BC |
390 | } |
391 | ||
40c5db35 | 392 | fov->io_complete = TRUE; |
67897036 BC |
393 | SetEvent(wd->iocomplete_event); |
394 | } while (ctx->wd->iothread_running); | |
395 | ||
396 | CloseHandle(ctx->iocp); | |
397 | free(ctx); | |
398 | return 0; | |
399 | } | |
400 | ||
c73ed246 JA |
401 | static void fio_windowsaio_io_u_free(struct thread_data *td, struct io_u *io_u) |
402 | { | |
403 | struct fio_overlapped *o = io_u->engine_data; | |
404 | ||
405 | if (o) { | |
406 | CloseHandle(o->o.hEvent); | |
407 | io_u->engine_data = NULL; | |
408 | free(o); | |
409 | } | |
410 | } | |
411 | ||
412 | static int fio_windowsaio_io_u_init(struct thread_data *td, struct io_u *io_u) | |
413 | { | |
414 | struct fio_overlapped *o; | |
415 | ||
416 | o = malloc(sizeof(*o)); | |
b0106419 | 417 | o->io_complete = FALSE; |
c73ed246 JA |
418 | o->io_u = io_u; |
419 | o->o.hEvent = CreateEvent(NULL, TRUE, FALSE, NULL); | |
03244c19 BC |
420 | if (o->o.hEvent == NULL) { |
421 | log_err("windowsaio: failed to create event handle\n"); | |
c73ed246 JA |
422 | free(o); |
423 | return 1; | |
424 | } | |
425 | ||
426 | io_u->engine_data = o; | |
427 | return 0; | |
428 | } | |
429 | ||
ecc314ba BC |
430 | static struct ioengine_ops ioengine = { |
431 | .name = "windowsaio", | |
432 | .version = FIO_IOOPS_VERSION, | |
433 | .init = fio_windowsaio_init, | |
434 | .queue = fio_windowsaio_queue, | |
ecc314ba BC |
435 | .getevents = fio_windowsaio_getevents, |
436 | .event = fio_windowsaio_event, | |
437 | .cleanup = fio_windowsaio_cleanup, | |
438 | .open_file = fio_windowsaio_open_file, | |
439 | .close_file = fio_windowsaio_close_file, | |
c73ed246 JA |
440 | .get_file_size = generic_get_file_size, |
441 | .io_u_init = fio_windowsaio_io_u_init, | |
442 | .io_u_free = fio_windowsaio_io_u_free, | |
ecc314ba BC |
443 | }; |
444 | ||
c874d188 | 445 | static void fio_init fio_windowsaio_register(void) |
ecc314ba BC |
446 | { |
447 | register_ioengine(&ioengine); | |
448 | } | |
449 | ||
c874d188 | 450 | static void fio_exit fio_windowsaio_unregister(void) |
ecc314ba BC |
451 | { |
452 | unregister_ioengine(&ioengine); | |
453 | } |