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