xref: /qemu/util/aio-win32.c (revision 6402cbbb)
1 /*
2  * QEMU aio implementation
3  *
4  * Copyright IBM Corp., 2008
5  * Copyright Red Hat Inc., 2012
6  *
7  * Authors:
8  *  Anthony Liguori   <aliguori@us.ibm.com>
9  *  Paolo Bonzini     <pbonzini@redhat.com>
10  *
11  * This work is licensed under the terms of the GNU GPL, version 2.  See
12  * the COPYING file in the top-level directory.
13  *
14  * Contributions after 2012-01-13 are licensed under the terms of the
15  * GNU GPL, version 2 or (at your option) any later version.
16  */
17 
18 #include "qemu/osdep.h"
19 #include "qemu-common.h"
20 #include "block/block.h"
21 #include "qemu/queue.h"
22 #include "qemu/sockets.h"
23 #include "qapi/error.h"
24 #include "qemu/rcu_queue.h"
25 
26 struct AioHandler {
27     EventNotifier *e;
28     IOHandler *io_read;
29     IOHandler *io_write;
30     EventNotifierHandler *io_notify;
31     GPollFD pfd;
32     int deleted;
33     void *opaque;
34     bool is_external;
35     QLIST_ENTRY(AioHandler) node;
36 };
37 
38 void aio_set_fd_handler(AioContext *ctx,
39                         int fd,
40                         bool is_external,
41                         IOHandler *io_read,
42                         IOHandler *io_write,
43                         AioPollFn *io_poll,
44                         void *opaque)
45 {
46     /* fd is a SOCKET in our case */
47     AioHandler *node;
48 
49     qemu_lockcnt_lock(&ctx->list_lock);
50     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
51         if (node->pfd.fd == fd && !node->deleted) {
52             break;
53         }
54     }
55 
56     /* Are we deleting the fd handler? */
57     if (!io_read && !io_write) {
58         if (node) {
59             /* If aio_poll is in progress, just mark the node as deleted */
60             if (qemu_lockcnt_count(&ctx->list_lock)) {
61                 node->deleted = 1;
62                 node->pfd.revents = 0;
63             } else {
64                 /* Otherwise, delete it for real.  We can't just mark it as
65                  * deleted because deleted nodes are only cleaned up after
66                  * releasing the list_lock.
67                  */
68                 QLIST_REMOVE(node, node);
69                 g_free(node);
70             }
71         }
72     } else {
73         HANDLE event;
74         long bitmask = 0;
75 
76         if (node == NULL) {
77             /* Alloc and insert if it's not already there */
78             node = g_new0(AioHandler, 1);
79             node->pfd.fd = fd;
80             QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
81         }
82 
83         node->pfd.events = 0;
84         if (node->io_read) {
85             node->pfd.events |= G_IO_IN;
86         }
87         if (node->io_write) {
88             node->pfd.events |= G_IO_OUT;
89         }
90 
91         node->e = &ctx->notifier;
92 
93         /* Update handler with latest information */
94         node->opaque = opaque;
95         node->io_read = io_read;
96         node->io_write = io_write;
97         node->is_external = is_external;
98 
99         if (io_read) {
100             bitmask |= FD_READ | FD_ACCEPT | FD_CLOSE;
101         }
102 
103         if (io_write) {
104             bitmask |= FD_WRITE | FD_CONNECT;
105         }
106 
107         event = event_notifier_get_handle(&ctx->notifier);
108         WSAEventSelect(node->pfd.fd, event, bitmask);
109     }
110 
111     qemu_lockcnt_unlock(&ctx->list_lock);
112     aio_notify(ctx);
113 }
114 
115 void aio_set_fd_poll(AioContext *ctx, int fd,
116                      IOHandler *io_poll_begin,
117                      IOHandler *io_poll_end)
118 {
119     /* Not implemented */
120 }
121 
122 void aio_set_event_notifier(AioContext *ctx,
123                             EventNotifier *e,
124                             bool is_external,
125                             EventNotifierHandler *io_notify,
126                             AioPollFn *io_poll)
127 {
128     AioHandler *node;
129 
130     qemu_lockcnt_lock(&ctx->list_lock);
131     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
132         if (node->e == e && !node->deleted) {
133             break;
134         }
135     }
136 
137     /* Are we deleting the fd handler? */
138     if (!io_notify) {
139         if (node) {
140             g_source_remove_poll(&ctx->source, &node->pfd);
141 
142             /* aio_poll is in progress, just mark the node as deleted */
143             if (qemu_lockcnt_count(&ctx->list_lock)) {
144                 node->deleted = 1;
145                 node->pfd.revents = 0;
146             } else {
147                 /* Otherwise, delete it for real.  We can't just mark it as
148                  * deleted because deleted nodes are only cleaned up after
149                  * releasing the list_lock.
150                  */
151                 QLIST_REMOVE(node, node);
152                 g_free(node);
153             }
154         }
155     } else {
156         if (node == NULL) {
157             /* Alloc and insert if it's not already there */
158             node = g_new0(AioHandler, 1);
159             node->e = e;
160             node->pfd.fd = (uintptr_t)event_notifier_get_handle(e);
161             node->pfd.events = G_IO_IN;
162             node->is_external = is_external;
163             QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
164 
165             g_source_add_poll(&ctx->source, &node->pfd);
166         }
167         /* Update handler with latest information */
168         node->io_notify = io_notify;
169     }
170 
171     qemu_lockcnt_unlock(&ctx->list_lock);
172     aio_notify(ctx);
173 }
174 
175 void aio_set_event_notifier_poll(AioContext *ctx,
176                                  EventNotifier *notifier,
177                                  EventNotifierHandler *io_poll_begin,
178                                  EventNotifierHandler *io_poll_end)
179 {
180     /* Not implemented */
181 }
182 
183 bool aio_prepare(AioContext *ctx)
184 {
185     static struct timeval tv0;
186     AioHandler *node;
187     bool have_select_revents = false;
188     fd_set rfds, wfds;
189 
190     /*
191      * We have to walk very carefully in case aio_set_fd_handler is
192      * called while we're walking.
193      */
194     qemu_lockcnt_inc(&ctx->list_lock);
195 
196     /* fill fd sets */
197     FD_ZERO(&rfds);
198     FD_ZERO(&wfds);
199     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
200         if (node->io_read) {
201             FD_SET ((SOCKET)node->pfd.fd, &rfds);
202         }
203         if (node->io_write) {
204             FD_SET ((SOCKET)node->pfd.fd, &wfds);
205         }
206     }
207 
208     if (select(0, &rfds, &wfds, NULL, &tv0) > 0) {
209         QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
210             node->pfd.revents = 0;
211             if (FD_ISSET(node->pfd.fd, &rfds)) {
212                 node->pfd.revents |= G_IO_IN;
213                 have_select_revents = true;
214             }
215 
216             if (FD_ISSET(node->pfd.fd, &wfds)) {
217                 node->pfd.revents |= G_IO_OUT;
218                 have_select_revents = true;
219             }
220         }
221     }
222 
223     qemu_lockcnt_dec(&ctx->list_lock);
224     return have_select_revents;
225 }
226 
227 bool aio_pending(AioContext *ctx)
228 {
229     AioHandler *node;
230     bool result = false;
231 
232     /*
233      * We have to walk very carefully in case aio_set_fd_handler is
234      * called while we're walking.
235      */
236     qemu_lockcnt_inc(&ctx->list_lock);
237     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
238         if (node->pfd.revents && node->io_notify) {
239             result = true;
240             break;
241         }
242 
243         if ((node->pfd.revents & G_IO_IN) && node->io_read) {
244             result = true;
245             break;
246         }
247         if ((node->pfd.revents & G_IO_OUT) && node->io_write) {
248             result = true;
249             break;
250         }
251     }
252 
253     qemu_lockcnt_dec(&ctx->list_lock);
254     return result;
255 }
256 
257 static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
258 {
259     AioHandler *node;
260     bool progress = false;
261     AioHandler *tmp;
262 
263     /*
264      * We have to walk very carefully in case aio_set_fd_handler is
265      * called while we're walking.
266      */
267     QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) {
268         int revents = node->pfd.revents;
269 
270         if (!node->deleted &&
271             (revents || event_notifier_get_handle(node->e) == event) &&
272             node->io_notify) {
273             node->pfd.revents = 0;
274             node->io_notify(node->e);
275 
276             /* aio_notify() does not count as progress */
277             if (node->e != &ctx->notifier) {
278                 progress = true;
279             }
280         }
281 
282         if (!node->deleted &&
283             (node->io_read || node->io_write)) {
284             node->pfd.revents = 0;
285             if ((revents & G_IO_IN) && node->io_read) {
286                 node->io_read(node->opaque);
287                 progress = true;
288             }
289             if ((revents & G_IO_OUT) && node->io_write) {
290                 node->io_write(node->opaque);
291                 progress = true;
292             }
293 
294             /* if the next select() will return an event, we have progressed */
295             if (event == event_notifier_get_handle(&ctx->notifier)) {
296                 WSANETWORKEVENTS ev;
297                 WSAEnumNetworkEvents(node->pfd.fd, event, &ev);
298                 if (ev.lNetworkEvents) {
299                     progress = true;
300                 }
301             }
302         }
303 
304         if (node->deleted) {
305             if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
306                 QLIST_REMOVE(node, node);
307                 g_free(node);
308                 qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
309             }
310         }
311     }
312 
313     return progress;
314 }
315 
316 void aio_dispatch(AioContext *ctx)
317 {
318     qemu_lockcnt_inc(&ctx->list_lock);
319     aio_bh_poll(ctx);
320     aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
321     qemu_lockcnt_dec(&ctx->list_lock);
322     timerlistgroup_run_timers(&ctx->tlg);
323 }
324 
325 bool aio_poll(AioContext *ctx, bool blocking)
326 {
327     AioHandler *node;
328     HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
329     bool progress, have_select_revents, first;
330     int count;
331     int timeout;
332 
333     progress = false;
334 
335     /* aio_notify can avoid the expensive event_notifier_set if
336      * everything (file descriptors, bottom halves, timers) will
337      * be re-evaluated before the next blocking poll().  This is
338      * already true when aio_poll is called with blocking == false;
339      * if blocking == true, it is only true after poll() returns,
340      * so disable the optimization now.
341      */
342     if (blocking) {
343         atomic_add(&ctx->notify_me, 2);
344     }
345 
346     qemu_lockcnt_inc(&ctx->list_lock);
347     have_select_revents = aio_prepare(ctx);
348 
349     /* fill fd sets */
350     count = 0;
351     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
352         if (!node->deleted && node->io_notify
353             && aio_node_check(ctx, node->is_external)) {
354             events[count++] = event_notifier_get_handle(node->e);
355         }
356     }
357 
358     first = true;
359 
360     /* ctx->notifier is always registered.  */
361     assert(count > 0);
362 
363     /* Multiple iterations, all of them non-blocking except the first,
364      * may be necessary to process all pending events.  After the first
365      * WaitForMultipleObjects call ctx->notify_me will be decremented.
366      */
367     do {
368         HANDLE event;
369         int ret;
370 
371         timeout = blocking && !have_select_revents
372             ? qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)) : 0;
373         ret = WaitForMultipleObjects(count, events, FALSE, timeout);
374         if (blocking) {
375             assert(first);
376             atomic_sub(&ctx->notify_me, 2);
377         }
378 
379         if (first) {
380             aio_notify_accept(ctx);
381             progress |= aio_bh_poll(ctx);
382             first = false;
383         }
384 
385         /* if we have any signaled events, dispatch event */
386         event = NULL;
387         if ((DWORD) (ret - WAIT_OBJECT_0) < count) {
388             event = events[ret - WAIT_OBJECT_0];
389             events[ret - WAIT_OBJECT_0] = events[--count];
390         } else if (!have_select_revents) {
391             break;
392         }
393 
394         have_select_revents = false;
395         blocking = false;
396 
397         progress |= aio_dispatch_handlers(ctx, event);
398     } while (count > 0);
399 
400     qemu_lockcnt_dec(&ctx->list_lock);
401 
402     progress |= timerlistgroup_run_timers(&ctx->tlg);
403     return progress;
404 }
405 
406 void aio_context_setup(AioContext *ctx)
407 {
408 }
409 
410 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
411                                  int64_t grow, int64_t shrink, Error **errp)
412 {
413     error_setg(errp, "AioContext polling is not implemented on Windows");
414 }
415