xref: /qemu/util/aio-posix.c (revision 7271a819)
1 /*
2  * QEMU aio implementation
3  *
4  * Copyright IBM, Corp. 2008
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  * Contributions after 2012-01-13 are licensed under the terms of the
13  * GNU GPL, version 2 or (at your option) any later version.
14  */
15 
16 #include "qemu/osdep.h"
17 #include "qemu-common.h"
18 #include "block/block.h"
19 #include "qemu/rcu_queue.h"
20 #include "qemu/sockets.h"
21 #include "qemu/cutils.h"
22 #include "trace.h"
23 #ifdef CONFIG_EPOLL_CREATE1
24 #include <sys/epoll.h>
25 #endif
26 
27 struct AioHandler
28 {
29     GPollFD pfd;
30     IOHandler *io_read;
31     IOHandler *io_write;
32     AioPollFn *io_poll;
33     IOHandler *io_poll_begin;
34     IOHandler *io_poll_end;
35     int deleted;
36     void *opaque;
37     bool is_external;
38     QLIST_ENTRY(AioHandler) node;
39 };
40 
41 #ifdef CONFIG_EPOLL_CREATE1
42 
43 /* The fd number threashold to switch to epoll */
44 #define EPOLL_ENABLE_THRESHOLD 64
45 
46 static void aio_epoll_disable(AioContext *ctx)
47 {
48     ctx->epoll_available = false;
49     if (!ctx->epoll_enabled) {
50         return;
51     }
52     ctx->epoll_enabled = false;
53     close(ctx->epollfd);
54 }
55 
56 static inline int epoll_events_from_pfd(int pfd_events)
57 {
58     return (pfd_events & G_IO_IN ? EPOLLIN : 0) |
59            (pfd_events & G_IO_OUT ? EPOLLOUT : 0) |
60            (pfd_events & G_IO_HUP ? EPOLLHUP : 0) |
61            (pfd_events & G_IO_ERR ? EPOLLERR : 0);
62 }
63 
64 static bool aio_epoll_try_enable(AioContext *ctx)
65 {
66     AioHandler *node;
67     struct epoll_event event;
68 
69     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
70         int r;
71         if (node->deleted || !node->pfd.events) {
72             continue;
73         }
74         event.events = epoll_events_from_pfd(node->pfd.events);
75         event.data.ptr = node;
76         r = epoll_ctl(ctx->epollfd, EPOLL_CTL_ADD, node->pfd.fd, &event);
77         if (r) {
78             return false;
79         }
80     }
81     ctx->epoll_enabled = true;
82     return true;
83 }
84 
85 static void aio_epoll_update(AioContext *ctx, AioHandler *node, bool is_new)
86 {
87     struct epoll_event event;
88     int r;
89     int ctl;
90 
91     if (!ctx->epoll_enabled) {
92         return;
93     }
94     if (!node->pfd.events) {
95         ctl = EPOLL_CTL_DEL;
96     } else {
97         event.data.ptr = node;
98         event.events = epoll_events_from_pfd(node->pfd.events);
99         ctl = is_new ? EPOLL_CTL_ADD : EPOLL_CTL_MOD;
100     }
101 
102     r = epoll_ctl(ctx->epollfd, ctl, node->pfd.fd, &event);
103     if (r) {
104         aio_epoll_disable(ctx);
105     }
106 }
107 
108 static int aio_epoll(AioContext *ctx, GPollFD *pfds,
109                      unsigned npfd, int64_t timeout)
110 {
111     AioHandler *node;
112     int i, ret = 0;
113     struct epoll_event events[128];
114 
115     assert(npfd == 1);
116     assert(pfds[0].fd == ctx->epollfd);
117     if (timeout > 0) {
118         ret = qemu_poll_ns(pfds, npfd, timeout);
119     }
120     if (timeout <= 0 || ret > 0) {
121         ret = epoll_wait(ctx->epollfd, events,
122                          sizeof(events) / sizeof(events[0]),
123                          timeout);
124         if (ret <= 0) {
125             goto out;
126         }
127         for (i = 0; i < ret; i++) {
128             int ev = events[i].events;
129             node = events[i].data.ptr;
130             node->pfd.revents = (ev & EPOLLIN ? G_IO_IN : 0) |
131                 (ev & EPOLLOUT ? G_IO_OUT : 0) |
132                 (ev & EPOLLHUP ? G_IO_HUP : 0) |
133                 (ev & EPOLLERR ? G_IO_ERR : 0);
134         }
135     }
136 out:
137     return ret;
138 }
139 
140 static bool aio_epoll_enabled(AioContext *ctx)
141 {
142     /* Fall back to ppoll when external clients are disabled. */
143     return !aio_external_disabled(ctx) && ctx->epoll_enabled;
144 }
145 
146 static bool aio_epoll_check_poll(AioContext *ctx, GPollFD *pfds,
147                                  unsigned npfd, int64_t timeout)
148 {
149     if (!ctx->epoll_available) {
150         return false;
151     }
152     if (aio_epoll_enabled(ctx)) {
153         return true;
154     }
155     if (npfd >= EPOLL_ENABLE_THRESHOLD) {
156         if (aio_epoll_try_enable(ctx)) {
157             return true;
158         } else {
159             aio_epoll_disable(ctx);
160         }
161     }
162     return false;
163 }
164 
165 #else
166 
167 static void aio_epoll_update(AioContext *ctx, AioHandler *node, bool is_new)
168 {
169 }
170 
171 static int aio_epoll(AioContext *ctx, GPollFD *pfds,
172                      unsigned npfd, int64_t timeout)
173 {
174     assert(false);
175 }
176 
177 static bool aio_epoll_enabled(AioContext *ctx)
178 {
179     return false;
180 }
181 
182 static bool aio_epoll_check_poll(AioContext *ctx, GPollFD *pfds,
183                           unsigned npfd, int64_t timeout)
184 {
185     return false;
186 }
187 
188 #endif
189 
190 static AioHandler *find_aio_handler(AioContext *ctx, int fd)
191 {
192     AioHandler *node;
193 
194     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
195         if (node->pfd.fd == fd)
196             if (!node->deleted)
197                 return node;
198     }
199 
200     return NULL;
201 }
202 
203 void aio_set_fd_handler(AioContext *ctx,
204                         int fd,
205                         bool is_external,
206                         IOHandler *io_read,
207                         IOHandler *io_write,
208                         AioPollFn *io_poll,
209                         void *opaque)
210 {
211     AioHandler *node;
212     bool is_new = false;
213     bool deleted = false;
214 
215     qemu_lockcnt_lock(&ctx->list_lock);
216 
217     node = find_aio_handler(ctx, fd);
218 
219     /* Are we deleting the fd handler? */
220     if (!io_read && !io_write && !io_poll) {
221         if (node == NULL) {
222             qemu_lockcnt_unlock(&ctx->list_lock);
223             return;
224         }
225 
226         /* If the GSource is in the process of being destroyed then
227          * g_source_remove_poll() causes an assertion failure.  Skip
228          * removal in that case, because glib cleans up its state during
229          * destruction anyway.
230          */
231         if (!g_source_is_destroyed(&ctx->source)) {
232             g_source_remove_poll(&ctx->source, &node->pfd);
233         }
234 
235         /* If the lock is held, just mark the node as deleted */
236         if (qemu_lockcnt_count(&ctx->list_lock)) {
237             node->deleted = 1;
238             node->pfd.revents = 0;
239         } else {
240             /* Otherwise, delete it for real.  We can't just mark it as
241              * deleted because deleted nodes are only cleaned up while
242              * no one is walking the handlers list.
243              */
244             QLIST_REMOVE(node, node);
245             deleted = true;
246         }
247 
248         if (!node->io_poll) {
249             ctx->poll_disable_cnt--;
250         }
251     } else {
252         if (node == NULL) {
253             /* Alloc and insert if it's not already there */
254             node = g_new0(AioHandler, 1);
255             node->pfd.fd = fd;
256             QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
257 
258             g_source_add_poll(&ctx->source, &node->pfd);
259             is_new = true;
260 
261             ctx->poll_disable_cnt += !io_poll;
262         } else {
263             ctx->poll_disable_cnt += !io_poll - !node->io_poll;
264         }
265 
266         /* Update handler with latest information */
267         node->io_read = io_read;
268         node->io_write = io_write;
269         node->io_poll = io_poll;
270         node->opaque = opaque;
271         node->is_external = is_external;
272 
273         node->pfd.events = (io_read ? G_IO_IN | G_IO_HUP | G_IO_ERR : 0);
274         node->pfd.events |= (io_write ? G_IO_OUT | G_IO_ERR : 0);
275     }
276 
277     aio_epoll_update(ctx, node, is_new);
278     qemu_lockcnt_unlock(&ctx->list_lock);
279     aio_notify(ctx);
280 
281     if (deleted) {
282         g_free(node);
283     }
284 }
285 
286 void aio_set_fd_poll(AioContext *ctx, int fd,
287                      IOHandler *io_poll_begin,
288                      IOHandler *io_poll_end)
289 {
290     AioHandler *node = find_aio_handler(ctx, fd);
291 
292     if (!node) {
293         return;
294     }
295 
296     node->io_poll_begin = io_poll_begin;
297     node->io_poll_end = io_poll_end;
298 }
299 
300 void aio_set_event_notifier(AioContext *ctx,
301                             EventNotifier *notifier,
302                             bool is_external,
303                             EventNotifierHandler *io_read,
304                             AioPollFn *io_poll)
305 {
306     aio_set_fd_handler(ctx, event_notifier_get_fd(notifier), is_external,
307                        (IOHandler *)io_read, NULL, io_poll, notifier);
308 }
309 
310 void aio_set_event_notifier_poll(AioContext *ctx,
311                                  EventNotifier *notifier,
312                                  EventNotifierHandler *io_poll_begin,
313                                  EventNotifierHandler *io_poll_end)
314 {
315     aio_set_fd_poll(ctx, event_notifier_get_fd(notifier),
316                     (IOHandler *)io_poll_begin,
317                     (IOHandler *)io_poll_end);
318 }
319 
320 static void poll_set_started(AioContext *ctx, bool started)
321 {
322     AioHandler *node;
323 
324     if (started == ctx->poll_started) {
325         return;
326     }
327 
328     ctx->poll_started = started;
329 
330     qemu_lockcnt_inc(&ctx->list_lock);
331     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
332         IOHandler *fn;
333 
334         if (node->deleted) {
335             continue;
336         }
337 
338         if (started) {
339             fn = node->io_poll_begin;
340         } else {
341             fn = node->io_poll_end;
342         }
343 
344         if (fn) {
345             fn(node->opaque);
346         }
347     }
348     qemu_lockcnt_dec(&ctx->list_lock);
349 }
350 
351 
352 bool aio_prepare(AioContext *ctx)
353 {
354     /* Poll mode cannot be used with glib's event loop, disable it. */
355     poll_set_started(ctx, false);
356 
357     return false;
358 }
359 
360 bool aio_pending(AioContext *ctx)
361 {
362     AioHandler *node;
363     bool result = false;
364 
365     /*
366      * We have to walk very carefully in case aio_set_fd_handler is
367      * called while we're walking.
368      */
369     qemu_lockcnt_inc(&ctx->list_lock);
370 
371     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
372         int revents;
373 
374         revents = node->pfd.revents & node->pfd.events;
375         if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR) && node->io_read &&
376             aio_node_check(ctx, node->is_external)) {
377             result = true;
378             break;
379         }
380         if (revents & (G_IO_OUT | G_IO_ERR) && node->io_write &&
381             aio_node_check(ctx, node->is_external)) {
382             result = true;
383             break;
384         }
385     }
386     qemu_lockcnt_dec(&ctx->list_lock);
387 
388     return result;
389 }
390 
391 static bool aio_dispatch_handlers(AioContext *ctx)
392 {
393     AioHandler *node, *tmp;
394     bool progress = false;
395 
396     QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) {
397         int revents;
398 
399         revents = node->pfd.revents & node->pfd.events;
400         node->pfd.revents = 0;
401 
402         if (!node->deleted &&
403             (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR)) &&
404             aio_node_check(ctx, node->is_external) &&
405             node->io_read) {
406             node->io_read(node->opaque);
407 
408             /* aio_notify() does not count as progress */
409             if (node->opaque != &ctx->notifier) {
410                 progress = true;
411             }
412         }
413         if (!node->deleted &&
414             (revents & (G_IO_OUT | G_IO_ERR)) &&
415             aio_node_check(ctx, node->is_external) &&
416             node->io_write) {
417             node->io_write(node->opaque);
418             progress = true;
419         }
420 
421         if (node->deleted) {
422             if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
423                 QLIST_REMOVE(node, node);
424                 g_free(node);
425                 qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
426             }
427         }
428     }
429 
430     return progress;
431 }
432 
433 void aio_dispatch(AioContext *ctx)
434 {
435     qemu_lockcnt_inc(&ctx->list_lock);
436     aio_bh_poll(ctx);
437     aio_dispatch_handlers(ctx);
438     qemu_lockcnt_dec(&ctx->list_lock);
439 
440     timerlistgroup_run_timers(&ctx->tlg);
441 }
442 
443 /* These thread-local variables are used only in a small part of aio_poll
444  * around the call to the poll() system call.  In particular they are not
445  * used while aio_poll is performing callbacks, which makes it much easier
446  * to think about reentrancy!
447  *
448  * Stack-allocated arrays would be perfect but they have size limitations;
449  * heap allocation is expensive enough that we want to reuse arrays across
450  * calls to aio_poll().  And because poll() has to be called without holding
451  * any lock, the arrays cannot be stored in AioContext.  Thread-local data
452  * has none of the disadvantages of these three options.
453  */
454 static __thread GPollFD *pollfds;
455 static __thread AioHandler **nodes;
456 static __thread unsigned npfd, nalloc;
457 static __thread Notifier pollfds_cleanup_notifier;
458 
459 static void pollfds_cleanup(Notifier *n, void *unused)
460 {
461     g_assert(npfd == 0);
462     g_free(pollfds);
463     g_free(nodes);
464     nalloc = 0;
465 }
466 
467 static void add_pollfd(AioHandler *node)
468 {
469     if (npfd == nalloc) {
470         if (nalloc == 0) {
471             pollfds_cleanup_notifier.notify = pollfds_cleanup;
472             qemu_thread_atexit_add(&pollfds_cleanup_notifier);
473             nalloc = 8;
474         } else {
475             g_assert(nalloc <= INT_MAX);
476             nalloc *= 2;
477         }
478         pollfds = g_renew(GPollFD, pollfds, nalloc);
479         nodes = g_renew(AioHandler *, nodes, nalloc);
480     }
481     nodes[npfd] = node;
482     pollfds[npfd] = (GPollFD) {
483         .fd = node->pfd.fd,
484         .events = node->pfd.events,
485     };
486     npfd++;
487 }
488 
489 static bool run_poll_handlers_once(AioContext *ctx)
490 {
491     bool progress = false;
492     AioHandler *node;
493 
494     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
495         if (!node->deleted && node->io_poll &&
496             aio_node_check(ctx, node->is_external) &&
497             node->io_poll(node->opaque)) {
498             progress = true;
499         }
500 
501         /* Caller handles freeing deleted nodes.  Don't do it here. */
502     }
503 
504     return progress;
505 }
506 
507 /* run_poll_handlers:
508  * @ctx: the AioContext
509  * @max_ns: maximum time to poll for, in nanoseconds
510  *
511  * Polls for a given time.
512  *
513  * Note that ctx->notify_me must be non-zero so this function can detect
514  * aio_notify().
515  *
516  * Note that the caller must have incremented ctx->list_lock.
517  *
518  * Returns: true if progress was made, false otherwise
519  */
520 static bool run_poll_handlers(AioContext *ctx, int64_t max_ns)
521 {
522     bool progress;
523     int64_t end_time;
524 
525     assert(ctx->notify_me);
526     assert(qemu_lockcnt_count(&ctx->list_lock) > 0);
527     assert(ctx->poll_disable_cnt == 0);
528 
529     trace_run_poll_handlers_begin(ctx, max_ns);
530 
531     end_time = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) + max_ns;
532 
533     do {
534         progress = run_poll_handlers_once(ctx);
535     } while (!progress && qemu_clock_get_ns(QEMU_CLOCK_REALTIME) < end_time);
536 
537     trace_run_poll_handlers_end(ctx, progress);
538 
539     return progress;
540 }
541 
542 /* try_poll_mode:
543  * @ctx: the AioContext
544  * @blocking: busy polling is only attempted when blocking is true
545  *
546  * ctx->notify_me must be non-zero so this function can detect aio_notify().
547  *
548  * Note that the caller must have incremented ctx->list_lock.
549  *
550  * Returns: true if progress was made, false otherwise
551  */
552 static bool try_poll_mode(AioContext *ctx, bool blocking)
553 {
554     if (blocking && ctx->poll_max_ns && ctx->poll_disable_cnt == 0) {
555         /* See qemu_soonest_timeout() uint64_t hack */
556         int64_t max_ns = MIN((uint64_t)aio_compute_timeout(ctx),
557                              (uint64_t)ctx->poll_ns);
558 
559         if (max_ns) {
560             poll_set_started(ctx, true);
561 
562             if (run_poll_handlers(ctx, max_ns)) {
563                 return true;
564             }
565         }
566     }
567 
568     poll_set_started(ctx, false);
569 
570     /* Even if we don't run busy polling, try polling once in case it can make
571      * progress and the caller will be able to avoid ppoll(2)/epoll_wait(2).
572      */
573     return run_poll_handlers_once(ctx);
574 }
575 
576 bool aio_poll(AioContext *ctx, bool blocking)
577 {
578     AioHandler *node;
579     int i;
580     int ret = 0;
581     bool progress;
582     int64_t timeout;
583     int64_t start = 0;
584 
585     /* aio_notify can avoid the expensive event_notifier_set if
586      * everything (file descriptors, bottom halves, timers) will
587      * be re-evaluated before the next blocking poll().  This is
588      * already true when aio_poll is called with blocking == false;
589      * if blocking == true, it is only true after poll() returns,
590      * so disable the optimization now.
591      */
592     if (blocking) {
593         atomic_add(&ctx->notify_me, 2);
594     }
595 
596     qemu_lockcnt_inc(&ctx->list_lock);
597 
598     if (ctx->poll_max_ns) {
599         start = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
600     }
601 
602     progress = try_poll_mode(ctx, blocking);
603     if (!progress) {
604         assert(npfd == 0);
605 
606         /* fill pollfds */
607 
608         if (!aio_epoll_enabled(ctx)) {
609             QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
610                 if (!node->deleted && node->pfd.events
611                     && aio_node_check(ctx, node->is_external)) {
612                     add_pollfd(node);
613                 }
614             }
615         }
616 
617         timeout = blocking ? aio_compute_timeout(ctx) : 0;
618 
619         /* wait until next event */
620         if (aio_epoll_check_poll(ctx, pollfds, npfd, timeout)) {
621             AioHandler epoll_handler;
622 
623             epoll_handler.pfd.fd = ctx->epollfd;
624             epoll_handler.pfd.events = G_IO_IN | G_IO_OUT | G_IO_HUP | G_IO_ERR;
625             npfd = 0;
626             add_pollfd(&epoll_handler);
627             ret = aio_epoll(ctx, pollfds, npfd, timeout);
628         } else  {
629             ret = qemu_poll_ns(pollfds, npfd, timeout);
630         }
631     }
632 
633     if (blocking) {
634         atomic_sub(&ctx->notify_me, 2);
635     }
636 
637     /* Adjust polling time */
638     if (ctx->poll_max_ns) {
639         int64_t block_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - start;
640 
641         if (block_ns <= ctx->poll_ns) {
642             /* This is the sweet spot, no adjustment needed */
643         } else if (block_ns > ctx->poll_max_ns) {
644             /* We'd have to poll for too long, poll less */
645             int64_t old = ctx->poll_ns;
646 
647             if (ctx->poll_shrink) {
648                 ctx->poll_ns /= ctx->poll_shrink;
649             } else {
650                 ctx->poll_ns = 0;
651             }
652 
653             trace_poll_shrink(ctx, old, ctx->poll_ns);
654         } else if (ctx->poll_ns < ctx->poll_max_ns &&
655                    block_ns < ctx->poll_max_ns) {
656             /* There is room to grow, poll longer */
657             int64_t old = ctx->poll_ns;
658             int64_t grow = ctx->poll_grow;
659 
660             if (grow == 0) {
661                 grow = 2;
662             }
663 
664             if (ctx->poll_ns) {
665                 ctx->poll_ns *= grow;
666             } else {
667                 ctx->poll_ns = 4000; /* start polling at 4 microseconds */
668             }
669 
670             if (ctx->poll_ns > ctx->poll_max_ns) {
671                 ctx->poll_ns = ctx->poll_max_ns;
672             }
673 
674             trace_poll_grow(ctx, old, ctx->poll_ns);
675         }
676     }
677 
678     aio_notify_accept(ctx);
679 
680     /* if we have any readable fds, dispatch event */
681     if (ret > 0) {
682         for (i = 0; i < npfd; i++) {
683             nodes[i]->pfd.revents = pollfds[i].revents;
684         }
685     }
686 
687     npfd = 0;
688 
689     progress |= aio_bh_poll(ctx);
690 
691     if (ret > 0) {
692         progress |= aio_dispatch_handlers(ctx);
693     }
694 
695     qemu_lockcnt_dec(&ctx->list_lock);
696 
697     progress |= timerlistgroup_run_timers(&ctx->tlg);
698 
699     return progress;
700 }
701 
702 void aio_context_setup(AioContext *ctx)
703 {
704     /* TODO remove this in final patch submission */
705     if (getenv("QEMU_AIO_POLL_MAX_NS")) {
706         fprintf(stderr, "The QEMU_AIO_POLL_MAX_NS environment variable has "
707                 "been replaced with -object iothread,poll-max-ns=NUM\n");
708         exit(1);
709     }
710 
711 #ifdef CONFIG_EPOLL_CREATE1
712     assert(!ctx->epollfd);
713     ctx->epollfd = epoll_create1(EPOLL_CLOEXEC);
714     if (ctx->epollfd == -1) {
715         fprintf(stderr, "Failed to create epoll instance: %s", strerror(errno));
716         ctx->epoll_available = false;
717     } else {
718         ctx->epoll_available = true;
719     }
720 #endif
721 }
722 
723 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
724                                  int64_t grow, int64_t shrink, Error **errp)
725 {
726     /* No thread synchronization here, it doesn't matter if an incorrect value
727      * is used once.
728      */
729     ctx->poll_max_ns = max_ns;
730     ctx->poll_ns = 0;
731     ctx->poll_grow = grow;
732     ctx->poll_shrink = shrink;
733 
734     aio_notify(ctx);
735 }
736