aio-posix.c

   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-common.h"
  17 #include "block/block.h"
  18 #include "qemu/queue.h"
  19 #include "qemu/sockets.h"
  20
  21 struct AioHandler
  22 {
  23     GPollFD pfd;
  24     IOHandler *io_read;
  25     IOHandler *io_write;
  26     AioFlushHandler *io_flush;
  27     int deleted;
  28     void *opaque;
  29     QLIST_ENTRY(AioHandler) node;
  30 };
  31
  32 static AioHandler *find_aio_handler(AioContext *ctx, int fd)
  33 {
  34     AioHandler *node;
  35
  36     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
  37         if (node->pfd.fd == fd)
  38             if (!node->deleted)
  39                 return node;
  40     }
  41
  42     return NULL;
  43 }
  44
  45 void aio_set_fd_handler(AioContext *ctx,
  46                         int fd,
  47                         IOHandler *io_read,
  48                         IOHandler *io_write,
  49                         AioFlushHandler *io_flush,
  50                         void *opaque)
  51 {
  52     AioHandler *node;
  53
  54     node = find_aio_handler(ctx, fd);
  55
  56     /* Are we deleting the fd handler? */
  57     if (!io_read && !io_write) {
  58         if (node) {
  59             g_source_remove_poll(&ctx->source, &node->pfd);
  60
  61             /* If the lock is held, just mark the node as deleted */
  62             if (ctx->walking_handlers) {
  63                 node->deleted = 1;
  64                 node->pfd.revents = 0;
  65             } else {
  66                 /* Otherwise, delete it for real.  We can't just mark it as
  67                  * deleted because deleted nodes are only cleaned up after
  68                  * releasing the walking_handlers lock.
  69                  */
  70                 QLIST_REMOVE(node, node);
  71                 g_free(node);
  72             }
  73         }
  74     } else {
  75         if (node == NULL) {
  76             /* Alloc and insert if it's not already there */
  77             node = g_malloc0(sizeof(AioHandler));
  78             node->pfd.fd = fd;
  79             QLIST_INSERT_HEAD(&ctx->aio_handlers, node, node);
  80
  81             g_source_add_poll(&ctx->source, &node->pfd);
  82         }
  83         /* Update handler with latest information */
  84         node->io_read = io_read;
  85         node->io_write = io_write;
  86         node->io_flush = io_flush;
  87         node->opaque = opaque;
  88
  89         node->pfd.events = (io_read ? G_IO_IN | G_IO_HUP : 0);
  90         node->pfd.events |= (io_write ? G_IO_OUT : 0);
  91     }
  92
  93     aio_notify(ctx);
  94 }
  95
  96 void aio_set_event_notifier(AioContext *ctx,
  97                             EventNotifier *notifier,
  98                             EventNotifierHandler *io_read,
  99                             AioFlushEventNotifierHandler *io_flush)
 100 {
 101     aio_set_fd_handler(ctx, event_notifier_get_fd(notifier),
 102                        (IOHandler *)io_read, NULL,
 103                        (AioFlushHandler *)io_flush, notifier);
 104 }
 105
 106 bool aio_pending(AioContext *ctx)
 107 {
 108     AioHandler *node;
 109
 110     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
 111         int revents;
 112
 113         /*
 114          * FIXME: right now we cannot get G_IO_HUP and G_IO_ERR because
 115          * main-loop.c is still select based (due to the slirp legacy).
 116          * If main-loop.c ever switches to poll, G_IO_ERR should be
 117          * tested too.  Dispatching G_IO_ERR to both handlers should be
 118          * okay, since handlers need to be ready for spurious wakeups.
 119          */
 120         revents = node->pfd.revents & node->pfd.events;
 121         if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR) && node->io_read) {
 122             return true;
 123         }
 124         if (revents & (G_IO_OUT | G_IO_ERR) && node->io_write) {
 125             return true;
 126         }
 127     }
 128
 129     return false;
 130 }
 131
 132 bool aio_poll(AioContext *ctx, bool blocking)
 133 {
 134     static struct timeval tv0;
 135     AioHandler *node;
 136     fd_set rdfds, wrfds;
 137     int max_fd = -1;
 138     int ret;
 139     bool busy, progress;
 140
 141     progress = false;
 142
 143     /*
 144      * If there are callbacks left that have been queued, we need to call then.
 145      * Do not call select in this case, because it is possible that the caller
 146      * does not need a complete flush (as is the case for qemu_aio_wait loops).
 147      */
 148     if (aio_bh_poll(ctx)) {
 149         blocking = false;
 150         progress = true;
 151     }
 152
 153     /*
 154      * Then dispatch any pending callbacks from the GSource.
 155      *
 156      * We have to walk very carefully in case qemu_aio_set_fd_handler is
 157      * called while we're walking.
 158      */
 159     node = QLIST_FIRST(&ctx->aio_handlers);
 160     while (node) {
 161         AioHandler *tmp;
 162         int revents;
 163
 164         ctx->walking_handlers++;
 165
 166         revents = node->pfd.revents & node->pfd.events;
 167         node->pfd.revents = 0;
 168
 169         /* See comment in aio_pending.  */
 170         if (revents & (G_IO_IN | G_IO_HUP | G_IO_ERR) && node->io_read) {
 171             node->io_read(node->opaque);
 172             progress = true;
 173         }
 174         if (revents & (G_IO_OUT | G_IO_ERR) && node->io_write) {
 175             node->io_write(node->opaque);
 176             progress = true;
 177         }
 178
 179         tmp = node;
 180         node = QLIST_NEXT(node, node);
 181
 182         ctx->walking_handlers--;
 183
 184         if (!ctx->walking_handlers && tmp->deleted) {
 185             QLIST_REMOVE(tmp, node);
 186             g_free(tmp);
 187         }
 188     }
 189
 190     if (progress && !blocking) {
 191         return true;
 192     }
 193
 194     ctx->walking_handlers++;
 195
 196     FD_ZERO(&rdfds);
 197     FD_ZERO(&wrfds);
 198
 199     /* fill fd sets */
 200     busy = false;
 201     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
 202         /* If there aren't pending AIO operations, don't invoke callbacks.
 203          * Otherwise, if there are no AIO requests, qemu_aio_wait() would
 204          * wait indefinitely.
 205          */
 206         if (!node->deleted && node->io_flush) {
 207             if (node->io_flush(node->opaque) == 0) {
 208                 continue;
 209             }
 210             busy = true;
 211         }
 212         if (!node->deleted && node->io_read) {
 213             FD_SET(node->pfd.fd, &rdfds);
 214             max_fd = MAX(max_fd, node->pfd.fd + 1);
 215         }
 216         if (!node->deleted && node->io_write) {
 217             FD_SET(node->pfd.fd, &wrfds);
 218             max_fd = MAX(max_fd, node->pfd.fd + 1);
 219         }
 220     }
 221
 222     ctx->walking_handlers--;
 223
 224     /* No AIO operations?  Get us out of here */
 225     if (!busy) {
 226         return progress;
 227     }
 228
 229     /* wait until next event */
 230     ret = select(max_fd, &rdfds, &wrfds, NULL, blocking ? NULL : &tv0);
 231
 232     /* if we have any readable fds, dispatch event */
 233     if (ret > 0) {
 234         /* we have to walk very carefully in case
 235          * qemu_aio_set_fd_handler is called while we're walking */
 236         node = QLIST_FIRST(&ctx->aio_handlers);
 237         while (node) {
 238             AioHandler *tmp;
 239
 240             ctx->walking_handlers++;
 241
 242             if (!node->deleted &&
 243                 FD_ISSET(node->pfd.fd, &rdfds) &&
 244                 node->io_read) {
 245                 node->io_read(node->opaque);
 246                 progress = true;
 247             }
 248             if (!node->deleted &&
 249                 FD_ISSET(node->pfd.fd, &wrfds) &&
 250                 node->io_write) {
 251                 node->io_write(node->opaque);
 252                 progress = true;
 253             }
 254
 255             tmp = node;
 256             node = QLIST_NEXT(node, node);
 257
 258             ctx->walking_handlers--;
 259
 260             if (!ctx->walking_handlers && tmp->deleted) {
 261                 QLIST_REMOVE(tmp, node);
 262                 g_free(tmp);
 263             }
 264         }
 265     }
 266
 267     assert(progress || busy);
 268     return true;
 269 }