include/block/aio.h

   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  */
  13
  14 #ifndef QEMU_AIO_H
  15 #define QEMU_AIO_H
  16
  17 #include "qemu/typedefs.h"
  18 #include "qemu-common.h"
  19 #include "qemu/queue.h"
  20 #include "qemu/event_notifier.h"
  21 #include "qemu/thread.h"
  22 #include "qemu/rfifolock.h"
  23 #include "qemu/timer.h"
  24
  25 typedef struct BlockAIOCB BlockAIOCB;
  26 typedef void BlockCompletionFunc(void *opaque, int ret);
  27
  28 typedef struct AIOCBInfo {
  29     void (*cancel_async)(BlockAIOCB *acb);
  30     AioContext *(*get_aio_context)(BlockAIOCB *acb);
  31     size_t aiocb_size;
  32 } AIOCBInfo;
  33
  34 struct BlockAIOCB {
  35     const AIOCBInfo *aiocb_info;
  36     BlockDriverState *bs;
  37     BlockCompletionFunc *cb;
  38     void *opaque;
  39     int refcnt;
  40 };
  41
  42 void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs,
  43                    BlockCompletionFunc *cb, void *opaque);
  44 void qemu_aio_unref(void *p);
  45 void qemu_aio_ref(void *p);
  46
  47 typedef struct AioHandler AioHandler;
  48 typedef void QEMUBHFunc(void *opaque);
  49 typedef void IOHandler(void *opaque);
  50
  51 struct AioContext {
  52     GSource source;
  53
  54     /* Protects all fields from multi-threaded access */
  55     RFifoLock lock;
  56
  57     /* The list of registered AIO handlers */
  58     QLIST_HEAD(, AioHandler) aio_handlers;
  59
  60     /* This is a simple lock used to protect the aio_handlers list.
  61      * Specifically, it's used to ensure that no callbacks are removed while
  62      * we're walking and dispatching callbacks.
  63      */
  64     int walking_handlers;
  65
  66     /* Used to avoid unnecessary event_notifier_set calls in aio_notify.
  67      * Writes protected by lock or BQL, reads are lockless.
  68      */
  69     bool dispatching;
  70
  71     /* lock to protect between bh's adders and deleter */
  72     QemuMutex bh_lock;
  73
  74     /* Anchor of the list of Bottom Halves belonging to the context */
  75     struct QEMUBH *first_bh;
  76
  77     /* A simple lock used to protect the first_bh list, and ensure that
  78      * no callbacks are removed while we're walking and dispatching callbacks.
  79      */
  80     int walking_bh;
  81
  82     /* Used for aio_notify.  */
  83     EventNotifier notifier;
  84
  85     /* Thread pool for performing work and receiving completion callbacks */
  86     struct ThreadPool *thread_pool;
  87
  88     /* TimerLists for calling timers - one per clock type */
  89     QEMUTimerListGroup tlg;
  90 };
  91
  92 /* Used internally to synchronize aio_poll against qemu_bh_schedule.  */
  93 void aio_set_dispatching(AioContext *ctx, bool dispatching);
  94
  95 /**
  96  * aio_context_new: Allocate a new AioContext.
  97  *
  98  * AioContext provide a mini event-loop that can be waited on synchronously.
  99  * They also provide bottom halves, a service to execute a piece of code
 100  * as soon as possible.
 101  */
 102 AioContext *aio_context_new(Error **errp);
 103
 104 /**
 105  * aio_context_ref:
 106  * @ctx: The AioContext to operate on.
 107  *
 108  * Add a reference to an AioContext.
 109  */
 110 void aio_context_ref(AioContext *ctx);
 111
 112 /**
 113  * aio_context_unref:
 114  * @ctx: The AioContext to operate on.
 115  *
 116  * Drop a reference to an AioContext.
 117  */
 118 void aio_context_unref(AioContext *ctx);
 119
 120 /* Take ownership of the AioContext.  If the AioContext will be shared between
 121  * threads, and a thread does not want to be interrupted, it will have to
 122  * take ownership around calls to aio_poll().  Otherwise, aio_poll()
 123  * automatically takes care of calling aio_context_acquire and
 124  * aio_context_release.
 125  *
 126  * Access to timers and BHs from a thread that has not acquired AioContext
 127  * is possible.  Access to callbacks for now must be done while the AioContext
 128  * is owned by the thread (FIXME).
 129  */
 130 void aio_context_acquire(AioContext *ctx);
 131
 132 /* Relinquish ownership of the AioContext. */
 133 void aio_context_release(AioContext *ctx);
 134
 135 /**
 136  * aio_bh_new: Allocate a new bottom half structure.
 137  *
 138  * Bottom halves are lightweight callbacks whose invocation is guaranteed
 139  * to be wait-free, thread-safe and signal-safe.  The #QEMUBH structure
 140  * is opaque and must be allocated prior to its use.
 141  */
 142 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque);
 143
 144 /**
 145  * aio_notify: Force processing of pending events.
 146  *
 147  * Similar to signaling a condition variable, aio_notify forces
 148  * aio_wait to exit, so that the next call will re-examine pending events.
 149  * The caller of aio_notify will usually call aio_wait again very soon,
 150  * or go through another iteration of the GLib main loop.  Hence, aio_notify
 151  * also has the side effect of recalculating the sets of file descriptors
 152  * that the main loop waits for.
 153  *
 154  * Calling aio_notify is rarely necessary, because for example scheduling
 155  * a bottom half calls it already.
 156  */
 157 void aio_notify(AioContext *ctx);
 158
 159 /**
 160  * aio_bh_poll: Poll bottom halves for an AioContext.
 161  *
 162  * These are internal functions used by the QEMU main loop.
 163  * And notice that multiple occurrences of aio_bh_poll cannot
 164  * be called concurrently
 165  */
 166 int aio_bh_poll(AioContext *ctx);
 167
 168 /**
 169  * qemu_bh_schedule: Schedule a bottom half.
 170  *
 171  * Scheduling a bottom half interrupts the main loop and causes the
 172  * execution of the callback that was passed to qemu_bh_new.
 173  *
 174  * Bottom halves that are scheduled from a bottom half handler are instantly
 175  * invoked.  This can create an infinite loop if a bottom half handler
 176  * schedules itself.
 177  *
 178  * @bh: The bottom half to be scheduled.
 179  */
 180 void qemu_bh_schedule(QEMUBH *bh);
 181
 182 /**
 183  * qemu_bh_cancel: Cancel execution of a bottom half.
 184  *
 185  * Canceling execution of a bottom half undoes the effect of calls to
 186  * qemu_bh_schedule without freeing its resources yet.  While cancellation
 187  * itself is also wait-free and thread-safe, it can of course race with the
 188  * loop that executes bottom halves unless you are holding the iothread
 189  * mutex.  This makes it mostly useless if you are not holding the mutex.
 190  *
 191  * @bh: The bottom half to be canceled.
 192  */
 193 void qemu_bh_cancel(QEMUBH *bh);
 194
 195 /**
 196  *qemu_bh_delete: Cancel execution of a bottom half and free its resources.
 197  *
 198  * Deleting a bottom half frees the memory that was allocated for it by
 199  * qemu_bh_new.  It also implies canceling the bottom half if it was
 200  * scheduled.
 201  * This func is async. The bottom half will do the delete action at the finial
 202  * end.
 203  *
 204  * @bh: The bottom half to be deleted.
 205  */
 206 void qemu_bh_delete(QEMUBH *bh);
 207
 208 /* Return whether there are any pending callbacks from the GSource
 209  * attached to the AioContext, before g_poll is invoked.
 210  *
 211  * This is used internally in the implementation of the GSource.
 212  */
 213 bool aio_prepare(AioContext *ctx);
 214
 215 /* Return whether there are any pending callbacks from the GSource
 216  * attached to the AioContext, after g_poll is invoked.
 217  *
 218  * This is used internally in the implementation of the GSource.
 219  */
 220 bool aio_pending(AioContext *ctx);
 221
 222 /* Dispatch any pending callbacks from the GSource attached to the AioContext.
 223  *
 224  * This is used internally in the implementation of the GSource.
 225  */
 226 bool aio_dispatch(AioContext *ctx);
 227
 228 /* Progress in completing AIO work to occur.  This can issue new pending
 229  * aio as a result of executing I/O completion or bh callbacks.
 230  *
 231  * Return whether any progress was made by executing AIO or bottom half
 232  * handlers.  If @blocking == true, this should always be true except
 233  * if someone called aio_notify.
 234  *
 235  * If there are no pending bottom halves, but there are pending AIO
 236  * operations, it may not be possible to make any progress without
 237  * blocking.  If @blocking is true, this function will wait until one
 238  * or more AIO events have completed, to ensure something has moved
 239  * before returning.
 240  */
 241 bool aio_poll(AioContext *ctx, bool blocking);
 242
 243 /* Register a file descriptor and associated callbacks.  Behaves very similarly
 244  * to qemu_set_fd_handler.  Unlike qemu_set_fd_handler, these callbacks will
 245  * be invoked when using aio_poll().
 246  *
 247  * Code that invokes AIO completion functions should rely on this function
 248  * instead of qemu_set_fd_handler[2].
 249  */
 250 void aio_set_fd_handler(AioContext *ctx,
 251                         int fd,
 252                         IOHandler *io_read,
 253                         IOHandler *io_write,
 254                         void *opaque);
 255
 256 /* Register an event notifier and associated callbacks.  Behaves very similarly
 257  * to event_notifier_set_handler.  Unlike event_notifier_set_handler, these callbacks
 258  * will be invoked when using aio_poll().
 259  *
 260  * Code that invokes AIO completion functions should rely on this function
 261  * instead of event_notifier_set_handler.
 262  */
 263 void aio_set_event_notifier(AioContext *ctx,
 264                             EventNotifier *notifier,
 265                             EventNotifierHandler *io_read);
 266
 267 /* Return a GSource that lets the main loop poll the file descriptors attached
 268  * to this AioContext.
 269  */
 270 GSource *aio_get_g_source(AioContext *ctx);
 271
 272 /* Return the ThreadPool bound to this AioContext */
 273 struct ThreadPool *aio_get_thread_pool(AioContext *ctx);
 274
 275 /**
 276  * aio_timer_new:
 277  * @ctx: the aio context
 278  * @type: the clock type
 279  * @scale: the scale
 280  * @cb: the callback to call on timer expiry
 281  * @opaque: the opaque pointer to pass to the callback
 282  *
 283  * Allocate a new timer attached to the context @ctx.
 284  * The function is responsible for memory allocation.
 285  *
 286  * The preferred interface is aio_timer_init. Use that
 287  * unless you really need dynamic memory allocation.
 288  *
 289  * Returns: a pointer to the new timer
 290  */
 291 static inline QEMUTimer *aio_timer_new(AioContext *ctx, QEMUClockType type,
 292                                        int scale,
 293                                        QEMUTimerCB *cb, void *opaque)
 294 {
 295     return timer_new_tl(ctx->tlg.tl[type], scale, cb, opaque);
 296 }
 297
 298 /**
 299  * aio_timer_init:
 300  * @ctx: the aio context
 301  * @ts: the timer
 302  * @type: the clock type
 303  * @scale: the scale
 304  * @cb: the callback to call on timer expiry
 305  * @opaque: the opaque pointer to pass to the callback
 306  *
 307  * Initialise a new timer attached to the context @ctx.
 308  * The caller is responsible for memory allocation.
 309  */
 310 static inline void aio_timer_init(AioContext *ctx,
 311                                   QEMUTimer *ts, QEMUClockType type,
 312                                   int scale,
 313                                   QEMUTimerCB *cb, void *opaque)
 314 {
 315     timer_init_tl(ts, ctx->tlg.tl[type], scale, cb, opaque);
 316 }
 317
 318 /**
 319  * aio_compute_timeout:
 320  * @ctx: the aio context
 321  *
 322  * Compute the timeout that a blocking aio_poll should use.
 323  */
 324 int64_t aio_compute_timeout(AioContext *ctx);
 325
 326 #endif