Add a condition variable for waiting on events, and wait for it to be…

… signalled if min_nr hasn't been reached

fs/aio.c

@@ -45,6 +45,7 @@ struct aioctx *aioctx_new(int events_capacity, pid_t pid) {
     memset(aioctx_events, 0, sizeof(struct aioctx_event) * events_capacity);
 
     lock_init(&aioctx->lock);
+    cond_init(&aioctx->cond);
 
     aioctx->refcount = 1;
     aioctx->events_capacity = events_capacity;
@@ -65,6 +66,7 @@ void aioctx_retain(struct aioctx *ctx) {
 
 static void _aioctx_decrement_ref(struct aioctx *ctx) {
     if (--ctx->refcount == 0) {
+        cond_destroy(ctx->cond);
         free(ctx->events);
         free(ctx);
     } else {
@@ -143,6 +145,7 @@ void aioctx_complete_event(struct aioctx *ctx, unsigned int index, int64_t resul
         ctx->events[index].data.as_complete = data;
     }
 
+    notify_once(&ctx->cond);
     unlock(&ctx->lock);
 }
 
@@ -171,6 +174,16 @@ bool aioctx_consume_completed_event(struct aioctx *ctx, uint64_t *user_data, add
     return result;
 }
 
+int aioctx_wait_for_completion(struct aioctx *ctx, struct timespec *timeout) {
+    if (ctx == NULL) return _EINVAL;
+
+    lock(&ctx->lock);
+    int err = wait_for(&ctx->cond, &ctx->lock, timeout);
+    unlock(&ctx->lock);
+
+    return err;
+}
+
 void aioctx_lock(struct aioctx* ctx) {
     if (ctx == NULL) return;
 

fs/aio.h

@@ -90,6 +90,7 @@ struct aioctx_event {
 struct aioctx {
     atomic_uint refcount;
     lock_t lock;
+    cond_t cond;
 
     // Indicates if this context is owned by a task.
     // 
@@ -192,6 +193,9 @@ void aioctx_cancel_event(struct aioctx *ctx, unsigned int index);
 // 
 // This accepts two result parameters, whose meaning is determined solely by
 // the event opcode.
+// 
+// This also signals any threads waiting on the context that an event has been
+// completed.
 void aioctx_complete_event(struct aioctx *ctx, unsigned int index, int64_t result0, int64_t result1);
 
 // Consume a completed I/O event.
@@ -205,6 +209,19 @@ void aioctx_complete_event(struct aioctx *ctx, unsigned int index, int64_t resul
 // from the queue, and the passed-in parameters should not be used.
 bool aioctx_consume_completed_event(struct aioctx *ctx, uint64_t *user_data, addr_t *iocbp, struct aioctx_event_complete *completed_data);
 
+// Wait for an event to complete.
+// 
+// This function blocks the current thread until an event completion is posted
+// to the context, or the timeout expires. When new events are completed, this
+// function will return 0. If the timeout expired, this function will return
+// _ETIMEDOUT. Any other error codes should be sent to client code.
+// 
+// Please note that this function returning with 0 is not a guarantee that
+// `aioctx_consume_completed_event` will yield data. This function may
+// spuriously return 0 or some other thread may have claimed the event in
+// between this function returning and the other function being called.
+int aioctx_wait_for_completion(struct aioctx *ctx, struct timespec *timeout);
+
 void aioctx_lock(struct aioctx* ctx);
 void aioctx_unlock(struct aioctx* ctx);
 

kernel/aio.c

@@ -3,6 +3,7 @@
 #include "kernel/task.h"
 #include "kernel/aio.h"
 #include "kernel/fs.h"
+#include "kernel/time.h"
 #include "fs/aio.h"
 #include "fs/fd.h"
 
@@ -60,12 +61,24 @@ dword_t sys_io_destroy(dword_t ctx_id) {
     return 0;
 }
 
-dword_t sys_io_getevents(dword_t ctx_id, dword_t min_nr, dword_t nr, addr_t events, addr_t timeout) {
-    STRACE("io_getevents(0x%x, %d, %d, 0x%x, 0x%x)", ctx_id, min_nr, nr, events, timeout);
+dword_t sys_io_getevents(dword_t ctx_id, dword_t min_nr, dword_t nr, addr_t events, addr_t timeout_addr) {
+    STRACE("io_getevents(0x%x, %d, %d, 0x%x, 0x%x)", ctx_id, min_nr, nr, events, timeout_addr);
 
     struct aioctx *ctx = aioctx_table_get_and_retain(current->aioctx, ctx_id);
     if (ctx == NULL) return _EINVAL;
     if (events == 0) return _EFAULT;
+
+    struct timespec_ guest_timeout;
+    struct timespec host_timeout;
+    struct timespec *timeout = &host_timeout;
+
+    if (timeout_addr != 0) {
+        if (user_get(timeout_addr, guest_timeout)) return _EFAULT;
+        host_timeout.tv_sec = guest_timeout.sec;
+        host_timeout.tv_nsec = guest_timeout.nsec;
+    } else {
+        timeout = NULL;
+    }
 
     dword_t i = 0;
     for (i = 0; i < nr; i += 1) {
@@ -74,8 +87,13 @@ dword_t sys_io_getevents(dword_t ctx_id, dword_t min_nr, dword_t nr, addr_t even
         struct aioctx_event_complete cdata;
 
         if (!aioctx_consume_completed_event(ctx, &user_data, &iocbp, &cdata)) {
-            //TODO: Block until min_nr events recieved or timeout exceeded
-            break;
+            if (i >= min_nr) break;
+
+            int err = aioctx_wait_for_completion(ctx, timeout);
+
+            if (err === _ETIMEDOUT) break;
+            if (err < 0) return err;
+            continue;
         }
 
         uint64_t obj = (uint64_t)iocbp;