FreeBSD and Linux AIO support

7 files changed, 372 insertions, 14 deletions

diff --git a/src/os/unix/ngx_file_aio_read.c b/src/os/unix/ngx_file_aio_read.c
new file mode 100644
index 000000000..bf089ec5d
--- /dev/null
+++ b/src/os/unix/ngx_file_aio_read.c
@@ -0,0 +1,210 @@
+
+/*
+ * Copyright (C) Igor Sysoev
+ */
+
+
+#include <ngx_config.h>
+#include <ngx_core.h>
+#include <ngx_event.h>
+
+
+/*
+ * FreeBSD file AIO features and quirks:
+ *
+ *    if an asked data are already in VM cache, then aio_error() returns 0,
+ *    and the data are already copied in buffer;
+ *
+ *    aio_read() preread in VM cache as minimum 32K;
+ *
+ *    aio_read/aio_error() may return EINPROGRESS for just written data;
+ *
+ *    kqueue EVFILT_AIO filter is level triggered only: an event repeats
+ *    until aio_return() will be called;
+ *
+ *    aio_cancel() can not cancel file AIO: it returns AIO_NOTCANCELED always.
+ */
+
+
+extern int  ngx_kqueue;
+
+
+static ssize_t ngx_file_aio_result(ngx_file_t *file, ngx_event_aio_t *aio,
+    ngx_event_t *ev);
+static void ngx_file_aio_event_handler(ngx_event_t *ev);
+
+
+ssize_t
+ngx_file_aio_read(ngx_file_t *file, u_char *buf, size_t size, off_t offset,
+    ngx_pool_t *pool)
+{
+    int                 n;
+    ngx_event_t        *ev;
+    ngx_event_aio_t    *aio;
+    static ngx_uint_t   enosys = 0;
+
+    if (enosys) {
+        return ngx_read_file(file, buf, size, offset);
+    }
+
+    aio = file->aio;
+
+    if (aio == NULL) {
+        aio = ngx_pcalloc(pool, sizeof(ngx_event_aio_t));
+        if (aio == NULL) {
+            return NGX_ERROR;
+        }
+
+        aio->file = file;
+        aio->fd = file->fd;
+        aio->event.data = aio;
+        aio->event.ready = 1;
+        aio->event.log = file->log;
+        file->aio = aio;
+    }
+
+    ev = &aio->event;
+
+    if (!ev->ready) {
+        ngx_log_error(NGX_LOG_ALERT, file->log, 0,
+                      "second aio post for \"%V\"", &file->name);
+        return NGX_AGAIN;
+    }
+
+    ngx_log_debug4(NGX_LOG_DEBUG_CORE, file->log, 0,
+                   "aio complete:%d @%O:%z %V",
+                   ev->complete, offset, size, &file->name);
+
+    if (ev->complete) {
+        ev->complete = 0;
+        ngx_set_errno(aio->err);
+
+        if (aio->err == 0) {
+            return aio->nbytes;
+        }
+
+        return NGX_ERROR;
+    }
+
+    ngx_memzero(&aio->aiocb, sizeof(struct aiocb));
+
+    aio->aiocb.aio_fildes = file->fd;
+    aio->aiocb.aio_offset = offset;
+    aio->aiocb.aio_buf = buf;
+    aio->aiocb.aio_nbytes = size;
+#if (NGX_HAVE_KQUEUE)
+    aio->aiocb.aio_sigevent.sigev_notify_kqueue = ngx_kqueue;
+    aio->aiocb.aio_sigevent.sigev_notify = SIGEV_KEVENT;
+    aio->aiocb.aio_sigevent.sigev_value.sigval_ptr = ev;
+#endif
+    ev->handler = ngx_file_aio_event_handler;
+
+    n = aio_read(&aio->aiocb);
+
+    if (n == -1) {
+        n = ngx_errno;
+
+        if (n == NGX_EAGAIN) {
+            return ngx_read_file(file, buf, size, offset);
+        }
+
+        ngx_log_error(NGX_LOG_CRIT, file->log, n,
+                      "aio_read(\"%V\") failed", &file->name);
+
+        if (n == NGX_ENOSYS) {
+            enosys = 1;
+            return ngx_read_file(file, buf, size, offset);
+        }
+
+        return NGX_ERROR;
+    }
+
+    ngx_log_debug2(NGX_LOG_DEBUG_CORE, file->log, 0,
+                   "aio_read: fd:%d %d", file->fd, n);
+
+    ev->active = 1;
+    ev->ready = 0;
+    ev->complete = 0;
+
+    return ngx_file_aio_result(aio->file, aio, ev);
+}
+
+
+static ssize_t
+ngx_file_aio_result(ngx_file_t *file, ngx_event_aio_t *aio, ngx_event_t *ev)
+{
+    int        n;
+    ngx_err_t  err;
+
+    n = aio_error(&aio->aiocb);
+
+    ngx_log_debug2(NGX_LOG_DEBUG_CORE, file->log, 0,
+                   "aio_error: fd:%d %d", file->fd, n);
+
+    if (n == -1) {
+        err = ngx_errno;
+        aio->err = err;
+
+        ngx_log_error(NGX_LOG_ALERT, file->log, err,
+                      "aio_error(\"%V\") failed", &file->name);
+        return NGX_ERROR;
+    }
+
+    if (n != 0) {
+        if (n == NGX_EINPROGRESS) {
+            if (ev->ready) {
+                ev->ready = 0;
+                ngx_log_error(NGX_LOG_ALERT, file->log, n,
+                              "aio_read(\"%V\") still in progress",
+                              &file->name);
+            }
+
+            return NGX_AGAIN;
+        }
+
+        aio->err = n;
+        ev->ready = 0;
+
+        ngx_log_error(NGX_LOG_CRIT, file->log, n,
+                      "aio_read(\"%V\") failed", &file->name);
+        return NGX_ERROR;
+    }
+
+    n = aio_return(&aio->aiocb);
+
+    if (n == -1) {
+        err = ngx_errno;
+        aio->err = err;
+        ev->ready = 0;
+
+        ngx_log_error(NGX_LOG_ALERT, file->log, err,
+                      "aio_return(\"%V\") failed", &file->name);
+        return NGX_ERROR;
+    }
+
+    aio->err = 0;
+    aio->nbytes = n;
+    ev->ready = 1;
+    ev->active = 0;
+
+    ngx_log_debug2(NGX_LOG_DEBUG_CORE, file->log, 0,
+                   "aio_return: fd:%d %d", file->fd, n);
+
+    return n;
+}
+
+
+static void
+ngx_file_aio_event_handler(ngx_event_t *ev)
+{
+    ngx_event_aio_t  *aio;
+
+    aio = ev->data;
+
+    ngx_log_debug2(NGX_LOG_DEBUG_CORE, ev->log, 0,
+                   "aio event handler fd:%d %V", aio->fd, &aio->file->name);
+
+    if (ngx_file_aio_result(aio->file, aio, ev) != NGX_AGAIN) {
+        aio->handler(ev);
+    }
+}
diff --git a/src/os/unix/ngx_files.h b/src/os/unix/ngx_files.h
index 5395acd03..993ce8f93 100644
--- a/src/os/unix/ngx_files.h
+++ b/src/os/unix/ngx_files.h
@@ -287,4 +287,12 @@ size_t ngx_fs_bsize(u_char *name);
 #define ngx_set_stderr_n         "dup2(STDERR_FILENO)"
 
 
+#if (NGX_HAVE_FILE_AIO)
+
+ssize_t ngx_file_aio_read(ngx_file_t *file, u_char *buf, size_t size,
+    off_t offset, ngx_pool_t *pool);
+
+#endif
+
+
 #endif /* _NGX_FILES_H_INCLUDED_ */
diff --git a/src/os/unix/ngx_freebsd_config.h b/src/os/unix/ngx_freebsd_config.h
index 997581a6b..20cb66640 100644
--- a/src/os/unix/ngx_freebsd_config.h
+++ b/src/os/unix/ngx_freebsd_config.h
@@ -73,13 +73,14 @@
 #endif
 
 
-#if (NGX_HAVE_AIO)
-#include <aio.h>
+#if (NGX_HAVE_KQUEUE)
+#include <sys/event.h>
 #endif
 
 
-#if (NGX_HAVE_KQUEUE)
-#include <sys/event.h>
+#if (NGX_HAVE_FILE_AIO || NGX_HAVE_AIO)
+#include <aio.h>
+typedef struct aiocb  ngx_aiocb_t;
 #endif
 
 
diff --git a/src/os/unix/ngx_linux_aio_read.c b/src/os/unix/ngx_linux_aio_read.c
new file mode 100644
index 000000000..3b4185c8b
--- /dev/null
+++ b/src/os/unix/ngx_linux_aio_read.c
@@ -0,0 +1,131 @@
+
+/*
+ * Copyright (C) Igor Sysoev
+ */
+
+
+#include <ngx_config.h>
+#include <ngx_core.h>
+#include <ngx_event.h>
+
+
+extern int            ngx_eventfd;
+extern aio_context_t  ngx_aio_ctx;
+
+
+static void ngx_file_aio_event_handler(ngx_event_t *ev);
+
+
+static long
+io_submit(aio_context_t ctx, long n, struct iocb **paiocb)
+{
+    return syscall(SYS_io_submit, ctx, n, paiocb);
+}
+
+
+ssize_t
+ngx_file_aio_read(ngx_file_t *file, u_char *buf, size_t size, off_t offset,
+    ngx_pool_t *pool)
+{
+    long                n;
+    struct iocb        *piocb[1];
+    ngx_event_t        *ev;
+    ngx_event_aio_t    *aio;
+    static ngx_uint_t   enosys = 0;
+
+    if (enosys) {
+        return ngx_read_file(file, buf, size, offset);
+    }
+
+    aio = file->aio;
+
+    if (aio == NULL) {
+        aio = ngx_pcalloc(pool, sizeof(ngx_event_aio_t));
+        if (aio == NULL) {
+            return NGX_ERROR;
+        }
+
+        aio->file = file;
+        aio->fd = file->fd;
+        aio->event.data = aio;
+        aio->event.ready = 1;
+        aio->event.log = file->log;
+        file->aio = aio;
+    }
+
+    ev = &aio->event;
+
+    if (!ev->ready) {
+        ngx_log_error(NGX_LOG_ALERT, file->log, 0,
+                      "second aio post for \"%V\"", &file->name);
+        return NGX_AGAIN;
+    }
+
+    ngx_log_debug4(NGX_LOG_DEBUG_CORE, file->log, 0,
+                   "aio complete:%d @%O:%z %V",
+                   ev->complete, offset, size, &file->name);
+
+    if (ev->complete) {
+        ev->active = 0;
+        ev->complete = 0;
+
+        if (aio->res >= 0) {
+            ngx_set_errno(0);
+            return aio->res;
+        }
+
+        ngx_set_errno(-aio->res);
+        return NGX_ERROR;
+    }
+
+    ngx_memzero(&aio->aiocb, sizeof(struct iocb));
+
+    aio->aiocb.aio_data = (uint64_t) (uintptr_t) ev;
+    aio->aiocb.aio_lio_opcode = IOCB_CMD_PREAD;
+    aio->aiocb.aio_fildes = file->fd;
+    aio->aiocb.aio_buf = (uint64_t) (uintptr_t) buf;
+    aio->aiocb.aio_nbytes = size;
+    aio->aiocb.aio_offset = offset;
+    aio->aiocb.aio_flags = IOCB_FLAG_RESFD;
+    aio->aiocb.aio_resfd = ngx_eventfd;
+
+    ev->handler = ngx_file_aio_event_handler;
+
+    piocb[0] = &aio->aiocb;
+
+    n = io_submit(ngx_aio_ctx, 1, piocb);
+
+    if (n == 1) {
+        return NGX_AGAIN;
+    }
+
+    n = -n;
+
+    if (n == NGX_EAGAIN) {
+        return ngx_read_file(file, buf, size, offset);
+    }
+
+    ngx_log_error(NGX_LOG_CRIT, file->log, n,
+                  "io_submit(\"%V\") failed", &file->name);
+
+    if (n == NGX_ENOSYS) {
+        enosys = 1;
+        return ngx_read_file(file, buf, size, offset);
+    }
+
+    return NGX_ERROR;
+}
+
+
+static void
+ngx_file_aio_event_handler(ngx_event_t *ev)
+{
+    ngx_event_aio_t  *aio;
+
+    aio = ev->data;
+
+    ngx_log_debug2(NGX_LOG_DEBUG_CORE, ev->log, 0,
+                   "aio event handler fd:%d %V", aio->fd, &aio->file->name);
+
+    aio->handler(ev);
+}
diff --git a/src/os/unix/ngx_linux_config.h b/src/os/unix/ngx_linux_config.h
index fb978c46e..cf45e54a1 100644
--- a/src/os/unix/ngx_linux_config.h
+++ b/src/os/unix/ngx_linux_config.h
@@ -81,6 +81,13 @@ extern ssize_t sendfile(int s, int fd, int32_t *offset, size_t size);
 #endif
 
 
+#if (NGX_HAVE_FILE_AIO)
+#include <sys/syscall.h>
+#include <linux/aio_abi.h>
+typedef struct iocb  ngx_aiocb_t;
+#endif
+
+
 #define NGX_LISTEN_BACKLOG        511
 
 
diff --git a/src/os/unix/ngx_posix_config.h b/src/os/unix/ngx_posix_config.h
index 3644151ee..86a80d701 100644
--- a/src/os/unix/ngx_posix_config.h
+++ b/src/os/unix/ngx_posix_config.h
@@ -112,6 +112,12 @@
 #endif
 
 
+#if (NGX_HAVE_FILE_AIO)
+#include <aio.h>
+typedef struct aiocb  ngx_aiocb_t;
+#endif
+
+
 #define NGX_LISTEN_BACKLOG  511
 
 
diff --git a/src/os/unix/ngx_solaris_config.h b/src/os/unix/ngx_solaris_config.h
index 663f26570..ae8212d8a 100644
--- a/src/os/unix/ngx_solaris_config.h
+++ b/src/os/unix/ngx_solaris_config.h
@@ -62,16 +62,6 @@
 #endif
 
 
-#if (NGX_HAVE_SENDFILE)
-#include <sys/sendfile.h>
-#endif
-
-
-#if (NGX_HAVE_AIO)
-#include <aio.h>
-#endif
-
-
 #if (NGX_HAVE_DEVPOLL)
 #include <sys/ioctl.h>
 #include <sys/devpoll.h>
@@ -83,6 +73,11 @@
 #endif
 
 
+#if (NGX_HAVE_SENDFILE)
+#include <sys/sendfile.h>
+#endif
+
+
 #define NGX_LISTEN_BACKLOG           511