s3-aio: Use correct locking context for SMB2
[samba.git] / source3 / smbd / aio.c
index 64d7928c92bb8b329492173f63c409f82b54b0d7..eec29f664f3379ed54f0144f2c041f5bc7082b1d 100644 (file)
 #include "smbd/smbd.h"
 #include "smbd/globals.h"
 #include "../lib/util/tevent_ntstatus.h"
-
-#if defined(HAVE_AIO)
-
-/* The signal we'll use to signify aio done. */
-#ifndef RT_SIGNAL_AIO
-#define RT_SIGNAL_AIO  (SIGRTMIN+3)
-#endif
-
-#ifndef HAVE_STRUCT_SIGEVENT_SIGEV_VALUE_SIVAL_PTR
-#ifdef HAVE_STRUCT_SIGEVENT_SIGEV_VALUE_SIGVAL_PTR
-#define sival_int      sigval_int
-#define sival_ptr      sigval_ptr
-#endif
-#endif
+#include "../lib/util/tevent_unix.h"
+#include "lib/tevent_wait.h"
 
 /****************************************************************************
  The buffer we keep around whilst an aio request is in process.
 *****************************************************************************/
 
 struct aio_extra {
-       struct aio_extra *next, *prev;
-       SMB_STRUCT_AIOCB acb;
        files_struct *fsp;
        struct smb_request *smbreq;
        DATA_BLOB outbuf;
        struct lock_struct lock;
+       size_t nbyte;
+       off_t offset;
        bool write_through;
-       int (*handle_completion)(struct aio_extra *ex, int errcode);
 };
 
 /****************************************************************************
@@ -61,56 +48,8 @@ bool aio_write_through_requested(struct aio_extra *aio_ex)
        return aio_ex->write_through;
 }
 
-/****************************************************************************
- Initialize the signal handler for aio read/write.
-*****************************************************************************/
-
-static void smbd_aio_signal_handler(struct tevent_context *ev_ctx,
-                                   struct tevent_signal *se,
-                                   int signum, int count,
-                                   void *_info, void *private_data)
-{
-       siginfo_t *info = (siginfo_t *)_info;
-       struct aio_extra *aio_ex = (struct aio_extra *)
-                               info->si_value.sival_ptr;
-
-       smbd_aio_complete_aio_ex(aio_ex);
-       TALLOC_FREE(aio_ex);
-}
-
-
-bool initialize_async_io_handler(void)
-{
-       static bool tried_signal_setup = false;
-
-       if (aio_signal_event) {
-               return true;
-       }
-       if (tried_signal_setup) {
-               return false;
-       }
-       tried_signal_setup = true;
-
-       aio_signal_event = tevent_add_signal(server_event_context(),
-                                            server_event_context(),
-                                            RT_SIGNAL_AIO, SA_SIGINFO,
-                                            smbd_aio_signal_handler,
-                                            NULL);
-       if (!aio_signal_event) {
-               DEBUG(10, ("Failed to setup RT_SIGNAL_AIO handler\n"));
-               return false;
-       }
-       return true;
-}
-
-static int handle_aio_read_complete(struct aio_extra *aio_ex, int errcode);
-static int handle_aio_write_complete(struct aio_extra *aio_ex, int errcode);
-static int handle_aio_smb2_read_complete(struct aio_extra *aio_ex, int errcode);
-static int handle_aio_smb2_write_complete(struct aio_extra *aio_ex, int errcode);
-
 static int aio_extra_destructor(struct aio_extra *aio_ex)
 {
-       DLIST_REMOVE(aio_list_head, aio_ex);
        outstanding_aio_calls--;
        return 0;
 }
@@ -141,13 +80,74 @@ static struct aio_extra *create_aio_extra(TALLOC_CTX *mem_ctx,
                        return NULL;
                }
        }
-       DLIST_ADD(aio_list_head, aio_ex);
        talloc_set_destructor(aio_ex, aio_extra_destructor);
        aio_ex->fsp = fsp;
        outstanding_aio_calls++;
        return aio_ex;
 }
 
+struct aio_req_fsp_link {
+       files_struct *fsp;
+       struct tevent_req *req;
+};
+
+static int aio_del_req_from_fsp(struct aio_req_fsp_link *lnk)
+{
+       unsigned i;
+       files_struct *fsp = lnk->fsp;
+       struct tevent_req *req = lnk->req;
+
+       for (i=0; i<fsp->num_aio_requests; i++) {
+               if (fsp->aio_requests[i] == req) {
+                       break;
+               }
+       }
+       if (i == fsp->num_aio_requests) {
+               DEBUG(1, ("req %p not found in fsp %p\n", req, fsp));
+               return 0;
+       }
+       fsp->num_aio_requests -= 1;
+       fsp->aio_requests[i] = fsp->aio_requests[fsp->num_aio_requests];
+
+       if (fsp->num_aio_requests == 0) {
+               tevent_wait_done(fsp->deferred_close);
+       }
+       return 0;
+}
+
+static bool aio_add_req_to_fsp(files_struct *fsp, struct tevent_req *req)
+{
+       size_t array_len;
+       struct aio_req_fsp_link *lnk;
+
+       lnk = talloc(req, struct aio_req_fsp_link);
+       if (lnk == NULL) {
+               return false;
+       }
+
+       array_len = talloc_array_length(fsp->aio_requests);
+       if (array_len <= fsp->num_aio_requests) {
+               struct tevent_req **tmp;
+
+               tmp = talloc_realloc(
+                       fsp, fsp->aio_requests, struct tevent_req *,
+                       fsp->num_aio_requests+1);
+               if (tmp == NULL) {
+                       TALLOC_FREE(lnk);
+                       return false;
+               }
+               fsp->aio_requests = tmp;
+       }
+       fsp->aio_requests[fsp->num_aio_requests] = req;
+       fsp->num_aio_requests += 1;
+
+       lnk->fsp = fsp;
+       lnk->req = req;
+       talloc_set_destructor(lnk, aio_del_req_from_fsp);
+
+       return true;
+}
+
 static void aio_pread_smb1_done(struct tevent_req *req);
 
 /****************************************************************************
@@ -160,7 +160,6 @@ NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
                             size_t smb_maxcnt)
 {
        struct aio_extra *aio_ex;
-       SMB_STRUCT_AIOCB *a;
        size_t bufsize;
        size_t min_aio_read_size = lp_aio_read_size(SNUM(conn));
        struct tevent_req *req;
@@ -203,7 +202,6 @@ NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
                DEBUG(10,("schedule_aio_read_and_X: malloc fail.\n"));
                return NT_STATUS_NO_MEMORY;
        }
-       aio_ex->handle_completion = handle_aio_read_complete;
 
        construct_reply_common_req(smbreq, (char *)aio_ex->outbuf.data);
        srv_set_message((char *)aio_ex->outbuf.data, 12, 0, True);
@@ -219,17 +217,8 @@ NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
                return NT_STATUS_FILE_LOCK_CONFLICT;
        }
 
-       a = &aio_ex->acb;
-
-       /* Now set up the aio record for the read call. */
-
-       a->aio_fildes = fsp->fh->fd;
-       a->aio_buf = smb_buf(aio_ex->outbuf.data);
-       a->aio_nbytes = smb_maxcnt;
-       a->aio_offset = startpos;
-       a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
-       a->aio_sigevent.sigev_signo  = RT_SIGNAL_AIO;
-       a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
+       aio_ex->nbyte = smb_maxcnt;
+       aio_ex->offset = startpos;
 
        req = SMB_VFS_PREAD_SEND(aio_ex, fsp->conn->sconn->ev_ctx,
                                 fsp, smb_buf(aio_ex->outbuf.data),
@@ -243,6 +232,13 @@ NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
        }
        tevent_req_set_callback(req, aio_pread_smb1_done, aio_ex);
 
+       if (!aio_add_req_to_fsp(fsp, req)) {
+               DEBUG(1, ("Could not add req to fsp\n"));
+               SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
+               TALLOC_FREE(aio_ex);
+               return NT_STATUS_RETRY;
+       }
+
        aio_ex->smbreq = talloc_move(aio_ex, &smbreq);
 
        DEBUG(10,("schedule_aio_read_and_X: scheduled aio_read for file %s, "
@@ -296,12 +292,12 @@ static void aio_pread_smb1_done(struct tevent_req *req)
                SSVAL(outbuf,smb_vwv7, ((nread >> 16) & 1));
                SSVAL(smb_buf(outbuf), -2, nread);
 
-               aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nread;
+               aio_ex->fsp->fh->pos = aio_ex->offset + nread;
                aio_ex->fsp->fh->position_information = aio_ex->fsp->fh->pos;
 
                DEBUG( 3, ("handle_aio_read_complete file %s max=%d "
                           "nread=%d\n", fsp_str_dbg(fsp),
-                          (int)aio_ex->acb.aio_nbytes, (int)nread ) );
+                          (int)aio_ex->nbyte, (int)nread ) );
 
        }
        smb_setlen(outbuf, outsize - 4);
@@ -315,12 +311,105 @@ static void aio_pread_smb1_done(struct tevent_req *req)
 
        DEBUG(10, ("handle_aio_read_complete: scheduled aio_read completed "
                   "for file %s, offset %.0f, len = %u\n",
-                  fsp_str_dbg(fsp), (double)aio_ex->acb.aio_offset,
+                  fsp_str_dbg(fsp), (double)aio_ex->offset,
                   (unsigned int)nread));
 
        TALLOC_FREE(aio_ex);
 }
 
+struct pwrite_fsync_state {
+       struct tevent_context *ev;
+       files_struct *fsp;
+       bool write_through;
+       ssize_t nwritten;
+};
+
+static void pwrite_fsync_write_done(struct tevent_req *subreq);
+static void pwrite_fsync_sync_done(struct tevent_req *subreq);
+
+static struct tevent_req *pwrite_fsync_send(TALLOC_CTX *mem_ctx,
+                                           struct tevent_context *ev,
+                                           struct files_struct *fsp,
+                                           const void *data,
+                                           size_t n, off_t offset,
+                                           bool write_through)
+{
+       struct tevent_req *req, *subreq;
+       struct pwrite_fsync_state *state;
+
+       req = tevent_req_create(mem_ctx, &state, struct pwrite_fsync_state);
+       if (req == NULL) {
+               return NULL;
+       }
+       state->ev = ev;
+       state->fsp = fsp;
+       state->write_through = write_through;
+
+       subreq = SMB_VFS_PWRITE_SEND(state, ev, fsp, data, n, offset);
+       if (tevent_req_nomem(subreq, req)) {
+               return tevent_req_post(req, ev);
+       }
+       tevent_req_set_callback(subreq, pwrite_fsync_write_done, req);
+       return req;
+}
+
+static void pwrite_fsync_write_done(struct tevent_req *subreq)
+{
+       struct tevent_req *req = tevent_req_callback_data(
+               subreq, struct tevent_req);
+       struct pwrite_fsync_state *state = tevent_req_data(
+               req, struct pwrite_fsync_state);
+       connection_struct *conn = state->fsp->conn;
+       int err;
+       bool do_sync;
+
+       state->nwritten = SMB_VFS_PWRITE_RECV(subreq, &err);
+       TALLOC_FREE(subreq);
+       if (state->nwritten == -1) {
+               tevent_req_error(req, err);
+               return;
+       }
+
+       do_sync = (lp_strict_sync(SNUM(conn)) &&
+                  (lp_syncalways(SNUM(conn)) || state->write_through));
+       if (!do_sync) {
+               tevent_req_done(req);
+               return;
+       }
+
+       subreq = SMB_VFS_FSYNC_SEND(state, state->ev, state->fsp);
+       if (tevent_req_nomem(subreq, req)) {
+               return;
+       }
+       tevent_req_set_callback(subreq, pwrite_fsync_sync_done, req);
+}
+
+static void pwrite_fsync_sync_done(struct tevent_req *subreq)
+{
+       struct tevent_req *req = tevent_req_callback_data(
+               subreq, struct tevent_req);
+       int ret, err;
+
+       ret = SMB_VFS_FSYNC_RECV(subreq, &err);
+       TALLOC_FREE(subreq);
+       if (ret == -1) {
+               tevent_req_error(req, err);
+               return;
+       }
+       tevent_req_done(req);
+}
+
+static ssize_t pwrite_fsync_recv(struct tevent_req *req, int *perr)
+{
+       struct pwrite_fsync_state *state = tevent_req_data(
+               req, struct pwrite_fsync_state);
+
+       if (tevent_req_is_unix_error(req, perr)) {
+               return -1;
+       }
+       return state->nwritten;
+}
+
 static void aio_pwrite_smb1_done(struct tevent_req *req);
 
 /****************************************************************************
@@ -334,7 +423,6 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
                              size_t numtowrite)
 {
        struct aio_extra *aio_ex;
-       SMB_STRUCT_AIOCB *a;
        size_t bufsize;
        size_t min_aio_write_size = lp_aio_write_size(SNUM(conn));
        struct tevent_req *req;
@@ -380,7 +468,6 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
                DEBUG(0,("schedule_aio_write_and_X: malloc fail.\n"));
                return NT_STATUS_NO_MEMORY;
        }
-       aio_ex->handle_completion = handle_aio_write_complete;
        aio_ex->write_through = BITSETW(smbreq->vwv+7,0);
 
        construct_reply_common_req(smbreq, (char *)aio_ex->outbuf.data);
@@ -397,20 +484,12 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
                return NT_STATUS_FILE_LOCK_CONFLICT;
        }
 
-       a = &aio_ex->acb;
-
-       /* Now set up the aio record for the write call. */
-
-       a->aio_fildes = fsp->fh->fd;
-       a->aio_buf = discard_const_p(char, data);
-       a->aio_nbytes = numtowrite;
-       a->aio_offset = startpos;
-       a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
-       a->aio_sigevent.sigev_signo  = RT_SIGNAL_AIO;
-       a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
+       aio_ex->nbyte = numtowrite;
+       aio_ex->offset = startpos;
 
-       req = SMB_VFS_PWRITE_SEND(aio_ex, fsp->conn->sconn->ev_ctx, fsp,
-                                 data, numtowrite, startpos);
+       req = pwrite_fsync_send(aio_ex, fsp->conn->sconn->ev_ctx, fsp,
+                               data, numtowrite, startpos,
+                               aio_ex->write_through);
        if (req == NULL) {
                DEBUG(3,("schedule_aio_wrote_and_X: aio_write failed. "
                         "Error %s\n", strerror(errno) ));
@@ -420,6 +499,13 @@ NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
        }
        tevent_req_set_callback(req, aio_pwrite_smb1_done, aio_ex);
 
+       if (!aio_add_req_to_fsp(fsp, req)) {
+               DEBUG(1, ("Could not add req to fsp\n"));
+               SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
+               TALLOC_FREE(aio_ex);
+               return NT_STATUS_RETRY;
+       }
+
        aio_ex->smbreq = talloc_move(aio_ex, &smbreq);
 
        /* This should actually be improved to span the write. */
@@ -460,11 +546,11 @@ static void aio_pwrite_smb1_done(struct tevent_req *req)
                req, struct aio_extra);
        files_struct *fsp = aio_ex->fsp;
        char *outbuf = (char *)aio_ex->outbuf.data;
-       ssize_t numtowrite = aio_ex->acb.aio_nbytes;
+       ssize_t numtowrite = aio_ex->nbyte;
        ssize_t nwritten;
        int err;
 
-       nwritten = SMB_VFS_PWRITE_RECV(req, &err);
+       nwritten = pwrite_fsync_recv(req, &err);
        TALLOC_FREE(req);
 
        DEBUG(10, ("pwrite_recv returned %d, err = %s\n", (int)nwritten,
@@ -481,6 +567,8 @@ static void aio_pwrite_smb1_done(struct tevent_req *req)
        /* Unlock now we're done. */
        SMB_VFS_STRICT_UNLOCK(fsp->conn, fsp, &aio_ex->lock);
 
+       mark_file_modified(fsp);
+
        if (fsp->aio_write_behind) {
 
                if (nwritten != numtowrite) {
@@ -519,8 +607,6 @@ static void aio_pwrite_smb1_done(struct tevent_req *req)
                ERROR_NT(map_nt_error_from_unix(err));
                srv_set_message(outbuf,0,0,true);
         } else {
-               NTSTATUS status;
-
                SSVAL(outbuf,smb_vwv2,nwritten);
                SSVAL(outbuf,smb_vwv4,(nwritten>>16)&1);
                if (nwritten < (ssize_t)numtowrite) {
@@ -530,17 +616,8 @@ static void aio_pwrite_smb1_done(struct tevent_req *req)
 
                DEBUG(3,("handle_aio_write: %s, num=%d wrote=%d\n",
                         fsp_fnum_dbg(fsp), (int)numtowrite, (int)nwritten));
-               status = sync_file(fsp->conn,fsp, aio_ex->write_through);
-               if (!NT_STATUS_IS_OK(status)) {
-                       ERROR_BOTH(map_nt_error_from_unix(errno),
-                                  ERRHRD, ERRdiskfull);
-                       srv_set_message(outbuf,0,0,true);
-                       DEBUG(5, ("handle_aio_write: sync_file for %s "
-                                 "returned %s\n",
-                                 fsp_str_dbg(fsp), nt_errstr(status)));
-               }
 
-               aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nwritten;
+               aio_ex->fsp->fh->pos = aio_ex->offset + nwritten;
        }
 
        show_msg(outbuf);
@@ -554,7 +631,7 @@ static void aio_pwrite_smb1_done(struct tevent_req *req)
 
        DEBUG(10, ("handle_aio_write_complete: scheduled aio_write completed "
                   "for file %s, offset %.0f, requested %u, written = %u\n",
-                  fsp_str_dbg(fsp), (double)aio_ex->acb.aio_offset,
+                  fsp_str_dbg(fsp), (double)aio_ex->offset,
                   (unsigned int)numtowrite, (unsigned int)nwritten));
 
        TALLOC_FREE(aio_ex);
@@ -564,7 +641,6 @@ bool cancel_smb2_aio(struct smb_request *smbreq)
 {
        struct smbd_smb2_request *smb2req = smbreq->smb2req;
        struct aio_extra *aio_ex = NULL;
-       int ret;
 
        if (smb2req) {
                aio_ex = talloc_get_type(smbreq->async_priv,
@@ -579,14 +655,17 @@ bool cancel_smb2_aio(struct smb_request *smbreq)
                return false;
        }
 
-       ret = SMB_VFS_AIO_CANCEL(aio_ex->fsp, &aio_ex->acb);
-       if (ret != AIO_CANCELED) {
-               return false;
-       }
+       /*
+        * We let the aio request run. Setting fsp to NULL has the
+        * effect that the _done routines don't send anything out.
+        */
 
+       aio_ex->fsp = NULL;
        return true;
 }
 
+static void aio_pread_smb2_done(struct tevent_req *req);
+
 /****************************************************************************
  Set up an aio request from a SMB2 read call.
 *****************************************************************************/
@@ -600,9 +679,8 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
                                size_t smb_maxcnt)
 {
        struct aio_extra *aio_ex;
-       SMB_STRUCT_AIOCB *a;
        size_t min_aio_read_size = lp_aio_read_size(SNUM(conn));
-       int ret;
+       struct tevent_req *req;
 
        if (fsp->base_fsp != NULL) {
                /* No AIO on streams yet */
@@ -641,9 +719,8 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
        if (!(aio_ex = create_aio_extra(smbreq->smb2req, fsp, 0))) {
                return NT_STATUS_NO_MEMORY;
        }
-       aio_ex->handle_completion = handle_aio_smb2_read_complete;
 
-       init_strict_lock_struct(fsp, (uint64_t)smbreq->smbpid,
+       init_strict_lock_struct(fsp, fsp->op->global->open_persistent_id,
                (uint64_t)startpos, (uint64_t)smb_maxcnt, READ_LOCK,
                &aio_ex->lock);
 
@@ -653,22 +730,22 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
                return NT_STATUS_FILE_LOCK_CONFLICT;
        }
 
-       a = &aio_ex->acb;
+       aio_ex->nbyte = smb_maxcnt;
+       aio_ex->offset = startpos;
 
-       /* Now set up the aio record for the read call. */
-
-       a->aio_fildes = fsp->fh->fd;
-       a->aio_buf = preadbuf->data;
-       a->aio_nbytes = smb_maxcnt;
-       a->aio_offset = startpos;
-       a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
-       a->aio_sigevent.sigev_signo  = RT_SIGNAL_AIO;
-       a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
+       req = SMB_VFS_PREAD_SEND(aio_ex, fsp->conn->sconn->ev_ctx, fsp,
+                                preadbuf->data, smb_maxcnt, startpos);
+       if (req == NULL) {
+               DEBUG(0, ("smb2: SMB_VFS_PREAD_SEND failed. "
+                         "Error %s\n", strerror(errno)));
+               SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
+               TALLOC_FREE(aio_ex);
+               return NT_STATUS_RETRY;
+       }
+       tevent_req_set_callback(req, aio_pread_smb2_done, aio_ex);
 
-       ret = SMB_VFS_AIO_READ(fsp, a);
-       if (ret == -1) {
-               DEBUG(0,("smb2: aio_read failed. "
-                       "Error %s\n", strerror(errno) ));
+       if (!aio_add_req_to_fsp(fsp, req)) {
+               DEBUG(1, ("Could not add req to fsp\n"));
                SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
                TALLOC_FREE(aio_ex);
                return NT_STATUS_RETRY;
@@ -687,6 +764,60 @@ NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
        return NT_STATUS_OK;
 }
 
+static void aio_pread_smb2_done(struct tevent_req *req)
+{
+       struct aio_extra *aio_ex = tevent_req_callback_data(
+               req, struct aio_extra);
+       struct tevent_req *subreq = aio_ex->smbreq->smb2req->subreq;
+       files_struct *fsp = aio_ex->fsp;
+       NTSTATUS status;
+       ssize_t nread;
+       int err = 0;
+
+       nread = SMB_VFS_PREAD_RECV(req, &err);
+       TALLOC_FREE(req);
+
+       DEBUG(10, ("pread_recv returned %d, err = %s\n", (int)nread,
+                  (nread == -1) ? strerror(err) : "no error"));
+
+       if (fsp == NULL) {
+               DEBUG( 3, ("aio_pread_smb2_done: file closed whilst "
+                          "aio outstanding (mid[%llu]).\n",
+                          (unsigned long long)aio_ex->smbreq->mid));
+               TALLOC_FREE(aio_ex);
+               return;
+       }
+
+       /* Unlock now we're done. */
+       SMB_VFS_STRICT_UNLOCK(fsp->conn, fsp, &aio_ex->lock);
+
+       /* Common error or success code processing for async or sync
+          read returns. */
+
+       status = smb2_read_complete(subreq, nread, err);
+
+       if (nread > 0) {
+               fsp->fh->pos = aio_ex->offset + nread;
+               fsp->fh->position_information = fsp->fh->pos;
+       }
+
+       DEBUG(10, ("smb2: scheduled aio_read completed "
+                  "for file %s, offset %.0f, len = %u "
+                  "(errcode = %d, NTSTATUS = %s)\n",
+                  fsp_str_dbg(aio_ex->fsp),
+                  (double)aio_ex->offset,
+                  (unsigned int)nread,
+                  err, nt_errstr(status)));
+
+       if (!NT_STATUS_IS_OK(status)) {
+               tevent_req_nterror(subreq, status);
+               return;
+       }
+       tevent_req_done(subreq);
+}
+
+static void aio_pwrite_smb2_done(struct tevent_req *req);
+
 /****************************************************************************
  Set up an aio request from a SMB2write call.
 *****************************************************************************/
@@ -699,9 +830,8 @@ NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
                                bool write_through)
 {
        struct aio_extra *aio_ex = NULL;
-       SMB_STRUCT_AIOCB *a = NULL;
        size_t min_aio_write_size = lp_aio_write_size(SNUM(conn));
-       int ret;
+       struct tevent_req *req;
 
        if (fsp->base_fsp != NULL) {
                /* No AIO on streams yet */
@@ -731,14 +861,18 @@ NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
                return NT_STATUS_RETRY;
        }
 
+       if (smbreq->unread_bytes) {
+               /* Can't do async with recvfile. */
+               return NT_STATUS_RETRY;
+       }
+
        if (!(aio_ex = create_aio_extra(smbreq->smb2req, fsp, 0))) {
                return NT_STATUS_NO_MEMORY;
        }
 
-       aio_ex->handle_completion = handle_aio_smb2_write_complete;
        aio_ex->write_through = write_through;
 
-       init_strict_lock_struct(fsp, (uint64_t)smbreq->smbpid,
+       init_strict_lock_struct(fsp, fsp->op->global->open_persistent_id,
                in_offset, (uint64_t)in_data.length, WRITE_LOCK,
                &aio_ex->lock);
 
@@ -748,22 +882,23 @@ NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
                return NT_STATUS_FILE_LOCK_CONFLICT;
        }
 
-       a = &aio_ex->acb;
+       aio_ex->nbyte = in_data.length;
+       aio_ex->offset = in_offset;
 
-       /* Now set up the aio record for the write call. */
-
-       a->aio_fildes = fsp->fh->fd;
-       a->aio_buf = in_data.data;
-       a->aio_nbytes = in_data.length;
-       a->aio_offset = in_offset;
-       a->aio_sigevent.sigev_notify = SIGEV_SIGNAL;
-       a->aio_sigevent.sigev_signo  = RT_SIGNAL_AIO;
-       a->aio_sigevent.sigev_value.sival_ptr = aio_ex;
+       req = pwrite_fsync_send(aio_ex, fsp->conn->sconn->ev_ctx, fsp,
+                               in_data.data, in_data.length, in_offset,
+                               write_through);
+       if (req == NULL) {
+               DEBUG(3, ("smb2: SMB_VFS_PWRITE_SEND failed. "
+                         "Error %s\n", strerror(errno)));
+               SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
+               TALLOC_FREE(aio_ex);
+               return NT_STATUS_RETRY;
+       }
+       tevent_req_set_callback(req, aio_pwrite_smb2_done, aio_ex);
 
-       ret = SMB_VFS_AIO_WRITE(fsp, a);
-       if (ret == -1) {
-               DEBUG(3,("smb2: aio_write failed. "
-                       "Error %s\n", strerror(errno) ));
+       if (!aio_add_req_to_fsp(fsp, req)) {
+               DEBUG(1, ("Could not add req to fsp\n"));
                SMB_VFS_STRICT_UNLOCK(conn, fsp, &aio_ex->lock);
                TALLOC_FREE(aio_ex);
                return NT_STATUS_RETRY;
@@ -796,469 +931,50 @@ NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
        return NT_STATUS_OK;
 }
 
-/****************************************************************************
- Complete the read and return the data or error back to the client.
- Returns errno or zero if all ok.
-*****************************************************************************/
-
-static int handle_aio_read_complete(struct aio_extra *aio_ex, int errcode)
+static void aio_pwrite_smb2_done(struct tevent_req *req)
 {
-       int outsize;
-       char *outbuf = (char *)aio_ex->outbuf.data;
-       char *data = smb_buf(outbuf);
-       ssize_t nread = SMB_VFS_AIO_RETURN(aio_ex->fsp,&aio_ex->acb);
-
-       if (nread < 0) {
-               /* We're relying here on the fact that if the fd is
-                  closed then the aio will complete and aio_return
-                  will return an error. Hopefully this is
-                  true.... JRA. */
-
-               DEBUG( 3,( "handle_aio_read_complete: file %s nread == %d. "
-                          "Error = %s\n",
-                          fsp_str_dbg(aio_ex->fsp), (int)nread, strerror(errcode)));
-
-               ERROR_NT(map_nt_error_from_unix(errcode));
-               outsize = srv_set_message(outbuf,0,0,true);
-       } else {
-               outsize = srv_set_message(outbuf,12,nread,False);
-               SSVAL(outbuf,smb_vwv2,0xFFFF); /* Remaining - must be * -1. */
-               SSVAL(outbuf,smb_vwv5,nread);
-               SSVAL(outbuf,smb_vwv6,smb_offset(data,outbuf));
-               SSVAL(outbuf,smb_vwv7,((nread >> 16) & 1));
-               SSVAL(smb_buf(outbuf),-2,nread);
-
-               aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nread;
-               aio_ex->fsp->fh->position_information = aio_ex->fsp->fh->pos;
-
-               DEBUG( 3, ( "handle_aio_read_complete file %s max=%d "
-                           "nread=%d\n",
-                           fsp_str_dbg(aio_ex->fsp),
-                           (int)aio_ex->acb.aio_nbytes, (int)nread ) );
-
-       }
-       smb_setlen(outbuf,outsize - 4);
-       show_msg(outbuf);
-       if (!srv_send_smb(aio_ex->smbreq->sconn, outbuf,
-                       true, aio_ex->smbreq->seqnum+1,
-                       IS_CONN_ENCRYPTED(aio_ex->fsp->conn), NULL)) {
-               exit_server_cleanly("handle_aio_read_complete: srv_send_smb "
-                                   "failed.");
-       }
-
-       DEBUG(10,("handle_aio_read_complete: scheduled aio_read completed "
-                 "for file %s, offset %.0f, len = %u\n",
-                 fsp_str_dbg(aio_ex->fsp), (double)aio_ex->acb.aio_offset,
-                 (unsigned int)nread ));
-
-       return errcode;
-}
-
-/****************************************************************************
- Complete the write and return the data or error back to the client.
- Returns error code or zero if all ok.
-*****************************************************************************/
-
-static int handle_aio_write_complete(struct aio_extra *aio_ex, int errcode)
-{
-       files_struct *fsp = aio_ex->fsp;
-       char *outbuf = (char *)aio_ex->outbuf.data;
-       ssize_t numtowrite = aio_ex->acb.aio_nbytes;
-       ssize_t nwritten = SMB_VFS_AIO_RETURN(fsp,&aio_ex->acb);
-
-       if (fsp->aio_write_behind) {
-               if (nwritten != numtowrite) {
-                       if (nwritten == -1) {
-                               DEBUG(5,("handle_aio_write_complete: "
-                                        "aio_write_behind failed ! File %s "
-                                        "is corrupt ! Error %s\n",
-                                        fsp_str_dbg(fsp), strerror(errcode)));
-                       } else {
-                               DEBUG(0,("handle_aio_write_complete: "
-                                        "aio_write_behind failed ! File %s "
-                                        "is corrupt ! Wanted %u bytes but "
-                                        "only wrote %d\n", fsp_str_dbg(fsp),
-                                        (unsigned int)numtowrite,
-                                        (int)nwritten ));
-                               errcode = EIO;
-                       }
-               } else {
-                       DEBUG(10,("handle_aio_write_complete: "
-                                 "aio_write_behind completed for file %s\n",
-                                 fsp_str_dbg(fsp)));
-               }
-               /* TODO: should no return 0 in case of an error !!! */
-               return 0;
-       }
-
-       /* We don't need outsize or set_message here as we've already set the
-          fixed size length when we set up the aio call. */
-
-       if(nwritten == -1) {
-               DEBUG( 3,( "handle_aio_write: file %s wanted %u bytes. "
-                          "nwritten == %d. Error = %s\n",
-                          fsp_str_dbg(fsp), (unsigned int)numtowrite,
-                          (int)nwritten, strerror(errcode) ));
-
-               ERROR_NT(map_nt_error_from_unix(errcode));
-               srv_set_message(outbuf,0,0,true);
-        } else {
-               NTSTATUS status;
-
-               SSVAL(outbuf,smb_vwv2,nwritten);
-               SSVAL(outbuf,smb_vwv4,(nwritten>>16)&1);
-               if (nwritten < (ssize_t)numtowrite) {
-                       SCVAL(outbuf,smb_rcls,ERRHRD);
-                       SSVAL(outbuf,smb_err,ERRdiskfull);
-               }
-
-               DEBUG(3,("handle_aio_write: %s, num=%d wrote=%d\n",
-                        fsp_fnum_dbg(fsp), (int)numtowrite, (int)nwritten));
-               status = sync_file(fsp->conn,fsp, aio_ex->write_through);
-               if (!NT_STATUS_IS_OK(status)) {
-                       errcode = errno;
-                       ERROR_BOTH(map_nt_error_from_unix(errcode),
-                                  ERRHRD, ERRdiskfull);
-                       srv_set_message(outbuf,0,0,true);
-                       DEBUG(5,("handle_aio_write: sync_file for %s returned %s\n",
-                                fsp_str_dbg(fsp), nt_errstr(status)));
-               }
-
-               aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nwritten;
-
-               mark_file_modified(aio_ex->fsp);
-       }
-
-       show_msg(outbuf);
-       if (!srv_send_smb(aio_ex->smbreq->sconn, outbuf,
-                         true, aio_ex->smbreq->seqnum+1,
-                         IS_CONN_ENCRYPTED(fsp->conn),
-                         NULL)) {
-               exit_server_cleanly("handle_aio_write_complete: "
-                                   "srv_send_smb failed.");
-       }
-
-       DEBUG(10,("handle_aio_write_complete: scheduled aio_write completed "
-                 "for file %s, offset %.0f, requested %u, written = %u\n",
-                 fsp_str_dbg(fsp), (double)aio_ex->acb.aio_offset,
-                 (unsigned int)numtowrite, (unsigned int)nwritten ));
-
-       return errcode;
-}
-
-/****************************************************************************
- Complete the read and return the data or error back to the client.
- Returns errno or zero if all ok.
-*****************************************************************************/
-
-static int handle_aio_smb2_read_complete(struct aio_extra *aio_ex, int errcode)
-{
-       NTSTATUS status;
+       struct aio_extra *aio_ex = tevent_req_callback_data(
+               req, struct aio_extra);
+       ssize_t numtowrite = aio_ex->nbyte;
        struct tevent_req *subreq = aio_ex->smbreq->smb2req->subreq;
-       ssize_t nread = SMB_VFS_AIO_RETURN(aio_ex->fsp,&aio_ex->acb);
-
-       /* Common error or success code processing for async or sync
-          read returns. */
-
-       status = smb2_read_complete(subreq, nread, errcode);
-
-       if (nread > 0) {
-               aio_ex->fsp->fh->pos = aio_ex->acb.aio_offset + nread;
-               aio_ex->fsp->fh->position_information = aio_ex->fsp->fh->pos;
-       }
-
-       DEBUG(10,("smb2: scheduled aio_read completed "
-               "for file %s, offset %.0f, len = %u "
-               "(errcode = %d, NTSTATUS = %s)\n",
-               fsp_str_dbg(aio_ex->fsp),
-               (double)aio_ex->acb.aio_offset,
-               (unsigned int)nread,
-               errcode,
-               nt_errstr(status) ));
-
-       if (!NT_STATUS_IS_OK(status)) {
-               tevent_req_nterror(subreq, status);
-               return errcode;
-       }
-
-       tevent_req_done(subreq);
-       return errcode;
-}
-
-/****************************************************************************
- Complete the SMB2 write and return the data or error back to the client.
- Returns error code or zero if all ok.
-*****************************************************************************/
-
-static int handle_aio_smb2_write_complete(struct aio_extra *aio_ex, int errcode)
-{
        files_struct *fsp = aio_ex->fsp;
-       ssize_t numtowrite = aio_ex->acb.aio_nbytes;
-       ssize_t nwritten = SMB_VFS_AIO_RETURN(fsp,&aio_ex->acb);
-       struct tevent_req *subreq = aio_ex->smbreq->smb2req->subreq;
        NTSTATUS status;
+       ssize_t nwritten;
+       int err = 0;
 
-       status = smb2_write_complete(subreq, nwritten, errcode);
-
-       DEBUG(10,("smb2: scheduled aio_write completed "
-               "for file %s, offset %.0f, requested %u, "
-               "written = %u (errcode = %d, NTSTATUS = %s)\n",
-               fsp_str_dbg(fsp),
-               (double)aio_ex->acb.aio_offset,
-               (unsigned int)numtowrite,
-               (unsigned int)nwritten,
-               errcode,
-               nt_errstr(status) ));
-
-       if (!NT_STATUS_IS_OK(status)) {
-               tevent_req_nterror(subreq, status);
-               return errcode;
-       }
-
-       mark_file_modified(fsp);
-
-       tevent_req_done(subreq);
-       return errcode;
-}
-
-/****************************************************************************
- Handle any aio completion. Returns True if finished (and sets *perr if err
- was non-zero), False if not.
-*****************************************************************************/
-
-static bool handle_aio_completed(struct aio_extra *aio_ex, int *perr)
-{
-       files_struct *fsp = NULL;
-       int err;
-
-       if(!aio_ex) {
-               DEBUG(3, ("handle_aio_completed: Non-existing aio_ex passed\n"));
-               return false;
-       }
-
-       if (!aio_ex->fsp) {
-               DEBUG(3, ("handle_aio_completed: aio_ex->fsp == NULL\n"));
-               return false;
-       }
-
-       fsp = aio_ex->fsp;
+       nwritten = pwrite_fsync_recv(req, &err);
+       TALLOC_FREE(req);
 
-       /* Ensure the operation has really completed. */
-       err = SMB_VFS_AIO_ERROR(fsp, &aio_ex->acb);
-       if (err == EINPROGRESS) {
-               DEBUG(10,( "handle_aio_completed: operation mid %llu still in "
-                       "process for file %s\n",
-                       (unsigned long long)aio_ex->smbreq->mid,
-                       fsp_str_dbg(aio_ex->fsp)));
-               return False;
-       }
+       DEBUG(10, ("pwrite_recv returned %d, err = %s\n", (int)nwritten,
+                  (nwritten == -1) ? strerror(err) : "no error"));
 
-       if (err == ECANCELED) {
-               DEBUG(10,( "handle_aio_completed: operation mid %llu canceled "
-                       "for file %s\n",
-                       (unsigned long long)aio_ex->smbreq->mid,
-                       fsp_str_dbg(aio_ex->fsp)));
+       if (fsp == NULL) {
+               DEBUG( 3, ("aio_pwrite_smb2_done: file closed whilst "
+                          "aio outstanding (mid[%llu]).\n",
+                          (unsigned long long)aio_ex->smbreq->mid));
+               TALLOC_FREE(aio_ex);
+               return;
        }
 
        /* Unlock now we're done. */
        SMB_VFS_STRICT_UNLOCK(fsp->conn, fsp, &aio_ex->lock);
 
-       err = aio_ex->handle_completion(aio_ex, err);
-       if (err) {
-               *perr = err; /* Only save non-zero errors. */
-       }
-
-       return True;
-}
-
-/****************************************************************************
- Handle any aio completion inline.
-*****************************************************************************/
-
-void smbd_aio_complete_aio_ex(struct aio_extra *aio_ex)
-{
-       files_struct *fsp = NULL;
-       int ret = 0;
+       mark_file_modified(fsp);
 
-       DEBUG(10,("smbd_aio_complete_mid: mid[%llu]\n",
-               (unsigned long long)aio_ex->smbreq->mid));
+        status = smb2_write_complete_nosync(subreq, nwritten, err);
 
-       fsp = aio_ex->fsp;
-       if (fsp == NULL) {
-               /* file was closed whilst I/O was outstanding. Just
-                * ignore. */
-               DEBUG( 3,( "smbd_aio_complete_mid: file closed whilst "
-                       "aio outstanding (mid[%llu]).\n",
-                       (unsigned long long)aio_ex->smbreq->mid));
-               return;
-       }
+       DEBUG(10, ("smb2: scheduled aio_write completed "
+                  "for file %s, offset %.0f, requested %u, "
+                  "written = %u (errcode = %d, NTSTATUS = %s)\n",
+                  fsp_str_dbg(fsp),
+                  (double)aio_ex->offset,
+                  (unsigned int)numtowrite,
+                  (unsigned int)nwritten,
+                  err, nt_errstr(status)));
 
-       if (!handle_aio_completed(aio_ex, &ret)) {
+       if (!NT_STATUS_IS_OK(status)) {
+               tevent_req_nterror(subreq, status);
                return;
        }
+       tevent_req_done(subreq);
 }
-
-/****************************************************************************
- We're doing write behind and the client closed the file. Wait up to 45
- seconds (my arbitrary choice) for the aio to complete. Return 0 if all writes
- completed, errno to return if not.
-*****************************************************************************/
-
-#define SMB_TIME_FOR_AIO_COMPLETE_WAIT 45
-
-int wait_for_aio_completion(files_struct *fsp)
-{
-       struct aio_extra *aio_ex;
-       const SMB_STRUCT_AIOCB **aiocb_list;
-       int aio_completion_count = 0;
-       time_t start_time = time_mono(NULL);
-       int seconds_left;
-
-       for (seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT;
-            seconds_left >= 0;) {
-               int err = 0;
-               int i;
-               struct timespec ts;
-
-               aio_completion_count = 0;
-               for( aio_ex = aio_list_head; aio_ex; aio_ex = aio_ex->next) {
-                       if (aio_ex->fsp == fsp) {
-                               aio_completion_count++;
-                       }
-               }
-
-               if (!aio_completion_count) {
-                       return 0;
-               }
-
-               DEBUG(3,("wait_for_aio_completion: waiting for %d aio events "
-                        "to complete.\n", aio_completion_count ));
-
-               aiocb_list = SMB_MALLOC_ARRAY(const SMB_STRUCT_AIOCB *,
-                                             aio_completion_count);
-               if (!aiocb_list) {
-                       return ENOMEM;
-               }
-
-               for( i = 0, aio_ex = aio_list_head;
-                    aio_ex;
-                    aio_ex = aio_ex->next) {
-                       if (aio_ex->fsp == fsp) {
-                               aiocb_list[i++] = &aio_ex->acb;
-                       }
-               }
-
-               /* Now wait up to seconds_left for completion. */
-               ts.tv_sec = seconds_left;
-               ts.tv_nsec = 0;
-
-               DEBUG(10,("wait_for_aio_completion: %d events, doing a wait "
-                         "of %d seconds.\n",
-                         aio_completion_count, seconds_left ));
-
-               err = SMB_VFS_AIO_SUSPEND(fsp, aiocb_list,
-                                         aio_completion_count, &ts);
-
-               DEBUG(10,("wait_for_aio_completion: returned err = %d, "
-                         "errno = %s\n", err, strerror(errno) ));
-
-               if (err == -1 && errno == EAGAIN) {
-                       DEBUG(0,("wait_for_aio_completion: aio_suspend timed "
-                                "out waiting for %d events after a wait of "
-                                "%d seconds\n", aio_completion_count,
-                                seconds_left));
-                       /* Timeout. */
-                       SAFE_FREE(aiocb_list);
-                       /* We're hosed here - IO may complete
-                          and trample over memory if we free
-                          the aio_ex struct, but if we don't
-                          we leak IO requests. I think smb_panic()
-                          if the right thing to do here. JRA.
-                       */
-                       smb_panic("AIO suspend timed out - cannot continue.");
-                       return EIO;
-               }
-
-               /* One or more events might have completed - process them if
-                * so. */
-               for( i = 0; i < aio_completion_count; i++) {
-                       aio_ex = (struct aio_extra *)aiocb_list[i]->aio_sigevent.sigev_value.sival_ptr;
-
-                       if (!handle_aio_completed(aio_ex, &err)) {
-                               continue;
-                       }
-                       TALLOC_FREE(aio_ex);
-               }
-
-               SAFE_FREE(aiocb_list);
-               seconds_left = SMB_TIME_FOR_AIO_COMPLETE_WAIT
-                       - (time_mono(NULL) - start_time);
-       }
-
-       /* We timed out - we don't know why. Return ret if already an error,
-        * else EIO. */
-       DEBUG(10,("wait_for_aio_completion: aio_suspend timed out waiting "
-                 "for %d events\n",
-                 aio_completion_count));
-
-       return EIO;
-}
-
-#else
-
-bool initialize_async_io_handler(void)
-{
-       return false;
-}
-
-NTSTATUS schedule_aio_read_and_X(connection_struct *conn,
-                            struct smb_request *smbreq,
-                            files_struct *fsp, off_t startpos,
-                            size_t smb_maxcnt)
-{
-       return NT_STATUS_RETRY;
-}
-
-NTSTATUS schedule_aio_write_and_X(connection_struct *conn,
-                             struct smb_request *smbreq,
-                             files_struct *fsp, const char *data,
-                             off_t startpos,
-                             size_t numtowrite)
-{
-       return NT_STATUS_RETRY;
-}
-
-bool cancel_smb2_aio(struct smb_request *smbreq)
-{
-       return false;
-}
-
-NTSTATUS schedule_smb2_aio_read(connection_struct *conn,
-                                struct smb_request *smbreq,
-                                files_struct *fsp,
-                               TALLOC_CTX *ctx,
-                               DATA_BLOB *preadbuf,
-                                off_t startpos,
-                                size_t smb_maxcnt)
-{
-       return NT_STATUS_RETRY;
-}
-
-NTSTATUS schedule_aio_smb2_write(connection_struct *conn,
-                               struct smb_request *smbreq,
-                               files_struct *fsp,
-                               uint64_t in_offset,
-                               DATA_BLOB in_data,
-                               bool write_through)
-{
-       return NT_STATUS_RETRY;
-}
-
-int wait_for_aio_completion(files_struct *fsp)
-{
-       return 0;
-}
-
-void smbd_aio_complete_mid(uint64_t mid);
-
-#endif