asmlinkage long sys_vmsplice(int fd, const struct iovec __user *iov,
unsigned long nr_segs, unsigned int flags)
{
struct file *file;
long error;
int fput;
error = -EBADF;
file = fget_light(fd, &fput);
if (file) {
if (file->f_mode & FMODE_WRITE)
error = do_vmsplice(file, iov, nr_segs, flags);
fput_light(file, fput);
}
return error;
}
int MPID_nem_lmt_vmsplice_start_send(MPIDI_VC_t *vc, MPID_Request *sreq, MPL_IOV r_cookie)
{
int mpi_errno = MPI_SUCCESS;
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_LMT_VMSPLICE_START_SEND);
int pipe_fd;
int complete;
struct lmt_vmsplice_node *node = NULL;
int (*reqFn)(MPIDI_VC_t *, MPID_Request *, int *);
MPIDI_CH3I_VC *vc_ch = &vc->ch;
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_LMT_VMSPLICE_START_SEND);
/* Must do this after the other side has opened for reading, otherwise we
will error out with ENXIO. This will be indicated by the receipt of a
CTS message. */
pipe_fd = open(vc_ch->lmt_copy_buf_handle, O_NONBLOCK|O_WRONLY);
MPIR_ERR_CHKANDJUMP1(pipe_fd < 0, mpi_errno, MPI_ERR_OTHER, "**open",
"**open %s", MPIU_Strerror(errno));
mpi_errno = populate_iov_from_req(sreq);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
/* send the first flight */
sreq->ch.vc = vc; /* XXX DJG is this already assigned? */
complete = 0;
mpi_errno = do_vmsplice(sreq, pipe_fd, sreq->dev.iov,
&sreq->dev.iov_offset, &sreq->dev.iov_count, &complete);
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
if (!complete) {
/* push for later progress */
node = MPL_malloc(sizeof(struct lmt_vmsplice_node));
node->pipe_fd = pipe_fd;
node->req = sreq;
node->next = outstanding_head;
outstanding_head = node;
++MPID_nem_local_lmt_pending;
}
fn_fail:
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_LMT_VMSPLICE_START_SEND);
return mpi_errno;
}
/* XXX DJG FIXME at some point this should poll, much like the newtcp module.
But then we have that whole pollfd array to manage, which we don't really
need until this proof-of-concept proves itself. */
int MPID_nem_lmt_vmsplice_progress(void)
{
int mpi_errno = MPI_SUCCESS;
struct lmt_vmsplice_node *prev = NULL;
struct lmt_vmsplice_node *free_me = NULL;
struct lmt_vmsplice_node *cur = outstanding_head;
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_LMT_VMSPLICE_PROGRESS);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_LMT_VMSPLICE_PROGRESS);
while (cur) {
int complete = 0;
switch (MPIDI_Request_get_type(cur->req)) {
case MPIDI_REQUEST_TYPE_RECV:
mpi_errno = do_readv(cur->req, cur->pipe_fd, cur->req->dev.iov,
&cur->req->dev.iov_offset,
&cur->req->dev.iov_count, &complete);
/* FIXME: set the error status of the req and complete it, rather than POP */
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
break;
case MPIDI_REQUEST_TYPE_SEND:
mpi_errno = do_vmsplice(cur->req, cur->pipe_fd, cur->req->dev.iov,
&cur->req->dev.iov_offset,
&cur->req->dev.iov_count, &complete);
/* FIXME: set the error status of the req and complete it, rather than POP */
if (mpi_errno) MPIR_ERR_POP(mpi_errno);
break;
default:
MPIR_ERR_INTERNALANDJUMP(mpi_errno, "unexpected request type");
break;
}
if (complete) {
MPL_DBG_MSG(MPIDI_CH3_DBG_CHANNEL, VERBOSE, ".... complete");
/* remove the node from the list */
if (cur == outstanding_head) {
outstanding_head = cur->next;
prev = NULL;
free_me = cur;
cur = cur->next;
}
else {
prev->next = cur->next;
prev = cur;
free_me = cur;
cur = cur->next;
}
if (free_me) MPL_free(free_me);
--MPID_nem_local_lmt_pending;
}
if (!cur) break; /* we might have made cur NULL above */
prev = cur;
cur = cur->next;
}
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_LMT_VMSPLICE_PROGRESS);
return mpi_errno;
fn_fail:
goto fn_exit;
}
