struct rds_message *rds_message_map_pages(unsigned long *page_addrs, unsigned int total_len)
{
struct rds_message *rm;
unsigned int i;
int num_sgs = ceil(total_len, PAGE_SIZE);
int extra_bytes = num_sgs * sizeof(struct scatterlist);
rm = rds_message_alloc(extra_bytes, GFP_NOWAIT);
if (!rm)
return ERR_PTR(-ENOMEM);
set_bit(RDS_MSG_PAGEVEC, &rm->m_flags);
rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
rm->data.op_nents = ceil(total_len, PAGE_SIZE);
rm->data.op_sg = rds_message_alloc_sgs(rm, num_sgs);
if (!rm->data.op_sg) {
rds_message_put(rm);
return ERR_PTR(-ENOMEM);
}
for (i = 0; i < rm->data.op_nents; ++i) {
sg_set_page(&rm->data.op_sg[i],
virt_to_page(page_addrs[i]),
PAGE_SIZE, 0);
}
return rm;
}
static void rds_ib_send_unmap_rm(struct rds_ib_connection *ic,
struct rds_ib_send_work *send,
int wc_status)
{
struct rds_message *rm = send->s_rm;
rdsdebug("ic %p send %p rm %p\n", ic, send, rm);
ib_dma_unmap_sg(ic->i_cm_id->device,
rm->m_sg, rm->m_nents,
DMA_TO_DEVICE);
if (rm->m_rdma_op != NULL) {
rds_ib_send_unmap_rdma(ic, rm->m_rdma_op);
/* If the user asked for a completion notification on this
* message, we can implement three different semantics:
* 1. Notify when we received the ACK on the RDS message
* that was queued with the RDMA. This provides reliable
* notification of RDMA status at the expense of a one-way
* packet delay.
* 2. Notify when the IB stack gives us the completion event for
* the RDMA operation.
* 3. Notify when the IB stack gives us the completion event for
* the accompanying RDS messages.
* Here, we implement approach #3. To implement approach #2,
* call rds_rdma_send_complete from the cq_handler. To implement #1,
* don't call rds_rdma_send_complete at all, and fall back to the notify
* handling in the ACK processing code.
*
* Note: There's no need to explicitly sync any RDMA buffers using
* ib_dma_sync_sg_for_cpu - the completion for the RDMA
* operation itself unmapped the RDMA buffers, which takes care
* of synching.
*/
rds_ib_send_rdma_complete(rm, wc_status);
if (rm->m_rdma_op->r_write)
rds_stats_add(s_send_rdma_bytes, rm->m_rdma_op->r_bytes);
else
rds_stats_add(s_recv_rdma_bytes, rm->m_rdma_op->r_bytes);
}
/* If anyone waited for this message to get flushed out, wake
* them up now */
rds_message_unmapped(rm);
rds_message_put(rm);
send->s_rm = NULL;
}
void rds_message_inc_free(struct rds_incoming *inc)
{
struct rds_message *rm = container_of(inc, struct rds_message, m_inc);
rds_message_put(rm);
}
struct rds_message *rds_message_copy_from_user(struct iovec *first_iov,
size_t total_len)
{
unsigned long to_copy;
unsigned long iov_off;
unsigned long sg_off;
struct rds_message *rm;
struct iovec *iov;
struct scatterlist *sg;
int ret;
rm = rds_message_alloc(ceil(total_len, PAGE_SIZE), GFP_KERNEL);
if (rm == NULL) {
ret = -ENOMEM;
goto out;
}
rm->m_inc.i_hdr.h_len = cpu_to_be32(total_len);
/*
* now allocate and copy in the data payload.
*/
sg = rm->m_sg;
iov = first_iov;
iov_off = 0;
sg_off = 0; /* Dear gcc, sg->page will be null from kzalloc. */
while (total_len) {
if (sg_page(sg) == NULL) {
ret = rds_page_remainder_alloc(sg, total_len,
GFP_HIGHUSER);
if (ret)
goto out;
rm->m_nents++;
sg_off = 0;
}
while (iov_off == iov->iov_len) {
iov_off = 0;
iov++;
}
to_copy = min(iov->iov_len - iov_off, sg->length - sg_off);
to_copy = min_t(size_t, to_copy, total_len);
rdsdebug("copying %lu bytes from user iov [%p, %zu] + %lu to "
"sg [%p, %u, %u] + %lu\n",
to_copy, iov->iov_base, iov->iov_len, iov_off,
(void *)sg_page(sg), sg->offset, sg->length, sg_off);
ret = rds_page_copy_from_user(sg_page(sg), sg->offset + sg_off,
iov->iov_base + iov_off,
to_copy);
if (ret)
goto out;
iov_off += to_copy;
total_len -= to_copy;
sg_off += to_copy;
if (sg_off == sg->length)
sg++;
}
ret = 0;
out:
if (ret) {
if (rm)
rds_message_put(rm);
rm = ERR_PTR(ret);
}
return rm;
}