提交 146b6df6 编写于 作者: T Tom Tucker

svcrdma: Modify the RPC recv path to use FRMR when available

RPCRDMA requests that specify a read-list are fetched with RDMA_READ. Using
an FRMR to map the data sink improves NFSRDMA security on transports that
place the RDMA_READ data sink LKEY on the wire because the valid lifetime
of the MR is only the duration of the RDMA_READ. The LKEY is invalidated
when the last RDMA_READ WR completes.

Mapping the data sink also allows for very large amounts to data to be
fetched with a single WR, so if the client is also using FRMR, the entire
RPC read-list can be fetched with a single WR.
Signed-off-by: NTom Tucker <tom@opengridcomputing.com>
上级 5b180a9a
...@@ -212,6 +212,7 @@ extern int svc_rdma_post_recv(struct svcxprt_rdma *); ...@@ -212,6 +212,7 @@ extern int svc_rdma_post_recv(struct svcxprt_rdma *);
extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *); extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *); extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int); extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
extern void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt);
extern struct svc_rdma_req_map *svc_rdma_get_req_map(void); extern struct svc_rdma_req_map *svc_rdma_get_req_map(void);
extern void svc_rdma_put_req_map(struct svc_rdma_req_map *); extern void svc_rdma_put_req_map(struct svc_rdma_req_map *);
extern int svc_rdma_fastreg(struct svcxprt_rdma *, struct svc_rdma_fastreg_mr *); extern int svc_rdma_fastreg(struct svcxprt_rdma *, struct svc_rdma_fastreg_mr *);
......
...@@ -116,7 +116,7 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp, ...@@ -116,7 +116,7 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
* *
* Assumptions: * Assumptions:
* - chunk[0]->position points to pages[0] at an offset of 0 * - chunk[0]->position points to pages[0] at an offset of 0
* - pages[] is not physically or virtually contigous and consists of * - pages[] is not physically or virtually contiguous and consists of
* PAGE_SIZE elements. * PAGE_SIZE elements.
* *
* Output: * Output:
...@@ -125,7 +125,7 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp, ...@@ -125,7 +125,7 @@ static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
* chunk in the read list * chunk in the read list
* *
*/ */
static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt, static int map_read_chunks(struct svcxprt_rdma *xprt,
struct svc_rqst *rqstp, struct svc_rqst *rqstp,
struct svc_rdma_op_ctxt *head, struct svc_rdma_op_ctxt *head,
struct rpcrdma_msg *rmsgp, struct rpcrdma_msg *rmsgp,
...@@ -211,26 +211,128 @@ static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt, ...@@ -211,26 +211,128 @@ static int rdma_rcl_to_sge(struct svcxprt_rdma *xprt,
return sge_no; return sge_no;
} }
static void rdma_set_ctxt_sge(struct svcxprt_rdma *xprt, /* Map a read-chunk-list to an XDR and fast register the page-list.
struct svc_rdma_op_ctxt *ctxt, *
struct kvec *vec, * Assumptions:
u64 *sgl_offset, * - chunk[0] position points to pages[0] at an offset of 0
int count) * - pages[] will be made physically contiguous by creating a one-off memory
* region using the fastreg verb.
* - byte_count is # of bytes in read-chunk-list
* - ch_count is # of chunks in read-chunk-list
*
* Output:
* - sge array pointing into pages[] array.
* - chunk_sge array specifying sge index and count for each
* chunk in the read list
*/
static int fast_reg_read_chunks(struct svcxprt_rdma *xprt,
struct svc_rqst *rqstp,
struct svc_rdma_op_ctxt *head,
struct rpcrdma_msg *rmsgp,
struct svc_rdma_req_map *rpl_map,
struct svc_rdma_req_map *chl_map,
int ch_count,
int byte_count)
{
int page_no;
int ch_no;
u32 offset;
struct rpcrdma_read_chunk *ch;
struct svc_rdma_fastreg_mr *frmr;
int ret = 0;
frmr = svc_rdma_get_frmr(xprt);
if (IS_ERR(frmr))
return -ENOMEM;
head->frmr = frmr;
head->arg.head[0] = rqstp->rq_arg.head[0];
head->arg.tail[0] = rqstp->rq_arg.tail[0];
head->arg.pages = &head->pages[head->count];
head->hdr_count = head->count; /* save count of hdr pages */
head->arg.page_base = 0;
head->arg.page_len = byte_count;
head->arg.len = rqstp->rq_arg.len + byte_count;
head->arg.buflen = rqstp->rq_arg.buflen + byte_count;
/* Fast register the page list */
frmr->kva = page_address(rqstp->rq_arg.pages[0]);
frmr->direction = DMA_FROM_DEVICE;
frmr->access_flags = (IB_ACCESS_LOCAL_WRITE|IB_ACCESS_REMOTE_WRITE);
frmr->map_len = byte_count;
frmr->page_list_len = PAGE_ALIGN(byte_count) >> PAGE_SHIFT;
for (page_no = 0; page_no < frmr->page_list_len; page_no++) {
frmr->page_list->page_list[page_no] =
ib_dma_map_single(xprt->sc_cm_id->device,
page_address(rqstp->rq_arg.pages[page_no]),
PAGE_SIZE, DMA_TO_DEVICE);
if (ib_dma_mapping_error(xprt->sc_cm_id->device,
frmr->page_list->page_list[page_no]))
goto fatal_err;
atomic_inc(&xprt->sc_dma_used);
head->arg.pages[page_no] = rqstp->rq_arg.pages[page_no];
}
head->count += page_no;
/* rq_respages points one past arg pages */
rqstp->rq_respages = &rqstp->rq_arg.pages[page_no];
/* Create the reply and chunk maps */
offset = 0;
ch = (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
for (ch_no = 0; ch_no < ch_count; ch_no++) {
rpl_map->sge[ch_no].iov_base = frmr->kva + offset;
rpl_map->sge[ch_no].iov_len = ch->rc_target.rs_length;
chl_map->ch[ch_no].count = 1;
chl_map->ch[ch_no].start = ch_no;
offset += ch->rc_target.rs_length;
ch++;
}
ret = svc_rdma_fastreg(xprt, frmr);
if (ret)
goto fatal_err;
return ch_no;
fatal_err:
printk("svcrdma: error fast registering xdr for xprt %p", xprt);
svc_rdma_put_frmr(xprt, frmr);
return -EIO;
}
static int rdma_set_ctxt_sge(struct svcxprt_rdma *xprt,
struct svc_rdma_op_ctxt *ctxt,
struct svc_rdma_fastreg_mr *frmr,
struct kvec *vec,
u64 *sgl_offset,
int count)
{ {
int i; int i;
ctxt->count = count; ctxt->count = count;
ctxt->direction = DMA_FROM_DEVICE; ctxt->direction = DMA_FROM_DEVICE;
for (i = 0; i < count; i++) { for (i = 0; i < count; i++) {
atomic_inc(&xprt->sc_dma_used); ctxt->sge[i].length = 0; /* in case map fails */
ctxt->sge[i].addr = if (!frmr) {
ib_dma_map_single(xprt->sc_cm_id->device, ctxt->sge[i].addr =
vec[i].iov_base, vec[i].iov_len, ib_dma_map_single(xprt->sc_cm_id->device,
DMA_FROM_DEVICE); vec[i].iov_base,
vec[i].iov_len,
DMA_FROM_DEVICE);
if (ib_dma_mapping_error(xprt->sc_cm_id->device,
ctxt->sge[i].addr))
return -EINVAL;
ctxt->sge[i].lkey = xprt->sc_dma_lkey;
atomic_inc(&xprt->sc_dma_used);
} else {
ctxt->sge[i].addr = (unsigned long)vec[i].iov_base;
ctxt->sge[i].lkey = frmr->mr->lkey;
}
ctxt->sge[i].length = vec[i].iov_len; ctxt->sge[i].length = vec[i].iov_len;
ctxt->sge[i].lkey = xprt->sc_phys_mr->lkey;
*sgl_offset = *sgl_offset + vec[i].iov_len; *sgl_offset = *sgl_offset + vec[i].iov_len;
} }
return 0;
} }
static int rdma_read_max_sge(struct svcxprt_rdma *xprt, int sge_count) static int rdma_read_max_sge(struct svcxprt_rdma *xprt, int sge_count)
...@@ -278,6 +380,7 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt, ...@@ -278,6 +380,7 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt,
struct svc_rdma_op_ctxt *hdr_ctxt) struct svc_rdma_op_ctxt *hdr_ctxt)
{ {
struct ib_send_wr read_wr; struct ib_send_wr read_wr;
struct ib_send_wr inv_wr;
int err = 0; int err = 0;
int ch_no; int ch_no;
int ch_count; int ch_count;
...@@ -301,9 +404,20 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt, ...@@ -301,9 +404,20 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt,
svc_rdma_rcl_chunk_counts(ch, &ch_count, &byte_count); svc_rdma_rcl_chunk_counts(ch, &ch_count, &byte_count);
if (ch_count > RPCSVC_MAXPAGES) if (ch_count > RPCSVC_MAXPAGES)
return -EINVAL; return -EINVAL;
sge_count = rdma_rcl_to_sge(xprt, rqstp, hdr_ctxt, rmsgp,
rpl_map, chl_map, if (!xprt->sc_frmr_pg_list_len)
ch_count, byte_count); sge_count = map_read_chunks(xprt, rqstp, hdr_ctxt, rmsgp,
rpl_map, chl_map, ch_count,
byte_count);
else
sge_count = fast_reg_read_chunks(xprt, rqstp, hdr_ctxt, rmsgp,
rpl_map, chl_map, ch_count,
byte_count);
if (sge_count < 0) {
err = -EIO;
goto out;
}
sgl_offset = 0; sgl_offset = 0;
ch_no = 0; ch_no = 0;
...@@ -312,13 +426,16 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt, ...@@ -312,13 +426,16 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt,
next_sge: next_sge:
ctxt = svc_rdma_get_context(xprt); ctxt = svc_rdma_get_context(xprt);
ctxt->direction = DMA_FROM_DEVICE; ctxt->direction = DMA_FROM_DEVICE;
ctxt->frmr = hdr_ctxt->frmr;
ctxt->read_hdr = NULL;
clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
clear_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags);
/* Prepare READ WR */ /* Prepare READ WR */
memset(&read_wr, 0, sizeof read_wr); memset(&read_wr, 0, sizeof read_wr);
ctxt->wr_op = IB_WR_RDMA_READ;
read_wr.wr_id = (unsigned long)ctxt; read_wr.wr_id = (unsigned long)ctxt;
read_wr.opcode = IB_WR_RDMA_READ; read_wr.opcode = IB_WR_RDMA_READ;
ctxt->wr_op = read_wr.opcode;
read_wr.send_flags = IB_SEND_SIGNALED; read_wr.send_flags = IB_SEND_SIGNALED;
read_wr.wr.rdma.rkey = ch->rc_target.rs_handle; read_wr.wr.rdma.rkey = ch->rc_target.rs_handle;
read_wr.wr.rdma.remote_addr = read_wr.wr.rdma.remote_addr =
...@@ -327,10 +444,15 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt, ...@@ -327,10 +444,15 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt,
read_wr.sg_list = ctxt->sge; read_wr.sg_list = ctxt->sge;
read_wr.num_sge = read_wr.num_sge =
rdma_read_max_sge(xprt, chl_map->ch[ch_no].count); rdma_read_max_sge(xprt, chl_map->ch[ch_no].count);
rdma_set_ctxt_sge(xprt, ctxt, err = rdma_set_ctxt_sge(xprt, ctxt, hdr_ctxt->frmr,
&rpl_map->sge[chl_map->ch[ch_no].start], &rpl_map->sge[chl_map->ch[ch_no].start],
&sgl_offset, &sgl_offset,
read_wr.num_sge); read_wr.num_sge);
if (err) {
svc_rdma_unmap_dma(ctxt);
svc_rdma_put_context(ctxt, 0);
goto out;
}
if (((ch+1)->rc_discrim == 0) && if (((ch+1)->rc_discrim == 0) &&
(read_wr.num_sge == chl_map->ch[ch_no].count)) { (read_wr.num_sge == chl_map->ch[ch_no].count)) {
/* /*
...@@ -339,6 +461,29 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt, ...@@ -339,6 +461,29 @@ static int rdma_read_xdr(struct svcxprt_rdma *xprt,
* the client and the RPC needs to be enqueued. * the client and the RPC needs to be enqueued.
*/ */
set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
if (hdr_ctxt->frmr) {
set_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags);
/*
* Invalidate the local MR used to map the data
* sink.
*/
if (xprt->sc_dev_caps &
SVCRDMA_DEVCAP_READ_W_INV) {
read_wr.opcode =
IB_WR_RDMA_READ_WITH_INV;
ctxt->wr_op = read_wr.opcode;
read_wr.ex.invalidate_rkey =
ctxt->frmr->mr->lkey;
} else {
/* Prepare INVALIDATE WR */
memset(&inv_wr, 0, sizeof inv_wr);
inv_wr.opcode = IB_WR_LOCAL_INV;
inv_wr.send_flags = IB_SEND_SIGNALED;
inv_wr.ex.invalidate_rkey =
hdr_ctxt->frmr->mr->lkey;
read_wr.next = &inv_wr;
}
}
ctxt->read_hdr = hdr_ctxt; ctxt->read_hdr = hdr_ctxt;
} }
/* Post the read */ /* Post the read */
......
...@@ -105,7 +105,7 @@ struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt) ...@@ -105,7 +105,7 @@ struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
return ctxt; return ctxt;
} }
static void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt) void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
{ {
struct svcxprt_rdma *xprt = ctxt->xprt; struct svcxprt_rdma *xprt = ctxt->xprt;
int i; int i;
...@@ -343,9 +343,12 @@ static void process_context(struct svcxprt_rdma *xprt, ...@@ -343,9 +343,12 @@ static void process_context(struct svcxprt_rdma *xprt,
break; break;
case IB_WR_RDMA_READ: case IB_WR_RDMA_READ:
case IB_WR_RDMA_READ_WITH_INV:
if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) { if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr; struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;
BUG_ON(!read_hdr); BUG_ON(!read_hdr);
if (test_bit(RDMACTXT_F_FAST_UNREG, &ctxt->flags))
svc_rdma_put_frmr(xprt, ctxt->frmr);
spin_lock_bh(&xprt->sc_rq_dto_lock); spin_lock_bh(&xprt->sc_rq_dto_lock);
set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags); set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
list_add_tail(&read_hdr->dto_q, list_add_tail(&read_hdr->dto_q,
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册