libceph: make method call data be a separate data item
authorAlex Elder <elder@inktank.com>
Fri, 5 Apr 2013 19:46:02 +0000 (14:46 -0500)
committerSage Weil <sage@inktank.com>
Thu, 2 May 2013 04:18:35 +0000 (21:18 -0700)
Right now the data for a method call is specified via a pointer and
length, and it's copied--along with the class and method name--into
a pagelist data item to be sent to the osd.  Instead, encode the
data in a data item separate from the class and method names.

This will allow large amounts of data to be supplied to methods
without copying.  Only rbd uses the class functionality right now,
and when it really needs this it will probably need to use a page
array rather than a page list.  But this simple implementation
demonstrates the functionality on the osd client, and that's enough
for now.

This resolves:
    http://tracker.ceph.com/issues/4104

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>
drivers/block/rbd.c
include/linux/ceph/osd_client.h
net/ceph/osd_client.c

index 6f7a52cf75c7a0c5432fa3aff635332a8a28289d..11b7987cb75f5690581c6c8c3760b6488a41cbc4 100644 (file)
@@ -1847,8 +1847,19 @@ static int rbd_obj_method_sync(struct rbd_device *rbd_dev,
                goto out;
 
        osd_req_op_cls_init(obj_request->osd_req, 0, CEPH_OSD_OP_CALL,
-                                       class_name, method_name,
-                                       outbound, outbound_size);
+                                       class_name, method_name);
+       if (outbound_size) {
+               struct ceph_pagelist *pagelist;
+
+               pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
+               if (!pagelist)
+                       goto out;
+
+               ceph_pagelist_init(pagelist);
+               ceph_pagelist_append(pagelist, outbound, outbound_size);
+               osd_req_op_cls_request_data_pagelist(obj_request->osd_req, 0,
+                                               pagelist);
+       }
        osd_req_op_cls_response_data_pages(obj_request->osd_req, 0,
                                        obj_request->pages, inbound_size,
                                        0, false, false);
index 4ec46c0ceaf76405dbeebd7c5b2551ff47474da9..2a68a7465c18db56be9e5563e902bf4216defe41 100644 (file)
@@ -92,10 +92,9 @@ struct ceph_osd_req_op {
                struct {
                        const char *class_name;
                        const char *method_name;
-                       const void *request_data;
                        struct ceph_osd_data request_info;
+                       struct ceph_osd_data request_data;
                        struct ceph_osd_data response_data;
-                       u32 request_data_len;
                        __u8 class_len;
                        __u8 method_len;
                        __u8 argc;
@@ -259,6 +258,9 @@ extern void osd_req_op_extent_osd_data_bio(struct ceph_osd_request *,
                                        struct bio *bio, size_t bio_length);
 #endif /* CONFIG_BLOCK */
 
+extern void osd_req_op_cls_request_data_pagelist(struct ceph_osd_request *,
+                                       unsigned int which,
+                                       struct ceph_pagelist *pagelist);
 extern void osd_req_op_cls_response_data_pages(struct ceph_osd_request *,
                                        unsigned int which,
                                        struct page **pages, u64 length,
@@ -267,9 +269,7 @@ extern void osd_req_op_cls_response_data_pages(struct ceph_osd_request *,
 
 extern void osd_req_op_cls_init(struct ceph_osd_request *osd_req,
                                        unsigned int which, u16 opcode,
-                                       const char *class, const char *method,
-                                       const void *request_data,
-                                       size_t request_data_size);
+                                       const char *class, const char *method);
 extern void osd_req_op_watch_init(struct ceph_osd_request *osd_req,
                                        unsigned int which, u16 opcode,
                                        u64 cookie, u64 version, int flag);
index 73227853d845e8cbd95be56aac61c4eef855fc5a..939be67199ca5fed987313fbb183eb16d66f58d0 100644 (file)
@@ -135,6 +135,16 @@ osd_req_op_cls_request_info(struct ceph_osd_request *osd_req,
 }
 EXPORT_SYMBOL(osd_req_op_cls_request_info);    /* ??? */
 
+struct ceph_osd_data *
+osd_req_op_cls_request_data(struct ceph_osd_request *osd_req,
+                       unsigned int which)
+{
+       BUG_ON(which >= osd_req->r_num_ops);
+
+       return &osd_req->r_ops[which].cls.request_data;
+}
+EXPORT_SYMBOL(osd_req_op_cls_request_data);    /* ??? */
+
 struct ceph_osd_data *
 osd_req_op_cls_response_data(struct ceph_osd_request *osd_req,
                        unsigned int which)
@@ -192,6 +202,17 @@ static void osd_req_op_cls_request_info_pagelist(
        ceph_osd_data_pagelist_init(osd_data, pagelist);
 }
 
+void osd_req_op_cls_request_data_pagelist(
+                       struct ceph_osd_request *osd_req,
+                       unsigned int which, struct ceph_pagelist *pagelist)
+{
+       struct ceph_osd_data *osd_data;
+
+       osd_data = osd_req_op_cls_request_data(osd_req, which);
+       ceph_osd_data_pagelist_init(osd_data, pagelist);
+}
+EXPORT_SYMBOL(osd_req_op_cls_request_data_pagelist);
+
 void osd_req_op_cls_response_data_pages(struct ceph_osd_request *osd_req,
                        unsigned int which, struct page **pages, u64 length,
                        u32 alignment, bool pages_from_pool, bool own_pages)
@@ -251,6 +272,7 @@ static void osd_req_op_data_release(struct ceph_osd_request *osd_req,
                break;
        case CEPH_OSD_OP_CALL:
                ceph_osd_data_release(&op->cls.request_info);
+               ceph_osd_data_release(&op->cls.request_data);
                ceph_osd_data_release(&op->cls.response_data);
                break;
        default:
@@ -492,8 +514,7 @@ void osd_req_op_extent_update(struct ceph_osd_request *osd_req,
 EXPORT_SYMBOL(osd_req_op_extent_update);
 
 void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
-                       u16 opcode, const char *class, const char *method,
-                       const void *request_data, size_t request_data_size)
+                       u16 opcode, const char *class, const char *method)
 {
        struct ceph_osd_req_op *op = osd_req_op_init(osd_req, which, opcode);
        struct ceph_pagelist *pagelist;
@@ -520,12 +541,6 @@ void osd_req_op_cls_init(struct ceph_osd_request *osd_req, unsigned int which,
        ceph_pagelist_append(pagelist, method, size);
        payload_len += size;
 
-       op->cls.request_data = request_data;
-       BUG_ON(request_data_size > (size_t) U32_MAX);
-       op->cls.request_data_len = (u32) request_data_size;
-       ceph_pagelist_append(pagelist, request_data, request_data_size);
-       payload_len += request_data_size;
-
        osd_req_op_cls_request_info_pagelist(osd_req, which, pagelist);
 
        op->cls.argc = 0;       /* currently unused */
@@ -576,7 +591,9 @@ static u64 osd_req_encode_op(struct ceph_osd_request *req,
                              struct ceph_osd_op *dst, unsigned int which)
 {
        struct ceph_osd_req_op *src;
+       struct ceph_osd_data *osd_data;
        u64 request_data_len = 0;
+       u64 data_length;
 
        BUG_ON(which >= req->r_num_ops);
        src = &req->r_ops[which];
@@ -599,22 +616,31 @@ static u64 osd_req_encode_op(struct ceph_osd_request *req,
                        cpu_to_le64(src->extent.truncate_size);
                dst->extent.truncate_seq =
                        cpu_to_le32(src->extent.truncate_seq);
+               osd_data = &src->extent.osd_data;
                if (src->op == CEPH_OSD_OP_WRITE)
-                       ceph_osdc_msg_data_add(req->r_request,
-                                               &src->extent.osd_data);
+                       ceph_osdc_msg_data_add(req->r_request, osd_data);
                else
-                       ceph_osdc_msg_data_add(req->r_reply,
-                                               &src->extent.osd_data);
+                       ceph_osdc_msg_data_add(req->r_reply, osd_data);
                break;
        case CEPH_OSD_OP_CALL:
                dst->cls.class_len = src->cls.class_len;
                dst->cls.method_len = src->cls.method_len;
-               dst->cls.indata_len = cpu_to_le32(src->cls.request_data_len);
-               ceph_osdc_msg_data_add(req->r_reply, &src->cls.response_data);
-               ceph_osdc_msg_data_add(req->r_request, &src->cls.request_info);
-               BUG_ON(src->cls.request_info.type !=
-                                       CEPH_OSD_DATA_TYPE_PAGELIST);
-               request_data_len = src->cls.request_info.pagelist->length;
+               osd_data = &src->cls.request_info;
+               ceph_osdc_msg_data_add(req->r_request, osd_data);
+               BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGELIST);
+               request_data_len = osd_data->pagelist->length;
+
+               osd_data = &src->cls.request_data;
+               data_length = ceph_osd_data_length(osd_data);
+               if (data_length) {
+                       BUG_ON(osd_data->type == CEPH_OSD_DATA_TYPE_NONE);
+                       dst->cls.indata_len = cpu_to_le32(data_length);
+                       ceph_osdc_msg_data_add(req->r_request, osd_data);
+                       src->payload_len += data_length;
+                       request_data_len += data_length;
+               }
+               osd_data = &src->cls.response_data;
+               ceph_osdc_msg_data_add(req->r_reply, osd_data);
                break;
        case CEPH_OSD_OP_STARTSYNC:
                break;