libceph: add data pointers in osd op structures
Alex Elder [Wed, 3 Apr 2013 06:28:58 +0000 (01:28 -0500)]
An extent type osd operation currently implies that there will
be corresponding data supplied in the data portion of the request
(for write) or response (for read) message.  Similarly, an osd class
method operation implies a data item will be supplied to receive
the response data from the operation.

Add a ceph_osd_data pointer to each of those structures, and assign
it to point to eithre the incoming or the outgoing data structure in
the osd message.  The data is not always available when an op is
initially set up, so add two new functions to allow setting them
after the op has been initialized.

Begin to make use of the data item pointer available in the osd
operation rather than the request data in or out structure in
places where it's convenient.  Add some assertions to verify
pointers are always set the way they're expected to be.

This is a sort of stepping stone toward really moving the data
into the osd request ops, to allow for some validation before
making that jump.

This is the first in a series of patches that resolve:
    http://tracker.ceph.com/issues/4657

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Josh Durgin <josh.durgin@inktank.com>

drivers/block/rbd.c
fs/ceph/addr.c
fs/ceph/file.c
include/linux/ceph/osd_client.h
net/ceph/osd_client.c

index c12b555..eb64ed0 100644 (file)
@@ -1315,23 +1315,39 @@ static void rbd_osd_req_format_op(struct rbd_obj_request *obj_request,
                                        bool write_request)
 {
        struct rbd_img_request *img_request = obj_request->img_request;
+       struct ceph_osd_request *osd_req = obj_request->osd_req;
+       struct ceph_osd_data *osd_data = NULL;
        struct ceph_snap_context *snapc = NULL;
        u64 snap_id = CEPH_NOSNAP;
        struct timespec *mtime = NULL;
        struct timespec now;
 
-       rbd_assert(obj_request->osd_req != NULL);
+       rbd_assert(osd_req != NULL);
 
        if (write_request) {
+               osd_data = &osd_req->r_data_out;
                now = CURRENT_TIME;
                mtime = &now;
                if (img_request)
                        snapc = img_request->snapc;
-       } else if (img_request) {
-               snap_id = img_request->snap_id;
+       } else {
+               osd_data = &osd_req->r_data_in;
+               if (img_request)
+                       snap_id = img_request->snap_id;
        }
+       if (obj_request->type != OBJ_REQUEST_NODATA) {
+               struct ceph_osd_req_op *op = &obj_request->osd_req->r_ops[0];
 
-       ceph_osdc_build_request(obj_request->osd_req, obj_request->offset,
+               /*
+                * If it has data, it's either a object class method
+                * call (cls) or it's an extent operation.
+                */
+               if (op->op == CEPH_OSD_OP_CALL)
+                       osd_req_op_cls_response_data(op, osd_data);
+               else
+                       osd_req_op_extent_osd_data(op, osd_data);
+       }
+       ceph_osdc_build_request(osd_req, obj_request->offset,
                        snapc, snap_id, mtime);
 }
 
index c9da074..0ac3a37 100644 (file)
@@ -343,7 +343,8 @@ static int start_read(struct inode *inode, struct list_head *page_list, int max)
                }
                pages[i] = page;
        }
-       ceph_osd_data_pages_init(&req->r_data_in, pages, len, 0,
+       BUG_ON(req->r_ops[0].extent.osd_data != &req->r_data_in);
+       ceph_osd_data_pages_init(req->r_ops[0].extent.osd_data, pages, len, 0,
                                        false, false);
        req->r_callback = finish_read;
        req->r_inode = inode;
@@ -916,8 +917,9 @@ get_more_pages:
                dout("writepages got %d pages at %llu~%llu\n",
                     locked_pages, offset, len);
 
-               ceph_osd_data_pages_init(&req->r_data_out, pages, len, 0,
-                                               !!pool, false);
+               BUG_ON(req->r_ops[0].extent.osd_data != &req->r_data_out);
+               ceph_osd_data_pages_init(req->r_ops[0].extent.osd_data, pages,
+                                               len, 0, !!pool, false);
 
                pages = NULL;   /* request message now owns the pages array */
                pool = NULL;
index a12f476..cddc10f 100644 (file)
@@ -574,8 +574,9 @@ more:
                        own_pages = true;
                }
        }
-       ceph_osd_data_pages_init(&req->r_data_out, pages, len, page_align,
-                                       false, own_pages);
+       BUG_ON(req->r_ops[0].extent.osd_data != &req->r_data_out);
+       ceph_osd_data_pages_init(req->r_ops[0].extent.osd_data, pages, len,
+                                       page_align, false, own_pages);
 
        /* BUG_ON(vino.snap != CEPH_NOSNAP); */
        ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
index a9c4089..ae51935 100644 (file)
@@ -87,12 +87,14 @@ struct ceph_osd_req_op {
                        u64 offset, length;
                        u64 truncate_size;
                        u32 truncate_seq;
+                       struct ceph_osd_data *osd_data;
                } extent;
                struct {
                        const char *class_name;
                        const char *method_name;
                        const void *request_data;
                        u32 request_data_len;
+                       struct ceph_osd_data *response_data;
                        __u8 class_len;
                        __u8 method_len;
                        __u8 argc;
@@ -236,10 +238,14 @@ extern void osd_req_op_extent_init(struct ceph_osd_req_op *op, u16 opcode,
                                        u64 offset, u64 length,
                                        u64 truncate_size, u32 truncate_seq);
 extern void osd_req_op_extent_update(struct ceph_osd_req_op *op, u64 length);
+extern void osd_req_op_extent_osd_data(struct ceph_osd_req_op *op,
+                                       struct ceph_osd_data *osd_data);
 extern void osd_req_op_cls_init(struct ceph_osd_req_op *op, u16 opcode,
                                        const char *class, const char *method,
                                        const void *request_data,
                                        size_t request_data_size);
+extern void osd_req_op_cls_response_data(struct ceph_osd_req_op *op,
+                                       struct ceph_osd_data *response_data);
 extern void osd_req_op_watch_init(struct ceph_osd_req_op *op, u16 opcode,
                                        u64 cookie, u64 version, int flag);
 
index 87fcf0b..23491e9 100644 (file)
@@ -372,6 +372,13 @@ void osd_req_op_extent_update(struct ceph_osd_req_op *op, u64 length)
 }
 EXPORT_SYMBOL(osd_req_op_extent_update);
 
+void osd_req_op_extent_osd_data(struct ceph_osd_req_op *op,
+                               struct ceph_osd_data *osd_data)
+{
+       op->extent.osd_data = osd_data;
+}
+EXPORT_SYMBOL(osd_req_op_extent_osd_data);
+
 void osd_req_op_cls_init(struct ceph_osd_req_op *op, u16 opcode,
                        const char *class, const char *method,
                        const void *request_data, size_t request_data_size)
@@ -406,6 +413,13 @@ void osd_req_op_cls_init(struct ceph_osd_req_op *op, u16 opcode,
 }
 EXPORT_SYMBOL(osd_req_op_cls_init);
 
+void osd_req_op_cls_response_data(struct ceph_osd_req_op *op,
+                               struct ceph_osd_data *response_data)
+{
+       op->cls.response_data = response_data;
+}
+EXPORT_SYMBOL(osd_req_op_cls_response_data);
+
 void osd_req_op_watch_init(struct ceph_osd_req_op *op, u16 opcode,
                                u64 cookie, u64 version, int flag)
 {
@@ -449,6 +463,10 @@ 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);
+               if (src->op == CEPH_OSD_OP_WRITE)
+                       WARN_ON(src->extent.osd_data != &req->r_data_out);
+               else
+                       WARN_ON(src->extent.osd_data != &req->r_data_in);
                break;
        case CEPH_OSD_OP_CALL:
                pagelist = kmalloc(sizeof (*pagelist), GFP_NOFS);
@@ -464,8 +482,9 @@ static u64 osd_req_encode_op(struct ceph_osd_request *req,
                                     src->cls.method_len);
                ceph_pagelist_append(pagelist, src->cls.request_data,
                                     src->cls.request_data_len);
-
                ceph_osd_data_pagelist_init(&req->r_data_out, pagelist);
+
+               WARN_ON(src->cls.response_data != &req->r_data_in);
                request_data_len = pagelist->length;
                break;
        case CEPH_OSD_OP_STARTSYNC:
@@ -609,6 +628,7 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
                                               bool use_mempool)
 {
        struct ceph_osd_request *req;
+       struct ceph_osd_data *osd_data;
        struct ceph_osd_req_op *op;
        u64 objnum = 0;
        u64 objoff = 0;
@@ -623,6 +643,8 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
                                        GFP_NOFS);
        if (!req)
                return ERR_PTR(-ENOMEM);
+       osd_data = opcode == CEPH_OSD_OP_WRITE ? &req->r_data_out
+                                              : &req->r_data_in;
 
        req->r_flags = flags;
 
@@ -646,6 +668,8 @@ struct ceph_osd_request *ceph_osdc_new_request(struct ceph_osd_client *osdc,
        op = &req->r_ops[0];
        osd_req_op_extent_init(op, opcode, objoff, objlen,
                                truncate_size, truncate_seq);
+       osd_req_op_extent_osd_data(op, osd_data);
+
        /*
         * A second op in the ops array means the caller wants to
         * also issue a include a 'startsync' command so that the