On Mon, Feb 24, 2014 at 4:59 PM, Alex Elder <elder@xxxxxxxx> wrote:
> On 02/21/2014 12:55 PM, Ilya Dryomov wrote:
>> In an effort to reduce fragmentation, prefix every rbd write with
>> a CEPH_OSD_OP_SETALLOCHINT osd op with an expected_write_size value set
>> to the object size (1 << order). Backwards compatibility is taken care
>> of on the libceph/osd side.
>
> If *every* write will include a hint, why even encode this as
> a distinct opcode? Why not just extend the definition of a
> write operation to include the write hint data? The server
> side could check expected_object_size, and if 0 (or some other
> invalid value) it means the client isn't supplying a hint.
>
> However, on the assumption you want this to be a distinct
> OSD op I think you generally did the right thing. See my
> comments below. For now I'm not indicating "Reviewed-by"
> because it sounds like the nature of this change is under
> discussion still. And I really do think that if the hint
> is not going to be made more generic (and possibly even if
> it is) I'd rather see this hinting done using an extension
> of the write operation (like I suggest above). In this
> case it is clearly directly tied to every write operation
> and separating it sort of obscures that.
Yes, the assumption is that we want to do this in a separate op. The
hint is durable, in that it's enough to do it once, so it doesn't make
much sense to fold it into the write op(s). The reason every rbd write
is prefixed is that rbd doesn't explicitly create objects and relies on
writes creating them implicitly, so there is no place to stick a single
hint op into. To get around that we decided to prefix every rbd write
with a hint (just like write and setattr ops, hint op will create an
object implicitly if it doesn't exist).
I'll add the above paragraph to the commit message.
>
> -Alex
>
>> Signed-off-by: Ilya Dryomov <ilya.dryomov@xxxxxxxxxxx>
>> ---
>> drivers/block/rbd.c | 66 ++++++++++++++++++++++++++++++++++++++-------------
>> 1 file changed, 49 insertions(+), 17 deletions(-)
>>
>> diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
>> index 6cf001ef00bc..14496f39c770 100644
>> --- a/drivers/block/rbd.c
>> +++ b/drivers/block/rbd.c
>> @@ -1662,7 +1662,12 @@ static void rbd_osd_req_callback(struct ceph_osd_request *osd_req,
>> */
>> obj_request->xferred = osd_req->r_reply_op_len[0];
>> rbd_assert(obj_request->xferred < (u64)UINT_MAX);
>> +
>> opcode = osd_req->r_ops[0].op;
>> + if (opcode == CEPH_OSD_OP_SETALLOCHINT) {
>> + BUG_ON(osd_req->r_ops[1].op != CEPH_OSD_OP_WRITE);
>> + opcode = CEPH_OSD_OP_WRITE;
>> + }
>
> I would rather see this handled as a distinct case below
> rather than fudging the opcode. Bury that BUG_ON() inside
> inside rbd_osd_write_callback() logic to avoid encoding those
> op-specific details here. And change the BUG_ON() to rbd_assert().
It can't be buried inside rbd_osd_write_callback() because the latter
takes obj_request and opcode is the osd_request property. I changed it
to the following though:
...
case CEPH_OSD_OP_SETALLOCHINT:
rbd_assert(osd_req->r_ops[1].op == CEPH_OSD_OP_WRITE);
/* fall through */
case CEPH_OSD_OP_WRITE:
...
>
>> switch (opcode) {
>> case CEPH_OSD_OP_READ:
>> rbd_osd_read_callback(obj_request);
>> @@ -1715,6 +1720,12 @@ static void rbd_osd_req_format_write(struct rbd_obj_request *obj_request)
>> snapc, CEPH_NOSNAP, &mtime);
>> }
>>
>> +/*
>> + * Create an osd request. A read request has one osd op (read).
>> + * A write request has either one (watch) or two (hint+write) osd ops.
>> + * (All rbd writes are prefixed with an allocation hint op, but
>
> Maybe "rbd data writes" or something. A watch changes state but
> doesn't write data.
Done.
>
>> + * technically osd watch is a write request, hence this distinction.)
>> + */
>> static struct ceph_osd_request *rbd_osd_req_create(
>> struct rbd_device *rbd_dev,
>> bool write_request,
>> @@ -1734,7 +1745,8 @@ static struct ceph_osd_request *rbd_osd_req_create(
>> snapc = img_request->snapc;
>> }
>>
>> - rbd_assert(num_ops == 1);
>> + rbd_assert((!write_request && num_ops == 1) ||
>> + (write_request && num_ops >= 1 && num_ops <= 2));
>
> Maybe:
> rbd_assert(num_ops == 1 || write_request && num_ops == 2);
Done. This isn't very clear but definitely better than what I had and
not so verbose as
if (write_request)
rbd_assert(num_ops == 1 || num_ops == 2);
else
rbd_assert(num_ops == 1);
I had to add extra parentheses around && though to silence gcc.
>
>>
>> /* Allocate and initialize the request, for the num_ops ops */
>>
>> @@ -1760,8 +1772,8 @@ static struct ceph_osd_request *rbd_osd_req_create(
>>
>> /*
>> * Create a copyup osd request based on the information in the
>> - * object request supplied. A copyup request has two osd ops,
>> - * a copyup method call, and a "normal" write request.
>> + * object request supplied. A copyup request has three osd ops,
>> + * a copyup method call, a hint op, and a write op.
>
> There's a comment in the caller that should be similarly updated:
> * We need a new one that can hold the two ops in a copyup
> ^^^
It already is, in this patch, below.
>> */
>> static struct ceph_osd_request *
>> rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
>> @@ -1777,12 +1789,12 @@ rbd_osd_req_create_copyup(struct rbd_obj_request *obj_request)
>> rbd_assert(img_request);
>> rbd_assert(img_request_write_test(img_request));
>>
>> - /* Allocate and initialize the request, for the two ops */
>> + /* Allocate and initialize the request, for the three ops */
>>
>> snapc = img_request->snapc;
>> rbd_dev = img_request->rbd_dev;
>> osdc = &rbd_dev->rbd_client->client->osdc;
>> - osd_req = ceph_osdc_alloc_request(osdc, snapc, 2, false, GFP_ATOMIC);
>> + osd_req = ceph_osdc_alloc_request(osdc, snapc, 3, false, GFP_ATOMIC);
>> if (!osd_req)
>> return NULL; /* ENOMEM */
>>
>> @@ -2159,12 +2171,10 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
>> struct page **pages = NULL;
>> u64 img_offset;
>> u64 resid;
>> - u16 opcode;
>
> Why move opcode inside the inner loop? It's not going
> to change, is it? (Maybe this was a leftover from an
> abandoned attempt to do this a different way.)
Probably, done.
>
>>
>> dout("%s: img %p type %d data_desc %p\n", __func__, img_request,
>> (int)type, data_desc);
>>
>> - opcode = write_request ? CEPH_OSD_OP_WRITE : CEPH_OSD_OP_READ;
>> img_offset = img_request->offset;
>> resid = img_request->length;
>> rbd_assert(resid > 0);
>> @@ -2183,6 +2193,8 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
>> const char *object_name;
>> u64 offset;
>> u64 length;
>> + unsigned int which;
>
> unsigned int which = 0;
>
>> + u16 opcode;
>>
>> object_name = rbd_segment_name(rbd_dev, img_offset);
>> if (!object_name)
>> @@ -2224,20 +2236,34 @@ static int rbd_img_request_fill(struct rbd_img_request *img_request,
>> pages += page_count;
>> }
>>
>> - osd_req = rbd_osd_req_create(rbd_dev, write_request, 1,
>> + osd_req = rbd_osd_req_create(rbd_dev, write_request,
>> + (write_request ? 2 : 1),
>> obj_request);
>> if (!osd_req)
>> goto out_partial;
>> obj_request->osd_req = osd_req;
>> obj_request->callback = rbd_img_obj_callback;
>>
>> - osd_req_op_extent_init(osd_req, 0, opcode, offset, length,
>> - 0, 0);
>> + if (write_request) {
>
>
>
>> + osd_req_op_hint_init(osd_req, 0,
>> + CEPH_OSD_OP_SETALLOCHINT,
>> + rbd_obj_bytes(&rbd_dev->header),
>> + rbd_obj_bytes(&rbd_dev->header),
>> + 0);
>> +
>> + which = 1;
> which++;
>
>> + opcode = CEPH_OSD_OP_WRITE;
>> + } else {
>
> And this block goes away (assign opcode outside the loop).
Done.
>
>> + which = 0;
>> + opcode = CEPH_OSD_OP_READ;
>> + }
>> + osd_req_op_extent_init(osd_req, which, opcode, offset, length,
>> + 0, 0);
>> if (type == OBJ_REQUEST_BIO)
>> - osd_req_op_extent_osd_data_bio(osd_req, 0,
>> + osd_req_op_extent_osd_data_bio(osd_req, which,
>> obj_request->bio_list, length);
>> else
>> - osd_req_op_extent_osd_data_pages(osd_req, 0,
>> + osd_req_op_extent_osd_data_pages(osd_req, which,
>> obj_request->pages, length,
>> offset & ~PAGE_MASK, false, false);
>>
>> @@ -2358,7 +2384,7 @@ rbd_img_obj_parent_read_full_callback(struct rbd_img_request *img_request)
>>
>> /*
>> * The original osd request is of no use to use any more.
>> - * We need a new one that can hold the two ops in a copyup
>> + * We need a new one that can hold the three ops in a copyup
Here ^^ (re: comment).
Thanks,
Ilya
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
