On Mon, 12 Mar 2012, Alex Elder wrote:
> Convert ceph_calc_file_object_mapping() so it uses 64-bit variables
> and divide operators. This is in anticipation of the next patch,
> to keep the the changes it incurs more focused.
>
> I made most of the variable names more verbose, and because it now
> uses do_div() some of the calculations get done in a different
> order. Net result is that the code does what it did before, but it
> looks quite a bit different...
I didn't double check your math, but he code certainly looks nicer. It
would be nice to have a test that runs some tool that verifies re-read
data (iozone?) run when non-default file layouts. Say, 1MB stripes over 9
13MB objects, something like that. The 'cephfs' command line utility can
be used to set the default layout for new files under a
subdirectory/subtree, so running it on . in a workunit bash script (e.g.
based on ceph.git/qa/workunits/suites/iozone.sh) should do the trick.
('cephfs' only works on the kclient, though, so it needs to be a distinct
script that we don't already run on ceph-fuse... probably stick it in the
qa/workunits/kclient directory).
Not that this code was thoroughly tested with weird layouts before or
anything :), but it would be nice to see that before we merge this in.
The preferred osd stuff I'm less worried about. It would actually be nice
to rip all that out, assuming we can get a hadoop person to convince us
that it doesn't matter. :/
sage
>
> Signed-off-by: Alex Elder <elder@xxxxxxxxxxxxx>
> ---
> net/ceph/osdmap.c | 112
> +++++++++++++++++++++++++++++++++--------------------
> 1 files changed, 70 insertions(+), 42 deletions(-)
>
> diff --git a/net/ceph/osdmap.c b/net/ceph/osdmap.c
> index f5f6e41..26c30e7 100644
> --- a/net/ceph/osdmap.c
> +++ b/net/ceph/osdmap.c
> @@ -933,59 +933,87 @@ bad:
>
>
> /*
> - * calculate file layout from given offset, length.
> - * fill in correct oid, logical length, and object extent
> - * offset, length.
> + * calculate file layout from given offset, length. fill in correct
> + * object number, logical length, and object extent offset and length.
> *
> - * for now, we write only a single su, until we can
> - * pass a stride back to the caller.
> + * for now, we write only a single stripe_unit, until we can pass a
> + * stride back to the caller.
> */
> void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
> - u64 off, u64 *plen,
> - u64 *ono,
> - u64 *oxoff, u64 *oxlen)
> + u64 off,
> + u64 *plen,
> + u64 *object_num,
> + u64 *object_ext_off,
> + u64 *object_ext_len)
> {
> - u32 osize = (u32) ceph_file_layout_object_size(layout);
> - u32 su = (u32) ceph_file_layout_stripe_unit(layout);
> - u32 sc = (u32) ceph_file_layout_stripe_count(layout);
> - u32 bl, stripeno, stripepos, objsetno;
> - u32 su_per_object;
> - u64 t, su_offset;
> -
> - dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
> - osize, su);
> - su_per_object = osize / su;
> - dout("osize %u / su %u = su_per_object %u\n", osize, su,
> - su_per_object);
> -
> - BUG_ON((su & ~PAGE_MASK) != 0);
> - /* bl = *off / su; */
> - t = off;
> - do_div(t, su);
> - bl = t;
> - dout("off %llu / su %u = bl %u\n", off, su, bl);
> -
> - stripeno = bl / sc;
> - stripepos = bl % sc;
> - objsetno = stripeno / su_per_object;
> -
> - *ono = objsetno * sc + stripepos;
> - dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
> -
> - /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
> - t = off;
> - su_offset = do_div(t, su);
> - *oxoff = su_offset + (stripeno % su_per_object) * su;
> + u64 object_size = (u64) ceph_file_layout_object_size(layout);
> + u64 stripe_unit = (u64) ceph_file_layout_stripe_unit(layout);
> + u64 stripe_count = (u64) ceph_file_layout_stripe_count(layout);
> + u64 stripe_unit_per_object;
> + u64 stripe_unit_num;
> + u64 stripe_unit_offset;
> + u64 stripe_num;
> + u64 stripe_pos; /* Which object within an object set
> */
> + u64 obj_set_num;
> + u64 obj_stripe_unit_num; /* Which stripe_unit within object */
> +
> + BUG_ON((stripe_unit & ~PAGE_MASK) != 0);
> +
> + dout("mapping %llu~%llu object_size %llu fl_stripe_unit %llu\n",
> + off, *plen, object_size, stripe_unit);
> +
> + /* stripe_unit_per_object = object_size / stripe_unit; */
> + stripe_unit_per_object = object_size;
> + do_div(stripe_unit_per_object, stripe_unit);
> + dout("object_size %llu / stripe_unit %llu "
> + "= stripe_unit_per_object %llu\n",
> + object_size, stripe_unit, stripe_unit_per_object);
> +
> + /*
> + * stripe_unit_num = off / stripe_unit;
> + * stripe_unit_offset = off % stripe_unit;
> + */
> + stripe_unit_num = off;
> + stripe_unit_offset = do_div(stripe_unit_num, stripe_unit);
> + dout("off %llu / stripe_unit %llu = "
> + "stripe_unit_num %llu rem stripe_unit_offset =
> %llu\n",
> + off, stripe_unit, stripe_unit_num, stripe_unit_offset);
> +
> + /*
> + * stripe_num = stripe_unit_num / stripe_count;
> + * stripe_pos = stripe_unit_num % stripe_count;
> + */
> + stripe_num = stripe_unit_num;
> + stripe_pos = do_div(stripe_num, stripe_count);
> + dout("stripe_unit_num %llu / stripe_count %llu = "
> + "stripe_num %llu rem stripe_pos %llu\n",
> + stripe_unit_num, stripe_count, stripe_num, stripe_pos);
> +
> + /*
> + * obj_set_num = stripe_num / stripe_unit_per_object;
> + * obj_stripe_unit_num = stripe_num % stripe_unit_per_object;
> + */
> + obj_set_num = stripe_num;
> + obj_stripe_unit_num = do_div(obj_set_num, stripe_unit_per_object);
> +
> + *object_num = obj_set_num * stripe_count + stripe_pos;
> + dout("obj_set_num %llu * stripe_count %llu = object_num %llu\n",
> + obj_set_num, stripe_count, *object_num);
> + *object_ext_off = stripe_unit * obj_stripe_unit_num
> + + stripe_unit_offset;
> + dout("obj_stripe_unit_num %llu * stripe_unit %llu = "
> + "object_ext_off %llu\n",
> + obj_stripe_unit_num, stripe_unit, *object_ext_off);
>
> /*
> * Calculate the length of the extent being written to the selected
> * object. This is the minimum of the full length requested (plen) or
> * the remainder of the current stripe being written to.
> */
> - *oxlen = min_t(u64, *plen, su - su_offset);
> - *plen = *oxlen;
> + *object_ext_len = min_t(u64, *plen, stripe_unit - stripe_unit_offset);
> + *plen = *object_ext_len;
>
> - dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
> + dout(" obj extent %llu~%llu\n", *object_ext_off, *object_ext_len);
> }
> EXPORT_SYMBOL(ceph_calc_file_object_mapping);
>
> --
> 1.7.5.4
>
> --
> 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
>
>
--
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
