| Message ID | 1394680896-8554-1-git-send-email-lucienchao@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 03/12/2014 10:21 PM, Guangliang Zhao wrote: > It need to copyup the parent's content when layered writing, > but an entire object write would overwrite it, so skip it. > > Signed-off-by: Guangliang Zhao <lucienchao@gmail.com> > --- > drivers/block/rbd.c | 49 ++++++++++++++++++++++++++++++++++--------------- > 1 file changed, 34 insertions(+), 15 deletions(-) > > diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c > index b365e0d..2d48858 100644 > --- a/drivers/block/rbd.c > +++ b/drivers/block/rbd.c > @@ -2624,34 +2624,53 @@ out: > return ret; > } > > -static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) > +static bool rbd_obj_request_simple(struct rbd_obj_request *obj_request) The only objects that will be passed to this functions will be associated with images. That fact can be reinforced by naming it rbd_img_obj_request_simple(). Other than that suggestion, this looks good. Reviewed-by: Alex Elder <elder@linaro.org> > { > struct rbd_img_request *img_request; > struct rbd_device *rbd_dev; > - bool known; > + u64 obj_size; > > rbd_assert(obj_request_img_data_test(obj_request)); > > img_request = obj_request->img_request; > rbd_assert(img_request); > rbd_dev = img_request->rbd_dev; > + obj_size = (u64) 1 << rbd_dev->header.obj_order; > > + /* Read requests didn't need special handling */ > + if (!img_request_write_test(img_request)) > + return true; > + /* No-layered writes are simple requests*/ > + if (!img_request_layered_test(img_request)) > + return true; > /* > - * Only writes to layered images need special handling. > - * Reads and non-layered writes are simple object requests. > * Layered writes that start beyond the end of the overlap > - * with the parent have no parent data, so they too are > - * simple object requests. Finally, if the target object is > - * known to already exist, its parent data has already been > - * copied, so a write to the object can also be handled as a > - * simple object request. > + * with the parent have no parent data, so they are simple > + * object requests. > */ > - if (!img_request_write_test(img_request) || > - !img_request_layered_test(img_request) || > - rbd_dev->parent_overlap <= obj_request->img_offset || > - ((known = obj_request_known_test(obj_request)) && > - obj_request_exists_test(obj_request))) { > + if (rbd_dev->parent_overlap <= obj_request->img_offset) > + return true; > + /* > + * If the obj_request aligns with the boundary and equals > + * to the size of an object, it doesn't need copyup, because > + * the obj_request will overwrite it finally. > + */ > + if ((!obj_request->offset) && (obj_request->length == obj_size)) > + return true; > + /* > + * If the target object is known to already exist, its parent > + * data has already been copied, so a write to the object can > + * also be handled as a simple object request > + */ > + if (obj_request_known_test(obj_request) && > + obj_request_exists_test(obj_request)) > + return true; > + return false; > +} > > +static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) > +{ > + if (rbd_obj_request_simple(obj_request)) { > struct rbd_device *rbd_dev; > struct ceph_osd_client *osdc; > > @@ -2667,7 +2686,7 @@ static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) > * start by reading the data for the full target object from > * the parent so we can use it for a copyup to the target. > */ > - if (known) > + if (obj_request_known_test(obj_request)) > return rbd_img_obj_parent_read_full(obj_request); > > /* We don't know whether the target exists. Go find out. */ > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 03/12/2014 08:21 PM, Guangliang Zhao wrote: > It need to copyup the parent's content when layered writing, > but an entire object write would overwrite it, so skip it. > > Signed-off-by: Guangliang Zhao <lucienchao@gmail.com> > --- This looks good to me. This situation is unlikely with normal I/O, but discards done e.g. by mkfs are more likely to benefit from this optimization. Nice refactoring of that conditional too (thanks for suggesting that Alex)! Reviewed-by: Josh Durgin <josh.durgin@inktank.com> > drivers/block/rbd.c | 49 ++++++++++++++++++++++++++++++++++--------------- > 1 file changed, 34 insertions(+), 15 deletions(-) > > diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c > index b365e0d..2d48858 100644 > --- a/drivers/block/rbd.c > +++ b/drivers/block/rbd.c > @@ -2624,34 +2624,53 @@ out: > return ret; > } > > -static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) > +static bool rbd_obj_request_simple(struct rbd_obj_request *obj_request) > { > struct rbd_img_request *img_request; > struct rbd_device *rbd_dev; > - bool known; > + u64 obj_size; > > rbd_assert(obj_request_img_data_test(obj_request)); > > img_request = obj_request->img_request; > rbd_assert(img_request); > rbd_dev = img_request->rbd_dev; > + obj_size = (u64) 1 << rbd_dev->header.obj_order; > > + /* Read requests didn't need special handling */ > + if (!img_request_write_test(img_request)) > + return true; > + /* No-layered writes are simple requests*/ > + if (!img_request_layered_test(img_request)) > + return true; > /* > - * Only writes to layered images need special handling. > - * Reads and non-layered writes are simple object requests. > * Layered writes that start beyond the end of the overlap > - * with the parent have no parent data, so they too are > - * simple object requests. Finally, if the target object is > - * known to already exist, its parent data has already been > - * copied, so a write to the object can also be handled as a > - * simple object request. > + * with the parent have no parent data, so they are simple > + * object requests. > */ > - if (!img_request_write_test(img_request) || > - !img_request_layered_test(img_request) || > - rbd_dev->parent_overlap <= obj_request->img_offset || > - ((known = obj_request_known_test(obj_request)) && > - obj_request_exists_test(obj_request))) { > + if (rbd_dev->parent_overlap <= obj_request->img_offset) > + return true; > + /* > + * If the obj_request aligns with the boundary and equals > + * to the size of an object, it doesn't need copyup, because > + * the obj_request will overwrite it finally. > + */ > + if ((!obj_request->offset) && (obj_request->length == obj_size)) > + return true; > + /* > + * If the target object is known to already exist, its parent > + * data has already been copied, so a write to the object can > + * also be handled as a simple object request > + */ > + if (obj_request_known_test(obj_request) && > + obj_request_exists_test(obj_request)) > + return true; > + return false; > +} > > +static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) > +{ > + if (rbd_obj_request_simple(obj_request)) { > struct rbd_device *rbd_dev; > struct ceph_osd_client *osdc; > > @@ -2667,7 +2686,7 @@ static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) > * start by reading the data for the full target object from > * the parent so we can use it for a copyup to the target. > */ > - if (known) > + if (obj_request_known_test(obj_request)) > return rbd_img_obj_parent_read_full(obj_request); > > /* We don't know whether the target exists. Go find out. */ > -- To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c index b365e0d..2d48858 100644 --- a/drivers/block/rbd.c +++ b/drivers/block/rbd.c @@ -2624,34 +2624,53 @@ out: return ret; } -static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) +static bool rbd_obj_request_simple(struct rbd_obj_request *obj_request) { struct rbd_img_request *img_request; struct rbd_device *rbd_dev; - bool known; + u64 obj_size; rbd_assert(obj_request_img_data_test(obj_request)); img_request = obj_request->img_request; rbd_assert(img_request); rbd_dev = img_request->rbd_dev; + obj_size = (u64) 1 << rbd_dev->header.obj_order; + /* Read requests didn't need special handling */ + if (!img_request_write_test(img_request)) + return true; + /* No-layered writes are simple requests*/ + if (!img_request_layered_test(img_request)) + return true; /* - * Only writes to layered images need special handling. - * Reads and non-layered writes are simple object requests. * Layered writes that start beyond the end of the overlap - * with the parent have no parent data, so they too are - * simple object requests. Finally, if the target object is - * known to already exist, its parent data has already been - * copied, so a write to the object can also be handled as a - * simple object request. + * with the parent have no parent data, so they are simple + * object requests. */ - if (!img_request_write_test(img_request) || - !img_request_layered_test(img_request) || - rbd_dev->parent_overlap <= obj_request->img_offset || - ((known = obj_request_known_test(obj_request)) && - obj_request_exists_test(obj_request))) { + if (rbd_dev->parent_overlap <= obj_request->img_offset) + return true; + /* + * If the obj_request aligns with the boundary and equals + * to the size of an object, it doesn't need copyup, because + * the obj_request will overwrite it finally. + */ + if ((!obj_request->offset) && (obj_request->length == obj_size)) + return true; + /* + * If the target object is known to already exist, its parent + * data has already been copied, so a write to the object can + * also be handled as a simple object request + */ + if (obj_request_known_test(obj_request) && + obj_request_exists_test(obj_request)) + return true; + return false; +} +static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) +{ + if (rbd_obj_request_simple(obj_request)) { struct rbd_device *rbd_dev; struct ceph_osd_client *osdc; @@ -2667,7 +2686,7 @@ static int rbd_img_obj_request_submit(struct rbd_obj_request *obj_request) * start by reading the data for the full target object from * the parent so we can use it for a copyup to the target. */ - if (known) + if (obj_request_known_test(obj_request)) return rbd_img_obj_parent_read_full(obj_request); /* We don't know whether the target exists. Go find out. */
It need to copyup the parent's content when layered writing, but an entire object write would overwrite it, so skip it. Signed-off-by: Guangliang Zhao <lucienchao@gmail.com> --- drivers/block/rbd.c | 49 ++++++++++++++++++++++++++++++++++--------------- 1 file changed, 34 insertions(+), 15 deletions(-)