On Mon, Jul 29, 2019 at 11:43 AM Dongsheng Yang
<dongsheng.yang@xxxxxxxxxxxx> wrote:
>
> Introduce RBD_IMG_APPEND_JOURNAL and __RBD_IMG_APPEND_JOURNAL in rbd_img_state.
> When a image request after RBD_IMG_EXCLUSIVE_LOCK, it will go into __RBD_IMG_APPEND_JOURNAL
> and then RBD_IMG_APPEND_JOURNAL. after that, it then would go into __RBD_IMG_OBJECT_REQUESTS.
>
> That means, we will append journal event before send the data object request for image request.
>
> Signed-off-by: Dongsheng Yang <dongsheng.yang@xxxxxxxxxxxx>
> ---
> drivers/block/rbd.c | 250 +++++++++++++++++++++++++++++++++++++++++++++++++++-
> 1 file changed, 249 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
> index 86008f2..89bc7b3 100644
> --- a/drivers/block/rbd.c
> +++ b/drivers/block/rbd.c
> @@ -121,6 +121,7 @@ static int atomic_dec_return_safe(atomic_t *v)
> #define RBD_FEATURE_OBJECT_MAP (1ULL<<3)
> #define RBD_FEATURE_FAST_DIFF (1ULL<<4)
> #define RBD_FEATURE_DEEP_FLATTEN (1ULL<<5)
> +#define RBD_FEATURE_JOURNALING (1ULL<<6)
> #define RBD_FEATURE_DATA_POOL (1ULL<<7)
> #define RBD_FEATURE_OPERATIONS (1ULL<<8)
>
> @@ -327,6 +328,8 @@ enum img_req_flags {
> enum rbd_img_state {
> RBD_IMG_START = 1,
> RBD_IMG_EXCLUSIVE_LOCK,
> + __RBD_IMG_APPEND_JOURNAL,
> + RBD_IMG_APPEND_JOURNAL,
> __RBD_IMG_OBJECT_REQUESTS,
> RBD_IMG_OBJECT_REQUESTS,
> };
> @@ -355,6 +358,7 @@ struct rbd_img_request {
> int work_result;
>
> struct completion completion;
> + uint64_t journaler_commit_tid;
>
> struct kref kref;
> };
> @@ -448,6 +452,8 @@ struct rbd_device {
> atomic_t parent_ref;
> struct rbd_device *parent;
>
> + struct rbd_journal *journal;
> +
> /* Block layer tags. */
> struct blk_mq_tag_set tag_set;
>
> @@ -3650,6 +3656,19 @@ static void rbd_img_object_requests(struct rbd_img_request *img_req)
> }
> }
>
> +static bool rbd_img_need_journal(struct rbd_img_request *img_req) {
> + struct rbd_device *rbd_dev = img_req->rbd_dev;
> +
> + if (img_req->op_type == OBJ_OP_READ)
> + return false;
> +
> + if (!(rbd_dev->header.features & RBD_FEATURE_JOURNALING))
> + return false;
> +
> + return true;
> +}
> +
> +static void rbd_img_journal_append(struct rbd_img_request *img_req);
> static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
> {
> struct rbd_device *rbd_dev = img_req->rbd_dev;
> @@ -3676,6 +3695,27 @@ static bool rbd_img_advance(struct rbd_img_request *img_req, int *result)
> rbd_assert(!need_exclusive_lock(img_req) ||
> __rbd_is_lock_owner(rbd_dev));
>
> + if (!rbd_img_need_journal(img_req)) {
> + img_req->state = RBD_IMG_APPEND_JOURNAL;
> + goto again;
> + }
> +
> + rbd_img_journal_append(img_req);
> + if (!img_req->pending.num_pending) {
> + *result = img_req->pending.result;
> + img_req->state = RBD_IMG_OBJECT_REQUESTS;
> + goto again;
> + }
> + img_req->state = __RBD_IMG_APPEND_JOURNAL;
> + return false;
> + case __RBD_IMG_APPEND_JOURNAL:
> + if (!pending_result_dec(&img_req->pending, result))
> + return false;
> + /* fall through */
> + case RBD_IMG_APPEND_JOURNAL:
> + if (*result)
> + return true;
> +
> rbd_img_object_requests(img_req);
> if (!img_req->pending.num_pending) {
> *result = img_req->pending.result;
> @@ -3744,9 +3784,15 @@ static void rbd_img_handle_request(struct rbd_img_request *img_req, int result)
> } else {
> struct request *rq = img_req->rq;
>
> + if (!result && img_req->journaler_commit_tid) {
> + ceph_journaler_client_committed(img_req->rbd_dev->journal->journaler,
> + img_req->journaler_commit_tid);
> + }
> +
> complete_all(&img_req->completion);
> rbd_img_request_put(img_req);
> - blk_mq_end_request(rq, errno_to_blk_status(result));
> + if (rq)
> + blk_mq_end_request(rq, errno_to_blk_status(result));
> }
> }
>
> @@ -6927,6 +6973,208 @@ static int rbd_dev_header_name(struct rbd_device *rbd_dev)
> return ret;
> }
>
> +enum rbd_journal_event_type {
> + EVENT_TYPE_AIO_DISCARD = 0,
> + EVENT_TYPE_AIO_WRITE = 1,
> + EVENT_TYPE_AIO_FLUSH = 2,
> + EVENT_TYPE_OP_FINISH = 3,
> + EVENT_TYPE_SNAP_CREATE = 4,
> + EVENT_TYPE_SNAP_REMOVE = 5,
> + EVENT_TYPE_SNAP_RENAME = 6,
> + EVENT_TYPE_SNAP_PROTECT = 7,
> + EVENT_TYPE_SNAP_UNPROTECT = 8,
> + EVENT_TYPE_SNAP_ROLLBACK = 9,
> + EVENT_TYPE_RENAME = 10,
> + EVENT_TYPE_RESIZE = 11,
> + EVENT_TYPE_FLATTEN = 12,
> + EVENT_TYPE_DEMOTE_PROMOTE = 13,
> + EVENT_TYPE_SNAP_LIMIT = 14,
> + EVENT_TYPE_UPDATE_FEATURES = 15,
> + EVENT_TYPE_METADATA_SET = 16,
> + EVENT_TYPE_METADATA_REMOVE = 17,
> + EVENT_TYPE_AIO_WRITESAME = 18,
> + EVENT_TYPE_AIO_COMPARE_AND_WRITE = 19,
> +};
> +
> +
> +// RBD_EVENT_FIXED_SIZE(10 = CEPH_ENCODING_START_BLK_LEN(6) + EVENT_TYPE(4))
> +static const uint32_t RBD_EVENT_FIXED_SIZE = 10;
> +
> +static struct bio_vec *setup_write_bvecs(void *buf, u64 offset, u64 length)
> +{
> + u32 i;
> + struct bio_vec *bvecs;
> + u32 bvec_count;
> +
> + bvec_count = calc_pages_for(offset, length);
> + bvecs = kcalloc(bvec_count, sizeof(*bvecs), GFP_NOIO);
> + if (!bvecs)
> + goto err;
> +
> + offset = offset % PAGE_SIZE;
> + for (i = 0; i < bvec_count; i++) {
> + unsigned int len = min(length, (u64)PAGE_SIZE - offset);
> +
> + bvecs[i].bv_page = alloc_page(GFP_NOIO);
> + if (!bvecs[i].bv_page)
> + goto free_bvecs;
> +
> + bvecs[i].bv_offset = offset;
> + bvecs[i].bv_len = len;
> + memcpy(page_address(bvecs[i].bv_page) + bvecs[i].bv_offset, buf, bvecs[i].bv_len);
> + length -= len;
> + buf += len;
> + offset = 0;
> + }
> +
> + rbd_assert(!length);
> +
> + return bvecs;
> +
> +free_bvecs:
> +err:
> + return NULL;
> +}
> +
> +static void rbd_journal_callback(struct ceph_journaler_ctx *journaler_ctx)
> +{
> + struct rbd_img_request *img_req = journaler_ctx->priv;
> + int result = journaler_ctx->result;
> +
> + ceph_journaler_ctx_put(journaler_ctx);
> + rbd_img_handle_request(img_req, result);
> +}
> +
> +static void img_journal_append_write_event(struct rbd_img_request *img_req)
> +{
> + struct rbd_journal *journal = img_req->rbd_dev->journal;
> + struct ceph_journaler_ctx *journaler_ctx;
> + u64 offset = (u64)blk_rq_pos(img_req->rq) << SECTOR_SHIFT;
> + u64 length = blk_rq_bytes(img_req->rq);
> + struct bio *bio = img_req->rq->bio;
> + uint64_t prefix_len = RBD_EVENT_FIXED_SIZE + 20;
> + uint64_t max_append_size = ceph_journaler_get_max_append_size(journal->journaler) - prefix_len;
> + uint64_t append_size = min(max_append_size, length);
> + uint64_t bio_offset = 0;
> + void *p;
> + int ret;
> +
> + rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
> +
> + while (length > 0) {
> + journaler_ctx = ceph_journaler_ctx_alloc();
> + if (!journaler_ctx) {
> + img_req->pending.result = -ENOMEM;
> + return;
> + }
> +
> + journaler_ctx->bio_iter.bio = bio;
> + journaler_ctx->bio_iter.iter = bio->bi_iter;
> +
> + ceph_bio_iter_advance(&journaler_ctx->bio_iter, bio_offset);
> + append_size = min(max_append_size, length);
> + journaler_ctx->bio_len = append_size;
> + bio_offset += append_size;
> + length -= append_size;
> +
> + // RBD_EVENT_FIXED_SIZE + offset(8) + length(8) + string_len(4) = 30
> + journaler_ctx->prefix_len = prefix_len;
> + journaler_ctx->prefix_offset = PAGE_SIZE - journaler_ctx->prefix_len;
> +
> + p = page_address(journaler_ctx->prefix_page) + journaler_ctx->prefix_offset;
> +
> + ceph_start_encoding(&p, 1, 1, journaler_ctx->prefix_len + journaler_ctx->bio_len - 6);
> +
> + ceph_encode_32(&p, EVENT_TYPE_AIO_WRITE);
> +
> + ceph_encode_64(&p, offset);
Looks like this is encoding the same offset for all append chunks?
This, coupled with always checksumming the entire bio, makes me think
that this code path hasn't been tested.
> + ceph_encode_64(&p, append_size);
> +
> + // first part of ceph_encode_string();
> + ceph_encode_32(&p, journaler_ctx->bio_len);
> +
> + journaler_ctx->priv = img_req;
> + journaler_ctx->callback = rbd_journal_callback;
> +
> + ret = ceph_journaler_append(journal->journaler, journal->tag_tid,
> + journaler_ctx);
> + if (ret) {
> + ceph_journaler_ctx_put(journaler_ctx);
> + img_req->pending.result = ret;
> + return;
> + }
> +
> + rbd_assert(!ret);
Bogus assert.
> + img_req->pending.num_pending++;
> + img_req->journaler_commit_tid = journaler_ctx->commit_tid;
> + }
> +}
> +
> +static void img_journal_append_discard_event(struct rbd_img_request *img_req)
> +{
> + struct rbd_journal *journal = img_req->rbd_dev->journal;
> + struct ceph_journaler_ctx *journaler_ctx;
> + u64 offset = (u64)blk_rq_pos(img_req->rq) << SECTOR_SHIFT;
> + u64 length = blk_rq_bytes(img_req->rq);
> + struct bio *bio = img_req->rq->bio;
> + void *p;
> + int ret;
> +
> + rbd_assert(!img_req->pending.result && !img_req->pending.num_pending);
Move this assert to rbd_img_journal_append() to avoid duplicating it
for write event and for discard event.
> +
> + journaler_ctx = ceph_journaler_ctx_alloc();
> + if (!journaler_ctx) {
> + img_req->pending.result = -ENOMEM;
> + return;
> + }
> +
> + journaler_ctx->bio_iter.bio = bio;
> + journaler_ctx->bio_iter.iter = bio->bi_iter;
> + journaler_ctx->bio_len = 0;
> +
> + // RBD_EVENT_FIXED_SIZE + offset(8) + length(8) = 26
> + journaler_ctx->prefix_len = RBD_EVENT_FIXED_SIZE + 16;
> + journaler_ctx->prefix_offset = PAGE_SIZE - journaler_ctx->prefix_len;
> +
> + p = page_address(journaler_ctx->prefix_page) + journaler_ctx->prefix_offset;
> +
> + ceph_start_encoding(&p, 1, 1, journaler_ctx->prefix_len + journaler_ctx->bio_len - 6);
> +
> + ceph_encode_32(&p, EVENT_TYPE_AIO_DISCARD);
> +
> + ceph_encode_64(&p, offset);
> + ceph_encode_64(&p, length);
> +
> + journaler_ctx->priv = img_req;
> + journaler_ctx->callback = rbd_journal_callback;
> +
> + ret = ceph_journaler_append(journal->journaler, journal->tag_tid,
> + journaler_ctx);
> + if (ret) {
> + ceph_journaler_ctx_put(journaler_ctx);
> + img_req->pending.result = ret;
> + return;
> + }
> +
> + rbd_assert(!ret);
Bogus assert.
> + img_req->pending.num_pending++;
> + img_req->journaler_commit_tid = journaler_ctx->commit_tid;
> +}
> +
> +static void rbd_img_journal_append(struct rbd_img_request *img_req)
> +{
> + switch (img_req->op_type) {
> + case OBJ_OP_WRITE:
> + img_journal_append_write_event(img_req);
> + break;
> + case OBJ_OP_DISCARD:
> + img_journal_append_discard_event(img_req);
> + break;
> + default:
> + img_req->pending.result = -ENOTSUPP;
What about zeroouts?
Thanks,
Ilya
