With journaling feature enabled, we need to append event
to journal before sending img_request.
Signed-off-by: Dongsheng Yang <dongsheng.yang@xxxxxxxxxxxx>
---
drivers/block/rbd.c | 204 +++++++++++++++++++++++++++++++++++++++++++++++++---
1 file changed, 196 insertions(+), 8 deletions(-)
diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index bd90c17..5b641f8 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -298,6 +298,7 @@ struct rbd_img_request {
u32 pending_count;
struct completion completion;
+ uint64_t journaler_commit_tid;
struct kref kref;
};
@@ -441,6 +442,7 @@ struct rbd_journal {
static struct kmem_cache *rbd_img_request_cache;
static struct kmem_cache *rbd_obj_request_cache;
+static struct kmem_cache *rbd_journal_ctx_cache;
static int rbd_major;
static DEFINE_IDA(rbd_dev_id_ida);
@@ -2616,12 +2618,20 @@ static void rbd_img_end_child_request(struct rbd_img_request *img_req)
static void rbd_img_end_request(struct rbd_img_request *img_req)
{
rbd_assert(!test_bit(IMG_REQ_CHILD, &img_req->flags));
- rbd_assert((!img_req->result &&
- img_req->xferred == blk_rq_bytes(img_req->rq)) ||
- (img_req->result < 0 && !img_req->xferred));
- blk_mq_end_request(img_req->rq,
- errno_to_blk_status(img_req->result));
+ if (img_req->rq) {
+ rbd_assert((!img_req->result &&
+ img_req->xferred == blk_rq_bytes(img_req->rq)) ||
+ (img_req->result < 0 && !img_req->xferred));
+ blk_mq_end_request(img_req->rq,
+ errno_to_blk_status(img_req->result));
+ }
+
+ if (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);
}
@@ -3689,6 +3699,21 @@ static int rbd_wait_state_locked(struct rbd_device *rbd_dev, bool may_acquire)
return ret;
}
+struct rbd_journal_ctx {
+ struct rbd_device *rbd_dev;
+ struct rbd_img_request *img_request;
+ struct request *rq;
+ struct ceph_snap_context *snapc;
+ int result;
+ bool must_be_locked;
+
+ struct ceph_bio_iter bio_iter;
+};
+
+static int rbd_journal_append(struct rbd_device *rbd_dev, struct bio *bio,
+ u64 offset, u64 length, enum obj_operation_type op_type,
+ struct rbd_journal_ctx *ctx);
+
static void rbd_queue_workfn(struct work_struct *work)
{
struct request *rq = blk_mq_rq_from_pdu(work);
@@ -3794,7 +3819,29 @@ static void rbd_queue_workfn(struct work_struct *work)
if (result)
goto err_img_request;
- rbd_img_request_submit(img_request);
+ if (!(rbd_dev->header.features & RBD_FEATURE_JOURNALING) ||
+ (op_type == OBJ_OP_READ)) {
+ rbd_img_request_submit(img_request);
+ } else {
+ struct rbd_journal_ctx *ctx = kmem_cache_zalloc(rbd_journal_ctx_cache, GFP_NOIO);
+
+ if (!ctx){
+ result = -ENOMEM;
+ goto err_unlock;
+ }
+
+ ctx->img_request = img_request;
+ ctx->rq = rq;
+ ctx->snapc = snapc;
+ ctx->must_be_locked = must_be_locked;
+ ctx->rbd_dev = rbd_dev;
+ result = rbd_journal_append(rbd_dev, rq->bio, offset, length, op_type, ctx);
+ if (result) {
+ rbd_warn(rbd_dev, "error in rbd_journal_append");
+ goto err_unlock;
+ }
+ }
+
if (must_be_locked)
up_read(&rbd_dev->lock_rwsem);
return;
@@ -5996,6 +6043,140 @@ static void rbd_dev_image_release(struct rbd_device *rbd_dev)
rbd_dev->spec->image_id = NULL;
}
+static void rbd_journal_callback(struct ceph_journaler_ctx *journaler_ctx)
+{
+ struct rbd_journal_ctx *ctx = journaler_ctx->priv;
+ int result = journaler_ctx->result;
+ struct rbd_device *rbd_dev = ctx->rbd_dev;
+ bool must_be_locked = ctx->must_be_locked;
+
+ if (result)
+ goto err_rq;
+
+ if (must_be_locked)
+ down_read(&rbd_dev->lock_rwsem);
+
+ rbd_img_request_submit(ctx->img_request);
+
+ if (must_be_locked)
+ up_read(&rbd_dev->lock_rwsem);
+
+ goto out;
+
+err_rq:
+ ceph_put_snap_context(ctx->snapc);
+ blk_mq_end_request(ctx->rq, errno_to_blk_status(result));
+ rbd_img_request_put(ctx->img_request);
+out:
+ kmem_cache_free(rbd_journal_ctx_cache, ctx);
+ ceph_journaler_ctx_put(journaler_ctx);
+}
+
+static int rbd_journal_append_write_event(struct rbd_device *rbd_dev, struct bio *bio,
+ u64 offset, u64 length, struct rbd_journal_ctx *ctx)
+{
+ void *p = NULL;
+ struct ceph_journaler_ctx *journaler_ctx;
+ int ret = 0;
+
+ journaler_ctx = ceph_journaler_ctx_alloc();
+ if (!journaler_ctx) {
+ return -ENOMEM;
+ }
+
+ ctx->bio_iter.bio = bio;
+ ctx->bio_iter.iter = bio->bi_iter;
+
+ journaler_ctx->bio_iter = &ctx->bio_iter;
+ journaler_ctx->bio_len = length;
+
+ // EVENT_FIXED_SIZE(10 = CEPH_ENCODING_START_BLK_LEN(6) + EVENT_TYPE(4)) +
+ // offset(8) + length(8) + string_len(4) = 30
+ journaler_ctx->prefix_len = 30;
+ 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);
+ ceph_encode_64(&p, length);
+
+ // first part of ceph_encode_string();
+ ceph_encode_32(&p, journaler_ctx->bio_len);
+
+ journaler_ctx->priv = ctx;
+ journaler_ctx->callback = rbd_journal_callback;
+
+ ret = ceph_journaler_append(rbd_dev->journal->journaler, rbd_dev->journal->tag_tid,
+ &ctx->img_request->journaler_commit_tid, journaler_ctx);
+ if (ret) {
+ ceph_journaler_ctx_put(journaler_ctx);
+ return ret;
+ }
+ return 0;
+}
+
+static int rbd_journal_append_discard_event(struct rbd_device *rbd_dev, struct bio *bio,
+ u64 offset, u64 length, struct rbd_journal_ctx *ctx)
+{
+ void *p = NULL;
+ struct ceph_journaler_ctx *journaler_ctx;
+ int ret = 0;
+
+ journaler_ctx = ceph_journaler_ctx_alloc();
+ if (!journaler_ctx) {
+ return -ENOMEM;
+ }
+
+ ctx->bio_iter.bio = bio;
+ ctx->bio_iter.iter = bio->bi_iter;
+
+ journaler_ctx->bio_iter = &ctx->bio_iter;
+ journaler_ctx->bio_len = 0;
+
+ // EVENT_FIXED_SIZE(10 = CEPH_ENCODING_START_BLK_LEN(6) + EVENT_TYPE(4)) +
+ // offset(8) + length(8) = 26
+ journaler_ctx->prefix_len = 26;
+ 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 = ctx;
+ journaler_ctx->callback = rbd_journal_callback;
+
+ ret = ceph_journaler_append(rbd_dev->journal->journaler, rbd_dev->journal->tag_tid,
+ &ctx->img_request->journaler_commit_tid, journaler_ctx);
+ if (ret) {
+ ceph_journaler_ctx_put(journaler_ctx);
+ return ret;
+ }
+ return 0;
+}
+
+static int rbd_journal_append(struct rbd_device *rbd_dev, struct bio *bio,
+ u64 offset, u64 length, enum obj_operation_type op_type,
+ struct rbd_journal_ctx *ctx)
+{
+ switch (op_type) {
+ case OBJ_OP_WRITE:
+ return rbd_journal_append_write_event(rbd_dev, bio, offset, length, ctx);
+ case OBJ_OP_DISCARD:
+ return rbd_journal_append_discard_event(rbd_dev, bio, offset, length, ctx);
+ default:
+ return 0;
+ }
+}
+
/*
* Probe for the existence of the header object for the given rbd
* device. If this image is the one being mapped (i.e., not a
@@ -6369,11 +6550,18 @@ static int __init rbd_slab_init(void)
rbd_assert(!rbd_obj_request_cache);
rbd_obj_request_cache = KMEM_CACHE(rbd_obj_request, 0);
if (!rbd_obj_request_cache)
- goto out_err;
+ goto destroy_img_request_cache;
+ rbd_assert(!rbd_journal_ctx_cache);
+ rbd_journal_ctx_cache = KMEM_CACHE(rbd_journal_ctx, 0);
+ if (!rbd_journal_ctx_cache)
+ goto destroy_obj_request_cache;
return 0;
-out_err:
+destroy_obj_request_cache:
+ kmem_cache_destroy(rbd_obj_request_cache);
+ rbd_obj_request_cache = NULL;
+destroy_img_request_cache:
kmem_cache_destroy(rbd_img_request_cache);
rbd_img_request_cache = NULL;
return -ENOMEM;
--
1.8.3.1
