When we found there is no client refre to the object set, we can remove the objects. Signed-off-by: Dongsheng Yang <dongsheng.yang@xxxxxxxxxxxx> --- net/ceph/journaler.c | 119 ++++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 118 insertions(+), 1 deletion(-) diff --git a/net/ceph/journaler.c b/net/ceph/journaler.c index e0dce2a..1a3651e 100644 --- a/net/ceph/journaler.c +++ b/net/ceph/journaler.c @@ -142,6 +142,10 @@ void ceph_journaler_destroy(struct ceph_journaler *journaler) } EXPORT_SYMBOL(ceph_journaler_destroy); +static int remove_set(struct ceph_journaler *journaler, uint64_t object_set); +static int set_minimum_set(struct ceph_journaler* journaler, + uint64_t minimum_set); + static int refresh(struct ceph_journaler *journaler, bool init) { int ret = 0; @@ -207,6 +211,34 @@ static int refresh(struct ceph_journaler *journaler, bool init) } spin_unlock(&journaler->meta_lock); + if (need_advance) { + spin_lock(&journaler->advancing_lock); + journaler->advancing = false; + journaler->active_set = active_set; + spin_unlock(&journaler->advancing_lock); + + queue_work(journaler->task_wq, &journaler->flush_work); + } + + // remove set if necessary + if (minimum_commit_set > minimum_set) { + uint64_t trim_set = minimum_set; + while (trim_set < minimum_commit_set) { + ret = remove_set(journaler, trim_set); + if (ret < 0 && ret != -ENOENT) { + pr_err("failed to trim object_set: %llu", trim_set); + return ret; + } + trim_set++; + } + + ret = set_minimum_set(journaler, minimum_commit_set); + if (ret < 0) { + pr_err("failed to set minimum set to %llu", minimum_commit_set); + return ret; + } + } + return 0; } @@ -1293,7 +1325,7 @@ static bool advance_object_set(struct ceph_journaler *journaler) pr_err("error in set active_set: %d", ret); } - queue_work(journaler->task_wq, &journaler->notify_update_work); + queue_work(journaler->notify_wq, &journaler->notify_update_work); return true; } @@ -1724,3 +1756,88 @@ int ceph_journaler_allocate_tag(struct ceph_journaler *journaler, return ret; } EXPORT_SYMBOL(ceph_journaler_allocate_tag); + +// trimming +static int ceph_journaler_obj_remove_sync(struct ceph_journaler *journaler, + struct ceph_object_id *oid, + struct ceph_object_locator *oloc) + +{ + struct ceph_osd_client *osdc = journaler->osdc; + struct ceph_osd_request *req; + int ret; + + req = ceph_osdc_alloc_request(osdc, NULL, 1, false, GFP_KERNEL); + if (!req) + return -ENOMEM; + + ceph_oid_copy(&req->r_base_oid, oid); + ceph_oloc_copy(&req->r_base_oloc, oloc); + req->r_flags = CEPH_OSD_FLAG_WRITE; + + osd_req_op_init(req, 0, CEPH_OSD_OP_DELETE, 0); + + ret = ceph_osdc_alloc_messages(req, GFP_KERNEL); + if (ret) + goto out_req; + + ceph_osdc_start_request(osdc, req, false); + ret = ceph_osdc_wait_request(osdc, req); + +out_req: + ceph_osdc_put_request(req); + return ret; +} + +static int remove_set(struct ceph_journaler *journaler, uint64_t object_set) +{ + uint64_t object_num = 0; + int splay_offset = 0; + struct ceph_object_id object_oid; + int ret = 0; + + ceph_oid_init(&object_oid); + for (splay_offset = 0; splay_offset < journaler->splay_width; splay_offset++) { + object_num = splay_offset + (object_set * journaler->splay_width); + if (!ceph_oid_empty(&object_oid)) { + ceph_oid_destroy(&object_oid); + ceph_oid_init(&object_oid); + } + ret = ceph_oid_aprintf(&object_oid, GFP_KERNEL, "%s%llu", + journaler->object_oid_prefix, object_num); + if (ret) { + pr_err("aprintf error : %d", ret); + goto out; + } + ret = ceph_journaler_obj_remove_sync(journaler, &object_oid, + &journaler->data_oloc); + if (ret < 0 && ret != -ENOENT) { + pr_err("%s: failed to remove object: %llu", + __func__, object_num); + goto out; + } + } + ret = 0; +out: + ceph_oid_destroy(&object_oid); + return ret; +} + +static int set_minimum_set(struct ceph_journaler* journaler, + uint64_t minimum_set) +{ + int ret = 0; + + ret = ceph_cls_journaler_set_minimum_set(journaler->osdc, + &journaler->header_oid, + &journaler->header_oloc, + minimum_set); + if (ret < 0) { + pr_err("%s: failed to set_minimum_set: %d", __func__, ret); + return ret; + } + + queue_work(journaler->notify_wq, &journaler->notify_update_work); + + return ret; +} -- 1.8.3.1