Whamcloud - gitweb
LU-2285 osp: Block new reservations during orphan recoveries
[fs/lustre-release.git] / lustre / osp / osp_internal.h
index e572552..ca2497d 100644 (file)
 #define _OSP_INTERNAL_H
 
 #include <obd.h>
+#include <obd_class.h>
 #include <dt_object.h>
 #include <lustre_fid.h>
 
+/*
+ * Infrastructure to support tracking of last committed llog record
+ */
+struct osp_id_tracker {
+       spinlock_t               otr_lock;
+       __u32                    otr_next_id;
+       __u32                    otr_committed_id;
+       /* callback is register once per diskfs -- that's the whole point */
+       struct dt_txn_callback   otr_tx_cb;
+       /* single node can run many clusters */
+       cfs_list_t               otr_wakeup_list;
+       cfs_list_t               otr_list;
+       /* underlying shared device */
+       struct dt_device        *otr_dev;
+       /* how many users of this tracker */
+       cfs_atomic_t             otr_refcount;
+};
+
 struct osp_device {
        struct dt_device                 opd_dt_dev;
        /* corresponded OST index */
@@ -53,8 +72,12 @@ struct osp_device {
        struct obd_export               *opd_storage_exp;
        struct dt_device                *opd_storage;
        struct dt_object                *opd_last_used_file;
-       /* protected by opd_pre_lock */
+
+       /* stored persistently in LE format, updated directly to/from disk
+        * and required le64_to_cpu() conversion before use.
+        * Protected by opd_pre_lock */
        volatile obd_id                  opd_last_used_id;
+
        obd_id                           opd_gap_start;
        int                              opd_gap_count;
        /* connection to OST */
@@ -76,6 +99,78 @@ struct osp_device {
         * reported via ->ldo_recovery_complete() */
        int                              opd_recovery_completed;
 
+       /*
+        * Precreation pool
+        */
+       spinlock_t                       opd_pre_lock;
+       /* last id assigned in creation */
+       __u64                            opd_pre_used_id;
+       /* last created id OST reported, next-created - available id's */
+       __u64                            opd_pre_last_created;
+       /* how many ids are reserved in declare, we shouldn't block in create */
+       __u64                            opd_pre_reserved;
+       /* dedicate precreate thread */
+       struct ptlrpc_thread             opd_pre_thread;
+       /* thread waits for signals about pool going empty */
+       cfs_waitq_t                      opd_pre_waitq;
+       /* consumers (who needs new ids) wait here */
+       cfs_waitq_t                      opd_pre_user_waitq;
+       /* current precreation status: working, failed, stopping? */
+       int                              opd_pre_status;
+       /* how many to precreate next time */
+       int                              opd_pre_grow_count;
+       int                              opd_pre_min_grow_count;
+       int                              opd_pre_max_grow_count;
+       /* whether to grow precreation window next time or not */
+       int                              opd_pre_grow_slow;
+       /* cleaning up orphans or recreating missing objects */
+       int                              opd_pre_recovering;
+
+       /*
+        * OST synchronization
+        */
+       spinlock_t                       opd_syn_lock;
+       /* unique generation, to recognize start of new records in the llog */
+       struct llog_gen                  opd_syn_generation;
+       /* number of changes to sync, used to wake up sync thread */
+       unsigned long                    opd_syn_changes;
+       /* processing of changes from previous mount is done? */
+       int                              opd_syn_prev_done;
+       /* found records */
+       struct ptlrpc_thread             opd_syn_thread;
+       cfs_waitq_t                      opd_syn_waitq;
+       /* list of remotely committed rpc */
+       cfs_list_t                       opd_syn_committed_there;
+       /* number of changes being under sync */
+       int                              opd_syn_sync_in_progress;
+       /* number of RPCs in flight - flow control */
+       int                              opd_syn_rpc_in_flight;
+       int                              opd_syn_max_rpc_in_flight;
+       /* number of RPC in processing (including non-committed by OST) */
+       int                              opd_syn_rpc_in_progress;
+       int                              opd_syn_max_rpc_in_progress;
+       /* osd api's commit cb control structure */
+       struct dt_txn_callback           opd_syn_txn_cb;
+       /* last used change number -- semantically similar to transno */
+       unsigned long                    opd_syn_last_used_id;
+       /* last committed change number -- semantically similar to
+        * last_committed */
+       unsigned long                    opd_syn_last_committed_id;
+       /* last processed (taken from llog) id */
+       unsigned long                    opd_syn_last_processed_id;
+       struct osp_id_tracker           *opd_syn_tracker;
+       cfs_list_t                       opd_syn_ontrack;
+
+       /*
+        * statfs related fields: OSP maintains it on its own
+        */
+       struct obd_statfs                opd_statfs;
+       cfs_time_t                       opd_statfs_fresh_till;
+       cfs_timer_t                      opd_statfs_timer;
+       int                              opd_statfs_update_in_progress;
+       /* how often to update statfs data */
+       int                              opd_statfs_maxage;
+
        cfs_proc_dir_entry_t            *opd_symlink;
 };
 
@@ -85,17 +180,28 @@ extern cfs_mem_cache_t *osp_object_kmem;
 struct osp_object {
        struct lu_object_header  opo_header;
        struct dt_object         opo_obj;
-       int                      opo_reserved;
+       int                      opo_reserved:1,
+                                opo_new:1;
 };
 
 extern struct lu_object_operations osp_lu_obj_ops;
+extern const struct dt_device_operations osp_dt_ops;
 
 struct osp_thread_info {
        struct lu_buf            osi_lb;
        struct lu_fid            osi_fid;
        struct lu_attr           osi_attr;
+       struct ost_id            osi_oi;
        obd_id                   osi_id;
        loff_t                   osi_off;
+       union {
+               struct llog_rec_hdr             osi_hdr;
+               struct llog_unlink64_rec        osi_unlink;
+               struct llog_setattr64_rec       osi_setattr;
+               struct llog_gen_rec             osi_gen;
+       };
+       struct llog_cookie       osi_cookie;
+       struct llog_catid        osi_cid;
 };
 
 static inline void osp_objid_buf_prep(struct osp_thread_info *osi,
@@ -190,7 +296,33 @@ static inline struct dt_object *osp_object_child(struct osp_object *o)
 /* osp_dev.c */
 void osp_update_last_id(struct osp_device *d, obd_id objid);
 
+/* osp_precreate.c */
+int osp_init_precreate(struct osp_device *d);
+int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d);
+__u64 osp_precreate_get_id(struct osp_device *d);
+void osp_precreate_fini(struct osp_device *d);
+int osp_object_truncate(const struct lu_env *env, struct dt_object *dt, __u64);
+void osp_pre_update_status(struct osp_device *d, int rc);
+void osp_statfs_need_now(struct osp_device *d);
+
 /* lproc_osp.c */
 void lprocfs_osp_init_vars(struct lprocfs_static_vars *lvars);
+void osp_lprocfs_init(struct osp_device *osp);
+
+/* osp_sync.c */
+int osp_sync_declare_add(const struct lu_env *env, struct osp_object *o,
+                        llog_op_type type, struct thandle *th);
+int osp_sync_add(const struct lu_env *env, struct osp_object *o,
+                llog_op_type type, struct thandle *th,
+                const struct lu_attr *attr);
+int osp_sync_init(const struct lu_env *env, struct osp_device *d);
+int osp_sync_fini(struct osp_device *d);
+void __osp_sync_check_for_work(struct osp_device *d);
+
+/* osp_ost.c */
+int osp_init_for_ost(const struct lu_env *env, struct osp_device *m,
+                    struct lu_device_type *ldt, struct lustre_cfg *cfg);
+int osp_disconnect(struct osp_device *d);
+int osp_fini_for_ost(struct osp_device *osp);
 
 #endif