Whamcloud - gitweb
LU-12930 various: use schedule_timeout_*interruptible
[fs/lustre-release.git] / lustre / osp / osp_internal.h
index fd2d41d..c78694d 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2014, Intel Corporation.
+ * Copyright (c) 2012, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #include <lustre_fid.h>
 #include <lustre_update.h>
 #include <lu_target.h>
-#include <lustre_mdc.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;
+       __u64                    otr_next_id;
+       __u64                    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 */
@@ -72,7 +67,6 @@ struct osp_precreate {
        /*
         * Precreation pool
         */
-       spinlock_t                       osp_pre_lock;
 
        /* last fid to assign in creation */
        struct lu_fid                    osp_pre_used_fid;
@@ -84,16 +78,79 @@ struct osp_precreate {
        wait_queue_head_t                osp_pre_user_waitq;
        /* current precreation status: working, failed, stopping? */
        int                              osp_pre_status;
-       /* how many to precreate next time */
-       int                              osp_pre_grow_count;
-       int                              osp_pre_min_grow_count;
-       int                              osp_pre_max_grow_count;
-       /* whether to grow precreation window next time or not */
-       int                              osp_pre_grow_slow;
+       /* how many objects to precreate next time */
+       int                              osp_pre_create_count;
+       int                              osp_pre_min_create_count;
+       int                              osp_pre_max_create_count;
+       /* whether to increase precreation window next time or not */
+       int                              osp_pre_create_slow;
        /* cleaning up orphans or recreating missing objects */
        int                              osp_pre_recovering;
 };
 
+struct osp_update_request_sub {
+       struct object_update_request    *ours_req; /* may be vmalloc'd */
+       size_t                          ours_req_size;
+       /* Linked to osp_update_request->our_req_list */
+       struct list_head                ours_list;
+};
+
+struct osp_update_request {
+       int                             our_flags;
+       /* update request result */
+       int                             our_rc;
+
+       /* List of osp_update_request_sub */
+       struct list_head                our_req_list;
+       int                             our_req_nr;
+       int                             our_update_nr;
+
+       struct list_head                our_cb_items;
+       struct list_head                our_invalidate_cb_list;
+
+       /* points to thandle if this update request belongs to one */
+       struct osp_thandle              *our_th;
+
+       __u64                           our_version;
+       __u64                           our_generation;
+       /* protect our_list and flag */
+       spinlock_t                      our_list_lock;
+       /* linked to the list(ou_list) in osp_updates */
+       struct list_head                our_list;
+       __u32                           our_batchid;
+       __u32                           our_req_ready:1;
+
+};
+
+struct osp_updates {
+       struct list_head        ou_list;
+       spinlock_t              ou_lock;
+       wait_queue_head_t       ou_waitq;
+
+       /* The next rpc version which supposed to be sent in
+        * osp_send_update_thread().*/
+       __u64                   ou_rpc_version;
+
+       /* The rpc version assigned to the osp thandle during (osp_md_write()),
+        * which will be sent by this order. Note: the osp_thandle has be sent
+        * by this order to make sure the remote update log will follow the
+        * llog format rule. XXX: these probably should be removed once we
+        * invent new llog format */
+       __u64                   ou_version;
+
+       /* The generation of current osp update RPC, which is used to make sure
+        * those stale RPC(with older generation) will not be sent, otherwise it
+        * will cause update lllog corruption */
+       __u64                   ou_generation;
+};
+
+struct osp_rpc_lock {
+       /** Lock protecting in-flight RPC concurrency. */
+       struct mutex            rpcl_mutex;
+       /** Used for MDS/RPC load testing purposes. */
+       unsigned int            rpcl_fakes;
+};
+
 struct osp_device {
        struct dt_device                 opd_dt_dev;
        /* corresponded OST index */
@@ -112,16 +169,16 @@ struct osp_device {
         * and required le64_to_cpu() conversion before use.
         * Protected by opd_pre_lock */
        struct lu_fid                   opd_last_used_fid;
+       /* on disk copy last_used_fid.f_oid or idif */
+       u64                             opd_last_id;
        struct lu_fid                   opd_gap_start_fid;
        int                              opd_gap_count;
        /* connection to OST */
+       struct osp_rpc_lock              opd_rpc_lock;
        struct obd_device               *opd_obd;
        struct obd_export               *opd_exp;
-       struct obd_uuid                  opd_cluuid;
        struct obd_connect_data         *opd_connect_data;
        int                              opd_connects;
-       struct proc_dir_entry           *opd_proc_entry;
-       struct lprocfs_stats            *opd_stats;
        /* connection status. */
        unsigned int                     opd_new_connection:1,
                                         opd_got_disconnected:1,
@@ -138,82 +195,96 @@ struct osp_device {
        struct osp_precreate            *opd_pre;
        /* dedicate precreate thread */
        struct ptlrpc_thread             opd_pre_thread;
+       spinlock_t                       opd_pre_lock;
        /* thread waits for signals about pool going empty */
        wait_queue_head_t                opd_pre_waitq;
 
+       /* send update thread */
+       struct osp_updates              *opd_update;
+       /* dedicate update thread */
+       struct ptlrpc_thread             opd_update_thread;
+
        /*
-        * OST synchronization
+        * OST synchronization thread
         */
-       spinlock_t                       opd_syn_lock;
+       spinlock_t                       opd_sync_lock;
        /* unique generation, to recognize start of new records in the llog */
-       struct llog_gen                  opd_syn_generation;
+       struct llog_gen                  opd_sync_generation;
        /* number of changes to sync, used to wake up sync thread */
-       unsigned long                    opd_syn_changes;
+       atomic_t                         opd_sync_changes;
        /* processing of changes from previous mount is done? */
-       int                              opd_syn_prev_done;
+       int                              opd_sync_prev_done;
        /* found records */
-       struct ptlrpc_thread             opd_syn_thread;
-       wait_queue_head_t                opd_syn_waitq;
+       struct ptlrpc_thread             opd_sync_thread;
+       wait_queue_head_t                opd_sync_waitq;
+       /* list of in flight rpcs */
+       struct list_head                 opd_sync_in_flight_list;
        /* list of remotely committed rpc */
-       struct list_head                 opd_syn_committed_there;
-       /* number of changes being under sync */
-       int                              opd_syn_sync_in_progress;
+       struct list_head                 opd_sync_committed_there;
        /* number of RPCs in flight - flow control */
-       int                              opd_syn_rpc_in_flight;
-       int                              opd_syn_max_rpc_in_flight;
+       atomic_t                         opd_sync_rpcs_in_flight;
+       int                              opd_sync_max_rpcs_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;
+       atomic_t                         opd_sync_rpcs_in_progress;
+       int                              opd_sync_max_rpcs_in_progress;
        /* osd api's commit cb control structure */
-       struct dt_txn_callback           opd_syn_txn_cb;
+       struct dt_txn_callback           opd_sync_txn_cb;
        /* last used change number -- semantically similar to transno */
-       unsigned long                    opd_syn_last_used_id;
+       unsigned long                    opd_sync_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;
-       struct list_head                 opd_syn_ontrack;
+       __u64                            opd_sync_last_committed_id;
+       /* last processed catalog index */
+       int                              opd_sync_last_catalog_idx;
+       /* number of processed records */
+       atomic64_t                       opd_sync_processed_recs;
        /* stop processing new requests until barrier=0 */
-       atomic_t                         opd_syn_barrier;
-       wait_queue_head_t                opd_syn_barrier_waitq;
+       atomic_t                         opd_sync_barrier;
+       wait_queue_head_t                opd_sync_barrier_waitq;
+       /* last generated id */
+       ktime_t                          opd_sync_next_commit_cb;
+       atomic_t                         opd_commits_registered;
 
        /*
         * statfs related fields: OSP maintains it on its own
         */
        struct obd_statfs                opd_statfs;
-       cfs_time_t                       opd_statfs_fresh_till;
-       struct timer_list                        opd_statfs_timer;
+       ktime_t                          opd_statfs_fresh_till;
+       struct timer_list                opd_statfs_timer;
        int                              opd_statfs_update_in_progress;
        /* how often to update statfs data */
-       int                              opd_statfs_maxage;
+       time64_t                         opd_statfs_maxage;
 
-       struct proc_dir_entry           *opd_symlink;
+       struct dentry                   *opd_debugfs;
 
        /* If the caller wants to do some idempotent async operations on
         * remote server, it can append the async remote requests on the
         * osp_device::opd_async_requests via declare() functions, these
         * requests can be packed together and sent to the remote server
         * via single OUT RPC later. */
-       struct dt_update_request        *opd_async_requests;
+       struct osp_update_request       *opd_async_requests;
        /* Protect current operations on opd_async_requests. */
        struct mutex                     opd_async_requests_mutex;
        struct list_head                 opd_async_updates;
        struct rw_semaphore              opd_async_updates_rwsem;
        atomic_t                         opd_async_updates_count;
+
+       /*
+        * Limit the object allocation using ENOSPC for opd_pre_status
+        */
+       int                             opd_reserved_mb_high;
+       int                             opd_reserved_mb_low;
 };
 
-#define opd_pre_lock                   opd_pre->osp_pre_lock
 #define opd_pre_used_fid               opd_pre->osp_pre_used_fid
 #define opd_pre_last_created_fid       opd_pre->osp_pre_last_created_fid
 #define opd_pre_reserved               opd_pre->osp_pre_reserved
 #define opd_pre_user_waitq             opd_pre->osp_pre_user_waitq
 #define opd_pre_status                 opd_pre->osp_pre_status
-#define opd_pre_grow_count             opd_pre->osp_pre_grow_count
-#define opd_pre_min_grow_count         opd_pre->osp_pre_min_grow_count
-#define opd_pre_max_grow_count         opd_pre->osp_pre_max_grow_count
-#define opd_pre_grow_slow              opd_pre->osp_pre_grow_slow
+#define opd_pre_create_count           opd_pre->osp_pre_create_count
+#define opd_pre_min_create_count       opd_pre->osp_pre_min_create_count
+#define opd_pre_max_create_count       opd_pre->osp_pre_max_create_count
+#define opd_pre_create_slow            opd_pre->osp_pre_create_slow
 #define opd_pre_recovering             opd_pre->osp_pre_recovering
 
 extern struct kmem_cache *osp_object_kmem;
@@ -232,24 +303,25 @@ struct osp_xattr_entry {
        char                     oxe_buf[0];
 };
 
-struct osp_object_attr {
-       struct lu_attr          ooa_attr;
-       struct list_head        ooa_xattr_list;
-};
-
 /* this is a top object */
 struct osp_object {
        struct lu_object_header opo_header;
        struct dt_object        opo_obj;
        unsigned int            opo_reserved:1,
-                               opo_non_exist:1;
+                               opo_non_exist:1,
+                               opo_stale:1;
 
        /* read/write lock for md osp object */
        struct rw_semaphore     opo_sem;
        const struct lu_env     *opo_owner;
-       struct osp_object_attr *opo_ooa;
+       struct lu_attr          opo_attr;
+       struct list_head        opo_xattr_list;
+       struct list_head        opo_invalidate_cb_list;
        /* Protect opo_ooa. */
        spinlock_t              opo_lock;
+       /* to implement in-flight invalidation */
+       atomic_t                opo_invalidate_seq;
+       struct rw_semaphore     opo_invalidate_sem;
 };
 
 extern struct lu_object_operations osp_lu_obj_ops;
@@ -264,12 +336,11 @@ struct osp_thread_info {
        struct lu_attr           osi_attr;
        struct ost_id            osi_oi;
        struct ost_id            osi_oi2;
-       u64                      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_setattr64_rec_v2    osi_setattr;
                struct llog_gen_rec             osi_gen;
        };
        struct llog_cookie       osi_cookie;
@@ -284,6 +355,7 @@ struct osp_it {
        __u32                     ooi_pos_page;
        __u32                     ooi_pos_lu_page;
        __u32                     ooi_attr;
+       __u32                     ooi_rec_size;
        int                       ooi_pos_ent;
        int                       ooi_total_npages;
        int                       ooi_valid_npages;
@@ -296,12 +368,17 @@ struct osp_it {
        struct page              **ooi_pages;
 };
 
+#define OSP_THANDLE_MAGIC      0x20141214
 struct osp_thandle {
        struct thandle           ot_super;
-       struct dt_update_request *ot_dur;
 
        /* OSP will use this thandle to update last oid*/
        struct thandle          *ot_storage_th;
+       __u32                    ot_magic;
+       struct list_head         ot_commit_dcb_list;
+       struct list_head         ot_stop_dcb_list;
+       struct osp_update_request *ot_our;
+       atomic_t                 ot_refcount;
 };
 
 static inline struct osp_thandle *
@@ -310,13 +387,13 @@ thandle_to_osp_thandle(struct thandle *th)
        return container_of(th, struct osp_thandle, ot_super);
 }
 
-static inline struct dt_update_request *
-thandle_to_dt_update_request(struct thandle *th)
+static inline struct osp_update_request *
+thandle_to_osp_update_request(struct thandle *th)
 {
        struct osp_thandle *oth;
 
        oth = thandle_to_osp_thandle(th);
-       return oth->ot_dur;
+       return oth->ot_our;
 }
 
 /* The transaction only include the updates on the remote node, and
@@ -327,7 +404,7 @@ static inline bool is_only_remote_trans(struct thandle *th)
 }
 
 static inline void osp_objid_buf_prep(struct lu_buf *buf, loff_t *off,
-                                     __u32 *id, int index)
+                                     __u64 *id, int index)
 {
        /* Note: through id is only 32 bits, it will also write 64 bits
         * for oid to keep compatibility with the previous version. */
@@ -354,19 +431,11 @@ extern struct lu_context_key osp_thread_key;
 
 static inline struct osp_thread_info *osp_env_info(const struct lu_env *env)
 {
-       struct osp_thread_info *info;
-
-       info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
-       if (info == NULL) {
-               lu_env_refill((struct lu_env *)env);
-               info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
-       }
-       LASSERT(info);
-       return info;
+       return lu_env_info(env, &osp_thread_key);
 }
 
 struct osp_txn_info {
-       __u32   oti_current_id;
+       __u64   oti_current_id;
 };
 
 extern struct lu_context_key osp_txn_key;
@@ -436,20 +505,79 @@ static inline struct seq_server_site *osp_seq_site(struct osp_device *osp)
        return osp->opd_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
 }
 
-#define osp_init_rpc_lock(lck) mdc_init_rpc_lock(lck)
+/**
+ * Serializes in-flight MDT-modifying RPC requests to preserve idempotency.
+ *
+ * This mutex is used to implement execute-once semantics on the MDT.
+ * The MDT stores the last transaction ID and result for every client in
+ * its last_rcvd file. If the client doesn't get a reply, it can safely
+ * resend the request and the MDT will reconstruct the reply being aware
+ * that the request has already been executed. Without this lock,
+ * execution status of concurrent in-flight requests would be
+ * overwritten.
+ *
+ * This imlpementation limits the extent to which we can keep a full pipeline
+ * of in-flight requests from a single client.  This limitation can be
+ * overcome by allowing multiple slots per client in the last_rcvd file,
+ * see LU-6864.
+ */
+#define OSP_FAKE_RPCL_IT ((void *)0x2c0012bfUL)
+
+static inline void osp_init_rpc_lock(struct osp_device *osp)
+{
+       struct osp_rpc_lock *lck = &osp->opd_rpc_lock;
+
+       mutex_init(&lck->rpcl_mutex);
+       lck->rpcl_fakes = 0;
+}
 
 static inline void osp_get_rpc_lock(struct osp_device *osp)
 {
-       struct mdc_rpc_lock *rpc_lock = osp->opd_obd->u.cli.cl_rpc_lock;
+       struct osp_rpc_lock *lck = &osp->opd_rpc_lock;
+
+       /* This would normally block until the existing request finishes.
+        * If fail_loc is set it will block until the regular request is
+        * done, then increment rpcl_fakes.  Once that is non-zero it
+        * will only be cleared when all fake requests are finished.
+        * Only when all fake requests are finished can normal requests
+        * be sent, to ensure they are recoverable again.
+        */
+ again:
+       mutex_lock(&lck->rpcl_mutex);
 
-       mdc_get_rpc_lock(rpc_lock, NULL);
+       if (CFS_FAIL_CHECK_QUIET(OBD_FAIL_MDC_RPCS_SEM) ||
+           CFS_FAIL_CHECK_QUIET(OBD_FAIL_OSP_RPCS_SEM)) {
+               lck->rpcl_fakes++;
+               mutex_unlock(&lck->rpcl_mutex);
+
+               return;
+       }
+
+       /* This will only happen when the CFS_FAIL_CHECK() was just turned
+        * off but there are still requests in progress.  Wait until they
+        * finish.  It doesn't need to be efficient in this extremely rare
+        * case, just have low overhead in the common case when it isn't true.
+        */
+       if (unlikely(lck->rpcl_fakes)) {
+               mutex_unlock(&lck->rpcl_mutex);
+               schedule_timeout_uninterruptible(cfs_time_seconds(1) / 4);
+
+               goto again;
+       }
 }
 
 static inline void osp_put_rpc_lock(struct osp_device *osp)
 {
-       struct mdc_rpc_lock *rpc_lock = osp->opd_obd->u.cli.cl_rpc_lock;
+       struct osp_rpc_lock *lck = &osp->opd_rpc_lock;
 
-       mdc_put_rpc_lock(rpc_lock, NULL);
+       if (lck->rpcl_fakes) { /* OBD_FAIL_OSP_RPCS_SEM */
+               mutex_lock(&lck->rpcl_mutex);
+
+               if (lck->rpcl_fakes) /* check again under lock */
+                       lck->rpcl_fakes--;
+       }
+
+       mutex_unlock(&lck->rpcl_mutex);
 }
 
 static inline int osp_fid_diff(const struct lu_fid *fid1,
@@ -470,16 +598,26 @@ static inline int osp_fid_diff(const struct lu_fid *fid1,
                       fid_idif_id(fid2->f_seq, fid2->f_oid, 0);
        }
 
-       LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID
-                ", fid2:"DFID"\n", PFID(fid1), PFID(fid2));
+       LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID", fid2:"DFID"\n",
+                PFID(fid1), PFID(fid2));
 
        return fid_oid(fid1) - fid_oid(fid2);
 }
 
+static inline void osp_fid_to_obdid(struct lu_fid *last_fid, u64 *osi_id)
+{
+       if (fid_is_idif((last_fid)))
+               *osi_id = fid_idif_id(fid_seq(last_fid), fid_oid(last_fid),
+                                     fid_ver(last_fid));
+       else
+               *osi_id = fid_oid(last_fid);
+}
 
 static inline void osp_update_last_fid(struct osp_device *d, struct lu_fid *fid)
 {
        int diff = osp_fid_diff(fid, &d->opd_last_used_fid);
+       struct lu_fid *gap_start = &d->opd_gap_start_fid;
+
        /*
         * we might have lost precreated objects due to VBR and precreate
         * orphans, the gap in objid can be calculated properly only here
@@ -487,12 +625,19 @@ static inline void osp_update_last_fid(struct osp_device *d, struct lu_fid *fid)
        if (diff > 0) {
                if (diff > 1) {
                        d->opd_gap_start_fid = d->opd_last_used_fid;
-                       d->opd_gap_start_fid.f_oid++;
+                       if (fid_oid(gap_start) == LUSTRE_DATA_SEQ_MAX_WIDTH) {
+                               gap_start->f_seq++;
+                               gap_start->f_oid = fid_is_idif(gap_start) ?
+                                                              0 : 1;
+                       } else {
+                               gap_start->f_oid++;
+                       }
                        d->opd_gap_count = diff - 1;
                        CDEBUG(D_HA, "Gap in objids: start="DFID", count =%d\n",
                               PFID(&d->opd_gap_start_fid), d->opd_gap_count);
                }
                d->opd_last_used_fid = *fid;
+               osp_fid_to_obdid(fid, &d->opd_last_id);
        }
 }
 
@@ -535,6 +680,74 @@ static inline int osp_is_fid_client(struct osp_device *osp)
        return imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_FID;
 }
 
+struct object_update *
+update_buffer_get_update(struct object_update_request *request,
+                        unsigned int index);
+
+int osp_extend_update_buffer(const struct lu_env *env,
+                            struct osp_update_request *our);
+
+struct osp_update_request_sub *
+osp_current_object_update_request(struct osp_update_request *our);
+
+int osp_object_update_request_create(struct osp_update_request *our,
+                                    size_t size);
+
+#define OSP_UPDATE_RPC_PACK(env, out_something_pack, our, ...)         \
+({                                                                     \
+       struct object_update *object_update;                            \
+       size_t max_update_length;                                       \
+       struct osp_update_request_sub *ours;                            \
+       int ret;                                                        \
+                                                                       \
+       while (1) {                                                     \
+               ours = osp_current_object_update_request(our);          \
+               LASSERT(ours != NULL);                                  \
+               max_update_length = ours->ours_req_size -               \
+                           object_update_request_size(ours->ours_req); \
+                                                                       \
+               object_update = update_buffer_get_update(ours->ours_req,\
+                                        ours->ours_req->ourq_count);   \
+               ret = out_something_pack(env, object_update,            \
+                                        &max_update_length,            \
+                                        __VA_ARGS__);                  \
+               if (ret == -E2BIG) {                                    \
+                       int rc1;                                        \
+                       /* Create new object update request */          \
+                       rc1 = osp_object_update_request_create(our,     \
+                               max_update_length  +                    \
+                               offsetof(struct object_update_request,  \
+                                        ourq_updates[0]) + 1);         \
+                       if (rc1 != 0) {                                 \
+                               ret = rc1;                              \
+                               break;                                  \
+                       }                                               \
+                       continue;                                       \
+               } else {                                                \
+                       if (ret == 0) {                                 \
+                               ours->ours_req->ourq_count++;           \
+                               (our)->our_update_nr++;                 \
+                               object_update->ou_batchid =             \
+                                                    (our)->our_batchid;\
+                               object_update->ou_flags |=              \
+                                                    (our)->our_flags;  \
+                       }                                               \
+                       break;                                          \
+               }                                                       \
+       }                                                               \
+       ret;                                                            \
+})
+
+static inline bool osp_send_update_thread_running(struct osp_device *osp)
+{
+       return osp->opd_update_thread.t_flags & SVC_RUNNING;
+}
+
+static inline bool osp_send_update_thread_stopped(struct osp_device *osp)
+{
+       return osp->opd_update_thread.t_flags & SVC_STOPPED;
+}
+
 typedef int (*osp_update_interpreter_t)(const struct lu_env *env,
                                        struct object_update_reply *rep,
                                        struct ptlrpc_request *req,
@@ -543,42 +756,60 @@ typedef int (*osp_update_interpreter_t)(const struct lu_env *env,
 
 /* osp_dev.c */
 void osp_update_last_id(struct osp_device *d, u64 objid);
-extern struct llog_operations osp_mds_ost_orig_logops;
 
 /* osp_trans.c */
 int osp_insert_async_request(const struct lu_env *env, enum update_type op,
                             struct osp_object *obj, int count, __u16 *lens,
-                            const void **bufs, void *data,
+                            const void **bufs, void *data, __u32 repsize,
                             osp_update_interpreter_t interpreter);
 
 int osp_unplug_async_request(const struct lu_env *env,
                             struct osp_device *osp,
-                            struct dt_update_request *update);
+                            struct osp_update_request *update);
 int osp_trans_update_request_create(struct thandle *th);
 struct thandle *osp_trans_create(const struct lu_env *env,
                                 struct dt_device *d);
 int osp_trans_start(const struct lu_env *env, struct dt_device *dt,
                    struct thandle *th);
 int osp_insert_update_callback(const struct lu_env *env,
-                              struct dt_update_request *update,
+                              struct osp_update_request *update,
                               struct osp_object *obj, void *data,
                               osp_update_interpreter_t interpreter);
-int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
-                       const struct object_update_request *ureq,
-                       struct ptlrpc_request **reqp);
-struct dt_update_request *dt_update_request_create(struct dt_device *dt);
-void dt_update_request_destroy(struct dt_update_request *dt_update);
+
+struct osp_update_request *osp_update_request_create(struct dt_device *dt);
+void osp_update_request_destroy(const struct lu_env *env,
+                               struct osp_update_request *update);
+
+int osp_send_update_thread(void *arg);
+int osp_check_and_set_rpc_version(struct osp_thandle *oth,
+                                 struct osp_object *obj);
+
+void osp_thandle_destroy(const struct lu_env *env, struct osp_thandle *oth);
+static inline void osp_thandle_get(struct osp_thandle *oth)
+{
+       atomic_inc(&oth->ot_refcount);
+}
+
+static inline void osp_thandle_put(const struct lu_env *env,
+                                  struct osp_thandle *oth)
+{
+       if (atomic_dec_and_test(&oth->ot_refcount))
+               osp_thandle_destroy(env, oth);
+}
 
 int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
-                       const struct object_update_request *ureq,
+                       struct osp_update_request *our,
                        struct ptlrpc_request **reqp);
 int osp_remote_sync(const struct lu_env *env, struct osp_device *osp,
-                   struct dt_update_request *update,
+                   struct osp_update_request *update,
                    struct ptlrpc_request **reqp);
 
 struct thandle *osp_get_storage_thandle(const struct lu_env *env,
                                        struct thandle *th,
                                        struct osp_device *osp);
+void osp_trans_callback(const struct lu_env *env,
+                       struct osp_thandle *oth, int rc);
+void osp_invalidate_request(struct osp_device *osp);
 /* osp_object.c */
 int osp_attr_get(const struct lu_env *env, struct dt_object *dt,
                 struct lu_attr *attr);
@@ -594,9 +825,12 @@ int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
                          const char *name, struct thandle *th);
 int osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
                  const char *name, struct thandle *th);
+int osp_invalidate(const struct lu_env *env, struct dt_object *dt);
+void osp_obj_invalidate_cache(struct osp_object *obj);
 
 int osp_trans_stop(const struct lu_env *env, struct dt_device *dt,
                   struct thandle *th);
+int osp_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb);
 
 struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt,
                          __u32 attr);
@@ -609,15 +843,12 @@ int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di,
                   void *key_rec);
 int osp_it_next_page(const struct lu_env *env, struct dt_it *di);
 /* osp_md_object.c */
-int osp_md_declare_object_create(const struct lu_env *env,
-                                struct dt_object *dt,
-                                struct lu_attr *attr,
-                                struct dt_allocation_hint *hint,
-                                struct dt_object_format *dof,
-                                struct thandle *th);
-int osp_md_object_create(const struct lu_env *env, struct dt_object *dt,
-                        struct lu_attr *attr, struct dt_allocation_hint *hint,
-                        struct dt_object_format *dof, struct thandle *th);
+int osp_md_declare_create(const struct lu_env *env, struct dt_object *dt,
+                         struct lu_attr *attr, struct dt_allocation_hint *hint,
+                         struct dt_object_format *dof, struct thandle *th);
+int osp_md_create(const struct lu_env *env, struct dt_object *dt,
+                 struct lu_attr *attr, struct dt_allocation_hint *hint,
+                 struct dt_object_format *dof, struct thandle *th);
 int osp_md_declare_attr_set(const struct lu_env *env, struct dt_object *dt,
                            const struct lu_attr *attr, struct thandle *th);
 int osp_md_attr_set(const struct lu_env *env, struct dt_object *dt,
@@ -638,22 +869,48 @@ int osp_reset_last_used(const struct lu_env *env, struct osp_device *osp);
 int osp_write_last_oid_seq_files(struct lu_env *env, struct osp_device *osp,
                                 struct lu_fid *fid, int sync);
 int osp_init_pre_fid(struct osp_device *osp);
+int osp_init_statfs(struct osp_device *osp);
+void osp_fini_statfs(struct osp_device *osp);
+void osp_statfs_fini(struct osp_device *d);
 
 /* lproc_osp.c */
-void osp_lprocfs_init(struct osp_device *osp);
+void osp_tunables_init(struct osp_device *osp);
+void osp_tunables_fini(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);
+                        enum 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,
+                enum 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);
+void osp_sync_check_for_work(struct osp_device *osp);
+void osp_sync_force(const struct lu_env *env, struct osp_device *d);
+int osp_sync_add_commit_cb_1s(const struct lu_env *env, struct osp_device *d,
+                             struct thandle *th);
 
 /* lwp_dev.c */
-extern struct obd_ops lwp_obd_device_ops;
+extern const struct obd_ops lwp_obd_device_ops;
 extern struct lu_device_type lwp_device_type;
 
+static inline struct lu_device *osp2top(const struct osp_device *osp)
+{
+       return osp->opd_dt_dev.dd_lu_dev.ld_site->ls_top_dev;
+}
+
+static inline void osp_set_req_replay(const struct osp_device *osp,
+                                     struct ptlrpc_request *req)
+{
+       struct obd_device *obd = osp2top(osp)->ld_obd;
+
+       /* The RPC must be recovery related for the cases:
+        *
+        * 1. sent during recovery, or
+        * 2. sent before the recovery thread target_recovery_thread() start,
+        *    such as triggered by lod_sub_recovery_thread(). */
+       if (obd->obd_recovering || (obd->obd_replayable && obd->obd_no_conn))
+               req->rq_allow_replay = 1;
+}
+
 #endif