*
* 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
*/
* 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, 2016, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
*/
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 */
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;
+ int osp_pre_delorphan_sent;
+};
+
+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_device {
/* 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
*/
/* 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;
+ atomic_t opd_syn_changes;
/* processing of changes from previous mount is done? */
int opd_syn_prev_done;
/* found records */
struct ptlrpc_thread opd_syn_thread;
wait_queue_head_t opd_syn_waitq;
+ /* list of inflight rpc */
+ struct list_head opd_syn_inflight_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;
/* number of RPCs in flight - flow control */
- int opd_syn_rpc_in_flight;
+ atomic_t 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;
+ atomic_t 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;
+ __u64 opd_syn_last_used_id;
/* last committed change number -- semantically similar to
* last_committed */
- unsigned long opd_syn_last_committed_id;
+ __u64 opd_syn_last_committed_id;
/* last processed (taken from llog) id */
- unsigned long opd_syn_last_processed_id;
+ volatile __u64 opd_syn_last_processed_id;
struct osp_id_tracker *opd_syn_tracker;
struct list_head opd_syn_ontrack;
/* stop processing new requests until barrier=0 */
* 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_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;
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;
};
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 *
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
}
struct osp_txn_info {
- __u32 oti_current_id;
+ __u64 oti_current_id;
};
extern struct lu_context_key osp_txn_key;
unsigned int index);
int osp_extend_update_buffer(const struct lu_env *env,
- struct update_buffer *ubuf);
-
-#define osp_update_rpc_pack(env, name, update, op, ...) \
-({ \
- struct object_update *object_update; \
- size_t max_update_length; \
- struct object_update_request *ureq; \
- int ret; \
- \
+ 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, name, our, op, ...) \
+({ \
+ struct object_update *object_update; \
+ size_t max_update_length; \
+ struct osp_update_request_sub *ours; \
+ int ret; \
+ \
while (1) { \
- ureq = update->dur_buf.ub_req; \
- max_update_length = update->dur_buf.ub_req_size - \
- object_update_request_size(ureq); \
+ 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(ureq, \
- ureq->ourq_count); \
- ret = out_##name##_pack(env, object_update, max_update_length, \
+ object_update = update_buffer_get_update(ours->ours_req,\
+ ours->ours_req->ourq_count); \
+ ret = out_##name##_pack(env, object_update, \
+ &max_update_length, \
__VA_ARGS__); \
if (ret == -E2BIG) { \
int rc1; \
- /* extend the buffer and retry */ \
- rc1 = osp_extend_update_buffer(env, &update->dur_buf); \
+ /* 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) { \
- object_update->ou_flags |= update->dur_flags; \
- ureq->ourq_count++; \
+ 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,
/* 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);
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);
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);
-int osp_oac_init(struct osp_object *obj);
/* osp_md_object.c */
int osp_md_declare_object_create(const struct lu_env *env,
struct dt_object *dt,
extern 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