* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2012, 2016, 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
/*
* Precreation pool
*/
- spinlock_t osp_pre_lock;
/* last fid to assign in creation */
struct lu_fid osp_pre_used_fid;
__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 */
* 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,
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;
atomic_t opd_sync_barrier;
wait_queue_head_t opd_sync_barrier_waitq;
/* last generated id */
- cfs_time_t opd_sync_next_commit_cb;
+ 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
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
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;
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;
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);
+
+ 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);
- mdc_get_rpc_lock(rpc_lock, NULL);
+ 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(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;
+
+ if (lck->rpcl_fakes) { /* OBD_FAIL_OSP_RPCS_SEM */
+ mutex_lock(&lck->rpcl_mutex);
- mdc_put_rpc_lock(rpc_lock, NULL);
+ 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,
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
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);
}
}
int osp_object_update_request_create(struct osp_update_request *our,
size_t size);
-#define osp_update_rpc_pack(env, name, our, op, ...) \
+#define OSP_UPDATE_RPC_PACK(env, out_something_pack, our, ...) \
({ \
- struct object_update *object_update; \
- size_t max_update_length; \
+ struct object_update *object_update; \
+ size_t max_update_length; \
struct osp_update_request_sub *ours; \
int ret; \
\
\
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__); \
+ ret = out_something_pack(env, object_update, \
+ &max_update_length, \
+ __VA_ARGS__); \
if (ret == -E2BIG) { \
int rc1; \
/* Create new object update request */ \
/* 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,
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 *osp);
void osp_sync_force(const struct lu_env *env, struct osp_device *d);
-int osp_sync_add_commit_cb(const struct lu_env *env, struct osp_device *d,
- struct thandle *th);
int osp_sync_add_commit_cb_1s(const struct lu_env *env, struct osp_device *d,
struct thandle *th);