*
* 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, 2013, Intel Corporation.
+ * Copyright (c) 2012, 2016, 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 */
- cfs_list_t otr_wakeup_list;
- cfs_list_t otr_list;
+ struct list_head otr_wakeup_list;
+ struct list_head otr_list;
/* underlying shared device */
struct dt_device *otr_dev;
/* how many users of this tracker */
- cfs_atomic_t otr_refcount;
+ atomic_t otr_refcount;
};
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;
+ 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 {
struct obd_uuid opd_cluuid;
struct obd_connect_data *opd_connect_data;
int opd_connects;
- cfs_proc_dir_entry_t *opd_proc_entry;
+ struct proc_dir_entry *opd_proc_entry;
struct lprocfs_stats *opd_stats;
/* connection status. */
unsigned int opd_new_connection:1,
/* 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 */
- cfs_list_t opd_syn_committed_there;
+ 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;
- cfs_list_t opd_syn_ontrack;
+ struct list_head opd_syn_ontrack;
+ /* stop processing new requests until barrier=0 */
+ atomic_t opd_syn_barrier;
+ wait_queue_head_t opd_syn_barrier_waitq;
/*
* statfs related fields: OSP maintains it on its own
/* how often to update statfs data */
int opd_statfs_maxage;
- cfs_proc_dir_entry_t *opd_symlink;
+ struct proc_dir_entry *opd_symlink;
/* 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 semaphore opd_async_fc_sem;
+ 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;
struct list_head oxe_list;
atomic_t oxe_ref;
void *oxe_value;
- int oxe_buflen;
- int oxe_namelen;
- int oxe_vallen;
+ size_t oxe_buflen;
+ size_t oxe_namelen;
+ size_t oxe_vallen;
unsigned int oxe_exist:1,
oxe_ready:1;
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_new:1,
- opo_empty: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 lu_attr osi_attr;
struct ost_id osi_oi;
struct ost_id osi_oi2;
- obd_id osi_id;
+ u64 osi_id;
loff_t osi_off;
union {
struct llog_rec_hdr osi_hdr;
struct obdo osi_obdo;
};
+/* Iterator for OSP */
+struct osp_it {
+ __u32 ooi_pos_page;
+ __u32 ooi_pos_lu_page;
+ __u32 ooi_attr;
+ int ooi_pos_ent;
+ int ooi_total_npages;
+ int ooi_valid_npages;
+ unsigned int ooi_swab:1;
+ __u64 ooi_next;
+ struct dt_object *ooi_obj;
+ void *ooi_ent;
+ struct page *ooi_cur_page;
+ struct lu_idxpage *ooi_cur_idxpage;
+ struct page **ooi_pages;
+};
+
+#define OSP_THANDLE_MAGIC 0x20141214
+struct osp_thandle {
+ struct thandle ot_super;
+
+ /* 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 *
+thandle_to_osp_thandle(struct thandle *th)
+{
+ return container_of(th, struct osp_thandle, ot_super);
+}
+
+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_our;
+}
+
/* The transaction only include the updates on the remote node, and
* no local updates at all */
static inline bool is_only_remote_trans(struct thandle *th)
{
- return th->th_dev != NULL && th->th_dev->dd_ops == &osp_dt_ops;
+ return th->th_top == NULL;
}
static inline void osp_objid_buf_prep(struct lu_buf *buf, loff_t *off,
__u32 *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. */
buf->lb_buf = (void *)id;
- buf->lb_len = sizeof(obd_id);
- *off = sizeof(obd_id) * index;
+ buf->lb_len = sizeof(u64);
+ *off = sizeof(u64) * index;
}
static inline void osp_objseq_buf_prep(struct lu_buf *buf, loff_t *off,
__u64 *seq, int index)
{
buf->lb_buf = (void *)seq;
- buf->lb_len = sizeof(obd_id);
- *off = sizeof(obd_id) * index;
+ buf->lb_len = sizeof(u64);
+ *off = sizeof(u64) * index;
}
static inline void osp_buf_prep(struct lu_buf *lb, void *buf, int buf_len)
}
struct osp_txn_info {
- __u32 oti_current_id;
+ __u64 oti_current_id;
};
extern struct lu_context_key osp_txn_key;
}
#define osp_init_rpc_lock(lck) mdc_init_rpc_lock(lck)
-#define osp_get_rpc_lock(lck, it) mdc_get_rpc_lock(lck, it)
-#define osp_put_rpc_lock(lck, it) mdc_put_rpc_lock(lck, it)
+
+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;
+
+ mdc_get_rpc_lock(rpc_lock, NULL);
+}
+
+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;
+
+ mdc_put_rpc_lock(rpc_lock, NULL);
+}
+
+static inline int osp_fid_diff(const struct lu_fid *fid1,
+ const struct lu_fid *fid2)
+{
+ /* In 2.6+ ost_idx is packed into IDIF FID, while in 2.4 and 2.5 IDIF
+ * is always FID_SEQ_IDIF(0x100000000ULL), which does not include OST
+ * index in the seq. So we can not compare IDIF FID seq here */
+ if (fid_is_idif(fid1) && fid_is_idif(fid2)) {
+ __u32 ost_idx1 = fid_idif_ost_idx(fid1);
+ __u32 ost_idx2 = fid_idif_ost_idx(fid2);
+
+ LASSERTF(ost_idx1 == 0 || ost_idx2 == 0 || ost_idx1 == ost_idx2,
+ "fid1: "DFID", fid2: "DFID"\n", PFID(fid1),
+ PFID(fid2));
+
+ return fid_idif_id(fid1->f_seq, fid1->f_oid, 0) -
+ 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));
+
+ return fid_oid(fid1) - fid_oid(fid2);
+}
+
static inline void osp_update_last_fid(struct osp_device *d, struct lu_fid *fid)
{
- int diff = lu_fid_diff(fid, &d->opd_last_used_fid);
+ int diff = osp_fid_diff(fid, &d->opd_last_used_fid);
/*
* we might have lost precreated objects due to VBR and precreate
* orphans, the gap in objid can be calculated properly only here
return imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_FID;
}
-typedef int (*osp_async_update_interpterer_t)(const struct lu_env *env,
- struct object_update_reply *reply,
- struct osp_object *obj,
- void *data, int index, int rc);
+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, name, our, op, ...) \
+({ \
+ 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_##name##_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,
+ struct osp_object *obj,
+ void *data, int index, int rc);
/* osp_dev.c */
-void osp_update_last_id(struct osp_device *d, obd_id objid);
+void osp_update_last_id(struct osp_device *d, u64 objid);
extern struct llog_operations osp_mds_ost_orig_logops;
/* osp_trans.c */
-struct dt_update_request *
-osp_find_or_create_async_update_request(struct osp_device *osp);
-int osp_insert_async_update(const struct lu_env *env,
- struct dt_update_request *update, int op,
- struct osp_object *obj, int count,
- int *lens, const char **bufs, void *data,
- osp_async_update_interpterer_t interpterer);
-int osp_unplug_async_update(const struct lu_env *env,
- struct osp_device *osp,
- struct dt_update_request *update);
+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, __u32 repsize,
+ osp_update_interpreter_t interpreter);
+
+int osp_unplug_async_request(const struct lu_env *env,
+ struct osp_device *osp,
+ 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 osp_update_request *update,
+ struct osp_object *obj, void *data,
+ osp_update_interpreter_t interpreter);
+
+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,
+ struct osp_update_request *our,
+ struct ptlrpc_request **reqp);
+int osp_remote_sync(const struct lu_env *env, struct osp_device *osp,
+ 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, struct lustre_capa *capa);
+ struct lu_attr *attr);
int osp_xattr_get(const struct lu_env *env, struct dt_object *dt,
- struct lu_buf *buf, const char *name,
- struct lustre_capa *capa);
+ struct lu_buf *buf, const char *name);
int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, const char *name,
int flag, struct thandle *th);
int osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, const char *name, int fl,
- struct thandle *th, struct lustre_capa *capa);
-int osp_declare_object_destroy(const struct lu_env *env,
- struct dt_object *dt, struct thandle *th);
-int osp_object_destroy(const struct lu_env *env, struct dt_object *dt,
- struct thandle *th);
+ struct thandle *th);
+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);
+void osp_it_fini(const struct lu_env *env, struct dt_it *di);
+int osp_it_get(const struct lu_env *env, struct dt_it *di,
+ const struct dt_key *key);
+void osp_it_put(const struct lu_env *env, struct dt_it *di);
+__u64 osp_it_store(const struct lu_env *env, const struct dt_it *di);
+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,
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,
- const struct lu_attr *attr, struct thandle *th,
- struct lustre_capa *capa);
+ const struct lu_attr *attr, struct thandle *th);
+extern const struct dt_index_operations osp_md_index_ops;
/* osp_precreate.c */
int osp_init_precreate(struct osp_device *d);
int osp_init_pre_fid(struct osp_device *osp);
/* lproc_osp.c */
-void lprocfs_osp_init_vars(struct lprocfs_static_vars *lvars);
void osp_lprocfs_init(struct osp_device *osp);
/* osp_sync.c */
void __osp_sync_check_for_work(struct osp_device *d);
/* lwp_dev.c */
-void lprocfs_lwp_init_vars(struct lprocfs_static_vars *lvars);
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