Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / include / obd.h
index 7ef0ee7..b42e2c3 100644 (file)
@@ -1,5 +1,37 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
+ *
+ * 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.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
 #ifndef __OBD_H
@@ -29,6 +61,7 @@
 
 #include <lustre/lustre_idl.h>
 #include <lu_object.h>
+#include <lu_ref.h>
 #include <lustre_lib.h>
 #include <lustre_export.h>
 #include <lustre_quota.h>
@@ -52,7 +85,7 @@ struct loi_oap_pages {
 struct osc_async_rc {
         int     ar_rc;
         int     ar_force_sync;
-        int     ar_min_xid;
+        __u64   ar_min_xid;
 };
 
 struct lov_oinfo {                 /* per-stripe data structure */
@@ -75,6 +108,12 @@ struct lov_oinfo {                 /* per-stripe data structure */
         struct osc_async_rc     loi_ar;
 };
 
+static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
+{
+        oinfo->loi_kms = kms;
+        oinfo->loi_kms_valid = 1;
+}
+
 static inline void loi_init(struct lov_oinfo *loi)
 {
         CFS_INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
@@ -105,7 +144,7 @@ struct lov_array_info {
 
 struct lov_stripe_md {
         spinlock_t       lsm_lock;
-        void            *lsm_lock_owner; /* debugging */
+        pid_t            lsm_lock_owner; /* debugging */
 
         struct {
                 /* Public members. */
@@ -118,6 +157,7 @@ struct lov_stripe_md {
                 __u32 lw_stripe_size;      /* size of the stripe */
                 __u32 lw_pattern;          /* striping pattern (RAID0, RAID1) */
                 unsigned lw_stripe_count;  /* number of objects being striped over */
+                char  lw_pool_name[LOV_MAXPOOLNAME]; /* pool name */
         } lsm_wire;
 
         struct lov_array_info *lsm_array; /*Only for joined file array info*/
@@ -131,6 +171,7 @@ struct lov_stripe_md {
 #define lsm_stripe_size  lsm_wire.lw_stripe_size
 #define lsm_pattern      lsm_wire.lw_pattern
 #define lsm_stripe_count lsm_wire.lw_stripe_count
+#define lsm_pool_name    lsm_wire.lw_pool_name
 
 struct obd_info;
 
@@ -142,8 +183,11 @@ struct obd_info {
          * OSC. (e.g. lov_prep_enqueue_set initialises extent of the policy,
          * and osc_enqueue passes it into ldlm_lock_match & ldlm_cli_enqueue. */
         ldlm_policy_data_t      oi_policy;
-        /* Flags used while lock handling. The flags obtained on the enqueue
-         * request are set here, therefore they are request specific. */
+        /* Flags used for set request specific flags:
+           - while lock handling, the flags obtained on the enqueue
+           request are set here.
+           - while stats, the flags used for control delay/resend.
+         */
         int                     oi_flags;
         /* Lock handle specific for every OSC lock. */
         struct lustre_handle   *oi_lockh;
@@ -252,31 +296,30 @@ struct obd_device_target {
         struct lustre_quota_ctxt  obt_qctxt;
 };
 
+typedef void (*obd_pin_extent_cb)(void *data);
+typedef int (*obd_page_removal_cb_t)(void *data, int discard);
+typedef int (*obd_lock_cancel_cb)(struct ldlm_lock *,struct ldlm_lock_desc *,
+                                   void *, int);
+
 /* llog contexts */
 enum llog_ctxt_id {
         LLOG_CONFIG_ORIG_CTXT  =  0,
-        LLOG_CONFIG_REPL_CTXT  =  1,
-        LLOG_MDS_OST_ORIG_CTXT =  2,
-        LLOG_MDS_OST_REPL_CTXT =  3,
-        LLOG_SIZE_ORIG_CTXT    =  4,
-        LLOG_SIZE_REPL_CTXT    =  5,
-        LLOG_MD_ORIG_CTXT      =  6,
-        LLOG_MD_REPL_CTXT      =  7,
-        LLOG_RD1_ORIG_CTXT     =  8,
-        LLOG_RD1_REPL_CTXT     =  9,
-        LLOG_TEST_ORIG_CTXT    = 10,
-        LLOG_TEST_REPL_CTXT    = 11,
-        LLOG_LOVEA_ORIG_CTXT   = 12,
-        LLOG_LOVEA_REPL_CTXT   = 13,
+        LLOG_CONFIG_REPL_CTXT,
+        LLOG_MDS_OST_ORIG_CTXT,
+        LLOG_MDS_OST_REPL_CTXT,
+        LLOG_SIZE_ORIG_CTXT,
+        LLOG_SIZE_REPL_CTXT,
+        LLOG_RD1_ORIG_CTXT,
+        LLOG_RD1_REPL_CTXT,
+        LLOG_TEST_ORIG_CTXT,
+        LLOG_TEST_REPL_CTXT,
+        LLOG_LOVEA_ORIG_CTXT,
+        LLOG_LOVEA_REPL_CTXT,
         LLOG_MAX_CTXTS
 };
 
 #define FILTER_SUBDIR_COUNT      32            /* set to zero for no subdirs */
 
-#define FILTER_GROUP_LLOG 1
-#define FILTER_GROUP_ECHO 2
-#define FILTER_GROUP_MDS0 3
-
 struct filter_subdirs {
        cfs_dentry_t *dentry[FILTER_SUBDIR_COUNT];
 };
@@ -287,17 +330,6 @@ struct filter_ext {
         __u64                fe_end;
 };
 
-struct obd_llogs {
-        struct llog_ctxt        *llog_ctxt[LLOG_MAX_CTXTS];
-};
-
-struct filter_group_llog {
-        struct list_head list;
-        int group;
-        struct obd_llogs *llogs;
-        struct obd_export *exp;
-};
-
 struct filter_obd {
         /* NB this field MUST be first */
         struct obd_device_target fo_obt;
@@ -331,6 +363,8 @@ struct filter_obd {
         obd_size             fo_tot_pending;
 
         obd_size             fo_readcache_max_filesize;
+        int                  fo_read_cache;
+        int                  fo_writethrough_cache;
 
         struct obd_import   *fo_mdc_imp;
         struct obd_uuid      fo_mdc_uuid;
@@ -378,6 +412,7 @@ struct filter_obd {
         unsigned int             fo_fl_oss_capa;
         struct list_head         fo_capa_keys;
         struct hlist_head       *fo_capa_hash;
+        struct llog_commit_master *fo_lcm;
 };
 
 #define OSC_MAX_RIF_DEFAULT       8
@@ -391,8 +426,9 @@ struct filter_obd {
 
 struct mdc_rpc_lock;
 struct obd_import;
+struct lustre_cache;
 struct client_obd {
-        struct semaphore         cl_sem;
+        struct rw_semaphore      cl_sem;
         struct obd_uuid          cl_target_uuid;
         struct obd_import       *cl_import; /* ptlrpc connection state */
         int                      cl_conn_count;
@@ -468,8 +504,12 @@ struct client_obd {
         atomic_t                 cl_mgc_refcount;
         struct obd_export       *cl_mgc_mgsexp;
 
-        /* Flags section */
-        unsigned long            cl_checksum:1; /* debug checksums */
+        /* checksumming for data sent over the network */
+        unsigned int             cl_checksum:1; /* 0 = disabled, 1 = enabled */
+        /* supported checksum types that are worked out at connect time */
+        __u32                    cl_supp_cksum_types;
+        /* checksum algorithm to be used */
+        cksum_type_t             cl_cksum_type;
 
         /* also protected by the poorly named _loi_list_lock lock above */
         struct osc_async_rc      cl_ar;
@@ -481,6 +521,10 @@ struct client_obd {
         struct lu_client_seq    *cl_seq;
 
         atomic_t                 cl_resends; /* resend count */
+
+        /* Cache of triples */
+        struct lustre_cache     *cl_cache;
+        obd_lock_cancel_cb       cl_ext_lock_cancel_cb;
 };
 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
 
@@ -520,7 +564,6 @@ struct mds_obd {
         cfs_dentry_t                    *mds_logs_dir;
         cfs_dentry_t                    *mds_objects_dir;
         struct llog_handle              *mds_cfg_llh;
-//        struct llog_handle              *mds_catalog;
         struct obd_device               *mds_osc_obd; /* XXX lov_obd */
         struct obd_uuid                  mds_lov_uuid;
         char                            *mds_profile;
@@ -555,6 +598,7 @@ struct mds_obd {
 
         /* for capability keys update */
         struct lustre_capa_key          *mds_capa_keys;
+        struct rw_semaphore             mds_notify_lock;
 };
 
 /* lov objid */
@@ -616,15 +660,32 @@ struct ltd_qos {
         unsigned int        ltq_usable:1;    /* usable for striping */
 };
 
+/* Generic subset of OSTs */
+struct ost_pool {
+        __u32              *op_array;        /* array of index of
+                                                lov_obd->lov_tgts */
+        unsigned int        op_count;        /* number of OSTs in the array */
+        unsigned int        op_size;         /* allocated size of lp_array */
+        rwlock_t            op_rwlock;       /* to protect lov_pool use */
+};
+
+/* Round-robin allocator data */
+struct lov_qos_rr {
+        __u32               lqr_start_idx;   /* start index of new inode */
+        __u32               lqr_offset_idx;  /* aliasing for start_idx  */
+        int                 lqr_start_count; /* reseed counter */
+        struct ost_pool     lqr_pool;        /* round-robin optimized list */
+        unsigned long       lqr_dirty:1;     /* recalc round-robin list */
+};
+
+/* Stripe placement optimization */
 struct lov_qos {
         struct list_head    lq_oss_list;    /* list of OSSs that targets use */
         struct rw_semaphore lq_rw_sem;
         __u32               lq_active_oss_count;
-        __u32              *lq_rr_array;    /* round-robin optimized list */
-        unsigned int        lq_rr_size;     /* rr array size */
         unsigned int        lq_prio_free;   /* priority for free space */
+        struct lov_qos_rr   lq_rr;          /* round robin qos data */
         unsigned long       lq_dirty:1,     /* recalc qos data */
-                            lq_dirty_rr:1,  /* recalc round-robin list */
                             lq_same_space:1,/* the ost's all have approx.
                                                the same space avail */
                             lq_reset:1;     /* zero current penalties */
@@ -641,9 +702,29 @@ struct lov_tgt_desc {
                             ltd_reap:1;  /* should this target be deleted */
 };
 
+/* Pool metadata */
+#define pool_tgt_size(_p)   _p->pool_obds.op_size
+#define pool_tgt_count(_p)  _p->pool_obds.op_count
+#define pool_tgt_array(_p)  _p->pool_obds.op_array
+#define pool_tgt_rwlock(_p) _p->pool_obds.op_rwlock
+#define pool_tgt(_p, _i)    _p->pool_lov->lov_tgts[_p->pool_obds.op_array[_i]]
+
+struct pool_desc {
+        char                  pool_name[LOV_MAXPOOLNAME + 1]; /* name of pool */
+        struct ost_pool       pool_obds;              /* pool members */
+        struct lov_qos_rr     pool_rr;                /* round robin qos */
+        struct hlist_node     pool_hash;              /* access by poolname */
+        struct list_head      pool_list;              /* serial access */
+        cfs_proc_dir_entry_t *pool_proc_entry;        /* file in /proc */
+        struct lov_obd       *pool_lov;               /* lov obd to which this
+                                                         pool belong */
+};
+
 struct lov_obd {
         struct lov_desc         desc;
-        struct lov_tgt_desc   **lov_tgts;
+        struct lov_tgt_desc   **lov_tgts;              /* sparse array */
+        struct ost_pool         lov_packed;            /* all OSTs in a packed
+                                                          array */
         struct semaphore        lov_lock;
         struct obd_connect_data lov_ocd;
         struct lov_qos          lov_qos;               /* qos info per lov */
@@ -652,10 +733,14 @@ struct lov_obd {
         __u32                   lov_active_tgt_count;  /* how many active */
         __u32                   lov_death_row;/* tgts scheduled to be deleted */
         __u32                   lov_tgt_size;   /* size of tgts array */
-        __u32                   lov_start_idx;  /* start index of new inode */
-        __u32                   lov_offset_idx; /* aliasing for start_idx  */
-        int                     lov_start_count;/* reseed counter */
         int                     lov_connects;
+        obd_page_removal_cb_t   lov_page_removal_cb;
+        obd_pin_extent_cb       lov_page_pin_cb;
+        obd_lock_cancel_cb      lov_lock_cancel_cb;
+        int                     lov_pool_count;
+        lustre_hash_t          *lov_pools_hash_body; /* used for key access */
+        struct list_head        lov_pool_list; /* used for sequential access */
+        cfs_proc_dir_entry_t   *lov_pool_proc_entry;
 };
 
 struct lmv_tgt_desc {
@@ -666,10 +751,20 @@ struct lmv_tgt_desc {
         struct semaphore        ltd_fid_sem;
 };
 
+enum placement_policy {
+        PLACEMENT_CHAR_POLICY   = 0,
+        PLACEMENT_NID_POLICY    = 1,
+        PLACEMENT_INVAL_POLICY  = 2,
+        PLACEMENT_MAX_POLICY
+};
+
+typedef enum placement_policy placement_policy_t;
+
 struct lmv_obd {
         int                     refcount;
         struct lu_client_fld    lmv_fld;
         spinlock_t              lmv_lock;
+        placement_policy_t      lmv_placement;
         struct lmv_desc         desc;
         struct obd_uuid         cluuid;
         struct obd_export       *exp;
@@ -750,6 +845,8 @@ struct obd_trans_info {
         /* initial thread handling transaction */
         int                      oti_thread_id;
         __u32                    oti_conn_cnt;
+
+        struct obd_uuid         *oti_ost_uuid;
 };
 
 static inline void oti_init(struct obd_trans_info *oti,
@@ -757,17 +854,18 @@ static inline void oti_init(struct obd_trans_info *oti,
 {
         if (oti == NULL)
                 return;
-        memset(oti, 0, sizeof *oti);
+        memset(oti, 0, sizeof(*oti));
 
         if (req == NULL)
                 return;
 
         oti->oti_xid = req->rq_xid;
 
-        if (req->rq_repmsg && req->rq_reqmsg != 0)
+        if (req->rq_repmsg != NULL)
                 oti->oti_transno = lustre_msg_get_transno(req->rq_repmsg);
         oti->oti_thread_id = req->rq_svc_thread ? req->rq_svc_thread->t_id : -1;
-        oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
+        if (req->rq_reqmsg != NULL)
+                oti->oti_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
 }
 
 static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
@@ -841,6 +939,23 @@ struct target_recovery_data {
         struct completion trd_finishing;
 };
 
+enum filter_groups {
+        FILTER_GROUP_LLOG = 1,
+        FILTER_GROUP_ECHO,
+        FILTER_GROUP_MDS0
+};
+
+struct obd_llog_group {
+        struct list_head   olg_list;
+        int                olg_group;
+        struct llog_ctxt  *olg_ctxts[LLOG_MAX_CTXTS];
+        cfs_waitq_t        olg_waitq;
+        spinlock_t         olg_lock;
+        struct obd_export *olg_exp;
+        int                olg_initializing;
+        struct semaphore   olg_cat_processing;
+};
+
 /* corresponds to one of the obd's */
 #define MAX_OBD_NAME 128
 #define OBD_DEVICE_MAGIC        0XAB5CD6EF
@@ -872,13 +987,17 @@ struct obd_device {
                       obd_inactive:1;      /* device active/inactive
                                            * (for /proc/status only!!) */
         /* uuid-export hash body */
-        struct lustre_class_hash_body *obd_uuid_hash_body;
+        struct lustre_hash     *obd_uuid_hash;
         /* nid-export hash body */
-        struct lustre_class_hash_body *obd_nid_hash_body; 
-        atomic_t obd_refcount;
+        struct lustre_hash     *obd_nid_hash;
+        /* nid stats body */
+        struct lustre_hash     *obd_nid_stats_hash;
+        struct list_head        obd_nid_stats;
+        atomic_t                obd_refcount;
         cfs_waitq_t             obd_refcount_waitq;
         struct list_head        obd_exports;
         int                     obd_num_exports;
+        spinlock_t              obd_nid_lock;
         struct ldlm_namespace  *obd_namespace;
         struct ptlrpc_client    obd_ldlm_client; /* XXX OST/MDS only */
         /* a spinlock is OK for what we do now, may need a semaphore later */
@@ -890,7 +1009,7 @@ struct obd_device {
         struct obd_statfs       obd_osfs;       /* locked by obd_osfs_lock */
         __u64                   obd_osfs_age;
         struct lvfs_run_ctxt    obd_lvfs_ctxt;
-        struct llog_ctxt        *obd_llog_ctxt[LLOG_MAX_CTXTS];
+        struct obd_llog_group   obd_olg; /* default llog group */
         struct obd_device       *obd_observer;
         struct obd_notify_upcall obd_upcall;
         struct obd_export       *obd_self_export;
@@ -901,7 +1020,7 @@ struct obd_device {
         /* XXX encapsulate all this recovery data into one struct */
         svc_handler_t                    obd_recovery_handler;
         pid_t                            obd_processing_task;
-        
+
         int                              obd_max_recoverable_clients;
         int                              obd_connected_clients;
         int                              obd_recoverable_clients;
@@ -914,9 +1033,10 @@ struct obd_device {
         spinlock_t                       obd_uncommitted_replies_lock;
         cfs_timer_t                      obd_recovery_timer;
         time_t                           obd_recovery_start; /* seconds */
-        time_t                           obd_recovery_end; /* seconds */
+        time_t                           obd_recovery_end; /* seconds, for lprocfs_status */
         time_t                           obd_recovery_max_time; /* seconds, bz13079 */
-        
+        int                              obd_recovery_timeout;
+
         /* new recovery stuff from CMD2 */
         struct target_recovery_data      obd_recovery_data;
         int                              obd_replayed_locks;
@@ -947,12 +1067,24 @@ struct obd_device {
         struct lprocfs_stats  *md_stats;
 
         cfs_proc_dir_entry_t  *obd_proc_entry;
-        cfs_proc_dir_entry_t  *obd_proc_exports;
+        cfs_proc_dir_entry_t  *obd_proc_exports_entry;
         cfs_proc_dir_entry_t  *obd_svc_procroot;
         struct lprocfs_stats  *obd_svc_stats;
-        struct semaphore       obd_proc_exp_sem;
         atomic_t               obd_evict_inprogress;
         cfs_waitq_t            obd_evict_inprogress_waitq;
+
+        /**
+         * Ldlm pool part. Save last calculated SLV and Limit.
+         */
+        rwlock_t               obd_pool_lock;
+        int                    obd_pool_limit;
+        __u64                  obd_pool_slv;
+
+        /**
+         * A list of outstanding class_incref()'s against this obd. For
+         * debugging.
+         */
+        struct lu_ref          obd_reference;
 };
 
 #define OBD_OPT_FORCE           0x0001
@@ -963,15 +1095,8 @@ struct obd_device {
 enum obd_cleanup_stage {
 /* Special case hack for MDS LOVs */
         OBD_CLEANUP_EARLY,
-/* Precleanup stage 1, we must make sure all exports (other than the
-   self-export) get destroyed. */
+/* can be directly mapped to .ldto_device_fini() */
         OBD_CLEANUP_EXPORTS,
-/* Precleanup stage 2,  do other type-specific cleanup requiring the
-   self-export. */
-        OBD_CLEANUP_SELF_EXP,
-/* FIXME we should eliminate the "precleanup" function and make them stages
-   of the "cleanup" function. */
-        OBD_CLEANUP_OBD,
 };
 
 /* get/set_info keys */
@@ -988,9 +1113,35 @@ enum obd_cleanup_stage {
 #define KEY_REVIMP_UPD          "revimp_update"
 #define KEY_LOV_IDX             "lov_idx"
 #define KEY_LAST_ID             "last_id"
+#define KEY_READONLY            "read-only"
+#define KEY_LOCK_TO_STRIPE      "lock_to_stripe"
+#define KEY_CHECKSUM            "checksum"
+#define KEY_UNLINKED            "unlinked"
+#define KEY_EVICT_BY_NID        "evict_by_nid"
+#define KEY_REGISTER_TARGET     "register_target"
+#define KEY_SET_FS              "set_fs"
+#define KEY_CLEAR_FS            "clear_fs"
+#define KEY_BLOCKSIZE           "blocksize"
+#define KEY_BLOCKSIZE_BITS      "blocksize_bits"
+#define KEY_FIEMAP              "FIEMAP"
+/* XXX unused ?*/
+#define KEY_INTERMDS            "inter_mds"
+#define KEY_ASYNC               "async"
 
 struct lu_context;
 
+static inline int it_to_lock_mode(struct lookup_intent *it)
+{
+        /* CREAT needs to be tested before open (both could be set) */
+        if (it->it_op & IT_CREAT)
+                return LCK_CW;
+        else if (it->it_op & (IT_READDIR | IT_GETATTR | IT_OPEN | IT_LOOKUP))
+                return LCK_CR;
+
+        LASSERTF(0, "Invalid it_op: %d\n", it->it_op);
+        return -EINVAL;
+}
+
 struct md_op_data {
         struct lu_fid           op_fid1; /* operation fid1 (usualy parent) */
         struct lu_fid           op_fid2; /* operation fid2 (usualy child) */
@@ -1007,7 +1158,7 @@ struct md_op_data {
         __u32                   op_suppgids[2];
         __u32                   op_fsuid;
         __u32                   op_fsgid;
-        __u32                   op_cap;
+        cfs_cap_t               op_cap;
         void                   *op_data;
 
         /* iattr fields and blocks. */
@@ -1034,12 +1185,28 @@ struct md_op_data {
         __u32                   op_opc;
 };
 
+struct md_enqueue_info;
+/* metadata stat-ahead */
+typedef int (* md_enqueue_cb_t)(struct ptlrpc_request *req,
+                                struct md_enqueue_info *minfo,
+                                int rc);
+
+struct md_enqueue_info {
+        struct md_op_data       mi_data;
+        struct lookup_intent    mi_it;
+        struct lustre_handle    mi_lockh;
+        struct dentry          *mi_dentry;
+        md_enqueue_cb_t         mi_cb;
+        unsigned int            mi_generation;
+        void                   *mi_cbdata;
+};
+
 struct obd_ops {
         struct module *o_owner;
         int (*o_iocontrol)(unsigned int cmd, struct obd_export *exp, int len,
                            void *karg, void *uarg);
         int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
-                          __u32 *vallen, void *val);
+                          __u32 *vallen, void *val, struct lov_stripe_md *lsm);
         int (*o_set_info_async)(struct obd_export *, __u32 keylen, void *key,
                                 __u32 vallen, void *val,
                                 struct ptlrpc_request_set *set);
@@ -1061,11 +1228,13 @@ struct obd_ops {
          * asked for. If @ocd == NULL, use default parameters. */
         int (*o_connect)(const struct lu_env *env,
                          struct lustre_handle *conn, struct obd_device *src,
-                         struct obd_uuid *cluuid, struct obd_connect_data *ocd);
+                         struct obd_uuid *cluuid, struct obd_connect_data *ocd,
+                         void *localdata);
         int (*o_reconnect)(const struct lu_env *env,
                            struct obd_export *exp, struct obd_device *src,
                            struct obd_uuid *cluuid,
-                           struct obd_connect_data *ocd);
+                           struct obd_connect_data *ocd,
+                           void *localdata);
         int (*o_disconnect)(struct obd_export *exp);
 
         /* Initialize/finalize fids infrastructure. */
@@ -1076,14 +1245,14 @@ struct obd_ops {
         int (*o_fid_alloc)(struct obd_export *exp, struct lu_fid *fid,
                            struct md_op_data *op_data);
 
-        /* 
+        /*
          * Object with @fid is getting deleted, we may want to do something
          * about this.
          */
         int (*o_fid_delete)(struct obd_export *exp, const struct lu_fid *fid);
 
         int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
-                        __u64 max_age);
+                        __u64 max_age, __u32 flags);
         int (*o_statfs_async)(struct obd_device *obd, struct obd_info *oinfo,
                               __u64 max_age, struct ptlrpc_request_set *set);
         int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
@@ -1121,7 +1290,16 @@ struct obd_ops {
                                  struct lov_oinfo *loi,
                                  cfs_page_t *page, obd_off offset,
                                  struct obd_async_page_ops *ops, void *data,
-                                 void **res);
+                                 void **res, int nocache,
+                                 struct lustre_handle *lockh);
+        int (*o_reget_short_lock)(struct obd_export *exp,
+                                  struct lov_stripe_md *lsm,
+                                  void **res, int rw,
+                                  obd_off start, obd_off end,
+                                  void **cookie);
+        int (*o_release_short_lock)(struct obd_export *exp,
+                                    struct lov_stripe_md *lsm, obd_off end,
+                                    void *cookie, int rw);
         int (*o_queue_async_io)(struct obd_export *exp,
                                 struct lov_stripe_md *lsm,
                                 struct lov_oinfo *loi, void *cookie,
@@ -1165,12 +1343,14 @@ struct obd_ops {
                          obd_id *startid, obd_gr group, void *data);
         int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
                         int objcount, struct obd_ioobj *obj,
-                        int niocount, struct niobuf_remote *remote,
-                        struct niobuf_local *local, struct obd_trans_info *oti,
+                        struct niobuf_remote *remote, int *nr_pages,
+                        struct niobuf_local *local,
+                        struct obd_trans_info *oti,
                         struct lustre_capa *capa);
         int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
                           int objcount, struct obd_ioobj *obj,
-                          int niocount, struct niobuf_local *local,
+                          struct niobuf_remote *remote, int pages,
+                          struct niobuf_local *local,
                           struct obd_trans_info *oti, int rc);
         int (*o_enqueue)(struct obd_export *, struct obd_info *oinfo,
                          struct ldlm_enqueue_info *einfo,
@@ -1184,15 +1364,13 @@ struct obd_ops {
                         __u32 mode, struct lustre_handle *);
         int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
                                int flags, void *opaque);
-        int (*o_join_lru)(struct obd_export *, struct lov_stripe_md *,
-                         int join);
         int (*o_init_export)(struct obd_export *exp);
         int (*o_destroy_export)(struct obd_export *exp);
         int (*o_extent_calc)(struct obd_export *, struct lov_stripe_md *,
                              int cmd, obd_off *);
 
         /* llog related obd_methods */
-        int (*o_llog_init)(struct obd_device *obd, struct obd_llogs *llog,
+        int (*o_llog_init)(struct obd_device *obd, struct obd_llog_group *grp,
                            struct obd_device *disk_obd, int count,
                            struct llog_catid *logid, struct obd_uuid *uuid);
         int (*o_llog_finish)(struct obd_device *obd, int count);
@@ -1210,12 +1388,30 @@ struct obd_ops {
                         enum obd_notify_event ev, void *data);
 
         int (*o_health_check)(struct obd_device *);
+        struct obd_uuid *(*o_get_uuid) (struct obd_export *exp);
 
         /* quota methods */
         int (*o_quotacheck)(struct obd_export *, struct obd_quotactl *);
         int (*o_quotactl)(struct obd_export *, struct obd_quotactl *);
 
         int (*o_ping)(struct obd_export *exp);
+
+        int (*o_register_page_removal_cb)(struct obd_export *exp,
+                                          obd_page_removal_cb_t cb,
+                                          obd_pin_extent_cb pin_cb);
+        int (*o_unregister_page_removal_cb)(struct obd_export *exp,
+                                            obd_page_removal_cb_t cb);
+        int (*o_register_lock_cancel_cb)(struct obd_export *exp,
+                                       obd_lock_cancel_cb cb);
+        int (*o_unregister_lock_cancel_cb)(struct obd_export *exp,
+                                         obd_lock_cancel_cb cb);
+        /* pools methods */
+        int (*o_pool_new)(struct obd_device *obd, char *poolname);
+        int (*o_pool_del)(struct obd_device *obd, char *poolname);
+        int (*o_pool_add)(struct obd_device *obd, char *poolname,
+                          char *ostname);
+        int (*o_pool_rem)(struct obd_device *obd, char *poolname,
+                          char *ostname);
         /*
          * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
          * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
@@ -1244,8 +1440,9 @@ enum {
 #define MEA_MAGIC_ALL_CHARS      0xb222a11c
 #define MEA_MAGIC_HASH_SEGMENT   0xb222a11b
 
-#define MAX_HASH_SIZE            0x7fffffffUL
-#define MAX_HASH_HIGHEST_BIT     0x10000000
+#define MAX_HASH_SIZE_32         0x7fffffffUL
+#define MAX_HASH_SIZE            0x7fffffffffffffffULL
+#define MAX_HASH_HIGHEST_BIT     0x1000000000000000ULL
 
 struct lustre_md {
         struct mdt_body         *body;
@@ -1274,13 +1471,14 @@ struct md_ops {
         int (*m_close)(struct obd_export *, struct md_op_data *,
                        struct md_open_data *, struct ptlrpc_request **);
         int (*m_create)(struct obd_export *, struct md_op_data *,
-                        const void *, int, int, __u32, __u32, __u32,
+                        const void *, int, int, __u32, __u32, cfs_cap_t,
                         __u64, struct ptlrpc_request **);
         int (*m_done_writing)(struct obd_export *, struct md_op_data  *,
                               struct md_open_data *);
         int (*m_enqueue)(struct obd_export *, struct ldlm_enqueue_info *,
                          struct lookup_intent *, struct md_op_data *,
-                         struct lustre_handle *, void *, int, int);
+                         struct lustre_handle *, void *, int,
+                         struct ptlrpc_request **, int);
         int (*m_getattr)(struct obd_export *, const struct lu_fid *,
                          struct obd_capa *, obd_valid, int,
                          struct ptlrpc_request **);
@@ -1324,7 +1522,7 @@ struct md_ops {
         int (*m_init_ea_size)(struct obd_export *, int, int, int);
 
         int (*m_get_lustre_md)(struct obd_export *, struct ptlrpc_request *,
-                               int, struct obd_export *, struct obd_export *,
+                               struct obd_export *, struct obd_export *,
                                struct lustre_md *);
 
         int (*m_free_lustre_md)(struct obd_export *, struct lustre_md *);
@@ -1351,6 +1549,14 @@ struct md_ops {
                                  struct obd_capa *, __u32,
                                  struct ptlrpc_request **);
 
+        int (*m_intent_getattr_async)(struct obd_export *,
+                                      struct md_enqueue_info *,
+                                      struct ldlm_enqueue_info *);
+
+        int (*m_revalidate_lock)(struct obd_export *,
+                                 struct lookup_intent *,
+                                 struct lu_fid *);
+
         /*
          * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
          * lprocfs_alloc_md_stats() in obdclass/lprocfs_status.c. Also, add a
@@ -1376,15 +1582,18 @@ struct lsm_operations {
                              struct lov_mds_md *lmm);
 };
 
-extern struct lsm_operations lsm_plain_ops;
-extern struct lsm_operations lsm_join_ops;
-static inline struct lsm_operations *lsm_op_find(int magic)
+extern const struct lsm_operations lsm_v1_ops;
+extern const struct lsm_operations lsm_join_ops;
+extern const struct lsm_operations lsm_v3_ops;
+static inline const struct lsm_operations *lsm_op_find(int magic)
 {
         switch(magic) {
-        case LOV_MAGIC:
-               return &lsm_plain_ops;
+        case LOV_MAGIC_V1:
+               return &lsm_v1_ops;
         case LOV_MAGIC_JOIN:
                return &lsm_join_ops;
+        case LOV_MAGIC_V3:
+               return &lsm_v3_ops;
         default:
                CERROR("Cannot recognize lsm_magic %d\n", magic);
                return NULL;
@@ -1401,15 +1610,18 @@ static inline void obd_transno_commit_cb(struct obd_device *obd, __u64 transno,
                                          int error)
 {
         if (error) {
-                CERROR("%s: transno "LPD64" commit error: %d\n",
+                CERROR("%s: transno "LPU64" commit error: %d\n",
                        obd->obd_name, transno, error);
                 return;
         }
-        CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
-               obd->obd_name, transno);
         if (transno > obd->obd_last_committed) {
+                CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
+                       obd->obd_name, transno);
                 obd->obd_last_committed = transno;
                 ptlrpc_commit_replies (obd);
+        } else {
+                CDEBUG(D_INFO, "%s: transno "LPD64" committed\n",
+                       obd->obd_name, transno);
         }
 }