Whamcloud - gitweb
b=22176 Add .sync_fs super block handler
[fs/lustre-release.git] / lustre / include / obd.h
index 5ef03cc..46f8bb7 100644 (file)
@@ -26,7 +26,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  */
 /*
@@ -54,9 +54,6 @@
 
 #define IOC_MDC_TYPE         'i'
 #define IOC_MDC_MIN_NR       20
-/* Moved to lustre_user.h
-#define IOC_MDC_LOOKUP       _IOWR(IOC_MDC_TYPE, 20, struct obd_ioctl_data *)
-#define IOC_MDC_GETSTRIPE    _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
 #define IOC_MDC_MAX_NR       50
 
 #include <lustre/lustre_idl.h>
@@ -88,8 +85,7 @@ struct osc_async_rc {
 };
 
 struct lov_oinfo {                 /* per-stripe data structure */
-        __u64 loi_id;              /* object ID on the target OST */
-        __u64 loi_gr;              /* object group on the target OST */
+        struct ost_id   loi_oi;    /* object ID/Sequence on the target OST */
         int loi_ost_idx;           /* OST stripe index in lov_tgt_desc->tgts */
         int loi_ost_gen;           /* generation of this loi_ost_idx */
 
@@ -100,12 +96,14 @@ struct lov_oinfo {                 /* per-stripe data structure */
         cfs_list_t loi_hp_ready_item;
         cfs_list_t loi_write_item;
         cfs_list_t loi_read_item;
-
+        cfs_list_t loi_sync_fs_item;
         unsigned long loi_kms_valid:1;
         __u64 loi_kms;             /* known minimum size */
         struct ost_lvb loi_lvb;
         struct osc_async_rc     loi_ar;
 };
+#define loi_id  loi_oi.oi_id
+#define loi_seq loi_oi.oi_seq
 
 static inline void loi_kms_set(struct lov_oinfo *oinfo, __u64 kms)
 {
@@ -125,6 +123,7 @@ static inline void loi_init(struct lov_oinfo *loi)
         CFS_INIT_LIST_HEAD(&loi->loi_hp_ready_item);
         CFS_INIT_LIST_HEAD(&loi->loi_write_item);
         CFS_INIT_LIST_HEAD(&loi->loi_read_item);
+        CFS_INIT_LIST_HEAD(&loi->loi_sync_fs_item);
 }
 
 struct lov_stripe_md {
@@ -134,7 +133,7 @@ struct lov_stripe_md {
         struct {
                 /* Public members. */
                 __u64 lw_object_id;        /* lov object id */
-                __u64 lw_object_gr;        /* lov object group */
+                __u64 lw_object_seq;       /* lov object seq */
                 __u64 lw_maxbytes;         /* maximum possible file size */
 
                 /* LOV-private members start here -- only for use in lov/. */
@@ -149,7 +148,7 @@ struct lov_stripe_md {
 };
 
 #define lsm_object_id    lsm_wire.lw_object_id
-#define lsm_object_gr    lsm_wire.lw_object_gr
+#define lsm_object_seq   lsm_wire.lw_object_seq
 #define lsm_maxbytes     lsm_wire.lw_maxbytes
 #define lsm_magic        lsm_wire.lw_magic
 #define lsm_stripe_size  lsm_wire.lw_stripe_size
@@ -161,6 +160,12 @@ struct obd_info;
 
 typedef int (*obd_enqueue_update_f)(void *cookie, int rc);
 
+struct osc_sync_fs_wait {
+        struct obd_info      *sfw_oi;
+        obd_enqueue_update_f  sfw_upcall;
+        int started;
+};
+
 /* obd info for a particular level (lov, osc). */
 struct obd_info {
         /* Lock policy. It keeps an extent which is specific for a particular
@@ -227,8 +232,10 @@ struct brw_page {
 
 struct ost_server_data;
 
+#define OBT_MAGIC       0xBDDECEAE
 /* hold common fields for "target" device */
 struct obd_device_target {
+        __u32                     obt_magic;
         struct super_block       *obt_sb;
         /** last_rcvd file */
         struct file              *obt_rcvd_filp;
@@ -300,6 +307,7 @@ struct filter_obd {
         int                  fo_tot_granted_clients;
 
         obd_size             fo_readcache_max_filesize;
+        cfs_spinlock_t       fo_flags_lock;
         int                  fo_read_cache:1,   /**< enable read-only cache */
                              fo_writethrough_cache:1,/**< read cache writes */
                              fo_mds_ost_sync:1, /**< MDS-OST orphan recovery*/
@@ -342,6 +350,9 @@ struct filter_obd {
 
         int                      fo_fmd_max_num; /* per exp filter_mod_data */
         int                      fo_fmd_max_age; /* jiffies to fmd expiry */
+        unsigned long            fo_syncjournal:1, /* sync journal on writes */
+                                 fo_sync_lock_cancel:2;/* sync on lock cancel */
+
 
         /* sptlrpc stuff */
         cfs_rwlock_t             fo_sptlrpc_lock;
@@ -370,6 +381,14 @@ struct timeout_item {
 #define OSC_MAX_DIRTY_MB_MAX   2048     /* arbitrary, but < MAX_LONG bytes */
 #define OSC_DEFAULT_RESENDS      10
 
+/* possible values for fo_sync_lock_cancel */
+enum {
+        NEVER_SYNC_ON_CANCEL = 0,
+        BLOCKING_SYNC_ON_CANCEL = 1,
+        ALWAYS_SYNC_ON_CANCEL = 2,
+        NUM_SYNC_ON_CANCEL_STATES
+};
+
 #define MDC_MAX_RIF_DEFAULT       8
 #define MDC_MAX_RIF_MAX         512
 
@@ -428,6 +447,7 @@ struct client_obd {
         cfs_list_t               cl_loi_hp_ready_list;
         cfs_list_t               cl_loi_write_list;
         cfs_list_t               cl_loi_read_list;
+        cfs_list_t               cl_loi_sync_fs_list;
         int                      cl_r_in_flight;
         int                      cl_w_in_flight;
         /* just a sum of the loi/lop pending numbers to be exported by /proc */
@@ -475,12 +495,15 @@ struct client_obd {
         struct lu_client_seq    *cl_seq;
 
         cfs_atomic_t             cl_resends; /* resend count */
+        struct osc_sync_fs_wait  cl_sf_wait;
 };
+
 #define obd2cli_tgt(obd) ((char *)(obd)->u.cli.cl_target_uuid.uuid)
 
 #define CL_NOT_QUOTACHECKED 1   /* client->cl_qchk_stat init value */
 
 struct mgs_obd {
+        struct obd_device_target         mgs_obt;
         struct ptlrpc_service           *mgs_service;
         struct vfsmount                 *mgs_vfsmnt;
         struct super_block              *mgs_sb;
@@ -508,10 +531,10 @@ struct mds_obd {
         cfs_dentry_t                    *mds_logs_dir;
         cfs_dentry_t                    *mds_objects_dir;
         struct llog_handle              *mds_cfg_llh;
-        struct obd_device               *mds_osc_obd; /* XXX lov_obd */
+        struct obd_device               *mds_lov_obd;
         struct obd_uuid                  mds_lov_uuid;
         char                            *mds_profile;
-        struct obd_export               *mds_osc_exp; /* XXX lov_exp */
+        struct obd_export               *mds_lov_exp;
         struct lov_desc                  mds_lov_desc;
         __u32                            mds_id;
 
@@ -626,6 +649,9 @@ struct lov_qos_rr {
         unsigned long       lqr_dirty:1;     /* recalc round-robin list */
 };
 
+/* allow statfs data caching for 1 second */
+#define OBD_STATFS_CACHE_SECONDS 1
+
 struct lov_statfs_data {
         struct obd_info   lsd_oi;
         struct obd_statfs lsd_statfs;
@@ -760,7 +786,7 @@ struct niobuf_local {
 
 #define LUSTRE_CMM_NAME         "cmm"
 #define LUSTRE_MDD_NAME         "mdd"
-#define LUSTRE_OSD_NAME         "osd"
+#define LUSTRE_OSD_NAME         "osd-ldiskfs"
 #define LUSTRE_VVP_NAME         "vvp"
 #define LUSTRE_LMV_NAME         "lmv"
 #define LUSTRE_CMM_MDC_NAME     "cmm-mdc"
@@ -872,6 +898,8 @@ static inline void oti_free_cookies(struct obd_trans_info *oti)
  * Events signalled through obd_notify() upcall-chain.
  */
 enum obd_notify_event {
+        /* target added */
+        OBD_NOTIFY_CREATE,
         /* Device connect start */
         OBD_NOTIFY_CONNECT,
         /* Device activated */
@@ -916,13 +944,6 @@ struct target_recovery_data {
         cfs_completion_t  trd_finishing;
 };
 
-enum filter_groups {
-        FILTER_GROUP_MDS0 = 0,
-        FILTER_GROUP_LLOG = 1,
-        FILTER_GROUP_ECHO = 2 ,
-        FILTER_GROUP_MDS1_N_BASE = 3
-};
-
 /**
   * In HEAD for CMD, the object is created in group number which is 3>=
   * or indexing starts from 3. To test this assertions are added to disallow
@@ -933,48 +954,30 @@ enum filter_groups {
   * 2. The group number indexing starts from 0 instead of 3
   */
 
-static inline int filter_group_is_mds(obd_gr group)
-{
-        return (group == FILTER_GROUP_MDS0 ||
-                group >= FILTER_GROUP_MDS1_N_BASE);
-}
-
-#define LASSERT_MDS_GROUP(group) LASSERT(filter_group_is_mds(group))
+#define LASSERT_SEQ_IS_MDT(seq) LASSERT(fid_seq_is_mdt(seq))
 
-static inline __u64 objgrp_to_mdsno(obd_gr group)
+static inline __u64 objseq_to_mdsno(obd_seq seq)
 {
-        LASSERT(filter_group_is_mds(group));
-        if (group == FILTER_GROUP_MDS0)
+        LASSERT_SEQ_IS_MDT(seq);
+        if (seq == FID_SEQ_OST_MDT0)
                 return 0;
-        return group - FILTER_GROUP_MDS1_N_BASE + 1;
+        return seq - FID_SEQ_OST_MDT1 + 1;
 }
 
-static inline int mdt_to_obd_objgrp(int mdtid)
+static inline int mdt_to_obd_objseq(int mdtid)
 {
         /**
-         * MDS0 uses group 0 always, other MDSes will use groups from
-         * FILTER_GROUP_MDS1_N_BASE
+         * MDS0 uses seq 0 pre FID-on-OST, other MDSes will use seq from
+         * FID_SEQ_OST_MDT1
          */
         if (mdtid)
-                return FILTER_GROUP_MDS1_N_BASE + mdtid - 1;
+                return FID_SEQ_OST_MDT1 + mdtid - 1;
         return 0;
 }
 
-static inline __u64 obdo_mdsno(struct obdo *oa)
-{
-        LASSERT((oa->o_valid & OBD_MD_FLGROUP));
-        return objgrp_to_mdsno(oa->o_gr);
-}
-
-static inline int obdo_is_mds(struct obdo *oa)
-{
-        LASSERT(oa->o_valid & OBD_MD_FLGROUP);
-        return filter_group_is_mds(oa->o_gr);
-}
-
 struct obd_llog_group {
         cfs_list_t         olg_list;
-        int                olg_group;
+        int                olg_seq;
         struct llog_ctxt  *olg_ctxts[LLOG_MAX_CTXTS];
         cfs_waitq_t        olg_waitq;
         cfs_spinlock_t     olg_lock;
@@ -987,6 +990,7 @@ struct obd_llog_group {
 #define MAX_OBD_NAME 128
 #define OBD_DEVICE_MAGIC        0XAB5CD6EF
 #define OBD_DEV_BY_DEVNAME      0xffffd0de
+
 struct obd_device {
         struct obd_type        *obd_type;
         __u32                   obd_magic;
@@ -998,16 +1002,15 @@ struct obd_device {
         struct lu_device       *obd_lu_dev;
 
         int                     obd_minor;
+        /* bitfield modification is protected by obd_dev_lock */
         unsigned long obd_attached:1,      /* finished attach */
                       obd_set_up:1,        /* finished setup */
                       obd_recovering:1,    /* there are recoverable clients */
                       obd_abort_recovery:1,/* recovery expired */
                       obd_version_recov:1, /* obd uses version checking */
-                      obd_recovery_expired:1,
                       obd_replayable:1,    /* recovery is enabled; inform clients */
                       obd_no_transno:1,    /* no committed-transno notification */
                       obd_no_recov:1,      /* fail instead of retry messages */
-                      obd_req_replaying:1, /* replaying requests */
                       obd_stopping:1,      /* started cleanup */
                       obd_starting:1,      /* started setup */
                       obd_force:1,         /* cleanup with > 0 obd refcount */
@@ -1017,6 +1020,9 @@ struct obd_device {
                       obd_inactive:1,      /* device active/inactive
                                            * (for /proc/status only!!) */
                       obd_process_conf:1;  /* device is processing mgs config */
+        /* use separate field as it is set in interrupt to don't mess with
+         * protection of other bits using _bh lock */
+        unsigned long obd_recovery_expired:1;
         /* uuid-export hash body */
         cfs_hash_t             *obd_uuid_hash;
         /* nid-export hash body */
@@ -1034,7 +1040,7 @@ struct obd_device {
         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 */
-        cfs_spinlock_t          obd_dev_lock;
+        cfs_spinlock_t          obd_dev_lock; /* protects obd bitfield above */
         cfs_semaphore_t         obd_dev_sem;
         __u64                   obd_last_committed;
         struct fsfilt_operations *obd_fsops;
@@ -1055,11 +1061,14 @@ struct obd_device {
         int                              obd_connected_clients;
         int                              obd_stale_clients;
         int                              obd_delayed_clients;
-        cfs_spinlock_t                   obd_processing_task_lock; /* BH lock (timer) */
+        /* this lock protects all recovery list_heads, timer and
+         * obd_next_recovery_transno value */
+        cfs_spinlock_t                   obd_recovery_task_lock;
         __u64                            obd_next_recovery_transno;
         int                              obd_replayed_requests;
         int                              obd_requests_queued_for_recovery;
         cfs_waitq_t                      obd_next_transno_waitq;
+        /* protected by obd_recovery_task_lock */
         cfs_timer_t                      obd_recovery_timer;
         time_t                           obd_recovery_start; /* seconds */
         time_t                           obd_recovery_end; /* seconds, for lprocfs_status */
@@ -1071,6 +1080,7 @@ struct obd_device {
         int                              obd_replayed_locks;
         cfs_atomic_t                     obd_req_replay_clients;
         cfs_atomic_t                     obd_lock_replay_clients;
+        /* all lists are protected by obd_recovery_task_lock */
         cfs_list_t                       obd_req_replay_queue;
         cfs_list_t                       obd_lock_replay_queue;
         cfs_list_t                       obd_final_req_queue;
@@ -1158,6 +1168,8 @@ enum obd_cleanup_stage {
 #define KEY_SET_FS              "set_fs"
 /*      KEY_SET_INFO in lustre_idl.h */
 #define KEY_SPTLRPC_CONF        "sptlrpc_conf"
+#define KEY_CONNECT_FLAG        "connect_flags"
+#define KEY_SYNC_LOCK_CANCEL    "sync_lock_cancel"
 
 
 struct lu_context;
@@ -1181,7 +1193,7 @@ struct md_op_data {
         struct lu_fid           op_fid4; /* to the operation locks. */
         mdsno_t                 op_mds;  /* what mds server open will go to */
         struct lustre_handle    op_handle;
-        __u64                   op_mod_time;
+        obd_time                op_mod_time;
         const char             *op_name;
         int                     op_namelen;
         __u32                   op_mode;
@@ -1333,8 +1345,8 @@ struct obd_ops {
                       struct lustre_handle *srconn, struct lov_stripe_md *src,
                       obd_size start, obd_size end, struct obd_trans_info *);
         int (*o_iterate)(struct lustre_handle *conn,
-                         int (*)(obd_id, obd_gr, void *),
-                         obd_id *startid, obd_gr group, void *data);
+                         int (*)(obd_id, obd_seq, void *),
+                         obd_id *startid, obd_seq seq, void *data);
         int (*o_preprw)(int cmd, struct obd_export *exp, struct obdo *oa,
                         int objcount, struct obd_ioobj *obj,
                         struct niobuf_remote *remote, int *nr_pages,
@@ -1389,7 +1401,8 @@ struct obd_ops {
                           struct obd_quotactl *);
         int (*o_quota_adjust_qunit)(struct obd_export *exp,
                                     struct quota_adjust_qunit *oqaq,
-                                    struct lustre_quota_ctxt *qctxt);
+                                    struct lustre_quota_ctxt *qctxt,
+                                    struct ptlrpc_request_set *rqset);
 
 
         int (*o_ping)(struct obd_export *exp);
@@ -1403,6 +1416,8 @@ struct obd_ops {
                           char *ostname);
         void (*o_getref)(struct obd_device *obd);
         void (*o_putref)(struct obd_device *obd);
+        int (*o_sync_fs)(struct obd_device *obd, struct obd_info *oinfo,
+                         int wait);
         /*
          * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
          * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
@@ -1642,4 +1657,7 @@ static inline struct md_open_data *obd_mod_alloc(void)
         }                                                       \
 })
 
+extern void obdo_from_inode(struct obdo *dst, struct inode *src,
+                            struct lu_fid *parent, obd_flag valid);
+
 #endif /* __OBD_H */