Whamcloud - gitweb
LU-80 lov: large stripe count support
[fs/lustre-release.git] / lustre / include / obd.h
index ebe7112..e670dba 100644 (file)
@@ -30,6 +30,9 @@
  * Use is subject to license terms.
  */
 /*
+ * Copyright (c) 2011 Whamcloud, Inc.
+ */
+/*
  * This file is part of Lustre, http://www.lustre.org/
  * Lustre is a trademark of Sun Microsystems, Inc.
  */
@@ -54,9 +57,6 @@
 
 #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>
@@ -132,11 +132,13 @@ struct lov_stripe_md {
         cfs_spinlock_t   lsm_lock;
         pid_t            lsm_lock_owner; /* debugging */
 
+        /* maximum possible file size, might change as OSTs status changes,
+         * e.g. disconnected, deactivated */
+        __u64            lsm_maxbytes;
         struct {
                 /* Public members. */
                 __u64 lw_object_id;        /* lov object id */
                 __u64 lw_object_seq;       /* lov object seq */
-                __u64 lw_maxbytes;         /* maximum possible file size */
 
                 /* LOV-private members start here -- only for use in lov/. */
                 __u32 lw_magic;
@@ -151,7 +153,6 @@ struct lov_stripe_md {
 
 #define lsm_object_id    lsm_wire.lw_object_id
 #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
 #define lsm_pattern      lsm_wire.lw_pattern
@@ -172,6 +173,7 @@ struct obd_info {
            - while lock handling, the flags obtained on the enqueue
            request are set here.
            - while stats, the flags used for control delay/resend.
+           - while setattr, the flags used for distinguish punch operation
          */
         int                     oi_flags;
         /* Lock handle specific for every OSC lock. */
@@ -228,8 +230,15 @@ struct brw_page {
 
 struct ost_server_data;
 
+struct osd_properties {
+        size_t osd_max_ea_size;
+};
+
+#define OBT_MAGIC       0xBDDECEAE
 /* hold common fields for "target" device */
 struct obd_device_target {
+        __u32                     obt_magic;
+        __u32                     obt_instance;
         struct super_block       *obt_sb;
         /** last_rcvd file */
         struct file              *obt_rcvd_filp;
@@ -241,6 +250,7 @@ struct obd_device_target {
         cfs_rw_semaphore_t        obt_rwsem;
         struct vfsmount          *obt_vfsmnt;
         struct file              *obt_health_check_filp;
+        struct osd_properties    obt_osd_properties;
 };
 
 /* llog contexts */
@@ -301,7 +311,8 @@ struct filter_obd {
         int                  fo_tot_granted_clients;
 
         obd_size             fo_readcache_max_filesize;
-        int                  fo_read_cache:1,   /**< enable read-only cache */
+        cfs_spinlock_t       fo_flags_lock;
+        unsigned 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*/
                              fo_raid_degraded:1;/**< RAID device degraded */
@@ -343,6 +354,9 @@ 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;
@@ -371,6 +385,14 @@ 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
 
@@ -448,7 +470,6 @@ struct client_obd {
         cfs_waitq_t              cl_destroy_waitq;
 
         struct mdc_rpc_lock     *cl_rpc_lock;
-        struct mdc_rpc_lock     *cl_setattr_lock;
         struct mdc_rpc_lock     *cl_close_lock;
         struct osc_creator       cl_oscc;
 
@@ -482,14 +503,15 @@ struct client_obd {
 #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;
         struct dentry                   *mgs_configs_dir;
-        struct dentry                   *mgs_fid_de;
         cfs_list_t                       mgs_fs_db_list;
         cfs_semaphore_t                  mgs_sem;
         cfs_proc_dir_entry_t            *mgs_proc_live;
+        cfs_time_t                       mgs_start_time;
 };
 
 struct mds_obd {
@@ -498,7 +520,6 @@ struct mds_obd {
         struct ptlrpc_service           *mds_service;
         struct ptlrpc_service           *mds_setattr_service;
         struct ptlrpc_service           *mds_readpage_service;
-        cfs_dentry_t                    *mds_fid_de;
         int                              mds_max_mdsize;
         int                              mds_max_cookiesize;
         __u64                            mds_io_epoch;
@@ -509,10 +530,10 @@ 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;
 
@@ -567,11 +588,12 @@ struct obd_id_info {
 /* */
 
 struct echo_obd {
-        struct obdo          eo_oa;
-        cfs_spinlock_t       eo_lock;
-        __u64                eo_lastino;
-        struct lustre_handle eo_nl_lock;
-        cfs_atomic_t         eo_prep;
+        struct obd_device_target eo_obt;
+        struct obdo              eo_oa;
+        cfs_spinlock_t           eo_lock;
+        __u64                    eo_lastino;
+        struct lustre_handle     eo_nl_lock;
+        cfs_atomic_t             eo_prep;
 };
 
 struct ost_obd {
@@ -627,6 +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;
@@ -761,7 +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"
@@ -849,8 +874,8 @@ static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
         if (num_cookies == 1)
                 oti->oti_logcookies = &oti->oti_onecookie;
         else
-                OBD_ALLOC(oti->oti_logcookies,
-                          num_cookies * sizeof(oti->oti_onecookie));
+                OBD_ALLOC_LARGE(oti->oti_logcookies,
+                                num_cookies * sizeof(oti->oti_onecookie));
 
         oti->oti_numcookies = num_cookies;
 }
@@ -863,8 +888,8 @@ static inline void oti_free_cookies(struct obd_trans_info *oti)
         if (oti->oti_logcookies == &oti->oti_onecookie)
                 LASSERT(oti->oti_numcookies == 1);
         else
-                OBD_FREE(oti->oti_logcookies,
-                         oti->oti_numcookies * sizeof(oti->oti_onecookie));
+                OBD_FREE_LARGE(oti->oti_logcookies,
+                               oti->oti_numcookies*sizeof(oti->oti_onecookie));
         oti->oti_logcookies = NULL;
         oti->oti_numcookies = 0;
 }
@@ -873,6 +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 */
@@ -889,7 +916,10 @@ enum obd_notify_event {
         /* Configuration event */
         OBD_NOTIFY_CONFIG,
         /* Trigger quota recovery */
-        OBD_NOTIFY_QUOTA
+        OBD_NOTIFY_QUOTA,
+        /* Administratively deactivate/activate event */
+        OBD_NOTIFY_DEACTIVATE,
+        OBD_NOTIFY_ACTIVATE
 };
 
 /* bit-mask flags for config events */
@@ -963,6 +993,7 @@ 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;
@@ -974,16 +1005,15 @@ 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 */
@@ -992,7 +1022,11 @@ struct obd_device {
                       obd_no_conn:1,       /* deny new connections */
                       obd_inactive:1,      /* device active/inactive
                                            * (for /proc/status only!!) */
+                      obd_no_ir:1,         /* no imperative recovery. */
                       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 */
@@ -1010,7 +1044,7 @@ 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;
@@ -1028,25 +1062,30 @@ struct obd_device {
         time_t                  obd_eviction_timer; /* for ping evictor */
 
         int                              obd_max_recoverable_clients;
-        int                              obd_connected_clients;
+        cfs_atomic_t                     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 */
-        time_t                           obd_recovery_time_hard;
+        int                              obd_recovery_time_hard;
         int                              obd_recovery_timeout;
+        int                              obd_recovery_ir_factor;
 
         /* new recovery stuff from CMD2 */
         struct target_recovery_data      obd_recovery_data;
         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;
@@ -1132,12 +1171,26 @@ enum obd_cleanup_stage {
 #define KEY_REGISTER_TARGET     "register_target"
 #define KEY_REVIMP_UPD          "revimp_update"
 #define KEY_SET_FS              "set_fs"
+#define KEY_TGT_COUNT           "tgt_count"
 /*      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;
 
+/* /!\ must be coherent with include/linux/namei.h on patched kernel */
+#define IT_OPEN     (1 << 0)
+#define IT_CREAT    (1 << 1)
+#define IT_READDIR  (1 << 2)
+#define IT_GETATTR  (1 << 3)
+#define IT_LOOKUP   (1 << 4)
+#define IT_UNLINK   (1 << 5)
+#define IT_TRUNC    (1 << 6)
+#define IT_GETXATTR (1 << 7)
+#define IT_EXEC     (1 << 8)
+#define IT_PIN      (1 << 9)
+
 static inline int it_to_lock_mode(struct lookup_intent *it)
 {
         /* CREAT needs to be tested before open (both could be set) */
@@ -1157,7 +1210,7 @@ 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;
@@ -1204,11 +1257,10 @@ struct md_enqueue_info {
         struct md_op_data       mi_data;
         struct lookup_intent    mi_it;
         struct lustre_handle    mi_lockh;
-        struct dentry          *mi_dentry;
         struct inode           *mi_dir;
         md_enqueue_cb_t         mi_cb;
+        __u64                   mi_cbdata;
         unsigned int            mi_generation;
-        void                   *mi_cbdata;
 };
 
 struct obd_ops {
@@ -1299,9 +1351,9 @@ struct obd_ops {
         int (*o_punch)(struct obd_export *exp, struct obd_info *oinfo,
                        struct obd_trans_info *oti,
                        struct ptlrpc_request_set *rqset);
-        int (*o_sync)(struct obd_export *exp, struct obdo *oa,
-                      struct lov_stripe_md *ea, obd_size start, obd_size end,
-                      void *capa);
+        int (*o_sync)(struct obd_export *exp, struct obd_info *oinfo,
+                      obd_size start, obd_size end,
+                      struct ptlrpc_request_set *set);
         int (*o_migrate)(struct lustre_handle *conn, struct lov_stripe_md *dst,
                          struct lov_stripe_md *src, obd_size start,
                          obd_size end, struct obd_trans_info *oti);
@@ -1365,7 +1417,8 @@ 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);
@@ -1463,8 +1516,8 @@ struct md_ops {
         int (*m_sync)(struct obd_export *, const struct lu_fid *,
                       struct obd_capa *, struct ptlrpc_request **);
         int (*m_readpage)(struct obd_export *, const struct lu_fid *,
-                          struct obd_capa *, __u64, struct page *,
-                          struct ptlrpc_request **);
+                          struct obd_capa *, __u64, struct page **,
+                          unsigned, struct ptlrpc_request **);
 
         int (*m_unlink)(struct obd_export *, struct md_op_data *,
                         struct ptlrpc_request **);
@@ -1492,7 +1545,7 @@ struct md_ops {
                                       struct ptlrpc_request *);
         int (*m_clear_open_replay_data)(struct obd_export *,
                                         struct obd_client_handle *);
-        int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u32 *);
+        int (*m_set_lock_data)(struct obd_export *, __u64 *, void *, __u64 *);
 
         ldlm_mode_t (*m_lock_match)(struct obd_export *, int,
                                     const struct lu_fid *, ldlm_type_t,
@@ -1516,7 +1569,7 @@ struct md_ops {
                                       struct ldlm_enqueue_info *);
 
         int (*m_revalidate_lock)(struct obd_export *, struct lookup_intent *,
-                                 struct lu_fid *, __u32 *);
+                                 struct lu_fid *, __u64 *bits);
 
         /*
          * NOTE: If adding ops, add another LPROCFS_MD_OP_INIT() line to
@@ -1621,4 +1674,11 @@ 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);
 
+/* return 1 if client should be resend request */
+static inline int client_should_resend(int resend, struct client_obd *cli)
+{
+        return cfs_atomic_read(&cli->cl_resends) ?
+               cfs_atomic_read(&cli->cl_resends) > resend : 1;
+}
+
 #endif /* __OBD_H */