Whamcloud - gitweb
Fixes and cleanups in lmv.
[fs/lustre-release.git] / lustre / include / linux / obd.h
index 05294a4..417cce0 100644 (file)
@@ -1,7 +1,7 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- * Copyright (C) 2001  Cluster File Systems, Inc.
+ * Copyright (C) 2001, 2002 Cluster File Systems, Inc.
  *
  * This code is issued under the GNU General Public License.
  * See the file COPYING in this distribution
 
 #ifndef __OBD_H
 #define __OBD_H
-#include <linux/fs.h>
-#include <linux/list.h>
-#include <linux/smp_lock.h>
-#include <linux/proc_fs.h>
 
+#define IOC_OSC_TYPE         'h'
+#define IOC_OSC_MIN_NR       20
+#define IOC_OSC_SET_ACTIVE   _IOWR(IOC_OSC_TYPE, 21, struct obd_device *)
+#define IOC_OSC_MAX_NR       50
+
+#define IOC_MDC_TYPE         'i'
+#define IOC_MDC_MIN_NR       20
+#define IOC_MDC_LOOKUP       _IOWR(IOC_MDC_TYPE, 20, struct obd_device *)
+/* Moved to lustre_user.h
+#define IOC_MDC_GETSTRIPE    _IOWR(IOC_MDC_TYPE, 21, struct lov_mds_md *) */
+#define IOC_MDC_FINISH_GNS   _IOWR(IOC_MDC_TYPE, 22, struct obd_device *)
+#define IOC_MDC_MAX_NR       50
+
+#ifdef __KERNEL__
+# include <linux/fs.h>
+# include <linux/list.h>
+# include <linux/sched.h> /* for struct task_struct, for current.h */
+# include <asm/current.h> /* for smp_lock.h */
+# include <linux/smp_lock.h>
+# include <linux/proc_fs.h>
+# include <linux/mount.h>
+#endif
+
+#include <linux/lvfs.h>
 #include <linux/lustre_lib.h>
 #include <linux/lustre_idl.h>
+#include <linux/lustre_export.h>
+
+/* this is really local to the OSC */
+struct loi_oap_pages {
+        struct list_head        lop_pending;
+        int                     lop_num_pending;
+        struct list_head        lop_urgent;
+        struct list_head        lop_pending_group;
+};
+
+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 */
+        int loi_ost_idx;           /* OST stripe index in lov_tgt_desc->tgts */
+        int loi_ost_gen;           /* generation of this loi_ost_idx */
+
+        /* used by the osc to keep track of what objects to build into rpcs */
+        struct loi_oap_pages loi_read_lop;
+        struct loi_oap_pages loi_write_lop;
+        /* _cli_ is poorly named, it should be _ready_ */
+        struct list_head loi_cli_item;
+        struct list_head loi_write_item;
+        struct list_head loi_read_item;
+
+        int loi_kms_valid:1;
+        __u64 loi_kms;             /* known minimum size */
+        __u64 loi_rss;             /* recently seen size */
+        __u64 loi_mtime;           /* recently seen mtime */
+        __u64 loi_blocks;          /* recently seen blocks */
+};
+
+static inline void loi_init(struct lov_oinfo *loi)
+{
+        INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending);
+        INIT_LIST_HEAD(&loi->loi_read_lop.lop_urgent);
+        INIT_LIST_HEAD(&loi->loi_read_lop.lop_pending_group);
+        INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending);
+        INIT_LIST_HEAD(&loi->loi_write_lop.lop_urgent);
+        INIT_LIST_HEAD(&loi->loi_write_lop.lop_pending_group);
+        INIT_LIST_HEAD(&loi->loi_cli_item);
+        INIT_LIST_HEAD(&loi->loi_write_item);
+        INIT_LIST_HEAD(&loi->loi_read_item);
+}
+
+struct lov_stripe_md {
+        /* Public members. */
+        __u64 lsm_object_id;        /* lov object id */
+        __u64 lsm_object_gr;        /* lov object id */
+        __u64 lsm_maxbytes;         /* maximum possible file size */
+        unsigned long lsm_xfersize; /* optimal transfer size */
+
+        /* LOV-private members start here -- only for use in lov/. */
+        __u32 lsm_magic;
+        __u32 lsm_stripe_size;      /* size of the stripe */
+        __u32 lsm_pattern;          /* striping pattern (RAID0, RAID1) */
+        unsigned lsm_stripe_count;  /* number of objects being striped over */
+        struct lov_oinfo lsm_oinfo[0];
+};
 
 struct obd_type {
         struct list_head typ_chain;
         struct obd_ops *typ_ops;
+        struct md_ops *typ_md_ops;
+        struct proc_dir_entry *typ_procroot;
         char *typ_name;
         int  typ_refcnt;
 };
 
-struct brw_page { 
-        struct page *pg;
-        obd_size count;
+struct brw_page {
         obd_off  off;
+        struct page *pg;
+        int count;
         obd_flag flag;
 };
 
-struct lov_oinfo { /* per-child structure */
-        __u64 loi_id;
-        __u64 loi_size;
+enum async_flags {
+        ASYNC_READY = 0x1, /* ap_make_ready will not be called before this
+                              page is added to an rpc */
+        ASYNC_URGENT = 0x2,
+        ASYNC_COUNT_STABLE = 0x4, /* ap_refresh_count will not be called
+                                     to give the caller a chance to update
+                                     or cancel the size of the io */
+        ASYNC_GROUP_SYNC = 0x8,  /* ap_completion will not be called, instead
+                                    the page is accounted for in the
+                                    obd_io_group given to 
+                                    obd_queue_group_io */
 };
 
-struct lov_stripe_md {
-        __u32 lmd_magic;
-        __u32 lmd_easize;          /* packed size for MDS of ea */
-        __u64 lmd_object_id;       /* lov object id */
-        __u64 lmd_stripe_offset;   /* offset of the stripe */ 
-        __u64 lmd_stripe_size;     /* size of the stripe */
-        __u32 lmd_stripe_count;    /* how many objects are being striped */
-        __u32 lmd_stripe_pattern;  /* per-lov object stripe pattern */
-        struct lov_oinfo lmd_oinfo[0];
-};
-
-struct lov_stripe_md_one {
-        __u32 lmd_magic;
-        __u32 lmd_easize;          /* packed size for MDS of ea */
-        __u64 lmd_object_id;       /* lov object id */
-        __u64 lmd_stripe_offset;   /* offset of the stripe */ 
-        __u64 lmd_stripe_size;     /* size of the stripe */
-        __u32 lmd_stripe_count;    /* how many objects are being striped */
-        __u32 lmd_stripe_pattern;  /* per-lov object stripe pattern */
-        struct lov_oinfo lmd_oinfo[1];
-};
-
-/* Individual type definitions */
-
-struct ext2_obd {
-        struct super_block *e2_sb;
-        struct vfsmount *e2_vfsmnt;
-};
-
-#define OBD_RUN_CTXT_MAGIC      0xC0FFEEAA
-#define OBD_CTXT_DEBUG          /* development-only debugging */
-struct obd_run_ctxt {
-        struct vfsmount *pwdmnt;
-        struct dentry   *pwd;
-        mm_segment_t     fs;
-#ifdef OBD_CTXT_DEBUG
-        __u32            magic;
-#endif
+struct obd_async_page_ops {
+        int  (*ap_make_ready)(void *data, int cmd);
+        int  (*ap_refresh_count)(void *data, int cmd);
+        void (*ap_fill_obdo)(void *data, int cmd, struct obdo *oa);
+        void (*ap_completion)(void *data, int cmd, struct obdo *oa, int rc);
 };
 
-#ifdef OBD_CTXT_DEBUG
-#define OBD_SET_CTXT_MAGIC(ctxt) (ctxt)->magic = OBD_RUN_CTXT_MAGIC
-#else
-#define OBD_SET_CTXT_MAGIC(ctxt) do {} while(0)
-#endif
+/* the `oig' is passed down from a caller of obd rw methods.  the callee
+ * records enough state such that the caller can sleep on the oig and
+ * be woken when all the callees have finished their work */
+struct obd_io_group {
+        spinlock_t      oig_lock;
+        atomic_t        oig_refcount;
+        int             oig_pending;
+        int             oig_rc;
+        struct list_head oig_occ_list;
+        wait_queue_head_t oig_waitq;
+};
+
+/* the oig callback context lets the callee of obd rw methods register
+ * for callbacks from the caller. */
+struct oig_callback_context {
+        struct list_head occ_oig_item;
+        /* called when the caller has received a signal while sleeping.
+         * callees of this method are encouraged to abort their state 
+         * in the oig.  This may be called multiple times. */
+        void (*occ_interrupted)(struct oig_callback_context *occ);
+};
+
+/* if we find more consumers this could be generalized */
+#define OBD_HIST_MAX 32
+struct obd_histogram {
+        spinlock_t      oh_lock;
+        unsigned long   oh_buckets[OBD_HIST_MAX];
+};
+
+struct ost_server_data;
+
+#define FILTER_SUBDIR_COUNT      32            /* set to zero for no subdirs */
+
+#define FILTER_GROUP_LLOG 1
+#define FILTER_GROUP_ECHO 2
+#define FILTER_GROUP_FIRST_MDS 3
+
+struct filter_subdirs {
+        struct dentry *dentry[FILTER_SUBDIR_COUNT];
+};
+
+struct filter_group_llog {
+        struct list_head list;
+        int group;
+        struct obd_llogs *llogs;
+};
 
 struct filter_obd {
-        char *fo_fstype;
-        struct super_block *fo_sb;
-        struct vfsmount *fo_vfsmnt;
-        struct obd_run_ctxt fo_ctxt;
-        struct dentry *fo_dentry_O;
-        struct dentry *fo_dentry_O_mode[16];
-        spinlock_t fo_lock;
-        __u64 fo_lastino;
+        const char          *fo_fstype;
+        struct super_block  *fo_sb;
+        struct vfsmount     *fo_vfsmnt;
+
+        int                    fo_group_count;
+        struct dentry         *fo_dentry_O; /* the "O"bject directory dentry */
+        struct dentry         **fo_groups;   /* dentries for each group dir */
+        struct filter_subdirs *fo_subdirs;  /* subdir array per group */
+        __u64                 *fo_last_objids; // per-group last created objid
+        struct file          **fo_last_objid_files;
+
+        spinlock_t           fo_objidlock; /* protect fo_lastobjid increment */
+        spinlock_t           fo_translock; /* protect fsd_last_rcvd increment */
+        struct file         *fo_rcvd_filp;
+        struct filter_server_data *fo_fsd;
+        unsigned long       *fo_last_rcvd_slots;
+        __u64                fo_mount_count;
+
+        unsigned int         fo_destroy_in_progress:1;
+        struct semaphore     fo_create_lock;
+
         struct file_operations *fo_fop;
         struct inode_operations *fo_iop;
         struct address_space_operations *fo_aops;
+
+        struct list_head     fo_export_list;
+        int                  fo_subdir_count;
+
+        obd_size             fo_tot_dirty;      /* protected by obd_osfs_lock */
+        obd_size             fo_tot_granted;    /* all values in bytes */
+        obd_size             fo_tot_pending;
+
+        obd_size             fo_readcache_max_filesize;
+
+        struct obd_import   *fo_mdc_imp;
+        struct obd_uuid      fo_mdc_uuid;
+        struct lustre_handle fo_mdc_conn;
+
+        struct semaphore     fo_alloc_lock;
+
+        struct obd_histogram     fo_r_pages;
+        struct obd_histogram     fo_w_pages;
+        struct obd_histogram     fo_r_discont_pages;
+        struct obd_histogram     fo_w_discont_pages;
+        struct obd_histogram     fo_r_discont_blocks;
+        struct obd_histogram     fo_w_discont_blocks;
+
+        struct list_head         fo_llog_list;
+        spinlock_t               fo_llog_list_lock;
 };
 
 struct mds_server_data;
 
+#define OSC_MAX_RIF_DEFAULT       4
+#define OSC_MAX_RIF_MAX          32
+#define OSC_MAX_DIRTY_DEFAULT     4
+#define OSC_MAX_DIRTY_MB_MAX    256     /* totally arbitrary */
+
+struct mdc_rpc_lock;
 struct client_obd {
-        struct ptlrpc_client *cl_client;
-        struct ptlrpc_client *cl_ldlm_client;
-        struct ptlrpc_connection *cl_conn;
-        struct lustre_handle cl_exporth;
-        struct semaphore cl_sem;
-        int cl_conn_count;
-        __u8 cl_target_uuid[37];
-        int cl_max_mdsize;
+        struct obd_import       *cl_import;
+        struct semaphore         cl_sem;
+        int                      cl_conn_count;
+        /* max_mds_easize is purely a performance thing so we don't have to
+         * call obd_size_wiremd() all the time. */
+        int                      cl_max_mds_easize;
+        int                      cl_max_mds_cookiesize;
+        kdev_t                   cl_sandev;
+
+        //struct llog_canceld_ctxt *cl_llcd; /* it's included by obd_llog_ctxt */
+        void                    *cl_llcd_offset;
+
+        struct obd_device       *cl_mgmtcli_obd;
+
+        /* the grant values are protected by loi_list_lock below */
+        long                     cl_dirty;         /* all _dirty_ in bytes */
+        long                     cl_dirty_max;     /* allowed w/o rpc */
+        long                     cl_avail_grant;   /* bytes of credit for ost */
+        long                     cl_lost_grant;    /* lost credits (trunc) */
+        struct list_head         cl_cache_waiters; /* waiting for cache/grant */
+
+        /* keep track of objects that have lois that contain pages which
+         * have been queued for async brw.  this lock also protects the
+         * lists of osc_client_pages that hang off of the loi */
+        spinlock_t               cl_loi_list_lock;
+        struct list_head         cl_loi_ready_list;
+        struct list_head         cl_loi_write_list;
+        struct list_head         cl_loi_read_list;
+        int                      cl_brw_in_flight;
+        /* just a sum of the loi/lop pending numbers to be exported by /proc */
+        int                      cl_pending_w_pages;
+        int                      cl_pending_r_pages;
+        int                      cl_max_pages_per_rpc;
+        int                      cl_max_rpcs_in_flight;
+        struct obd_histogram     cl_read_rpc_hist;
+        struct obd_histogram     cl_write_rpc_hist;
+        struct obd_histogram     cl_read_page_hist;
+        struct obd_histogram     cl_write_page_hist;
+
+        struct mdc_rpc_lock     *cl_rpc_lock;
+        struct mdc_rpc_lock     *cl_setattr_lock;
+        struct osc_creator      cl_oscc;
 };
 
+/* Like a client, with some hangers-on.  Keep mc_client_obd first so that we
+ * can reuse the various client setup/connect functions. */
+struct mgmtcli_obd {
+        struct client_obd        mc_client_obd; /* nested */
+        struct ptlrpc_thread    *mc_ping_thread;
+        struct obd_export       *mc_ping_exp; /* XXX single-target */
+        struct list_head         mc_registered;
+        void                    *mc_hammer;
+};
+
+#define mc_import mc_client_obd.cl_import
+
 struct mds_obd {
-        struct ptlrpc_service *mds_service;
-
-        char *mds_fstype;
-        struct super_block *mds_sb;
-        struct super_operations *mds_sop;
-        struct vfsmount *mds_vfsmnt;
-        struct obd_run_ctxt mds_ctxt;
-        struct file_operations *mds_fop;
-        struct inode_operations *mds_iop;
-        struct address_space_operations *mds_aops;
-        struct mds_fs_operations *mds_fsops;
-        int mds_max_mdsize;
-        struct file *mds_rcvd_filp;
-        spinlock_t mds_last_lock;
-        __u64 mds_last_committed;
-        __u64 mds_last_rcvd;
-        __u64 mds_mount_count;
-        struct ll_fid mds_rootfid;
-        struct mds_server_data *mds_server_data;
-};
-
-struct ldlm_obd {
-        struct ptlrpc_service *ldlm_service;
-        struct ptlrpc_client *ldlm_client;
-        struct ptlrpc_connection *ldlm_server_conn;
+        struct ptlrpc_service           *mds_service;
+        struct ptlrpc_service           *mds_setattr_service;
+        struct ptlrpc_service           *mds_readpage_service;
+        struct super_block              *mds_sb;
+        struct vfsmount                 *mds_vfsmnt;
+        struct dentry                   *mds_fid_de;
+        int                              mds_max_mdsize;
+        int                              mds_max_cookiesize;
+        struct file                     *mds_rcvd_filp;
+        spinlock_t                       mds_transno_lock;
+        __u64                            mds_last_transno;
+        __u64                            mds_mount_count;
+        __u64                            mds_io_epoch;
+        struct semaphore                 mds_epoch_sem;
+        struct ll_fid                    mds_rootfid;
+        struct mds_server_data          *mds_server_data;
+        struct dentry                   *mds_pending_dir;
+        struct dentry                   *mds_logs_dir;
+        struct dentry                   *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;
+        struct obd_export               *mds_osc_exp; /* XXX lov_exp */
+        int                              mds_has_lov_desc;
+        struct lov_desc                  mds_lov_desc;
+        obd_id                          *mds_lov_objids;
+        int                              mds_lov_objids_valid;
+        int                              mds_lov_nextid_set;
+        struct file                     *mds_lov_objid_filp;
+        unsigned long                   *mds_client_bitmap;
+        struct semaphore                 mds_orphan_recovery_sem;
+        /*add mds num here for real mds and cache mds create
+          FIXME later will be totally fixed by b_cmd*/
+        int                              mds_num;
+        atomic_t                         mds_open_count;
+
+        char                            *mds_lmv_name;
+        struct obd_device               *mds_lmv_obd; /* XXX lmv_obd */
+        struct obd_export               *mds_lmv_exp; /* XXX lov_exp */
+        struct ptlrpc_service           *mds_create_service;
+        atomic_t                         mds_real_clients;
+        struct obd_uuid                  mds_lmv_uuid;
+        struct dentry                   *mds_fids_dir;
 };
 
 struct echo_obd {
-        char *eo_fstype;
-        struct super_block *eo_sb;
-        struct vfsmount *eo_vfsmnt;
-        struct obd_run_ctxt eo_ctxt;
+        struct obdo oa;
         spinlock_t eo_lock;
         __u64 eo_lastino;
-        struct file_operations *eo_fop;
-        struct inode_operations *eo_iop;
-        struct address_space_operations *eo_aops;
+        atomic_t eo_getattr;
+        atomic_t eo_setattr;
+        atomic_t eo_create;
+        atomic_t eo_destroy;
+        atomic_t eo_prep;
+        atomic_t eo_read;
+        atomic_t eo_write;
+};
+
+/*
+ * this struct does double-duty acting as either a client or
+ * server instance .. maybe not wise.
+ */
+struct ptlbd_obd {
+        /* server's */
+        struct ptlrpc_service *ptlbd_service;
+        struct file *filp;
+        /* client's */
+        struct ptlrpc_client    bd_client;
+        struct obd_import       *bd_import;
+        struct obd_uuid         bd_server_uuid;
+        struct obd_export       *bd_exp;
+        int refcount; /* XXX sigh */
 };
 
 struct recovd_obd {
-        time_t                recovd_waketime;
-        time_t                recovd_timeout;
-        __u32                 recovd_flags; 
-        __u32                 recovd_wakeup_flag; 
         spinlock_t            recovd_lock;
         struct list_head      recovd_managed_items; /* items managed  */
-        struct list_head      recovd_troubled_items; /* items in trouble */
+        struct list_head      recovd_troubled_items; /* items in recovery */
+
         wait_queue_head_t     recovd_recovery_waitq;
         wait_queue_head_t     recovd_ctl_waitq;
         wait_queue_head_t     recovd_waitq;
-        struct task_struct    *recovd_thread;
+        struct task_struct   *recovd_thread;
+        __u32                 recovd_state;
 };
 
-struct trace_obd {
-        struct obdtrace_opstats *stats;
+struct ost_obd {
+        struct ptlrpc_service *ost_service;
+        struct ptlrpc_service *ost_create_service;
 };
 
-#if 0
-struct snap_obd {
-        unsigned int snap_index;  /* which snapshot index are we accessing */
-        int snap_tableno;
+struct echo_client_obd {
+        struct obd_export   *ec_exp;   /* the local connection to osc/lov */
+        spinlock_t           ec_lock;
+        struct list_head     ec_objects;
+        int                  ec_nstripes;
+        __u64                ec_unique;
 };
 
-#endif
-
-struct ost_obd {
-        struct ptlrpc_service *ost_service;
-        struct lustre_handle ost_conn;   /* the local connection to the OBD */
+struct cache_obd {
+        struct obd_export *cobd_real_exp;/* local connection to target obd */
+        struct obd_export *cobd_cache_exp; /* local connection to cache obd */
+        int    refcount;
+        int    cache_on;
 };
 
-
-struct lov_tgt_desc { 
-        uuid_t uuid;
-        struct lustre_handle conn; 
+struct lov_tgt_desc {
+        struct obd_uuid          uuid;
+        struct obd_export       *ltd_exp;
+        int                      active; /* is this target up for requests */
 };
 
 struct lov_obd {
-        struct obd_device *mdcobd;
+        spinlock_t lov_lock;
         struct lov_desc desc;
-        int bufsize; 
+        int bufsize;
+        int refcount;
+        int lo_catalog_loaded:1;
         struct lov_tgt_desc *tgts;
 };
 
+struct lmv_tgt_desc {
+        struct obd_uuid         uuid;
+        struct obd_export       *ltd_exp;
+        int                      active; /* is this target up for requests */
+};
+
+struct lmv_obd {
+        spinlock_t              lmv_lock;
+        struct lmv_desc         desc;
+        int                     bufsize;
+        int                     refcount;
+        struct lmv_tgt_desc     *tgts;
+        struct obd_uuid         cluuid;
+        struct obd_export       *exp;
+        int                     connected;
+        int                     max_easize;
+        int                     max_cookiesize;
+        int                     server_timeout;
+};
+
 struct niobuf_local {
         __u64 offset;
         __u32 len;
-        __u32 xid;
         __u32 flags;
-        void *addr;
         struct page *page;
-        void *target_private;
         struct dentry *dentry;
+        int lnb_grant_used;
+        int rc;
 };
 
-#define N_LOCAL_TEMP_PAGE 0x00000001
+struct cache_manager_obd {
+        struct obd_device *cm_master_obd;       /* master lov */
+        struct obd_export *cm_master_exp;
+        struct obd_device *cm_cache_obd;        /* cache obdfilter */
+        struct obd_export *cm_cache_exp;
+        int    cm_master_group;                 /* master group*/        
+        struct cmobd_write_service   *cm_write_srv;
+};
+        
+
+/* Don't conflict with on-wire flags OBD_BRW_WRITE, etc */
+#define N_LOCAL_TEMP_PAGE 0x10000000
+
+struct obd_trans_info {
+        __u64                    oti_transno;
+        __u64                   *oti_objid;
+        /* Only used on the server side for tracking acks. */
+        struct oti_req_ack_lock {
+                struct lustre_handle lock;
+                __u32                mode;
+        }                        oti_ack_locks[4];
+        void                    *oti_handle;
+        struct llog_cookie       oti_onecookie;
+        struct llog_cookie      *oti_logcookies;
+        int                      oti_numcookies;
+};
+
+static inline void oti_alloc_cookies(struct obd_trans_info *oti,int num_cookies)
+{
+        if (!oti)
+                return;
+
+        if (num_cookies == 1)
+                oti->oti_logcookies = &oti->oti_onecookie;
+        else
+                OBD_ALLOC(oti->oti_logcookies,
+                          num_cookies * sizeof(oti->oti_onecookie));
+
+        oti->oti_numcookies = num_cookies;
+}
+
+static inline void oti_free_cookies(struct obd_trans_info *oti)
+{
+        if (!oti || !oti->oti_logcookies)
+                return;
+
+        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));
+        oti->oti_logcookies = NULL;
+        oti->oti_numcookies = 0;
+}
+
+/* llog contexts */
+enum llog_ctxt_id {
+        LLOG_CONFIG_ORIG_CTXT =  0,
+        LLOG_CONFIG_REPL_CTXT =  1,
+        LLOG_UNLINK_ORIG_CTXT =  2,
+        LLOG_UNLINK_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_REINT_ORIG_CTXT  = 12,
+        LLOG_MAX_CTXTS
+};
+
+struct obd_llogs {
+        struct llog_ctxt        *llog_ctxt[LLOG_MAX_CTXTS];
+};
 
 /* corresponds to one of the obd's */
 struct obd_device {
@@ -212,108 +532,306 @@ struct obd_device {
 
         /* common and UUID name of this device */
         char *obd_name;
-        __u8 obd_uuid[37];
+        struct obd_uuid obd_uuid;
 
         int obd_minor;
-        int obd_flags;
+        unsigned int obd_attached:1, obd_set_up:1, obd_recovering:1,
+                obd_abort_recovery:1, obd_replayable:1, obd_no_transno:1,
+                obd_no_recov:1, obd_stopping:1;
+        atomic_t obd_refcount;
+        wait_queue_head_t obd_refcount_waitq;
         struct proc_dir_entry *obd_proc_entry;
-        struct list_head obd_exports;
-        struct list_head obd_imports;
+        struct list_head       obd_exports;
+        int                    obd_num_exports;
         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 */
-        spinlock_t obd_dev_lock;
+        spinlock_t             obd_dev_lock;
+        __u64                  obd_last_committed;
+        struct fsfilt_operations *obd_fsops;
+        spinlock_t              obd_osfs_lock;
+        struct obd_statfs       obd_osfs;
+        unsigned long           obd_osfs_age;
+        struct lvfs_run_ctxt    obd_lvfs_ctxt;
+        struct obd_llogs        obd_llogs;
+        struct llog_ctxt        *obd_llog_ctxt[LLOG_MAX_CTXTS];
+        struct obd_device       *obd_observer;
+        struct obd_export       *obd_self_export;
+
+        /* XXX encapsulate all this recovery data into one struct */
+        svc_handler_t                    obd_recovery_handler;
+        int                              obd_max_recoverable_clients;
+        int                              obd_connected_clients;
+        int                              obd_recoverable_clients;
+        spinlock_t                       obd_processing_task_lock;
+        pid_t                            obd_processing_task;
+        __u64                            obd_next_recovery_transno;
+        int                              obd_replayed_requests;
+        int                              obd_requests_queued_for_recovery;
+        wait_queue_head_t                obd_next_transno_waitq;
+        struct list_head                 obd_uncommitted_replies;
+        spinlock_t                       obd_uncommitted_replies_lock;
+        struct timer_list                obd_recovery_timer;
+        struct list_head                 obd_recovery_queue;
+        struct list_head                 obd_delayed_reply_queue;
+
         union {
-                struct ext2_obd ext2;
                 struct filter_obd filter;
                 struct mds_obd mds;
                 struct client_obd cli;
                 struct ost_obd ost;
-                //                struct osc_obd osc;
-                struct ldlm_obd ldlm;
+                struct echo_client_obd echo_client;
                 struct echo_obd echo;
                 struct recovd_obd recovd;
-                struct trace_obd trace;
                 struct lov_obd lov;
-#if 0
-                struct snap_obd snap;
-#endif
+                struct cache_obd cobd;
+                struct ptlbd_obd ptlbd;
+                struct mgmtcli_obd mgmtcli;
+                struct lmv_obd lmv;
+                struct cache_manager_obd cmobd;
         } u;
+       /* Fields used by LProcFS */
+        unsigned int           obd_cntr_base;
+        struct lprocfs_stats  *obd_stats;
+        struct proc_dir_entry *obd_svc_procroot;
+        struct lprocfs_stats  *obd_svc_stats;
 };
 
-struct io_cb_data;
+#define OBD_OPT_FORCE           0x0001
+#define OBD_OPT_FAILOVER        0x0002
+#define OBD_OPT_REAL_CLIENT     0x0004
+
+#define OBD_LLOG_FL_SENDNOW     0x0001
+
+struct mdc_op_data;
 
 struct obd_ops {
-        int (*o_iocontrol)(long cmd, struct lustre_handle *, int len,
+        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 lustre_handle *, obd_count keylen, void *key,
-                          obd_count *vallen, void **val);
-        int (*o_set_info)(struct lustre_handle *, obd_count keylen, void *key,
-                          obd_count vallen, void *val);
+        int (*o_get_info)(struct obd_export *, __u32 keylen, void *key,
+                          __u32 *vallen, void *val);
+        int (*o_set_info)(struct obd_export *, __u32 keylen, void *key,
+                          __u32 vallen, void *val);
         int (*o_attach)(struct obd_device *dev, obd_count len, void *data);
         int (*o_detach)(struct obd_device *dev);
         int (*o_setup) (struct obd_device *dev, obd_count len, void *data);
-        int (*o_cleanup)(struct obd_device *dev);
+        int (*o_precleanup)(struct obd_device *dev, int flags);
+        int (*o_cleanup)(struct obd_device *dev, int flags);
+        int (*o_postrecov)(struct obd_device *dev);
         int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
-                         char *cluuid);
-        int (*o_disconnect)(struct lustre_handle *conn);
-
-
-        int (*o_statfs)(struct lustre_handle *conn, struct statfs *statfs);
+                         struct obd_uuid *cluuid);
+        int (*o_disconnect)(struct obd_export *exp, int flags);
+
+        int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
+                        unsigned long max_age);
+        int (*o_packmd)(struct obd_export *exp, struct lov_mds_md **disk_tgt,
+                        struct lov_stripe_md *mem_src);
+        int (*o_unpackmd)(struct obd_export *exp,struct lov_stripe_md **mem_tgt,
+                          struct lov_mds_md *disk_src, int disk_len);
         int (*o_preallocate)(struct lustre_handle *, obd_count *req,
                              obd_id *ids);
-        int (*o_create)(struct lustre_handle *conn,  struct obdo *oa,
-                        struct lov_stripe_md **ea);
-        int (*o_destroy)(struct lustre_handle *conn, struct obdo *oa,
-                         struct lov_stripe_md *ea);
-        int (*o_setattr)(struct lustre_handle *conn, struct obdo *oa,
+        int (*o_create)(struct obd_export *exp,  struct obdo *oa,
+                        struct lov_stripe_md **ea, struct obd_trans_info *oti);
+        int (*o_destroy)(struct obd_export *exp, struct obdo *oa,
+                         struct lov_stripe_md *ea, struct obd_trans_info *oti);
+        int (*o_setattr)(struct obd_export *exp, struct obdo *oa,
+                         struct lov_stripe_md *ea, struct obd_trans_info *oti);
+        int (*o_getattr)(struct obd_export *exp, struct obdo *oa,
                          struct lov_stripe_md *ea);
-        int (*o_getattr)(struct lustre_handle *conn, struct obdo *oa,
-                         struct lov_stripe_md *ea);
-        int (*o_open)(struct lustre_handle *conn, struct obdo *oa,
-                      struct lov_stripe_md *);
-        int (*o_close)(struct lustre_handle *conn, struct obdo *oa,
-                       struct lov_stripe_md *);
-        int (*o_brw)(int rw, struct lustre_handle *conn,
-                     struct lov_stripe_md *md, obd_count oa_bufs,
-                     struct brw_page *pgarr, brw_callback_t callback, 
-                     struct io_cb_data *data);
-        int (*o_punch)(struct lustre_handle *conn, struct obdo *tgt,
-                       struct lov_stripe_md *md, obd_size count,
-                       obd_off offset);
-        int (*o_sync)(struct lustre_handle *conn, struct obdo *tgt,
-                      obd_size count, obd_off offset);
-        int (*o_migrate)(struct lustre_handle *conn, struct obdo *dst,
-                         struct obdo *src, obd_size count, obd_off offset);
-        int (*o_copy)(struct lustre_handle *dstconn, struct obdo *dst,
-                      struct lustre_handle *srconn, struct obdo *src,
-                      obd_size count, obd_off offset);
+        int (*o_getattr_async)(struct obd_export *exp, struct obdo *oa,
+                               struct lov_stripe_md *ea,
+                               struct ptlrpc_request_set *set);
+        int (*o_brw)(int rw, struct obd_export *exp, struct obdo *oa,
+                     struct lov_stripe_md *ea, obd_count oa_bufs,
+                     struct brw_page *pgarr, struct obd_trans_info *oti);
+        int (*o_brw_async)(int rw, struct obd_export *exp, struct obdo *oa,
+                           struct lov_stripe_md *ea, obd_count oa_bufs,
+                           struct brw_page *pgarr, struct ptlrpc_request_set *,
+                           struct obd_trans_info *oti);
+        int (*o_prep_async_page)(struct obd_export *exp, 
+                                 struct lov_stripe_md *lsm,
+                                 struct lov_oinfo *loi, 
+                                 struct page *page, obd_off offset, 
+                                 struct obd_async_page_ops *ops, void *data,
+                                 void **res);
+        int (*o_queue_async_io)(struct obd_export *exp, 
+                                struct lov_stripe_md *lsm, 
+                                struct lov_oinfo *loi, void *cookie, 
+                                int cmd, obd_off off, int count, 
+                                obd_flag brw_flags, obd_flag async_flags);
+        int (*o_queue_group_io)(struct obd_export *exp, 
+                                struct lov_stripe_md *lsm, 
+                                struct lov_oinfo *loi, 
+                                struct obd_io_group *oig, 
+                                void *cookie, int cmd, obd_off off, int count, 
+                                obd_flag brw_flags, obd_flag async_flags);
+        int (*o_trigger_group_io)(struct obd_export *exp, 
+                                  struct lov_stripe_md *lsm, 
+                                  struct lov_oinfo *loi, 
+                                  struct obd_io_group *oig);
+        int (*o_set_async_flags)(struct obd_export *exp,
+                                struct lov_stripe_md *lsm,
+                                struct lov_oinfo *loi, void *cookie,
+                                obd_flag async_flags);
+        int (*o_teardown_async_page)(struct obd_export *exp,
+                                     struct lov_stripe_md *lsm,
+                                     struct lov_oinfo *loi, void *cookie);
+        int (*o_punch)(struct obd_export *exp, struct obdo *oa,
+                       struct lov_stripe_md *ea, obd_size start,
+                       obd_size end, struct obd_trans_info *oti);
+        int (*o_sync)(struct obd_export *exp, struct obdo *oa,
+                      struct lov_stripe_md *ea, obd_size start, obd_size end);
+        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);
+        int (*o_copy)(struct lustre_handle *dstconn, struct lov_stripe_md *dst,
+                      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 (*o_preprw)(int cmd, struct lustre_handle *conn,
+        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, void **desc_private);
-        int (*o_commitrw)(int cmd, struct lustre_handle *conn,
+                        struct niobuf_local *local, struct obd_trans_info *oti);
+        int (*o_commitrw)(int cmd, struct obd_export *exp, struct obdo *oa,
                           int objcount, struct obd_ioobj *obj,
                           int niocount, struct niobuf_local *local,
-                          void *desc_private);
-        int (*o_enqueue)(struct lustre_handle *conn, struct lov_stripe_md *md,
-                         struct lustre_handle *parent_lock, 
-                         __u32 type, void *cookie, int cookielen, __u32 mode,
-                         int *flags, void *cb, void *data, int datalen,
+                          struct obd_trans_info *oti, int rc);
+        int (*o_write_extents)(struct obd_export *exp, struct obd_ioobj *obj,
+                               int niocount, struct niobuf_local *local,int rc);
+        int (*o_enqueue)(struct obd_export *, struct lov_stripe_md *,
+                         __u32 type, ldlm_policy_data_t *, __u32 mode,
+                         int *flags, void *bl_cb, void *cp_cb, void *gl_cb,
+                         void *data, __u32 lvb_len, void *lvb_swabber,
                          struct lustre_handle *lockh);
-        int (*o_cancel)(struct lustre_handle *, struct lov_stripe_md *md, __u32 mode, struct lustre_handle *);
+        int (*o_match)(struct obd_export *, struct lov_stripe_md *, __u32 type,
+                       ldlm_policy_data_t *, __u32 mode, int *flags, void *data,
+                       struct lustre_handle *lockh);
+        int (*o_change_cbdata)(struct obd_export *, struct lov_stripe_md *,
+                               ldlm_iterator_t it, void *data);
+        int (*o_cancel)(struct obd_export *, struct lov_stripe_md *md,
+                        __u32 mode, struct lustre_handle *);
+        int (*o_cancel_unused)(struct obd_export *, struct lov_stripe_md *,
+                               int flags, void *opaque);
+        int (*o_san_preprw)(int cmd, struct obd_export *exp,
+                            struct obdo *oa, int objcount,
+                            struct obd_ioobj *obj, int niocount,
+                            struct niobuf_remote *remote);
+        int (*o_init_export)(struct obd_export *exp);
+        int (*o_destroy_export)(struct obd_export *exp);
+
+        /* llog related obd_methods */
+        int (*o_llog_init)(struct obd_device *, struct obd_llogs *,
+                           struct obd_device *, int, struct llog_catid *);
+        int (*o_llog_finish)(struct obd_device *, struct obd_llogs *, int);
+        int (*o_llog_connect)(struct obd_device *, struct llogd_conn_body *);
+
+       
+        /* metadata-only methods */
+        int (*o_pin)(struct obd_export *, obd_id ino, __u32 gen, int type,
+                     struct obd_client_handle *, int flag);
+        int (*o_unpin)(struct obd_export *, struct obd_client_handle *, int);
+
+        int (*o_import_event)(struct obd_device *, struct obd_import *,
+                              enum obd_import_event);
+
+        int (*o_notify)(struct obd_device *obd, struct obd_device *watched,
+                        int active);
+        int (*o_init_ea_size)(struct obd_export *, int, int);
+
+        /* 
+         * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
+         * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
+         * Also, add a wrapper function in include/linux/obd_class.h.
+         */
 };
 
-#if BITS_PER_LONG > 32
-#define LPU64 "%lu"
-#define LPD64 "%ld"
-#define LPX64 "%lx"
-#else
-#define LPU64 "%Lu"
-#define LPD64 "%Ld"
-#define LPX64 "%Lx"
-#endif
+struct ll_uctxt;
+
+struct md_ops {
+        int (*m_getstatus)(struct obd_export *, struct ll_fid *);
+        int (*m_change_cbdata)(struct obd_export *, struct ll_fid *,
+                               ldlm_iterator_t, void *);
+        int (*m_change_cbdata_name)(struct obd_export *, struct ll_fid *,
+                                    char *, int, struct ll_fid *,
+                                    ldlm_iterator_t, void *);
+        int (*m_close)(struct obd_export *, struct obdo *,
+                       struct obd_client_handle *,
+                       struct ptlrpc_request **);
+        int (*m_create)(struct obd_export *, struct mdc_op_data *,
+                        const void *, int, int, __u32, __u32,
+                        __u64, struct ptlrpc_request **);
+        int (*m_done_writing)(struct obd_export *, struct obdo *);
+        int (*m_enqueue)(struct obd_export *, int, struct lookup_intent *,
+                         int, struct mdc_op_data *, struct lustre_handle *,
+                         void *, int, ldlm_completion_callback,
+                         ldlm_blocking_callback, void *);
+        int (*m_getattr)(struct obd_export *, struct ll_fid *,
+                         unsigned long, unsigned int,
+                         struct ptlrpc_request **);
+        int (*m_getattr_name)(struct obd_export *, struct ll_fid *,
+                              char *, int, unsigned long,
+                              unsigned int, struct ptlrpc_request **);
+        int (*m_intent_lock)(struct obd_export *, struct ll_uctxt *,
+                             struct ll_fid *, const char *, int,
+                             void *, int, struct ll_fid *,
+                             struct lookup_intent *, int,
+                             struct ptlrpc_request **,
+                             ldlm_blocking_callback);
+        int (*m_link)(struct obd_export *, struct mdc_op_data *,
+                      struct ptlrpc_request **);
+        int (*m_rename)(struct obd_export *, struct mdc_op_data *,
+                        const char *, int, const char *, int,
+                        struct ptlrpc_request **);
+        int (*m_setattr)(struct obd_export *, struct mdc_op_data *,
+                         struct iattr *, void *, int , void *, int,
+                         struct ptlrpc_request **);
+        int (*m_sync)(struct obd_export *, struct ll_fid *,
+                      struct ptlrpc_request **);
+        int (*m_readpage)(struct obd_export *, struct ll_fid *,
+                          __u64, struct page *, struct ptlrpc_request **);
+        int (*m_unlink)(struct obd_export *, struct mdc_op_data *,
+                        struct ptlrpc_request **);
+        int (*m_valid_attrs)(struct obd_export *, struct ll_fid *);
+        struct obd_device * (*m_get_real_obd)(struct obd_export *,
+                             char *name, int len);
+        
+        int (*m_req2lustre_md)(struct obd_export *exp, 
+                               struct ptlrpc_request *req, unsigned int offset,
+                               struct obd_export *osc_exp, struct lustre_md *md);
+        int (*m_set_open_replay_data)(struct obd_export *exp,
+                                      struct obd_client_handle *och,
+                                      struct ptlrpc_request *open_req);
+        int (*m_clear_open_replay_data)(struct obd_export *exp,
+                                        struct obd_client_handle *och);
+        int (*m_store_inode_generation)(struct obd_export *exp, 
+                                        struct ptlrpc_request *req, int reqoff,
+                                        int repoff);
+        int (*m_set_lock_data)(struct obd_export *exp, __u64 *l, void *data);
+
+        /* 
+         * NOTE: If adding ops, add another LPROCFS_OBD_OP_INIT() line
+         * to lprocfs_alloc_obd_stats() in obdclass/lprocfs_status.c.
+         * Also, add a wrapper function in include/linux/obd_class.h.
+         */
+};
 
-#endif
+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",
+                       obd->obd_name, transno, error);
+                return;
+        }
+        CDEBUG(D_HA, "%s: transno "LPD64" committed\n",
+               obd->obd_name, transno);
+        if (transno > obd->obd_last_committed) {
+                obd->obd_last_committed = transno;
+                ptlrpc_commit_replies (obd);
+        }
+}
+
+#endif /* __OBD_H */