/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Copyright (C) 2001 Cluster File Systems, Inc. <info@clusterfs.com>
+ * Copyright (C) 2001-2003 Cluster File Systems, Inc. <info@clusterfs.com>
*
* This file is part of Lustre, http://www.lustre.org.
*
#define _LUSTRE_MDS_H
#ifdef __KERNEL__
+# include <linux/fs.h>
+# include <linux/dcache.h>
+#endif
-#include <linux/fs.h>
+#include <linux/lustre_handles.h>
+#include <linux/kp30.h>
#include <linux/lustre_idl.h>
+#include <linux/lustre_lib.h>
+#include <linux/lustre_dlm.h>
+#include <linux/lustre_log.h>
+#include <linux/lustre_export.h>
struct ldlm_lock_desc;
-struct lov_stripe_md;
struct mds_obd;
struct ptlrpc_connection;
struct ptlrpc_client;
struct obd_export;
struct ptlrpc_request;
struct obd_device;
+struct ll_file_data;
#define LUSTRE_MDS_NAME "mds"
#define LUSTRE_MDT_NAME "mdt"
#define LUSTRE_MDC_NAME "mdc"
struct mds_update_record {
- __u32 ur_fsuid;
- __u32 ur_fsgid;
__u32 ur_opcode;
struct ll_fid *ur_fid1;
struct ll_fid *ur_fid2;
char *ur_name;
int ur_tgtlen;
char *ur_tgt;
+ int ur_eadatalen;
+ void *ur_eadata;
+ int ur_cookielen;
+ struct llog_cookie *ur_logcookies;
struct iattr ur_iattr;
+ struct lvfs_ucred ur_uc;
__u64 ur_rdev;
__u32 ur_mode;
- __u32 ur_uid;
- __u32 ur_gid;
__u64 ur_time;
+ __u32 ur_flags;
};
-#define MDS_LR_CLIENT 8192
-#define MDS_LR_SIZE 128
+#define _ur_fsuid ur_uc.luc_fsuid
+#define _ur_fsgid ur_uc.luc_fsgid
+#define _ur_cap ur_uc.luc_cap
+#define _ur_suppgid1 ur_uc.luc_suppgid1
+#define _ur_suppgid2 ur_uc.luc_suppgid2
+
+/* i_attr_flags holds the open count in the inode in 2.4 */
+//XXX Alex implement on 2.4 with i_attr_flags and find soln for 2.5 please
+#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
+# define mds_inode_oatomic(inode) ((atomic_t *)&(inode)->i_cindex)
+#else
+# define mds_inode_oatomic(inode) ((atomic_t *)&(inode)->i_attr_flags)
+#endif
+
+# define mds_open_orphan_count(inode) \
+ atomic_read(mds_inode_oatomic(inode))
+# define mds_open_orphan_inc(inode) \
+ atomic_inc(mds_inode_oatomic(inode))
+# define mds_open_orphan_dec_test(inode) \
+ atomic_dec_and_test(mds_inode_oatomic(inode))
+
+#define mds_inode_is_orphan(inode) ((inode)->i_flags & 0x4000000)
+#define mds_inode_set_orphan(inode) (inode)->i_flags |= 0x4000000
+
+#define MDS_LR_SERVER_SIZE 512
+
+#define MDS_LR_CLIENT_START 8192
+#define MDS_LR_CLIENT_SIZE 128
+#if MDS_LR_CLIENT_START < MDS_LR_SERVER_SIZE
+#error "Can't have MDS_LR_CLIENT_START < MDS_LR_SERVER_SIZE"
+#endif
#define MDS_CLIENT_SLOTS 17
-#define MDS_MOUNT_RECOV 2
+#define MDS_ROCOMPAT_LOVOBJID 0x00000001
+#define MDS_ROCOMPAT_SUPP (MDS_ROCOMPAT_LOVOBJID)
-/* Data stored per server at the head of the last_rcvd file. In le32 order. */
+#define MDS_INCOMPAT_SUPP (0)
+
+#define REAL_MDS_NUMBER 1
+#define CACHE_MDS_NUMBER 0
+
+/*flags for indicate the record are come from cmobd reint or
+ mdc create */
+#define REC_REINT_CREATE 0x0001
+
+/* Data stored per server at the head of the last_rcvd file. In le32 order.
+ * Try to keep this the same as fsd_server_data so we might one day merge. */
struct mds_server_data {
- __u8 msd_uuid[37]; /* server UUID */
- __u8 uuid_padding[3]; /* unused */
- __u64 msd_last_rcvd; /* last completed transaction ID */
- __u64 msd_mount_count; /* MDS incarnation number */
- __u8 padding[512 - 56];
+ __u8 msd_uuid[40]; /* server UUID */
+ __u64 msd_last_transno; /* last completed transaction ID */
+ __u64 msd_mount_count; /* MDS incarnation number */
+ __u64 msd_unused;
+ __u32 msd_feature_compat; /* compatible feature flags */
+ __u32 msd_feature_rocompat;/* read-only compatible feature flags */
+ __u32 msd_feature_incompat;/* incompatible feature flags */
+ __u32 msd_server_size; /* size of server data area */
+ __u32 msd_client_start; /* start of per-client data area */
+ __u16 msd_client_size; /* size of per-client data area */
+ __u16 msd_subdir_count; /* number of subdirectories for objects */
+ __u64 msd_catalog_oid; /* recovery catalog object id */
+ __u32 msd_catalog_ogen; /* recovery catalog inode generation */
+ __u8 msd_peeruuid[40]; /* UUID of LOV/OSC associated with MDS */
+ __u8 msd_padding[MDS_LR_SERVER_SIZE - 140];
};
/* Data stored per client in the last_rcvd file. In le32 order. */
struct mds_client_data {
- __u8 mcd_uuid[37]; /* client UUID */
- __u8 uuid_padding[3]; /* unused */
- __u64 mcd_last_rcvd; /* last completed transaction ID */
- __u64 mcd_mount_count; /* MDS incarnation number */
- __u64 mcd_last_xid; /* client RPC xid for the last transaction */
- __u8 padding[MDS_LR_SIZE - 64];
-};
-
-/* In-memory access to client data from MDS struct */
-struct mds_export_data {
- struct list_head med_open_head;
- spinlock_t med_open_lock;
- struct mds_client_data *med_mcd;
- int med_off;
+ __u8 mcd_uuid[40]; /* client UUID */
+ __u64 mcd_last_transno; /* last completed transaction ID */
+ __u64 mcd_last_xid; /* xid for the last transaction */
+ __u32 mcd_last_result; /* result from last RPC */
+ __u32 mcd_last_data; /* per-op data (disposition for open &c.) */
+ __u8 mcd_padding[MDS_LR_CLIENT_SIZE - 64];
};
/* file data for open files on MDS */
struct mds_file_data {
- struct list_head mfd_list;
- struct lustre_handle mfd_clienthandle;
- __u64 mfd_servercookie;
- struct file *mfd_file;
+ struct portals_handle mfd_handle; /* must be first */
+ atomic_t mfd_refcount;
+ struct list_head mfd_list;
+ __u64 mfd_xid;
+ int mfd_mode;
+ struct dentry *mfd_dentry;
};
/* mds/mds_reint.c */
int mds_reint_rec(struct mds_update_record *r, int offset,
- struct ptlrpc_request *req);
-
-/* lib/mds_updates.c */
-void mds_unpack_body(struct mds_body *b);
-void mds_unpack_fid(struct ll_fid *fid);
-void mds_pack_fid(struct ll_fid *fid);
-void mds_pack_req_body(struct ptlrpc_request *);
-void mds_pack_rep_body(struct ptlrpc_request *);
-int mds_update_unpack(struct ptlrpc_request *, int offset,
- struct mds_update_record *);
-
-void mds_getattr_pack(struct ptlrpc_request *req, int offset,
- struct inode *inode, const char *name, int namelen);
-void mds_setattr_pack(struct ptlrpc_request *, int offset, struct inode *,
- struct iattr *, const char *name, int namelen);
-void mds_create_pack(struct ptlrpc_request *, int offset, struct inode *,
- __u32 mode, __u64 id, __u32 uid, __u32 gid, __u64 time,
- const char *name, int namelen, const char *tgt,
- int tgtlen);
-void mds_unlink_pack(struct ptlrpc_request *, int offset, struct inode *inode,
- struct inode *child, __u32 mode, const char *name,
- int namelen);
-void mds_link_pack(struct ptlrpc_request *, int offset, struct inode *ino,
- struct inode *dir, const char *name, int namelen);
-void mds_rename_pack(struct ptlrpc_request *, int offset, struct inode *srcdir,
- struct inode *tgtdir, const char *name, int namelen,
- const char *tgt, int tgtlen);
-void mds_pack_inode2fid(struct ll_fid *fid, struct inode *inode);
-void mds_pack_inode2body(struct mds_body *body, struct inode *inode);
+ struct ptlrpc_request *req, struct lustre_handle *);
/* mds/handler.c */
-struct dentry *mds_name2locked_dentry(struct obd_device *, struct dentry *dir,
- struct vfsmount **mnt, char *name,
- int namelen, int lock_mode,
- struct lustre_handle *lockh,
- int dir_lock_mode);
+#ifdef __KERNEL__
struct dentry *mds_fid2locked_dentry(struct obd_device *obd, struct ll_fid *fid,
struct vfsmount **mnt, int lock_mode,
- struct lustre_handle *lockh);
+ struct lustre_handle *lockh, int *pmode,
+ char *name, int namelen, __u64 lockpart);
struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid,
struct vfsmount **mnt);
-int mds_reint(int offset, struct ptlrpc_request *req);
+int mds_update_server_data(struct obd_device *, int force_sync);
+
+/* mds/mds_fs.c */
+int mds_fs_setup(struct obd_device *obddev, struct vfsmount *mnt);
+int mds_fs_cleanup(struct obd_device *obddev, int failover);
+#endif
+
+/* mds/mds_lov.c */
+
+/* mdc/mdc_locks.c */
+int it_disposition(struct lookup_intent *it, int flag);
+void it_set_disposition(struct lookup_intent *it, int flag);
+int it_open_error(int phase, struct lookup_intent *it);
+int mdc_set_lock_data(struct obd_export *exp, __u64 *lockh, void *data);
+int mdc_change_cbdata(struct obd_export *exp, struct ll_fid *fid,
+ ldlm_iterator_t it, void *data);
+int mdc_intent_lock(struct obd_export *exp, struct ll_uctxt *,
+ struct ll_fid *parent,
+ const char *name, int len, void *lmm, int lmmsize,
+ struct ll_fid *child,
+ struct lookup_intent *, int,
+ struct ptlrpc_request **reqp,
+ ldlm_blocking_callback cb_blocking);
+int mdc_enqueue(struct obd_export *exp,
+ int lock_type,
+ struct lookup_intent *it,
+ int lock_mode,
+ struct mdc_op_data *data,
+ struct lustre_handle *lockh,
+ void *lmm,
+ int lmmlen,
+ ldlm_completion_callback cb_completion,
+ ldlm_blocking_callback cb_blocking,
+ void *cb_data);
/* mdc/mdc_request.c */
-int mdc_enqueue(struct lustre_handle *conn, int lock_type,
- struct lookup_intent *it, int lock_mode, struct inode *dir,
- struct dentry *de, struct lustre_handle *lockh, char *tgt,
- int tgtlen, void *data, int datalen);
-int mdc_cancel_unused(struct lustre_handle *conn, struct inode *, int flags);
-int mdc_getlovinfo(struct obd_device *obd, struct lustre_handle *mdc_connh,
- struct ptlrpc_request **request);
-int mdc_getstatus(struct lustre_handle *conn, struct ll_fid *rootfid);
-int mdc_getattr(struct lustre_handle *conn,
- obd_id ino, int type, unsigned long valid, size_t ea_size,
+int mdc_req2lustre_md(struct obd_export *exp_mdc, struct ptlrpc_request *req,
+ unsigned int offset, struct obd_export *exp_osc,
+ struct lustre_md *md);
+int mdc_getstatus(struct obd_export *exp, struct ll_fid *rootfid);
+int mdc_getattr(struct obd_export *exp, struct ll_fid *fid,
+ unsigned long valid, unsigned int ea_size,
struct ptlrpc_request **request);
-int mdc_setattr(struct lustre_handle *conn,
- struct inode *, struct iattr *iattr, struct ptlrpc_request **);
-int mdc_open(struct lustre_handle *conn, obd_id ino, int type, int flags,
- struct lov_stripe_md *, struct lustre_handle *fh,
+int mdc_getattr_name(struct obd_export *exp, struct ll_fid *fid,
+ char *filename, int namelen, unsigned long valid,
+ unsigned int ea_size, struct ptlrpc_request **request);
+int mdc_setattr(struct obd_export *exp, struct mdc_op_data *data,
+ struct iattr *iattr, void *ea, int ealen, void *ea2, int ea2len,
+ struct ptlrpc_request **request);
+int mdc_open(struct obd_export *exp, obd_id ino, int type, int flags,
+ struct lov_mds_md *lmm, int lmm_size, struct lustre_handle *fh,
+ struct ptlrpc_request **);
+struct obd_client_handle;
+int mdc_set_open_replay_data(struct obd_export *exp,
+ struct obd_client_handle *och,
+ struct ptlrpc_request *open_req);
+int mdc_clear_open_replay_data(struct obd_export *exp,
+ struct obd_client_handle *och);
+int mdc_close(struct obd_export *, struct obdo *, struct obd_client_handle *,
+ struct ptlrpc_request **);
+int mdc_readpage(struct obd_export *exp, struct ll_fid *mdc_fid,
+ __u64, struct page *, struct ptlrpc_request **);
+int mdc_create(struct obd_export *exp, struct mdc_op_data *op_data,
+ const void *data, int datalen, int mode, __u32 uid, __u32 gid,
+ __u64 rdev, struct ptlrpc_request **request);
+int mdc_unlink(struct obd_export *exp, struct mdc_op_data *data,
+ struct ptlrpc_request **request);
+int mdc_link(struct obd_export *exp, struct mdc_op_data *data,
struct ptlrpc_request **);
-int mdc_close(struct lustre_handle *conn, obd_id ino, int type,
- struct lustre_handle *fh, struct ptlrpc_request **req);
-int mdc_readpage(struct lustre_handle *conn, obd_id ino,
- int type, __u64 offset, char *addr, struct ptlrpc_request **);
-int mdc_create(struct lustre_handle *conn,
- struct inode *dir, const char *name, int namelen,
- const char *tgt, int tgtlen, int mode, __u32 uid, __u32 gid,
- __u64 time, __u64 rdev, struct lov_stripe_md *md,
- struct ptlrpc_request **);
-int mdc_unlink(struct lustre_handle *, struct inode *dir, struct inode *child,
- __u32 mode, const char *name, int namelen,
- struct ptlrpc_request **);
-int mdc_link(struct lustre_handle *conn,
- struct dentry *src, struct inode *dir, const char *name,
- int namelen, struct ptlrpc_request **);
-int mdc_rename(struct lustre_handle *conn,
- struct inode *src, struct inode *tgt, const char *old,
- int oldlen, const char *new, int newlen,
- struct ptlrpc_request **);
-int mdc_create_client(obd_uuid_t uuid, struct ptlrpc_client *cl);
+int mdc_rename(struct obd_export *exp, struct mdc_op_data *data,
+ const char *old, int oldlen, const char *new, int newlen,
+ struct ptlrpc_request **request);
+int mdc_sync(struct obd_export *exp, struct ll_fid *fid,
+ struct ptlrpc_request **);
+int mdc_create_client(struct obd_uuid uuid, struct ptlrpc_client *cl);
/* Store the generation of a newly-created inode in |req| for replay. */
-void mdc_store_inode_generation(struct ptlrpc_request *req, int reqoff,
- int repoff);
-
-extern int mds_client_add(struct mds_export_data *med, int cl_off);
-extern int mds_client_free(struct obd_export *exp);
-
-/* mds/mds_fs.c */
-struct mds_fs_operations {
- struct module *fs_owner;
- void *(* fs_start)(struct inode *inode, int op);
- int (* fs_commit)(struct inode *inode, void *handle);
- int (* fs_setattr)(struct dentry *dentry, void *handle,
- struct iattr *iattr);
- int (* fs_set_md)(struct inode *inode, void *handle,
- struct lov_mds_md *md);
- int (* fs_get_md)(struct inode *inode,
- struct lov_mds_md *md);
- ssize_t (* fs_readpage)(struct file *file, char *buf, size_t count,
- loff_t *offset);
- void (* fs_delete_inode)(struct inode *inode);
- void (* cl_delete_inode)(struct inode *inode);
- int (* fs_journal_data)(struct file *file);
- int (* fs_set_last_rcvd)(struct mds_obd *mds, void *handle);
- int (* fs_statfs)(struct super_block *sb, struct statfs *sfs);
-};
-
-extern int mds_register_fs_type(struct mds_fs_operations *op, const char *name);
-extern void mds_unregister_fs_type(const char *name);
-extern int mds_fs_setup(struct obd_device *obddev, struct vfsmount *mnt);
-extern void mds_fs_cleanup(struct obd_device *obddev);
-
-#define MDS_FSOP_UNLINK 1
-#define MDS_FSOP_RMDIR 2
-#define MDS_FSOP_RENAME 3
-#define MDS_FSOP_CREATE 4
-#define MDS_FSOP_MKDIR 5
-#define MDS_FSOP_SYMLINK 6
-#define MDS_FSOP_MKNOD 7
-#define MDS_FSOP_SETATTR 8
-#define MDS_FSOP_LINK 9
-
-#endif /* __KERNEL__ */
+int mdc_store_inode_generation(struct obd_export *exp, struct ptlrpc_request *req,
+ int reqoff, int repoff);
+int mdc_llog_process(struct obd_export *, char *logname, llog_cb_t, void *data);
+int mdc_done_writing(struct obd_export *exp, struct obdo *);
+int mdc_reint(struct ptlrpc_request *request, struct mdc_rpc_lock *rpc_lock,
+ int level);
+static inline void mdc_pack_fid(struct ll_fid *fid, obd_id ino, __u32 gen,
+ int type)
+{
+ fid->id = ino;
+ fid->generation = gen;
+ fid->f_type = type;
+}
/* ioctls for trying requests */
#define IOC_REQUEST_TYPE 'f'
#define IOC_REQUEST_CLOSE _IOWR('f', 35, long)
#define IOC_REQUEST_MAX_NR 35
+#define MDS_CHECK_RESENT(req, reconstruct) \
+{ \
+ if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) { \
+ struct mds_client_data *mcd = \
+ req->rq_export->exp_mds_data.med_mcd; \
+ if (mcd->mcd_last_xid == req->rq_xid) { \
+ reconstruct; \
+ RETURN(0); \
+ } \
+ DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")", \
+ mcd->mcd_last_xid); \
+ } \
+}
+
#endif