X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Finclude%2Flinux%2Flustre_mds.h;h=2d816058600b835e433898f7a1abd8a1ce985285;hb=e44e9b278432a1df83482b1cd83b2081fabe94dc;hp=e7f942b81baf3cad4b7e4bf7a6e92a09bff421ee;hpb=8e49c56982fd7690df8456e8ab12a66d2ac9d186;p=fs%2Flustre-release.git diff --git a/lustre/include/linux/lustre_mds.h b/lustre/include/linux/lustre_mds.h index e7f942b..2d81605 100644 --- a/lustre/include/linux/lustre_mds.h +++ b/lustre/include/linux/lustre_mds.h @@ -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) 2001-2003 Cluster File Systems, Inc. * * This file is part of Lustre, http://www.lustre.org. * @@ -18,7 +18,7 @@ * along with Lustre; if not, write to the Free Software * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. * - * MDS data structures. + * MDS data structures. * See also lustre_idl.h for wire formats of requests. * */ @@ -27,14 +27,32 @@ #define _LUSTRE_MDS_H #ifdef __KERNEL__ +# include +# include +#endif -#include +#include +#include #include -#include +#include +#include +#include +#include + +struct ldlm_lock_desc; +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 { +struct mds_update_record { __u32 ur_opcode; struct ll_fid *ur_fid1; struct ll_fid *ur_fid2; @@ -42,200 +60,207 @@ struct mds_update_record { 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; - __u64 ur_id; + struct lvfs_ucred ur_uc; + __u64 ur_rdev; __u32 ur_mode; - __u32 ur_uid; - __u32 ur_gid; __u64 ur_time; -}; + __u32 ur_flags; +}; + +#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_CLIENT 8192 -#define MDS_LR_SIZE 128 +#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) + +#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. */ +/* 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 */ - __u32 mcd_last_xid; /* client RPC xid for the last transaction */ - __u8 padding[MDS_LR_SIZE - 60]; -}; - -/* In-memory access to client data from MDS struct */ -struct mds_client_info { - struct list_head mci_list; - struct list_head mci_open_head; - struct mds_client_data *mci_mcd; - int mci_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 file * mfd_file; - __u64 mfd_clientfd; - __u32 mfd_clientcookie; +struct mds_file_data { + 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, struct ptlrpc_request *req); -struct mds_client_info *mds_uuid_to_mci(struct mds_obd *mds, __u8 *uuid); - -/* lib/mds_updates.c */ -void mds_pack_req_body(struct ptlrpc_request *); -void mds_pack_rep_body(struct ptlrpc_request *); -void mds_unpack_req_body(struct ptlrpc_request *); -void mds_unpack_rep_body(struct ptlrpc_request *); -int mds_update_unpack(struct ptlrpc_request *, struct mds_update_record *); - -void mds_setattr_pack(struct mds_rec_setattr *, struct inode *, struct iattr *); -void mds_create_pack(struct mds_rec_create *, struct inode *, __u32 mode, - __u64 id, __u32 uid, __u32 gid, __u64 time); -void mds_unlink_pack(struct mds_rec_unlink *, struct inode *inode, - struct inode *child); -void mds_link_pack(struct mds_rec_link *, struct inode *ino, struct inode *dir); -void mds_rename_pack(struct mds_rec_rename *, struct inode *srcdir, - struct inode *tgtdir); +int mds_reint_rec(struct mds_update_record *r, int offset, + struct ptlrpc_request *req, struct lustre_handle *); /* mds/handler.c */ -struct dentry *mds_fid2dentry(struct mds_obd *mds, struct ll_fid *fid, struct vfsmount **mnt); - -/* mdc/mdc_request.c */ -int mdc_connect(struct ptlrpc_client *, struct ptlrpc_connection *, - struct ll_fid *rootfid, __u64 *last_committed, __u64 *last_rcvd, - __u32 *last_xid, struct ptlrpc_request **); -int mdc_getattr(struct ptlrpc_client *, struct ptlrpc_connection *, ino_t ino, - int type, unsigned long valid, size_t ea_size, - struct ptlrpc_request **); -int mdc_setattr(struct ptlrpc_client *, struct ptlrpc_connection *, - struct inode *, struct iattr *iattr, struct ptlrpc_request **); -int mdc_open(struct ptlrpc_client *, struct ptlrpc_connection *, ino_t ino, - int type, int flags, __u64 cookie, __u64 *fh, struct ptlrpc_request **req); -int mdc_close(struct ptlrpc_client *cl, struct ptlrpc_connection *peer, - ino_t ino, int type, __u64 fh, struct ptlrpc_request **req); -int mdc_readpage(struct ptlrpc_client *, struct ptlrpc_connection *, ino_t ino, - int type, __u64 offset, char *addr, struct ptlrpc_request **); -int mdc_create(struct ptlrpc_client *, struct ptlrpc_connection *, - struct inode *dir, const char *name, int namelen, - const char *tgt, int tgtlen, - int mode, __u64 id, __u32 uid, __u32 gid, __u64 time, - struct ptlrpc_request **); -int mdc_unlink(struct ptlrpc_client *, struct ptlrpc_connection *, - struct inode *dir, struct inode *child, const char *name, - int namelen, struct ptlrpc_request **); -int mdc_link(struct ptlrpc_client *, struct ptlrpc_connection *, - struct dentry *src, struct inode *dir, const char *name, - int namelen, struct ptlrpc_request **); -int mdc_rename(struct ptlrpc_client *, struct ptlrpc_connection *, - struct inode *src, struct inode *tgt, const char *old, - int oldlen, const char *new, int newlen, - struct ptlrpc_request **); -int mdc_create_client(char *uuid, struct ptlrpc_client *cl); - -struct mds_fs_operations { - 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_objid)(struct inode *inode, void *handle, obd_id id); - void (* fs_get_objid)(struct inode *inode, obd_id *id); - 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); -}; - -#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 - -static inline void *mds_fs_start(struct mds_obd *mds, struct inode *inode, - int op) -{ - return mds->mds_fsops->fs_start(inode, op); -} - -static inline int mds_fs_commit(struct mds_obd *mds, struct inode *inode, - void *handle) -{ - return mds->mds_fsops->fs_commit(inode, handle); -} - -static inline int mds_fs_setattr(struct mds_obd *mds, struct dentry *dentry, - void *handle, struct iattr *iattr) -{ - /* - * NOTE: we probably don't need to take i_sem here when changing - * ATTR_SIZE because the MDS never needs to truncate a file. - * The ext2/ext3 code never truncates a directory, and files - * stored on the MDS are entirely sparse (no data blocks). - * If we do need to get it, we can do it here. - */ - return mds->mds_fsops->fs_setattr(dentry, handle, iattr); -} - -static inline int mds_fs_set_objid(struct mds_obd *mds, struct inode *inode, - void *handle, __u64 id) -{ - return mds->mds_fsops->fs_set_objid(inode, handle, id); -} - -static inline void mds_fs_get_objid(struct mds_obd *mds, struct inode *inode, - __u64 *id) -{ - mds->mds_fsops->fs_get_objid(inode, id); -} - -static inline ssize_t mds_fs_readpage(struct mds_obd *mds, struct file *file, - char *buf, size_t count, loff_t *offset) -{ - return mds->mds_fsops->fs_readpage(file, buf, count, offset); -} +#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, 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_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 -/* Set up callback to update mds->mds_last_committed with the current - * value of mds->mds_last_recieved when this transaction is on disk. - */ -static inline int mds_fs_set_last_rcvd(struct mds_obd *mds, void *handle) -{ - return mds->mds_fsops->fs_set_last_rcvd(mds, handle); -} +/* 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); -/* Enable data journaling on the given file */ -static inline ssize_t mds_fs_journal_data(struct mds_obd *mds, - struct file *file) +/* mdc/mdc_request.c */ +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_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_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. */ +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) { - return mds->mds_fsops->fs_journal_data(file); + fid->id = ino; + fid->generation = gen; + fid->f_type = type; } -extern struct mds_fs_operations mds_ext2_fs_ops; -extern struct mds_fs_operations mds_ext3_fs_ops; -extern struct mds_fs_operations mds_extN_fs_ops; - -#endif /* __KERNEL__ */ - /* ioctls for trying requests */ #define IOC_REQUEST_TYPE 'f' #define IOC_REQUEST_MIN_NR 30 @@ -248,4 +273,18 @@ extern struct mds_fs_operations mds_extN_fs_ops; #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