source: trunk/kernel/fs/vfs.h @ 649

/*
 * vfs.h - Virtual File System definition.
 *
 * Author  Mohamed Lamine Karaoui (2014,2015)
 *         Alain Greiner (2016,2017,2018)
 *
 * Copyright (c) UPMC Sorbonne Universites
 *
 * This file is part of ALMOS-MKH.
 *
 * ALMOS-MKH is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by
 * the Free Software Foundation; version 2.0 of the License.
 *
 * ALMOS-MKH is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with ALMOS-MKH; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

#ifndef _VFS_H_
#define _VFS_H_

#include <kernel_config.h>
#include <hal_kernel_types.h>
#include <hal_atomic.h>
#include <remote_rwlock.h>
#include <remote_busylock.h>
#include <busylock.h>
#include <list.h>
#include <xlist.h>
#include <bits.h>
#include <xhtab.h>
#include <errno.h>
#include <shared_syscalls.h>
#include <fatfs.h>
#include <ramfs.h>
#include <devfs.h>
#include <dev_ioc.h>

/****  Forward declarations  ***/

struct vfs_inode_s;
struct vfs_dentry_s;
struct vfs_ctx_s;
struct vfs_file_s;

struct mapper_s;
struct process_s;
struct device_s;
struct vseg_s;
struct page_s;

/******************************************************************************************
 * These flags are used to define the working mode of the vfs_lookup() function.  
 *****************************************************************************************/

#define VFS_LOOKUP_DIR      0x01     /* the searched inode must be a directory           */
#define VFS_LOOKUP_OPEN         0x02     /* the search is for an open/opendir                */
#define VFS_LOOKUP_PARENT       0x04     /* return the parent inode (not the inode itself)   */
#define VFS_LOOKUP_CREATE   0x10     /* file must be created if missing                  */
#define VFS_LOOKUP_EXCL     0x20     /* file cannot previously exist                     */

/******************************************************************************************
 * This structure defines a VFS context, that contains informations common to all inodes
 * and dentries for a given file system. As it is declared as a global variable in the
 * kdata segment (fs_context[] array), it is replicated in all clusters.
 * The <extend> field is a pointer on the FS specific context extension. 
 * This extension is dynamically allocated by kernel_init in all clusters.
 * In each cluster, both this VFS context and the FS specific context are handled as
 * private by the local OS intance.
 *****************************************************************************************/

typedef enum
{
        FS_TYPE_DEVFS = 0,
        FS_TYPE_FATFS = 1,
        FS_TYPE_RAMFS = 2,
 
        FS_TYPES_NR   = 3,
}
vfs_fs_type_t;

typedef enum
{
    CTX_ATTR_READ_ONLY    = 0x01,            /*! write access prohibited                 */
    CTX_ATTR_SYNC         = 0x10,            /*! synchronise FS on each write            */
}
vfs_ctx_attr_t;

typedef struct vfs_ctx_s
{
        vfs_fs_type_t  type;                     /*! File System type                        */
        uint32_t           attr;                     /*! global attributes for all files in FS   */
        uint32_t       total_clusters;           /*! total number of clusters on device      */
        uint32_t       cluster_size;             /*! cluster size on device (bytes)          */
        xptr_t         vfs_root_xp;              /*! extended pointer on VFS root inode      */
    busylock_t     lock;                     /*! lock protecting inum allocator          */
    uint32_t       bitmap[BITMAP_SIZE(CONFIG_VFS_MAX_INODES)];  /* inum allocator        */
    void         * extend;                   /*! FS specific context extension           */
}
vfs_ctx_t;

/******************************************************************************************
 * This structure define a VFS inode.
 * An inode can have several children dentries (if it is a directory), an can have several
 * parents dentries (if it hass several aliases links):
 * - The "parents" field is the root of the xlist of parents dentries, and the "links"
 *   fiels define the number of aliases parent dentries. only a FILE inode can have
 *   several parents (no hard links for directories). 
 * - The "children" field is an embedded xhtab containing pointers on all local children
 *   dentries. This set of children is empty for a FILE inode.
 * Synchronisation:
 * - the main_lock (remote_rwlock) is used during the inode tree traversal,
 *   or for inode modification (add/remove children in xhtab).
 * - the size_lock (remote_rwlock) is used during read/write accesses to the size
 *   field in the mapper. 
 * - access to the data stored in the associated mapper use the mapper remote_rwlock 
 *   protecting radix tree traversal and modifications.
 *****************************************************************************************/

/* this enum define the VFS inode types values                                           */
/* WARNING : this enum must be kept consistent with macros in <shared_stat.h> file       */
/*           and with types in <shared_dirent.h> file.                                   */

typedef enum   
{
    INODE_TYPE_FILE  =     0,           /*! regular file                                 */
    INODE_TYPE_DIR   =     1,           /*! directory                                    */
    INODE_TYPE_FIFO  =     2,           /*! POSIX named pipe                             */
    INODE_TYPE_PIPE  =     3,           /*! POSIX anonymous pipe                         */
    INODE_TYPE_SOCK  =     4,           /*! POSIX socket                                 */
    INODE_TYPE_DEV   =     5,           /*! character device                             */
    INODE_TYPE_BLK   =     6,           /*! block device                                 */
    INODE_TYPE_SYML  =     7,           /*! symbolic link                                */
}
vfs_inode_type_t;

/* this enum define the VFS inode attributes values */

typedef enum 
{
    INODE_ATTR_DIRTY   =     0x01,       /* modified versus the value on device          */ 
    INODE_ATTR_INLOAD  =     0x04,       /* loading from device in progress              */
    INODE_ATTR_NEW     =     0x08,       /* not saved on device yet                      */
}
vfs_inode_attr_t;

typedef struct vfs_inode_s
{
    struct vfs_ctx_s * ctx;              /*! local pointer on FS context                 */
    uint32_t           inum;             /*! inode identifier (unique in file system)    */
    uint32_t           attr;             /*! inode attributes (see above)                */
    vfs_inode_type_t   type;             /*! inode type (see above)                      */
    uint32_t           size;             /*! number of bytes                             */
    uint32_t           uid;              /*! user owner identifier                       */
    uint32_t           gid;              /*! group owner identifier                      */
    uint32_t           rights;           /*! access rights                               */
    xlist_entry_t      parents;          /*! root of list of parents dentries            */
    uint32_t           links;            /*! number of parent dentries (hard links)      */
    xhtab_t            children;         /*! embedded xhtab of children dentries         */
    remote_rwlock_t    size_lock;        /*! protect read/write to size                  */
    remote_rwlock_t    main_lock;        /*! protect inode tree traversal and modifs     */
    struct mapper_s  * mapper;           /*! associated file cache                       */
    void             * extend;           /*! fs_type_specific inode extension            */
}
vfs_inode_t;

/* This define the masks for the inode <rights> field  */

#define VFS_ISUID          0x0004000
#define VFS_ISGID          0x0002000
#define VFS_ISVTX          0x0001000

#define VFS_IRWXU      0x0000700
#define VFS_IRUSR      0x0000400
#define VFS_IWUSR      0x0000200
#define VFS_IXUSR      0x0000100

#define VFS_IRWXG      0x0000070
#define VFS_IRGRP      0x0000040
#define VFS_IWGRP      0x0000020
#define VFS_IXGRP      0x0000010

#define VFS_IRWXO      0x0000007
#define VFS_IROTH      0x0000004
#define VFS_IWOTH      0x0000002
#define VFS_IXOTH      0x0000001

/******************************************************************************************
 Rpt* This structure defines a directory entry.
 * A dentry contains the name of a remote file/dir, an extended pointer on the 
 * inode representing this file/dir, a local pointer on the inode representing 
 * the parent directory.
 * A dentry can be member of the set of children of a given directory inode (xhtab).
 * A dentry can be member of the set of parents  of a given inode (xlist).
 *****************************************************************************************/

typedef struct vfs_dentry_s
{
    struct vfs_ctx_s   * ctx;            /*! local pointer on FS context                 */
        char                 name[CONFIG_VFS_MAX_NAME_LENGTH];
        uint32_t             length;         /*! name length (bytes)                         */
    struct vfs_inode_s * parent;         /*! local pointer on parent inode               */
    xptr_t               child_xp;       /*! extended pointer on child inode             */
    xlist_entry_t        children;       /*! member of set of children dentries          */
    xlist_entry_t        parents;        /*! member of set of parent dentries            */
        void               * extend;         /*! FS specific extension                       */
}
vfs_dentry_t;

/******************************************************************************************
 * This file structure describes an open file/directory for a given process.
 * It is not replicated, and is dynamically allocated in the cluster that contains
 * the inode, when a thread makes an open() or opendir() system call.
 * It cannot exist a file structure without an inode structure in same cluster.
 * As the fd_array (containing extended pointers on the open file descriptors)
 * is replicated in all process descriptors, we need a references counter.
 *****************************************************************************************/

typedef enum
{
    FD_ATTR_READ_ENABLE    = 0x01,     /*! read access possible                         */
    FD_ATTR_WRITE_ENABLE   = 0x02,     /*! write access possible                        */
    FD_ATTR_APPEND         = 0x04,     /*! append on each write                         */
    FD_ATTR_CLOSE_EXEC     = 0x08,     /*! close file on exec                           */
    FD_ATTR_SYNC           = 0x10,     /*! synchronise FS on each write                 */
    FD_ATTR_IS_DIR         = 0x20,     /*! this is a directory                          */
}
vfs_file_attr_t;

typedef struct vfs_file_s
{
        struct vfs_ctx_s      * ctx;        /*! local pointer on FS context.                 */
        uint32_t                gc;         /*! generation counter                           */
        vfs_file_attr_t         attr;       /*! file attributes bit vector (see above)       */
        vfs_inode_type_t        type;       /*! same type as inode                           */
        uint32_t                offset;     /*! seek position in file                        */
        uint32_t                refcount;   /*! all pointers on this file descriptor         */
        remote_rwlock_t         lock;       /*! protect offset modifications                 */
        struct mapper_s       * mapper;     /*! associated file cache                        */
        struct vfs_inode_s    * inode;      /*! local pointer on associated inode            */
        void                  * extend;     /*! FS specific extension                        */
}
vfs_file_t;


/******************************************************************************************
 *        These functions access / modify  a VFS context.
 *****************************************************************************************/

/****************************************************************************************** 
 * This function initialise a (statically allocated) VFS context in local cluster. 
 ******************************************************************************************
 * @ fs_type        : file system type.
 * @ attr           : global attributes (for all files in FS.
 * @ total_clusters : total number of clusters on device.
 * @ cluster_size   : cluster size on device (bytes).
 * @ vfs_root_xp    : extended pointer on VFS root inode.
 * @ extend         : fs_type_specific extension.
 *****************************************************************************************/
void vfs_ctx_init( vfs_fs_type_t   type,
                   uint32_t        attr,
                       uint32_t        total_clusters,
                       uint32_t        cluster_size,
                       xptr_t          vfs_root_xp,
                   void          * extend );

/****************************************************************************************** 
 * This function allocates an inode identifier from the local cluster inum allocator.
 * The inum respects a fixed format:
 * - the 16 MSB bits contain the cluster identifier : cxy
 * - the 16 LSB bits contains the local inode identifier  : lid
 ******************************************************************************************
 * @ ctx      : local pointer on file system context.
 * @ inum     : [ou] buffer for allocated inode identifier.
 * @ return 0 if success / return non-zero if error.
 *****************************************************************************************/
error_t vfs_ctx_inum_alloc( vfs_ctx_t * ctx,
                            uint32_t  * inum );

/****************************************************************************************** 
 * This function release an inode identifier.                                  
 ******************************************************************************************
 * @ ctx      : local pointer on file system context.
 * @ inum     : released inode identifier.
 *****************************************************************************************/
void vfs_ctx_inum_release( vfs_ctx_t * ctx,
                           uint32_t    inum );



/******************************************************************************************
 *        These low-level functions access / modify a VFS inode descriptor
 *****************************************************************************************/

/****************************************************************************************** 
 * This function returns a printable string for the inode type.
 *****************************************************************************************/
const char * vfs_inode_type_str( vfs_inode_type_t type );

/****************************************************************************************** 
 * This function allocates memory from local cluster for an inode descriptor and the 
 * associated mapper, and partially initialise this inode from arguments values.
 * It does NOT link it to the Inode Tree, as this is done by add_child_in_parent(). 
 * It must called by a local thread. Use the RPC_INODE_CREATE if client thread is remote.
 ******************************************************************************************
 * @ fs_type    : file system type.
 * @ inode_type : inode type.
 * @ attr       : inode attributes.
 * @ rights     : inode access rights.
 * @ uid        : user owner ID.
 * @ gid        : group owner ID.
 * @ inode_xp   : [out] buffer for extended pointer on created inode.
 * @ return 0 if success / return ENOMEM or EINVAL if error.
 *****************************************************************************************/
error_t vfs_inode_create( vfs_fs_type_t     fs_type,
                          uint32_t          attr,
                          uint32_t          rights,
                          uid_t             uid,
                          gid_t             gid,
                          xptr_t          * inode_xp );

/****************************************************************************************** 
 * This function releases memory allocated to an inode descriptor, including
 * all memory allocated to the mapper (both mapper descriptor and radix tree).
 * The mapper should not contain any dirty page (should be synchronized before deletion).
 * It must be executed by a thread running in the cluster containing the inode.
 * Use the rpc_vfs_inode_destroy_client() function if required.
 ****************************************************************************************** 
 * @ inode  : local pointer on inode descriptor.
 *****************************************************************************************/
void vfs_inode_destroy( vfs_inode_t *  inode );  

/****************************************************************************************** 
 * This function returns the <size> of a file/dir from a remote inode,
 * taking the remote_rwlock protecting <size> in READ_MODE.
 *****************************************************************************************
 * @ inode_xp  : extended pointer on the remote inode.
 * @ return the current size.
 *****************************************************************************************/
uint32_t vfs_inode_get_size( xptr_t inode_xp );

/****************************************************************************************** 
 * This function updates the "size" field of a remote inode identified by <inode_xp>.
 * It takes the rwlock protecting the file size in WRITE_MODE, and set the "size" field
 * when the current size is smaller than the requested <size> argument. 
 *****************************************************************************************
 * @ inode_xp  : extended pointer on the remote inode.
 * @ size      : requested size value.
 *****************************************************************************************/
void vfs_inode_update_size( xptr_t   inode_xp,
                            uint32_t size );

/****************************************************************************************** 
 * This function takes the main lock of a remote inode.
 * This lock protect all inode fields, including the children dentries.
 *****************************************************************************************
 * @ inode_xp  : extended pointer on the remote inode.
 *****************************************************************************************/
void vfs_inode_lock( xptr_t inode_xp );

/****************************************************************************************** 
 * This function releases the main lock of a remote inode.
 * This lock protect all inode fiels, including the children dentries.
 *****************************************************************************************
 * @ inode_xp  : extended pointer on the remote inode.
 *****************************************************************************************/
void vfs_inode_unlock( xptr_t inode_xp );

/****************************************************************************************** 
 * This debug function copies the name of a remote inode identified by the <inode_xp>
 * argument to a local buffer identified by the <name> argument.
 * The local buffer size must be at least CONFIG_VFS_MAX_NAME_LENGTH. 
 ******************************************************************************************
 * @ inode_xp  : extended pointer on the remote inode.
 * @ name      : local buffer pointer.
 *****************************************************************************************/
void vfs_inode_get_name( xptr_t inode_xp,
                         char * name );

/******************************************************************************************
 * This function accesses successively all pages of a file identified by the <inode>,
 * argument, to force misses, and load all pages into mapper.
 * The target inode can be a directory or a file, but this function is mainly used 
 * to prefetch a complete directory to the mapper.
 * It must be executed by a thread running in the cluster containing the inode.
 * This function does NOT take any lock.
 ******************************************************************************************
 * @ inode  : local pointer on inode.
 * @ return 0 if success / return -1 if device access failure.
 *****************************************************************************************/
error_t vfs_inode_load_all_pages( vfs_inode_t * inode );

/****************************************************************************************** 
 * This debug function display the curren state of an inode descriptor identified by
 * the <inode_xp> argument.
 *****************************************************************************************/
void vfs_inode_display( xptr_t inode_xp );

/******************************************************************************************
 *        These low-level functions access / modify a VFS dentry descriptor
 *****************************************************************************************/

/****************************************************************************************** 
 * This function allocates memory from local cluster for a dentry descriptor, 
 * initialises it from  arguments values, and returns the extended pointer on dentry.
 * It must called by a local thread. Use the RPC_DENTRY_CREATE if client thread is remote.
 ******************************************************************************************
 * @ fs_type    : [in]  file system type.
 * @ name       : [in]  directory entry file/dir name.
 * @ dentry_xp  : [out] buffer for extended pointer on created dentry.
 * @ return 0 if success / return ENOMEM or EINVAL if error.
 *****************************************************************************************/
error_t vfs_dentry_create( vfs_fs_type_t   fs_type,
                           char          * name,
                           xptr_t        * dentry_xp );
 
/****************************************************************************************** 
 * This function removes the dentry from the parent inode xhtab, and releases the memory
 * allocated to the dentry descriptor.
 * It must be executed by a thread running in the cluster containing the dentry.
 * Use the RPC_DENTRY_DESTROY if required.
 ****************************************************************************************** 
 * @ dentry  : [in] local pointer on dentry descriptor.
 *****************************************************************************************/
void vfs_dentry_destroy( vfs_dentry_t *  dentry );  


/******************************************************************************************
 *        These low-level functions access / modify a VFS file descriptor
 *****************************************************************************************/

/****************************************************************************************** 
 * This function allocates memory and initializes a new local file descriptor.
 * It must be executed in the cluster containing the inode.
 * If the client thread is not running in the owner cluster, it must use the
 * rpc_vfs_file_create_client() function.
 ****************************************************************************************** 
 * @ inode    : local pointer on associated inode.
 * @ attr     : file descriptor attributes.
 * @ file_xp  : [out] buffer for extended pointer on created file descriptor.
 * @ return 0 if success / return ENOMEM if error.
 *****************************************************************************************/
error_t vfs_file_create( vfs_inode_t * inode,
                         uint32_t      attr,
                         xptr_t      * file_xp );  

/****************************************************************************************** 
 * This function releases memory allocated to a local file descriptor.
 * It must be executed by a thread running in the cluster containing the inode,
 * and the file refcount must be zero. Use the RPC_VFS_FILE_DESTROY if required.
 ****************************************************************************************** 
 * @ file  : local pointer on file descriptor.
 *****************************************************************************************/
void vfs_file_destroy( vfs_file_t *  file );  

/****************************************************************************************** 
 * These functions increment (resp. decrement) the count field in a remote file
 * descriptor, using a remote_atomic access. 
 *****************************************************************************************/
void vfs_file_count_up  ( xptr_t   file_xp );
void vfs_file_count_down( xptr_t   file_xp );

/****************************************************************************************** 
 * This debug function copies the name of a the file identified by <file_xp>
 * argument to a local buffer identified by the <name> argument.
 * The local buffer size must be at least CONFIG_VFS_MAX_NAME_LENGTH. 
 *****************************************************************************************
 * @ file_xp  : extended pointer on the remote inode.
 * @ name     : local buffer pointer.
 *****************************************************************************************/
void vfs_file_get_name( xptr_t inode_xp,
                        char * name );




/******************************************************************************************
 *        These functions access / modify the distributed VFS Inode Tree
 *****************************************************************************************/

/****************************************************************************************** 
 * This function returns in a kernel <buffer> allocated by the caller function,
 * the pathname of a file/dir identified by the <inode_xp> argument. 
 * It traverse the Inode Tree from the target node to the root.
 * It can be called by any thread running in any cluster.
 * As this buffer if filled in "reverse order" (i.e. from the target inode to the root),
 * the pathname is stored in the higher part of the buffer. 
 * A pointer on the first character of the pathname is returned in <first> buffer.
 *
 * WARNING : This function takes & releases the remote_rwlock protecting the Inode Tree.
 ******************************************************************************************
 * @ inode_xp    : [in]  extended pointer on target inode descriptor.
 * @ buffer      : [in]  kernel buffer for pathname (allocated by caller).
 * @ first       : [out] pointer on first character in buffer.
 * @ max_size    : [in]  max number of characters in buffer.
 * @ return 0 if success / return EINVAL if buffer too small.
 *****************************************************************************************/
error_t vfs_get_path( xptr_t    inode_xp,
                      char    * buffer,
                      char   ** first,
                      uint32_t  max_size );

/******************************************************************************************
 * This function traverses the the Inode Tree, from inode identified by the <root_xp> 
 * argument, and returns in <inode_xp> the inode identified by the < pathname> argument. 
 * It can be called by a thread running in any cluster.
 * It supports the following flags that define the lookup modes :
 * - VFS_LOOKUP_DIR    : the searched inode must be a directory 
 * - VFS_LOOKUP_OPEN   : the search is for an open/opendir
 * - VFS_LOOKUP_PARENT : return the parent inode (not the inode itself) 
 * - VFS_LOOKUP_CREATE : file/directory must be created if missing on IOC 
 * - VFS_LOOKUP_EXCL   : file cannot previously exist
 * As the inode Tree is a cache, the search policy is the following :
 * - If a given directory name in the path is not found in the inode tree, it try to load
 *   the missing dentry/inode couple, from informations found in the parent directory.
 * - If this directory entry does not exist on IOC, it returns an error.
 * - If the the file identified by the pathname does not exist on IOC but the
 *   flag CREATE is set, the inode is created. It returns an error otherwise.
 * - If the the file identified by the pathname exist on device, but both flags EXCL
 *   and CREATE are set, an error is returned.
 * - If the PARENT flag is set, it returns in <inode_xp> an extended pointer on the parent
 *   inode, and copies in <last_name> buffer a string containing the last name in path.
 *
 * WARNING : The remote_rwlock protecting the Inode Tree must be taken by the caller. 
 *
 * TODO the access rights are not checked yet.
 ******************************************************************************************
 * @ root_xp     : [in]  extended pointer on root inode (can be root of a subtree). 
 * @ pathname    : [in]  path (can be relative or absolute).
 * @ lookup_mode : [in]  flags defining the searching mode.
 * @ inode_xp    : [out] buffer for extended pointer on searched inode.
 * @ last_name   : [out] pointer on buffer for last name in path.
 * @ return 0 if success / ENOENT if inode not found , EACCES if permisson denied,
 *****************************************************************************************/
error_t vfs_lookup( xptr_t             root_xp,
                    char             * pathname,
                    uint32_t           lookup_mode,
                                        xptr_t           * inode_xp,
                    char             * last_name );

/******************************************************************************************
 * This function creates a new couple dentry/inode, and insert it in the Inode-Tree.
 * Only the distributed Inode Tree is modified: it does NOT modify the parent mapper, 
 * and does NOT update the FS on IOC device.
 * It set the inode type to the default INODE_TYPE_FILE value
 * It can be executed by any thread running in any cluster (can be different from both
 * the child cluster and the parent cluster).
 * The new child inode and the parent inode can have different FS types.
 * [Implementation note]
 * As there are cross-references between inode and dentry, this function implements
 * a five steps scenario :
 * 1) The dentry descriptor is created in the cluster containing the existing <parent_xp>
 *    inode, and partially initialized, using the RPC_VFS_CREATE DENTRY if required.
 * 2) The inode and its associated mapper are created in cluster identified by <child_cxy>,
 *    and partially initialised, using the RPC_VFS_CREATE_INODE if required.
 * 3) The pointers on dentry in parent inode are updated, using remote access.
 * 4) The pointers on dentry in child inode are updated, using remotes access.
 * 5) The pointers on parent and child inode in dentry are updated, using remotes access.
 *****************************************************************************************
 * @ child_inode_cxy  : [in]  target cluster for child inode.
 * @ fs_type          : [in]  child inode FS type.
 * @ parent_inode_xp  : [in]  extended pointer on parent inode.
 * @ name             : [in]  new directory entry name.
 * @ dentry_xp        : [out] buffer for extended pointer on dentry
 * @ child_inode_xp   : [out] buffer for extended pointer on child inode.
 * @ return 0 if success / -1 if dentry or inode cannot be created.
 *****************************************************************************************/
error_t vfs_add_child_in_parent( cxy_t              child_inode_cxy,
                                 vfs_fs_type_t      fs_type,
                                 xptr_t             parent_inode_xp,
                                 char             * name,
                                 xptr_t           * dentry_xp,   
                                 xptr_t           * child_inode_xp );

/******************************************************************************************
 * This function removes a remote dentry from the Inode-Tree.
 * - It removes the dentry from the parent inode xhtab ("children" field), and from the
 *   child inode xlist ("parents" field).
 * - It releases the memory allocated to the dentry descriptor.
 * - If the number of parents of the child inode is one, it also releases the memory
 *   allocated to the child inode.
 * Only the Inode Tree is modified: it does NOT modify the parent mapper, 
 * and does NOT update the FS on IOC device.
 * It can be executed by any thread running in any cluster (can be different from both
 * the inode cluster and the dentry cluster).
 ******************************************************************************************
 * @ dentry_xp   : extended pointer on removed dentry.
 *****************************************************************************************/
void vfs_remove_child_from_parent( xptr_t dentry_xp );

/******************************************************************************************
 * This function is called by the vfs_lookup() function when a new (dentry/inode) must
 * be created from scratch and introduced in both the parent mapper and the IOC device.
 * The dentry and inode descriptors must have been previously created by the caller.
 * 1. It allocates one cluster from the relevant FS, updates the FAT mapper,
 *    and synchronously update the IOC device).
 * 2. It set the "size", and "extend" fields in child inode descriptor.
 * 3. It updates the parent directory mapper to introduce the new child,
 *    and synchronously update the IOC device.
 * 4. It set the "extend" field in dentry descriptor.
 * It can be called by a thread running in any cluster.
 ******************************************************************************************
 * @ parent_xp   : extended pointer on parent inode descriptor.
 * @ dentry_xp   : extended pointer on new dentry descriptor.
 * @ child_xp    : extended pointer on child inode descriptor.
 * @ return 0 if success / -1 if failure.
 *****************************************************************************************/
error_t vfs_new_dentry_init( xptr_t   parent_xp,
                             xptr_t   dentry_xp,
                             xptr_t   child_xp );

/******************************************************************************************
 * This function creates in the directory identified by the <child_xp> argument, 
 * the two special dentries <.> and <..>. The parent directory inode is defined
 * by the <parent_xp> argument. The two dentries are introduced in the Inode Tree.
 * They are also introduced in the child directory mapper, and the IOC device is updated.
 * This function is called by all functions creating a brand new directory : vfs_mkdir(),
 * devfs_global_init(), and devfs_local_init().
 ******************************************************************************************
 * @ child_xp    : extended pointer on new directory inode.
 * @ parent_xp   : extended pointer on parent directory inode.
 * @ return 0 if success / -1 if failure.
 *****************************************************************************************/
error_t vfs_add_special_dentries( xptr_t  child_xp,
                                  xptr_t  parent_xp );

/******************************************************************************************
 * This recursive function diplays a complete inode/dentry sub-tree.
 * Any inode can be selected as the sub-tree root.
 * WARNING : this function is not protected against a concurrent inode/dentry removal...
 ******************************************************************************************
 * @ inode_xp   : extended pointer on sub-tree root inode.
 *****************************************************************************************/
void vfs_display( xptr_t   inode_xp );

/****************************************************************************************** 
 * This function mount a given file system type for a given process 
 * TODO non implemented yet [AG].     
 *****************************************************************************************/
error_t vfs_mount_fs_root( struct device_s  * device,
                                       uint32_t           fs_type,
                                       struct process_s * process );




/******************************************************************************************
 *        These functions define the VFS "syscalls" API (user commands)
 *****************************************************************************************/

/****************************************************************************************** 
 * This function moves <size> bytes between a remote file mapper, identified by the 
 * <file_xp> argument, and a - possibly distributed - user space <buffer>, taken into 
 * account the offset in <file_xp>. The transfer direction is defined by <to_buffer>.
 * It is called by the sys_read() and sys_write() functions.
 ******************************************************************************************
 * @ to_buffer : mapper -> buffer if true / buffer -> mapper if false. 
 * @ file_xp   : extended pointer on the remote file descriptor.
 * @ buffer    : user space pointer on buffer (can be physically distributed).
 * @ size      : requested number of bytes from offset.
 * @ returns number of bytes actually moved if success / -1 if error.
 *****************************************************************************************/
int vfs_user_move( bool_t   to_buffer,
                   xptr_t   file_xp, 
                   void   * buffer,
                   uint32_t size );

/****************************************************************************************** 
 * This function moves <size> bytes between a remote file mapper, identified by the 
 * <file_xp> argument, and a - possibly remote - kernel <buffer_xp>, taken into 
 * account the offset in <file_xp>. The transfer direction is defined by <to_buffer>.
 * It is called by the elf_load_process() function.
 ******************************************************************************************
 * @ to_buffer : mapper -> buffer if true / buffer -> mapper if false. 
 * @ file_xp   : extended pointer on the remote file descriptor.
 * @ buffer_xp : user space pointer on buffer (can be physically distributed).
 * @ size      : requested number of bytes from offset.
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_kernel_move( bool_t   to_buffer,
                         xptr_t   file_xp, 
                         xptr_t   buffer_xp,
                         uint32_t size );

/****************************************************************************************** 
 * This function allocates a vfs_file_t structure in the cluster containing the inode
 * identified by the <root_xp> & <path> arguments.
 * It initializes it, register it in the reference process fd_array identified by the
 * <process_xp> argument, and returns both the extended pointer on the file descriptor,
 * and the allocated index in the <file_xp> and <file_id> buffers. 
 * The pathname can be relative to current directory or absolute.
 * If the inode does not exist in the inode cache, it try to find the file on the IOC
 * device, and creates an inode on a pseudo randomly selected cluster if found.
 * It the requested file does not exist on device, it creates a new inode if the
 * O_CREAT flag is set, and return an error otherwise. 
 *
 * WARNING : this function takes & releases the remote_rwlock protecting the Inode Tree.
 ****************************************************************************************** 
 * @ root_xp     : extended pointer on path root inode.
 * @ path        : file pathname (absolute or relative to current directory).
 * @ process_xp  : extended pointer on client reference process.
 * @ flags       : defined in vfs_file_t structure.
 * @ mode        : access rights (as defined by chmod).
 * @ file_xp     : [out] buffer for extended pointer on created remote file descriptor.
 * @ file_id     : [out] buffer for created file descriptor index in reference fd_array.
 * @ return 0 if success / return non-zero if error.
 *****************************************************************************************/
error_t vfs_open( xptr_t             root_xp,
                          char             * path,
                  xptr_t             process_xp,
                          uint32_t           flags,
                  uint32_t           mode,
                          xptr_t           * file_xp,
                          uint32_t         * file_id );

/****************************************************************************************** 
 * This function set a new value in the offset of the open file descriptor <file_xp>.
 * This value depends on the <whence> argument:
 * - if VFS_SEEK_SET, new value is <offset>
 * - if VFS_SEEK_CUR, new value is current_offset + <offset>
 * - if VFS_SEEK_END, new value is file_size + <offset>
 ****************************************************************************************** 
 * @ file_xp   : extended pointer on the remote open file descriptor.
 * @ offset    : local pointer on source kernel buffer.
 * @ whence    : VFS_SEEK_SET / VFS_SEEK_CUR / VFS_SEEK_END.
 * @ new_offset : [out] buffer for new offset value.
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_lseek( xptr_t     file_xp,
                   uint32_t   offset,
                   uint32_t   whence, 
                   uint32_t * new_offset );

/****************************************************************************************** 
 * This function close the - non-replicated - file descriptor identified by the <file_xp>
 * and <file_id> arguments. The <file_id> is required to reset the fd_array[] slot.
 * It can be called by a thread running in any cluster, and executes the following actions:
 * 1) It access the block device to updates all dirty pages from the mapper associated 
 *    to the file, and removes these pages from the dirty list, using an RPC if required.
 * 2) It updates the file size in all parent directory mapper(s), and update the modified
 *    pages on the block device, using RPCs if required.
 * 3) All entries in the fd_array copies are directly reset by the calling thread,
 *    using remote accesses. 
 * 4) The memory allocated to file descriptor in cluster containing the inode is released,
 *    using an RPC if cluster containing the file descriptor is remote.
 ******************************************************************************************
 * @ file_xp     : extended pointer on the file descriptor.
 * @ file_id     : file descriptor index in fd_array[].
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_close( xptr_t    file_xp,
                   uint32_t  file_id );

/****************************************************************************************** 
 * This function is called by the kernel to create in the file system a new directory 
 * identified by the <root_xp> & <path> arguments, with the access permission defined
 * by the <rights> argument. All nodes in the path - but the last -  must exist.
 *
 * WARNING : this function takes & releases the remote_rwlock protecting the Inode Tree.
 ******************************************************************************************
 * @ root_xp : extended pointer on path root inode (any inode in Inode Tree).
 * @ path    : pathname (absolute or relative to current directory).
 * @ rights  : access rights.
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_mkdir( xptr_t   root_xp,
                   char   * path,
                   uint32_t rights );

/****************************************************************************************** 
 * This function is called by the kernel to create in the file system a new directory 
 * entry identified by the <new_root_xp> & <new_path> arguments, to be linked to an
 * existing inode, identified by the  <old_root_xp> & <old_path> arguments.
 * If the link is successful, the link count of the target inode is incremented.
 * The <new_path> and <old_path> share equal access rights to the underlying inode.
 * Both the IOC device and the Inode Tree are modified. 
 $
 * TODO This function should handle any type of node, but the current implementation
 * handles only the FILE and DIR types.
 *
 * WARNING : this function takes & releases the remote_rwlock protecting the Inode Tree.
 ******************************************************************************************
 * @ old_root_xp : extended pointer on old path root inode (any inode in Inode Tree).
 * @ old_path    : old pathname (absolute or relative to current directory).
 * @ nld_root_xp : extended pointer on new path root inode (any inode in Inode Tree).
 * @ new_path    : new pathname (absolute or relative to current directory).
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_link( xptr_t   old_root_xp,
                  char   * old_path,
                  xptr_t   new_root_xp,
                  char   * new_path );

/****************************************************************************************** 
 * This function is called by the kernel to remove from the file system a directory entry
 * identified by the  <root_xp> & <path> arguments.
 * The link count of the target node is decremented.
 * If the removed link is the last, the target inode is deleted.
 * Both the IOC device and the Inode Tree are modified. 
 *
 * TODO This function should handle any type of node, but the current implementation
 * handles only only the FILE and DIR types.
 *
 * WARNING : this function takes & releases the remote_rwlock protecting the Inode Tree.
 ******************************************************************************************
 * @ root_xp  : extended pointer on root inode (can be any inode in Inode Tree).
 * @ path     : pathname (absolute or relative to current directory).
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_unlink( xptr_t   root_xp,
                    char   * path );

/****************************************************************************************** 
 * This function returns, in the structure pointed by the <st> pointer, various
 * informations on the inode identified by the <root_inode_xp> and <patname> arguments.
 *
 * TODO : only partially implemented yet (only size and inum fields).
 *
 * WARNING : this function takes & releases the remote_rwlock protecting the Inode Tree.
 ******************************************************************************************
 * @ root_xp    : extended pointer on path root inode (any inode in Inode Tree)
 * @ pathname   : pathname to target inode.
 * @ st         : local pointer on the stat structure in kernel space.
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_stat( xptr_t        root_xp,
                  char        * pathname,
                  struct stat * st );

/****************************************************************************************** 
 * This function  creates a new directory as defined by the <root_xp> & <path> arguments.
 * TODO not implemented yet...
 ******************************************************************************************
 * @ root_xp  : extended pointer on the path root directory.
 * @ path     : pathname (absolute or relative to CWD).                     
 * @ mode     : access rights (as defined by chmod)
 * @ returns 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_mkdir( xptr_t     root_xp,
                   char     * path,
                   uint32_t   mode );

/****************************************************************************************** 
 * This function makes the directory identified by the <root_xp and <path> arguments 
 * to become the working directory for the calling process.
 ****************************************************************************************** 
 * @ root_xp  : extended pointer on the path root directory.
 * @ path     : pathname (absolute or relative to CWD).
 * return 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_chdir( xptr_t   root_xp,
                   char   * path );

/****************************************************************************************** 
 * This function change the access rigths for the file/directory identified by the 
 * <root_xp> and <path> arguments as defined by the <mode> argument value.
 ****************************************************************************************** 
 * @ root_xp  : extended pointer on the path root directory.
 * @ path     : pathname (absolute or relative to CWD).
 * @ mode     : access rights
 * return 0 if success / -1 if error.
 *****************************************************************************************/
error_t vfs_chmod( xptr_t        root_xp,
                   char        * path,
                   uint32_t      mode );

/****************************************************************************************** 
 * This function creates a named FIFO file.
 * TODO not implemented yet                                                          
 ****************************************************************************************** 
 * @ root_xp  : extended pointer on the path root directory.
 * @ path     : pathname (absolute or relative to CWD).
 * @ mode     : access rights new value.
 *****************************************************************************************/
error_t vfs_mkfifo( xptr_t       root_xp,
                    char       * path,
                    uint32_t     mode );





/******************************************************************************************
 *        These functions define the VFS "FS" API to a specific File System
 *****************************************************************************************/

/******************************************************************************************
 * This function makes the I/O operation to move one page identified by the <page_xp>
 * argument to/from the IOC device from/to the mapper, as defined by the <cmd_type>.
 * Depending on the file system type, it calls the proper, FS specific function.
 * It is used in case of MISS on the mapper, or when a dirty page in the mapper must
 * be updated in the File System.
 * The mapper pointer is obtained from the page descriptor.
 * It can be executed by any thread running in any cluster.
 * This function does NOT take any lock.
 ******************************************************************************************
 * @ page_xp   : extended pointer on page descriptor (for mapper and page_id).
 * @ cmd_type  : IOC_READ / IOC_WRITE / IOC_SYNC_READ / IOC_SYNC_WRITE
 * @ returns 0 if success / return -1 if device access failure.
 *****************************************************************************************/
error_t vfs_fs_move_page( xptr_t      page_xp,
                          cmd_type_t  cmd_type );

/******************************************************************************************
 * This function updates the mapper associated to a directory inode identified by the 
 * <parent> argument, to add a new entry identified by the <dentry> argument. 
 * The directory inode descriptor and the dentry descriptor are in the same cluster.
 * Depending on the file system type, it calls the proper, FS specific function.
 * It also updates the dentry descriptor and/or the inode descriptor extensions
 * as required by the specific file system type.
 * Finally, it synchronously updates the parent directory on IOC device.
 *
 * It must be executed by a thread running in the cluster containing the parent directory.
 * It can be the RPC_VFS_FS_ADD_DENTRY. This function does NOT take any lock.
 ******************************************************************************************
 * @ parent  : local pointer on parent (directory) inode.
 * @ dentry  : local pointer on dentry containing name.
 * @ return 0 if success / return -1 if device access failure.
 *****************************************************************************************/
error_t vfs_fs_add_dentry( vfs_inode_t  * parent,
                           vfs_dentry_t * dentry );

/******************************************************************************************
 * This function updates the mapper associated to a directory inode identified by the 
 * <parent> argument, to remove an entry identified by the <dentry> argument. 
 * The directory inode descriptor and the dentry descriptor are in the same cluster.
 * Depending on the file system type, it calls the proper, FS specific function.
 * Finally, it synchronously updates the parent directory on IOC device.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It must be executed by a thread running in the cluster containing the parent directory.
 * It can be the RPC_VFS_FS_REMOVE_DENTRY. This function does NOT take any lock.
 ******************************************************************************************
 * @ parent  : local pointer on parent (directory) inode.
 * @ dentry  : local pointer on dentry containing name.
 * @ return 0 if success / return -1 if device access failure.
 *****************************************************************************************/
error_t vfs_fs_remove_dentry( vfs_inode_t  * parent,
                              vfs_dentry_t * dentry );

/******************************************************************************************
 * This function scan the mapper of an an existing parent inode directory, identified by
 * the <parent> argument to find a directory entry identified by the <name> argument,
 * and updates both the child inode descriptor, identified by the <child_xp> argument,
 * and the associated dentry descriptor :
 * - It set the "size", "type", and "extend" fields in inode descriptor.
 * - It set the "extend" field in dentry descriptor.
 * It is called by the vfs_lookup() function in case of miss.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It must be called by a thread running in the cluster containing the parent inode.
 * It can be the RPC_VFS_FS_NEW_DENTRY. This function does NOT take any lock.
 ******************************************************************************************
 * @ parent    : local pointer on parent inode (directory).
 * @ name      : child name.
 * @ child_xp  : extended pointer on remote child inode (file or directory)
 * @ return 0 if success / return -1 if dentry not found.
 *****************************************************************************************/
error_t vfs_fs_new_dentry( vfs_inode_t * parent,
                           char        * name,
                           xptr_t        child_xp );

/******************************************************************************************
 * This function scan the mapper of an an existing inode directory, identified by
 * the <inode> argument, to find a directory entry identified by the <dentry> argument,
 * and update the size for this directory entry in mapper, as defined by <size>.
 * The parent directory on device is synchronously updated.
 * It is called by the vfs_close() function.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It must be called by a thread running in the cluster containing the parent inode.
 * It can be the RPC_VFS_FS_UPDATE_DENTRY. This function does NOT take any lock.
 ******************************************************************************************
 * @ parent    : local pointer on parent inode (directory).
 * @ dentry    : local pointer on dentry.
 * @ size      : new size value (bytes).
 * @ return 0 if success / return ENOENT if not found.
 *****************************************************************************************/
error_t vfs_fs_update_dentry( vfs_inode_t  * inode,
                              vfs_dentry_t * dentry,
                              uint32_t       size );

/******************************************************************************************
 * This function scan the mapper of an an existing parent inode directory, identified by
 * the <inode> argument and copy up to <max_dirent> valid dentries to a 
 * local dirent array, defined by the <array> argument. The <min_dentry> argument defines
 * the index of the first dentry to copied to the target dirent array.
 * This function returns in the <entries> buffer the number of dentries actually written,
 * and signals in the <done> buffer when the last valid entry has been found.
 * If the <detailed> argument is true, a dentry/inode couple that does not exist in
 * the Inode Tree is dynamically created, and all dirent fiels are documented in the
 * dirent array. Otherwise, only the dentry name is documented.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It must be called by a thread running in the cluster containing the parent inode.
 * This function does NOT take any lock.
 ******************************************************************************************
 * @ inode      : [in]  local pointer on directory inode.
 * @ array      : [in]  local pointer on array of dirents.
 * @ max_dirent : [in]  max number of slots in dirent array.
 * @ min_dentry : [in]  index of first dentry to be copied into array.
 * @ detailed   : [in]  dynamic inode creation if true.
 * @ entries    : [out] number of dentries actually copied into array.
 * @ done       : [out] Boolean true when last entry found.
 * @ return 0 if success / return -1 if failure.
 *****************************************************************************************/
error_t vfs_fs_get_user_dir( vfs_inode_t   * inode,
                             struct dirent * array,
                             uint32_t        max_dirent,
                             uint32_t        min_dentry,
                             bool_t          detailed,
                             uint32_t      * entries,
                             bool_t        * done );
 
/*****************************************************************************************
 * This function  updates the FS on the IOC device for a given inode identified by 
 * the <inode> argument. It scan all pages registered in the associated mapper,
 * and copies from mapper to device each page marked as dirty.
 * WARNING : The target <inode> cannot be a directory, because all modifications in a 
 * directory are synchronously done on the IOC device by the two vfs_fs_add_dentry() 
 * and vfs_fs_remove_dentry() functions.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It must be called by a thread running in the inode cluster.
 *****************************************************************************************
 * @ inode   : local pointer on inode.
 * @ return 0 if success / return EIO if failure during device access. 
 ****************************************************************************************/
error_t vfs_fs_sync_inode( struct vfs_inode_s * inode );

/*****************************************************************************************
 * This function updates the FS defined by the <fs_type> argument on the IOC device
 * for the FAT itself. It scan all clusters registered in the FAT mapper, and copies 
 * to device each page marked as dirty.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It can be called by a thread running in any cluster.
 *****************************************************************************************
 * @ fs_type   : specific file system type.
 * @ return 0 if success / return EIO if failure during device access.
 ****************************************************************************************/
error_t vfs_fs_sync_fat( vfs_fs_type_t fs_type );

/*****************************************************************************************
 * This function updates the free clusters info on the IOC device for the FS defined
 * by the <fs_type> argument. 
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It can be called by a thread running in any cluster.
 *****************************************************************************************
 * @ fs_type   : specific file system type.
 * @ return 0 if success / return EIO if failure during device access.
 ****************************************************************************************/
error_t vfs_fs_sync_free_info( vfs_fs_type_t fs_type );

/******************************************************************************************
 * This function allocates a free cluster from the FS identified by the <fs_type>
 * argument. It updates the selected FS File Allocation Table.
 *
 * Depending on the file system type, it calls the relevant, FS specific function.
 * It can be called by a thread running in any cluster.
 ******************************************************************************************
 * @ fs_type   : [in]  File System type.
 * @ cluster   : [out] cluster index in File system.
 * @ return 0 if success / return -1 if no free cluster
 *****************************************************************************************/
error_t vfs_fs_cluster_alloc( uint32_t   fs_type,
                              uint32_t * cluster );

/******************************************************************************************
 * This function makes all I/O operations required to release all clusters allocated
 * on IOC device to a given inode, identified by the <inode_xp> argument.
 * Depending on the file system type, it calls the proper, FS specific function.
 * It is called by the vfs_unlink() function.
 * It can be executed by a thread running in any cluster.
 * This function does NOT take any lock.
 ******************************************************************************************
 * @ inode_xp  : extended pointer on inode.
 * @ return 0 if success / return -1 if device access failure.
 *****************************************************************************************/
error_t vfs_fs_release_inode( xptr_t  inode_xp ); 


#endif  /* _VFS_H_ */