source: trunk/kernel/kern/process.h @ 423

/*
 * process.h - process related management functions
 *
 * Authors  Ghassan Almaless (2008,2009,2010,2011,2012)
 *          Mohamed Lamine Karaoui (2015)
 *          Alain Greiner (2016,2017)
 *
 * 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 _PROCESS_H_
#define _PROCESS_H_

#include <kernel_config.h>
#include <errno.h>
#include <hal_types.h>
#include <list.h>
#include <xlist.h>
#include <bits.h>
#include <spinlock.h>
#include <hal_atomic.h>
#include <vmm.h>
#include <signal.h>
#include <cluster.h>
#include <vfs.h>

/****  Forward declarations  ****/

struct thread_s;

/*********************************************************************************************
 * These macros are used to compose or decompose global process identifier (PID)
 * to or from cluster identifier / local process index (CXY , LPID)
 ********************************************************************************************/

#define LPID_FROM_PID( pid )       (lpid_t)(pid & 0x0000FFFF)
#define CXY_FROM_PID( pid )        (cxy_t)(pid >> 16)
#define PID( cxy , lpid )          (pid_t)((cxy << 16) | lpid )

/*********************************************************************************************
 * This enum defines the actions that can be executed by the process_signal() function.
 ********************************************************************************************/

enum process_sigactions
{
    BLOCK_ALL_THREADS    = 11,
    UNBLOCK_ALL_THREADS  = 22,
    DELETE_ALL_THREADS   = 33, 
};

/*********************************************************************************************
 * This structure defines an array of extended pointers on the open file descriptors
 * for a given process. We use an extended pointer because the open file descriptor
 * is always stored in the same cluster as the inode associated to the file.
 * A free entry in this array contains the XPTR_NULL value.
 * The array size is defined by a the CONFIG_PROCESS_FILE_MAX_NR parameter.
 * All modifications (open/close) in this structure must be done by the reference cluster,
 * and reported in process copies.
 ********************************************************************************************/

typedef struct fd_array_s
{
        remote_spinlock_t lock;                               /*! lock protecting fd_array      */
    uint32_t          current;                            /*! current number of open files  */
        xptr_t            array[CONFIG_PROCESS_FILE_MAX_NR];  /*! xptr on open file descriptors */
}
fd_array_t;

/*********************************************************************************************
 * This structure defines a process descriptor.
 * A process is identified by a unique PID (process identifier):
 * - The PID 16 LSB bits contain the LPID (Local Process Index)
 * - The PID 16 MSB bits contain the owner cluster CXY.
 * In each cluster, the process manager allocates  the LPID values for the process that
 * are owned by this cluster.
 * The process descriptor is replicated in all clusters containing at least one thread 
 * of the PID process, with the following rules :
 * 1) The <pid>, <ppid>, <ref_xp>, <vfs_root_xp>, <vfs_bin_xp>  fields are defined
 *    in all process descriptor copies.
 * 2) The <vfs_cwd_xp> and associated <cwd_lock>, that can be dynamically modified,
 *    are only defined in the reference process descriptor.
 * 2) The <vmm>, containing the VSL (list of registered vsegs), and the GPT (generic
 *    page table), are only complete in the reference process cluster, other copies 
 *    are actually use as read-only caches.
 * 3) the <fd_array>, containing extended pointers on the open file descriptors, is only
 *    complete in the reference process cluster, other copies are read-only caches.
 * 4) The <sem_root>, <mutex_root>, <barrier_root>, <condvar_root>, and the associated
 *    <sync_lock>, that are dynamically allocated, are only defined in the reference cluster.
 * 5) The <children_root>, and <children_nr> fields are only defined in the reference
 *    cluster, and are undefined in other clusters.
 * 6) The <brothers_list>, <local_list>, <copies_list>, <th_tbl>, <th_nr>, <th_lock> fields
 *    are defined in all process descriptors copies.
 ********************************************************************************************/

typedef struct process_s
{
        vmm_t             vmm;              /*! embedded virtual memory manager                 */

        fd_array_t        fd_array;         /*! embedded open file descriptors array            */

        xptr_t            vfs_root_xp;      /*! extended pointer on current VFS root inode      */
        xptr_t            vfs_bin_xp;       /*! extended pointer on .elf file inode             */
        pid_t             pid;              /*! process identifier                              */
        pid_t             ppid;             /*! parent process identifier                       */
    xptr_t            ref_xp;           /*! extended pointer on reference process           */

        xptr_t            vfs_cwd_xp;       /*! extended pointer on current working dir inode   */
        remote_rwlock_t   cwd_lock;         /*! lock protecting working directory changes       */

        xlist_entry_t     children_root;    /*! root of the children process xlist              */
    uint32_t          children_nr;      /*! number of children processes                    */

        xlist_entry_t     brothers_list;    /*! member of list of children of same parent       */
    xlist_entry_t     local_list;       /*! member of list of process in same cluster       */
    xlist_entry_t     copies_list;      /*! member of list of copies of same process        */

        spinlock_t        th_lock;          /*! lock protecting th_tbl[] concurrent access      */
        uint32_t          th_nr;            /*! number of threads in this cluster               */
        struct thread_s * th_tbl[CONFIG_THREAD_MAX_PER_CLUSTER]; /*! pointers on local threads  */

    xlist_entry_t     sem_root;         /*! root of the process semaphore list              */
    xlist_entry_t     mutex_root;       /*! root of the process mutex list                  */
    xlist_entry_t     barrier_root;     /*! root of the process barrier list                */
    xlist_entry_t     condvar_root;     /*! root of the process condvar list                */

    remote_spinlock_t sync_lock;        /*! lock protecting sem,mutex,barrier,condvar lists */
}
process_t;

/*********************************************************************************************
 * This structure defines the information required by the process_make_exec() function
 * to create a new reference process descriptor, and the associated main thread.
 ********************************************************************************************/

typedef struct exec_info_s
{
    pid_t              pid;            /*! process identifier (both parent and child)       */

    char               path[CONFIG_VFS_MAX_PATH_LENGTH];   /*!  .elf file path              */

    char            ** args_pointers;  /*! physical base address of array of pointers       */
    char             * args_buf_base;  /*! physical base address of kernel args buffer      */
    uint32_t           args_nr;        /*! actual number of arguments                       */

    char            ** envs_pointers;  /*! physical base address of array of pointers       */
    char             * envs_buf_base;  /*! physical base address of kernel args buffer      */
    char             * envs_buf_free;  /*! physical address of first free slot in envs_buf  */
    uint32_t           envs_nr;        /*! actual number of environment variables           */
}
exec_info_t;

/***************   Process Descriptor Operations    *****************************************/

/*********************************************************************************************
 * This function allocates memory in local cluster for a process descriptor.
 *********************************************************************************************
 * @ returns pointer on process descriptor if success / return NULL if failure
 ********************************************************************************************/
process_t * process_alloc();

/*********************************************************************************************
 * This function releases memory in local cluster for a process descriptor.
 *********************************************************************************************
 * @ process      : pointer on process descriptor to release.
 ********************************************************************************************/
void process_free( process_t * process );

/*********************************************************************************************
 * This function allocates memory and initializes the "process_init" descriptor and the
 * associated "thread_init" descriptor. It is called once at the end of the kernel 
 * initialisation procedure, by the kernel process in cluster_IO.
 * The "process_init" is the first user process, and all other user processes will be forked
 * from this process. The code executed by "process_init" is stored in a .elf file, whose
 * pathname is defined by the CONFIG_PROCESS_INIT_PATH configuration variable. 
 * The process_init streams are defined  by the CONFIG_INIT_[STDIN/STDOUT/STDERR] variables.
 * Its local process identifier is 1, and parent process is the local kernel process_zero.
 ********************************************************************************************/
void process_init_create();

/*********************************************************************************************
 * This function initialize, in each cluster, the kernel "process_zero", that is the owner
 * of all kernel threads in a given cluster. It is called by the kernel_init() function.
 * Both the PID and PPID fields are set to zero, and the ref_xp is the local process_zero.
 * The th_tbl[] is initialized as empty.
 *********************************************************************************************
 * @ process      : [in] pointer on local process descriptor to initialize.
 ********************************************************************************************/
void process_zero_init( process_t * process );

/*********************************************************************************************
 * This function initializes a local, reference user process descriptor from another process
 * descriptor, defined by the <model_xp> argument. The <process> descriptor, the <pid>, and
 * the <ppid> arguments must be previously defined by the caller.
 * It can be called by two functions, depending on the process type:
 * 1) if "process" is the "process_init", the parent is the kernel process. It is
 *    called once, by the process_init_create() function in cluster[xmax-1][ymax-1].
 * 2) if the caller is the process_make_fork() function, the model is generally a remote
 *    process, that is also the parent process.

 * 3) if the caller is the process_make_exec() function, the model is always a local process,
 *    and the parent is the parent of the model process. DEPRECATED [AG]

 * The following fields are initialised (for all process but process_zero).
 * - It set the pid / ppid / ref_xp fields.
 * - It initializes the VMM (register the kentry, args, envs vsegs in VSL)
 * - It initializes the FDT, defining the three pseudo files STDIN / STDOUT / STDERR.
 * - It set the root_xp, bin_xp, cwd_xp fields.
 * - It reset the children list as empty, but does NOT register it in parent children list.
 * - It reset the TH_TBL list of threads as empty.
 * - It reset the semaphore / mutex / barrier / condvar lists as empty.
 * - It registers the process in the local_list, rooted in the local cluster manager.
 * - It registers the process in the copies_list, rooted in the owner cluster manager.
 * - It registers the process extended pointer in the local pref_tbl[] array.
 *********************************************************************************************
 * @ process      : [in] pointer on local process descriptor to initialize.
 * @ pid          : [in] process identifier.
 * @ ppid         : [in] parent process identifier.
 * @ model_xp     : [in] extended pointer on model process descriptor (local or remote).
 ********************************************************************************************/
void process_reference_init( process_t * process,
                             pid_t       pid,
                             pid_t       ppid,
                             xptr_t      model_xp );

/*********************************************************************************************
 * This function initializes a copy process descriptor, in the local cluster,
 * from information defined in the reference remote process descriptor.
 *********************************************************************************************
 * @ process              : [in] local pointer on process descriptor to initialize.
 * @ reference_process_xp : [in] extended pointer on reference process descriptor.
 * @ return 0 if success / return ENOMEM if failure
 ********************************************************************************************/
error_t process_copy_init( process_t * local_process,
                           xptr_t      reference_process_xp );

/*********************************************************************************************
 * This function releases all memory allocated for a process descriptor in the local cluster,
 * including memory allocated for embedded substructures (fd_array, vmm, etc).
 * The local th_tbl[] array must be empty.
 *********************************************************************************************
 * @ process     : pointer on the process descriptor.
 ********************************************************************************************/
void process_destroy( process_t * process );

/*********************************************************************************************
 * This function returns a printable string defining the action for process_signa().
 *********************************************************************************************
 * @ action_type   : BLOCK_ALL_THREADS / UNBLOCK_ALL_THREADS / DELETE_ALL_THREADS
 * @ return a string pointer.
 ********************************************************************************************/
char * process_action_str( uint32_t action_type );

/*********************************************************************************************
 * This function allows a client thread running in the owner cluster of a process identified
 * by the <process> argument to block, unblock or delete all threads of the target process,
 * depending on the <action_type> argument.  The scenario is the following:
 * - It uses the multicast, non blocking rpc_process_sigaction_client() function to send
 *   parallel requests to all remote clusters containing a process copy. Then it blocks
 $   and deschedule to wait completion of these parrallel requests.
 * - In each remote cluster, the rpc_process_sigaction_server() function, calls directly
 *   the relevant process_block(), process_unblock(), or process_delete() function, and
 *   decrement the responses counter to signal completion. The last server unblock
 *   the client thread.
 * - Finally, the client thread calls directly the process_block(), process_unblock(), or
 *   process_delete() function in the owner cluster.
 * It is used by the sys_kill() & sys_exit() functions to handle the "kill" & "exit" syscalls.
 * It is also used by the process_make_exec() function to handle the "exec" syscall.
 * WARNING : the DELETE and the BLOCK actions are NOT executed on the client thread.
 *********************************************************************************************
 * @ process     : pointer on the process descriptor in owner cluster.
 * @ action_type : BLOCK_ALL_THREADS / UNBLOCK_ALL_THREADS / DELETE_ALL_THREADS
 ********************************************************************************************/
void process_sigaction( process_t * process,
                        uint32_t    action_type );

/*********************************************************************************************
 * This function blocks all threads of a given user process in a given cluster.
 * It loops on all local threads of the process, set the THREAD_BLOCKED_GLOBAL bit, 
 * and request the relevant schedulers to acknowledge the blocking, using IPI if required. 
 * The threads are not detached from the scheduler, and not detached from the local process.
 * This function returns only when all blockable threads in cluster are actually blocked.
 * WARNING : the client thread defined by the <client_xp> argument is NOT blocked.
 *********************************************************************************************
 * @ process     : pointer on the target process descriptor.
 * @ client_xp   : extended pointer on the client thread, that should not be blocked.
 ********************************************************************************************/
void process_block( process_t * process,
                    xptr_t      client_xp );

/*********************************************************************************************
 * This function unblocks all threads of a given user process in a given cluster.
 *********************************************************************************************
 * @ process     : pointer on the process descriptor.
 ********************************************************************************************/
void process_unblock( process_t * process );

/*********************************************************************************************
 * This function delete all threads, of a given user process in a given cluster.
 * It loops on all local threads of the process, and set the THREAD_FLAG_REQ_DELETE bit.
 * For each marked thread, the following actions will be done by the scheduler at the next
 * scheduling point: 
 * - the thread will be detached from the scheduler.
 * - the thread will be detached from the local process descriptor.
 * - the thread will be detached from parent if required.
 * - the memory allocated to the thread descriptor is released.
 * - the memory allocated to the process descriptor is released, if it is the last thread.
 * WARNING : the client thread defined by the <client_xp> argument is NOT deleted.
 *********************************************************************************************
 * @ process     : pointer on the process descriptor.
 * @ client_xp   : extended pointer on the client thread, that should not be deleted.
 ********************************************************************************************/
void process_delete( process_t * process,
                     xptr_t      client_xp );

/*********************************************************************************************
 * This function returns a pointer on the local copy of a process identified by its PID.
 * If this local copy does not exist yet, it is dynamically created, from the reference
 * process descriptor, registered in the global copies_list, and registered in the local_list.
 * This function is used by the thread_user_create() function.
 *********************************************************************************************
 * @ pid     : searched process identifier.
 * @ returns pointer on the local process descriptor if success / returns NULL if failure.
 ********************************************************************************************/
process_t * process_get_local_copy( pid_t pid );

/*********************************************************************************************
 * This function implements the "exec" system call, and is called by the sys_exec() function.
 * The "new" process keep the "old" process PID and PPID, all open files, and env variables,
 * the vfs_root and vfs_cwd, but build a brand new memory image (new VMM from the new .elf).
 * It actually creates a "new" reference process descriptor, and copies all relevant 
 * information from the "old" process descriptor to the "new" process descriptor.
 * It completes the "new" process descriptor, from information found in the <exec_info> 
 * structure (defined in the process.h file), that must be built by the caller.
 * It creates and initializes the associated main thread. It finally destroys all copies 
 * of the "old" process in all clusters, and destroys all old associated threads.
 * It is executed in the local cluster, that becomes both the "owner" and the "reference"
 * cluster for the "new" process.
 *********************************************************************************************
 * @ exec_info   : [in]  pointer on the exec_info structure.
 * @ return 0 if success / return non-zero if error.
 ********************************************************************************************/
error_t process_make_exec( exec_info_t * exec_info );

/*********************************************************************************************
 * This function implements the "fork" system call, and is called by the sys_fork() function. 
 * It allocates memory and initializes a new "child" process descriptor, and the
 * associated "child" thread descriptor in the local cluster. This function can involve
 * up to three different clusters :
 * - the local (child) cluster can be any cluster defined by the sys_fork function.
 * - the parent cluster must be the reference cluster for the parent process.
 * - the client cluster containing the thread requesting the fork can be any cluster.
 * The new "child" process descriptor is initialised from informations found in the "parent"
 * reference process descriptor, containing the complete process description.
 * The new "child" thread descriptor is initialised from informations found in the "parent"
 * thread descriptor.
 *********************************************************************************************
 * @ parent_process_xp  : extended pointer on the reference parent process.
 * @ parent_thread_xp   : extended pointer on the parent thread requesting the fork.
 * @ child_pid          : [out] child process identifier.
 * @ child_thread_ptr   : [out] local pointer on child thread in target cluster.
 * @ return 0 if success / return non-zero if error.
 ********************************************************************************************/
error_t process_make_fork(  xptr_t             parent_process_xp,
                            xptr_t             parent_thread_xp,
                            pid_t            * child_pid, 
                            struct thread_s ** child_thread_ptr );

/*********************************************************************************************
 * This function implement the "exit" system call, and is called by the sys_exit() function.
 * It must be executed by a thread running in the calling process owner cluster.
 * It uses twice the multicast RPC_PROCESS_SIGNAL to first block all process threads
 * in all clusters, and then delete all threads and process descriptors.
 *********************************************************************************************
 * @ pid      : process identifier.
 * @ status   : exit return value.
 ********************************************************************************************/
void process_make_exit( pid_t       pid,
                        uint32_t    status );

/*********************************************************************************************
 * This function implement the "kill" system call, and is called by the sys_kill() function.
 * It must be executed by a thread running in the target process owner cluster.
 * Only the SIGKILL, SIGSTOP, and SIGCONT signals are supported.
 * User defined handlers are not supported.
 * It uses once or twice the multicast RPC_PROCESS_SIGNAL to block, unblock or delete
 * all process threads in all clusters, and then delete process descriptors.
 *********************************************************************************************
 * @ pid     : process identifier.
 * @ sig_id  : signal type.
 ********************************************************************************************/
void process_make_kill( pid_t     pid,
                        uint32_t  sig_id );


/********************   File Management Operations   ****************************************/

/*********************************************************************************************
 * This function initializes all entries of the local fd_array as empty.
 *********************************************************************************************
 * @ process  : pointer on the local process descriptor.
 ********************************************************************************************/
void process_fd_init( process_t * process );

/*********************************************************************************************
 * This function uses as many remote accesses as required, to reset an entry in fd_array[],
 * in all clusters containing a copy. The entry is identified by the <fdid> argument.
 * This function must be executed by a thread running reference cluster, that contains
 * the complete list of process descriptors copies.
 *********************************************************************************************
 * @ process  : pointer on the local process descriptor.
 * @ fdid     : file descriptor index in the fd_array.
 ********************************************************************************************/
void process_fd_remove( process_t * process,
                        uint32_t    fdid );

/*********************************************************************************************
 * This function returns an extended pointer on a file descriptor identified by its index
 * in fd_array. It can be called by any thread running in any cluster.
 * It accesses first the local process descriptor. In case of local miss, it uses remote
 * access to access the reference process descriptor.
 * It updates the local fd_array when the file descriptor exists in reference cluster.
 * The file descriptor refcount is not incremented.
 *********************************************************************************************
 * @ process  : pointer on the local process descriptor.
 * @ fdid     : file descriptor index in the fd_array.
 * @ return extended pointer on file descriptor if success / return XPTR_NULL if not found.
 ********************************************************************************************/
xptr_t process_fd_get_xptr( process_t * process,
                            uint32_t    fdid );

/*********************************************************************************************
 * This function checks the number of open files for a given process.
 * It can be called by any thread in any cluster, because it uses portable remote access
 * primitives to access the reference process descriptor.
 *********************************************************************************************
 * @ returns true if file descriptor array full.
 ********************************************************************************************/
bool_t process_fd_array_full();

/*********************************************************************************************
 * This function allocates a free slot in the fd_array of the reference process,
 * register the <file_xp> argument in the allocated slot, and return the slot index.
 * It can be called by any thread in any cluster, because it uses portable remote access
 * primitives to access the reference process descriptor.
 *********************************************************************************************
 * @ file_xp  : extended pointer on the file descriptor to be registered.
 * @ fdid     : [out] buffer for fd_array slot index.
 * @ return 0 if success / return EMFILE if array full.
 ********************************************************************************************/
error_t process_fd_register( process_t * process,
                             xptr_t      file_xp,
                             uint32_t  * fdid );

/*********************************************************************************************
 * This function copies all non-zero entries (other than the three first stdin/stdout/stderr)
 * from a remote <src_xp> fd_array, embedded in a process descriptor, to another remote 
 * <dst_xp> fd_array, embedded in another process descriptor. 
 * The calling thread can be running in any cluster.
 * It takes the remote lock protecting the <src_xp> fd_array during the copy.
 * For each involved file descriptor, the refcount is incremented.
 *********************************************************************************************
 * @ dst_xp   : extended pointer on the destination fd_array_t.
 * @ src_xp   : extended pointer on the source fd_array_t.
 ********************************************************************************************/
void process_fd_remote_copy( xptr_t dst_xp,
                             xptr_t src_xp );



/********************   Thread Related Operations   *****************************************/

/*********************************************************************************************
 * This function registers a new thread in the local process descriptor.
 * It checks that there is an available slot in the local th_tbl[] array,
 * allocates a new LTID, and registers the new thread in the th_tbl[].
 * WARNING : the lock protecting the th_tbl[] must be taken by the caller.
 *********************************************************************************************
 * @ process  : pointer on the local process descriptor.
 * @ thread   : pointer on new thread to be registered.
 * @ trdid    : [out] address of buffer for allocated trdid.
 * @ returns 0 if success / returns non zero if no slot available.
 ********************************************************************************************/
error_t process_register_thread( process_t       * process,
                                 struct thread_s * thread,
                                 trdid_t         * trdid );

/*********************************************************************************************
 * This function removes a thread registration from the local process descriptor.
 * WARNING : the lock protecting the th_tbl[] must be taken by the caller.
 *********************************************************************************************
 * @ thread   : local pointer on thread to be removed.
 ********************************************************************************************/
void process_remove_thread( struct thread_s * thread );



#endif  /* _PROCESS_H_ */