/** \file fishd.c

The universal variable server. fishd is automatically started by fish
if a fishd server isn't already running. fishd reads any saved
variables from ~/.fishd, and takes care of communication between fish
instances. When no clients are running, fishd will automatically shut
down and save.

\section fishd-commands Commands

Fishd works by sending and receiving commands. Each command is ended
with a newline. These are the commands supported by fishd:

<pre>set KEY:VALUE
set_export KEY:VALUE
</pre>

These commands update the value of a variable. The only difference
between the two is that <tt>set_export</tt>-variables should be
exported to children of the process using them. When sending messages,
all values below 32 or above 127 must be escaped using C-style
backslash escapes. This means that the over the wire protocol is
ASCII. However, any conforming reader must also accept non-ascii
characters and interpret them as UTF-8. Lines containing invalid UTF-8
escape sequences must be ignored entirely.

<pre>erase KEY
</pre>

Erase the variable with the specified name.

<pre>barrier
barrier_reply
</pre>

A \c barrier command will result in a barrier_reply being added to
the end of the senders queue of unsent messages. These commands are
used to synchronize clients, since once the reply for a barrier
message returns, the sender can know that any updates available at the
time the original barrier request was sent have been received.

*/

#include "config.h"


#include <stdio.h>
#include <stdlib.h>
#include <wchar.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <pwd.h>
#include <fcntl.h>

#ifdef HAVE_GETOPT_H
#include <getopt.h>
#endif

#include <errno.h>
#include <locale.h>
#include <signal.h>

#include "fallback.h"
#include "util.h"

#include "common.h"
#include "wutil.h"
#include "env_universal_common.h"
#include "path.h"
#include "print_help.h"

#ifndef HOST_NAME_MAX
/**
 Maximum length of hostname return. It is ok if this is too short,
 getting the actual hostname is not critical, so long as the string
 is unique in the filesystem namespace.
 */
#define HOST_NAME_MAX 255
#endif

/**
   Maximum length of socket filename
*/
#ifndef UNIX_PATH_MAX 
#define UNIX_PATH_MAX 100
#endif

/**
   Fallback if MSG_DONTWAIT isn't defined. That's actually prerry bad,
   and may lead to strange fishd behaviour, but at least it should
   work most of the time.
*/
#ifndef MSG_DONTWAIT
#define MSG_DONTWAIT 0
#endif

/**
   Small greeting to show that fishd is running
*/
#define GREETING "# Fish universal variable daemon\n"

/**
   Small not about not editing ~/.fishd manually. Inserted at the top of all .fishd files.
*/
#define SAVE_MSG "# This file is automatically generated by the fishd universal variable daemon.\n# Do NOT edit it directly, your changes will be overwritten.\n"

/**
   The name of the save file. The hostname is appended to this.
*/
#define FILE "fishd."

/**
   Maximum length of hostname. Longer hostnames are truncated
*/
#define HOSTNAME_LEN 32

/**
   The string to append to the socket name to name the lockfile
*/
#define LOCKPOSTFIX ".lock"

/**
   The timeout in seconds on the lockfile for critical section
*/
#define LOCKTIMEOUT 1

/**
   Getopt short switches for fishd
*/
#define GETOPT_STRING "hv"

/**
   The list of connections to clients
*/
static connection_t *conn;

/**
   The socket to accept new clients on
*/
static int sock;

/**
   Set to one when fishd should save and exit
*/
static int quit=0;

/**
   Constructs the fish socket filename
*/
static char *get_socket_filename()
{
	char *name;
	const char *dir = getenv( "FISHD_SOCKET_DIR" );
	char *uname = getenv( "USER" );

	if( dir == NULL )
	{
		dir = "/tmp";
	}

	if( uname == NULL )
	{
		struct passwd *pw;
		pw = getpwuid( getuid() );
		uname = strdup( pw->pw_name );
	}

	name = (char *)malloc( strlen(dir)+ strlen(uname)+ strlen(SOCK_FILENAME) + 2 );
	if( name == NULL )
	{
		wperror( L"get_socket_filename" );
		exit( EXIT_FAILURE );
	}
	strcpy( name, dir );
	strcat( name, "/" );
	strcat( name, SOCK_FILENAME );
	strcat( name, uname );

	if( strlen( name ) >= UNIX_PATH_MAX )
	{
		debug( 1, L"Filename too long: '%s'", name );
		exit( EXIT_FAILURE );
	}
	return name;
}

/**
   Signal handler for the term signal. 
*/
static void handle_term( int signal )
{
	quit=1;
}


/**
 Writes a pid_t in decimal representation to str.
 str must contain sufficient space.
 The conservatively approximate maximum number of characters a pid_t will 
 represent is given by: (int)(0.31 * sizeof(pid_t) + CHAR_BIT + 1)
 Returns the length of the string
 */
static int sprint_pid_t( pid_t pid, char *str )
{
	int len, i = 0;
	int dig;
    
	/* Store digits in reverse order into string */
	while( pid != 0 )
	{
		dig = pid % 10;
		str[i] = '0' + dig;
		pid = ( pid - dig ) / 10;
		i++;
	}
	len = i;
	/* Reverse digits */
	i /= 2;
	while( i )
	{
		i--;
		dig = str[i];
		str[i] = str[len -  1 - i];
		str[len - 1 - i] = dig;
	}
	return len;
}



/**
 Writes a pseudo-random number (between one and maxlen) of pseudo-random
 digits into str.
 str must point to an allocated buffer of size of at least maxlen chars.
 Returns the number of digits written.
 Since the randomness in part depends on machine time it has _some_ extra
 strength but still not enough for use in concurrent locking schemes on a
 single machine because gettimeofday may not return a different value on
 consecutive calls when:
 a) the OS does not support fine enough resolution
 b) the OS is running on an SMP machine.
 Additionally, gettimeofday errors are ignored.
 Excludes chars other than digits since ANSI C only guarantees that digits
 are consecutive.
 */
static int sprint_rand_digits( char *str, int maxlen )
{
	int i, max;
	struct timeval tv;
	
	/* 
     Seed the pseudo-random generator based on time - this assumes
     that consecutive calls to gettimeofday will return different values
     and ignores errors returned by gettimeofday.
     Cast to unsigned so that wrapping occurs on overflow as per ANSI C.
     */
	(void)gettimeofday( &tv, NULL );
	srand( (unsigned int)tv.tv_sec + (unsigned int)tv.tv_usec * 1000000UL );
	max = 1 + (maxlen - 1) * (rand() / (RAND_MAX + 1.0));
	for( i = 0; i < max; i++ )
	{
		str[i] = '0' + 10 * (rand() / (RAND_MAX + 1.0));
	}
	return i;
}


/**
 Generate a filename unique in an NFS namespace by creating a copy of str and
 appending .{hostname}.{pid} to it.  If gethostname() fails then a pseudo-
 random string is substituted for {hostname} - the randomness of the string
 should be strong enough across different machines.  The main assumption 
 though is that gethostname will not fail and this is just a "safe enough"
 fallback.
 The memory returned should be freed using free().
 */
static char *gen_unique_nfs_filename( const char *filename )
{
	int pidlen, hnlen, orglen = strlen( filename );
	char hostname[HOST_NAME_MAX + 1];
	char *newname;
	
	if ( gethostname( hostname, HOST_NAME_MAX + 1 ) == 0 )
	{
		hnlen = strlen( hostname );
	}
	else
	{
		hnlen = sprint_rand_digits( hostname, HOST_NAME_MAX );
		hostname[hnlen] = '\0';
	}
	newname = (char *)malloc( orglen + 1 /* period */ + hnlen + 1 /* period */ +
                             /* max possible pid size: 0.31 ~= log(10)2 */
                             (int)(0.31 * sizeof(pid_t) * CHAR_BIT + 1)
                             + 1 /* '\0' */ );
	
	if ( newname == NULL )
	{
		debug( 1, L"gen_unique_nfs_filename: %s", strerror( errno ) );
		return newname;
	}
	memcpy( newname, filename, orglen );
	newname[orglen] = '.';
	memcpy( newname + orglen + 1, hostname, hnlen );
	newname[orglen + 1 + hnlen] = '.';
	pidlen = sprint_pid_t( getpid(), newname + orglen + 1 + hnlen + 1 );
	newname[orglen + 1 + hnlen + 1 + pidlen] = '\0';
    /*	debug( 1, L"gen_unique_nfs_filename returning with: newname = \"%s\"; "
     L"HOST_NAME_MAX = %d; hnlen = %d; orglen = %d; "
     L"sizeof(pid_t) = %d; maxpiddigits = %d; malloc'd size: %d", 
     newname, (int)HOST_NAME_MAX, hnlen, orglen, 
     (int)sizeof(pid_t), 
     (int)(0.31 * sizeof(pid_t) * CHAR_BIT + 1),
     (int)(orglen + 1 + hnlen + 1 + 
     (int)(0.31 * sizeof(pid_t) * CHAR_BIT + 1) + 1) ); */
	return newname;
}


/**
 The number of milliseconds to wait between polls when attempting to acquire 
 a lockfile
 */
#define LOCKPOLLINTERVAL 10

/**
 Attempt to acquire a lock based on a lockfile, waiting LOCKPOLLINTERVAL 
 milliseconds between polls and timing out after timeout seconds, 
 thereafter forcibly attempting to obtain the lock if force is non-zero.
 Returns 1 on success, 0 on failure.
 To release the lock the lockfile must be unlinked.
 A unique temporary file named by appending characters to the lockfile name 
 is used; any pre-existing file of the same name is subject to deletion.
 */
static int acquire_lock_file( const char *lockfile, const int timeout, int force )
{
	int fd, timed_out = 0;
	int ret = 0; /* early exit returns failure */
	struct timespec pollint;
	struct timeval start, end;
	double elapsed;
	struct stat statbuf;
    
	/*
     (Re)create a unique file and check that it has one only link.
     */
	char *linkfile = gen_unique_nfs_filename( lockfile );
	if( linkfile == NULL )
	{
		goto done;
	}
	(void)unlink( linkfile );
    /* OK to not use CLO_EXEC here because fishd is single threaded */
	if( ( fd = open( linkfile, O_CREAT|O_RDONLY, 0600 ) ) == -1 )
	{
		debug( 1, L"acquire_lock_file: open: %s", strerror( errno ) );
		goto done;
	}
	/*
     Don't need to check exit status of close on read-only file descriptors
     */
	close( fd );
	if( stat( linkfile, &statbuf ) != 0 )
	{
		debug( 1, L"acquire_lock_file: stat: %s", strerror( errno ) );
		goto done;
	}
	if ( statbuf.st_nlink != 1 )
	{
		debug( 1, L"acquire_lock_file: number of hardlinks on unique "
              L"tmpfile is %d instead of 1.", (int)statbuf.st_nlink );
		goto done;
	}
	if( gettimeofday( &start, NULL ) != 0 )
	{
		debug( 1, L"acquire_lock_file: gettimeofday: %s", strerror( errno ) );
		goto done;
	}
	end = start;
	pollint.tv_sec = 0;
	pollint.tv_nsec = LOCKPOLLINTERVAL * 1000000;
	do 
	{
		/*
         Try to create a hard link to the unique file from the 
         lockfile.  This will only succeed if the lockfile does not 
         already exist.  It is guaranteed to provide race-free 
         semantics over NFS which the alternative of calling 
         open(O_EXCL|O_CREAT) on the lockfile is not.  The lock 
         succeeds if the call to link returns 0 or the link count on 
         the unique file increases to 2.
         */
		if( link( linkfile, lockfile ) == 0 ||
           ( stat( linkfile, &statbuf ) == 0 && 
            statbuf.st_nlink == 2 ) )
		{
			/* Successful lock */
			ret = 1;
			break;
		}
		elapsed = end.tv_sec + end.tv_usec/1000000.0 - 
        ( start.tv_sec + start.tv_usec/1000000.0 );
		/* 
         The check for elapsed < 0 is to deal with the unlikely event 
         that after the loop is entered the system time is set forward
         past the loop's end time.  This would otherwise result in a 
         (practically) infinite loop.
         */
		if( timed_out || elapsed >= timeout || elapsed < 0 )
		{
			if ( timed_out == 0 && force )
			{
				/*
                 Timed out and force was specified - attempt to
                 remove stale lock and try a final time
                 */
				(void)unlink( lockfile );
				timed_out = 1;
				continue;
			}
			else
			{
				/*
                 Timed out and final try was unsuccessful or 
                 force was not specified
                 */
				debug( 1, L"acquire_lock_file: timed out "
                      L"trying to obtain lockfile %s using "
                      L"linkfile %s", lockfile, linkfile );
				break;
			}
		}
		nanosleep( &pollint, NULL );
	} while( gettimeofday( &end, NULL ) == 0 );
done:
	/* The linkfile is not needed once the lockfile has been created */
	(void)unlink( linkfile );
	free( linkfile );
	return ret;
}

/**
   Acquire the lock for the socket
   Returns the name of the lock file if successful or 
   NULL if unable to obtain lock.
   The returned string must be free()d after unlink()ing the file to release 
   the lock
*/
static char *acquire_socket_lock( const char *sock_name )
{
	int len = strlen( sock_name );
	char *lockfile = (char *)malloc( len + strlen( LOCKPOSTFIX ) + 1 );
	
	if( lockfile == NULL ) 
	{
		wperror( L"acquire_socket_lock" );
		exit( EXIT_FAILURE );
	}
	strcpy( lockfile, sock_name );
	strcpy( lockfile + len, LOCKPOSTFIX );
	if ( !acquire_lock_file( lockfile, LOCKTIMEOUT, 1 ) )
	{
		free( lockfile );
		lockfile = NULL;
	}
	return lockfile;
}

/**
   Connects to the fish socket and starts listening for connections
*/
static int get_socket()
{
	int s, len, doexit = 0;
	int exitcode = EXIT_FAILURE;
	struct sockaddr_un local;
	char *sock_name = get_socket_filename();

	/*
	   Start critical section protected by lock
	*/
	char *lockfile = acquire_socket_lock( sock_name );
	if( lockfile == NULL )
	{
		debug( 0, L"Unable to obtain lock on socket, exiting" );
		exit( EXIT_FAILURE );
	}
	debug( 4, L"Acquired lockfile: %s", lockfile );
	
	local.sun_family = AF_UNIX;
	strcpy( local.sun_path, sock_name );
	len = sizeof(local);
	
	debug(1, L"Connect to socket at %s", sock_name);
	
	if( ( s = socket( AF_UNIX, SOCK_STREAM, 0 ) ) == -1 )
	{
		wperror( L"socket" );
		doexit = 1;
		goto unlock;
	}

	/*
	   First check whether the socket has been opened by another fishd;
	   if so, exit with success status
	*/
	if( connect( s, (struct sockaddr *)&local, len ) == 0 )
	{
		debug( 1, L"Socket already exists, exiting" );
		doexit = 1;
		exitcode = 0;
		goto unlock;
	}
	
	unlink( local.sun_path );
	if( bind( s, (struct sockaddr *)&local, len ) == -1 )
	{
		wperror( L"bind" );
		doexit = 1;
		goto unlock;
	}

	if( fcntl( s, F_SETFL, O_NONBLOCK ) != 0 )
	{
		wperror( L"fcntl" );
		close( s );
		doexit = 1;
	} else if( listen( s, 64 ) == -1 )
	{
		wperror( L"listen" );
		doexit = 1;
	}

unlock:
	(void)unlink( lockfile );
	debug( 4, L"Released lockfile: %s", lockfile );
	/*
	   End critical section protected by lock
	*/
	
	free( lockfile );
	
	free( sock_name );

	if( doexit )
	{
		exit( exitcode );
	}

	return s;
}

/**
   Event handler. Broadcasts updates to all clients.
*/
static void broadcast( int type, const wchar_t *key, const wchar_t *val )
{
	connection_t *c;
	message_t *msg;

	if( !conn )
		return;
	
	msg = create_message( type, key, val );
	
	/*
	  Don't merge these loops, or try_send_all can free the message
	  prematurely
	*/
	
	for( c = conn; c; c=c->next )
	{
		msg->count++;
        c->unsent->push(msg);
	}	
	
	for( c = conn; c; c=c->next )
	{
		try_send_all( c );
	}	
}

/**
   Make program into a creature of the night.
*/
static void daemonize()
{
	/*
	  Fork, and let parent exit
	*/
	switch( fork() )
	{
		case -1:
			debug( 0, L"Could not put fishd in background. Quitting" );
			wperror( L"fork" );
			exit(1);
			
		case 0:
		{
			/*
			  Make fishd ignore the HUP signal.
			*/
			struct sigaction act;
			sigemptyset( & act.sa_mask );
			act.sa_flags=0;
			act.sa_handler=SIG_IGN;
			sigaction( SIGHUP, &act, 0);

			/*
			  Make fishd save and exit on the TERM signal.
			*/
			sigfillset( & act.sa_mask );
			act.sa_flags=0;
			act.sa_handler=&handle_term;
			sigaction( SIGTERM, &act, 0);
			break;
		}
		
		default:
		{			
			debug( 0, L"Parent process exiting (This is normal)" );
			exit(0);
		}		
	}
	
	/*
	  Put ourself in out own processing group
	*/
	setsid();
	
	/*
	  Close stdin and stdout. We only use stderr, anyway.
	*/
	close( 0 );
	close( 1 );

}

/**
   Get environment variable value. The resulting string needs to be free'd.
*/
static wchar_t *fishd_env_get( const wchar_t *key )
{
	char *nres, *nkey;
	wchar_t *res;
	
	nkey = wcs2str( key );
	nres = getenv( nkey );
	free( nkey );
	if( nres )
	{
		return str2wcs( nres );
	}
	else
	{
		res = env_universal_common_get( key );
		if( res )
			res = wcsdup( res );		
		return res;
	}
}

/**
   Get the configuration directory. The resulting string needs to be
   free'd. This is mostly the same code as path_get_config(), but had
   to be rewritten to avoid dragging in additional library
   dependencies.
*/
static wcstring fishd_get_config()
{
	wchar_t *xdg_dir, *home;
	bool done = false;
	wcstring result;
	
	xdg_dir = fishd_env_get( L"XDG_CONFIG_HOME" );
	if (xdg_dir)
	{
        result = xdg_dir;
        append_path_component(result, L"/fish");
		if (!create_directory(result))
		{
			done = true;
		}
		free(xdg_dir);
	}
	else
	{		
		home = fishd_env_get( L"HOME" );
		if( home )
		{
            result = home;
            append_path_component(result, L"/.config/fish");
			if (!create_directory(result))
			{
				done = 1;
			}
			free( home );
		}
	}
    
    if (! done) {
        /* Bad juju */
		debug( 0, _(L"Unable to create a configuration directory for fish. Your personal settings will not be saved. Please set the $XDG_CONFIG_HOME variable to a directory where the current user has write access." ));
        result.clear();        
    }
    
    return result;
}

/**
   Load or save all variables
*/
static void load_or_save( int save)
{
	const wcstring wdir = fishd_get_config();
	char hostname[HOSTNAME_LEN];
	connection_t c;
	int fd;
	
	if (wdir.empty())
		return;
	
	std::string dir = wcs2string( wdir );
	
	gethostname( hostname, HOSTNAME_LEN );
	
    std::string name;
    name.append(dir);
    name.append("/");
    name.append(FILE);
    name.append(hostname);
	
	debug( 4, L"Open file for %s: '%s'", 
		   save?"saving":"loading", 
		   name.c_str() );
	
    /* OK to not use CLO_EXEC here because fishd is single threaded */
	fd = open(name.c_str(), save?(O_CREAT | O_TRUNC | O_WRONLY):O_RDONLY, 0600);
	
	if( fd == -1 )
	{
		debug( 1, L"Could not open load/save file. No previous saves?" );
		wperror( L"open" );
		return;		
	}
	debug( 4, L"File open on fd %d", c.fd );

	connection_init( &c, fd );

	if( save )
	{
		
		write_loop( c.fd, SAVE_MSG, strlen(SAVE_MSG) );
		enqueue_all( &c );
	}
	else
		read_message( &c );

	connection_destroy( &c );	
}

/**
   Load variables from disk
*/
static void load()
{
	load_or_save(0);
}

/**
   Save variables to disk
*/
static void save()
{
	load_or_save(1);
}

/**
   Do all sorts of boring initialization.
*/
static void init()
{

	sock = get_socket();
	daemonize();	
	env_universal_common_init( &broadcast );
	
	load();	
}

/**
   Main function for fishd
*/
int main( int argc, char ** argv )
{
	int child_socket;
	struct sockaddr_un remote;
	socklen_t t;
	int max_fd;
	int update_count=0;
	
	fd_set read_fd, write_fd;

	set_main_thread();
    setup_fork_guards();
	
	program_name=L"fishd";
	wsetlocale( LC_ALL, L"" );	

	/*
	  Parse options
	*/
	while( 1 )
	{
		static struct option
			long_options[] =
			{
				{
					"help", no_argument, 0, 'h' 
				}
				,
				{
					"version", no_argument, 0, 'v' 
				}
				,
				{ 
					0, 0, 0, 0 
				}
			}
		;
		
		int opt_index = 0;
		
		int opt = getopt_long( argc,
							   argv, 
							   GETOPT_STRING,
							   long_options, 
							   &opt_index );
		
		if( opt == -1 )
			break;
		
		switch( opt )
		{
			case 0:
				break;				

			case 'h':
				print_help( argv[0], 1 );
				exit(0);				
								
			case 'v':
				debug( 0, L"%ls, version %s\n", program_name, PACKAGE_VERSION );
				exit( 0 );				
				
			case '?':
				return 1;
				
		}		
	}
	
	init();
	while(1) 
	{
		connection_t *c;
		int res;

		t = sizeof( remote );		
		
		FD_ZERO( &read_fd );
		FD_ZERO( &write_fd );
		FD_SET( sock, &read_fd );
		max_fd = sock+1;
		for( c=conn; c; c=c->next )
		{
			FD_SET( c->fd, &read_fd );
			max_fd = maxi( max_fd, c->fd+1);
			
			if( ! c->unsent->empty() )
			{
				FD_SET( c->fd, &write_fd );
			}
		}

		while( 1 )
		{
			res=select( max_fd, &read_fd, &write_fd, 0, 0 );

			if( quit )
			{
				save();
				exit(0);
			}
			
			if( res != -1 )
				break;
			
			if( errno != EINTR )
			{
				wperror( L"select" );
				exit(1);
			}
		}
				
		if( FD_ISSET( sock, &read_fd ) )
		{
			if( (child_socket = 
				 accept( sock, 
						 (struct sockaddr *)&remote, 
						 &t) ) == -1) {
				wperror( L"accept" );
				exit(1);
			}
			else
			{
				debug( 4, L"Connected with new child on fd %d", child_socket );

				if( fcntl( child_socket, F_SETFL, O_NONBLOCK ) != 0 )
				{
					wperror( L"fcntl" );
					close( child_socket );		
				}
				else
				{
					connection_t *newc = (connection_t *)malloc( sizeof(connection_t));
					connection_init( newc, child_socket );					
					newc->next = conn;
					send( newc->fd, GREETING, strlen(GREETING), MSG_DONTWAIT );
					enqueue_all( newc );				
					conn=newc;
				}
			}
		}
		
		for( c=conn; c; c=c->next )
		{
			if( FD_ISSET( c->fd, &write_fd ) )
			{
				try_send_all( c );
			}
		}
		
		for( c=conn; c; c=c->next )
		{
			if( FD_ISSET( c->fd, &read_fd ) )
			{
				read_message( c );

				/*
				  Occasionally we save during normal use, so that we
				  won't lose everything on a system crash
				*/
				update_count++;
				if( update_count >= 64 )
				{
					save();
					update_count = 0;
				}
			}
		}
		
		connection_t *prev=0;
		c=conn;
		
		while( c )
		{
			if( c->killme )
			{
				debug( 4, L"Close connection %d", c->fd );

				while( ! c->unsent->empty() )
				{
					message_t *msg = c->unsent->front();
                    c->unsent->pop();
					msg->count--;
					if( !msg->count )
						free( msg );
				}
				
				connection_destroy( c );
				if( prev )
				{
					prev->next=c->next;
				}
				else
				{
					conn=c->next;
				}
				
				free(c);
				
				c=(prev?prev->next:conn);
				
			}
			else
			{
				prev=c;
				c=c->next;
			}
		}

		if( !conn )
		{
			debug( 0, L"No more clients. Quitting" );
			save();			
			env_universal_common_destroy();
			break;
		}		

	}
}