.TH OPEN 2 
.UC 4
.SH NAME
creat, open, mkdir, mknod \- create or open a file, special file, or directory
.SH SYNOPSIS
.nf
.B #include <sys/types.h>
.B #include <unistd>
.B int creat (const char *file_name, mode_t mode);
.B int mknod (const char *file_name, mode_t mode, dev_t dev);
.B int mkdir (const char *file_name, mode_t mode);
.B int open (const char *file_name, int flag, ...);
.fi
.SH DESCRIPTION
.B open()
opens the file specified by 
.I file_name
, with modes set according to the 
.I  flag 
parameter.
.PP
The bit fields of 
.I flag 
are : 
.PP 
.B O_RDONLY	
.PP
.B O_WRONLY
.PP
.B O_RDWR
.PP
.B O_CREAT
- create if it doesn't exist.  Note that if this flag is specified, and additional parameter 
.I mode_t mode
must be passed to open specifying permissions on the created file.  This is modified by 
.B umask
in the usual way, with  the permissions of the created file becoming 
(
.B ~umask  
& 
.I mode
)
.PP
.B O_EXCL
.PP
.B O_NOCTTY
- not implemented.
.PP
.B O_TRUNC	
.PP
.B O_APPEND
.PP
.B O_NONBLOCK
.PP
.B O_NDELAY
- not implemented.
.PP
.B creat()
is a more specific version of
.B open()
which is used to create 
.I file_name.
The behavior of   
.B creat 
(
.I file_name, mode
)
is the same as 
.PP
.B open 
(
.I file_name,
O_CREAT | O_WRONLY, O_TRUNC, 
.I mode
);
.PP
.B mkdir()
is identical to 
.B creat(),
only a directory is created instead of a normal file.
.PP
.B mknod()
is identical to 
.B creat(),
only
.I dev
specifies the device number that is associated  with
.I file_name
.SH ERRORS
.B -EACCESS
is returned if the directory does not have execute permissions
.PP
.B -EINVAL
is returned if there are no more filedescriptors available
in the system or for the process.  These maximums
are specified in
.B NR_OPEN 
and 
.B NR_FILE
, both defined in 
.I linux/include/linux/fs.h    
with defaults of 20 and 64 respectively.
.PP
.B -EISDIR
is returned if you attempt to open a directory with access flags of
anything but O_RDONLY, or with
O_CREATE, or O_TRUNC flags set.
.PP
.B -ELOOP
is returned if 
circular links are encountered.
.PP
.B -ENOENT
is returned if a portion of the specified path is missing.
.PP
.B -EPERM
is returned if the permissions for the file or directory do not
allow the requested action.
.SH FILES
linux/fs/open.c
.br
linux/fs/namei.c
.br
/usr/include/linux/sys.h
.br
/usr/include/unistd.h
.SH SEE ALSO
close(2), fopen(3), link(2), unlink(2), umask(2)
.SH BUGS
.B O_NCTTY
,
.B O_NDELAY 
are not yet implemented.   
.PP
The stock distribution of linux supports a maximum of
20 open file descriptors per process, as specified in the 
.B NR_OPEN 
macro in 
.I linux/include/linux/fs.h
.
This is easily expanded to 32,
but since certain internal bitmaps are implemented in 32 bit integers,
further expansion will prove difficult, although changing the 
.B close_on_exec
field of 
.B task_struct, which is defined in 
.I linux/include/linux/sched.h 
and other related fields 
to be long long instead of long will allow a maximum of
64 file descriptors per process.
.PP
The stock persystem limit is 64 open file descriptors, as specified in the 
.B NR_FILES
macro in 
.I linux/include/linux/fs.h.  
This is an artificial limit,
and is easily circumvented by recompiling
the kernel with a larger value.