OSF/1 does this as follows:
struct file *uf_ofile[NOFILE_IN_U];/* file structs of open files */
char uf_pofile[NOFILE_IN_U]; /* per-process flags of open files */
int uf_lastfile; /* high-water mark of uf_ofile */
/*
* If greater than NOFILE_IN_U file descriptors are allocated,
* uf_ofile_of and uf_pofile_of are used to reference the KALLOC'ed
* buffers which store the additional entries.
*/
u_int uf_of_count;
struct file **uf_ofile_of; /* Pointer to KALLOC'ed buffer */
char *uf_pofile_of; /* Pointer to KALLOC'ed buffer */
NOFILE_IN_U is 64 so you have 64 descriptors available directly in the
struct utask. Further ones (up to a configurable systemwide max of
default 4096) are dynamically allocated in chunks and hang off of_ofile_of.
uf_pofile[] and uf_ofile handle per-file flags by having one char for
each file rather than the bit field that Linux uses even at kernel level
in struct files_struct. OSF/1 <sys/select.h> says
/*
* Select uses bit masks of file descriptors in longs.
* These macros manipulate such bit fields (the filesystem macros use chars).
* FD_SETSIZE may be defined by the user, but the default here is equal
* to OPEN_MAX_SYSTEM in param.h since that defines the absolute maximum
* number of file descriptors that a process can open.
*/
/* We don't really want to include all of limits.h to get the
real value of this, so we leave this ugly constant here */
#define NBBY 8 /* number of bits in a byte */
#define MAX_NOFILE 4096
#ifndef FD_SETSIZE
#define FD_SETSIZE MAX_NOFILE
#endif
so the default is for 4096 bits but the programmer can set FD_SETSIZE
manually if desperate for a smaller fd_set.
--Malcolm
-- Malcolm Beattie <mbeattie@sable.ox.ac.uk> Unix Systems Programmer Oxford University Computing Services