...routines

By definition, a reentrant routine is a piece of code that can be entered before its completion and still execute correctly. This implies that a reentrant routine must not be allowed to modify its instructions. Yet self-modification is sometimes a very useful technique, especially for low-level routines, so why it doesn't work with reentrant code?

Also, you may have already encountered reentrant routines when dealing with recursive procedures written in a programming language like Pascal or C. Why is some form of reentrance mandatory for implementing these constructs? In what these implementations differ from reentrance between programs in a multiprogramming environment?

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...record

This pointer is often implicit, in the sense that a stack of records is mantained, with the the record of the newest activation on top. Try to describe to yourself how this mechanism works...

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...code

...or the fragment that's being executed at a given time, as is often the case for systems that make use of virtual memory

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...OS

or rather in the scheduler of an OS, which is the subsystem dealing with the policy of process dispatching

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...involved

...and because of the undefeated programmers' tendency to use all the available resources for their applications, which implies that more memory usually means larger processes, and not more processes instead...

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...value

Depending on the algorithms used to implement a priority scheme for sharing CPU usage among processes, more than one ``priority'' value for the process may be recorded, and some of them may change with time

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...delay

...according to the scheduling policy in use: the CPU resources are invariably limited, hence there's no way to satisfy all possible scheduling wishes

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...them

In the jargon this condition is expressively called trashing.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...routine

Note that this is worse than having a slow, but guaranteed, response time. It means that the controller is time-variant: sometimes it has a huge inertia, sometimes it hasn't; sometimes it's so slow that the plant goes astray before the controller gets to know about, sometimes it's so fast that the plant is forced to jump around like crazy and never reaches a nice steady regime.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...networks

See the Dec. 94 issue of em Byte for an interesting discussion of threaded server implementations under MS WindowsNT, using the Winsock protocols.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...termination)

This is different than having a parent terminated before the child. In this case the child is simply adopted by process 1, init.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...UNIX

except process 0, swapper, and process -1, pager

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...API

API = Application Programmer's Interface, i.e. the interface (usually a layer of library functions) through which an application programmer can specify the use of the services provided by of some software system in his/her high-level language source code

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...available

A simpler, more rigid but portable, way to generate error message is provided by the perror() library function (which is standard in ANSI C, while sys_errlist and errno are UNIX specific). You may want to check its manual page.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...flavors,

Well, actually four, but nobody cares about System V's wait4().

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...exit()

..or to the return statement of the main() function, if the child is a complete C program

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...assumed

...at the very least, you can simply think that all the files we are talking about are just sequences of bytes stored on a disk, and that they are in the same disk directory as the program that uses them: only a file's name, and not its physical disk location, is then relevant to the discussion

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...so

BSD and System V guarantee only that each process has at least 20 slots in the object reference table, but this limit can be configured on system installation. the getdtablesize() system call can be used to ask the system what the actual table size is.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...functions

More precisely, since in UNIX the stream-oriented functions are implemented in terms of the block-oriented ones, a file descriptor is one of the fields of a FILE structure. So, if you have opened a FILE *fp in the stream-oriented mode with fopen(), and then want to perform some low-level operations on it, you will first need to find its associated file descriptor using the fileno(fp) macro. Conversely, the fdopen(fd, "r") stream library is used to ``convert'' a file descriptor fd into a FILE pointer open for reading ("w" for writing, etc.). Once more, be warned that these calls are UNIX specific, though some other systems use a similar file descriptor mechanism for low level I/O.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...them

The precise header names can change with the UNIX flavor you are using. As usual, the ultimate reference is the manual page.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...mask

actually there's more than that, but we won't deal with the so called special files until much later in the course.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...optional

An often overlooked characteristic of C is that a function can be arranged so to have a number of parameters not specified at compile time. Yet every C programmer routinely uses functions like printf(), which can take in pronciple any number of parameters. Check the section on the ``variable argument lists'' or ``va_lists'' in your C book to learn more about.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...read

...which might be less than those requested, if there weren't that many left in the file, starting from the current seek position onward...

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...located

...actually not: there are security reasons for never ever putting the current directory, ".", in the PATH. But this is just the simplest security hole that an undergrad wishing to peek in instructors' assignment files must know about. An easily blocked one, unfortunately...

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...element

So, for example, if envp contains ten environment variables, then envp[10]==NULL.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...optional

...so optional that it's not even in the ANSI standard for C, and this is why your compiler may yell at you if you try to use a main() like the above. The ANSI and Posix committees declared that every C program must access the environment through a global external pointer, declared as extern char *environ[], whose content is exactly as envp above. Amen.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...O_WRONLY?)

Posix defines in unistd.h three constants for these file descriptors, STDIN_FILENO for standard input, STDOUT_FILENO for standard output, STDERR_FILENO for standard error. The use of theses named constants is to be preferred to the numbers for portability, since system implementors free to change the numbers as they wish.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...linker

The system can figure out if it's an executable by looking for a ``magic number'' in the first few bytes of the file: man a.out, for those eager to know more about magic cookies.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...script

If you are not familiar with UNIX shell scripts yet, think of them as akin to ``.BAT'' files in M$-DOG, only much better, and then go fetch yourself Kernighan & Pike's gospel down at the library.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...constraint

There's actually one, but it's much looser. For efficiency reasons, the system has to use tables of fixed size to hold the arguments, so there's a limit to the size that the whole argv string array can attain, and it's specified via the ARG_MAX constant, in the sys/limit.h header. This is several kilobytes at least, so if you happen to have to worry about, it probably means that you are trying to shoot yourself in a foot with a battleship cannon: you know, ``arguments'' can be read from a file too ...

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...it

Warning: a naive implementation, i.e. one that just substitutes read() with fgets(), write() with fputs(), and pipe() with popen(), will possibly hang doing nothing. If this happens, the problem is in the stream buffering: check out the manpage for setvbuf().

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...file

...hence as an octal value encoding the appropriate access bits in owner-group-world order. Constants are provided for this purpose too:
SEM_R for owner read and SEM_W for owner write;
SEM_R>>3 for group read and SEM_W>>3 for group write;
SEM_R>>6 for world read and SEM_W>>6 for world write;

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...sleep

...hence the sequence of steps for accessing a shared resource controlled by a semaphore becomes: (1) test the semaphore that controls the resource; (2) if the counter is positive, decrement it by one and access the resource; (3) if, otherwise, it's zero, sleep until it gets positive, and return to step 1 on wakeup; (4) increment by one the counter when done using the resource.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...usage

...and if you find that those details make sense and have the elegance their authors claim, please let me know: one personal idea of common sense and elegance would need revising, either mine or yours.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...possible

...where ``possible'' refers more to execution capability than to size: we'll meet later a phenomenon called ``trashing'' that imposes a limit on the number of active processes way below the physical capabilities of main memory

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...UNIX

What follows is largely based on the article ``4.2BSD and 4.3BSD as Examples of the UNIX System'', by J. S. Quaterman, A. Silbershatz and J. L. Peterson, appeared on Computing Surveys, 17(4), Dec. 1985.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...run

As we'll see later, System V R4 uses the opposite convention, otherwise users' life would be too easy...

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...data.

Here the term relation is used strictly in its set-theoretical meaning: a relation R among the sets is, by definition, a subset of their cartesian product

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...FMS

...not to be confused with the file system, which is a data structure used to organize files on disk.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...search

A dicotomic (or binary, or divide et impera) search for the record with a given key proceeds as follows: (1) Define the `` section'' of the file as the entire file; (2) compare the search key with the one of the record at the middle of the section; (3) if they are equal return the record; (4) otherwise, if the search key is higher, define the section to be the next half of the current one; (4.1) otherwise, define the section to be the previous half; (5) if the section has zero size, return with ``not found'' error; (6) goto 2.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...system

In other words, the FMS offers to users/applications directory modification services in the form of commands or system calls, so that it can protect the fyle system integrity by allowing only ``correct'' operations.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...layers

From lowest to highest, phisical, datalink, network, transport, session, presentation, application.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Unix).

By extension, the term server is also used to indicate a computer which, through some server application, provides some service for other computers connected to it via a network. The most common example is a file server which has a local disk and services requests from remote clients to read and write files on that disk, often using Sun's Network Filing System (NFS) protocol or Novell Netware on IBM PCs.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.