Originally Posted by porter

Okay, what you need to understand is what a process is in UNIX. Basically a process is

1. a memory image

2. a set of file descriptors 0,1,2 etc.

3. a point of execution

The memory image exists outside the kernel, the file descriptors live inside the kernel. A single file descriptor points to a file. That file may be referenced by multiple file descriptors.

Fact 1. All close() does is closes a single file descriptor for the current process. That drops the usage count on the actual file, when the usage count on the file drops to zero then it is really closed.

Fact 2. When a process terminates, all it's file descriptors that are still open are closed.

Fact 3. All "dup" and "dup2" do are allocate another file descriptor to point to the same file and increase the usage count on the underlying file.

Fact 4. Fork() duplicates the entire memory image AND the file descriptor table. Both processes now continue executing at the same point, the only difference being the parent got the pid of the child, the child got nothing. Duplicating the file descriptor table is like dup() has been called on each open file descriptor.

Fact 5. Exec() replaces the entire memory image with the new program but the file descriptors do not change EXCEPT for the file descriptors marked as FD_CLOEXEC, which are closed.

Fact 6. Pipe() returns two file descriptors, one is readable, one is writeable. A pipe will remain readable until all writeable file descriptors have been closed. Remember if you create a pipe then call fork there are two descriptors that are writeable.

Hints:

Get into the habit of putting -1 in variables that hold file descriptors when they are not holding a value, when you close it, put -1 back in the file descriptor variable.