What does an "exec" command do?

man bash says:

exec [-cl] [-a name] [command [arguments]]
      If command is specified, it replaces the shell.  No new  process
      is  created.  The arguments become the arguments to command.  If
      the -l option is supplied,  the  shell  places  a  dash  at  the
      beginning  of  the  zeroth  argument passed to command.  This is
      what login(1) does.  The -c option causes command to be executed
      with  an empty environment.  If -a is supplied, the shell passes
      name as the zeroth argument to the executed command.  If command
      cannot  be  executed  for  some  reason, a non-interactive shell
      exits, unless the execfail shell option  is  enabled.   In  that
      case,  it returns failure.  An interactive shell returns failure
      if the file cannot be executed.  If command  is  not  specified,
      any  redirections  take  effect  in  the  current shell, and the
      return status is 0.  If there is a redirection error, the return
      status is 1.

The last two lines are what is important: If you run exec by itself, without a command, it will simply make the redirections apply to the current shell. You probably know that when you run command > file, the output of command is written to file instead of to your terminal (this is called a redirection). If you run exec > file instead, then the redirection applies to the entire shell: Any output produced by the shell is written to file instead of to your terminal. For example here

bash-3.2$ bash
bash-3.2$ exec > file
bash-3.2$ date
bash-3.2$ exit
bash-3.2$ cat file
Thu 18 Sep 2014 23:56:25 CEST

I first start a new bash shell. Then, in this new shell I run exec > file, so that all output is redirected to file. Indeed, after that I run date but I get no output, because the output is redirected to file. Then I exit my shell (so that the redirection no longer applies) and I see that file indeed contains the output of the date command I ran earlier.

exec is a command with two very distinct behaviors, depending on whether at least one argument is used with it, or no argument is used at all.

• If at least one argument is passed, the first one is taken as a command name and exec try to execute it as a command passing the remaining arguments, if any, to that command and managing the redirections, if any.

• If the command passed as first argument doesn't exist, the current shell, not only the exec command, exits in error, unless the shell is interactive or the bash option execfail is set (shopt -s execfail). See also https://superuser.com/questions/992204/why-does-exec-non-existent-file-exits-the-shell-when-in-a-script-that-is-sourc

• If the command exists and is executable, it replaces the current shell. That means that if exec appears in a script, the instructions following the exec call will never be executed (unless exec is itself in a subshell). A successful exec never returns. Shell traps like "EXIT" won't get triggered either.

• If no argument is passed, exec is only used to redefine the current shell file descriptors. The shell continue after the exec, unlike with the previous case, but the standard input, output, error or whatever file descriptor has been redirected take effect.

• If some of the redirections uses /dev/null, any input from it will return EOF and any output to it will be discarded.

• You can close file descriptors by using - as source or destination, e.g. exec <&-. Subsequent read or writes will then fail.

Here are two examples:

echo foo > /tmp/bar
exec < /tmp/bar # exec has no arguments, will only affect current shell descriptors, here stdin
cat # simple command that read stdin and write it to stdout

This script will output "foo" as the cat command, instead of waiting for user input as it would have done in the usual case will take its input from the /tmp/bar file which contains foo.

echo foo > /tmp/bar
exec wc -c < /tmp/bar # exec has two arguments, the control flow will switch to the wc command
cat

This script will display 4 (the number of bytes in /tmp/bar) and immediately ends. The cat command won't be executed.

To understand exec you need to first understand fork. I am trying to keep it short.

• When you come to a fork in the road you generally have two options. Linux programs reach this fork in the road when they hit a fork() system call.

• Normal programs are system commands that exist in a compiled form on your system. When such a program is executed, a new process is created. This child process has the same environment as its parent, only the process ID number is different. This procedure is called forking.

• Forking provides a way for an existing process to start a new one. However, there may be situations where a child process is not the part of the same program as parent process. In this case exec is used. exec will replace the contents of the currently running process with the information from a program binary.
• After the forking process, the address space of the child process is overwritten with the new process data. This is done through an exec call to the system.

In bash, if you do help exec:

$ help exec
exec: exec [-cl] [-a name] [command [arguments ...]] [redirection ...]
    Replace the shell with the given command.

    Execute COMMAND, replacing this shell with the specified program.
    ARGUMENTS become the arguments to COMMAND.  If COMMAND is not specified,
    any redirections take effect in the current shell.

    Options:
      -a name   pass NAME as the zeroth argument to COMMAND
      -c        execute COMMAND with an empty environment
      -l        place a dash in the zeroth argument to COMMAND

    If the command cannot be executed, a non-interactive shell exits, unless
    the shell option `execfail' is set.

    Exit Status:
    Returns success unless COMMAND is not found or a redirection error occurs.

The relevant bit:

If COMMAND is not specified, any redirections take effect in the current shell.

exec is a shell builtin, which is the shell equivalent of the exec family of system calls that G_P speaks of (and whose manpages you appear to have read). It just has the POSIX mandated functionality of affecting the current shell if no command is specified.