That sounds like a perfect use for Parse::RecDescent.

Get it (togther with copious amounts of documentation and examples) from CPAN at http://search.cpan.org/search?dist=Parse-RecDescent.

#!/usr/bin/perl -w
use strict;
package Item;
sub new {
    my $proto = shift;
    my $class = ref($proto) || $proto;
    my $self  = {};
    bless ($self, $class);
    return $self;
}

sub add_attribute {
    my $self = shift;
    my ($key,$value) = @_;
    $self->{$key} = $value;
}

1;  #included in case Item becomes a real Module someday.

package Section;
sub new {
    my $proto = shift;
    my $class = ref($proto) || $proto;
    my $self  = {};
    $self->{Items} = [];
    bless ($self, $class);
    return $self;
}

sub add_attribute {
    my $self = shift;
    my ($key,$value) = @_;
    $self->{$key} = $value;
}

sub add_item {
    my $self = shift;
    my $item = Item->new();
    push @{$self->{Items}},$item;
    return $item;
}

1; #Ditto

package UberObject;
sub new {
    my $proto = shift;
    my $class = ref($proto) || $proto;
    my $self  = [];
    bless ($self, $class);
    return $self;
}

sub add_section {
    my $self = shift;
    my $section = Section->new();
    push @$self,$section;
    return $section;
}

1; #again

package Main;
use Data::Dumper;
my $uo = new UberObject();
my ($section,$item) = (undef,undef);

while (<>) {  #dammit Jim, parse the data already!
    if (/begin_section/) {
    $section = $uo->add_section();
    } elsif (/end_section/) {
    undef $section; 
    } elsif (/begin_item/) {
    $item = $section->add_item();
    } elsif (/end_item/) {
    undef $item;
    } elsif (/=/) {
        if (defined($item)) {
            $item->add_attribute(split(/=/));
        } elsif (defined($section)) {
            $section->add_attribute(split(/=/));
        } else { 
            die "Bad attribute at line $.";
        }
    }
}

#probably should do something interesting .....
#        .... nah Just Dump it all!
my $d=Data::Dumper->new([$uo],['UberObject']);
$d->Indent(1);
print $d->Dump();
[download]

begin_section
     name=section1
     type=section_type1

     begin_item
          name=item1
          desc=item_description1
          link=item_link1
     end_item

     begin_item
          name=item2
          desc=item_description2
          link=item_link2
     end_item

end_section

begin_section
     name=section2
     type=section_type2

     begin_item
          name=item21
          desc=item_description21
          link=item_link21
     end_item

     begin_item
          name=item22
          desc=item_description22
          link=item_link22
     end_item
end_section
[download]

"The correct thing" (Exercise left to reader)
[download]

Important things to remember are things like error_handling. i.e. in your example

  if (!/end_item/) {
    if(/begin_item/) {
      while (!/end_item/) {
      ...deal with data members here...
      }
    }
  }
  else {
    ... print or die an error message here about an end_item without b
+egin ...
  }
[download]

Some of these things that one must think of you get for "free" using the module that davorg suggested...

if(/begin_item/) {
    while (!/end_item/) {
    ...deal with data members here...
    }
}
[download]

while (my $range = /begin_item/i.../end_item/i) {
    # skip the first and last lines if you want...
    next unless $range > 1 && substr($range, -2) ne 'E0';

    # deal with data member
}
[download]

Another option would be to get an in-memory representation of the object (which I realize is part of the problem :), then use Data::Dumper to serialize it to disk.

I realize that these don't help you with the data file format that you have, but in case you're investigating other options, these are two good ones.