How to replace characters in an nginx variable string?

Question

Is there a way I can replace non alphanumeric characters returned with $request_uri with a space (or a +)?

What I'm trying to do is redirect all 404's in one of my sites to it's search engine, where the query is the uri requested. So, I have a block in my nginx.conf containing:

error_page 404 = @notfound;
location @notfound {
    return 301 $scheme://$host/?s=$request_uri;
}

While this does indeed work, the url's it's returning are the actual uri's complete with -_/ characters causing the search to always return 0 results

For instance... give this url: https://example.com/my-articles, the redirect ends up as this: https://example.com/?s=/my-articles

What I would like is to end up (ultimately) like this: https://example.com/?s=my+articles (tho, the + at the beginning works fine too... https://example.com/?s=+my+articles

I will need to do this without LUA or Perl modules. So, how can I accomplish this?

Answer 1

You may need to tweak this depending upon how far down your directory structure you want the replacement to go, but this is the basic concept.

Named location for initial capture of 404s:

location @notfound {
  rewrite (.*) /search$1 last;
}

Named locations are a bit limiting, so all this does is add /search/ to the beginning of the URI which returned 404. The last flag tells Nginx to break out of the current location and select the best location to process the request based on the rewritten URI, so we need a block to catch that:

location ^~ /search/ {
  internal;
  rewrite ^/search/(.*)([^a-z0-9\+])(.*)$ /search/$1+$3 last;
  rewrite ^/search/(.*)$ /?s=$1 permanent;
}

The internal directive makes this location only accessible to the Nginx process itself, any client requests to this block will return 404.

The first rewrite will change the last non text, digit or + character into a + and then ask Nginx to reevaluate the rewritten URI.

The location block is defined with the ^~ modifier, which means requests matching this location will not be evaluated against any regex defined location blocks, so this block should keep catching the rewritten requests.

Once all the non word characters are gone the first rewrite will no longer match so the request will be passed to the next rewrite, which removes the /search from the front of the URI and adds the query string.

My logs look like this:

>> curl -L -v http://127.0.0.1/users-forum-name.1
<<  "GET /?s=users+forum+name+1 HTTP/1.1"

>> curl -L -v http://127.0.0.1/users-forum-name/long-story/some_underscore
<< "GET /?s=users+forum+name+long+story+some+underscore"

You get the idea..