Lab Test: Bot Traffic & Mitigation Playbook Development (WordPress + Cloudflare)

Executive summary

Our experiment began on 04 Sep 2025, example.com we simulated abusive bot traffic targeting /w/ and /w/index.php (paths used by MediaWiki). Because the site runs WordPress, these requests fell through to PHP, driving PHP-FPM CPU to 100% and slowing responses.

What we did:

• Containment: Cloudflare WAF rule that blocks /w, /w/, and anything under /w/….
• Hardening:
  • Block /xmlrpc.php (safe if Jetpack/remote publishing is not used).
  • Protect /wp-login.php with a Managed Challenge (or IP allow-list).
• (Optional) Performance: Cloudflare Cache Rule to serve HTML from the edge for anonymous users.

Result: /w/* no longer reaches origin; PHP-FPM load normalized; key pages serve with CF-Cache-Status: HIT.

Impact

• Symptoms: High CPU, pegged PHP-FPM workers for the example.com pool.
• User experience: Intermittent slowdowns/timeouts during spikes.
• Blast radius: Mainly example.com. example-two.com also showed noisy login/XML-RPC hits.

Indicators & evidence

Hot PHP-FPM pool

ps -eo pcpu,pmem,args --no-headers \
| awk '/php-fpm: pool/ {pool=$NF; cpu[pool]+=$1; mem[pool]+=$2; n[pool]++} \
       END {for (p in cpu) printf "%6.1f%% CPU  %6.1f%% MEM  %2d procs  %s\n", cpu[p], mem[p], n[p], p}' \
| sort -nr

Access-log hotspots (top endpoints)

tail -n 20000 /var/www/vhosts/system/example.com/logs/*access* 2>/dev/null \
| awk '{print $7}' | cut -d? -f1 | sort | uniq -c | sort -nr | head

Spike fingerprint (time window)

FROM="04/Sep/2025:16:55"; TO="04/Sep/2025:17:01"
awk -v f="$FROM" -v t="$TO" '$4~/\[/ {ts=substr($4,2,20); if(ts>=f&&ts<=t) print $7}' \
  /var/www/vhosts/system/example.com/logs/access_log \
  /var/www/vhosts/system/example.com/logs/access_ssl_log \
| cut -d? -f1 | sort | uniq -c | sort -nr | head

Finding: Clustered /w/ and /w/index.php requests via Cloudflare edge IPs (bot traffic through the proxy).

Root cause

Automated crawl/attack requested MediaWiki edit paths on a WordPress site. Each request invoked PHP (Apache → PHP-FPM), saturating CPU.

Remediation steps

(http.request.uri.path in {"/w", "/w/"} or starts_with(http.request.uri.path, "/w/"))

not starts_with(http.request.uri.path, "/wp-admin")
and http.request.uri.path ne "/wp-login.php"
and http.request.uri.path ne "/xmlrpc.php"
and not (http.request.uri.query contains "preview=true")
and not any(http.request.headers["cookie"][*] contains "wordpress_logged_in_")
and not any(http.request.headers["cookie"][*] contains "comment_author_")

Validation & results

• /w/* eliminated:

  tail -n 500 /var/www/vhosts/system/example.com/logs/access_* 2>/dev/null \
  | awk '{print $7}' | cut -d? -f1 | grep -E '^/w(/|$)' | wc -l  

• Lower PHP-FPM CPU:

  ps -eo pcpu,pmem,args --no-headers \
  | awk '/php-fpm: pool/ {p=$NF; c[p]+=$1} END{for(p in c) printf "%5.1f%% CPU  %s\n", c[p], p}' \
  | sort -nr

• Edge cache hits:

  curl -sI https://example.com/ | egrep -i 'CF-Cache-Status|Age'  

Ongoing monitoring

• Top endpoints (recent):

  tail -n 600 /var/www/vhosts/system/example.com/logs/access_* 2>/dev/null \
  | awk '{print $7}' | cut -d? -f1 | sort | uniq -c | sort -nr | head

• Which site is hot:

  ps -eo pcpu,pmem,args --no-headers \
  | awk '/php-fpm: pool/ {pool=$NF; cpu[pool]+=$1} END {for (p in cpu) printf "%6.1f%% CPU  %s\n", cpu[p], p}' \
  | sort -nr

• Cloudflare Security → Events: filter by rule names (e.g., “Block /w and /w/*”, “Block xmlrpc”, “Protect wp-login”) to confirm mitigations are firing.

Recommendations

1. Keep the /w/* block permanently (zero cost on WordPress).
2. Leave /xmlrpc.php blocked unless a product explicitly requires it.
3. Protect /wp-login.php with Managed Challenge or IP allow-list.
4. (Optional) Keep the HTML cache rule; pair with Cloudflare’s WP plugin for automatic purges on content updates.
5. Use the quick-response runbook below for future spikes.

Appendix: Quick-response runbook

Identify the hot site/pool

ps -eo pcpu,pmem,args --no-headers \
| awk '/php-fpm: pool/ {p=$NF; c[p]+=$1} END{for(p in c) printf "%5.1f%% CPU  %s\n", c[p], p}' \
| sort -nr

Identify hot URLs

tail -n 20000 /var/www/vhosts/system/<DOMAIN>/logs/*access* 2>/dev/null \
| awk '{print $7}' | cut -d? -f1 | sort | uniq -c | sort -nr | head

Slice by time window

FROM="dd/Mon/yyyy:HH:MM"; TO="dd/Mon/yyyy:HH:MM"
awk -v f="$FROM" -v t="$TO" '$4~/\[/ {ts=substr($4,2,20); if(ts>=f&&ts<=t) print $7}' \
  /var/www/vhosts/system/<DOMAIN>/logs/access_log \
  /var/www/vhosts/system/<DOMAIN>/logs/access_ssl_log \
| cut -d? -f1 | sort | uniq -c | sort -nr | head

Cloudflare WAF expressions (copyable)

• Block MediaWiki paths on WordPress:

  (http.request.uri.path in {"/w", "/w/"} or starts_with(http.request.uri.path, "/w/"))

• Block XML-RPC:

  (http.request.uri.path eq "/xmlrpc.php")

• Protect Login (allow-list or challenge):

  (http.request.uri.path eq "/wp-login.php") and not ip.src in {YOUR.PUBLIC.IP}

• Cache HTML for anonymous users (Cache Rule; AND all):

  not starts_with(http.request.uri.path, "/wp-admin")
  and http.request.uri.path ne "/wp-login.php"
  and http.request.uri.path ne "/xmlrpc.php"
  and not (http.request.uri.query contains "preview=true")
  and not any(http.request.headers["cookie"][*] contains "wordpress_logged_in_")
  and not any(http.request.headers["cookie"][*] contains "comment_author_")

References & further reading

• MediaWiki — Manual: Short URL (why /w/ and /w/index.php exist): mediawiki.org/wiki/Manual:Short_URL
• MediaWiki — Manual: Short URL / Apache: mediawiki.org/wiki/Manual:Short_URL/Apache
• Cloudflare docs — Caching static HTML with WordPress/WooCommerce: developers.cloudflare.com/.../caching-static-html
• WordPress — Hardening & Brute Force guidance: developer.wordpress.org/.../security/hardening/ & .../security/brute-force/
• XML-RPC abuse & mitigation overviews: wpbeginner.com/.../disable-xml-rpc, liquidweb.com/.../brute-force-attacks/