diff --git a/man/coredns-shed.7 b/man/coredns-shed.7
new file mode 100644
index 000000000..471ca551e
--- /dev/null
+++ b/man/coredns-shed.7
@@ -0,0 +1,106 @@
+.\" Generated by Mmark Markdown Processor - mmark.miek.nl
+.TH "COREDNS-SHED" 7 "August 2026" "CoreDNS" "CoreDNS Plugins"
+
+.SH "NAME"
+.PP
+\fIshed\fP \- serializes UDP response writes per listener socket and sheds load when the socket cannot keep up.
+
+.SH "DESCRIPTION"
+.PP
+UDP responses written back through one listener socket serialize on the Go runtime's internal
+fdMutex, which allows at most 2^20\-1 concurrent operations (holders plus waiters) per file
+descriptor and terminates the process with
+
+.PP
+.RS
+
+.nf
+panic: too many concurrent operations on a single file or socket (max 1048575)
+
+.fi
+.RE
+
+.PP
+when that is exceeded. CoreDNS serves UDP with one goroutine per query, all writing back through
+the shared packet connection, so when queries arrive faster than the socket's serialized writes
+drain, every excess in\-flight query parks its goroutine in that wait queue and nothing bounds the
+pile. Observed in production: ~2.8M goroutines and 60GiB RSS before the panic.
+
+.PP
+The \fIshed\fP plugin makes that panic structurally unreachable, per UDP listener socket:
+
+.IP \(bu 4
+\fBSingle writer\fP \- responses are not written by the handler goroutine. The packed response is
+pushed onto a bounded per\-socket stack (fixed depth 1024) and one writer goroutine per socket
+performs the wire writes, so the file descriptor never sees more than one writer. The stack
+evicts the oldest entry when full and the writer pops the newest first, so under overload the
+socket's residual capacity always goes to the freshest response. The depth is a fixed burst
+budget (roughly 12\-16ms of a typical socket's drain rate), not a tunable.
+.IP \(bu 4
+\fBCoupled shedding\fP \- while a socket's stack is full, arriving queries on that socket are
+dropped before any plugin runs; work admitted then would only produce a response destined for
+eviction. There is no configuration: the stack's fullness is the signal.
+
+
+.PP
+Drops are silent \- no response is written, so the client's resolver retries against another
+server, the standard load\-shedding contract for UDP DNS. Every drop is counted.
+
+.PP
+The plugin only acts on UDP; TCP queries pass through untouched. It can only be used in plain DNS
+server blocks (not \fItls\fP, \fIgrpc\fP, \fIhttps\fP or \fIquic\fP), which is enforced at startup. It should be
+listed before (above) the \fIprometheus\fP plugin in the plugin chain, so that shed drops are never
+counted as handled requests by the \fIprometheus\fP plugin \- which is where this plugin sits by
+default.
+
+.PP
+When several server blocks share a listener, any block with \fIshed\fP installs the write discipline
+for every write on that socket, while the pre\-chain shedding only runs in blocks that carry the
+directive \- keep it uniform across blocks sharing a listener. The discipline covers every response
+written through \fB\fCWriteMsg\fR, which is how every plugin responds; a plugin writing raw bytes with
+\fB\fCResponseWriter.Write\fR would bypass it.
+
+.SH "SYNTAX"
+.PP
+.RS
+
+.nf
+shed
+
+.fi
+.RE
+
+.PP
+The plugin takes no arguments.
+
+.SH "METRICS"
+.PP
+If monitoring is enabled (via the \fIprometheus\fP plugin) then the following metric is exported:
+
+.IP \(bu 4
+\fB\fCcoredns_shed_dropped_total{server, reason}\fR \- counter of dropped queries and responses. The
+\fB\fCreason\fR label is \fB\fCquery\fR for queries dropped before the plugin chain because the socket's
+stack was full, and \fB\fCresponse\fR for responses dropped at the write boundary (evicted by a newer
+response, failed to reach the wire, or arriving during shutdown).
+
+
+.SH "EXAMPLES"
+.PP
+Protect the UDP listener while forwarding:
+
+.PP
+.RS
+
+.nf
+\&. {
+    shed
+    forward . 8.8.8.8
+}
+
+.fi
+.RE
+
+.SH "SEE ALSO"
+.PP
+The fdMutex limit is enforced in \fB\fCGOROOT/src/internal/poll/fd_mutex.go\fR.
+