From: Willy Tarreau <w@1wt.eu>
Date: Tue, 24 Jan 2023 11:11:41 +0000 (+0100)
Subject: BUG/MINOR: sink: make sure to always properly unmap a file-backed ring
X-Git-Tag: v2.8-dev3~98
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=fb9a4765b74cc18f25a6b732a66aae952262a7e0;p=thirdparty%2Fhaproxy.git

BUG/MINOR: sink: make sure to always properly unmap a file-backed ring

The munmap() call performed on exit was incorrect since it used to apply
to the buffer instead of the area, so neither the pointer nor the size
were page-aligned. This patches corrects this and also adds a call to
msync() since munmap() alone doesn't guarantee that data will be dumped.

This should be backported to 2.6.
---

diff --git a/src/sink.c b/src/sink.c
index 862a1ae322..0aae5f2a7e 100644
--- a/src/sink.c
+++ b/src/sink.c
@@ -1392,8 +1392,13 @@ static void sink_deinit()
 
 	list_for_each_entry_safe(sink, sb, &sink_list, sink_list) {
 		if (sink->type == SINK_TYPE_BUFFER) {
-			if (sink->store)
-				munmap(sink->ctx.ring->buf.area, sink->ctx.ring->buf.size);
+			if (sink->store) {
+				size_t size = (sink->ctx.ring->buf.size + 4095UL) & -4096UL;
+				void *area = (sink->ctx.ring->buf.area - sizeof(*sink->ctx.ring));
+
+				msync(area, size, MS_SYNC);
+				munmap(area, size);
+			}
 			else
 				ring_free(sink->ctx.ring);
 		}