kumquat-buildroot/package/fakeroot/0002-communicate-check-return-status-of-msgrcv.patch
Yann E. MORIN eff989bab8 package/fakeroot: fix highly parallel uses
Although the issue can very well occur with low-paralle builds, or even
with non-parallel builds, the conditions are so strict that the ocasion
it breaks is extremely rare, to the point where a failure would go
unnoticed.

Fixes #10141.

Reported-by: Clayton Shotwell <clayton.shotwell@rockwellcollins.com>
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
Cc: Clayton Shotwell <clayton.shotwell@rockwellcollins.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
2017-08-15 17:39:26 +02:00

47 lines
1.4 KiB
Diff

From a853f21633693f9eefc4949660253a5328d2d2f3 Mon Sep 17 00:00:00 2001
From: "Yann E. MORIN" <yann.morin.1998@free.fr>
Date: Sun, 13 Aug 2017 23:21:54 +0200
Subject: [PATCH 1/1] communicate: check return status of msgrcv()
msgrcv can return with -1 to indicate an error condition.
One such error is to have been interrupted by a signal.
Being interrupted by a signal is very rare in this code, except in a
very special condition: a highly-parallel (1000 jobs!) mksquashfs on
a filesystem with extended attributes, where we see errors like (those
are mksquashfs errors):
llistxattr for titi/603/883 failed in read_attrs, because Unknown
error 1716527536
See: https://bugs.busybox.net/show_bug.cgi?id=10141
In this case, we just have to retry the call to msgrcv().
Signed-off-by: "Yann E. MORIN" <yann.morin.1998@free.fr>
---
communicate.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/communicate.c b/communicate.c
index 293f404..787bb63 100644
--- a/communicate.c
+++ b/communicate.c
@@ -553,10 +553,13 @@ void send_get_fakem(struct fake_msg *buf)
l=msgrcv(msg_get,
(struct my_msgbuf*)buf,
sizeof(*buf)-sizeof(buf->mtype),0,0);
- while((buf->serial!=serial)||buf->pid!=pid);
+ while(((l==-1)&&(errno==EINTR))||(buf->serial!=serial)||buf->pid!=pid);
semaphore_down();
+ if(l==-1)
+ buf->xattr.flags_rc=errno;
+
/*
(nah, may be wrong, due to allignment)
--
2.11.0