Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <CA+8F9hicnF=kvjXPZFQy=Pa2HJUS3JS+G9VswFHNQQynPMHGVQ@mail.gmail.com>
 <20190424193903.swlfmfuo6cqnpkwa@dcvr> <20190427093319.sgicqik2oqkez3wk@dcvr>
 <CABeXuvrY9QdvF1gTfiMt-eVp7VtobwG9xzjQFkErq+3wpW_P3Q@mail.gmail.com>
 <20190428004858.el3yk6hljloeoxza@dcvr> <20190429204754.hkz7z736tdk4ucum@linux-r8p5>
 <20190429210427.dmfemfft2t2gdwko@dcvr>
In-Reply-To: <20190429210427.dmfemfft2t2gdwko@dcvr>
From:   Deepa Dinamani <deepa.kernel@gmail.com>
Date:   Tue, 30 Apr 2019 14:07:07 -0700
Message-ID: <CABeXuvqpAjk8ocRUabVU4Yviv7kgRkMneLE1Xy-jAtHdXAHBVw@mail.gmail.com>
Subject: Re: Strange issues with epoll since 5.0
To:     Eric Wong <e@80x24.org>
Cc:     Davidlohr Bueso <dave@stgolabs.net>, Arnd Bergmann <arnd@arndb.de>,
        Al Viro <viro@zeniv.linux.org.uk>,
        Jason Baron <jbaron@akamai.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Omar Kilani <omar.kilani@gmail.com>,
        Linux FS-devel Mailing List <linux-fsdevel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

I was also not able to reproduce this.
Arnd and I were talking about this today morning. Here is something
Arnd noticed:

If there was a signal after do_epoll_wait(), we never were not
entering the if (err = -EINTR) at all before. But, now we do.
We could try with the below patch:

diff --git a/fs/eventpoll.c b/fs/eventpoll.c
index 4a0e98d87fcc..5cfb800cf598 100644
--- a/fs/eventpoll.c
+++ b/fs/eventpoll.c
@@ -2330,7 +2330,7 @@ SYSCALL_DEFINE6(epoll_pwait, int, epfd, struct
epoll_event __user *, events,

        error = do_epoll_wait(epfd, events, maxevents, timeout);

-       restore_user_sigmask(sigmask, &sigsaved);
+       restore_user_sigmask(sigmask, &sigsaved, error == -EITNR);

        return error;
 }

diff --git a/kernel/signal.c b/kernel/signal.c
index 3a9e41197d46..4a8f96f5c1c0 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2849,7 +2849,7 @@ EXPORT_SYMBOL(set_compat_user_sigmask);
  * This is useful for syscalls such as ppoll, pselect, io_pgetevents and
  * epoll_pwait where a new sigmask is passed in from userland for the syscalls.
  */
-void restore_user_sigmask(const void __user *usigmask, sigset_t *sigsaved)
+void restore_user_sigmask(const void __user *usigmask, sigset_t
*sigsaved, int sig_pending)
 {

        if (!usigmask)
@@ -2859,7 +2859,7 @@ void restore_user_sigmask(const void __user
*usigmask, sigset_t *sigsaved)
         * Restoring sigmask here can lead to delivering signals that the above
         * syscalls are intended to block because of the sigmask passed in.
         */
-       if (signal_pending(current)) {
+       if (sig_pending) {
                current->saved_sigmask = *sigsaved;
                set_restore_sigmask();


If this works that means we know what is busted.
I'm not sure what the hang in the userspace is about. Is it because
the syscall did not return an error or the particular signal was
blocked etc.

There are also a few timing differences also. But, can we try this first?

-Deepa