From mboxrd@z Thu Jan 1 00:00:00 1970 From: =?ISO-8859-15?Q?Jere_Lepp=E4nen?= Date: Thu, 30 Apr 2020 08:54:11 +0000 Subject: Re: [PATCHv2] sctp: Don't add the shutdown timer if its already been added Message-Id: MIME-Version: 1 Content-Type: multipart/mixed; boundary="352996365-1736932537-1588236862=:629159" List-Id: References: <1458334046-26323-1-git-send-email-nhorman@tuxdriver.com> In-Reply-To: <1458334046-26323-1-git-send-email-nhorman@tuxdriver.com> To: linux-sctp@vger.kernel.org --352996365-1736932537-1588236862=:629159 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 8bit On Wed, 29 Apr 2020, Neil Horman wrote: > This BUG halt was reported a while back, but the patch somehow got > missed: > > PID: 2879 TASK: c16adaa0 CPU: 1 COMMAND: "sctpn" > #0 [f418dd28] crash_kexec at c04a7d8c > #1 [f418dd7c] oops_end at c0863e02 > #2 [f418dd90] do_invalid_op at c040aaca > #3 [f418de28] error_code (via invalid_op) at c08631a5 > EAX: f34baac0 EBX: 00000090 ECX: f418deb0 EDX: f5542950 EBP: 00000000 > DS: 007b ESI: f34ba800 ES: 007b EDI: f418dea0 GS: 00e0 > CS: 0060 EIP: c046fa5e ERR: ffffffff EFLAGS: 00010286 > #4 [f418de5c] add_timer at c046fa5e > #5 [f418de68] sctp_do_sm at f8db8c77 [sctp] > #6 [f418df30] sctp_primitive_SHUTDOWN at f8dcc1b5 [sctp] > #7 [f418df48] inet_shutdown at c080baf9 > #8 [f418df5c] sys_shutdown at c079eedf > #9 [f418df70] sys_socketcall at c079fe88 > EAX: ffffffda EBX: 0000000d ECX: bfceea90 EDX: 0937af98 > DS: 007b ESI: 0000000c ES: 007b EDI: b7150ae4 > SS: 007b ESP: bfceea7c EBP: bfceeaa8 GS: 0033 > CS: 0073 EIP: b775c424 ERR: 00000066 EFLAGS: 00000282 > > It appears that the side effect that starts the shutdown timer was processed > multiple times, which can happen as multiple paths can trigger it. This of > course leads to the BUG halt in add_timer getting called. > > Fix seems pretty straightforward, just check before the timer is added if its > already been started. If it has mod the timer instead to min(current > expiration, new expiration) > > Its been tested but not confirmed to fix the problem, as the issue has only > occured in production environments where test kernels are enjoined from being > installed. It appears to be a sane fix to me though. Also, recentely, > Jere found a reproducer posted on list to confirm that this resolves the > issues > > Signed-off-by: Neil Horman > CC: Vlad Yasevich > CC: "David S. Miller" > CC: Jere Leppänen > CC: marcelo.leitner@gmail.com > > --- > Change notes: > V2) Updated to use timer_reduce > --- > net/sctp/sm_sideeffect.c | 14 +++++++++++--- > 1 file changed, 11 insertions(+), 3 deletions(-) > > diff --git a/net/sctp/sm_sideeffect.c b/net/sctp/sm_sideeffect.c > index 2bc29463e1dc..9f36fe911d08 100644 > --- a/net/sctp/sm_sideeffect.c > +++ b/net/sctp/sm_sideeffect.c > @@ -1523,9 +1523,17 @@ static int sctp_cmd_interpreter(enum sctp_event_type event_type, > timeout = asoc->timeouts[cmd->obj.to]; > BUG_ON(!timeout); > > - timer->expires = jiffies + timeout; > - sctp_association_hold(asoc); > - add_timer(timer); > + /* > + * SCTP has a hard time with timer starts. Because we process > + * timer starts as side effects, it can be hard to tell if we > + * have already started a timer or not, which leads to BUG > + * halts when we call add_timer. So here, instead of just starting > + * a timer, if the timer is already started, and just mod > + * the timer with the shorter of the two expiration times > + */ > + if (!timer_pending(timer)) > + sctp_association_hold(asoc); > + timer_reduce(timer, jiffies + timeout); > break; > > case SCTP_CMD_TIMER_RESTART: > -- > 2.25.4 Thanks Neil. This patch works for me, but I only tested the one case I posted about. --352996365-1736932537-1588236862=:629159--