2016-11-11 17:53:36

by Wolfgang Walter

[permalink] [raw]
Subject: DRBD: ASSERT in drbd_actlog.c:259

Hello,

when I execute

mkfs.ext4 -J device=UUID=625d871f-c278-4acb-916d-774dc78dbd8a -v -b 4096 -E stride=$((512/4)),stripe_width=$((512/4*3)),lazy_itable_init=0 -O inline_data,mmp -L dyn -m .01 /dev/test/big

the command hangs and the kernel (4.4.30) complains:

block drbd1: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259
block drbd2: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259
block drbd1: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259
block drbd3: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259
block drbd2: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259
block drbd3: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259

This does _not_ happen if I use

mkfs.ext4 -J device=UUID=625d871f-c278-4acb-916d-774dc78dbd8a -v -b 4096 -E stride=$((512/4)),stripe_width=$((512/4*3)),lazy_itable_init=0, nodiscard -O inline_data,mmp -L dyn -m .01 /dev/export/dyn

drbd1, drbd2 and drbd3 are three volumes of one resource. These three drbd devices are physical volumes of a volume group test with an LV big. test/big uses drbd1 ... drbd3 as backing devices (striped; stripesize ist 512k.


Regards,
--
Wolfgang Walter
Studentenwerk M?nchen
Anstalt des ?ffentlichen Rechts


2016-11-14 10:11:25

by Lars Ellenberg

[permalink] [raw]
Subject: Re: [Drbd-dev] DRBD: ASSERT in drbd_actlog.c:259

On Fri, Nov 11, 2016 at 06:45:18PM +0100, Wolfgang Walter wrote:
> Hello,
>
> when I execute
>
> mkfs.ext4 -J device=UUID=625d871f-c278-4acb-916d-774dc78dbd8a -v -b 4096 -E stride=$((512/4)),stripe_width=$((512/4*3)),lazy_itable_init=0 -O inline_data,mmp -L dyn -m .01 /dev/test/big
>
> the command hangs

It should not hang, really, but it is probably too busy spewing below
message to the console over and over again.

There have been some problems with our trim/discard handling in some situations.
That's also why this does not happen if you say nodiscard.

> and the kernel (4.4.30) complains:
>
> block drbd1: ASSERT( (unsigned)(last - first) <= 1 ) in drivers/block/drbd/drbd_actlog.c:259

see:
commit 505675f96cf0f169647a18c3dda1f373eca957b1
Author: Lars Ellenberg <[email protected]>
Date: Tue Jun 14 00:26:23 2016 +0200

drbd: allow larger max_discard_sectors

Make sure we have at least 67 (> AL_UPDATES_PER_TRANSACTION)
al-extents available, and allow up to half of that to be
discarded in one bio.

Signed-off-by: Philipp Reisner <[email protected]>
Signed-off-by: Lars Ellenberg <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index d524973..265b2b6 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -258,7 +258,7 @@ bool drbd_al_begin_io_fastpath(struct drbd_device *device, struct drbd_interval
unsigned first = i->sector >> (AL_EXTENT_SHIFT-9);
unsigned last = i->size == 0 ? first : (i->sector + (i->size >> 9) - 1) >> (AL_EXTENT_SHIFT-9);

- D_ASSERT(device, (unsigned)(last - first) <= 1);
+ D_ASSERT(device, first <= last);
D_ASSERT(device, atomic_read(&device->local_cnt) > 0);

/* FIXME figure out a fast path for bios crossing AL extent boundaries */

...
and some more chunks, but you should not cherry-pick stuff "just so",
if it may change semantics ...

kernel v4.4 is roughly equivalent to out-of-tree drbd 8.4.5,
we are at 8.4.9 meanwhile, so you may want to try that as well.

--
: Lars Ellenberg
: LINBIT | Keeping the Digital World Running
: DRBD -- Heartbeat -- Corosync -- Pacemaker
: R&D, Integration, Ops, Consulting, Support

DRBD? and LINBIT? are registered trademarks of LINBIT