Re: [PATCH] nfs: don't call blocking operations while !TASK_RUNNING

[Jeff Layton <jeff.layton@primarydata.com>]

On Wed, 14 Jan 2015 13:06:17 -0500
Trond Myklebust <trond.myklebust@primarydata.com> wrote:

> On Wed, Jan 14, 2015 at 12:27 PM, Jeff Layton
> <jeff.layton@primarydata.com> wrote:
> > On Wed, 14 Jan 2015 12:13:43 -0500
> > Trond Myklebust <trond.myklebust@primarydata.com> wrote:
> >
> >> On Wed, Jan 14, 2015 at 10:32 AM, Jeff Layton <jlayton@primarydata.com> wrote:
> >> > Bruce reported seeing this warning pop when mounting using v4.1:
> >> >
> >> >      ------------[ cut here ]------------
> >> >      WARNING: CPU: 1 PID: 1121 at kernel/sched/core.c:7300 __might_sleep+0xbd/0xd0()
> >> >     do not call blocking ops when !TASK_RUNNING; state=1 set at [<ffffffff810ff58f>] prepare_to_wait+0x2f/0x90
> >> >     Modules linked in: rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver nfs lockd grace sunrpc fscache ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 xt_conntrack ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw snd_hda_codec_generic snd_hda_intel snd_hda_controller snd_hda_codec snd_hwdep snd_pcm snd_timer ppdev joydev snd virtio_console virtio_balloon pcspkr serio_raw parport_pc parport pvpanic floppy soundcore i2c_piix4 virtio_blk virtio_net qxl drm_kms_helper ttm drm virtio_pci virtio_ring ata_generic virtio pata_acpi
> >> >     CPU: 1 PID: 1121 Comm: nfsv4.1-svc Not tainted 3.19.0-rc4+ #25
> >> >     Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.7.5-20140709_153950- 04/01/2014
> >> >      0000000000000000 000000004e5e3f73 ffff8800b998fb48 ffffffff8186ac78
> >> >      0000000000000000 ffff8800b998fba0 ffff8800b998fb88 ffffffff810ac9da
> >> >      ffff8800b998fb68 ffffffff81c923e7 00000000000004d9 0000000000000000
> >> >     Call Trace:
> >> >      [<ffffffff8186ac78>] dump_stack+0x4c/0x65
> >> >      [<ffffffff810ac9da>] warn_slowpath_common+0x8a/0xc0
> >> >      [<ffffffff810aca65>] warn_slowpath_fmt+0x55/0x70
> >> >      [<ffffffff810ff58f>] ? prepare_to_wait+0x2f/0x90
> >> >      [<ffffffff810ff58f>] ? prepare_to_wait+0x2f/0x90
> >> >      [<ffffffff810dd2ad>] __might_sleep+0xbd/0xd0
> >> >      [<ffffffff8124c973>] kmem_cache_alloc_trace+0x243/0x430
> >> >      [<ffffffff810d941e>] ? groups_alloc+0x3e/0x130
> >> >      [<ffffffff810d941e>] groups_alloc+0x3e/0x130
> >> >      [<ffffffffa0301b1e>] svcauth_unix_accept+0x16e/0x290 [sunrpc]
> >> >      [<ffffffffa0300571>] svc_authenticate+0xe1/0xf0 [sunrpc]
> >> >      [<ffffffffa02fc564>] svc_process_common+0x244/0x6a0 [sunrpc]
> >> >      [<ffffffffa02fd044>] bc_svc_process+0x1c4/0x260 [sunrpc]
> >> >      [<ffffffffa03d5478>] nfs41_callback_svc+0x128/0x1f0 [nfsv4]
> >> >      [<ffffffff810ff970>] ? wait_woken+0xc0/0xc0
> >> >      [<ffffffffa03d5350>] ? nfs4_callback_svc+0x60/0x60 [nfsv4]
> >> >      [<ffffffff810d45bf>] kthread+0x11f/0x140
> >> >      [<ffffffff810ea815>] ? local_clock+0x15/0x30
> >> >      [<ffffffff810d44a0>] ? kthread_create_on_node+0x250/0x250
> >> >      [<ffffffff81874bfc>] ret_from_fork+0x7c/0xb0
> >> >      [<ffffffff810d44a0>] ? kthread_create_on_node+0x250/0x250
> >> >     ---[ end trace 675220a11e30f4f2 ]---
> >> >
> >> > nfs41_callback_svc does most of its work while in TASK_INTERRUPTIBLE,
> >> > which is just wrong. Fix that by finishing the wait immediately if we've
> >> > found that the list has something on it.
> >>
> >> ACK.
> >>
> >> > Also, we don't expect this kthread to accept signals, so we should be
> >> > using a TASK_UNINTERRUPTIBLE sleep instead.
> >>
> >> Umm... Won't that end up triggering the hung task watchdog for every
> >> 120seconds with no callback activity?
> >>
> >
> > Doh! You're correct.
> >
> > What's the right way to do this then? Do we need to use
> > schedule_timeout and wake up every 100s or so? Using TASK_INTERRUPTIBLE
> > to work around the watchdog seems wrong but I guess we can live with
> > that in the short term if it's the only way.
> 
> The alternative is to use TASK_KILLABLE. That's a little more
> restrictive, but still a PITA. Note the both TASK_INTERRUPTIBLE and
> TASK_KILLABLE will need something to handle that signalled() case,
> otherwise we end up with a permanent busy-waiting loop.
> 

Heh, our emails appear to have crossed. I went with a schedule_timeout
for v2 so we can sidestep the whole issue of signals.

It's simple enough to add a flush_signals() call at the end of the
loop, but I'm a little leery of what effects we might see if the thread
catches a signal while in the middle of doing real work.

If you really think that's the better approach, I can do a v3 patch
that does that however.

-- 
Jeff Layton <jlayton@primarydata.com>