From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 8E3442676F9; Thu, 13 Mar 2025 13:15:46 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=10.30.226.201 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1741871746; cv=none; b=aVQit0QNT1Vop4YBqswKrA3/sW5IInevxXlRqlrdNUJGufdMGakWzKN8uoeob2CYI9X4T9sPWbos8BJ7/RJzR6zvB0alBCd64ivOejC9cAejU6SA9YVVYmfjVW0uf6sUP9C6iwLoTb0B72DdABIXFqXCwIqyokHMtlTyk4dQnhw= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1741871746; c=relaxed/simple; bh=ccTjPdt350IulInM4Wn/+yqS5sSqqhAdPy79Zmp2csM=; h=Message-ID:Subject:From:To:Cc:Date:In-Reply-To:References: Content-Type:MIME-Version; b=R1ttl2vuYf1xaEERoPyA9lcVuh7rQMwpvXZBviu2EfYfJ76fGsUa7h3Gag3ZiPxUTcVWzB5NqXba1P7Arp7Y9XJXXfvbsXNmYvWJyiGcW+95flAgNroWwxGN2zj7rhgPrGBL+Q0cO6AU+4lDOF9T90i2K0vI40OB2ZH91x456uo= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=bj6IXNwe; arc=none smtp.client-ip=10.30.226.201 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="bj6IXNwe" Received: by smtp.kernel.org (Postfix) with ESMTPSA id 9537CC4CEDD; Thu, 13 Mar 2025 13:15:45 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1741871746; bh=ccTjPdt350IulInM4Wn/+yqS5sSqqhAdPy79Zmp2csM=; h=Subject:From:To:Cc:Date:In-Reply-To:References:From; b=bj6IXNweAqpmG2FzKjyNhFoR+K/eE/MTr4uhH1UjDENBsjvQAcJi/aQkpwY8kiw3q WtshW9vcjTjZyxk+70y4CnOxn6NPtuV+J8OtIw51oAprxH6RTem58fr0EDums+jNvt C8OGsI8SxfS84fUkITtNIZ/SHm66sIaOw7ScYHkTE36ncER1IrcIb2UYh1NpMDrasA 9NfuTScxjBTsRDebvWdhk+UlmC6ONAP/HGyRqIRTrAXe+XY6UjA6OdPu7sZ1vFO+1p TutJ/nMPl6PunXY+TWGyJPgH63uDnSSHTZW+Dp4RsZWwPVm1Fl0shJEaiGWVyYuNbU db+8iKi8ozBIw== Message-ID: Subject: Re: [PATCH] sunrpc: add a rpc_clnt shutdown control in debugfs From: Jeff Layton To: Trond Myklebust , "bcodding@redhat.com" Cc: "anna@kernel.org" , "linux-nfs@vger.kernel.org" , "linux-kernel@vger.kernel.org" Date: Thu, 13 Mar 2025 09:15:44 -0400 In-Reply-To: <8bf55a6fb64ba9e21a1ec8c5644355ffd6496c6e.camel@hammerspace.com> References: <20250312-rpc-shutdown-v1-1-cc90d79a71c2@kernel.org> <7906109F-91D2-4ECF-B868-5519B56D2CEE@redhat.com> <997f992f951bd235953c5f0e2959da6351a65adb.camel@kernel.org> <8bf55a6fb64ba9e21a1ec8c5644355ffd6496c6e.camel@hammerspace.com> Autocrypt: addr=jlayton@kernel.org; prefer-encrypt=mutual; keydata=mQINBE6V0TwBEADXhJg7s8wFDwBMEvn0qyhAnzFLTOCHooMZyx7XO7dAiIhDSi7G1NPxw n8jdFUQMCR/GlpozMFlSFiZXiObE7sef9rTtM68ukUyZM4pJ9l0KjQNgDJ6Fr342Htkjxu/kFV1Wv egyjnSsFt7EGoDjdKqr1TS9syJYFjagYtvWk/UfHlW09X+jOh4vYtfX7iYSx/NfqV3W1D7EDi0PqV T2h6v8i8YqsATFPwO4nuiTmL6I40ZofxVd+9wdRI4Db8yUNA4ZSP2nqLcLtFjClYRBoJvRWvsv4lm 0OX6MYPtv76hka8lW4mnRmZqqx3UtfHX/hF/zH24Gj7A6sYKYLCU3YrI2Ogiu7/ksKcl7goQjpvtV YrOOI5VGLHge0awt7bhMCTM9KAfPc+xL/ZxAMVWd3NCk5SamL2cE99UWgtvNOIYU8m6EjTLhsj8sn VluJH0/RcxEeFbnSaswVChNSGa7mXJrTR22lRL6ZPjdMgS2Km90haWPRc8Wolcz07Y2se0xpGVLEQ cDEsvv5IMmeMe1/qLZ6NaVkNuL3WOXvxaVT9USW1+/SGipO2IpKJjeDZfehlB/kpfF24+RrK+seQf CBYyUE8QJpvTZyfUHNYldXlrjO6n5MdOempLqWpfOmcGkwnyNRBR46g/jf8KnPRwXs509yAqDB6sE LZH+yWr9LQZEwARAQABtCVKZWZmIExheXRvbiA8amxheXRvbkBwb29jaGllcmVkcy5uZXQ+iQI7BB MBAgAlAhsDBgsJCAcDAgYVCAIJCgsEFgIDAQIeAQIXgAUCTpXWPAIZAQAKCRAADmhBGVaCFc65D/4 gBLNMHopQYgG/9RIM3kgFCCQV0pLv0hcg1cjr+bPI5f1PzJoOVi9s0wBDHwp8+vtHgYhM54yt43uI 7Htij0RHFL5eFqoVT4TSfAg2qlvNemJEOY0e4daljjmZM7UtmpGs9NN0r9r50W82eb5Kw5bc/r0km R/arUS2st+ecRsCnwAOj6HiURwIgfDMHGPtSkoPpu3DDp/cjcYUg3HaOJuTjtGHFH963B+f+hyQ2B rQZBBE76ErgTDJ2Db9Ey0kw7VEZ4I2nnVUY9B5dE2pJFVO5HJBMp30fUGKvwaKqYCU2iAKxdmJXRI ONb7dSde8LqZahuunPDMZyMA5+mkQl7kpIpR6kVDIiqmxzRuPeiMP7O2FCUlS2DnJnRVrHmCljLkZ Wf7ZUA22wJpepBligemtSRSbqCyZ3B48zJ8g5B8xLEntPo/NknSJaYRvfEQqGxgk5kkNWMIMDkfQO lDSXZvoxqU9wFH/9jTv1/6p8dHeGM0BsbBLMqQaqnWiVt5mG92E1zkOW69LnoozE6Le+12DsNW7Rj iR5K+27MObjXEYIW7FIvNN/TQ6U1EOsdxwB8o//Yfc3p2QqPr5uS93SDDan5ehH59BnHpguTc27Xi QQZ9EGiieCUx6Zh2ze3X2UW9YNzE15uKwkkuEIj60NvQRmEDfweYfOfPVOueC+iFifbQgSmVmZiBM YXl0b24gPGpsYXl0b25AcmVkaGF0LmNvbT6JAjgEEwECACIFAk6V0q0CGwMGCwkIBwMCBhUIAgkKC wQWAgMBAh4BAheAAAoJEAAOaEEZVoIViKUQALpvsacTMWWOd7SlPFzIYy2/fjvKlfB/Xs4YdNcf9q LqF+lk2RBUHdR/dGwZpvw/OLmnZ8TryDo2zXVJNWEEUFNc7wQpl3i78r6UU/GUY/RQmOgPhs3epQC 3PMJj4xFx+VuVcf/MXgDDdBUHaCTT793hyBeDbQuciARDJAW24Q1RCmjcwWIV/pgrlFa4lAXsmhoa c8UPc82Ijrs6ivlTweFf16VBc4nSLX5FB3ls7S5noRhm5/Zsd4PGPgIHgCZcPgkAnU1S/A/rSqf3F LpU+CbVBDvlVAnOq9gfNF+QiTlOHdZVIe4gEYAU3CUjbleywQqV02BKxPVM0C5/oVjMVx3bri75n1 TkBYGmqAXy9usCkHIsG5CBHmphv9MHmqMZQVsxvCzfnI5IO1+7MoloeeW/lxuyd0pU88dZsV/riHw 87i2GJUJtVlMl5IGBNFpqoNUoqmvRfEMeXhy/kUX4Xc03I1coZIgmwLmCSXwx9MaCPFzV/dOOrju2 xjO+2sYyB5BNtxRqUEyXglpujFZqJxxau7E0eXoYgoY9gtFGsspzFkVNntamVXEWVVgzJJr/EWW0y +jNd54MfPRqH+eCGuqlnNLktSAVz1MvVRY1dxUltSlDZT7P2bUoMorIPu8p7ZCg9dyX1+9T6Muc5d Hxf/BBP/ir+3e8JTFQBFOiLNdFtB9KZWZmIExheXRvbiA8amxheXRvbkBzYW1iYS5vcmc+iQI4BBM BAgAiBQJOldK9AhsDBgsJCAcDAgYVCAIJCgsEFgIDAQIeAQIXgAAKCRAADmhBGVaCFWgWD/0ZRi4h N9FK2BdQs9RwNnFZUr7JidAWfCrs37XrA/56olQl3ojn0fQtrP4DbTmCuh0SfMijB24psy1GnkPep naQ6VRf7Dxg/Y8muZELSOtsv2CKt3/02J1BBitrkkqmHyni5fLLYYg6fub0T/8Kwo1qGPdu1hx2BQ RERYtQ/S5d/T0cACdlzi6w8rs5f09hU9Tu4qV1JLKmBTgUWKN969HPRkxiojLQziHVyM/weR5Reu6 FZVNuVBGqBD+sfk/c98VJHjsQhYJijcsmgMb1NohAzwrBKcSGKOWJToGEO/1RkIN8tqGnYNp2G+aR 685D0chgTl1WzPRM6mFG1+n2b2RR95DxumKVpwBwdLPoCkI24JkeDJ7lXSe3uFWISstFGt0HL8Eew P8RuGC8s5h7Ct91HMNQTbjgA+Vi1foWUVXpEintAKgoywaIDlJfTZIl6Ew8ETN/7DLy8bXYgq0Xzh aKg3CnOUuGQV5/nl4OAX/3jocT5Cz/OtAiNYj5mLPeL5z2ZszjoCAH6caqsF2oLyAnLqRgDgR+wTQ T6gMhr2IRsl+cp8gPHBwQ4uZMb+X00c/Amm9VfviT+BI7B66cnC7Zv6Gvmtu2rEjWDGWPqUgccB7h dMKnKDthkA227/82tYoFiFMb/NwtgGrn5n2vwJyKN6SEoygGrNt0SI84y6hEVbQlSmVmZiBMYXl0b 24gPGpsYXl0b25AcHJpbWFyeWRhdGEuY29tPokCOQQTAQIAIwUCU4xmKQIbAwcLCQgHAwIBBhUIAg kKCwQWAgMBAh4BAheAAAoJEAAOaEEZVoIV1H0P/j4OUTwFd7BBbpoSp695qb6HqCzWMuExsp8nZjr uymMaeZbGr3OWMNEXRI1FWNHMtcMHWLP/RaDqCJil28proO+PQ/yPhsr2QqJcW4nr91tBrv/MqItu AXLYlsgXqp4BxLP67bzRJ1Bd2x0bWXurpEXY//VBOLnODqThGEcL7jouwjmnRh9FTKZfBDpFRaEfD FOXIfAkMKBa/c9TQwRpx2DPsl3eFWVCNuNGKeGsirLqCxUg5kWTxEorROppz9oU4HPicL6rRH22Ce 6nOAON2vHvhkUuO3GbffhrcsPD4DaYup4ic+DxWm+DaSSRJ+e1yJvwi6NmQ9P9UAuLG93S2MdNNbo sZ9P8k2mTOVKMc+GooI9Ve/vH8unwitwo7ORMVXhJeU6Q0X7zf3SjwDq2lBhn1DSuTsn2DbsNTiDv qrAaCvbsTsw+SZRwF85eG67eAwouYk+dnKmp1q57LDKMyzysij2oDKbcBlwB/TeX16p8+LxECv51a sjS9TInnipssssUDrHIvoTTXWcz7Y5wIngxDFwT8rPY3EggzLGfK5Zx2Q5S/N0FfmADmKknG/D8qG IcJE574D956tiUDKN4I+/g125ORR1v7bP+OIaayAvq17RP+qcAqkxc0x8iCYVCYDouDyNvWPGRhbL UO7mlBpjW9jK9e2fvZY9iw3QzIPGKtClKZWZmIExheXRvbiA8amVmZi5sYXl0b25AcHJpbWFyeWRh dGEuY29tPokCOQQTAQIAIwUCU4xmUAIbAwcLCQgHAwIBBhUIAgkKCwQWAgMBAh4BAheAAAoJEAAOa EEZVoIVzJoQALFCS6n/FHQS+hIzHIb56JbokhK0AFqoLVzLKzrnaeXhE5isWcVg0eoV2oTScIwUSU apy94if69tnUo4Q7YNt8/6yFM6hwZAxFjOXR0ciGE3Q+Z1zi49Ox51yjGMQGxlakV9ep4sV/d5a50 M+LFTmYSAFp6HY23JN9PkjVJC4PUv5DYRbOZ6Y1+TfXKBAewMVqtwT1Y+LPlfmI8dbbbuUX/kKZ5d dhV2736fgyfpslvJKYl0YifUOVy4D1G/oSycyHkJG78OvX4JKcf2kKzVvg7/Rnv+AueCfFQ6nGwPn 0P91I7TEOC4XfZ6a1K3uTp4fPPs1Wn75X7K8lzJP/p8lme40uqwAyBjk+IA5VGd+CVRiyJTpGZwA0 jwSYLyXboX+Dqm9pSYzmC9+/AE7lIgpWj+3iNisp1SWtHc4pdtQ5EU2SEz8yKvDbD0lNDbv4ljI7e flPsvN6vOrxz24mCliEco5DwhpaaSnzWnbAPXhQDWb/lUgs/JNk8dtwmvWnqCwRqElMLVisAbJmC0 BhZ/Ab4sph3EaiZfdXKhiQqSGdK4La3OTJOJYZphPdGgnkvDV9Pl1QZ0ijXQrVIy3zd6VCNaKYq7B AKidn5g/2Q8oio9Tf4XfdZ9dtwcB+bwDJFgvvDYaZ5bI3ln4V3EyW5i2NfXazz/GA/I/ZtbsigCFc 8ftCBKZWZmIExheXRvbiA8amxheXRvbkBrZXJuZWwub3JnPokCOAQTAQIAIgUCWe8u6AIbAwYLCQg HAwIGFQgCCQoLBBYCAwECHgECF4AACgkQAA5oQRlWghUuCg/+Lb/xGxZD2Q1oJVAE37uW308UpVSD 2tAMJUvFTdDbfe3zKlPDTuVsyNsALBGclPLagJ5ZTP+Vp2irAN9uwBuacBOTtmOdz4ZN2tdvNgozz uxp4CHBDVzAslUi2idy+xpsp47DWPxYFIRP3M8QG/aNW052LaPc0cedYxp8+9eiVUNpxF4SiU4i9J DfX/sn9XcfoVZIxMpCRE750zvJvcCUz9HojsrMQ1NFc7MFT1z3MOW2/RlzPcog7xvR5ENPH19ojRD CHqumUHRry+RF0lH00clzX/W8OrQJZtoBPXv9ahka/Vp7kEulcBJr1cH5Wz/WprhsIM7U9pse1f1g Yy9YbXtWctUz8uvDR7shsQxAhX3qO7DilMtuGo1v97I/Kx4gXQ52syh/w6EBny71CZrOgD6kJwPVV AaM1LRC28muq91WCFhs/nzHozpbzcheyGtMUI2Ao4K6mnY+3zIuXPygZMFr9KXE6fF7HzKxKuZMJO aEZCiDOq0anx6FmOzs5E6Jqdpo/mtI8beK+BE7Va6ni7YrQlnT0i3vaTVMTiCThbqsB20VrbMjlhp f8lfK1XVNbRq/R7GZ9zHESlsa35ha60yd/j3pu5hT2xyy8krV8vGhHvnJ1XRMJBAB/UYb6FyC7S+m QZIQXVeAA+smfTT0tDrisj1U5x6ZB9b3nBg65kc= Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable User-Agent: Evolution 3.54.3 (3.54.3-1.fc41) Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 On Wed, 2025-03-12 at 22:31 +0000, Trond Myklebust wrote: > On Wed, 2025-03-12 at 10:37 -0400, Jeff Layton wrote: > > On Wed, 2025-03-12 at 09:52 -0400, Benjamin Coddington wrote: > > > On 12 Mar 2025, at 9:36, Jeff Layton wrote: > > >=20 > > > > There have been confirmed reports where a container with an NFS > > > > mount > > > > inside it dies abruptly, along with all of its processes, but the > > > > NFS > > > > client sticks around and keeps trying to send RPCs after the > > > > networking > > > > is gone. > > > >=20 > > > > We have a reproducer where if we SIGKILL a container with an NFS > > > > mount, > > > > the RPC clients will stick around indefinitely. The orchestrator > > > > does a MNT_DETACH unmount on the NFS mount, and then tears down > > > > the > > > > networking while there are still RPCs in flight. > > > >=20 > > > > Recently new controls were added[1] that allow shutting down an > > > > NFS > > > > mount. That doesn't help here since the mount namespace is > > > > detached from > > > > any tasks at this point. > > >=20 > > > That's interesting - seems like the orchestrator could just reorder > > > its > > > request to shutdown before detaching the mount namespace.=C2=A0 Not a= n > > > objection, > > > just wondering why the MNT_DETACH must come first. > > >=20 > >=20 > > The reproducer we have is to systemd-nspawn a container, mount up an > > NFS mount inside it, start some I/O on it with fio and then kill -9 > > the > > systemd running inside the container. There isn't much the > > orchestrator > > (root-level systemd) can do to at that point other than clean up > > what's > > left. > >=20 > > I'm still working on a way to reliably detect when this has happened. > > For now, we just have to notice that some clients aren't dying. > >=20 > > > > Transplant shutdown_client() to the sunrpc module, and give it a > > > > more > > > > distinct name. Add a new debugfs sunrpc/rpc_clnt/*/shutdown knob > > > > that > > > > allows the same functionality as the one in /sys/fs/nfs, but at > > > > the > > > > rpc_clnt level. > > > >=20 > > > > [1]: commit d9615d166c7e ("NFS: add sysfs shutdown knob"). > > > >=20 > > > > Signed-off-by: Jeff Layton > > >=20 > > > I have a TODO to patch Documentation/ for this knob mostly to write > > > warnings > > > because there are some potential "gotchas" here - for example you > > > can have > > > shared RPC clients and shutting down one of those can cause > > > problems for a > > > different mount (this is true today with the > > > /sys/fs/nfs/[bdi]/shutdown > > > knob).=C2=A0 Shutting down aribitrary clients will definitely break > > > things in > > > weird ways, its not a safe place to explore. > > >=20 > >=20 > > Yes, you really do need to know what you're doing. 0200 permissions > > are > > essential for this file, IOW. Thanks for the R-b! >=20 > Sorry, but NACK! We should not be adding control mechanisms to debugfs. >=20 Ok. Would adding sunrpc controls under sysfs be more acceptable? I do agree that this is a potential footgun, however. It would be nicer to clean this situation up automagically. > One thing that might work in situations like this is perhaps to make > use of the fact that we are monitoring whether or not rpc_pipefs is > mounted. So if the mount is containerised, and the orchestrator > unmounts everything, including rpc_pipefs, we might take that as a hint > that we should treat any future connection errors as being fatal. >=20 rpc_pipefs isn't being mounted at all in the container I'm using. I think that's not going to be a reliable test for this. > Otherwise, we'd have to be able to monitor the root task, and check if > it is still alive in order to figure out if out containerised world has > collapsed. >=20 If by the root task, you mean the initial task in the container, then that method seems a little sketchy too. How would we determine that from the RPC layer? To be clear: the situation here is that we have a container with a veth device that is communicating with the outside world. Once all of the processes in the container exit, the veth device in the container disappears. The rpc_xprt holds a ref on the netns though, so that sticks around trying to retransmit indefinitely. I think what we really need is a lightweight reference on the netns. Something where we can tell that there are no userland tasks that care about it anymore, so we can be more aggressive about giving up on it. There is a "passive" refcount inside struct net, but that's not quite what we need as it won't keep the sunrpc_net in place. What if instead of holding a netns reference in the xprt, we have it hold a reference on a new refcount_t that lives in sunrpc_net? Then, we add a pre_exit pernet_ops callback that does a shutdown_client() on all of the rpc_clnt's attached to the xprts in that netns. The pre_exit can then just block until the sunrpc_net refcount goes to 0. I think that would allow everything to be cleaned up properly? --=20 Jeff Layton