From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id C633FC433EF for ; Thu, 27 Jan 2022 11:59:21 +0000 (UTC) Received: from localhost ([::1]:48278 helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1nD3Qu-0001aq-OM for qemu-devel@archiver.kernel.org; Thu, 27 Jan 2022 06:59:20 -0500 Received: from eggs.gnu.org ([209.51.188.92]:46042) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1nD3BO-0000N4-T1 for qemu-devel@nongnu.org; Thu, 27 Jan 2022 06:43:20 -0500 Received: from us-smtp-delivery-44.mimecast.com ([205.139.111.44]:56368) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1nD3BK-0002W3-SW for qemu-devel@nongnu.org; Thu, 27 Jan 2022 06:43:17 -0500 Received: from mimecast-mx01.redhat.com (mimecast-mx01.redhat.com [209.132.183.4]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id us-mta-470-6npUO3xQODWU9tQU7ofu5g-1; Thu, 27 Jan 2022 06:43:01 -0500 X-MC-Unique: 6npUO3xQODWU9tQU7ofu5g-1 Received: from smtp.corp.redhat.com (int-mx05.intmail.prod.int.phx2.redhat.com [10.5.11.15]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx01.redhat.com (Postfix) with ESMTPS id 515BA81A274 for ; Thu, 27 Jan 2022 11:43:00 +0000 (UTC) Received: from bahia (unknown [10.39.192.166]) by smtp.corp.redhat.com (Postfix) with ESMTP id 70EC04F856; Thu, 27 Jan 2022 11:42:49 +0000 (UTC) Date: Thu, 27 Jan 2022 12:42:48 +0100 From: Greg Kurz To: Vivek Goyal Subject: Re: [Virtio-fs] [PATCH v4 1/2] virtiofsd: Track mounts Message-ID: <20220127124248.1c33492c@bahia> In-Reply-To: References: <20220125141213.361930-1-groug@kaod.org> <20220125141213.361930-2-groug@kaod.org> MIME-Version: 1.0 X-Scanned-By: MIMEDefang 2.79 on 10.5.11.15 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: kaod.org Content-Type: text/plain; charset=WINDOWS-1252 Content-Transfer-Encoding: quoted-printable Received-SPF: softfail client-ip=205.139.111.44; envelope-from=groug@kaod.org; helo=us-smtp-delivery-44.mimecast.com X-Spam_score_int: -18 X-Spam_score: -1.9 X-Spam_bar: - X-Spam_report: (-1.9 / 5.0 requ) BAYES_00=-1.9, RCVD_IN_DNSWL_LOW=-0.7, SPF_HELO_NONE=0.001, SPF_SOFTFAIL=0.665 autolearn=no autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: virtio-fs@redhat.com, qemu-devel@nongnu.org Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: "Qemu-devel" On Wed, 26 Jan 2022 18:02:46 -0500 Vivek Goyal wrote: > On Wed, Jan 26, 2022 at 05:47:09PM -0500, Vivek Goyal wrote: > > On Tue, Jan 25, 2022 at 03:12:11PM +0100, Greg Kurz wrote: > > > The upcoming implementation of ->sync_fs() needs to know about all > > > submounts in order to call syncfs() on them when virtiofsd is started > > > without '-o announce_submounts'. > > >=20 > > > Track every inode that comes up with a new mount id in a GHashTable. > > > If the mount id isn't available, e.g. no statx() on the host, fallbac= k > > > on the device id for the key. This is done during lookup because we > > > only care for the submounts that the client knows about. The inode > > > is removed from the hash table when ultimately unreferenced. This > > > can happen on a per-mount basis when the client posts a FUSE_FORGET > > > request or for all submounts at once with FUSE_DESTROY. > > >=20 > > > Signed-off-by: Greg Kurz > > > --- > > > tools/virtiofsd/passthrough_ll.c | 43 +++++++++++++++++++++++++++++-= -- > > > 1 file changed, 40 insertions(+), 3 deletions(-) > > >=20 > > > diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passt= hrough_ll.c > > > index 64b5b4fbb186..7bf31fc129c8 100644 > > > --- a/tools/virtiofsd/passthrough_ll.c > > > +++ b/tools/virtiofsd/passthrough_ll.c > > > @@ -117,6 +117,7 @@ struct lo_inode { > > > GHashTable *posix_locks; /* protected by lo_inode->plock_mutex *= / > > > =20 > > > mode_t filetype; > > > + bool is_mnt; > > > }; > > > =20 > > > struct lo_cred { > > > @@ -164,6 +165,7 @@ struct lo_data { > > > bool use_statx; > > > struct lo_inode root; > > > GHashTable *inodes; /* protected by lo->mutex */ > > > + GHashTable *mnt_inodes; /* protected by lo->mutex */ > > > struct lo_map ino_map; /* protected by lo->mutex */ > > > struct lo_map dirp_map; /* protected by lo->mutex */ > > > struct lo_map fd_map; /* protected by lo->mutex */ > > > @@ -1000,6 +1002,31 @@ static int do_statx(struct lo_data *lo, int di= rfd, const char *pathname, > > > return 0; > > > } > > > =20 > >=20 > > Hi Greg, > >=20 > > Thanks for the patches. Had a quick look. Overall these patches look > > pretty good to me. I will spend more time testing and having a=20 > > closer look. Some quick thoughts below. > >=20 > > > +static uint64_t mnt_inode_key(struct lo_inode *inode) > > > +{ > > > + /* Prefer mnt_id, fallback on dev */ > > > + return inode->key.mnt_id ? inode->key.mnt_id : inode->key.dev; > > > +} > >=20 > > I am not sure if we should use inode->key.dev. This might create proble= m > > if same file system is bind mounted at two paths in shared dir. So > > say /dev/sdb is mounted at foo1/ and then bind mounted at foo2/ in > > shared dir. A user looks up foo1/ and does some writes. Then we > > lookup foo2/ and release that inode. Release of foo2 will let go > > inode from the hash. And that means if later another write happens > > in foo1/ followed by syncfs(), we will not issue syncfs() on filesystem > > backed by /dev/sdb. > >=20 > > So what are the options. > >=20 > > A. Make mnt_id mandatory and do not implement it if mnt_id is not > > available. > >=20 > > B. Don't do anything and live with this. It is a corner case and > > still better than not implement submount syncfs at all. > >=20 > > C. Instead of adding lo_inode to hash, create another kind of object > > and reference count that. It could be a mount fd which we open > > when we add object for the first time. So when foo1/ inode is > > instantiated, create mountfd object, add it to hash table using > > device id as the key. When foo2 comes along, we find the object > > in the hash and just bump up the ref. Now this mountfd object > > will go away when both foo1 and foo2 inodes have been evicted > > and will take care of the issue I am referring to. >=20 > And we could take a ref on mountfd object only when we find an > inode whose parent's device id/mnt_id is different from us. That > way for every inode in the system we don't go through this exercise. > Just only those dir inodes which are a mount point. >=20 Good idea ! > Vivek >=20 > >=20 > > I guess B is little extra complexity but probably not too bad. > > WDYT. It sounds litter better than option A and B. > >=20 > >=20 > > > + > > > +static void add_mnt_inode(struct lo_data *lo, struct lo_inode *inode= ) > > > +{ > > > + uint64_t mnt_key =3D mnt_inode_key(inode); > > > + > > > + if (!g_hash_table_contains(lo->mnt_inodes, &mnt_key)) { > > > + inode->is_mnt =3D true; > > > + g_hash_table_insert(lo->mnt_inodes, &mnt_key, inode); > > > + } > > > +} > > > + > > > +static void remove_mnt_inode(struct lo_data *lo, struct lo_inode *in= ode) > > > +{ > > > + uint64_t mnt_key =3D mnt_inode_key(inode); > > > + > > > + if (inode->is_mnt) { > > > + g_hash_table_remove(lo->mnt_inodes, &mnt_key); > > > + } > > > +} > >=20 > > Should we issue syncfs() on this inode when we are removing it? It > > is possible guest did some writes, let go inode and later issued > > a syncfs(). By that time inode is gone and we will not issue any > > syncfs() on this filesystem. Hence leaving data in host page cache. > >=20 > > Thanks > > Vivek > >=20 > > > + > > > /* > > > * Increments nlookup on the inode on success. unref_inode_lolocked(= ) must be > > > * called eventually to decrement nlookup again. If inodep is non-NU= LL, the > > > @@ -1086,10 +1113,15 @@ static int lo_do_lookup(fuse_req_t req, fuse_= ino_t parent, const char *name, > > > pthread_mutex_lock(&lo->mutex); > > > inode->fuse_ino =3D lo_add_inode_mapping(req, inode); > > > g_hash_table_insert(lo->inodes, &inode->key, inode); > > > + add_mnt_inode(lo, inode); > > > pthread_mutex_unlock(&lo->mutex); > > > } > > > e->ino =3D inode->fuse_ino; > > > =20 > > > + fuse_log(FUSE_LOG_DEBUG, " %lli/%s -> %lli%s\n", > > > + (unsigned long long) parent, name, (unsigned long long)= e->ino, > > > + inode->is_mnt ? " (submount)" : ""); > > > + > > > /* Transfer ownership of inode pointer to caller or drop it */ > > > if (inodep) { > > > *inodep =3D inode; > > > @@ -1099,9 +1131,6 @@ static int lo_do_lookup(fuse_req_t req, fuse_in= o_t parent, const char *name, > > > =20 > > > lo_inode_put(lo, &dir); > > > =20 > > > - fuse_log(FUSE_LOG_DEBUG, " %lli/%s -> %lli\n", (unsigned long l= ong)parent, > > > - name, (unsigned long long)e->ino); > > > - > > > return 0; > > > =20 > > > out_err: > > > @@ -1563,6 +1592,7 @@ static void unref_inode(struct lo_data *lo, str= uct lo_inode *inode, uint64_t n) > > > g_hash_table_destroy(inode->posix_locks); > > > pthread_mutex_destroy(&inode->plock_mutex); > > > } > > > + remove_mnt_inode(lo, inode); > > > /* Drop our refcount from lo_do_lookup() */ > > > lo_inode_put(lo, &inode); > > > } > > > @@ -3337,6 +3367,7 @@ static void lo_destroy(void *userdata) > > > struct lo_data *lo =3D (struct lo_data *)userdata; > > > =20 > > > pthread_mutex_lock(&lo->mutex); > > > + g_hash_table_remove_all(lo->mnt_inodes); > > > while (true) { > > > GHashTableIter iter; > > > gpointer key, value; > > > @@ -3850,6 +3881,7 @@ static void setup_root(struct lo_data *lo, stru= ct lo_inode *root) > > > root->posix_locks =3D g_hash_table_new_full( > > > g_direct_hash, g_direct_equal, NULL, posix_locks_value_d= estroy); > > > } > > > + add_mnt_inode(lo, root); > > > } > > > =20 > > > static guint lo_key_hash(gconstpointer key) > > > @@ -3869,6 +3901,10 @@ static gboolean lo_key_equal(gconstpointer a, = gconstpointer b) > > > =20 > > > static void fuse_lo_data_cleanup(struct lo_data *lo) > > > { > > > + if (lo->mnt_inodes) { > > > + g_hash_table_destroy(lo->mnt_inodes); > > > + } > > > + > > > if (lo->inodes) { > > > g_hash_table_destroy(lo->inodes); > > > } > > > @@ -3931,6 +3967,7 @@ int main(int argc, char *argv[]) > > > lo.root.fd =3D -1; > > > lo.root.fuse_ino =3D FUSE_ROOT_ID; > > > lo.cache =3D CACHE_AUTO; > > > + lo.mnt_inodes =3D g_hash_table_new(g_int64_hash, g_int64_equal); > > > =20 > > > /* > > > * Set up the ino map like this: > > > --=20 > > > 2.34.1 > > >=20 > >=20 > > _______________________________________________ > > Virtio-fs mailing list > > Virtio-fs@redhat.com > > https://listman.redhat.com/mailman/listinfo/virtio-fs > >=20 >=20