From patchwork Wed Mar 6 01:28:11 2013 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Asias He X-Patchwork-Id: 2222691 Return-Path: X-Original-To: patchwork-kvm@patchwork.kernel.org Delivered-To: patchwork-process-083081@patchwork1.kernel.org Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by patchwork1.kernel.org (Postfix) with ESMTP id D90D83FCF6 for ; Wed, 6 Mar 2013 01:28:35 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751074Ab3CFB20 (ORCPT ); Tue, 5 Mar 2013 20:28:26 -0500 Received: from mx1.redhat.com ([209.132.183.28]:38697 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750712Ab3CFB2Z (ORCPT ); Tue, 5 Mar 2013 20:28:25 -0500 Received: from int-mx12.intmail.prod.int.phx2.redhat.com (int-mx12.intmail.prod.int.phx2.redhat.com [10.5.11.25]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id r261SEia008372 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK); Tue, 5 Mar 2013 20:28:15 -0500 Received: from localhost (vpn1-113-55.nay.redhat.com [10.66.113.55]) by int-mx12.intmail.prod.int.phx2.redhat.com (8.14.4/8.14.4) with ESMTP id r261SCKH019745; Tue, 5 Mar 2013 20:28:13 -0500 Date: Wed, 6 Mar 2013 09:28:11 +0800 From: Asias He To: "Michael S. Tsirkin" Cc: Nicholas Bellinger , Paolo Bonzini , Stefan Hajnoczi , Rusty Russell , kvm@vger.kernel.org, virtualization@lists.linux-foundation.org, target-devel@vger.kernel.org Subject: Re: [PATCH] tcm_vhost: Add hotplug/hotunplug support Message-ID: <20130306012811.GA20372@hj.localdomain> References: <1362475027-12018-1-git-send-email-asias@redhat.com> <20130305121125.GA823@redhat.com> MIME-Version: 1.0 Content-Disposition: inline In-Reply-To: <20130305121125.GA823@redhat.com> User-Agent: Mutt/1.5.21 (2010-09-15) X-Scanned-By: MIMEDefang 2.68 on 10.5.11.25 Sender: kvm-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: kvm@vger.kernel.org On Tue, Mar 05, 2013 at 02:11:25PM +0200, Michael S. Tsirkin wrote: > On Tue, Mar 05, 2013 at 05:17:07PM +0800, Asias He wrote: > > In commit 365a7150094 ([SCSI] virtio-scsi: hotplug support for > > virtio-scsi), hotplug support is added to virtio-scsi. > > > > This patch adds hotplug and hotunplug support to tcm_vhost. > > > > You can create or delate a LUN in targetcli to hotplug or hotplug a LUN > > in guest. > > > > Signed-off-by: Asias He > > --- > > drivers/vhost/tcm_vhost.c | 171 ++++++++++++++++++++++++++++++++++++++++++++-- > > drivers/vhost/tcm_vhost.h | 9 +++ > > 2 files changed, 175 insertions(+), 5 deletions(-) > > > > diff --git a/drivers/vhost/tcm_vhost.c b/drivers/vhost/tcm_vhost.c > > index 9951297..6693695 100644 > > --- a/drivers/vhost/tcm_vhost.c > > +++ b/drivers/vhost/tcm_vhost.c > > @@ -63,6 +63,8 @@ enum { > > #define VHOST_SCSI_MAX_TARGET 256 > > #define VHOST_SCSI_MAX_VQ 128 > > > > +#define VHOST_SCSI_FEATURES (VHOST_FEATURES | (1ULL << VIRTIO_SCSI_F_HOTPLUG)) > > + > > struct vhost_scsi { > > /* Protected by vhost_scsi->dev.mutex */ > > struct tcm_vhost_tpg *vs_tpg[VHOST_SCSI_MAX_TARGET]; > > @@ -74,6 +76,11 @@ struct vhost_scsi { > > > > struct vhost_work vs_completion_work; /* cmd completion work item */ > > struct llist_head vs_completion_list; /* cmd completion queue */ > > + > > + struct vhost_work vs_event_work; /* evt injection work item */ > > + struct llist_head vs_event_list; /* evt injection queue */ > > + > > + bool vs_events_dropped; > > Documentation pls. > Also - how is this handled during migration? > Don't we need a way for userspace to retrieve this bit? Yes, but migration is not supported atm. > > }; > > > > /* Local pointer to allocated TCM configfs fabric module */ > > @@ -341,6 +348,23 @@ static int tcm_vhost_queue_tm_rsp(struct se_cmd *se_cmd) > > return 0; > > } > > > > +static void tcm_vhost_free_evt(struct tcm_vhost_evt *evt) > > +{ > > + kfree(evt); > > +} > > + > > +static struct tcm_vhost_evt *tcm_vhost_allocate_evt(u32 event, u32 reason) > > +{ > > + struct tcm_vhost_evt *evt; > > + > > + evt = kzalloc(sizeof(*evt), GFP_KERNEL); > > And if this fails? The check was dropped when I moving around the allocate helper. Will add it back. > > + > > + evt->event.event = event; > > + evt->event.reason = reason; > > + > > + return evt; > > +} > > + > > static void vhost_scsi_free_cmd(struct tcm_vhost_cmd *tv_cmd) > > { > > struct se_cmd *se_cmd = &tv_cmd->tvc_se_cmd; > > @@ -359,6 +383,71 @@ static void vhost_scsi_free_cmd(struct tcm_vhost_cmd *tv_cmd) > > kfree(tv_cmd); > > } > > > > +static void tcm_vhost_do_evt_work(struct vhost_scsi *vs, > > + struct virtio_scsi_event *event) > > +{ > > + struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT]; > > + struct virtio_scsi_event __user *eventp; > > + unsigned out, in; > > + int head, ret; > > + > > + if (!vs || !vs->vs_endpoint) > > + return; > > + > > + mutex_lock(&vq->mutex); > > +again: > > + vhost_disable_notify(&vs->dev, vq); > > + head = vhost_get_vq_desc(&vs->dev, vq, vq->iov, > > + ARRAY_SIZE(vq->iov), &out, &in, > > + NULL, NULL); > > + if (head < 0) { > > + vs->vs_events_dropped = true; > > + goto out; > > + } > > + if (head == vq->num) { > > + if (vhost_enable_notify(&vs->dev, vq)) > > + goto again; > > Could you code this up using loop, without goto please? This is not a loop actaully. We only need to grap one buffer at a time. IMHO, It is even uglier to user a loop: 1) you need to exit the loop arbitary at the end. 2) Extra indention. > > + vs->vs_events_dropped = true; > > + goto out; > > + } > > + > > + if ((vq->iov[out].iov_len != sizeof(struct virtio_scsi_event))) { > > We should avoid making layout assumptions. Please don't. Other places is doing this as well, e.g vhost_scsi_handle_vq(). Will do the conversion to no layout assumptions mode in further patches to convert all of them. > > + vq_err(vq, "Expecting virtio_scsi_event, got %zu bytes\n", > > + vq->iov[out].iov_len); > > + goto out; > > + } > > + > > + if (vs->vs_events_dropped) { > > + event->event |= VIRTIO_SCSI_T_EVENTS_MISSED; > > + vs->vs_events_dropped = false; > > + } > > + > > + eventp = vq->iov[out].iov_base; > > + ret = __copy_to_user(eventp, event, sizeof(*event)); > > + if (!ret) > > + vhost_add_used_and_signal(&vs->dev, vq, head, 0); > > + else > > + pr_err("Faulted on tcm_vhost_send_event\n"); > > vq_err please, this is guest triggerable. Okay. Will fix other places too. > > +out: > > + mutex_unlock(&vq->mutex); > > +} > > + > > +static void tcm_vhost_evt_work(struct vhost_work *work) > > +{ > > + struct vhost_scsi *vs = container_of(work, struct vhost_scsi, > > + vs_event_work); > > + struct tcm_vhost_evt *evt; > > + struct llist_node *llnode; > > + > > + llnode = llist_del_all(&vs->vs_event_list); > > + while (llnode) { > > + evt = llist_entry(llnode, struct tcm_vhost_evt, list); > > + llnode = llist_next(llnode); > > + tcm_vhost_do_evt_work(vs, &evt->event); > > + tcm_vhost_free_evt(evt); > > + } > > +} > > + > > /* Fill in status and signal that we are done processing this command > > * > > * This is scheduled in the vhost work queue so we are called with the owner > > @@ -757,9 +846,41 @@ static void vhost_scsi_ctl_handle_kick(struct vhost_work *work) > > pr_debug("%s: The handling func for control queue.\n", __func__); > > } > > > > +static int tcm_vhost_send_evt(struct vhost_scsi *vs, struct tcm_vhost_tpg *tpg, > > + struct se_lun *lun, u32 event, u32 reason) > > Align ) on ( please. What is the preferred alignment here? Can you elaborate? > > +{ > > + struct tcm_vhost_evt *evt; > > + > > + if (!vs->vs_endpoint) > > + return -EOPNOTSUPP; > > + > > Pls add a comment explaining the abive. > Is this dereference safe without any locking? vs->vs_endpoint is protected by vhost_scsi->dev.mutex. It is ok to use the lock for event vq. But for the cmd vq, it is not optimal to take the lock in the data path. We have this in vhost_scsi_handle_vq () { /* Must use ioctl VHOST_SCSI_SET_ENDPOINT */ if (unlikely(!vs->vs_endpoint)) return; } > > + evt = tcm_vhost_allocate_evt(event, reason); > > + if (!evt) > > + return -ENOMEM; > > And what happens then? How about we set event missed flag too? Okay. > > + > > + if (tpg && lun) { > > + evt->event.lun[0] = 0x01; > > + evt->event.lun[1] = tpg->tport_tpgt & 0xFF; > > + if (lun->unpacked_lun >= 256) > > + evt->event.lun[2] = lun->unpacked_lun >> 8 | 0x40 ; > > + evt->event.lun[3] = lun->unpacked_lun & 0xFF; > > I know it's not your fault but we really should share this code with > virtio scsi. Pls add TODO now. Ok. > > + } > > + > > + llist_add(&evt->list, &vs->vs_event_list); > > This can queue up quite a bit of memory if the handler thread > is delayed, no? Can we limit the # of outstanding events? > Will guest recover from a missed event? Hmm, good point. Will limit the number. The size of 'struct tcm_vhost_evt' is around 20 bytes. So if we limit it to 128, it is ~2.5K of memory. Paolo, if we limit the number of outstanding events and set vs->vs_events_dropped, the guest will recover from a missed event, right? > > + vhost_work_queue(&vs->dev, &vs->vs_event_work); > > + > > + return 0; > > +} > > + > > static void vhost_scsi_evt_handle_kick(struct vhost_work *work) > > { > > - pr_debug("%s: The handling func for event queue.\n", __func__); > > + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, > > + poll.work); > > + struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); > > + > > + if (vs->vs_events_dropped) > > + tcm_vhost_send_evt(vs, NULL, NULL, VIRTIO_SCSI_T_NO_EVENT, 0); > > + > > } > > > > static void vhost_scsi_handle_kick(struct vhost_work *work) > > @@ -815,6 +936,7 @@ static int vhost_scsi_set_endpoint( > > return -EEXIST; > > } > > tv_tpg->tv_tpg_vhost_count++; > > + tv_tpg->vhost_scsi = vs; > > vs->vs_tpg[tv_tpg->tport_tpgt] = tv_tpg; > > smp_mb__after_atomic_inc(); > > match = true; > > @@ -875,6 +997,7 @@ static int vhost_scsi_clear_endpoint( > > goto err; > > } > > tv_tpg->tv_tpg_vhost_count--; > > + tv_tpg->vhost_scsi = NULL; > > vs->vs_tpg[target] = NULL; > > vs->vs_endpoint = false; > > } > > @@ -896,6 +1019,7 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) > > return -ENOMEM; > > > > vhost_work_init(&s->vs_completion_work, vhost_scsi_complete_cmd_work); > > + vhost_work_init(&s->vs_event_work, tcm_vhost_evt_work); > > > > s->vqs[VHOST_SCSI_VQ_CTL].handle_kick = vhost_scsi_ctl_handle_kick; > > s->vqs[VHOST_SCSI_VQ_EVT].handle_kick = vhost_scsi_evt_handle_kick; > > @@ -941,7 +1065,7 @@ static void vhost_scsi_flush(struct vhost_scsi *vs) > > > > static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features) > > { > > - if (features & ~VHOST_FEATURES) > > + if (features & ~VHOST_SCSI_FEATURES) > > return -EOPNOTSUPP; > > > > mutex_lock(&vs->dev.mutex); > > @@ -987,7 +1111,7 @@ static long vhost_scsi_ioctl(struct file *f, unsigned int ioctl, > > return -EFAULT; > > return 0; > > case VHOST_GET_FEATURES: > > - features = VHOST_FEATURES; > > + features = VHOST_SCSI_FEATURES; > > if (copy_to_user(featurep, &features, sizeof features)) > > return -EFAULT; > > return 0; > > @@ -1057,6 +1181,40 @@ static char *tcm_vhost_dump_proto_id(struct tcm_vhost_tport *tport) > > return "Unknown"; > > } > > > > +static int tcm_vhost_hotplug(struct tcm_vhost_tpg *tpg, struct se_lun *lun) > > +{ > > + struct vhost_scsi *vs = tpg->vhost_scsi; > > + u64 features; > > + > > + if (!vs) > > + return -EOPNOTSUPP; > > + > > + features = vs->dev.acked_features; > > + if (!(features & 1ULL << VIRTIO_SCSI_F_HOTPLUG)) > > + return -EOPNOTSUPP; > > + > > + return tcm_vhost_send_evt(vs, tpg, lun, > > + VIRTIO_SCSI_T_TRANSPORT_RESET, > > + VIRTIO_SCSI_EVT_RESET_RESCAN); > > +} > > + > > +static int tcm_vhost_hotunplug(struct tcm_vhost_tpg *tpg, struct se_lun *lun) > > +{ > > + struct vhost_scsi *vs = tpg->vhost_scsi; > > + u64 features; > > + > > + if (!vs) > > + return -EOPNOTSUPP; > > + > > What are we checking for here, and why is it safe to do > outside any lock? tcm_vhost_hotplug and tcm_vhost_hotunplug might be called when vhost-scsi is not opened. Will add the locking. > > + features = vs->dev.acked_features; > > + if (!(features & 1ULL << VIRTIO_SCSI_F_HOTPLUG)) > > + return -EOPNOTSUPP; > > + > > + return tcm_vhost_send_evt(vs, tpg, lun, > > + VIRTIO_SCSI_T_TRANSPORT_RESET, > > + VIRTIO_SCSI_EVT_RESET_REMOVED); > > +} > > + > > static int tcm_vhost_port_link(struct se_portal_group *se_tpg, > > struct se_lun *lun) > > { > > @@ -1067,18 +1225,21 @@ static int tcm_vhost_port_link(struct se_portal_group *se_tpg, > > tv_tpg->tv_tpg_port_count++; > > mutex_unlock(&tv_tpg->tv_tpg_mutex); > > > > + tcm_vhost_hotplug(tv_tpg, lun); > > + > > return 0; > > } > > > > static void tcm_vhost_port_unlink(struct se_portal_group *se_tpg, > > - struct se_lun *se_lun) > > + struct se_lun *lun) > > { > > struct tcm_vhost_tpg *tv_tpg = container_of(se_tpg, > > struct tcm_vhost_tpg, se_tpg); > > - > > mutex_lock(&tv_tpg->tv_tpg_mutex); > > tv_tpg->tv_tpg_port_count--; > > mutex_unlock(&tv_tpg->tv_tpg_mutex); > > + > > + tcm_vhost_hotunplug(tv_tpg, lun); > > } > > > > static struct se_node_acl *tcm_vhost_make_nodeacl( > > diff --git a/drivers/vhost/tcm_vhost.h b/drivers/vhost/tcm_vhost.h > > index 1d2ae7a..191a945 100644 > > --- a/drivers/vhost/tcm_vhost.h > > +++ b/drivers/vhost/tcm_vhost.h > > @@ -70,6 +70,8 @@ struct tcm_vhost_tpg { > > struct tcm_vhost_tport *tport; > > /* Returned by tcm_vhost_make_tpg() */ > > struct se_portal_group se_tpg; > > + /* Pointer back to struct vhost_scsi*/ > > + void *vhost_scsi; > > Does it have to be void? Any why? > What lock protects this field? Please add a comment. struct vhost_scsi is defined in tcm_vhost.c. tv_tpg_mutex will protect it. Does this look good? > > }; > > > > struct tcm_vhost_tport { > > @@ -83,6 +85,13 @@ struct tcm_vhost_tport { > > struct se_wwn tport_wwn; > > }; > > > > +struct tcm_vhost_evt { > > + /* virtio_scsi event */ > > + struct virtio_scsi_event event; > > + /* virtio_scsi event list, serviced from vhost worker thread */ > > + struct llist_node list; > > +}; > > + > > /* > > * As per request from MST, keep TCM_VHOST related ioctl defines out of > > * linux/vhost.h (user-space) for now.. > > -- > > 1.8.1.4 diff --git a/drivers/vhost/tcm_vhost.h b/drivers/vhost/tcm_vhost.h index 191a945..8941a54 100644 --- a/drivers/vhost/tcm_vhost.h +++ b/drivers/vhost/tcm_vhost.h @@ -53,6 +53,7 @@ struct tcm_vhost_nacl { struct se_node_acl se_node_acl; }; +struct vhost_scsi; struct tcm_vhost_tpg { /* Vhost port target portal group tag for TCM */ u16 tport_tpgt; @@ -71,7 +72,7 @@ struct tcm_vhost_tpg { /* Returned by tcm_vhost_make_tpg() */ struct se_portal_group se_tpg; /* Pointer back to struct vhost_scsi*/ - void *vhost_scsi; + struct vhost_scsi *vhost_scsi; }; struct tcm_vhost_tport {