#define VHOST_SCSI_MAX_VQ 128
#define VHOST_SCSI_MAX_EVENT 128
+struct vhost_scsi_inflight {
+ /* Wait for the flush operation to finish */
+ struct completion comp;
+ /* Refcount for the inflight reqs */
+ struct kref kref;
+};
+
+struct vhost_scsi_virtqueue {
+ struct vhost_virtqueue vq;
+ /*
+ * Reference counting for inflight reqs, used for flush operation. At
+ * each time, one reference tracks new commands submitted, while we
+ * wait for another one to reach 0.
+ */
+ struct vhost_scsi_inflight inflights[2];
+ /*
+ * Indicate current inflight in use, protected by vq->mutex.
+ * Writers must also take dev mutex and flush under it.
+ */
+ int inflight_idx;
+};
+
struct vhost_scsi {
/* Protected by vhost_scsi->dev.mutex */
struct tcm_vhost_tpg **vs_tpg;
char vs_vhost_wwpn[TRANSPORT_IQN_LEN];
struct vhost_dev dev;
- struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ];
+ struct vhost_scsi_virtqueue vqs[VHOST_SCSI_MAX_VQ];
struct vhost_work vs_completion_work; /* cmd completion work item */
struct llist_head vs_completion_list; /* cmd completion queue */
((unsigned long)iov->iov_base & PAGE_MASK)) >> PAGE_SHIFT;
}
+void tcm_vhost_done_inflight(struct kref *kref)
+{
+ struct vhost_scsi_inflight *inflight;
+
+ inflight = container_of(kref, struct vhost_scsi_inflight, kref);
+ complete(&inflight->comp);
+}
+
+static void tcm_vhost_init_inflight(struct vhost_scsi *vs,
+ struct vhost_scsi_inflight *old_inflight[])
+{
+ struct vhost_scsi_inflight *new_inflight;
+ struct vhost_virtqueue *vq;
+ int idx, i;
+
+ for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
+ vq = &vs->vqs[i].vq;
+
+ mutex_lock(&vq->mutex);
+
+ /* store old infight */
+ idx = vs->vqs[i].inflight_idx;
+ if (old_inflight)
+ old_inflight[i] = &vs->vqs[i].inflights[idx];
+
+ /* setup new infight */
+ vs->vqs[i].inflight_idx = idx ^ 1;
+ new_inflight = &vs->vqs[i].inflights[idx ^ 1];
+ kref_init(&new_inflight->kref);
+ init_completion(&new_inflight->comp);
+
+ mutex_unlock(&vq->mutex);
+ }
+}
+
+static struct vhost_scsi_inflight *
+tcm_vhost_get_inflight(struct vhost_virtqueue *vq)
+{
+ struct vhost_scsi_inflight *inflight;
+ struct vhost_scsi_virtqueue *svq;
+
+ svq = container_of(vq, struct vhost_scsi_virtqueue, vq);
+ inflight = &svq->inflights[svq->inflight_idx];
+ kref_get(&inflight->kref);
+
+ return inflight;
+}
+
+static void tcm_vhost_put_inflight(struct vhost_scsi_inflight *inflight)
+{
+ kref_put(&inflight->kref, tcm_vhost_done_inflight);
+}
+
static int tcm_vhost_check_true(struct se_portal_group *se_tpg)
{
return 1;
static struct tcm_vhost_evt *tcm_vhost_allocate_evt(struct vhost_scsi *vs,
u32 event, u32 reason)
{
- struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT];
+ struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
struct tcm_vhost_evt *evt;
if (vs->vs_events_nr > VHOST_SCSI_MAX_EVENT) {
kfree(tv_cmd->tvc_sgl);
}
+ tcm_vhost_put_inflight(tv_cmd->inflight);
+
kfree(tv_cmd);
}
static void tcm_vhost_do_evt_work(struct vhost_scsi *vs,
struct tcm_vhost_evt *evt)
{
- struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT];
+ struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
struct virtio_scsi_event *event = &evt->event;
struct virtio_scsi_event __user *eventp;
unsigned out, in;
{
struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
vs_event_work);
- struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT];
+ struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
struct tcm_vhost_evt *evt;
struct llist_node *llnode;
v_rsp.sense_len);
ret = copy_to_user(tv_cmd->tvc_resp, &v_rsp, sizeof(v_rsp));
if (likely(ret == 0)) {
+ struct vhost_scsi_virtqueue *q;
vhost_add_used(tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc, 0);
- vq = tv_cmd->tvc_vq - vs->vqs;
+ q = container_of(tv_cmd->tvc_vq, struct vhost_scsi_virtqueue, vq);
+ vq = q - vs->vqs;
__set_bit(vq, signal);
} else
pr_err("Faulted on virtio_scsi_cmd_resp\n");
vq = -1;
while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1))
< VHOST_SCSI_MAX_VQ)
- vhost_signal(&vs->dev, &vs->vqs[vq]);
+ vhost_signal(&vs->dev, &vs->vqs[vq].vq);
}
static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(
+ struct vhost_virtqueue *vq,
struct tcm_vhost_tpg *tv_tpg,
struct virtio_scsi_cmd_req *v_req,
u32 exp_data_len,
tv_cmd->tvc_exp_data_len = exp_data_len;
tv_cmd->tvc_data_direction = data_direction;
tv_cmd->tvc_nexus = tv_nexus;
+ tv_cmd->inflight = tcm_vhost_get_inflight(vq);
return tv_cmd;
}
for (i = 0; i < data_num; i++)
exp_data_len += vq->iov[data_first + i].iov_len;
- tv_cmd = vhost_scsi_allocate_cmd(tv_tpg, &v_req,
+ tv_cmd = vhost_scsi_allocate_cmd(vq, tv_tpg, &v_req,
exp_data_len, data_direction);
if (IS_ERR(tv_cmd)) {
vq_err(vq, "vhost_scsi_allocate_cmd failed %ld\n",
static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index)
{
- vhost_poll_flush(&vs->dev.vqs[index].poll);
+ vhost_poll_flush(&vs->vqs[index].vq.poll);
}
+/* Callers must hold dev mutex */
static void vhost_scsi_flush(struct vhost_scsi *vs)
{
+ struct vhost_scsi_inflight *old_inflight[VHOST_SCSI_MAX_VQ];
int i;
+ /* Init new inflight and remember the old inflight */
+ tcm_vhost_init_inflight(vs, old_inflight);
+
+ /*
+ * The inflight->kref was initialized to 1. We decrement it here to
+ * indicate the start of the flush operation so that it will reach 0
+ * when all the reqs are finished.
+ */
+ for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
+ kref_put(&old_inflight[i]->kref, tcm_vhost_done_inflight);
+
+ /* Flush both the vhost poll and vhost work */
for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
vhost_scsi_flush_vq(vs, i);
vhost_work_flush(&vs->dev, &vs->vs_completion_work);
vhost_work_flush(&vs->dev, &vs->vs_event_work);
+
+ /* Wait for all reqs issued before the flush to be finished */
+ for (i = 0; i < VHOST_SCSI_MAX_VQ; i++)
+ wait_for_completion(&old_inflight[i]->comp);
}
/*
/* Verify that ring has been setup correctly. */
for (index = 0; index < vs->dev.nvqs; ++index) {
/* Verify that ring has been setup correctly. */
- if (!vhost_vq_access_ok(&vs->vqs[index])) {
+ if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
ret = -EFAULT;
goto out;
}
memcpy(vs->vs_vhost_wwpn, t->vhost_wwpn,
sizeof(vs->vs_vhost_wwpn));
for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
- vq = &vs->vqs[i];
+ vq = &vs->vqs[i].vq;
/* Flushing the vhost_work acts as synchronize_rcu */
mutex_lock(&vq->mutex);
rcu_assign_pointer(vq->private_data, vs_tpg);
mutex_lock(&vs->dev.mutex);
/* Verify that ring has been setup correctly. */
for (index = 0; index < vs->dev.nvqs; ++index) {
- if (!vhost_vq_access_ok(&vs->vqs[index])) {
+ if (!vhost_vq_access_ok(&vs->vqs[index].vq)) {
ret = -EFAULT;
goto err_dev;
}
}
if (match) {
for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) {
- vq = &vs->vqs[i];
+ vq = &vs->vqs[i].vq;
/* Flushing the vhost_work acts as synchronize_rcu */
mutex_lock(&vq->mutex);
rcu_assign_pointer(vq->private_data, NULL);
static int vhost_scsi_open(struct inode *inode, struct file *f)
{
struct vhost_scsi *s;
+ struct vhost_virtqueue **vqs;
int r, i;
s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s)
return -ENOMEM;
+ vqs = kmalloc(VHOST_SCSI_MAX_VQ * sizeof(*vqs), GFP_KERNEL);
+ if (!vqs) {
+ kfree(s);
+ 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->vs_events_nr = 0;
s->vs_events_missed = false;
- 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;
- for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++)
- s->vqs[i].handle_kick = vhost_scsi_handle_kick;
- r = vhost_dev_init(&s->dev, s->vqs, VHOST_SCSI_MAX_VQ);
+ vqs[VHOST_SCSI_VQ_CTL] = &s->vqs[VHOST_SCSI_VQ_CTL].vq;
+ vqs[VHOST_SCSI_VQ_EVT] = &s->vqs[VHOST_SCSI_VQ_EVT].vq;
+ s->vqs[VHOST_SCSI_VQ_CTL].vq.handle_kick = vhost_scsi_ctl_handle_kick;
+ s->vqs[VHOST_SCSI_VQ_EVT].vq.handle_kick = vhost_scsi_evt_handle_kick;
+ for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) {
+ vqs[i] = &s->vqs[i].vq;
+ s->vqs[i].vq.handle_kick = vhost_scsi_handle_kick;
+ }
+ r = vhost_dev_init(&s->dev, vqs, VHOST_SCSI_MAX_VQ);
+
+ tcm_vhost_init_inflight(s, NULL);
+
if (r < 0) {
+ kfree(vqs);
kfree(s);
return r;
}
vhost_dev_cleanup(&s->dev, false);
/* Jobs can re-queue themselves in evt kick handler. Do extra flush. */
vhost_scsi_flush(s);
+ kfree(s->dev.vqs);
kfree(s);
return 0;
}
u32 events_missed;
u64 features;
int r, abi_version = VHOST_SCSI_ABI_VERSION;
- struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT];
+ struct vhost_virtqueue *vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
switch (ioctl) {
case VHOST_SCSI_SET_ENDPOINT:
else
reason = VIRTIO_SCSI_EVT_RESET_REMOVED;
- vq = &vs->vqs[VHOST_SCSI_VQ_EVT];
+ vq = &vs->vqs[VHOST_SCSI_VQ_EVT].vq;
mutex_lock(&vq->mutex);
tcm_vhost_send_evt(vs, tpg, lun,
VIRTIO_SCSI_T_TRANSPORT_RESET, reason);