On 2018/4/5 11:23, Mike Christie wrote:
On 04/04/2018 09:38 PM, Xiubo Li wrote:
On 2018/4/5 8:47, Mike Christie wrote:
On 04/02/2018 06:42 AM, xiubli@xxxxxxxxxx wrote:
From: Xiubo Li <xiubli@xxxxxxxxxx>
This patch adds 1 tcmu attr to reset and complete all the blocked
netlink waiting threads. It's used when the userspace daemon like
tcmu-runner has crashed or forced to shutdown just before the
netlink requests be replied to the kernel, then the netlink requeting
threads will get stuck forever. We must reboot the machine to recover
from it and by this the rebootng is not a must then.
The netlink reset operation should be done before the userspace daemon
could receive and handle the netlink requests to be safe.
Signed-off-by: Xiubo Li <xiubli@xxxxxxxxxx>
---
drivers/target/target_core_user.c | 99
++++++++++++++++++++++++++++++++++++---
1 file changed, 93 insertions(+), 6 deletions(-)
diff --git a/drivers/target/target_core_user.c
b/drivers/target/target_core_user.c
index 4ad89ea..dc8879d 100644
--- a/drivers/target/target_core_user.c
+++ b/drivers/target/target_core_user.c
@@ -103,9 +103,13 @@ struct tcmu_hba {
#define TCMU_CONFIG_LEN 256
+static spinlock_t nl_complete_lock;
+static struct idr complete_wait_udevs = IDR_INIT;
+
struct tcmu_nl_cmd {
/* wake up thread waiting for reply */
- struct completion complete;
+ bool complete;
+
int cmd;
int status;
};
@@ -159,12 +163,17 @@ struct tcmu_dev {
spinlock_t nl_cmd_lock;
struct tcmu_nl_cmd curr_nl_cmd;
- /* wake up threads waiting on curr_nl_cmd */
+ /* wake up threads waiting on nl_cmd_wq */
wait_queue_head_t nl_cmd_wq;
+ /* complete thread waiting complete_wq */
+ wait_queue_head_t complete_wq;
+
char dev_config[TCMU_CONFIG_LEN];
int nl_reply_supported;
+
+ uint32_t dev_id;
};
#define TCMU_DEV(_se_dev) container_of(_se_dev, struct tcmu_dev,
se_dev)
@@ -251,6 +260,56 @@ static int tcmu_get_global_max_data_area(char
*buffer,
"Max MBs allowed to be allocated to all the tcmu device's "
"data areas.");
+static void tcmu_complete_wake_up(struct tcmu_dev *udev)
+{
+ struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
+
+ spin_lock(&nl_complete_lock);
+ nl_cmd->complete = true;
+ wake_up(&udev->complete_wq);
+ spin_unlock(&nl_complete_lock);
+}
+
+static void tcmu_complete_wake_up_all(void)
+{
+ struct tcmu_nl_cmd *nl_cmd;
+ struct tcmu_dev *udev;
+ int i;
+
+ spin_lock(&nl_complete_lock);
+ idr_for_each_entry(&complete_wait_udevs, udev, i) {
+ nl_cmd = &udev->curr_nl_cmd;
+ nl_cmd->complete = true;
+ wake_up(&udev->complete_wq);
+ }
+ spin_unlock(&nl_complete_lock);
+}
+
+static int tcmu_complete_wait(struct tcmu_dev *udev)
+{
+ struct tcmu_nl_cmd *nl_cmd = &udev->curr_nl_cmd;
+ uint32_t dev_id;
+
+ spin_lock(&nl_complete_lock);
+ dev_id = idr_alloc(&complete_wait_udevs, udev, 1, USHRT_MAX,
GFP_NOWAIT);
+ if (dev_id < 0) {
+ pr_err("tcmu: Could not allocate dev id.\n");
+ return dev_id;
+ }
+ udev->dev_id = dev_id;
dev_id is never used.
It will be used when the device is being removed.
Ah yeah, you are right. Bad comment on my part. I was thinking/writing
wrt if we used a list or helper around devices_idr.
I think if you just wanted to loop over all the devices you could just
use a list.
Or,
Just add a helper around target_core_device.c:devices_idr that just
gives you the tcmu devices.
+ spin_unlock(&nl_complete_lock);
+
+ pr_debug("sleeping for nl reply\n");
+ wait_event(udev->complete_wq, nl_cmd->complete);
I don't think you will need the complete field then or this function.
+
+ spin_lock(&nl_complete_lock);
+ nl_cmd->complete = false;
+ idr_remove(&complete_wait_udevs, dev_id);
+ spin_unlock(&nl_complete_lock);
+
+ return 0;
+}
+
/* multicast group */
enum tcmu_multicast_groups {
TCMU_MCGRP_CONFIG,
@@ -311,7 +370,7 @@ static int tcmu_genl_cmd_done(struct genl_info
*info, int completed_cmd)
if (!is_removed)
target_undepend_item(&dev->dev_group.cg_item);
if (!ret)
- complete(&nl_cmd->complete);
+ tcmu_complete_wake_up(udev);
return ret;
}
@@ -1258,6 +1317,7 @@ static struct se_device
*tcmu_alloc_device(struct se_hba *hba, const char *name)
timer_setup(&udev->cmd_timer, tcmu_cmd_timedout, 0);
init_waitqueue_head(&udev->nl_cmd_wq);
+ init_waitqueue_head(&udev->complete_wq);
spin_lock_init(&udev->nl_cmd_lock);
INIT_RADIX_TREE(&udev->data_blocks, GFP_KERNEL);
@@ -1462,7 +1522,11 @@ static void tcmu_dev_call_rcu(struct rcu_head *p)
kfree(udev->uio_info.name);
kfree(udev->name);
+
+ spin_lock(&nl_complete_lock);
+ idr_remove(&complete_wait_udevs, udev->dev_id);
kfree(udev);
+ spin_unlock(&nl_complete_lock);
}
static int tcmu_check_and_free_pending_cmd(struct tcmu_cmd *cmd)
@@ -1555,7 +1619,6 @@ static void tcmu_init_genl_cmd_reply(struct
tcmu_dev *udev, int cmd)
memset(nl_cmd, 0, sizeof(*nl_cmd));
nl_cmd->cmd = cmd;
- init_completion(&nl_cmd->complete);
spin_unlock(&udev->nl_cmd_lock);
}
@@ -1572,8 +1635,9 @@ static int tcmu_wait_genl_cmd_reply(struct
tcmu_dev *udev)
if (udev->nl_reply_supported <= 0)
return 0;
- pr_debug("sleeping for nl reply\n");
- wait_for_completion(&nl_cmd->complete);
+ ret = tcmu_complete_wait(udev);
+ if (ret)
+ return ret;
spin_lock(&udev->nl_cmd_lock);
nl_cmd->cmd = TCMU_CMD_UNSPEC;
@@ -2323,6 +2387,26 @@ static ssize_t tcmu_block_dev_store(struct
config_item *item, const char *page,
}
CONFIGFS_ATTR(tcmu_, block_dev);
+static ssize_t tcmu_reset_netlink_store(struct config_item *item,
const char *page,
+ size_t count)
+{
+ u8 val;
+ int ret;
+
+ ret = kstrtou8(page, 0, &val);
+ if (ret < 0)
+ return ret;
+
+ if (val != 1) {
+ pr_err("Invalid block value %d\n", val);
+ return -EINVAL;
+ }
+
+ tcmu_complete_wake_up_all();
+ return count;
+}
+CONFIGFS_ATTR_WO(tcmu_, reset_netlink);
If it's on the device it should only reset the device its on, so if 2
daemons/apps are managing different devices it doesn't mess up the other.
Or we could just assume that there is only 1 daemon type and just do a
global attr at the module level. Probably just the per device is best in
case we end up with people running gluster + qemu+tcmu and ceph +
tcmu-runner.
If you do the per device then you can just take the insides of
tcmu_genl_cmd_done and make it into a helper so that you can do the
refcount/target_undepend_item properly and it would do the wake up. In
the reset configfs function then grab the nl_cmd_lock, and set the
curr_nl_cmd status to some or pass it into the helper, and then call
your helper which does the common stuff.
I thought there should only 1 daemon will be exist in user space and at
I think the most common will be the 1 daemon type setup (it would be
rare to run qemu-tcmu and tcmu-runner at the same time) but we could
have multiple daemon instances of the same type due to containers in the
future.
Yes, it's possible.
the same time to simplify it by resetting only one device will also
reset all the others, or we need to reset all the devices one by one.
I think we want it per device to match the other code like ring reset
and nl reply supported for example. We then get support for everything
very cheaply.
No problem, will update the patch later.
If so, I will just do the per device resetting, then the patch will be
very simple.
+
static ssize_t tcmu_reset_ring_store(struct config_item *item,
const char *page,
size_t count)
{
@@ -2363,6 +2447,7 @@ static ssize_t tcmu_reset_ring_store(struct
config_item *item, const char *page,
static struct configfs_attribute *tcmu_action_attrs[] = {
&tcmu_attr_block_dev,
&tcmu_attr_reset_ring,
+ &tcmu_attr_reset_netlink,
NULL,
};
@@ -2519,6 +2604,8 @@ static int __init tcmu_module_init(void)
}
tcmu_ops.tb_dev_attrib_attrs = tcmu_attrs;
+ spin_lock_init(&nl_complete_lock);
+
ret = transport_backend_register(&tcmu_ops);
if (ret)
goto out_attrs;
--
To unsubscribe from this list: send the line "unsubscribe
target-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
