Re: [PATCH v4 01/18] nbd: Fix might_sleep warning on xmit timeout

From: Pranay Srivastava
Date: Wed May 25 2016 - 13:15:18 EST


On Mon, May 23, 2016 at 4:02 PM, Pranay Srivastava <pranjas@xxxxxxxxx> wrote:
> Hi Markus
>
> On Fri, May 20, 2016 at 1:52 PM, Markus Pargmann <mpa@xxxxxxxxxxxxxx> wrote:
>> On Friday 20 May 2016 02:05:36 Pranay Srivastava wrote:
>>> On Thu, May 19, 2016 at 11:52 AM, Markus Pargmann <mpa@xxxxxxxxxxxxxx> wrote:
>>> > Hi,
>>> >
>>> > On Wed, May 11, 2016 at 11:18:29AM +0300, Pranay Kr. Srivastava wrote:
>>> >> This patch fixes the warning generated when a timeout occurs
>>> >> on the request and socket is closed from a non-sleep context
>>> >> by
>>> >>
>>> >> 1. Moving the socket closing on a timeout to nbd_thread_send
>>> >
>>> > What happens if a send blocks?
>>>
>>> socket closing needs to be moved to a non-atomic context and,
>>> sender thread seemed to be a good place to do this. If you mean
>>> send blocks just before calling sock_shutdown[?] in nbd_thread_send
>>> then yes I think that should be removed. I need to re-check how nbd-server
>>> behaves in that case.
>>
>> No that's not what I meant. Your approach uses the sender thread as a
>> worker to close the socket. You are using waiting_wq to notify the
>> sender thread. That's fine so far.
>>
>> But what happens if the sender thread is at this point trying to send on
>> the socket which blocks? Then the timeout triggers and waiting_wq will
>> notify the sending thread as soon as it left the sending routine. But it
>> will not interrupt the thread that is waiting in kernel_sendmsg() and
>> the sending thread will be stuck much longer than specified in the
>> timeout.
>
> So socket shutdown must be triggered immediately. I've done a version using
> system_wq for this and appears to be good. I'll be sending that soon after doing
> cleanup and applying your sock_shutdown patch you sent earlier.
>
>>
>>>
>>> >
>>> >>
>>> >> 2. Make sock lock to be a mutex instead of a spin lock, since
>>> >> nbd_xmit_timeout doesn't need to hold it anymore.
>>> >
>>> > I can't see why we need a mutex instead of a spinlock?
>>>
>>> you are right, with your earlier patch we don't need it to be a mutex.
>>>
>>> >
>>> >>
>>> >> Signed-off-by: Pranay Kr. Srivastava <pranjas@xxxxxxxxx>
>>> >> ---
>>> >> drivers/block/nbd.c | 65 ++++++++++++++++++++++++++++++++---------------------
>>> >> 1 file changed, 39 insertions(+), 26 deletions(-)
>>> >>
>>> >> diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
>>> >> index 31e73a7..c79bcd7 100644
>>> >> --- a/drivers/block/nbd.c
>>> >> +++ b/drivers/block/nbd.c
>>> >> @@ -57,12 +57,12 @@ struct nbd_device {
>>> >> int blksize;
>>> >> loff_t bytesize;
>>> >> int xmit_timeout;
>>> >> - bool timedout;
>>> >> + atomic_t timedout;
>>> >> bool disconnect; /* a disconnect has been requested by user */
>>> >>
>>> >> struct timer_list timeout_timer;
>>> >> /* protects initialization and shutdown of the socket */
>>> >> - spinlock_t sock_lock;
>>> >> + struct mutex sock_lock;
>>> >> struct task_struct *task_recv;
>>> >> struct task_struct *task_send;
>>> >>
>>> >> @@ -172,10 +172,9 @@ static void nbd_end_request(struct nbd_device *nbd, struct request *req)
>>> >> */
>>> >> static void sock_shutdown(struct nbd_device *nbd)
>>> >> {
>>> >> - spin_lock_irq(&nbd->sock_lock);
>>> >> -
>>> >> + mutex_lock(&nbd->sock_lock);
>>> >> if (!nbd->sock) {
>>> >> - spin_unlock_irq(&nbd->sock_lock);
>>> >> + mutex_unlock(&nbd->sock_lock);
>>> >> return;
>>> >> }
>>> >>
>>> >> @@ -183,27 +182,19 @@ static void sock_shutdown(struct nbd_device *nbd)
>>> >> kernel_sock_shutdown(nbd->sock, SHUT_RDWR);
>>> >> sockfd_put(nbd->sock);
>>> >> nbd->sock = NULL;
>>> >> - spin_unlock_irq(&nbd->sock_lock);
>>> >> -
>>> >> + mutex_unlock(&nbd->sock_lock);
>>> >> del_timer(&nbd->timeout_timer);
>>> >> }
>>> >>
>>> >> static void nbd_xmit_timeout(unsigned long arg)
>>> >> {
>>> >> struct nbd_device *nbd = (struct nbd_device *)arg;
>>> >> - unsigned long flags;
>>> >>
>>> >> if (list_empty(&nbd->queue_head))
>>> >> return;
>>> >>
>>> >> - spin_lock_irqsave(&nbd->sock_lock, flags);
>>> >> -
>>> >> - nbd->timedout = true;
>>> >> -
>>> >> - if (nbd->sock)
>>> >> - kernel_sock_shutdown(nbd->sock, SHUT_RDWR);
>>> >> -
>>> >> - spin_unlock_irqrestore(&nbd->sock_lock, flags);
>>> >> + atomic_inc(&nbd->timedout);
>>> >> + wake_up(&nbd->waiting_wq);
>>> >>
>>> >> dev_err(nbd_to_dev(nbd), "Connection timed out, shutting down connection\n");
>>> >> }
>>> >> @@ -579,7 +570,27 @@ static int nbd_thread_send(void *data)
>>> >> /* wait for something to do */
>>> >> wait_event_interruptible(nbd->waiting_wq,
>>> >> kthread_should_stop() ||
>>> >> - !list_empty(&nbd->waiting_queue));
>>> >> + !list_empty(&nbd->waiting_queue) ||
>>> >> + atomic_read(&nbd->timedout));
>>> >> +
>>> >> + if (atomic_read(&nbd->timedout)) {
>>> >> + mutex_lock(&nbd->sock_lock);
>>> >> + if (nbd->sock) {
>>> >> + struct request sreq;
>>> >> +
>>> >> + blk_rq_init(NULL, &sreq);
>>> >> + sreq.cmd_type = REQ_TYPE_DRV_PRIV;
>>> >> + mutex_lock(&nbd->tx_lock);
>>> >> + nbd->disconnect = true;
>>> >> + nbd_send_req(nbd, &sreq);
>>> >> + mutex_unlock(&nbd->tx_lock);
>>> >> + dev_err(disk_to_dev(nbd->disk),
>>> >> + "Device Timeout occured.Shutting down"
>>> >> + " socket.");
>>> >> + }
>>> >> + mutex_unlock(&nbd->sock_lock);
>>> >> + sock_shutdown(nbd);
>>> >
>>> > Why are you trying to send something on a connection that timed out
>>> > (nbd_send_req())? And afterwards you execute a socket shutdown so in most
>>> > timeout cases this won't reach the server and we risk a blocking send on
>>> > a timedout connection.
>>>
>>> Ok. I get it. But shouldn't the server side also close it's socket as
>>> well? I don't
>>> think the timeout value is propagated to server or like server can
>>> "ping" to check
>>> if client is there right?
>>>
>>> I agree on nbd_send_req in timedout, it shouldn't be there, just a
>>> sock_shutdown should
>>> do. Can you confirm if I'm right about nbd-server side as well like it
>>> won't timeout and close
>>> that socket or did I miss any option while starting it?
>>
>> If the socket is closed the server will notice at some point in the
>> future at least after the TCP timeout. I am not sure how we could notify
>> the server without running into the next connection issues.
>>
>> Best Regards,
>>
>> Markus
>>
>>>
>>> >
>>> > Regards,
>>> >
>>> > Markus
>>> >
>>> >> + }
>>> >>
>>> >> /* extract request */
>>> >> if (list_empty(&nbd->waiting_queue))
>>> >> @@ -592,7 +603,11 @@ static int nbd_thread_send(void *data)
>>> >> spin_unlock_irq(&nbd->queue_lock);
>>> >>
>>> >> /* handle request */
>>> >> - nbd_handle_req(nbd, req);
>>> >> + if (atomic_read(&nbd->timedout)) {
>>> >> + req->errors++;
>>> >> + nbd_end_request(nbd, req);
>>> >> + } else
>>> >> + nbd_handle_req(nbd, req);
>>> >> }
>>> >>
>>> >> nbd->task_send = NULL;
>>> >> @@ -647,7 +662,7 @@ static int nbd_set_socket(struct nbd_device *nbd, struct socket *sock)
>>> >> {
>>> >> int ret = 0;
>>> >>
>>> >> - spin_lock_irq(&nbd->sock_lock);
>>> >> + mutex_lock(&nbd->sock_lock);
>>> >>
>>> >> if (nbd->sock) {
>>> >> ret = -EBUSY;
>>> >> @@ -657,7 +672,7 @@ static int nbd_set_socket(struct nbd_device *nbd, struct socket *sock)
>>> >> nbd->sock = sock;
>>> >>
>>> >> out:
>>> >> - spin_unlock_irq(&nbd->sock_lock);
>>> >> + mutex_unlock(&nbd->sock_lock);
>>> >>
>>> >> return ret;
>>> >> }
>>> >> @@ -666,7 +681,7 @@ out:
>>> >> static void nbd_reset(struct nbd_device *nbd)
>>> >> {
>>> >> nbd->disconnect = false;
>>> >> - nbd->timedout = false;
>>> >> + atomic_set(&nbd->timedout, 0);
>>> >> nbd->blksize = 1024;
>>> >> nbd->bytesize = 0;
>>> >> set_capacity(nbd->disk, 0);
>>> >> @@ -803,17 +818,15 @@ static int __nbd_ioctl(struct block_device *bdev, struct nbd_device *nbd,
>>> >> error = nbd_thread_recv(nbd, bdev);
>>> >> nbd_dev_dbg_close(nbd);
>>> >> kthread_stop(thread);
>>> >> -
>>> >> - mutex_lock(&nbd->tx_lock);
>>> >> -
>>> >> sock_shutdown(nbd);
>>> >> + mutex_lock(&nbd->tx_lock);
>>> >> nbd_clear_que(nbd);
>>> >> kill_bdev(bdev);
>>> >> nbd_bdev_reset(bdev);
>>> >>
>>> >> if (nbd->disconnect) /* user requested, ignore socket errors */
>>> >> error = 0;
>>> >> - if (nbd->timedout)
>>> >> + if (atomic_read(&nbd->timedout))
>>> >> error = -ETIMEDOUT;
>>> >>
>>> >> nbd_reset(nbd);
>>> >> @@ -1075,7 +1088,7 @@ static int __init nbd_init(void)
>>> >> nbd_dev[i].magic = NBD_MAGIC;
>>> >> INIT_LIST_HEAD(&nbd_dev[i].waiting_queue);
>>> >> spin_lock_init(&nbd_dev[i].queue_lock);
>>> >> - spin_lock_init(&nbd_dev[i].sock_lock);
>>> >> + mutex_init(&nbd_dev[i].sock_lock);
>>> >> INIT_LIST_HEAD(&nbd_dev[i].queue_head);
>>> >> mutex_init(&nbd_dev[i].tx_lock);
>>> >> init_timer(&nbd_dev[i].timeout_timer);
>>> >> --
>>> >> 2.6.2
>>> >>
>>> >>
>>> >
>>> > --
>>> > Pengutronix e.K. | |
>>> > Industrial Linux Solutions | http://www.pengutronix.de/ |
>>> > Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
>>> > Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
>>>
>>>
>>>
>>>
>>
>> --
>> Pengutronix e.K. | |
>> Industrial Linux Solutions | http://www.pengutronix.de/ |
>> Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 |
>> Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
>
>
>
> --
> ---P.K.S

I've sent 4 new patches for your review. It would be great if you can
let me know about them so I
can work on any required changes over weekend.

--
---P.K.S