fixes for hangs and error reporting in CDC_WDM
RFC 1/1 fixes for hangs and error reporting in CDC_WDM
Stress testing has shown that CDC-WDM has some issues with hangs and error reporting
1. wakeups are not correctly handled in multhreaded environments
2. unresponsive hardware is not handled
3. errors are not correctly reported. This needs flush() to be implemented.
This version makes wdm_flush() use interruptible sleep.
For easier review all squashed together:
drivers/usb/class/cdc-wdm.c +55 -8
@@ -58,6 +58,9 @@ MODULE_DEVICE_TABLE (usb, wdm_ids);#define WDM_MAX 16 +/* flush() is uninterruptible, but we cannot wait forever */ +#define WDM_FLUSH_TIMEOUT (30 * HZ) +/* CDC-WMC r1.1 requires wMaxCommand to be "at least 256 decimal (0x100)" */ #define WDM_DEFAULT_BUFSIZE 256
@@ -151,7 +154,7 @@ static void wdm_out_callback(struct urb *urb)kfree (desc->outbuf); desc->outbuf =NULL ;clear_bit (WDM_IN_USE , &desc->flags); -wake_up (&desc->wait); +wake_up_all (&desc->wait); }static void wdm_in_callback (struct urb *urb)
@@ -393,6 +396,9 @@ static ssize_t wdm_writeif (test_bit (WDM_RESETTING , &desc->flags)) r = -EIO ; +if (test_bit (WDM_DISCONNECTING , &desc->flags)) + r = -ENODEV ; +if (r <0 ) { rv = r; goto out_free_mem_pm;
@@ -424,6 +430,7 @@ static ssize_t wdm_writeif (rv <0 ) { desc->outbuf =NULL ;clear_bit (WDM_IN_USE , &desc->flags); +wake_up_all (&desc->wait);/* for flush() */ dev_err (&desc->intf->dev,"Tx URB error: %d\n" , rv); rv =usb_translate_errors (rv); goto out_free_mem_pm;
@@ -583,11 +590,39 @@ static ssize_t wdm_readreturn rv; } +/* +* The difference to flush is that we wait forever. If you don't like +* that behavior, you need to send a signal. +*/ + +static int wdm_fsync (struct file *file,loff_t start,loff_t end,int datasync) +{ +struct wdm_device *desc = file->private_data; +int rv; + + rv =wait_event_interruptible (desc->wait, + !test_bit (WDM_IN_USE , &desc->flags) || +test_bit (WDM_DISCONNECTING , &desc->flags)); + +if (test_bit (WDM_DISCONNECTING , &desc->flags)) +return -ENODEV ; +if (rv <0 ) +return -EINTR ; + +spin_lock_irq (&desc->iuspin); + rv = desc->werr; + desc->werr =0 ; +spin_unlock_irq (&desc->iuspin); + +return usb_translate_errors (rv); +} +static int wdm_flush (struct file *file,fl_owner_t id) {struct wdm_device *desc = file->private_data; +int rv; - wait_event(desc->wait, + rv = wait_event_interruptible_timeout(desc->wait,/* * needs both flags. We cannot do with one * because resetting it would cause a race
@@ -595,16 +630,27 @@ static int wdm_flush(struct file *file, fl_owner_t id)* a disconnect */ !test_bit (WDM_IN_USE , &desc->flags) || - test_bit(WDM_DISCONNECTING, &desc->flags)); +test_bit (WDM_DISCONNECTING , &desc->flags), + WDM_FLUSH_TIMEOUT); -/* cannot dereference desc->intf if WDM_DISCONNECTING */ +/* +* to report the correct error. +* This is best effort +* We are inevitably racing with the hardware. +*/ if (test_bit (WDM_DISCONNECTING , &desc->flags))return -ENODEV ; -if (desc->werr <0 ) -dev_err (&desc->intf->dev,"Error in flush path: %d\n" , - desc->werr); +if (!rv) +return -EIO ; +if (rv <0 ) +return -EINTR ; + +spin_lock_irq (&desc->iuspin); + rv = desc->werr; + desc->werr =0 ; +spin_unlock_irq (&desc->iuspin); -return usb_translate_errors (desc->werr); +return usb_translate_errors (rv); }static __poll_t wdm_poll (struct file *file,struct poll_table_struct *wait)
@@ -729,6 +775,7 @@ static const struct file_operations wdm_fops = { .owner = THIS_MODULE, .read = wdm_read, .write = wdm_write, + .fsync = wdm_fsync, .open = wdm_open, .flush = wdm_flush, .release = wdm_release,
RFC 1/7 CDC-WDM: fix hangs in flush() in multithreaded cases
In a multithreaded scenario an arbitrary number of threads can be in wdm_write() and in wdm_flush(). Hence whenever WDM_IN_USE is reset, all possible waiters need to be notified. This is true whether this is due to IO completing or to an error starting IO.
--- drivers/usb/class/cdc-wdm.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-)
drivers/usb/class/cdc-wdm.c +7 -4
@@ -151,7 +151,7 @@ static void wdm_out_callback(struct urb *urb)kfree (desc->outbuf); desc->outbuf =NULL ;clear_bit (WDM_IN_USE , &desc->flags); -wake_up (&desc->wait); +wake_up_all (&desc->wait); }static void wdm_in_callback (struct urb *urb)
@@ -424,6 +424,7 @@ static ssize_t wdm_writeif (rv <0 ) { desc->outbuf =NULL ;clear_bit (WDM_IN_USE , &desc->flags); +wake_up_all (&desc->wait);/* for flush() */ dev_err (&desc->intf->dev,"Tx URB error: %d\n" , rv); rv =usb_translate_errors (rv); goto out_free_mem_pm;
@@ -586,6 +587,7 @@ static ssize_t wdm_readstatic int wdm_flush (struct file *file,fl_owner_t id) {struct wdm_device *desc = file->private_data; +int rv; wait_event(desc->wait,/*
@@ -600,11 +602,12 @@ static int wdm_flush(struct file *file, fl_owner_t id)/* cannot dereference desc->intf if WDM_DISCONNECTING */ if (test_bit (WDM_DISCONNECTING , &desc->flags))return -ENODEV ; -if (desc->werr <0 ) + rv = desc->werr; +if (rv <0 )dev_err (&desc->intf->dev,"Error in flush path: %d\n" , - desc->werr); + rv); -return usb_translate_errors (desc->werr); +return usb_translate_errors (rv); }static __poll_t wdm_poll (struct file *file,struct poll_table_struct *wait)
Signature · 1 line
2.16.4
RFC 2/7 CDC-WDM: introduce a timeout in flush()
In reply to Oliver NeukumMalicious or defective hardware may cease communication while flush() is running. In last consequence we need a timeout, as hardware that remains silent must be ignored.
Making the wait for IO interruptible would not solve the issue. While it would avoid a hang, it would not allow any progress and we would end up with an unclosable fd.
The 30 seconds are coming out of thin air.
--- drivers/usb/class/cdc-wdm.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
drivers/usb/class/cdc-wdm.c +9 -2
@@ -58,6 +58,9 @@ MODULE_DEVICE_TABLE (usb, wdm_ids);#define WDM_MAX 16 +/* flush() is uninterruptible, but we cannot wait forever */ +#define WDM_FLUSH_TIMEOUT (30 * HZ) +/* CDC-WMC r1.1 requires wMaxCommand to be "at least 256 decimal (0x100)" */ #define WDM_DEFAULT_BUFSIZE 256
@@ -589,7 +592,7 @@ static int wdm_flush(struct file *file, fl_owner_t id)struct wdm_device *desc = file->private_data;int rv; - wait_event(desc->wait, + rv = wait_event_timeout(desc->wait,/* * needs both flags. We cannot do with one * because resetting it would cause a race
@@ -597,11 +600,15 @@ static int wdm_flush(struct file *file, fl_owner_t id)* a disconnect */ !test_bit (WDM_IN_USE , &desc->flags) || - test_bit(WDM_DISCONNECTING, &desc->flags)); +test_bit (WDM_DISCONNECTING , &desc->flags), + WDM_FLUSH_TIMEOUT);/* cannot dereference desc->intf if WDM_DISCONNECTING */ if (test_bit (WDM_DISCONNECTING , &desc->flags))return -ENODEV ; +if (!rv) +return -EIO ; + rv = desc->werr;if (rv <0 ) dev_err(&desc->intf->dev, "Error in flush path: %d\n",
Signature · 1 line
2.16.4
RFC 3/7 CDC-WDM: remove use of intf->dev after potential disconnect
In reply to Oliver NeukumAfter a disconnect intf->dev is not a valid pointer any longer. As flush() uses it only for logging purposes logging is not worth it. Remove the dev_err()
--- drivers/usb/class/cdc-wdm.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-)
drivers/usb/class/cdc-wdm.c +5 -4
@@ -603,16 +603,17 @@ static int wdm_flush(struct file *file, fl_owner_t id)test_bit (WDM_DISCONNECTING , &desc->flags), WDM_FLUSH_TIMEOUT); -/* cannot dereference desc->intf if WDM_DISCONNECTING */ +/* +* to report the correct error. +* This is best effort +* We are inevitably racing with the hardware. +*/ if (test_bit (WDM_DISCONNECTING , &desc->flags))return -ENODEV ;if (!rv)return -EIO ; rv = desc->werr; -if (rv <0 ) -dev_err (&desc->intf->dev,"Error in flush path: %d\n" , - rv);return usb_translate_errors (rv); }
Signature · 1 line
2.16.4
RFC 4/7 CDC-WDM: fix race reporting errors in flush
In reply to Oliver NeukumIn case a race was lost and multiple fds used,
an error could be reported multiple times. To fix
this a spinlock must be taken.
--- drivers/usb/class/cdc-wdm.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/usb/class/cdc-wdm.c +3 -0
@@ -613,7 +613,10 @@ static int wdm_flush(struct file *file, fl_owner_t id)if (!rv)return -EIO ; +spin_lock_irq (&desc->iuspin); rv = desc->werr; + desc->werr =0 ; +spin_unlock_irq (&desc->iuspin);return usb_translate_errors (rv); }
Signature · 1 line
2.16.4
RFC 5/7 CDC-WDM: correct error reporting in write()
In reply to Oliver NeukumIn case our wait was interrupted by a disconnect, we should report that.
--- drivers/usb/class/cdc-wdm.c | 3 +++ 1 file changed, 3 insertions(+)
drivers/usb/class/cdc-wdm.c +3 -0
@@ -396,6 +396,9 @@ static ssize_t wdm_writeif (test_bit (WDM_RESETTING , &desc->flags)) r = -EIO ; +if (test_bit (WDM_DISCONNECTING , &desc->flags)) + r = -ENODEV ; +if (r <0 ) { rv = r; goto out_free_mem_pm;
Signature · 1 line
2.16.4
RFC 6/7 CDC-WDM: implement fsync
In reply to Oliver NeukumSome users want to be very sure that data has gone out to the device. This needs fsync.
--- drivers/usb/class/cdc-wdm.c | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+)
drivers/usb/class/cdc-wdm.c +28 -0
@@ -590,6 +590,33 @@ static ssize_t wdm_readreturn rv; } +/* +* The difference to flush is that we wait forever. If you don't like +* that behavior, you need to send a signal. +*/ + +static int wdm_fsync (struct file *file,loff_t start,loff_t end,int datasync) +{ +struct wdm_device *desc = file->private_data; +int rv; + + rv =wait_event_interruptible (desc->wait, + !test_bit (WDM_IN_USE , &desc->flags) || +test_bit (WDM_DISCONNECTING , &desc->flags)); + +if (test_bit (WDM_DISCONNECTING , &desc->flags)) +return -ENODEV ; +if (rv <0 ) +return -EINTR ; + +spin_lock_irq (&desc->iuspin); + rv = desc->werr; + desc->werr =0 ; +spin_unlock_irq (&desc->iuspin); + +return usb_translate_errors (rv); +} +static int wdm_flush (struct file *file,fl_owner_t id) {struct wdm_device *desc = file->private_data;
@@ -746,6 +773,7 @@ static const struct file_operations wdm_fops = { .owner = THIS_MODULE, .read = wdm_read, .write = wdm_write, + .fsync = wdm_fsync, .open = wdm_open, .flush = wdm_flush, .release = wdm_release,
Signature · 1 line
2.16.4
RFC 7/7 CDC-WDM: making flush() interruptible
In reply to Oliver NeukumThere is no need for flush() to be uninterruptible. close(2) is allowed to return -EINTR.
30 seconds is quite long a time to sleep in an uninterruptible state. Change it to an interruptible sleep.
--- drivers/usb/class/cdc-wdm.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/usb/class/cdc-wdm.c +3 -1
@@ -622,7 +622,7 @@ static int wdm_flush(struct file *file, fl_owner_t id)struct wdm_device *desc = file->private_data;int rv; - rv = wait_event_timeout(desc->wait, + rv = wait_event_interruptible_timeout(desc->wait,/* * needs both flags. We cannot do with one * because resetting it would cause a race
@@ -642,6 +642,8 @@ static int wdm_flush(struct file *file, fl_owner_t id)return -ENODEV ;if (!rv)return -EIO ; +if (rv <0 ) +return -EINTR ;spin_lock_irq (&desc->iuspin); rv = desc->werr;
Signature · 1 line
2.16.4
Re: [RFC 7/7] CDC-WDM: making flush() interruptible
Quoted 5 lines · Oliver Neukum
There is no need for
flush()to be uninterruptible. close(2) is allowed to return-EINTR.30 seconds is quite long a time to sleep in an uninterruptible state. Change it to an interruptible sleep.
Doesn't this conflict with
Making the wait for IO interruptible would not solve the issue. While it would avoid a hang, it would not allow any progress and we would end up with an unclosable fd.
in [RFC 2/7] ? I suggested killable version, for giving up upon SIGKILL implies automatically closing fds by terminating that userspace process.
Re: [RFC 7/7] CDC-WDM: making flush() interruptible
Am Dienstag, den 22.09.2020, 20:38 +0900 schrieb Tetsuo Handa:
Quoted 7 lines · Oliver Neukum
Quoted 5 lines
There is no need for
flush()to be uninterruptible. close(2) is allowed to return-EINTR.30 seconds is quite long a time to sleep in an uninterruptible state. Change it to an interruptible sleep.
Quoted 8 lines
Doesn't this conflict with
Making the wait for IO interruptible would not solve the issue. While it would avoid a hang, it would not allow any progress and we would end up with an unclosable fd.
in [RFC 2/7] ? I suggested killable version, for giving up upon SIGKILL implies automatically closing fds by terminating that userspace process.
No. That we state in an earlier patch that an issue needs a timeout instead of an interruptible sleep, does not mean that another issue could not be solved by using an interruptible sleep.
Regards
Oliver
Quoted 12 lines · Oliver Neukum
Stress testing has shown that CDC-WDM has some issues with hangs and error reporting
1. wakeups are not correctly handled in multhreaded environments
2. unresponsive hardware is not handled
3. errors are not correctly reported. This needsflush()to be implemented.This version makes
wdm_flush()use interruptible sleep.For easier review all squashed together:
I have like 3 or 4 different "RFC" series here from you for this driver, which one is the "newest"?
And can you send a series that isn't RFC so that I can know you feel it is good enough to be merged?
thanks,
greg k-h
Quoted 9 lines · Greg KH
Quoted 8 lines · Oliver Neukum
Stress testing has shown that CDC-WDM has some issues with hangs and error reporting
1. wakeups are not correctly handled in multhreaded environments
2. unresponsive hardware is not handled
3. errors are not correctly reported. This needsflush()to be implemented.
Quoted 7 lines
Quoted 4 lines
This version makes
wdm_flush()use interruptible sleep.For easier review all squashed together:
I have like 3 or 4 different "RFC" series here from you for this driver, which one is the "newest"?
https://lkml.kernel.org/r/[email protected]
is the newest series from Oliver. But
https://lkml.kernel.org/r/[email protected]
is the squashed version with updated comments and deduplicated code.
Quoted 3 lines
And can you send a series that isn't RFC so that I can know you feel it is good enough to be merged?
Do you want this fix as a series of patches (the former link)?
Since I think that the changeset should be atomically applied, I posted the latter link.
Quoted 24 lines · Tetsuo Handa
Quoted 9 lines · Greg KH
Quoted 8 lines · Oliver Neukum
Stress testing has shown that CDC-WDM has some issues with hangs and error reporting
1. wakeups are not correctly handled in multhreaded environments
2. unresponsive hardware is not handled
3. errors are not correctly reported. This needsflush()to be implemented.Quoted 7 lines
Quoted 4 lines
This version makes
wdm_flush()use interruptible sleep.For easier review all squashed together:
I have like 3 or 4 different "RFC" series here from you for this driver, which one is the "newest"?
https://lkml.kernel.org/r/[email protected]
is the newest series from Oliver. But
Quoted 8 lines
is the squashed version with updated comments and deduplicated code.
Quoted 3 lines
And can you send a series that isn't RFC so that I can know you feel it is good enough to be merged?
Do you want this fix as a series of patches (the former link)?
Since I think that the changeset should be atomically applied, I posted the latter link.
A single patch would be good to send to me again, burried at the end of a long thread is hard to dig out.
Also with proper authorship is needed, did Oliver write this, or did you?
There is the co-developed-by: tag, which looks like it might be relevant here, can you do that?
thanks,
greg k-h
Quoted 8 lines · Greg KH
Quoted 2 lines
Do you want this fix as a series of patches (the former link)?
Since I think that the changeset should be atomically applied, I posted the latter link.A single patch would be good to send to me again, burried at the end of a long thread is hard to dig out.
Also with proper authorship is needed, did Oliver write this, or did you?
Oliver wrote the majority part. I just squashed the series and updated comments and deduplicated the code. Thus, I think main authorship should be given to Oliver.
Quoted 3 lines
There is the co-developed-by: tag, which looks like it might be relevant here, can you do that?
If you prefer the co-developed-by: tag, you can add:
Oliver, any comments?
Quoted 14 lines · Tetsuo Handa
Quoted 8 lines · Greg KH
Quoted 2 lines
Do you want this fix as a series of patches (the former link)?
Since I think that the changeset should be atomically applied, I posted the latter link.A single patch would be good to send to me again, burried at the end of a long thread is hard to dig out.
Also with proper authorship is needed, did Oliver write this, or did you?
Oliver wrote the majority part. I just squashed the series and updated comments and deduplicated the code. Thus, I think main authorship should be given to Oliver.
Quoted 1 line
Quoted 6 lines
Quoted 2 lines
There is the co-developed-by: tag, which looks like it might be relevant here, can you do that?
If you prefer the co-developed-by: tag, you can add:
Co-developed-by Tetsuo Handa <[email protected]>
That seems reasonable, then put Oliver's address on the first line with a "From:" line to make this all work correctly when I apply it.
thanks,
greg k-h
Am Samstag, den 26.09.2020, 07:40 +0200 schrieb Greg KH:
Quoted 17 lines · Tetsuo Handa
Quoted 16 lines · Greg KH
Quoted 8 lines
Quoted 2 lines
Do you want this fix as a series of patches (the former link)?
Since I think that the changeset should be atomically applied, I posted the latter link.A single patch would be good to send to me again, burried at the end of a long thread is hard to dig out.
Also with proper authorship is needed, did Oliver write this, or did you?
Oliver wrote the majority part. I just squashed the series and updated comments and deduplicated the code. Thus, I think main authorship should be given to Oliver.
Quoted 2 lines
There is the co-developed-by: tag, which looks like it might be relevant here, can you do that?
Quoted 6 lines
Quoted 3 lines
If you prefer the co-developed-by: tag, you can add:
Co-developed-by Tetsuo Handa <[email protected]>That seems reasonable, then put Oliver's address on the first line with a "From:" line to make this all work correctly when I apply it.
Hi,
I did sent out the series with a PATCH label last week. Should I resend?
Regards
Oliver
Quoted 8 lines · Oliver Neukum
Quoted 2 lines
That seems reasonable, then put Oliver's address on the first line with a "From:" line to make this all work correctly when I apply it.
Hi,
I did sent out the series with a PATCH label last week. Should I resend?
No need to resend. I already sent a squashed version as
https://lkml.kernel.org/r/[email protected] .