Re: [PATCH] dvb-core: Fix several locking related problems.
Em Sáb, 2007-03-10 às 02:49 +0100, Johannes Stezenbach escreveu: > On Sun, Mar 04, 2007 at 05:45:54PM +, Simon Arlott wrote: > > Fix several instances of dvb-core functions using mutex_lock_interruptible > > and returning -ERESTARTSYS where the calling function will either never > > retry or never check the return value. > > > > These cause a race condition with dvb_dmxdev_filter_free and > > dvb_dvr_release, both of which are filesystem release functions whose > > return value is ignored and will never be retried. When this happens it > > becomes impossible to open dvr0 again (-EBUSY) since it has not been > > released properly. > > > > Signed-off-by: Simon Arlott <[EMAIL PROTECTED]> > > Acked-By: Johannes Stezenbach <[EMAIL PROTECTED]> > > I can't test this but to me it looks good. > Mauro, could you please pick it up and keep it in the > linuxtv.org repository for a while for testing? Done. Thanks, Johannes. -- Cheers, Mauro - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] dvb-core: Fix several locking related problems.
Em Sáb, 2007-03-10 às 02:49 +0100, Johannes Stezenbach escreveu: On Sun, Mar 04, 2007 at 05:45:54PM +, Simon Arlott wrote: Fix several instances of dvb-core functions using mutex_lock_interruptible and returning -ERESTARTSYS where the calling function will either never retry or never check the return value. These cause a race condition with dvb_dmxdev_filter_free and dvb_dvr_release, both of which are filesystem release functions whose return value is ignored and will never be retried. When this happens it becomes impossible to open dvr0 again (-EBUSY) since it has not been released properly. Signed-off-by: Simon Arlott [EMAIL PROTECTED] Acked-By: Johannes Stezenbach [EMAIL PROTECTED] I can't test this but to me it looks good. Mauro, could you please pick it up and keep it in the linuxtv.org repository for a while for testing? Done. Thanks, Johannes. -- Cheers, Mauro - To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH] dvb-core: Fix several locking related problems.
On Sun, Mar 04, 2007 at 05:45:54PM +, Simon Arlott wrote: > Fix several instances of dvb-core functions using mutex_lock_interruptible > and returning -ERESTARTSYS where the calling function will either never > retry or never check the return value. > > These cause a race condition with dvb_dmxdev_filter_free and > dvb_dvr_release, both of which are filesystem release functions whose > return value is ignored and will never be retried. When this happens it > becomes impossible to open dvr0 again (-EBUSY) since it has not been > released properly. > > Signed-off-by: Simon Arlott <[EMAIL PROTECTED]> Acked-By: Johannes Stezenbach <[EMAIL PROTECTED]> I can't test this but to me it looks good. Mauro, could you please pick it up and keep it in the linuxtv.org repository for a while for testing? Thanks, Johannes > --- > On 04/03/07 15:41, Andreas Oberritter wrote: > >please send an updated patch together with > >Signed-off-by line to Mauro <[EMAIL PROTECTED]> and ask him to apply > >it for inclusion into the -mm tree for further testing. > > Unless there are other -mm trees I've not heard about, presumably I should > just do this myself. Doesn't linux-dvb have it's own development tree this > would get better tested in? > > The dvb_dvr_release change has been working for me for 6 months and the > dvb_dmxdev_filter_free (dvb_dmxdev_filter_free) change looks equivalent. > See http://www.linuxtv.org/pipermail/linux-dvb/2007-February/016120.html > for an example of the bug before and after fixing. > > All the other changes run ok for me but should have lockdep enabled when > testing (if there's a possible deadlock somewhere, using _interruptible > will hide it). > > drivers/media/dvb/dvb-core/dmxdev.c| 12 +++- > drivers/media/dvb/dvb-core/dvb_demux.c | 21 +++-- > drivers/media/dvb/dvb-core/dvbdev.c|9 +++-- > 3 files changed, 13 insertions(+), 29 deletions(-) > > diff --git a/drivers/media/dvb/dvb-core/dmxdev.c > b/drivers/media/dvb/dvb-core/dmxdev.c > index fc77de4..a5c0e1a 100644 > --- a/drivers/media/dvb/dvb-core/dmxdev.c > +++ b/drivers/media/dvb/dvb-core/dmxdev.c > @@ -180,8 +180,7 @@ static int dvb_dvr_release(struct inode *inode, struct > file *file) > struct dvb_device *dvbdev = file->private_data; > struct dmxdev *dmxdev = dvbdev->priv; > > - if (mutex_lock_interruptible(>mutex)) > - return -ERESTARTSYS; > + mutex_lock(>mutex); > > if ((file->f_flags & O_ACCMODE) == O_WRONLY) { > dmxdev->demux->disconnect_frontend(dmxdev->demux); > @@ -673,13 +672,8 @@ static int dvb_demux_open(struct inode *inode, struct > file *file) > static int dvb_dmxdev_filter_free(struct dmxdev *dmxdev, > struct dmxdev_filter *dmxdevfilter) > { > - if (mutex_lock_interruptible(>mutex)) > - return -ERESTARTSYS; > - > - if (mutex_lock_interruptible(>mutex)) { > - mutex_unlock(>mutex); > - return -ERESTARTSYS; > - } > + mutex_lock(>mutex); > + mutex_lock(>mutex); > > dvb_dmxdev_filter_stop(dmxdevfilter); > dvb_dmxdev_filter_reset(dmxdevfilter); > diff --git a/drivers/media/dvb/dvb-core/dvb_demux.c > b/drivers/media/dvb/dvb-core/dvb_demux.c > index fcff5ea..6d8d1c3 100644 > --- a/drivers/media/dvb/dvb-core/dvb_demux.c > +++ b/drivers/media/dvb/dvb-core/dvb_demux.c > @@ -673,8 +673,7 @@ static int dmx_ts_feed_stop_filtering(struct > dmx_ts_feed *ts_feed) > struct dvb_demux *demux = feed->demux; > int ret; > > - if (mutex_lock_interruptible(>mutex)) > - return -ERESTARTSYS; > + mutex_lock(>mutex); > > if (feed->state < DMX_STATE_GO) { > mutex_unlock(>mutex); > @@ -748,8 +747,7 @@ static int dvbdmx_release_ts_feed(struct dmx_demux *dmx, > struct dvb_demux *demux = (struct dvb_demux *)dmx; > struct dvb_demux_feed *feed = (struct dvb_demux_feed *)ts_feed; > > - if (mutex_lock_interruptible(>mutex)) > - return -ERESTARTSYS; > + mutex_lock(>mutex); > > if (feed->state == DMX_STATE_FREE) { > mutex_unlock(>mutex); > @@ -916,8 +914,7 @@ static int dmx_section_feed_stop_filtering(struct > dmx_section_feed *feed) > struct dvb_demux *dvbdmx = dvbdmxfeed->demux; > int ret; > > - if (mutex_lock_interruptible(>mutex)) > - return -ERESTARTSYS; > + mutex_lock(>mutex); > > if (!dvbdmx->stop_feed) { > mutex_unlock(>mutex); > @@ -942,8 +939,7 @@ static int dmx_section_feed_release_filter(struct > dmx_section_feed *feed, > struct dvb_demux_feed *dvbdmxfeed = (struct dvb_demux_feed *)feed; > struct dvb_demux *dvbdmx = dvbdmxfeed->demux; > > - if (mutex_lock_interruptible(>mutex)) > - return -ERESTARTSYS; > + mutex_lock(>mutex); > > if (dvbdmxfilter->feed != dvbdmxfeed) { > mutex_unlock(>mutex); > @@ -1016,8 +1012,7 @@
Re: [PATCH] dvb-core: Fix several locking related problems.
On Sun, Mar 04, 2007 at 05:45:54PM +, Simon Arlott wrote: Fix several instances of dvb-core functions using mutex_lock_interruptible and returning -ERESTARTSYS where the calling function will either never retry or never check the return value. These cause a race condition with dvb_dmxdev_filter_free and dvb_dvr_release, both of which are filesystem release functions whose return value is ignored and will never be retried. When this happens it becomes impossible to open dvr0 again (-EBUSY) since it has not been released properly. Signed-off-by: Simon Arlott [EMAIL PROTECTED] Acked-By: Johannes Stezenbach [EMAIL PROTECTED] I can't test this but to me it looks good. Mauro, could you please pick it up and keep it in the linuxtv.org repository for a while for testing? Thanks, Johannes --- On 04/03/07 15:41, Andreas Oberritter wrote: please send an updated patch together with Signed-off-by line to Mauro [EMAIL PROTECTED] and ask him to apply it for inclusion into the -mm tree for further testing. Unless there are other -mm trees I've not heard about, presumably I should just do this myself. Doesn't linux-dvb have it's own development tree this would get better tested in? The dvb_dvr_release change has been working for me for 6 months and the dvb_dmxdev_filter_free (dvb_dmxdev_filter_free) change looks equivalent. See http://www.linuxtv.org/pipermail/linux-dvb/2007-February/016120.html for an example of the bug before and after fixing. All the other changes run ok for me but should have lockdep enabled when testing (if there's a possible deadlock somewhere, using _interruptible will hide it). drivers/media/dvb/dvb-core/dmxdev.c| 12 +++- drivers/media/dvb/dvb-core/dvb_demux.c | 21 +++-- drivers/media/dvb/dvb-core/dvbdev.c|9 +++-- 3 files changed, 13 insertions(+), 29 deletions(-) diff --git a/drivers/media/dvb/dvb-core/dmxdev.c b/drivers/media/dvb/dvb-core/dmxdev.c index fc77de4..a5c0e1a 100644 --- a/drivers/media/dvb/dvb-core/dmxdev.c +++ b/drivers/media/dvb/dvb-core/dmxdev.c @@ -180,8 +180,7 @@ static int dvb_dvr_release(struct inode *inode, struct file *file) struct dvb_device *dvbdev = file-private_data; struct dmxdev *dmxdev = dvbdev-priv; - if (mutex_lock_interruptible(dmxdev-mutex)) - return -ERESTARTSYS; + mutex_lock(dmxdev-mutex); if ((file-f_flags O_ACCMODE) == O_WRONLY) { dmxdev-demux-disconnect_frontend(dmxdev-demux); @@ -673,13 +672,8 @@ static int dvb_demux_open(struct inode *inode, struct file *file) static int dvb_dmxdev_filter_free(struct dmxdev *dmxdev, struct dmxdev_filter *dmxdevfilter) { - if (mutex_lock_interruptible(dmxdev-mutex)) - return -ERESTARTSYS; - - if (mutex_lock_interruptible(dmxdevfilter-mutex)) { - mutex_unlock(dmxdev-mutex); - return -ERESTARTSYS; - } + mutex_lock(dmxdev-mutex); + mutex_lock(dmxdevfilter-mutex); dvb_dmxdev_filter_stop(dmxdevfilter); dvb_dmxdev_filter_reset(dmxdevfilter); diff --git a/drivers/media/dvb/dvb-core/dvb_demux.c b/drivers/media/dvb/dvb-core/dvb_demux.c index fcff5ea..6d8d1c3 100644 --- a/drivers/media/dvb/dvb-core/dvb_demux.c +++ b/drivers/media/dvb/dvb-core/dvb_demux.c @@ -673,8 +673,7 @@ static int dmx_ts_feed_stop_filtering(struct dmx_ts_feed *ts_feed) struct dvb_demux *demux = feed-demux; int ret; - if (mutex_lock_interruptible(demux-mutex)) - return -ERESTARTSYS; + mutex_lock(demux-mutex); if (feed-state DMX_STATE_GO) { mutex_unlock(demux-mutex); @@ -748,8 +747,7 @@ static int dvbdmx_release_ts_feed(struct dmx_demux *dmx, struct dvb_demux *demux = (struct dvb_demux *)dmx; struct dvb_demux_feed *feed = (struct dvb_demux_feed *)ts_feed; - if (mutex_lock_interruptible(demux-mutex)) - return -ERESTARTSYS; + mutex_lock(demux-mutex); if (feed-state == DMX_STATE_FREE) { mutex_unlock(demux-mutex); @@ -916,8 +914,7 @@ static int dmx_section_feed_stop_filtering(struct dmx_section_feed *feed) struct dvb_demux *dvbdmx = dvbdmxfeed-demux; int ret; - if (mutex_lock_interruptible(dvbdmx-mutex)) - return -ERESTARTSYS; + mutex_lock(dvbdmx-mutex); if (!dvbdmx-stop_feed) { mutex_unlock(dvbdmx-mutex); @@ -942,8 +939,7 @@ static int dmx_section_feed_release_filter(struct dmx_section_feed *feed, struct dvb_demux_feed *dvbdmxfeed = (struct dvb_demux_feed *)feed; struct dvb_demux *dvbdmx = dvbdmxfeed-demux; - if (mutex_lock_interruptible(dvbdmx-mutex)) - return -ERESTARTSYS; + mutex_lock(dvbdmx-mutex); if (dvbdmxfilter-feed != dvbdmxfeed) { mutex_unlock(dvbdmx-mutex); @@ -1016,8 +1012,7 @@ static int