containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

PATCH New [email protected] 3 participants 10 messages Oct 30, 2015 – Nov 10, 2015

lore

Share & tools

lore

[email protected]

author

2015-10-30 08:57 UTC #0 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

From: Shuang Qiu <[email protected]>

If dummy is compiled as module in kernel,it is loaded dynamically when adding dummy device.And it will also create a default dummy interface.So the sysfs_after will have one more interface than sysfs_before which makes testcase #3 failed.Loading dummy module before collecting sysfs interface to workaround such issue.

Signed-off-by Shuang Qiu <[email protected]>

---
 testcases/kernel/containers/netns/netns_sysfs.sh |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

testcases/kernel/containers/netns/netns_sysfs.sh +4 -0

index 7dea52b..b5791bd 100755
@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
 	tst_brkm TBROK "unable to create a new network namespace"
 fi
 TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"
 ls /sys/class/net >sysfs_before
 
 ns_exec $NS_HANDLE $NS_TYPE mount --make-rprivate /sys

Signature · 1 line

1.7.7

Cyril Hrubis

2015-11-04 12:49 UTC #1 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Hi!

testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 53

Quoted 12 lines

testcases/kernel/containers/netns/netns_sysfs.sh +4 -0

index 7dea52b..b5791bd 100755
@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
 	tst_brkm TBROK "unable to create a new network namespace"
 fi
 TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"

But this will fail when dummy wasn't compiled in the kernel, right?

Since modprobe will exit with 1 when module couldn't be found. So we should either just call 'modprobe dummy' and ignore the return value or change the code to ignore the default dummy interface.

Signature · 2 lines

Cyril Hrubis
[email protected]

Shuang Qiu

author

2015-11-05 04:46 UTC #2 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Hi Cyril,

Thanks for review.

Quoted 14 lines · Cyril Hrubis

Hi!
testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 53
Quoted 12 lines
testcases/kernel/containers/netns/netns_sysfs.sh +4 -0
index 7dea52b..b5791bd 100755
@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
  	tst_brkm TBROK "unable to create a new network namespace"
  fi
  TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"
But this will fail when dummy wasn't compiled in the kernel, right?

Yes,if dummy wasn't compiled in the kernel,it will also fail when adding dummy device in the next step:
===cut===
ns_exec $NS_HANDLE $NS_TYPE ip link add $DUMMYDEV type dummy || \
tst_brkm TBROK "failed to add a new dummy device"
===cut===

Quoted 4 lines

Since modprobe will exit with 1 when module couldn't be found. So we should either just call 'modprobe dummy' and ignore the return value or change the code to ignore the default dummy interface.

I prefer 'modprobe dummy' and ignore the return value,because I'm not sure whether there is any other default dummy interface name except "dummy0".

Thanks
Shuang

Quoted 1 line

Cyril Hrubis

2015-11-05 14:05 UTC #3 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Hi!

Quoted 13 lines

Quoted 5 lines
Quoted 5 lines
testcases/kernel/containers/netns/netns_sysfs.sh +0 -0
index 7dea52b..b5791bd 100755
@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
Quoted 8 lines
testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 53
Quoted 7 lines
testcases/kernel/containers/netns/netns_sysfs.sh +4 -0
  	tst_brkm TBROK "unable to create a new network namespace"
  fi
  TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"
But this will fail when dummy wasn't compiled in the kernel, right?

Quoted 6 lines

Yes,if dummy wasn't compiled in the kernel,it will also fail when adding dummy device in the next step:
===cut===
ns_exec $NS_HANDLE $NS_TYPE ip link add $DUMMYDEV type dummy || \
tst_brkm TBROK "failed to add a new dummy device"
===cut===

I meaned "wasn't compiled as module" i.e. compiled in kernel, sorry for the confusion. In that case this would work fine but the modprobe would fail.

Quoted 6 lines

Quoted 3 lines
Since modprobe will exit with 1 when module couldn't be found. So we should either just call 'modprobe dummy' and ignore the return value or change the code to ignore the default dummy interface.
I prefer 'modprobe dummy' and ignore the return value,because I'm not sure whether there is any other default dummy interface name except "dummy0".

Sounds good to me.

Signature · 2 lines

Cyril Hrubis
[email protected]

Shuang Qiu

author

2015-11-06 10:28 UTC #4 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Quoted 20 lines · Cyril Hrubis

Hi!
Quoted 13 lines
Quoted 5 lines
Quoted 5 lines
testcases/kernel/containers/netns/netns_sysfs.sh +0 -0
index 7dea52b..b5791bd 100755
@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
Quoted 8 lines
testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 53
Quoted 7 lines
testcases/kernel/containers/netns/netns_sysfs.sh +4 -0
   	tst_brkm TBROK "unable to create a new network namespace"
   fi
   TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"
But this will fail when dummy wasn't compiled in the kernel, right?
Quoted 6 lines
Yes,if dummy wasn't compiled in the kernel,it will also fail when adding dummy device in the next step:
===cut===
ns_exec $NS_HANDLE $NS_TYPE ip link add $DUMMYDEV type dummy || \
tst_brkm TBROK "failed to add a new dummy device"
===cut===

Quoted 3 lines

I meaned "wasn't compiled as module" i.e. compiled in kernel, sorry for the confusion. In that case this would work fine but the modprobe would fail.

Aha,yes,thanks for clarification.

Quoted 8 lines

Quoted 7 lines
Quoted 4 lines
Since modprobe will exit with 1 when module couldn't be found. So we should either just call 'modprobe dummy' and ignore the return value or change the code to ignore the default dummy interface.
I prefer 'modprobe dummy' and ignore the return value,because I'm not sure whether there is any other default dummy interface name except "dummy0".
Sounds good to me.

ok,I will submit another patch.

Thanks
Shuang

Quoted 1 line

Jiri Jaburek

2015-11-06 17:16 UTC #5 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

In reply to [email protected]

Quoted 7 lines · On 10/30/2015 09:57 AM, [email protected] wrote:

From: Shuang Qiu <[email protected]>
If dummy is compiled as module in kernel,it is loaded dynamically when adding dummy device.And it will also create a default dummy interface.So the sysfs_after will have one more interface than sysfs_before which makes testcase #3 failed.Loading dummy module before collecting sysfs interface to workaround such issue.

I cannot reproduce this on any RHEL or Fedora system, but I can reproduce it on Debian (Jessie). If we accounted only for the module version of 'dummy', it could be countered by adding

options dummy numdummies=0

to any /etc/modprobe.d/* config (temporarily), however for the built-in version of the module, such option would need to be passed via kernel cmdline.

We could check [ -d /sys/module/dummy ] and if it exists, presume that no extra setup is necessary, because

1) if it's a module, our creation of additional dummy interface won't affect existing interfaces as dummy0 is created only on insmod (dummy_init_one() is called during __init with numdummies=1)
2) if it's built in, well .. see (1)

otherwise we can try modprobing it and if it fails, end with CONF.

However - wouldn't it be simply easier to add an exception for dummy0 during the comparison check? The test operates with iface names other than dummy0, so a possible namespace bug would be found anyway. (See attached diff.)

However the (quite possibly) best solution would be a one which cleans up after itself, so maybe the [ -d /sys/module/dummy ] solution with explicit modprobe / modprobe -r would work better.

Your choice. :)

Jiri

Quoted 11 lines

Signed-off-by Shuang Qiu <[email protected]>
---
 testcases/kernel/containers/netns/netns_sysfs.sh |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)
testcases/kernel/containers/netns/netns_sysfs.sh +0 -0
index 7dea52b..b5791bd 100755

testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 56

Quoted 10 lines

testcases/kernel/containers/netns/netns_sysfs.sh +4 -0

@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
 	tst_brkm TBROK "unable to create a new network namespace"
 fi
 TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"
 ls /sys/class/net >sysfs_before
 
 ns_exec $NS_HANDLE $NS_TYPE mount --make-rprivate /sys

-------------- next part --------------

testcases/kernel/containers/netns/netns_sysfs.sh +5 -2

index 7dea52b..e725a50 100755
@@ -47,7 +47,10 @@ if [ $? -eq 1 ]; then
 	tst_brkm TBROK "unable to create a new network namespace"
 fi
 TST_CLEANUP=cleanup
-ls /sys/class/net >sysfs_before
+
+# exclude dummy0 from comparison as it gets automatically created by the
+# dummy device driver upon insmod/modprobe (during ip link add)
+ls /sys/class/net | grep -v dummy0 >sysfs_before
 
 ns_exec $NS_HANDLE $NS_TYPE mount --make-rprivate /sys
 ns_exec $NS_HANDLE $NS_TYPE ip link add $DUMMYDEV type dummy || \

@@ -84,7 +87,7 @@ fi
 
 
 # TEST CASE #3
-ls /sys/class/net >sysfs_after
+ls /sys/class/net | grep -v dummy0 >sysfs_after
 diff sysfs_before sysfs_after
 if [ $? -eq 0 ]; then
 	tst_resm TPASS "sysfs not affected by a separate namespace"

Cyril Hrubis

2015-11-09 14:14 UTC #6 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Hi!

Quoted 8 lines

However - wouldn't it be simply easier to add an exception for dummy0 during the comparison check? The test operates with iface names other than dummy0, so a possible namespace bug would be found anyway. (See attached diff.)
However the (quite possibly) best solution would be a one which cleans up after itself, so maybe the [ -d /sys/module/dummy ] solution with explicit modprobe / modprobe -r would work better.

Either one works for me.

testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 53

Quoted 13 lines

testcases/kernel/containers/netns/netns_sysfs.sh +4 -1

index 7dea52b..e725a50 100755
@@ -47,7 +47,10 @@ if [ $? -eq 1 ]; then
 	tst_brkm TBROK "unable to create a new network namespace"
 fi
 TST_CLEANUP=cleanup
-ls /sys/class/net >sysfs_before
+
+# exclude dummy0 from comparison as it gets automatically created by the
+# dummy device driver upon insmod/modprobe (during ip link add)
+ls /sys/class/net | grep -v dummy0 >sysfs_before

Maybe we should do grep -v dummy[0-9].* in order to skipp all dummy interfaces if someone changes the number of interfaces created. If we decide to go this way.

Signature · 2 lines

Cyril Hrubis
[email protected]

Jiri Jaburek

2015-11-09 15:20 UTC #7 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Quoted 11 lines · Cyril Hrubis

Hi!
Quoted 8 lines
However - wouldn't it be simply easier to add an exception for dummy0 during the comparison check? The test operates with iface names other than dummy0, so a possible namespace bug would be found anyway. (See attached diff.)
However the (quite possibly) best solution would be a one which cleans up after itself, so maybe the [ -d /sys/module/dummy ] solution with explicit modprobe / modprobe -r would work better.
Either one works for me.

I would send a patch with the more "clean" solution, but I'm not sure how to modularize the cleanup function here as I need to add modprobe -r only if the modprobe was previously done.
Furthermore, doing it that way would make the test impossible to run multiple times in parallel.

So it boils down to either doing a preemptive modprobe 2>/dev/null or ignoring dummy[0-9]+, it's up to you. :)

Quoted 10 lines

testcases/kernel/containers/netns/netns_sysfs.sh +0 -1

index 7dea52b..e725a50 100755
@@ -47,7 +47,10 @@ if [ $? -eq 1 ]; then
 	tst_brkm TBROK "unable to create a new network namespace"
 fi
 TST_CLEANUP=cleanup
-ls /sys/class/net >sysfs_before

Quoted 9 lines

testcases/kernel/containers/netns/netns_sysfs.sh if [ $? -eq 1 ]; then · line 53
Quoted 4 lines
testcases/kernel/containers/netns/netns_sysfs.sh +4 -0
+
+# exclude dummy0 from comparison as it gets automatically created by the
+# dummy device driver upon insmod/modprobe (during ip link add)
+ls /sys/class/net | grep -v dummy0 >sysfs_before
Maybe we should do grep -v dummy[0-9].* in order to skipp all dummy interfaces if someone changes the number of interfaces created. If we decide to go this way.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-containers-netns-exclude-dummy0-from-before-after-co.patch
Type: text/x-patch
Size: 1338 bytes
Desc: not available
URL: <http://lists.linux.it/pipermail/ltp/attachments/20151109/7f8a8c7c/attachment-0001.bin>

Cyril Hrubis

2015-11-09 15:26 UTC #8 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

Hi!

Quoted 2 lines

So it boils down to either doing a preemptive modprobe 2>/dev/null or ignoring dummy[0-9]+, it's up to you. :)

I don't really care :), both are valid solutions. Maybe we should flip a coin and decide.

Well since you attached a patch to this email allready I will spare us from doubt and simply apply it, thanks.

Signature · 2 lines

Cyril Hrubis
[email protected]

Shuang Qiu

author

2015-11-10 04:03 UTC #9 lore raw

PATCH 1/1 containers/netns/netns_sysfs.sh:load dummy module before collecting sysfs interface

In reply to Jiri Jaburek

Hi Jiri,

Quoted 26 lines · Jiri Jaburek

Quoted 7 lines · On 10/30/2015 09:57 AM, [email protected] wrote:
From: Shuang Qiu <[email protected]>
If dummy is compiled as module in kernel,it is loaded dynamically when adding dummy device.And it will also create a default dummy interface.So the sysfs_after will have one more interface than sysfs_before which makes testcase #3 failed.Loading dummy module before collecting sysfs interface to workaround such issue.
I cannot reproduce this on any RHEL or Fedora system, but I can reproduce it on Debian (Jessie). If we accounted only for the module version of 'dummy', it could be countered by adding
options dummy numdummies=0
to any /etc/modprobe.d/* config (temporarily), however for the built-in version of the module, such option would need to be passed via kernel cmdline.
We could check [ -d /sys/module/dummy ] and if it exists, presume that no extra setup is necessary, because
1) if it's a module, our creation of additional dummy interface won't affect existing interfaces as dummy0 is created only on insmod (dummy_init_one() is called during __init with numdummies=1)
2) if it's built in, well .. see (1)

Quoted 6 lines

otherwise we can try modprobing it and if it fails, end with CONF.
However - wouldn't it be simply easier to add an exception for dummy0 during the comparison check? The test operates with iface names other than dummy0, so a possible namespace bug would be found anyway. (See attached diff.)

I did not learn dummy module much before and not sure if there is a specify iface name.
As you explained,we will have a expect name and may find a namespace bug,so I also prefer your way.

Thanks
Shuang

Quoted 30 lines

However the (quite possibly) best solution would be a one which cleans up after itself, so maybe the [ -d /sys/module/dummy ] solution with explicit modprobe / modprobe -r would work better.
Your choice. :)
Jiri
Quoted 21 lines
Signed-off-by Shuang Qiu <[email protected]>
---
  testcases/kernel/containers/netns/netns_sysfs.sh |    4 ++++
  1 files changed, 4 insertions(+), 0 deletions(-)
testcases/kernel/containers/netns/netns_sysfs.sh +4 -0
index 7dea52b..b5791bd 100755
@@ -47,6 +47,10 @@ if [ $? -eq 1 ]; then
  	tst_brkm TBROK "unable to create a new network namespace"
  fi
  TST_CLEANUP=cleanup
+
+#Load dummy module before collecting sysfs interface
+lsmod | grep dummy || modprobe dummy
+[ $? -eq 0 ] || tst_brkm TBROK "failed to load dummy module"
  ls /sys/class/net >sysfs_before
 
  ns_exec $NS_HANDLE $NS_TYPE mount --make-rprivate /sys