[PATCH] Hardware iSCSI support

[WANG Chao]

Currently we retrieve all the necessary information of an iSCSI session
and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
because dracut has already had its own infrastructure to do bring up any
iscsi device configured in firmware of an HBA (Host Bus Bridge).

This patch will determine if an iscsi session is established by firmware
configured iscsi. If it does, then we don't pass anything down to 2nd
kernel. dracut will handle it.

I read some iscsi code and I also talked with storage QE. It's confirmed
that software iscsi will use "tcp" transport class while hardware iscsi
will use like "be2iscsi" "bnx2i" transport class. It's ok to use
the transport name of an iface of a scsi session to determine hardware
iscsi or not. Additionally using "iscsi_firmware" "rd.iscsi.firmware"
to
determine if the system is iscsi boot, because dracut can only handle
this case.

Along this patch, I add the infrastructure of determine iBFT, but it's
not complete yet. Mark the part as "FIXME". Once customer is asking or
maybe I get hold of iBFT box, I can start working on it.

Signed-off-by: WANG Chao <chaowang(a)redhat.com&gt;
---
 dracut-module-setup.sh | 77 +++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 76 insertions(+), 1 deletion(-)

diff --git a/dracut-module-setup.sh b/dracut-module-setup.sh
index dcebc47..9616866 100755
--- a/dracut-module-setup.sh
+++ b/dracut-module-setup.sh
@@ -411,7 +411,78 @@ kdump_get_iscsi_initiator() {
     return 1
 }
 
-# No ibft handling yet.
+# Determine hardware iscsi by an iface's transport class.
+# If transport class is tcp, it's software iscsi.
+# If it's be2iscsi, bnx2i, cxgb3i, cxgb4i, it's hardware iscsi.
+kdump_is_iscsi_hw_session() {
+    if [ "$(kdump_iscsi_get_rec_val $1 "iface.transport_name")" = tcp
]; then
+        return 1
+    fi
+
+    return 0
+}
+
+# FIXME: determine iBFT
+kdump_is_iscsi_ibft_session() {
+    return 1
+}
+
+# kernel cmdline must contain these either of the following arguments
+# for booting off hardware iSCSI:
+# rd.iscsi.firmware[=1], iscsi_firmware
+kdump_is_iscsi_hw_cmdline() {
+    local cmdline
+
+    cmdline="$(cat /proc/cmdline)"
+
+    for arg in $cmdline; do
+        case "$arg" in
+            "rd.iscsi.firmware" |\
+            "rd.iscsi.firmware=1" |\
+            "iscsi_firmware")
+                return 0
+                ;;
+        esac
+    done
+
+    return 1
+}
+
+# kernel cmdline must contain these either of the following arguments
+# for booting off iBFT:
+# rd.iscsi.ibft[=1], ip=ibft
+kdump_is_iscsi_ibft_cmdline() {
+    local cmdline
+
+    cmdline="$(cat /proc/cmdline)"
+
+    for arg in $cmdline; do
+        case "$arg" in
+            "rd.iscsi.ibft" |\
+            "rd.iscsi.ibft=1" |\
+            "ip=ibft")
+                return 0
+                ;;
+        esac
+    done
+
+    return 1
+}
+
+# Sanity check if an iscsi session is established by iBFT or firmware
+kdump_is_iscsi_fw() {
+
+    if kdump_is_iscsi_hw_cmdline && kdump_is_iscsi_hw_session $1; then
+        return 0
+    fi
+
+    if kdump_is_iscsi_ibft_cmdline && kdump_is_iscsi_ibft_session $1; then
+        return 0
+    fi
+
+    return 1
+}
+
 kdump_setup_iscsi_device() {
     local path=$1
     local tgt_name; local tgt_ipaddr;
@@ -434,6 +505,10 @@ kdump_setup_iscsi_device() {
         return 1
     fi
 
+    if kdump_is_iscsi_fw ${path}; then
+        return 0;
+    fi
+
     tgt_name=$(kdump_iscsi_get_rec_val ${path} "node.name")
     tgt_ipaddr=$(kdump_iscsi_get_rec_val ${path} "node.conn\[0\].address")
 
-- 
1.9.3

[Dave Young]

Hi,

Well done, thanks Chao.

Vivek/Bao/Minfei, could you review the patch as well?

On 10/28/14 at 06:43pm, WANG Chao wrote:
>  Currently we retrieve all the necessary information of an iSCSI
> session
>  and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  because dracut has already had its own infrastructure to do bring up any
>  iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  
>  This patch will determine if an iscsi session is established by firmware
>  configured iscsi. If it does, then we don't pass anything down to 2nd
>  kernel. dracut will handle it.
>  
>  I read some iscsi code and I also talked with storage QE. It's confirmed
>  that software iscsi will use "tcp" transport class while hardware iscsi
>  will use like "be2iscsi" "bnx2i" transport class. It's ok to use
>  the transport name of an iface of a scsi session to determine hardware
>  iscsi or not. Additionally using "iscsi_firmware" "rd.iscsi.firmware"
> to
>  determine if the system is iscsi boot, because dracut can only handle
>  this case.
>  
>  Along this patch, I add the infrastructure of determine iBFT, but it's
>  not complete yet. Mark the part as "FIXME". Once customer is asking or
>  maybe I get hold of iBFT box, I can start working on it. 
For non-root hwiscsi there's no kernel cmdline, you can put it in todo list
We can support root device firstly and see if it works well.

>  
>  Signed-off-by: WANG Chao <chaowang(a)redhat.com&gt;
>  ---
>   dracut-module-setup.sh | 77 +++++++++++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 76 insertions(+), 1 deletion(-)
>  
>  diff --git a/dracut-module-setup.sh b/dracut-module-setup.sh
>  index dcebc47..9616866 100755
>  --- a/dracut-module-setup.sh
>  +++ b/dracut-module-setup.sh
>  @@ -411,7 +411,78 @@ kdump_get_iscsi_initiator() {
>       return 1
>   }
>   
>  -# No ibft handling yet.
>  +# Determine hardware iscsi by an iface's transport class.
>  +# If transport class is tcp, it's software iscsi.
>  +# If it's be2iscsi, bnx2i, cxgb3i, cxgb4i, it's hardware iscsi.
>  +kdump_is_iscsi_hw_session() {
>  +    if [ "$(kdump_iscsi_get_rec_val $1 "iface.transport_name")" =
> tcp ]; then
>  +        return 1
>  +    fi
>  +
>  +    return 0
>  +}
>  +
>  +# FIXME: determine iBFT
>  +kdump_is_iscsi_ibft_session() {
>  +    return 1
>  +}
>  +
>  +# kernel cmdline must contain these either of the following arguments
>  +# for booting off hardware iSCSI:
>  +# rd.iscsi.firmware[=1], iscsi_firmware
>  +kdump_is_iscsi_hw_cmdline() {
>  +    local cmdline
>  +
>  +    cmdline="$(cat /proc/cmdline)"
>  +
>  +    for arg in $cmdline; do
>  +        case "$arg" in
>  +            "rd.iscsi.firmware" |\
>  +            "rd.iscsi.firmware=1" |\
>  +            "iscsi_firmware") 
dracut.cmdline manpage shows below:
       iscsi_firmware
           rd.iscsi.firmware=0

So looks like iscsi_firmware == rd.iscsi.firmware=0

>  +                return 0
>  +                ;;
>  +        esac
>  +    done
>  +
>  +    return 1
>  +}
>  +
>  +# kernel cmdline must contain these either of the following arguments
>  +# for booting off iBFT:
>  +# rd.iscsi.ibft[=1], ip=ibft
>  +kdump_is_iscsi_ibft_cmdline() {
>  +    local cmdline
>  +
>  +    cmdline="$(cat /proc/cmdline)"
>  +
>  +    for arg in $cmdline; do
>  +        case "$arg" in
>  +            "rd.iscsi.ibft" |\
>  +            "rd.iscsi.ibft=1" |\
>  +            "ip=ibft")
>  +                return 0
>  +                ;;
>  +        esac
>  +    done
>  +
>  +    return 1
>  +}
>  +
>  +# Sanity check if an iscsi session is established by iBFT or firmware
>  +kdump_is_iscsi_fw() {
>  +
>  +    if kdump_is_iscsi_hw_cmdline && kdump_is_iscsi_hw_session $1; then
>  +        return 0
>  +    fi
>  +
>  +    if kdump_is_iscsi_ibft_cmdline && kdump_is_iscsi_ibft_session $1; then
>  +        return 0
>  +    fi
>  +
>  +    return 1
>  +}
>  +
>   kdump_setup_iscsi_device() {
>       local path=$1
>       local tgt_name; local tgt_ipaddr;
>  @@ -434,6 +505,10 @@ kdump_setup_iscsi_device() {
>           return 1
>       fi
>   
>  +    if kdump_is_iscsi_fw ${path}; then
>  +        return 0;
>  +    fi
>  +
>       tgt_name=$(kdump_iscsi_get_rec_val ${path} "node.name")
>       tgt_ipaddr=$(kdump_iscsi_get_rec_val ${path} "node.conn\[0\].address")
>   
>  -- 
>  1.9.3
>  
>  _______________________________________________
>  kexec mailing list
>  kexec(a)lists.fedoraproject.org
>  https://lists.fedoraproject.org/mailman/listinfo/kexec

[WANG Chao]

On 11/03/14 at 10:33am, Dave Young wrote:
>  Hi,
>  
>  Well done, thanks Chao.
>  
>  Vivek/Bao/Minfei, could you review the patch as well?
>  
>  On 10/28/14 at 06:43pm, WANG Chao wrote:
>  > Currently we retrieve all the necessary information of an iSCSI session
>  > and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  > because dracut has already had its own infrastructure to do bring up any
>  > iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  > 
>  > This patch will determine if an iscsi session is established by firmware
>  > configured iscsi. If it does, then we don't pass anything down to 2nd
>  > kernel. dracut will handle it.
>  > 
>  > I read some iscsi code and I also talked with storage QE. It's confirmed
>  > that software iscsi will use "tcp" transport class while hardware iscsi
>  > will use like "be2iscsi" "bnx2i" transport class. It's ok to
> use
>  > the transport name of an iface of a scsi session to determine hardware
>  > iscsi or not. Additionally using "iscsi_firmware"
> "rd.iscsi.firmware" to
>  > determine if the system is iscsi boot, because dracut can only handle
>  > this case.
>  > 
>  > Along this patch, I add the infrastructure of determine iBFT, but it's
>  > not complete yet. Mark the part as "FIXME". Once customer is asking or
>  > maybe I get hold of iBFT box, I can start working on it.
>  
>  For non-root hwiscsi there's no kernel cmdline, you can put it in todo list
>  We can support root device firstly and see if it works well. 
Yes, that's TODO and nice-to-have. My concern is that in 2nd kernel, not all
hardware/driver works as fine as it does in 1st kernel. I've heard a few
cases that certain driver won't work at 2nd kernel. I found one bug
myself:

https://bugzilla.redhat.com/show_bug.cgi?id=1158014
[kdump] [be2iscsi]: IP: [<ffffffffa00cf700>]
hwi_get_async_handle.isra.23.constprop.39+0x90/0x1d0 [be2iscsi]

So I think it can be safer to use iscsi_tcp (software driver, no
offload) instead of using hardware associated offload driver. What do
you think?

[..]

>  > +# kernel cmdline must contain these either of the following
> arguments
>  > +# for booting off hardware iSCSI:
>  > +# rd.iscsi.firmware[=1], iscsi_firmware
>  > +kdump_is_iscsi_hw_cmdline() {
>  > +    local cmdline
>  > +
>  > +    cmdline="$(cat /proc/cmdline)"
>  > +
>  > +    for arg in $cmdline; do
>  > +        case "$arg" in
>  > +            "rd.iscsi.firmware" |\
>  > +            "rd.iscsi.firmware=1" |\
>  > +            "iscsi_firmware")
>  
>  dracut.cmdline manpage shows below:
>         iscsi_firmware
>             rd.iscsi.firmware=0
>  
>  So looks like iscsi_firmware == rd.iscsi.firmware=0 
That's a documentation issue in dracut. iscsi_firmware actually is
deprecated by rd.iscsi.firmware=1. You can send a patch to Harald to fix
it.

Thanks
WANG Chao

[Dave Young]

On 11/03/14 at 11:09am, WANG Chao wrote:
>  On 11/03/14 at 10:33am, Dave Young wrote:
>  > Hi,
>  > 
>  > Well done, thanks Chao.
>  > 
>  > Vivek/Bao/Minfei, could you review the patch as well?
>  > 
>  > On 10/28/14 at 06:43pm, WANG Chao wrote:
>  > > Currently we retrieve all the necessary information of an iSCSI session
>  > > and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  > > because dracut has already had its own infrastructure to do bring up any
>  > > iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  > > 
>  > > This patch will determine if an iscsi session is established by firmware
>  > > configured iscsi. If it does, then we don't pass anything down to 2nd
>  > > kernel. dracut will handle it.
>  > > 
>  > > I read some iscsi code and I also talked with storage QE. It's confirmed
>  > > that software iscsi will use "tcp" transport class while hardware
> iscsi
>  > > will use like "be2iscsi" "bnx2i" transport class. It's
> ok to use
>  > > the transport name of an iface of a scsi session to determine hardware
>  > > iscsi or not. Additionally using "iscsi_firmware"
> "rd.iscsi.firmware" to
>  > > determine if the system is iscsi boot, because dracut can only handle
>  > > this case.
>  > > 
>  > > Along this patch, I add the infrastructure of determine iBFT, but it's
>  > > not complete yet. Mark the part as "FIXME". Once customer is asking
> or
>  > > maybe I get hold of iBFT box, I can start working on it.
>  > 
>  > For non-root hwiscsi there's no kernel cmdline, you can put it in todo list
>  > We can support root device firstly and see if it works well.
>  
>  Yes, that's TODO and nice-to-have. My concern is that in 2nd kernel, not all
>  hardware/driver works as fine as it does in 1st kernel. I've heard a few
>  cases that certain driver won't work at 2nd kernel. I found one bug
>  myself:
>  
>  https://bugzilla.redhat.com/show_bug.cgi?id=1158014
>  [kdump] [be2iscsi]: IP: [<ffffffffa00cf700>]
> hwi_get_async_handle.isra.23.constprop.39+0x90/0x1d0 [be2iscsi]
>  
>  So I think it can be safer to use iscsi_tcp (software driver, no
>  offload) instead of using hardware associated offload driver. What do
>  you think? 
I think we can open bugs in case such issues. Just what we did for drm module
you know nvidia card does not work well.

Agree that we can enable root device support as the first step. BTW can you
update documentation in the patch? 

>  
>  [..]
>  
>  > > +# kernel cmdline must contain these either of the following arguments
>  > > +# for booting off hardware iSCSI:
>  > > +# rd.iscsi.firmware[=1], iscsi_firmware
>  > > +kdump_is_iscsi_hw_cmdline() {
>  > > +    local cmdline
>  > > +
>  > > +    cmdline="$(cat /proc/cmdline)"
>  > > +
>  > > +    for arg in $cmdline; do
>  > > +        case "$arg" in
>  > > +            "rd.iscsi.firmware" |\
>  > > +            "rd.iscsi.firmware=1" |\
>  > > +            "iscsi_firmware")
>  > 
>  > dracut.cmdline manpage shows below:
>  >        iscsi_firmware
>  >            rd.iscsi.firmware=0
>  > 
>  > So looks like iscsi_firmware == rd.iscsi.firmware=0
>  
>  That's a documentation issue in dracut. iscsi_firmware actually is
>  deprecated by rd.iscsi.firmware=1. You can send a patch to Harald to fix
>  it. 
Sure, here is a patch to fix the manpage:
http://rss.gmane.org/messages/excerpts/gmane.linux.kernel.initramfs

[WANG Chao]

On 11/03/14 at 01:17pm, Dave Young wrote:
>  On 11/03/14 at 11:09am, WANG Chao wrote:
>  > On 11/03/14 at 10:33am, Dave Young wrote:
>  > > Hi,
>  > > 
>  > > Well done, thanks Chao.
>  > > 
>  > > Vivek/Bao/Minfei, could you review the patch as well?
>  > > 
>  > > On 10/28/14 at 06:43pm, WANG Chao wrote:
>  > > > Currently we retrieve all the necessary information of an iSCSI session
>  > > > and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  > > > because dracut has already had its own infrastructure to do bring up any
>  > > > iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  > > > 
>  > > > This patch will determine if an iscsi session is established by firmware
>  > > > configured iscsi. If it does, then we don't pass anything down to 2nd
>  > > > kernel. dracut will handle it.
>  > > > 
>  > > > I read some iscsi code and I also talked with storage QE. It's
> confirmed
>  > > > that software iscsi will use "tcp" transport class while
> hardware iscsi
>  > > > will use like "be2iscsi" "bnx2i" transport class.
> It's ok to use
>  > > > the transport name of an iface of a scsi session to determine hardware
>  > > > iscsi or not. Additionally using "iscsi_firmware"
> "rd.iscsi.firmware" to
>  > > > determine if the system is iscsi boot, because dracut can only handle
>  > > > this case.
>  > > > 
>  > > > Along this patch, I add the infrastructure of determine iBFT, but
> it's
>  > > > not complete yet. Mark the part as "FIXME". Once customer is
> asking or
>  > > > maybe I get hold of iBFT box, I can start working on it.
>  > > 
>  > > For non-root hwiscsi there's no kernel cmdline, you can put it in todo
> list
>  > > We can support root device firstly and see if it works well.
>  > 
>  > Yes, that's TODO and nice-to-have. My concern is that in 2nd kernel, not all
>  > hardware/driver works as fine as it does in 1st kernel. I've heard a few
>  > cases that certain driver won't work at 2nd kernel. I found one bug
>  > myself:
>  > 
>  > https://bugzilla.redhat.com/show_bug.cgi?id=1158014
>  > [kdump] [be2iscsi]: IP: [<ffffffffa00cf700>]
> hwi_get_async_handle.isra.23.constprop.39+0x90/0x1d0 [be2iscsi]
>  > 
>  > So I think it can be safer to use iscsi_tcp (software driver, no
>  > offload) instead of using hardware associated offload driver. What do
>  > you think?
>  
>  I think we can open bugs in case such issues. Just what we did for drm module
>  you know nvidia card does not work well.
>  
>  Agree that we can enable root device support as the first step. BTW can you
>  update documentation in the patch?  
What documentation?

[Dave Young]

>  > Agree that we can enable root device support as the first step.
> BTW can you
>  > update documentation in the patch? 
>  
>  What documentation? 
Probably in kdump howto in our tree about the support status and todos.
A standalone patch is also good because it seems we did not add the support list
in Fedora tree.

[WANG Chao]

On 11/03/14 at 03:46pm, Dave Young wrote:
>  > > Agree that we can enable root device support as the first
> step. BTW can you
>  > > update documentation in the patch? 
>  > 
>  > What documentation?
>  
>  Probably in kdump howto in our tree about the support status and todos.
>  A standalone patch is also good because it seems we did not add the support list
>  in Fedora tree. 
I'd not add this until QA approves this one.

[Dave Young]

On 11/03/14 at 03:52pm, WANG Chao wrote:
>  On 11/03/14 at 03:46pm, Dave Young wrote:
>  > > > Agree that we can enable root device support as the first step. BTW can
> you
>  > > > update documentation in the patch? 
>  > > 
>  > > What documentation?
>  > 
>  > Probably in kdump howto in our tree about the support status and todos.
>  > A standalone patch is also good because it seems we did not add the support list
>  > in Fedora tree.
>  
>  I'd not add this until QA approves this one. 
Sure, that's fine. Just keep it in mind.

[Minfei Huang]

I lost the original mail, So reply it following this mail.
Chao's patch look good for me.

On 11/03/14 at 10:33am, Dave Young wrote:
>  Hi,
>  
>  Well done, thanks Chao.
>  
>  Vivek/Bao/Minfei, could you review the patch as well?
>  
>  On 10/28/14 at 06:43pm, WANG Chao wrote:
>  > Currently we retrieve all the necessary information of an iSCSI session
>  > and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  > because dracut has already had its own infrastructure to do bring up any
>  > iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  > 
>  >  kdump_setup_iscsi_device() {
>  >      local path=$1
>  >      local tgt_name; local tgt_ipaddr;
>  > @@ -434,6 +505,10 @@ kdump_setup_iscsi_device() {
>  >          return 1
>  >      fi
>  >  
>  > +    if kdump_is_iscsi_fw ${path}; then
>  > +        return 0;
>  > +    fi
>  > +
>  >      tgt_name=$(kdump_iscsi_get_rec_val ${path} "node.name")
>  >      tgt_ipaddr=$(kdump_iscsi_get_rec_val ${path}
> "node.conn\[0\].address")
>  >  
>  > -- 
>  > 1.9.3
>  > 
>  > _______________________________________________
>  > kexec mailing list
>  > kexec(a)lists.fedoraproject.org
>  > https://lists.fedoraproject.org/mailman/listinfo/kexec

[Baoquan He]

On 10/28/14 at 06:43pm, WANG Chao wrote:
>  Currently we retrieve all the necessary information of an iSCSI
> session
>  and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  because dracut has already had its own infrastructure to do bring up any
>  iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  
>  This patch will determine if an iscsi session is established by firmware
>  configured iscsi. If it does, then we don't pass anything down to 2nd
>  kernel. dracut will handle it.
>  
>  I read some iscsi code and I also talked with storage QE. It's confirmed
>  that software iscsi will use "tcp" transport class while hardware iscsi
>  will use like "be2iscsi" "bnx2i" transport class. It's ok to use
>  the transport name of an iface of a scsi session to determine hardware
>  iscsi or not. Additionally using "iscsi_firmware" "rd.iscsi.firmware"
> to
>  determine if the system is iscsi boot, because dracut can only handle
>  this case.
>  
>  Along this patch, I add the infrastructure of determine iBFT, but it's
>  not complete yet. Mark the part as "FIXME". Once customer is asking or
>  maybe I get hold of iBFT box, I can start working on it.
>  
>  Signed-off-by: WANG Chao <chaowang(a)redhat.com&gt;
>  ---
>   dracut-module-setup.sh | 77 +++++++++++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 76 insertions(+), 1 deletion(-)
>  
>  diff --git a/dracut-module-setup.sh b/dracut-module-setup.sh
>  index dcebc47..9616866 100755
>  --- a/dracut-module-setup.sh
>  +++ b/dracut-module-setup.sh
>  @@ -411,7 +411,78 @@ kdump_get_iscsi_initiator() {
>       return 1
>   }
>   
>  -# No ibft handling yet.
>  +# Determine hardware iscsi by an iface's transport class.
>  +# If transport class is tcp, it's software iscsi.
>  +# If it's be2iscsi, bnx2i, cxgb3i, cxgb4i, it's hardware iscsi.
>  +kdump_is_iscsi_hw_session() {
>  +    if [ "$(kdump_iscsi_get_rec_val $1 "iface.transport_name")" =
> tcp ]; then
>  +        return 1
>  +    fi
>  +
>  +    return 0
>  +}
>  +
>  +# FIXME: determine iBFT
>  +kdump_is_iscsi_ibft_session() {
>  +    return 1
>  +}
>  +
>  +# kernel cmdline must contain these either of the following arguments
>  +# for booting off hardware iSCSI:
>  +# rd.iscsi.firmware[=1], iscsi_firmware
>  +kdump_is_iscsi_hw_cmdline() { 
How about name it as kdump_check_iscsi_hw_cmdline?  Have the same
concern about above 2 functions.

>  +    local cmdline
>  +
>  +    cmdline="$(cat /proc/cmdline)"
>  +
>  +    for arg in $cmdline; do
>  +        case "$arg" in
>  +            "rd.iscsi.firmware" |\
>  +            "rd.iscsi.firmware=1" |\
>  +            "iscsi_firmware")
>  +                return 0
>  +                ;;
>  +        esac
>  +    done
>  +
>  +    return 1
>  +}
>  +
>  +# kernel cmdline must contain these either of the following arguments
>  +# for booting off iBFT:
>  +# rd.iscsi.ibft[=1], ip=ibft
>  +kdump_is_iscsi_ibft_cmdline() { 
ditto.

>  +    local cmdline
>  +
>  +    cmdline="$(cat /proc/cmdline)"
>  +
>  +    for arg in $cmdline; do
>  +        case "$arg" in
>  +            "rd.iscsi.ibft" |\
>  +            "rd.iscsi.ibft=1" |\
>  +            "ip=ibft")
>  +                return 0
>  +                ;;
>  +        esac
>  +    done
>  +
>  +    return 1
>  +}
>  +
>  +# Sanity check if an iscsi session is established by iBFT or firmware
>  +kdump_is_iscsi_fw() {
>  +
>  +    if kdump_is_iscsi_hw_cmdline && kdump_is_iscsi_hw_session $1; then
>  +        return 0
>  +    fi
>  +
>  +    if kdump_is_iscsi_ibft_cmdline && kdump_is_iscsi_ibft_session $1; then
>  +        return 0
>  +    fi
>  +
>  +    return 1
>  +}
>  +
>   kdump_setup_iscsi_device() {
>       local path=$1
>       local tgt_name; local tgt_ipaddr;
>  @@ -434,6 +505,10 @@ kdump_setup_iscsi_device() {
>           return 1
>       fi
>    
Could you add the reason why we needn't take care of hw iscsi and ibft
cases in kdump? This can make code more readable.


>  +    if kdump_is_iscsi_fw ${path}; then
>  +        return 0;
>  +    fi
>  +
>       tgt_name=$(kdump_iscsi_get_rec_val ${path} "node.name")
>       tgt_ipaddr=$(kdump_iscsi_get_rec_val ${path} "node.conn\[0\].address")
Besides above several small concerns, this patch looks good to me.

Thanks, Chao.


>   
>  -- 
>  1.9.3
>  
>  _______________________________________________
>  kexec mailing list
>  kexec(a)lists.fedoraproject.org
>  https://lists.fedoraproject.org/mailman/listinfo/kexec

[Vivek Goyal]

On Tue, Oct 28, 2014 at 06:43:39PM +0800, WANG Chao wrote:
>  Currently we retrieve all the necessary information of an iSCSI
> session
>  and pass them to 2nd kernel. This is wrong way to go for hardware iscsi,
>  because dracut has already had its own infrastructure to do bring up any
>  iscsi device configured in firmware of an HBA (Host Bus Bridge).
>  
>  This patch will determine if an iscsi session is established by firmware
>  configured iscsi. If it does, then we don't pass anything down to 2nd
>  kernel. dracut will handle it.
>  
>  I read some iscsi code and I also talked with storage QE. It's confirmed
>  that software iscsi will use "tcp" transport class while hardware iscsi
>  will use like "be2iscsi" "bnx2i" transport class. It's ok to use
>  the transport name of an iface of a scsi session to determine hardware
>  iscsi or not. Additionally using "iscsi_firmware" "rd.iscsi.firmware"
> to
>  determine if the system is iscsi boot, because dracut can only handle
>  this case.
>  
>  Along this patch, I add the infrastructure of determine iBFT, but it's
>  not complete yet. Mark the part as "FIXME". Once customer is asking or
>  maybe I get hold of iBFT box, I can start working on it.
>  
>  Signed-off-by: WANG Chao <chaowang(a)redhat.com&gt; 
Chao,

I am wondering that why do iscsi module of dracut should not do the work
of setting up iscsi devices.

Our general kdump theme so far has been (at least in terms of block
devices) that we pass in top level fs information to dracut and then
dracut automatically figures out the device stack beneath it and pull
in all the required dependencies.

So my general impression is we should do this work in two steps.

- First move all/most of the iscsi logic from dracut-module-setup.sh
  to dracut iscsi module.

- And then modify dracut iscsi module to do also support hardware and
  ibft iscsi connections.

What do you think?

Thanks
Vivek

>  ---
>   dracut-module-setup.sh | 77 +++++++++++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 76 insertions(+), 1 deletion(-)
>  
>  diff --git a/dracut-module-setup.sh b/dracut-module-setup.sh
>  index dcebc47..9616866 100755
>  --- a/dracut-module-setup.sh
>  +++ b/dracut-module-setup.sh
>  @@ -411,7 +411,78 @@ kdump_get_iscsi_initiator() {
>       return 1
>   }
>   
>  -# No ibft handling yet.
>  +# Determine hardware iscsi by an iface's transport class.
>  +# If transport class is tcp, it's software iscsi.
>  +# If it's be2iscsi, bnx2i, cxgb3i, cxgb4i, it's hardware iscsi.
>  +kdump_is_iscsi_hw_session() {
>  +    if [ "$(kdump_iscsi_get_rec_val $1 "iface.transport_name")" =
> tcp ]; then
>  +        return 1
>  +    fi
>  +
>  +    return 0
>  +}
>  +
>  +# FIXME: determine iBFT
>  +kdump_is_iscsi_ibft_session() {
>  +    return 1
>  +}
>  +
>  +# kernel cmdline must contain these either of the following arguments
>  +# for booting off hardware iSCSI:
>  +# rd.iscsi.firmware[=1], iscsi_firmware
>  +kdump_is_iscsi_hw_cmdline() {
>  +    local cmdline
>  +
>  +    cmdline="$(cat /proc/cmdline)"
>  +
>  +    for arg in $cmdline; do
>  +        case "$arg" in
>  +            "rd.iscsi.firmware" |\
>  +            "rd.iscsi.firmware=1" |\
>  +            "iscsi_firmware")
>  +                return 0
>  +                ;;
>  +        esac
>  +    done
>  +
>  +    return 1
>  +}
>  +
>  +# kernel cmdline must contain these either of the following arguments
>  +# for booting off iBFT:
>  +# rd.iscsi.ibft[=1], ip=ibft
>  +kdump_is_iscsi_ibft_cmdline() {
>  +    local cmdline
>  +
>  +    cmdline="$(cat /proc/cmdline)"
>  +
>  +    for arg in $cmdline; do
>  +        case "$arg" in
>  +            "rd.iscsi.ibft" |\
>  +            "rd.iscsi.ibft=1" |\
>  +            "ip=ibft")
>  +                return 0
>  +                ;;
>  +        esac
>  +    done
>  +
>  +    return 1
>  +}
>  +
>  +# Sanity check if an iscsi session is established by iBFT or firmware
>  +kdump_is_iscsi_fw() {
>  +
>  +    if kdump_is_iscsi_hw_cmdline && kdump_is_iscsi_hw_session $1; then
>  +        return 0
>  +    fi
>  +
>  +    if kdump_is_iscsi_ibft_cmdline && kdump_is_iscsi_ibft_session $1; then
>  +        return 0
>  +    fi
>  +
>  +    return 1
>  +}
>  +
>   kdump_setup_iscsi_device() {
>       local path=$1
>       local tgt_name; local tgt_ipaddr;
>  @@ -434,6 +505,10 @@ kdump_setup_iscsi_device() {
>           return 1
>       fi
>   
>  +    if kdump_is_iscsi_fw ${path}; then
>  +        return 0;
>  +    fi
>  +
>       tgt_name=$(kdump_iscsi_get_rec_val ${path} "node.name")
>       tgt_ipaddr=$(kdump_iscsi_get_rec_val ${path} "node.conn\[0\].address")
>   
>  -- 
>  1.9.3
>  
>  _______________________________________________
>  kexec mailing list
>  kexec(a)lists.fedoraproject.org
>  https://lists.fedoraproject.org/mailman/listinfo/kexec