GitHub/LineageOS/android_kernel_samsung_universal7580.git
14 years ago[SCSI] lpfc 8.3.5: Update the lpfc driver version to 8.3.5
James Smart [Fri, 2 Oct 2009 19:17:08 +0000 (15:17 -0400)]
[SCSI] lpfc 8.3.5: Update the lpfc driver version to 8.3.5

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] lpfc 8.3.5: fix fcp command polling, add FIP mode, performance optimisations...
James Smart [Fri, 2 Oct 2009 19:17:02 +0000 (15:17 -0400)]
[SCSI] lpfc 8.3.5: fix fcp command polling, add FIP mode, performance optimisations and devloss timout fixes

This patch includes the following changes:
- Fixed Panic/Hang when using polling mode for fcp commands
- Added support for Read_rev mbox bits indicating FIP mode of HBA
- Optimize performance of slow-path handling of els responses
- Add code to cleanup orphaned unsolicited receive sequences
- Fixed Devloss timeout when multiple initiators are in same zone

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] lpfc 8.3.5: Add AER support
James Smart [Fri, 2 Oct 2009 19:16:56 +0000 (15:16 -0400)]
[SCSI] lpfc 8.3.5: Add AER support

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] lpfc 8.3.5: fix sysfs parameters, vport creation and other bugs and update...
James Smart [Fri, 2 Oct 2009 19:16:51 +0000 (15:16 -0400)]
[SCSI] lpfc 8.3.5: fix sysfs parameters, vport creation and other bugs and update logging

This patch include the following fixes and changes:
- Fix crash when "error" is echoed to board_mode sysfs parameter
- Fix FCoE Parameter parsing in regions 23
- Fix driver crash when creating vport with large number of targets on SLI4
- Fix bug with npiv message being logged when it is not supported by the adapter
- Fix a potential dereferencing mailbox structure after free bug
- Fix firmware crash after vport create with high target count
- Error out requests to set board_mode to warm restart via sysfs on SLI4 HBAs
- Fix Block guard logging
- Fix a memory corruption issue during GID_FT IO prep
- Fix crash while processing unsolicited FC frames
- Fix failed to allocate XRI message is not a critical failure
- Update and fix formatting in some log messages
- Fix missing new line characters in log messages
- Removed the use of the locally defined FC transport layer related macros
- Check the rsplen in lpfc_handle_fcp_err function before using rsplen

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] lpfc 8.3.5: fix VPI registration, error clean up and add support for vlink...
James Smart [Fri, 2 Oct 2009 19:16:45 +0000 (15:16 -0400)]
[SCSI] lpfc 8.3.5: fix VPI registration, error clean up and add support for vlink events

This patch includes the following fixes and new features:
- Fix mask size for CT field in WQE
- Fix VPI base not used when unregistering VPI on port 1.
- Fix UNREG_VPI mailbox command to unreg the correct VPI
- Fixed Check for aborted els command
- Fix error when trying to load driver with wrong firmware on FCoE HBA.
- Fix bug with probe_one routines not putting the Scsi_Host back upon error
- Add support for Clear Virtual Link Async Events
- Add support for unsolicited CT exchange sequence abort
- Add 0x0714 OCeXXXXX PCI ID

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] lpfc 8.3.5: fix reset path, ELS ordering and discovery issues
James Smart [Fri, 2 Oct 2009 19:16:39 +0000 (15:16 -0400)]
[SCSI] lpfc 8.3.5: fix reset path, ELS ordering and discovery issues

This patch includes the following fixes:
- Fixed panic during HBA reset.
- Fixed FCoE event tag passed in resume_rpi.
- Fix out of order ELS commands
- Fixed discovery issues found during VLAN testing.
- Fix UNREG_VPI failure on extended link pull
- Fixed crash while processing unsolicited FC frames.
- Clear retry count in the delayed ELS handler
- Fixed discovery failure during quick link bounce.

Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libiscsi: hook into ramp up/down handling
Mike Christie [Wed, 11 Nov 2009 22:34:36 +0000 (16:34 -0600)]
[SCSI] libiscsi: hook into ramp up/down handling

It is rare to get a queue full with iscsi, because targets seem to
just reduce the iscsi cmd window. However, there is at least
one iscsi target that will throw a queue full when overloaded.
This hooks the iscsi code in to the ramp up/down code, so we
can handle it.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] iser: set tgt and lu reset timeout
Mike Christie [Wed, 11 Nov 2009 22:34:35 +0000 (16:34 -0600)]
[SCSI] iser: set tgt and lu reset timeout

When iser enabled lu reset support it did not set the
bit to allow userspace to get/set the timeout. This
sets the tgt and lu reset timeout bits.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] iscsi class: modify handling of replacement timeout
Mike Christie [Wed, 11 Nov 2009 22:34:34 +0000 (16:34 -0600)]
[SCSI] iscsi class: modify handling of replacement timeout

This patch modifies the replacement/recovery_timeout so it works
more like the fc fast io fail tmo.

If userspace tries to set the replacement/recovery_timeout to less than
zero, we will turn off the forced recovery cleanup.

If userspace sets the value to 0 then we will force the recovery
cleanup immediately.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libiscsi: add warm target reset tmf support
Mike Christie [Wed, 11 Nov 2009 22:34:33 +0000 (16:34 -0600)]
[SCSI] libiscsi: add warm target reset tmf support

This implements warm target reset tmf support for
the scsi-ml target reset callback. Previously we would
just drop the session in that callback. This patch will
now try a target reset and if that fails drop the session.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libiscsi: Check TMF state before sending PDU
Mike Christie [Wed, 11 Nov 2009 22:34:32 +0000 (16:34 -0600)]
[SCSI] libiscsi: Check TMF state before sending PDU

Patch and mail from both MikeC and HannesR:

Before we're trying to send a PDU we have to check whether a TMF
is active. If so and if the PDU will be affected by the TMF
we should allow only Data-out PDUs to be sent.

If fast_abort is set, no Data-out PDUs will be sent while
a LUN reset is being processed for a affected LUN.

fast_abort is now ingored during a ABORT TASK tmf. We will not
send any Data-outs for a task if the task is being aborted.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libiscsi: fix login/text checks in pdu injection code
Mike Christie [Wed, 11 Nov 2009 22:34:31 +0000 (16:34 -0600)]
[SCSI] libiscsi: fix login/text checks in pdu injection code

For some reason we used to check for the the immediate bit
set and the opcocde in many places instead of just masking
the opcode. In the passthrough code this is a problem
because userspace may or may not have set the immediate bit
and it does not have to. This fixes up the opcode checks
in the passthrough code, so we mask off the opcode then
check against the iscsi proto definition like is done in
other places.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] bnx2i: use common iscsi suspend queue
Mike Christie [Wed, 11 Nov 2009 22:34:30 +0000 (16:34 -0600)]
[SCSI] bnx2i: use common iscsi suspend queue

This just has bnx2i use the iscsi_suspend_queue helper.

The suspend works as follows:

When ep_poll has succeeed iscsid will call conn_bind, the LLD will
then call iscsi_conn_bind which will clear the suspend bit.
When ep_disconnect is called (or if there is a conn error) we set
the suspend bit. For the ep_disconnect case I  added a helper
in the previous kernel that will take the session lock to make sure
iscsi_queuecommand/xmit_task is not running and it will set
the suspend bit.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Acked-by: Anil Veerabhadrappa <anilgv@broadcom.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_devinfo: update Hitachi entries (v2)
Takahiro Yasui [Tue, 10 Nov 2009 21:22:19 +0000 (16:22 -0500)]
[SCSI] scsi_devinfo: update Hitachi entries (v2)

Four models, OPEN-/DF400/DF500/DISK-SUBSYSTEM, can handle REPORT_LUN,
and the BLIST_REPORTLUN2 flag needs to be set. And DF600 doesn't require
any flags because it returns ANSI 03h (SPC).

Signed-off-by: Takahiro Yasui <tyasui@redhat.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] add scsi target reset support to scsi ioctl
Mike Christie [Thu, 5 Nov 2009 17:37:28 +0000 (11:37 -0600)]
[SCSI] add scsi target reset support to scsi ioctl

The scsi ioctl code path was missing scsi target reset
support. This patch just adds it.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fc class: fail fast bsg requests
Mike Christie [Thu, 5 Nov 2009 17:18:09 +0000 (11:18 -0600)]
[SCSI] fc class: fail fast bsg requests

If the port state is blocked and the fast io fail tmo has
fired then this patch will fail bsg requests immediately.
This is needed if userspace is sending IOs to test the transport
like with fcping, so it will not have to wait for the dev loss tmo.
With this patch he bsg req fast io fail code behaves like the normal
and sg io/passthrough fast io fail.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Acked-By: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] pm8001: Fixes for tag alloc, error goto and code cleanup
jack_wang [Thu, 5 Nov 2009 14:33:51 +0000 (22:33 +0800)]
[SCSI] pm8001: Fixes for tag alloc, error goto and code cleanup

Allocate right size for bitmap tag,fix error goto and cleanup print
message and undocable commemts. patch attached.

Signed-off-by: Lindar Liu <lindar_liu@usish.com>
Signed-off-by: Jack Wang <jack_wang@usish.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] pm8001: enhance IOMB process modules
jack_wang [Thu, 5 Nov 2009 14:33:35 +0000 (22:33 +0800)]
[SCSI] pm8001: enhance IOMB process modules

We set interupt cascading count of outbound queue to get better
performance, correct some unnecessary return values and some noisy
print messages.  patch attached.

Signed-off-by: Jack Wang <jack_wang@usish.com>
Signed-off-by: Lindar Liu <lindar_liu@usish.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] pm8001: add reinitialize SPC parameters before phy start
jack_wang [Thu, 5 Nov 2009 14:32:31 +0000 (22:32 +0800)]
[SCSI] pm8001: add reinitialize SPC parameters before phy start

Signed-off-by: Jack Wang <jack_wang@usish.com>
Signed-off-by: Lindar Liu <lindar_liu@usish.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_lib_dma: fix bug with dma maps on nested scsi objects
James Bottomley [Thu, 5 Nov 2009 19:33:12 +0000 (13:33 -0600)]
[SCSI] scsi_lib_dma: fix bug with dma maps on nested scsi objects

Some of our virtual SCSI hosts don't have a proper bus parent at the
top, which can be a problem for doing DMA on them

This patch makes the host device cache a pointer to the physical bus
device and provides an extra API for setting it (the normal API picks
it up from the parent).  This patch also modifies the qla2xxx and lpfc
vport logic to use the new DMA host setting API.

Acked-By: James Smart <james.smart@emulex.com>
Cc: Stable Tree <stable@kernel.org>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_dh: create sysfs file, dh_state for all SCSI disk devices
Chandra Seetharaman [Fri, 11 Sep 2009 17:20:35 +0000 (10:20 -0700)]
[SCSI] scsi_dh: create sysfs file, dh_state for all SCSI disk devices

Create the sysfs file, dh_state even if the new SCSI device is not
in the any of the device handler's internal lists.

Signed-Off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Acked-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] qlogicpti: add missing parentheses
Roel Kluin [Tue, 3 Nov 2009 23:38:44 +0000 (00:38 +0100)]
[SCSI] qlogicpti: add missing parentheses

`+' has a higher precedence than `?' so the condition always
evaluates to true and this is preprocessed to `7*((ql) - 1)'

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc fcoe: increase ELS and CT timeouts
Joe Eykholt [Tue, 3 Nov 2009 19:50:21 +0000 (11:50 -0800)]
[SCSI] libfc fcoe: increase ELS and CT timeouts

The FC-LS spec. says ELS timeouts should be 2 x R_A_TOV.
The FC-GS spec. says CT timeouts should be 3 x R_A_TOV.

We've been using E_D_TOV for both of those.

Change for all ELS and CT requests except FLOGI, which we
leave at 2 seconds (using E_D_TOV).  One could argue that
R_A_TOV is locally determined until after FLOGI succeeds.

This does change FLOGI for vports which becomes FDISC.
This does not change the REC/SRR timeout which is 2 seconds.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: Do not pad FIP keep-alive to full frame size
Yi Zou [Tue, 3 Nov 2009 19:50:16 +0000 (11:50 -0800)]
[SCSI] libfcoe: Do not pad FIP keep-alive to full frame size

According to the FC-BB-5 Rev2.0, 7.8.6.2, we should not pad FIP keep-alive
frames.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: fix an issue of pending exch/es after i/f destroyed or rmmod fcoe
Vasu Dev [Tue, 3 Nov 2009 19:50:10 +0000 (11:50 -0800)]
[SCSI] libfc: fix an issue of pending exch/es after i/f destroyed or rmmod fcoe

All exches must be freed before its EM mempool destroyed in this
case but currently some exches could be still pending in their
scheduled delayed work after EM mempool is destroyed causing
this issue discussed and reported in this latest email thread:-

 http://www.open-fcoe.org/pipermail/devel/2009-October/004788.html

This patch fixes this issue by adding dedicated work queue thread
fc_exch_workqueue for exch delayed work and then flush this work
queue before destroying EM mempool.

The cancel_delayed_work_sync cannot be called during final
fc_exch_reset due to lport and exch locking ordering, so removes
related comment block not relevant any more with this patch.

Reported-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc, fcoe: fixes for highmem skb linearize panics
Chris Leech [Tue, 3 Nov 2009 19:50:05 +0000 (11:50 -0800)]
[SCSI] libfc, fcoe: fixes for highmem skb linearize panics

There are cases outside of our control that may result in a transmit
skb being linearized in dev_queue_xmit.  There are a couple of bugs
in libfc/fcoe that can result in a panic at that point.  This patch
contains two fixes to prevent those panics.

1) use fast cloning instead of shared skbs with dev_queue_xmit

dev_queue_xmit doen't want shared skbuffs being passed in, and
__skb_linearize will BUG if the skb is shared.  FCoE is holding an extra
reference around the call to dev_queue_xmit, so that when it returns an
error code indicating the frame has been dropped it can maintain it's
own backlog and retransmit.  Switch to using fast skb cloning for this
instead.

2) don't append compound pages as > PAGE_SIZE skb fragments

fc_fcp_send_data will append pages from a scatterlist to the nr_frags[]
if the netdev supports it.  But, it's using > PAGE_SIZE compound pages
as a single skb_frag.  In the highmem linearize case that page will be
passed to kmap_atomic to get a mapping to copy out of, but
kmap_atomic will only allow access to the first PAGE_SIZE part.
The memcpy will keep going and cause a page fault once is crosses the
first boundary.

If fc_fcp_send_data uses linear buffers from the start, it calls
kmap_atomic one PAGE_SIZE at a time.  That same logic needs to be
applied when setting up skb_frags.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: Fix using VLAN ID in creating lport's WWWN/WWPN
Yi Zou [Tue, 3 Nov 2009 19:49:59 +0000 (11:49 -0800)]
[SCSI] fcoe: Fix using VLAN ID in creating lport's WWWN/WWPN

If the underlying netdev is a VLAN device, make sure the VLAN ID is integrated
into the WWNN/WWPN name generation. Also added/updated the comments to reflect
this change.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: do not use DID_NO_CONNECT for pkt alloc failures.
Mike Christie [Tue, 3 Nov 2009 19:49:54 +0000 (11:49 -0800)]
[SCSI] libfc: do not use DID_NO_CONNECT for pkt alloc failures.

DID_NO_CONNECT is not a nice value to use for pkt alloc failures,
because you can probably retry and IO will become available again.
For the device reset callout, we do not want to set the scsi command
result for the above reason, and because we do not need to set
the scsi_cmd->result in this path. We and other drivers do not set it
for success for example, and we do not set it for other failure.
And scsi-ml does not send every command through this path, and it is
not expecting us to use the scsi_cmnd struct like a cmd coming thruogh
queuecommand. I think it is more for storage in case we need a cmd
struct for a tmf and to give us certain params like the LUN.

Patch was made over scsi-misc today.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: Fix setting lport's WWNN/WWPN to use san mac address
Yi Zou [Tue, 3 Nov 2009 19:49:49 +0000 (11:49 -0800)]
[SCSI] fcoe: Fix setting lport's WWNN/WWPN to use san mac address

We are still using netdev->dev_addr to generate lport's WWNN/WWPN even if the
LLD has support for NETDEV_HW_ADDR_T_SAN. Instead, we should just use the
fip->ctl_src_addr, which is the NETDEV_HW_ADDR_T_SAN if LLD supports it or it
is just the netdev->dev_addr if it does not.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: Fix getting san mac for VLAN interface
Yi Zou [Tue, 3 Nov 2009 19:49:43 +0000 (11:49 -0800)]
[SCSI] fcoe: Fix getting san mac for VLAN interface

Make sure we are get the SAN MAC address from the real netdev if the input
netdev is a VLAN device.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: Fix checking san mac address
Yi Zou [Tue, 3 Nov 2009 19:49:38 +0000 (11:49 -0800)]
[SCSI] fcoe: Fix checking san mac address

This was fixed before in 7a7f0c7 but it's introduced again recently.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fnic: enable bsg pass-thru for fcping
Joe Eykholt [Tue, 3 Nov 2009 19:49:32 +0000 (11:49 -0800)]
[SCSI] fnic: enable bsg pass-thru for fcping

Add initialization of .bsg_request in the scsi_transport_fc
template so that fcping works.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: register FC4 features with the FC switch
Joe Eykholt [Tue, 3 Nov 2009 19:49:27 +0000 (11:49 -0800)]
[SCSI] libfc: register FC4 features with the FC switch

Customers and certification tests have pointed out that we don't
show up on the switch management software as an initiator.

On some MDS switches 'show fcns database' command shows libfc
initiators as 'fcp' not 'fcp:init' like other initiators.

On others switches, I think the switch gets the features by doing a PRLI,
but it may be only certain models or under certain configurations.

Fix this by registering our FC4 features with the RFF_ID CT request
after local port login and after the RFT_ID.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fnic: Add FIP support to the fnic driver
Joe Eykholt [Tue, 3 Nov 2009 19:49:22 +0000 (11:49 -0800)]
[SCSI] fnic: Add FIP support to the fnic driver

Use libfcoe as a common FIP implementation with fcoe.
FIP or non-FIP mode is fully automatic if the firmware
supports and enables it.

Even if FIP is not supported, this uses libfcoe for the non-FIP
handling of FLOGI and its response.

Use the new lport_set_port_id() notification to capture
successful FLOGI responses and port_id resets.

While transitioning between Ethernet and FC mode, all rx and
tx FC frames are queued.  In Ethernet mode, all frames are
passed to the exchange manager to capture FLOGI responses.

Change to set data_src_addr to the ctl_src_addr whenever it
would have previously been zero because we're not logged in.
This seems safer so we'll never send a frame with a 0 source MAC.
This also eliminates a special case for sending FLOGI frames.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: fcoe: simplify receive FLOGI response
Joe Eykholt [Tue, 3 Nov 2009 19:49:16 +0000 (11:49 -0800)]
[SCSI] libfcoe: fcoe: simplify receive FLOGI response

There was a locking problem where the fip->lock was held during
the call to update_mac().  The rtnl_lock() must be taken before
the fip->lock, not the other way around.  This fixes that.

Now that fcoe_ctlr_recv_flog() is called only from the response handler
to a FLOGI request, some checking can be eliminated.  Instead of calling
update_mac(), just fill in the granted_mac address for the passed-in
frame (skb).

Eliminate the passed-in source MAC address since it is also in the skb.

Also, in fcoe, call fcoe_set_src_mac() directly instead of going thru
the fip function pointer.  This will generate less code.
Then, since fip isn't needed for LOGO response, use lport as the arg.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: add host number to lport link up/down messages.
Joe Eykholt [Tue, 3 Nov 2009 19:49:11 +0000 (11:49 -0800)]
[SCSI] libfc: add host number to lport link up/down messages.

The libfc link up/down messages don't indicate which port is changing.
The Port ID will often be 0.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: add set_fid function to libfc template
Joe Eykholt [Tue, 3 Nov 2009 19:49:05 +0000 (11:49 -0800)]
[SCSI] libfc: add set_fid function to libfc template

This is to notify the LLD when an FC_ID is assigned to the local port.

The fnic driver needs to push the assigned FC_ID to firmware.
It currently does this by intercepting the FLOGI responses, and
in order to make that code more common with FIP and NPIV, it
makes more sense to wait until the local port has completely
handled the FLOGI or FDISC response.  Also, when we fix
point-to-point FC_ID assignment, we'll need this callback as well.

Add a call to the libfc template, which is called whenever
the local port FC_ID is being assigned.  It defaults to
fc_lport_set_fid(), supplied by libfc.

As additional benefit of this function, the LLD may determine
the MAC address that caused the change by looking at the received frame.

We also print the assigned port ID as long as it isn't 0.
Setting port ID to 0 happens often in reset while retrying FLOGI,
and would be uninteresting.  This replaces the previous message
which didn't identify the host adapter instance.

patch v2 note: changed one word in a comment.  "intercepted" -> "provided".

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: fix fc_els_resp_type to correct display of CT responses
Joe Eykholt [Tue, 3 Nov 2009 19:49:00 +0000 (11:49 -0800)]
[SCSI] libfc: fix fc_els_resp_type to correct display of CT responses

Local port debug messages were using fc_els_resp_type() which showed
all CT responses as rejects.

Handle CT responses correctly based by inspecting fh_type.

I decided not to rename the function to keep the patch smaller.
We could call it just fc_resp_type() or fc_elsct_resp_type().

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: fix symbolic name registrations smashing skb data
Joe Eykholt [Tue, 3 Nov 2009 19:48:55 +0000 (11:48 -0800)]
[SCSI] libfc: fix symbolic name registrations smashing skb data

The strncpy for RSPN_ID and RSNN_NN requests was padding
past the allocated frame size.

Get the string length before filling in the ct header.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: fix RNN_ID smashing skb payload
Joe Eykholt [Tue, 3 Nov 2009 19:48:50 +0000 (11:48 -0800)]
[SCSI] libfc: fix RNN_ID smashing skb payload

The code that filled in the name server RNN_ID (register node name)
request had somehow gotten a line in it from the RFT_ID code
which copies 32 bytes of data over the relatively short payload.
This caused some corruption and hangs.

Simply deleted the extraneous line.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: add check to fail gracefully in bonding mode
john fastabend [Tue, 3 Nov 2009 19:48:44 +0000 (11:48 -0800)]
[SCSI] fcoe: add check to fail gracefully in bonding mode

This patch adds a check to fail gracefully when the netdevice
is bonded.  Previously, the error was detected but the stack
would continue to load.  This resulted in a partially enabled
fcoe intance and errors when the fcoe instance was destroy.

Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: remove extra function decalrations
Yi Zou [Tue, 3 Nov 2009 19:48:39 +0000 (11:48 -0800)]
[SCSI] fcoe: remove extra function decalrations

Remove the two extra function decalartions in fcoe.c.

Signed-off-by: Yi Zou <yi.zou@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: don't send ELS in FIP mode if no FCF selected
Joe Eykholt [Tue, 3 Nov 2009 19:48:32 +0000 (11:48 -0800)]
[SCSI] libfcoe: don't send ELS in FIP mode if no FCF selected

If link is up, but no FCF is selected, don't send any ELS frames.

This came up when an fnic received a multicast advertisement but
no solitited advertisments, so no FCF was selected.  It tried
to send FLOGIs anyway.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: FIP should report link to libfc whether selected or not
Joe Eykholt [Tue, 3 Nov 2009 19:48:27 +0000 (11:48 -0800)]
[SCSI] libfcoe: FIP should report link to libfc whether selected or not

The fnic driver with FIP is reporting link up, even though it's down.

When the interface is shut down by the switch, we receive a clear
virtual link, and set the state reported to libfc as down, although
we still report it up.  Clearly wrong.  That causes the subsequent
link down event not to be reported, and /sys shows the host "Online".

Currently, in FIP mode, if an FCF times out, then link to libfc
is reported as down, to stop FLOGIs.  That interferes with the LLD
link down being reported.

Users really need to know the physical link information, to diagnose
cabling issues, so physical link status should be reported to libfc.

If the selected FCF needs to be reported, that should be done
separately, in a later patch.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: fip: allow FIP receive to be called from IRQ.
Joe Eykholt [Tue, 3 Nov 2009 19:48:22 +0000 (11:48 -0800)]
[SCSI] libfcoe: fip: allow FIP receive to be called from IRQ.

FIP's fcoe_ctlr_recv() function was previously only called from
the soft IRQ in FCoE.  It's not performance critical and is more
convenient for some drivers to call it from the IRQ level.  Just
Change to use skb_queue()/dequeue() which uses spinlock_irqsave
instead of separate locking with _bh locks.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: fip: use SCSI host number to identify debug messages.
Joe Eykholt [Tue, 3 Nov 2009 19:48:16 +0000 (11:48 -0800)]
[SCSI] libfcoe: fip: use SCSI host number to identify debug messages.

Use scsi host number to identify debug messages.
Previously, no instance information was given, so if multiple
ports were active, it became confusing.
Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: Allow FIP to be disabled by the driver
Joe Eykholt [Tue, 3 Nov 2009 19:48:11 +0000 (11:48 -0800)]
[SCSI] libfcoe: Allow FIP to be disabled by the driver

Allow FIP to be disabled by the driver for devices
that want to use libfcoe in non-FIP mode.

The driver merely sets the fcoe_ctlr mode to the state which
should be entered when the link comes up.  The default is auto.
No change is needed for fcoe.c which uses auto mode.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: adds can_queue ramp up
Vasu Dev [Tue, 3 Nov 2009 19:48:06 +0000 (11:48 -0800)]
[SCSI] libfc: adds can_queue ramp up

Adds last_can_queue_ramp_down_time and updates this on every
ramp down. If last_can_queue_ramp_down_time is not zero then
do ramp up on any IO completion in added fc_fcp_can_queue_ramp_up.

Reset last_can_queue_ramp_down_time to zero once can_queue
is ramped up to added max_can_queue limit, this is to avoid any
more ramp up attempts on subsequent IO completion.

The ramp down and up are skipped for FC_CAN_QUEUE_PERIOD
to avoid infrequent changes to can_queue, this required
keeping track of ramp up time also in last_can_queue_ramp_up_time.

Adds code to ramp down can_queue if lp->qfull is set, with added
new ramp up code the can_queue will be increased after
FC_CAN_QUEUE_PERIOD, therefore it is safe to do ramp down
without fsp in this case and will avoid thrash. This required
fc_fcp_can_queue_ramp_down locking change so that it can be
called with Scsi_Host lock held.

Removes si->throttled and fsp state FC_SRB_NOMEM, not needed with
added ramp up code.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: reduce can_queue for all FCP frame allocation failures
Vasu Dev [Tue, 3 Nov 2009 19:48:00 +0000 (11:48 -0800)]
[SCSI] libfc: reduce can_queue for all FCP frame allocation failures

Currently can_queue is reduced only if frame alloc fails
during fc_fcp_send_data but frame alloc can fail at several
other places in FCP data path and can_queue needs to be
reduced for any FCP frame alloc failure.

This patch adds fc_fcp_frame_alloc for all FCP frame allocations
and if fc_frame_alloc fails in fc_fcp_frame_alloc then reduce
can_queue in fc_fcp_frame_alloc, this will reduce can_queue for
all FCP frame alloc failures.

This required moving fc_fcp_reduce_can_queue up, to build without
adding its prototype. Also renamed fc_fcp_reduce_can_queue to
fc_fcp_can_queue_ramp_down.

Removes fc_fcp_reduce_can_queue calling from fc_fcp_recv since
not needed with added fc_fcp_frame_alloc reducing can_queue.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe, libfc: use single frame allocation API
Vasu Dev [Tue, 3 Nov 2009 19:47:55 +0000 (11:47 -0800)]
[SCSI] fcoe, libfc: use single frame allocation API

Cleans up frame allocation APIs to have just single fc_frame_alloc API.

Removes _fc_frame_alloc, renames __fc_frame_alloc to _fc_frame_alloc.

Modifies fc_fcp_send_data for removed _fc_frame_alloc, fc_fcp_send_data
was the only user of removed _fc_frame_alloc.

Also Adds check in fc_frame_alloc to do mod by 4 for only non-zero
len value.

This patch is prep work to fix can_queue reducing in next patch.
Single fc_frame_alloc API helps in fixing can_queue reducing in
next patch.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: Formatting cleanups and commenting
Robert Love [Tue, 3 Nov 2009 19:47:50 +0000 (11:47 -0800)]
[SCSI] fcoe: Formatting cleanups and commenting

Added kernel-doc comment blocks to all structures and functions.

Renamed fc_lport instances rom lp to lport to be inline with our
naming convention.

Renamed all misnamed net_device instances to netdev to be inline
with our naming convention.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe: formatting and comment cleanups
Robert Love [Tue, 3 Nov 2009 19:47:45 +0000 (11:47 -0800)]
[SCSI] libfcoe: formatting and comment cleanups

Ensures that there are kernel-doc style comments for all
routines and structures.

There were also a few instances of fc_lport's named 'lp'
which were switched to 'lport' as per the libfc/libfcoe/fcoe
naming convention.

Also, emacs 'indent-region' and 'tabify' were ran on libfcoe.c.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Formatting cleanups across libfc
Robert Love [Tue, 3 Nov 2009 19:47:39 +0000 (11:47 -0800)]
[SCSI] libfc: Formatting cleanups across libfc

This patch makes a variety of cleanup changes to all libfc files.

This patch adds kernel-doc headers to all functions lacking them
and attempts to better format existing headers. It also add kernel-doc
headers to structures.

This patch ensures that the current naming conventions for local ports,
remote ports and remote port private data is upheld in the following
manner.

struct               instance (i.e. variable name)
--------------------------------------------------
fc_lport                      lport
fc_rport                      rport
fc_rport_libfc_priv           rpriv
fc_rport_priv                 rdata

I also renamed dns_rp and ptp_rp to dns_rdata and ptp_rdata
respectively.

I used emacs 'indent-region' and 'tabify' on all libfc files
to correct spacing alignments.

I feel sorry for anyone attempting to review this patch.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc, fcoe: Add FC passthrough support
Steve Ma [Tue, 3 Nov 2009 19:47:34 +0000 (11:47 -0800)]
[SCSI] libfc, fcoe: Add FC passthrough support

This is the Open-FCoE implementation of the FC
passthrough support via bsg interface.

Passthrough support is added to both N_Ports and
VN_Ports.

Signed-off-by: Steve Ma <steve.ma@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Add routine to copy data from a buffer to a SG list
Robert Love [Tue, 3 Nov 2009 19:47:28 +0000 (11:47 -0800)]
[SCSI] libfc: Add routine to copy data from a buffer to a SG list

When handling the multi-frame responses of fc pass-thru requests,
a code segment similar to fc_fcp_recv_data (routine to receive
inbound SCSI data) is used in the response handler. This patch
is to add a routine, called fc_copy_buffer_to_sglist(), to handle
the common function of copying data from a buffer to a scatter-
gather list in order to avoid code duplication.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Export FC headers
Robert Love [Tue, 3 Nov 2009 19:47:23 +0000 (11:47 -0800)]
[SCSI] libfc: Export FC headers

Export fc_els.h, fc_fs.h, fc_gs.h and fc_ns.h so that they
may be used by applications.

This will be needed for FC Passthrough applications like fcping,
but could be used by other applications.

Fix to include <linux/types.h> to exported files provided by
Chris Leech <christopher.leech@intel.com>.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: vport symbolic name support
Chris Leech [Tue, 3 Nov 2009 19:47:18 +0000 (11:47 -0800)]
[SCSI] fcoe: vport symbolic name support

Allow a vport specific string to be appended to the port symbolic
name.  The new symbolic name is sent to the name server after it
is set.

This currently messes with libhbalinux, which is looking for
the fcoe "fcoe <ver> over <ethX>" string and expects whatever
comes after the "over" to be a network interface name only.

Adds an EXPORT_SYMBOL to libfc for fc_frame_alloc_fill, which is
needed to allow fcoe to allocate a frame of variable length for
the RSPN request.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: combine name server registration request functions
Chris Leech [Tue, 3 Nov 2009 19:47:12 +0000 (11:47 -0800)]
[SCSI] libfc: combine name server registration request functions

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: combine name server registration response handlers
Chris Leech [Tue, 3 Nov 2009 19:47:07 +0000 (11:47 -0800)]
[SCSI] libfc: combine name server registration response handlers

They all do the same thing, so combine them into a single function.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Register Symbolic Port Name (RSPN_ID)
Chris Leech [Tue, 3 Nov 2009 19:47:01 +0000 (11:47 -0800)]
[SCSI] libfc: Register Symbolic Port Name (RSPN_ID)

Register the fc_host symbolic name as the symbolic port name
with the fabric name server.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Register Symbolic Node Name (RSNN_NN)
Chris Leech [Tue, 3 Nov 2009 19:46:56 +0000 (11:46 -0800)]
[SCSI] libfc: Register Symbolic Node Name (RSNN_NN)

Register the fc_host symbolic name as the symbolic node name
with the fabric name server.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: RNN_ID may be required before RSNN_NN with some switches
Chris Leech [Tue, 3 Nov 2009 19:46:51 +0000 (11:46 -0800)]
[SCSI] libfc: RNN_ID may be required before RSNN_NN with some switches

One could interpret FC-GS-5 to say that an explicit RNN_ID is required
before RSNN_NN is allowed to succeed, which is why RNN_ID was not obsoleted
along with RPN_ID acording to this document:
ftp://ftp.t11.org/t11/member/fc/gs-5/05-546v2.pdf

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: RPN_ID is obsolete and unnecessary
Chris Leech [Tue, 3 Nov 2009 19:46:46 +0000 (11:46 -0800)]
[SCSI] libfc: RPN_ID is obsolete and unnecessary

RPN_ID has been obsolete per FC-GS-5 for several years.  The port name is
registered implicitly as part of FLOGI, and it is undesirable for ports to
change a registered port name using RPN_ID while logged into the fabric.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: NPIV vport create/destroy
Chris Leech [Tue, 3 Nov 2009 19:46:40 +0000 (11:46 -0800)]
[SCSI] fcoe: NPIV vport create/destroy

Add NPIV vport create and destroy handlers and register them with the
FC transport.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: add a separate scsi transport template for NPIV vports
Chris Leech [Tue, 3 Nov 2009 19:46:34 +0000 (11:46 -0800)]
[SCSI] fcoe: add a separate scsi transport template for NPIV vports

Right now it's exactly the same as the physical port template,
and there is no way to create a port on anything other than the
netdev.  When the vport_create entry point gets hooked up it will
create lports on top of vport devices, which will use this.

Rename scsi_transport_fcoe_sw to fcoe_transport_template to be more
clear with naming now that there are two templates.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfcoe, fcoe: libfcoe NPIV support
Chris Leech [Tue, 3 Nov 2009 19:46:29 +0000 (11:46 -0800)]
[SCSI] libfcoe, fcoe: libfcoe NPIV support

The FIP code in libfcoe needed several changes to support NPIV

1) dst_src_addr needs to be managed per-n_port-ID for FPMA fabrics with NPIV
   enabled.  Managing the MAC address is now handled in fcoe, with some slight
   changes to update_mac() and a new get_src_addr() function pointer.

2) The libfc elsct_send() hook is used to setup FCoE specific response
   handlers for FIP encapsulated ELS exchanges.  This lets the FCoE specific
   handling know which VN_Port the exchange is for, and doesn't require
   tracking OX_IDs.  It might be possible to roll back to the full FIP frame
   in these, but for now I've just stashed the contents of the MAC address
   descriptor in the skb context block for later use.  Also, because
   fcoe_elsct_send() just passes control on to fc_elsct_send(), all transmits
   still come through the normal frame_send() path.

3) The NPIV changes added a mutex hold in the keep alive sending, the lport
   mutex is protecting the vport list.  We can't take a mutex from a timer,
   so move the FIP keep alive logic to the link work struct.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc, libfcoe: FDISC ELS for NPIV
Chris Leech [Tue, 3 Nov 2009 19:46:24 +0000 (11:46 -0800)]
[SCSI] libfc, libfcoe: FDISC ELS for NPIV

Add FDISC ELS handling to libfc and libfcoe, treat it the same as FLOGI where
appropriate.

Add checking for NPIV support in the FLOGI LS_ACC service parameters.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: vport link handling and fc_vport state managment
Chris Leech [Tue, 3 Nov 2009 19:46:19 +0000 (11:46 -0800)]
[SCSI] libfc: vport link handling and fc_vport state managment

NPIV vports are managed in libfc by changing their virtual link state
when the parent N_Ports internal state changes.  The vport link is only
online when the N_Port is in a ready state (logged into the fabric).

vport_state is updated as needed in this patch as well, currently the states
LINKDOWN, INITIALIZING, ACTIVE, DSIABLED, and NO_FABRIC_SUPP are used.

This also changes the fc_host port_state handling to differentiate between
LINKDOWN and OFFLINE.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: add some generic NPIV support routines to libfc
Chris Leech [Tue, 3 Nov 2009 19:46:14 +0000 (11:46 -0800)]
[SCSI] libfc: add some generic NPIV support routines to libfc

Adds a function to create a new VN_Port instances, which share the EM
list with the N_Port, VN_Port lookup by fabric ID when responding to a new
request (otherwise the exchange lookup from the N_Ports EM list is trusted to
return an exchange with a cached lport value for the correct VN_Port),
a pointer to a fc_vport structure for VN_Ports, and flags to indicate if an
N_Port supports NPIV and if the switch/fabric allows it.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: changes to libfc_host_alloc to consolidate initialization with allocation
Chris Leech [Tue, 3 Nov 2009 19:46:08 +0000 (11:46 -0800)]
[SCSI] libfc: changes to libfc_host_alloc to consolidate initialization with allocation

I'd like to keep basic initialization together with allocation, which means
this can't just be a tail-call to scsi_host_alloc.

This is needed to create a generic libfc host allocation routine for NPIV
VN_Ports, which will share the exchange ID space (through sharing exchange
manager structures) with the parent lport.  In order to clone the exchange
manager list when the lport is allocated, the list head must be initialized
earlier.

Also, update fnic to use the libfc_host_alloc so that later changes do not break
it. (contribution by Joe Eykholt)

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Move libfc_init and libfc_exit to fc_libfc.c
Robert Love [Tue, 3 Nov 2009 19:46:03 +0000 (11:46 -0800)]
[SCSI] libfc: Move libfc_init and libfc_exit to fc_libfc.c

These routines are for the libfc kernel module and should be in
the libfc .c file.

Moving the libfc __init routine into fc_libfc.c caused the creation
of the fc_setup_fcp() and fc_destroy_fcp() routines so that
scsi_pkt_cachep was not exposed outside of fc_fcp.c.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Add libfc/fc_libfc.[ch] for libfc internal routines
Robert Love [Tue, 3 Nov 2009 19:45:58 +0000 (11:45 -0800)]
[SCSI] libfc: Add libfc/fc_libfc.[ch] for libfc internal routines

include/scsi/libfc.h is currently loaded with common code
shared between libfc's sub-modules as well as shared between
libfc and fcoe. Previous patches attempted to move out
non-common code. This patch creates two files for common
libfc routines that will not be shared with fcoe, fnic or
any other LLDs.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Remove fc_fcp_complete
Robert Love [Tue, 3 Nov 2009 19:45:52 +0000 (11:45 -0800)]
[SCSI] libfc: Remove fc_fcp_complete

This function is never used, let's remove it.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: Move non-common routines and prototypes out of libfc.h
Robert Love [Tue, 3 Nov 2009 19:45:47 +0000 (11:45 -0800)]
[SCSI] libfc: Move non-common routines and prototypes out of libfc.h

This patch moves all non-common routines and function prototypes
out of libfc.h and into the appropriate .c files. It makes these
routines 'static' when necessary and removes any unnecessary EXPORT_SYMBOL
statements.

A result of moving the fc_exch_seq_send, fc_seq_els_rsp_send, fc_exch_alloc
and fc_seq_start_next prototypes out of libfc.h is that they were no longer
being imported into fc_exch.c when libfc.h was included. This caused errors
where routines in fc_exch.c were looking for undefined symbols. To fix this
this patch reorganizes fc_seq_alloc, fc_seq_start_next and
fc_seq_start_next_locked. This move also made it so that
fc_seq_start_next_locked did not need to be prototyped at the top of
fc_exch.c.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe: Increase FCOE_MAX_LUN to 0xFFFF (65535)
Robert Love [Tue, 3 Nov 2009 19:45:42 +0000 (11:45 -0800)]
[SCSI] fcoe: Increase FCOE_MAX_LUN to 0xFFFF (65535)

The maximum number of LUNs was far too low. This value is
what most other FC HBAs are using.

Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fnic: Allocate OS interrupt resources just before enabling interrupts
Abhijeet Joglekar [Tue, 3 Nov 2009 19:45:37 +0000 (11:45 -0800)]
[SCSI] fnic: Allocate OS interrupt resources just before enabling interrupts

The OS interrupt vectors were getting allocated before the interrupt
resources were mapped from hardware. For Legacy interrupts, since
they are shared with other devices, as soon as an interrupt is
registered with the OS, it can fire while the fnic isr resource is
still unmapped. This can cause crash because of access to unmapped resources.
For MSIX and MSI, since interrupts are not shared with other devices,
this problem didnt happen, because the interrupt is enabled as the last
step before returning from _probe. For Legacy however, since the
interrupt is shared, the handler can be called as soon as it is registered.

Solution is to register interrupt handlers with OS as last step before
enabling device interrupts.

Signed-off-by: Abhijeet Joglekar <abjoglek@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_transport_fc: Introduce helper function for blocking scsi_eh
Christof Schmitt [Fri, 30 Oct 2009 16:59:29 +0000 (17:59 +0100)]
[SCSI] scsi_transport_fc: Introduce helper function for blocking scsi_eh

Move the duplicated code from FC LLDs to SCSI FC transport class.

Acked-by: James Smart <james.smart@emulex.com>
Acked-by: Giridhar Malavali <giridhar.malavali@qlogic.com>
Acked-by: Abhijeet Joglekar <abjoglek@cisco.com>
Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_lib: fix potential NULL dereference
Jiri Slaby [Wed, 23 Sep 2009 14:15:35 +0000 (16:15 +0200)]
[SCSI] scsi_lib: fix potential NULL dereference

Stanse found a potential NULL dereference in scsi_kill_request.

Instead of triggering BUG() in 'if (unlikely(cmd == NULL))' branch,
the kernel will Oops earlier on cmd dereference.

Move the dereferences after the if.

Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] 3w-sas: Add new driver for LSI 3ware 9750
adam radford [Fri, 23 Oct 2009 21:52:33 +0000 (14:52 -0700)]
[SCSI] 3w-sas: Add new driver for LSI 3ware 9750

[jejb: fix up for new queue depth code]
Signed-off-by: Adam Radford <aradford@gmail.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_debug: fix Thin provisioning support
Douglas Gilbert [Thu, 29 Oct 2009 05:48:31 +0000 (01:48 -0400)]
[SCSI] scsi_debug: fix Thin provisioning support

While testing scsi_debug with these patches I found a
problem with the Block Limits VPD page function. The
length returned by the inquiry_evpd_b0() function was
too short. A patch to fix that and a cosmetic change
(that the form factor of scsi_debug is less than 1.8
inches) is attached.

Signed-off-by: Douglas Gilbert <dgilbert@interlog.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] vmw_pvscsi: SCSI driver for VMware's virtual HBA.
Alok Kataria [Tue, 13 Oct 2009 21:51:05 +0000 (14:51 -0700)]
[SCSI] vmw_pvscsi: SCSI driver for VMware's virtual HBA.

This is a driver for VMware's paravirtualized SCSI device,
which should improve disk performance for guests running
under control of VMware hypervisors that support such devices.

Signed-off-by: Alok N Kataria <akataria@vmware.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_dh: Make alua hardware handler's activate() async
Chandra Seetharaman [Wed, 21 Oct 2009 16:23:04 +0000 (09:23 -0700)]
[SCSI] scsi_dh: Make alua hardware handler's activate() async

Make the activate function asynchronous by using blk_execute_rq_nowait()

Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_dh: Make hp hardware handler's activate() async
Chandra Seetharaman [Wed, 21 Oct 2009 16:22:58 +0000 (09:22 -0700)]
[SCSI] scsi_dh: Make hp hardware handler's activate() async

Make the activate function asynchronous by using blk_execute_rq_nowait()

Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_dh: Make rdac hardware handler's activate() async
Chandra Seetharaman [Wed, 21 Oct 2009 16:22:51 +0000 (09:22 -0700)]
[SCSI] scsi_dh: Make rdac hardware handler's activate() async

Batch up MODE_SELECT in rdac device handler.

LSI RDAC storage has the capability of handling mode selects for
multiple luns in a same command. Make use of that ability to send
as few MODE SELECTs as possible to the storage controller as possible.

This patch creates a work queue and queues up activate requests
when a MODE SELECT is sent down the wire. When that MODE SELECT
completes, it compiles queued up activate requests for multiple
luns into a single MODE SELECT.

This reduces the time to do failover/failback of large number of LUNS.

Signed-off-by: Babu Moger <babu.moger@lsi.com>
Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi_dh: Change the scsidh_activate interface to be asynchronous
Chandra Seetharaman [Wed, 21 Oct 2009 16:22:46 +0000 (09:22 -0700)]
[SCSI] scsi_dh: Change the scsidh_activate interface to be asynchronous

Make scsi_dh_activate() function asynchronous, by taking in two additional
parameters, one is the callback function and the other is the data to call
the callback function with.

Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] zfcp: Adapt change_queue_depth for queue full tracking
Christof Schmitt [Fri, 16 Oct 2009 00:47:11 +0000 (17:47 -0700)]
[SCSI] zfcp: Adapt change_queue_depth for queue full tracking

Adapt the change_queue_depth callback in zfcp for the new reason
parameter. Simply pass each call back to the SCSI midlayer, there are
no resource adjustments necessary for zfcp.

Signed-off-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Removes check for (depth <= default_depth) in case of
SCSI_QDEPTH_RAMP_UP call back, not needed after added
max_queue_depth per sdev.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: add queue_depth ramp up
Vasu Dev [Fri, 16 Oct 2009 00:47:06 +0000 (17:47 -0700)]
[SCSI] libfc: add queue_depth ramp up

Adjust queue_depth on fc_change_queue_depth call back
with reason SCSI_QDEPTH_RAMP_UP, no additional resource
adjustments necessary for libfc.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Acked-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] add queue_depth ramp up code
Vasu Dev [Thu, 22 Oct 2009 22:46:33 +0000 (15:46 -0700)]
[SCSI] add queue_depth ramp up code

Current FC HBA queue_depth ramp up code depends on last queue
full time. The sdev already  has last_queue_full_time field to
track last queue full time but stored value is truncated by
last four bits.

So this patch updates last_queue_full_time without truncating
last 4 bits to store full value and then updates its only
current usages in scsi_track_queue_full to ignore last four bits
to keep current usages same while also use this field
in added ramp up code.

Adds scsi_handle_queue_ramp_up to ramp up queue_depth on
successful completion of IO. The scsi_handle_queue_ramp_up will
do ramp up on all luns of a target, just same as ramp down done
on all luns on a target.

The ramp up is skipped in case the change_queue_depth is not
supported by LLD or already reached to added max_queue_depth.

Updates added max_queue_depth on every new update to default
queue_depth value.

The ramp up is also skipped if lapsed time since either last
queue ramp up or down is less than LLD specified
queue_ramp_up_period.

Adds queue_ramp_up_period to sysfs but only if change_queue_depth
is supported since ramp up and queue_ramp_up_period is needed only
in case change_queue_depth is supported first.

Initializes queue_ramp_up_period to 120HZ jiffies as initial
default value, it is same as used in existing lpfc and qla2xxx.

-v2
 Combined all ramp code into this single patch.

-v3
 Moves max_queue_depth initialization after slave_configure is
called from after slave_alloc calling done. Also adjusted
max_queue_depth check to skip ramp up if current queue_depth
is >= max_queue_depth.

-v4
 Changes sdev->queue_ramp_up_period unit to ms when using sysfs i/f
to store or show its value.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Tested-by: Christof Schmitt <christof.schmitt@de.ibm.com>
Tested-by: Giridhar Malavali <giridhar.malavali@qlogic.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fcoe, libfc: fix an libfc issue with queue ramp down in libfc
Vasu Dev [Fri, 16 Oct 2009 00:46:55 +0000 (17:46 -0700)]
[SCSI] fcoe, libfc: fix an libfc issue with queue ramp down in libfc

The cmd_per_lun value is used by scsi-ml as fall back lowest
queue_depth value but in case of libfc cmd_per_lun is set to
same value as max queue_depth = 32.

So this patch reduces cmd_per_lun value to 3 and configures
each lun with default max queue_depth 32 in fc_slave_alloc.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Acked-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: convert to scsi_track_queue_full
Mike Christie [Fri, 16 Oct 2009 00:46:50 +0000 (17:46 -0700)]
[SCSI] libfc: convert to scsi_track_queue_full

This converts the libfc using scsi_track_queue_full to
track the queue full from the change_queue_depth callback.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Acked-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] scsi error: have scsi-ml call change_queue_depth to handle QUEUE_FULL
Mike Christie [Fri, 16 Oct 2009 00:46:44 +0000 (17:46 -0700)]
[SCSI] scsi error: have scsi-ml call change_queue_depth to handle QUEUE_FULL

This has scsi-ml call the change_queue_depth functions when
we get a QUEUE_FULL. It will only change the queue depth if
change_queue_depth is set because the LLD may have to
modify some internal resources, so I thought this would
be the safest route.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
-v2
Limits change_queue_depth to only all luns of target by adding
channel check while iterating for all luns of Scsi_Host. This is
same as currently qla2xxx FC HBA does on QUEUE_FULL event.

Signed-off-by: Vasu Dev <vasu.dev@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] modify change_queue_depth to take in reason why it is being called
Mike Christie [Fri, 16 Oct 2009 00:46:39 +0000 (17:46 -0700)]
[SCSI] modify change_queue_depth to take in reason why it is being called

This patch modifies scsi_host_template->change_queue_depth so that
it takes an argument indicating why it is being called. This will be
used so that if a LLD needs to do some extra processing when
handling queue fulls or later ramp ups, it can do so.

This is a simple port of the drivers setting a change_queue_depth
callback. In the patch I just have these LLDs adjust the queue depth
if the user was requesting it.

Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
[Vasu.Dev: v2
Also converted pmcraid_change_queue_depth and then verified
all modules compile  using "make allmodconfig" for any new build
warnings on X86_64.

Updated original description after combing two original
patches from Mike to make this patch git bisectable.]
Signed-off-by: Vasu Dev <vasu.dev@intel.com>
[jejb: fixed up 53c700]
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] pm8001: add SAS/SATA HBA driver
jack wang [Wed, 14 Oct 2009 08:19:21 +0000 (16:19 +0800)]
[SCSI] pm8001: add SAS/SATA HBA driver

This driver supports PMC-Sierra PCIe SAS/SATA 8x6G SPC 8001 chip based
host adapters.

Signed-off-by: Jack Wang <jack_wang@usish.com>
Signed-off-by: Lindar Liu <lindar_liu@usish.com>
Signed-off-by: Tom Peng <tom_peng@usish.com>
Signed-off-by: Kevin Ao <aoqingyun@usish.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] be2iscsi: Adding Ring Mode Wrb's V3
Jayamohan Kallickal [Fri, 23 Oct 2009 06:23:49 +0000 (11:53 +0530)]
[SCSI] be2iscsi: Adding Ring Mode Wrb's V3

This patch adds support for ring based wrbs

Signed-off-by: Jayamohan Kallickal <jayamohank@serverengines.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] be2iscsi: Adding msix and mcc_rings V3
Jayamohan Kallickal [Fri, 23 Oct 2009 06:22:33 +0000 (11:52 +0530)]
[SCSI] be2iscsi: Adding msix and mcc_rings V3

This patch enables msix for be2iscsi. It also enables use
of mcc_rings for fw commands. Since the mcc eq creation is
dependent on msix I am sending as one patch

Signed-off-by: Jayamohan Kallickal <jayamohank@serverengines.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: fix free of fc_rport_priv with timer pending
Joe Eykholt [Wed, 21 Oct 2009 23:28:30 +0000 (16:28 -0700)]
[SCSI] libfc: fix free of fc_rport_priv with timer pending

Timer crashes were caused by freeing a struct fc_rport_priv
with a timer pending, causing the timer facility list to be
corrupted.  This was during FC uplink flap tests with a lot
of targets.

After discovery, we were doing an PLOGI on an rdata that was
in DELETE state but not yet removed from the lookup list.
This moved the rdata from DELETE state to PLOGI state.
If the PLOGI exchange allocation failed and needed to be
retried, the timer scheduling could race with the free
being done by fc_rport_work().

When fc_rport_login() is called on a rport in DELETE state,
move it to a new state RESTART.  In fc_rport_work, when
handling a LOGO, STOPPED or FAILED event, look for restart
state.  In the RESTART case, don't take the rdata off the
list and after the transport remote port is deleted and
exchanges are reset, re-login to the remote port.

Note that the new RESTART state also corrects a problem we
had when re-discovering a port that had moved to DELETE state.
In that case, a new rdata was created, but the old rdata
would do an exchange manager reset affecting the FC_ID
for both the new rdata and old rdata.  With the new state,
the new port isn't logged into until after any old exchanges
are reset.

Signed-off-by: Joe Eykholt <jeykholt@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fnic: Pad the unused bytes of CDB to 0s
Abhijeet Joglekar [Wed, 21 Oct 2009 23:28:25 +0000 (16:28 -0700)]
[SCSI] fnic: Pad the unused bytes of CDB to 0s

Signed-off-by: Abhijeet Joglekar <abjoglek@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fnic: Set max_cmd_len to driver supported CDB length
Abhijeet Joglekar [Wed, 21 Oct 2009 23:28:19 +0000 (16:28 -0700)]
[SCSI] fnic: Set max_cmd_len to driver supported CDB length

Signed-off-by: Abhijeet Joglekar <abjoglek@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] fnic: Process all cq entries per ISR
Abhijeet Joglekar [Wed, 21 Oct 2009 23:28:14 +0000 (16:28 -0700)]
[SCSI] fnic: Process all cq entries per ISR

Driver was processing a fixed max number of cq descriptors per ISR. For
instance, for the SCSI IO queue, number of IOs processed per ISR were 8.
If hardware writes 9 cq descriptors to the cq and generates an interrupt,
driver would process only 8 descriptors and decrement the outstanding
credit count by 8. Unless another interrupt event happens, the hw does
not generate any additional interrupt. This results in the cq descriptor
sitting in the queue without being procesed and can cause IO timeouts
and aborts.

Modify all ISR functions to process all queued cq descriptors in one shot.
Since bulk of ELS frame processing is done in thread context and bulk
of SCSI IO processing is done in soft ISR deferred context, the cycles
spent in the ISR per cq descriptor is small.

Signed-off-by: Herman Lee <hermlee@cisco.com>
Signed-off-by: Abhijeet Joglekar <abjoglek@cisco.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
14 years ago[SCSI] libfc: fix memory corruption caused by double frees and bad error handling
Chris Leech [Wed, 21 Oct 2009 23:28:09 +0000 (16:28 -0700)]
[SCSI] libfc: fix memory corruption caused by double frees and bad error handling

I was running into several different panics under stress, which I traced down
to a few different possible slab corruption issues in error handling paths.
I have not yet looked into why these exchange sends fail, but with these
fixes my test system is much more stable under stress than before.

fc_elsct_send() could fail and either leave the passed in frame intact
(failure in fc_ct/els_fill) or the frame could have been freed if the
failure was is fc_exch_seq_send().  The caller had no way of knowing, and
there was a potential double free in the error handling in fc_fcp_rec().

Make fc_elsct_send() always free the frame before returning, and remove the
fc_frame_free() call in fc_fcp_rec().

While fc_exch_seq_send() did always consume the frame, there were double free
bugs in the error handling of fc_fcp_cmd_send() and fc_fcp_srr() as well.

Numerous calls to error handling routines (fc_disc_error(),
fc_lport_error(), fc_rport_error_retry() ) were passing in a frame pointer that
had already been freed in the case of an error.  I have changed the call
sites to pass in a NULL pointer, but there may be more appropriate error
codes to use.

Question:  Why do these error routines take a frame pointer anyway?  I
understand passing in a pointer encoded error to the response handlers, but
the error routines take no action on a valid pointer and should never be
called that way.

Signed-off-by: Chris Leech <christopher.leech@intel.com>
Signed-off-by: Robert Love <robert.w.love@intel.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>