[ofa-general] Re: [GIT PULL] please pull infiniband.git for-linus

[David Dillow]

Subject: IB/srp: Fix list corruption/oops on module reload

ib_srp doesn't clean up the transport attributes properly when
unloading, so it leaves references around to free'd memory.

The srp_remove_host() cannot go before the scsi_remove_host() call as
the documented call sequence suggests, as it will cause an oops when the
SRP transport code will free up an object that is not its own. So,
temporarily reorder the calls in srp_remove_work() to avoid the problem
as well until the transport code can be fixed.

Signed-off-by: David Dillow <dillowda at ornl.gov>
---
On Thu, 2008-01-03 at 15:13 -0500, David Dillow wrote:
> I'll send a new patch to you for just the ib_srp.c side.

Should've just sent it with the last email.

diff --git a/drivers/infiniband/ulp/srp/ib_srp.c b/drivers/infiniband/ulp/srp/ib_srp.c
index 950228f..6e7e3c8 100644
--- a/drivers/infiniband/ulp/srp/ib_srp.c
+++ b/drivers/infiniband/ulp/srp/ib_srp.c
@@ -423,8 +423,8 @@ static void srp_remove_work(struct work_struct *work)
 	list_del(&target->list);
 	spin_unlock(&target->srp_host->target_lock);
 
-	srp_remove_host(target->scsi_host);
 	scsi_remove_host(target->scsi_host);
+	srp_remove_host(target->scsi_host);
 	ib_destroy_cm_id(target->cm_id);
 	srp_free_target_ib(target);
 	scsi_host_put(target->scsi_host);
@@ -2054,6 +2054,7 @@ static void srp_remove_one(struct ib_device *device)
 		list_for_each_entry_safe(target, tmp_target,
 					 &host->target_list, list) {
 			scsi_remove_host(target->scsi_host);
+			srp_remove_host(target->scsi_host);
 			srp_disconnect_target(target);
 			ib_destroy_cm_id(target->cm_id);
 			srp_free_target_ib(target);