RDMA/srp: Rework SCSI device reset handling
authorBart Van Assche <bvanassche@acm.org>
Wed, 30 Jan 2019 22:05:55 +0000 (14:05 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 27 Feb 2019 09:08:07 +0000 (10:08 +0100)
commit 48396e80fb6526ea5ed267bd84f028bae56d2f9e upstream.

Since .scsi_done() must only be called after scsi_queue_rq() has
finished, make sure that the SRP initiator driver does not call
.scsi_done() while scsi_queue_rq() is in progress. Although
invoking sg_reset -d while I/O is in progress works fine with kernel
v4.20 and before, that is not the case with kernel v5.0-rc1. This
patch avoids that the following crash is triggered with kernel
v5.0-rc1:

BUG: unable to handle kernel NULL pointer dereference at 0000000000000138
CPU: 0 PID: 360 Comm: kworker/0:1H Tainted: G    B             5.0.0-rc1-dbg+ #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.10.2-1 04/01/2014
Workqueue: kblockd blk_mq_run_work_fn
RIP: 0010:blk_mq_dispatch_rq_list+0x116/0xb10
Call Trace:
 blk_mq_sched_dispatch_requests+0x2f7/0x300
 __blk_mq_run_hw_queue+0xd6/0x180
 blk_mq_run_work_fn+0x27/0x30
 process_one_work+0x4f1/0xa20
 worker_thread+0x67/0x5b0
 kthread+0x1cf/0x1f0
 ret_from_fork+0x24/0x30

Cc: <stable@vger.kernel.org>
Fixes: 94a9174c630c ("IB/srp: reduce lock coverage of command completion")
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/infiniband/ulp/srp/ib_srp.c

index ade98c234dcb39e1c860bb8b7a115964ac61d540..3f5b5893792cd4cb5341cbd551100dba87a169f0 100644 (file)
@@ -2669,7 +2669,6 @@ static int srp_reset_device(struct scsi_cmnd *scmnd)
 {
        struct srp_target_port *target = host_to_target(scmnd->device->host);
        struct srp_rdma_ch *ch;
-       int i, j;
        u8 status;
 
        shost_printk(KERN_ERR, target->scsi_host, "SRP reset_device called\n");
@@ -2681,15 +2680,6 @@ static int srp_reset_device(struct scsi_cmnd *scmnd)
        if (status)
                return FAILED;
 
-       for (i = 0; i < target->ch_count; i++) {
-               ch = &target->ch[i];
-               for (j = 0; j < target->req_ring_size; ++j) {
-                       struct srp_request *req = &ch->req_ring[j];
-
-                       srp_finish_req(ch, req, scmnd->device, DID_RESET << 16);
-               }
-       }
-
        return SUCCESS;
 }