scsi: ibmvfc: Remove BUG_ON in the case of an empty event pool
authorTyrel Datwyler <tyreld@linux.ibm.com>
Thu, 21 Sep 2023 22:54:25 +0000 (17:54 -0500)
committerMartin K. Petersen <martin.petersen@oracle.com>
Fri, 22 Sep 2023 02:14:20 +0000 (22:14 -0400)
In practice the driver should never send more commands than are allocated
to a queue's event pool. In the unlikely event that this happens, the code
asserts a BUG_ON, and in the case that the kernel is not configured to
crash on panic returns a junk event pointer from the empty event list
causing things to spiral from there. This BUG_ON is a historical artifact
of the ibmvfc driver first being upstreamed, and it is well known now that
the use of BUG_ON is bad practice except in the most unrecoverable
scenario. There is nothing about this scenario that prevents the driver
from recovering and carrying on.

Remove the BUG_ON in question from ibmvfc_get_event() and return a NULL
pointer in the case of an empty event pool. Update all call sites to
ibmvfc_get_event() to check for a NULL pointer and perfrom the appropriate
failure or recovery action.

Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
Link: https://lore.kernel.org/r/20230921225435.3537728-2-tyreld@linux.ibm.com
Reviewed-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/ibmvscsi/ibmvfc.c

index ce9eb00e2ca04d6b1a500a39df6bcee19fd71e0f..712d109cb558947d1e8fd2e106c7abca87c7d6a1 100644 (file)
@@ -1519,7 +1519,11 @@ static struct ibmvfc_event *ibmvfc_get_event(struct ibmvfc_queue *queue)
        unsigned long flags;
 
        spin_lock_irqsave(&queue->l_lock, flags);
-       BUG_ON(list_empty(&queue->free));
+       if (list_empty(&queue->free)) {
+               ibmvfc_log(queue->vhost, 4, "empty event pool on queue:%ld\n", queue->hwq_id);
+               spin_unlock_irqrestore(&queue->l_lock, flags);
+               return NULL;
+       }
        evt = list_entry(queue->free.next, struct ibmvfc_event, queue_list);
        atomic_set(&evt->free, 0);
        list_del(&evt->queue_list);
@@ -1948,9 +1952,15 @@ static int ibmvfc_queuecommand(struct Scsi_Host *shost, struct scsi_cmnd *cmnd)
        if (vhost->using_channels) {
                scsi_channel = hwq % vhost->scsi_scrqs.active_queues;
                evt = ibmvfc_get_event(&vhost->scsi_scrqs.scrqs[scsi_channel]);
+               if (!evt)
+                       return SCSI_MLQUEUE_HOST_BUSY;
+
                evt->hwq = hwq % vhost->scsi_scrqs.active_queues;
-       } else
+       } else {
                evt = ibmvfc_get_event(&vhost->crq);
+               if (!evt)
+                       return SCSI_MLQUEUE_HOST_BUSY;
+       }
 
        ibmvfc_init_event(evt, ibmvfc_scsi_done, IBMVFC_CMD_FORMAT);
        evt->cmnd = cmnd;
@@ -2038,6 +2048,11 @@ static int ibmvfc_bsg_timeout(struct bsg_job *job)
 
        vhost->aborting_passthru = 1;
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               spin_unlock_irqrestore(vhost->host->host_lock, flags);
+               return -ENOMEM;
+       }
+
        ibmvfc_init_event(evt, ibmvfc_bsg_timeout_done, IBMVFC_MAD_FORMAT);
 
        tmf = &evt->iu.tmf;
@@ -2096,6 +2111,10 @@ static int ibmvfc_bsg_plogi(struct ibmvfc_host *vhost, unsigned int port_id)
                goto unlock_out;
 
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               rc = -ENOMEM;
+               goto unlock_out;
+       }
        ibmvfc_init_event(evt, ibmvfc_sync_completion, IBMVFC_MAD_FORMAT);
        plogi = &evt->iu.plogi;
        memset(plogi, 0, sizeof(*plogi));
@@ -2214,6 +2233,11 @@ static int ibmvfc_bsg_request(struct bsg_job *job)
        }
 
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               spin_unlock_irqrestore(vhost->host->host_lock, flags);
+               rc = -ENOMEM;
+               goto out;
+       }
        ibmvfc_init_event(evt, ibmvfc_sync_completion, IBMVFC_MAD_FORMAT);
        mad = &evt->iu.passthru;
 
@@ -2302,6 +2326,11 @@ static int ibmvfc_reset_device(struct scsi_device *sdev, int type, char *desc)
                else
                        evt = ibmvfc_get_event(&vhost->crq);
 
+               if (!evt) {
+                       spin_unlock_irqrestore(vhost->host->host_lock, flags);
+                       return -ENOMEM;
+               }
+
                ibmvfc_init_event(evt, ibmvfc_sync_completion, IBMVFC_CMD_FORMAT);
                tmf = ibmvfc_init_vfc_cmd(evt, sdev);
                iu = ibmvfc_get_fcp_iu(vhost, tmf);
@@ -2505,6 +2534,8 @@ static struct ibmvfc_event *ibmvfc_init_tmf(struct ibmvfc_queue *queue,
        struct ibmvfc_tmf *tmf;
 
        evt = ibmvfc_get_event(queue);
+       if (!evt)
+               return NULL;
        ibmvfc_init_event(evt, ibmvfc_sync_completion, IBMVFC_MAD_FORMAT);
 
        tmf = &evt->iu.tmf;
@@ -2561,6 +2592,11 @@ static int ibmvfc_cancel_all_mq(struct scsi_device *sdev, int type)
 
                if (found_evt && vhost->logged_in) {
                        evt = ibmvfc_init_tmf(&queues[i], sdev, type);
+                       if (!evt) {
+                               spin_unlock(queues[i].q_lock);
+                               spin_unlock_irqrestore(vhost->host->host_lock, flags);
+                               return -ENOMEM;
+                       }
                        evt->sync_iu = &queues[i].cancel_rsp;
                        ibmvfc_send_event(evt, vhost, default_timeout);
                        list_add_tail(&evt->cancel, &cancelq);
@@ -2774,6 +2810,10 @@ static int ibmvfc_abort_task_set(struct scsi_device *sdev)
 
        if (vhost->state == IBMVFC_ACTIVE) {
                evt = ibmvfc_get_event(&vhost->crq);
+               if (!evt) {
+                       spin_unlock_irqrestore(vhost->host->host_lock, flags);
+                       return -ENOMEM;
+               }
                ibmvfc_init_event(evt, ibmvfc_sync_completion, IBMVFC_CMD_FORMAT);
                tmf = ibmvfc_init_vfc_cmd(evt, sdev);
                iu = ibmvfc_get_fcp_iu(vhost, tmf);
@@ -4032,6 +4072,12 @@ static void ibmvfc_tgt_send_prli(struct ibmvfc_target *tgt)
 
        kref_get(&tgt->kref);
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_NONE);
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               return;
+       }
        vhost->discovery_threads++;
        ibmvfc_init_event(evt, ibmvfc_tgt_prli_done, IBMVFC_MAD_FORMAT);
        evt->tgt = tgt;
@@ -4139,6 +4185,12 @@ static void ibmvfc_tgt_send_plogi(struct ibmvfc_target *tgt)
        kref_get(&tgt->kref);
        tgt->logo_rcvd = 0;
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_NONE);
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               return;
+       }
        vhost->discovery_threads++;
        ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_INIT_WAIT);
        ibmvfc_init_event(evt, ibmvfc_tgt_plogi_done, IBMVFC_MAD_FORMAT);
@@ -4215,6 +4267,8 @@ static struct ibmvfc_event *__ibmvfc_tgt_get_implicit_logout_evt(struct ibmvfc_t
 
        kref_get(&tgt->kref);
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt)
+               return NULL;
        ibmvfc_init_event(evt, done, IBMVFC_MAD_FORMAT);
        evt->tgt = tgt;
        mad = &evt->iu.implicit_logout;
@@ -4242,6 +4296,13 @@ static void ibmvfc_tgt_implicit_logout(struct ibmvfc_target *tgt)
        vhost->discovery_threads++;
        evt = __ibmvfc_tgt_get_implicit_logout_evt(tgt,
                                                   ibmvfc_tgt_implicit_logout_done);
+       if (!evt) {
+               vhost->discovery_threads--;
+               ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_NONE);
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               return;
+       }
 
        ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_INIT_WAIT);
        if (ibmvfc_send_event(evt, vhost, default_timeout)) {
@@ -4381,6 +4442,12 @@ static void ibmvfc_tgt_move_login(struct ibmvfc_target *tgt)
 
        kref_get(&tgt->kref);
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_DEL_RPORT);
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               return;
+       }
        vhost->discovery_threads++;
        ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_INIT_WAIT);
        ibmvfc_init_event(evt, ibmvfc_tgt_move_login_done, IBMVFC_MAD_FORMAT);
@@ -4547,6 +4614,14 @@ static void ibmvfc_adisc_timeout(struct timer_list *t)
        vhost->abort_threads++;
        kref_get(&tgt->kref);
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               tgt_err(tgt, "Failed to get cancel event for ADISC.\n");
+               vhost->abort_threads--;
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               spin_unlock_irqrestore(vhost->host->host_lock, flags);
+               return;
+       }
        ibmvfc_init_event(evt, ibmvfc_tgt_adisc_cancel_done, IBMVFC_MAD_FORMAT);
 
        evt->tgt = tgt;
@@ -4597,6 +4672,12 @@ static void ibmvfc_tgt_adisc(struct ibmvfc_target *tgt)
 
        kref_get(&tgt->kref);
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_NONE);
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               return;
+       }
        vhost->discovery_threads++;
        ibmvfc_init_event(evt, ibmvfc_tgt_adisc_done, IBMVFC_MAD_FORMAT);
        evt->tgt = tgt;
@@ -4700,6 +4781,12 @@ static void ibmvfc_tgt_query_target(struct ibmvfc_target *tgt)
 
        kref_get(&tgt->kref);
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               ibmvfc_set_tgt_action(tgt, IBMVFC_TGT_ACTION_NONE);
+               kref_put(&tgt->kref, ibmvfc_release_tgt);
+               __ibmvfc_reset_host(vhost);
+               return;
+       }
        vhost->discovery_threads++;
        evt->tgt = tgt;
        ibmvfc_init_event(evt, ibmvfc_tgt_query_target_done, IBMVFC_MAD_FORMAT);
@@ -4872,6 +4959,13 @@ static void ibmvfc_discover_targets(struct ibmvfc_host *vhost)
 {
        struct ibmvfc_discover_targets *mad;
        struct ibmvfc_event *evt = ibmvfc_get_event(&vhost->crq);
+       int level = IBMVFC_DEFAULT_LOG_LEVEL;
+
+       if (!evt) {
+               ibmvfc_log(vhost, level, "Discover Targets failed: no available events\n");
+               ibmvfc_hard_reset_host(vhost);
+               return;
+       }
 
        ibmvfc_init_event(evt, ibmvfc_discover_targets_done, IBMVFC_MAD_FORMAT);
        mad = &evt->iu.discover_targets;
@@ -4949,8 +5043,15 @@ static void ibmvfc_channel_setup(struct ibmvfc_host *vhost)
        struct ibmvfc_scsi_channels *scrqs = &vhost->scsi_scrqs;
        unsigned int num_channels =
                min(vhost->client_scsi_channels, vhost->max_vios_scsi_channels);
+       int level = IBMVFC_DEFAULT_LOG_LEVEL;
        int i;
 
+       if (!evt) {
+               ibmvfc_log(vhost, level, "Channel Setup failed: no available events\n");
+               ibmvfc_hard_reset_host(vhost);
+               return;
+       }
+
        memset(setup_buf, 0, sizeof(*setup_buf));
        if (num_channels == 0)
                setup_buf->flags = cpu_to_be32(IBMVFC_CANCEL_CHANNELS);
@@ -5012,6 +5113,13 @@ static void ibmvfc_channel_enquiry(struct ibmvfc_host *vhost)
 {
        struct ibmvfc_channel_enquiry *mad;
        struct ibmvfc_event *evt = ibmvfc_get_event(&vhost->crq);
+       int level = IBMVFC_DEFAULT_LOG_LEVEL;
+
+       if (!evt) {
+               ibmvfc_log(vhost, level, "Channel Enquiry failed: no available events\n");
+               ibmvfc_hard_reset_host(vhost);
+               return;
+       }
 
        ibmvfc_init_event(evt, ibmvfc_channel_enquiry_done, IBMVFC_MAD_FORMAT);
        mad = &evt->iu.channel_enquiry;
@@ -5134,6 +5242,12 @@ static void ibmvfc_npiv_login(struct ibmvfc_host *vhost)
        struct ibmvfc_npiv_login_mad *mad;
        struct ibmvfc_event *evt = ibmvfc_get_event(&vhost->crq);
 
+       if (!evt) {
+               ibmvfc_dbg(vhost, "NPIV Login failed: no available events\n");
+               ibmvfc_hard_reset_host(vhost);
+               return;
+       }
+
        ibmvfc_gather_partition_info(vhost);
        ibmvfc_set_login_info(vhost);
        ibmvfc_init_event(evt, ibmvfc_npiv_login_done, IBMVFC_MAD_FORMAT);
@@ -5198,6 +5312,12 @@ static void ibmvfc_npiv_logout(struct ibmvfc_host *vhost)
        struct ibmvfc_event *evt;
 
        evt = ibmvfc_get_event(&vhost->crq);
+       if (!evt) {
+               ibmvfc_dbg(vhost, "NPIV Logout failed: no available events\n");
+               ibmvfc_hard_reset_host(vhost);
+               return;
+       }
+
        ibmvfc_init_event(evt, ibmvfc_npiv_logout_done, IBMVFC_MAD_FORMAT);
 
        mad = &evt->iu.npiv_logout;