From 65080c51fde468465e3547d72e4a593b1361c0f4 Mon Sep 17 00:00:00 2001 From: Dmitry Bogdanov Date: Tue, 7 Jun 2022 16:19:53 +0300 Subject: scsi: iscsi: Prefer xmit of DataOut over new commands iscsi_data_xmit() (TX worker) is iterating over the queue of new SCSI commands concurrently with the queue being replenished. Only after the queue is emptied will we start sending pending DataOut PDUs. That leads to DataOut timeout on the target side and to connection reinstatement. Give priority to pending DataOut commands over new commands. Link: https://lore.kernel.org/r/20220607131953.11584-1-d.bogdanov@yadro.com Reviewed-by: Konstantin Shelekhin Reviewed-by: Mike Christie Signed-off-by: Dmitry Bogdanov Signed-off-by: Martin K. Petersen --- drivers/scsi/libiscsi.c | 44 ++++++++++++++++++++++++-------------------- 1 file changed, 24 insertions(+), 20 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 797abf4f5399..8d78559ae94a 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -1567,6 +1567,28 @@ check_mgmt: goto done; } +check_requeue: + while (!list_empty(&conn->requeue)) { + /* + * we always do fastlogout - conn stop code will clean up. + */ + if (conn->session->state == ISCSI_STATE_LOGGING_OUT) + break; + + task = list_entry(conn->requeue.next, struct iscsi_task, + running); + + if (iscsi_check_tmf_restrictions(task, ISCSI_OP_SCSI_DATA_OUT)) + break; + + list_del_init(&task->running); + rc = iscsi_xmit_task(conn, task, true); + if (rc) + goto done; + if (!list_empty(&conn->mgmtqueue)) + goto check_mgmt; + } + /* process pending command queue */ while (!list_empty(&conn->cmdqueue)) { task = list_entry(conn->cmdqueue.next, struct iscsi_task, @@ -1594,28 +1616,10 @@ check_mgmt: */ if (!list_empty(&conn->mgmtqueue)) goto check_mgmt; + if (!list_empty(&conn->requeue)) + goto check_requeue; } - while (!list_empty(&conn->requeue)) { - /* - * we always do fastlogout - conn stop code will clean up. - */ - if (conn->session->state == ISCSI_STATE_LOGGING_OUT) - break; - - task = list_entry(conn->requeue.next, struct iscsi_task, - running); - - if (iscsi_check_tmf_restrictions(task, ISCSI_OP_SCSI_DATA_OUT)) - break; - - list_del_init(&task->running); - rc = iscsi_xmit_task(conn, task, true); - if (rc) - goto done; - if (!list_empty(&conn->mgmtqueue)) - goto check_mgmt; - } spin_unlock_bh(&conn->session->frwd_lock); return -ENODATA; -- cgit v1.2.3 From 31500e902759322ba3c64b60dabae2704e738df8 Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:27:38 -0500 Subject: scsi: iscsi: Fix session removal on shutdown When the system is shutting down, iscsid is not running so we will not get a response to the ISCSI_ERR_INVALID_HOST error event. The system shutdown will then hang waiting on userspace to remove the session. This has libiscsi force the destruction of the session from the kernel when iscsi_host_remove() is called from a driver's shutdown callout. This fixes a regression added in qedi boot with commit d1f2ce77638d ("scsi: qedi: Fix host removal with running sessions") which made qedi use the common session removal function that waits on userspace instead of rolling its own kernel based removal. Link: https://lore.kernel.org/r/20220616222738.5722-7-michael.christie@oracle.com Fixes: d1f2ce77638d ("scsi: qedi: Fix host removal with running sessions") Tested-by: Nilesh Javali Reviewed-by: Lee Duncan Reviewed-by: Nilesh Javali Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/infiniband/ulp/iser/iscsi_iser.c | 4 ++-- drivers/scsi/be2iscsi/be_main.c | 2 +- drivers/scsi/bnx2i/bnx2i_iscsi.c | 2 +- drivers/scsi/cxgbi/libcxgbi.c | 2 +- drivers/scsi/iscsi_tcp.c | 4 ++-- drivers/scsi/libiscsi.c | 9 +++++++-- drivers/scsi/qedi/qedi_main.c | 9 ++++++--- include/scsi/libiscsi.h | 2 +- 8 files changed, 21 insertions(+), 13 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/infiniband/ulp/iser/iscsi_iser.c b/drivers/infiniband/ulp/iser/iscsi_iser.c index 321949a570ed..620ae5b2d80d 100644 --- a/drivers/infiniband/ulp/iser/iscsi_iser.c +++ b/drivers/infiniband/ulp/iser/iscsi_iser.c @@ -568,7 +568,7 @@ static void iscsi_iser_session_destroy(struct iscsi_cls_session *cls_session) struct Scsi_Host *shost = iscsi_session_to_shost(cls_session); iscsi_session_teardown(cls_session); - iscsi_host_remove(shost); + iscsi_host_remove(shost, false); iscsi_host_free(shost); } @@ -685,7 +685,7 @@ iscsi_iser_session_create(struct iscsi_endpoint *ep, return cls_session; remove_host: - iscsi_host_remove(shost); + iscsi_host_remove(shost, false); free_host: iscsi_host_free(shost); return NULL; diff --git a/drivers/scsi/be2iscsi/be_main.c b/drivers/scsi/be2iscsi/be_main.c index 3bb0adefbe06..02026476c39c 100644 --- a/drivers/scsi/be2iscsi/be_main.c +++ b/drivers/scsi/be2iscsi/be_main.c @@ -5745,7 +5745,7 @@ static void beiscsi_remove(struct pci_dev *pcidev) cancel_work_sync(&phba->sess_work); beiscsi_iface_destroy_default(phba); - iscsi_host_remove(phba->shost); + iscsi_host_remove(phba->shost, false); beiscsi_disable_port(phba, 1); /* after cancelling boot_work */ diff --git a/drivers/scsi/bnx2i/bnx2i_iscsi.c b/drivers/scsi/bnx2i/bnx2i_iscsi.c index 15fbd09baa94..a3c800e04a2e 100644 --- a/drivers/scsi/bnx2i/bnx2i_iscsi.c +++ b/drivers/scsi/bnx2i/bnx2i_iscsi.c @@ -909,7 +909,7 @@ void bnx2i_free_hba(struct bnx2i_hba *hba) { struct Scsi_Host *shost = hba->shost; - iscsi_host_remove(shost); + iscsi_host_remove(shost, false); INIT_LIST_HEAD(&hba->ep_ofld_list); INIT_LIST_HEAD(&hba->ep_active_list); INIT_LIST_HEAD(&hba->ep_destroy_list); diff --git a/drivers/scsi/cxgbi/libcxgbi.c b/drivers/scsi/cxgbi/libcxgbi.c index 4365d52c6430..32abdf0fa9aa 100644 --- a/drivers/scsi/cxgbi/libcxgbi.c +++ b/drivers/scsi/cxgbi/libcxgbi.c @@ -328,7 +328,7 @@ void cxgbi_hbas_remove(struct cxgbi_device *cdev) chba = cdev->hbas[i]; if (chba) { cdev->hbas[i] = NULL; - iscsi_host_remove(chba->shost); + iscsi_host_remove(chba->shost, false); pci_dev_put(cdev->pdev); iscsi_host_free(chba->shost); } diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c index 9fee70d6434a..52c6f70d60ec 100644 --- a/drivers/scsi/iscsi_tcp.c +++ b/drivers/scsi/iscsi_tcp.c @@ -898,7 +898,7 @@ iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max, remove_session: iscsi_session_teardown(cls_session); remove_host: - iscsi_host_remove(shost); + iscsi_host_remove(shost, false); free_host: iscsi_host_free(shost); return NULL; @@ -915,7 +915,7 @@ static void iscsi_sw_tcp_session_destroy(struct iscsi_cls_session *cls_session) iscsi_tcp_r2tpool_free(cls_session->dd_data); iscsi_session_teardown(cls_session); - iscsi_host_remove(shost); + iscsi_host_remove(shost, false); iscsi_host_free(shost); } diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 8d78559ae94a..3894decf8fbf 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -2832,11 +2832,12 @@ static void iscsi_notify_host_removed(struct iscsi_cls_session *cls_session) /** * iscsi_host_remove - remove host and sessions * @shost: scsi host + * @is_shutdown: true if called from a driver shutdown callout * * If there are any sessions left, this will initiate the removal and wait * for the completion. */ -void iscsi_host_remove(struct Scsi_Host *shost) +void iscsi_host_remove(struct Scsi_Host *shost, bool is_shutdown) { struct iscsi_host *ihost = shost_priv(shost); unsigned long flags; @@ -2845,7 +2846,11 @@ void iscsi_host_remove(struct Scsi_Host *shost) ihost->state = ISCSI_HOST_REMOVED; spin_unlock_irqrestore(&ihost->lock, flags); - iscsi_host_for_each_session(shost, iscsi_notify_host_removed); + if (!is_shutdown) + iscsi_host_for_each_session(shost, iscsi_notify_host_removed); + else + iscsi_host_for_each_session(shost, iscsi_force_destroy_session); + wait_event_interruptible(ihost->session_removal_wq, ihost->num_sessions == 0); if (signal_pending(current)) diff --git a/drivers/scsi/qedi/qedi_main.c b/drivers/scsi/qedi/qedi_main.c index deebe62e2b41..cecfb2cb4c7b 100644 --- a/drivers/scsi/qedi/qedi_main.c +++ b/drivers/scsi/qedi/qedi_main.c @@ -2414,9 +2414,12 @@ static void __qedi_remove(struct pci_dev *pdev, int mode) int rval; u16 retry = 10; - if (mode == QEDI_MODE_NORMAL || mode == QEDI_MODE_SHUTDOWN) { - iscsi_host_remove(qedi->shost); + if (mode == QEDI_MODE_NORMAL) + iscsi_host_remove(qedi->shost, false); + else if (mode == QEDI_MODE_SHUTDOWN) + iscsi_host_remove(qedi->shost, true); + if (mode == QEDI_MODE_NORMAL || mode == QEDI_MODE_SHUTDOWN) { if (qedi->tmf_thread) { destroy_workqueue(qedi->tmf_thread); qedi->tmf_thread = NULL; @@ -2791,7 +2794,7 @@ remove_host: #ifdef CONFIG_DEBUG_FS qedi_dbg_host_exit(&qedi->dbg_ctx); #endif - iscsi_host_remove(qedi->shost); + iscsi_host_remove(qedi->shost, false); stop_iscsi_func: qedi_ops->stop(qedi->cdev); stop_slowpath: diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index c0703cd20a99..9758a4a9923f 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -411,7 +411,7 @@ extern int iscsi_host_add(struct Scsi_Host *shost, struct device *pdev); extern struct Scsi_Host *iscsi_host_alloc(struct scsi_host_template *sht, int dd_data_size, bool xmit_can_sleep); -extern void iscsi_host_remove(struct Scsi_Host *shost); +extern void iscsi_host_remove(struct Scsi_Host *shost, bool is_shutdown); extern void iscsi_host_free(struct Scsi_Host *shost); extern int iscsi_target_alloc(struct scsi_target *starget); extern int iscsi_host_get_max_scsi_cmds(struct Scsi_Host *shost, -- cgit v1.2.3 From 4b9f8ce4d5e823e42944c5a0a4842b0f936365ad Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:45:49 -0500 Subject: scsi: iscsi: Rename iscsi_conn_queue_work() Rename iscsi_conn_queue_work() to iscsi_conn_queue_xmit() to reflect that it handles queueing of xmits only. Link: https://lore.kernel.org/r/20220616224557.115234-2-michael.christie@oracle.com Reviewed-by: Lee Duncan Reviewed-by: Wu Bo Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/scsi/cxgbi/libcxgbi.c | 2 +- drivers/scsi/iscsi_tcp.c | 2 +- drivers/scsi/libiscsi.c | 12 ++++++------ include/scsi/libiscsi.h | 2 +- 4 files changed, 9 insertions(+), 9 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/cxgbi/libcxgbi.c b/drivers/scsi/cxgbi/libcxgbi.c index 32abdf0fa9aa..af281e271f88 100644 --- a/drivers/scsi/cxgbi/libcxgbi.c +++ b/drivers/scsi/cxgbi/libcxgbi.c @@ -1455,7 +1455,7 @@ void cxgbi_conn_tx_open(struct cxgbi_sock *csk) if (conn) { log_debug(1 << CXGBI_DBG_SOCK, "csk 0x%p, cid %d.\n", csk, conn->id); - iscsi_conn_queue_work(conn); + iscsi_conn_queue_xmit(conn); } } EXPORT_SYMBOL_GPL(cxgbi_conn_tx_open); diff --git a/drivers/scsi/iscsi_tcp.c b/drivers/scsi/iscsi_tcp.c index 52c6f70d60ec..da1dc345b873 100644 --- a/drivers/scsi/iscsi_tcp.c +++ b/drivers/scsi/iscsi_tcp.c @@ -205,7 +205,7 @@ static void iscsi_sw_tcp_write_space(struct sock *sk) old_write_space(sk); ISCSI_SW_TCP_DBG(conn, "iscsi_write_space\n"); - iscsi_conn_queue_work(conn); + iscsi_conn_queue_xmit(conn); } static void iscsi_sw_tcp_conn_set_callbacks(struct iscsi_conn *conn) diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 3894decf8fbf..a6859805d530 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -83,7 +83,7 @@ MODULE_PARM_DESC(debug_libiscsi_eh, "%s " dbg_fmt, __func__, ##arg); \ } while (0); -inline void iscsi_conn_queue_work(struct iscsi_conn *conn) +inline void iscsi_conn_queue_xmit(struct iscsi_conn *conn) { struct Scsi_Host *shost = conn->session->host; struct iscsi_host *ihost = shost_priv(shost); @@ -91,7 +91,7 @@ inline void iscsi_conn_queue_work(struct iscsi_conn *conn) if (ihost->workq) queue_work(ihost->workq, &conn->xmitwork); } -EXPORT_SYMBOL_GPL(iscsi_conn_queue_work); +EXPORT_SYMBOL_GPL(iscsi_conn_queue_xmit); static void __iscsi_update_cmdsn(struct iscsi_session *session, uint32_t exp_cmdsn, uint32_t max_cmdsn) @@ -765,7 +765,7 @@ __iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr, goto free_task; } else { list_add_tail(&task->running, &conn->mgmtqueue); - iscsi_conn_queue_work(conn); + iscsi_conn_queue_xmit(conn); } return task; @@ -1513,7 +1513,7 @@ void iscsi_requeue_task(struct iscsi_task *task) */ iscsi_put_task(task); } - iscsi_conn_queue_work(conn); + iscsi_conn_queue_xmit(conn); spin_unlock_bh(&conn->session->frwd_lock); } EXPORT_SYMBOL_GPL(iscsi_requeue_task); @@ -1786,7 +1786,7 @@ int iscsi_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc) } } else { list_add_tail(&task->running, &conn->cmdqueue); - iscsi_conn_queue_work(conn); + iscsi_conn_queue_xmit(conn); } session->queued_cmdsn++; @@ -1967,7 +1967,7 @@ EXPORT_SYMBOL_GPL(iscsi_suspend_tx); static void iscsi_start_tx(struct iscsi_conn *conn) { clear_bit(ISCSI_CONN_FLAG_SUSPEND_TX, &conn->flags); - iscsi_conn_queue_work(conn); + iscsi_conn_queue_xmit(conn); } /* diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index 9758a4a9923f..51fb0c17815e 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -453,7 +453,7 @@ extern int iscsi_conn_get_addr_param(struct sockaddr_storage *addr, enum iscsi_param param, char *buf); extern void iscsi_suspend_tx(struct iscsi_conn *conn); extern void iscsi_suspend_queue(struct iscsi_conn *conn); -extern void iscsi_conn_queue_work(struct iscsi_conn *conn); +extern void iscsi_conn_queue_xmit(struct iscsi_conn *conn); #define iscsi_conn_printk(prefix, _c, fmt, a...) \ iscsi_cls_conn_printk(prefix, ((struct iscsi_conn *)_c)->cls_conn, \ -- cgit v1.2.3 From 8af809966c0b34cfacd8da9a412689b8e9910354 Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:45:50 -0500 Subject: scsi: iscsi: Add recv workqueue helpers Add helpers to allow the drivers to run their recv paths from libiscsi's workqueue. Link: https://lore.kernel.org/r/20220616224557.115234-3-michael.christie@oracle.com Reviewed-by: Lee Duncan Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/scsi/libiscsi.c | 29 +++++++++++++++++++++++++++-- include/scsi/libiscsi.h | 4 ++++ 2 files changed, 31 insertions(+), 2 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index a6859805d530..20156fb33646 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -93,6 +93,16 @@ inline void iscsi_conn_queue_xmit(struct iscsi_conn *conn) } EXPORT_SYMBOL_GPL(iscsi_conn_queue_xmit); +inline void iscsi_conn_queue_recv(struct iscsi_conn *conn) +{ + struct Scsi_Host *shost = conn->session->host; + struct iscsi_host *ihost = shost_priv(shost); + + if (ihost->workq && !test_bit(ISCSI_CONN_FLAG_SUSPEND_RX, &conn->flags)) + queue_work(ihost->workq, &conn->recvwork); +} +EXPORT_SYMBOL_GPL(iscsi_conn_queue_recv); + static void __iscsi_update_cmdsn(struct iscsi_session *session, uint32_t exp_cmdsn, uint32_t max_cmdsn) { @@ -1947,7 +1957,7 @@ EXPORT_SYMBOL_GPL(iscsi_suspend_queue); /** * iscsi_suspend_tx - suspend iscsi_data_xmit - * @conn: iscsi conn tp stop processing IO on. + * @conn: iscsi conn to stop processing IO on. * * This function sets the suspend bit to prevent iscsi_data_xmit * from sending new IO, and if work is queued on the xmit thread @@ -1960,7 +1970,7 @@ void iscsi_suspend_tx(struct iscsi_conn *conn) set_bit(ISCSI_CONN_FLAG_SUSPEND_TX, &conn->flags); if (ihost->workq) - flush_workqueue(ihost->workq); + flush_work(&conn->xmitwork); } EXPORT_SYMBOL_GPL(iscsi_suspend_tx); @@ -1970,6 +1980,21 @@ static void iscsi_start_tx(struct iscsi_conn *conn) iscsi_conn_queue_xmit(conn); } +/** + * iscsi_suspend_rx - Prevent recvwork from running again. + * @conn: iscsi conn to stop. + */ +void iscsi_suspend_rx(struct iscsi_conn *conn) +{ + struct Scsi_Host *shost = conn->session->host; + struct iscsi_host *ihost = shost_priv(shost); + + set_bit(ISCSI_CONN_FLAG_SUSPEND_RX, &conn->flags); + if (ihost->workq) + flush_work(&conn->recvwork); +} +EXPORT_SYMBOL_GPL(iscsi_suspend_rx); + /* * We want to make sure a ping is in flight. It has timed out. * And we are not busy processing a pdu that is making diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index 51fb0c17815e..5a10e5acfad2 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -213,6 +213,8 @@ struct iscsi_conn { struct list_head cmdqueue; /* data-path cmd queue */ struct list_head requeue; /* tasks needing another run */ struct work_struct xmitwork; /* per-conn. xmit workqueue */ + /* recv */ + struct work_struct recvwork; unsigned long flags; /* ISCSI_CONN_FLAGs */ /* negotiated params */ @@ -452,8 +454,10 @@ extern int iscsi_conn_get_param(struct iscsi_cls_conn *cls_conn, extern int iscsi_conn_get_addr_param(struct sockaddr_storage *addr, enum iscsi_param param, char *buf); extern void iscsi_suspend_tx(struct iscsi_conn *conn); +extern void iscsi_suspend_rx(struct iscsi_conn *conn); extern void iscsi_suspend_queue(struct iscsi_conn *conn); extern void iscsi_conn_queue_xmit(struct iscsi_conn *conn); +extern void iscsi_conn_queue_recv(struct iscsi_conn *conn); #define iscsi_conn_printk(prefix, _c, fmt, a...) \ iscsi_cls_conn_printk(prefix, ((struct iscsi_conn *)_c)->cls_conn, \ -- cgit v1.2.3 From 533ac412fdb4cbc8ce282b1ee1991e03b2717625 Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:45:54 -0500 Subject: scsi: iscsi: Remove unneeded task state check Commit 5923d64b7ab6 ("scsi: libiscsi: Drop taskqueuelock") added an extra task->state because for commit 6f8830f5bbab ("scsi: libiscsi: add lock around task lists to fix list corruption regression") we didn't know why we ended up with cmds on the list and thought it might have been a bad target sending a response while we were still sending the cmd. We were never able to get a target to send us a response early, because it turns out the bug was just a race in libiscsi/libiscsi_tcp where: 1. iscsi_tcp_r2t_rsp() queues a r2t to tcp_task->r2tqueue. 2. iscsi_tcp_task_xmit() runs iscsi_tcp_get_curr_r2t() and sees we have a r2t. It dequeues it and iscsi_tcp_task_xmit() starts to process it. 3. iscsi_tcp_r2t_rsp() runs iscsi_requeue_task() and puts the task on the requeue list. 4. iscsi_tcp_task_xmit() sends the data for r2t. This is the final chunk of data, so the cmd is done. 5. target sends the response. 6. On a different CPU from #3, iscsi_complete_task() processes the response. Since there was no common lock for the list, the lists/tasks pointers are not fully in sync, so could end up with list corruption. Since it was just a race on our side, remove the extra check and fix up the comments. Link: https://lore.kernel.org/r/20220616224557.115234-7-michael.christie@oracle.com Reviewed-by: Wu Bo Reviewed-by: Lee Duncan Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/scsi/libiscsi.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 20156fb33646..fce9f9e5b00b 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -567,16 +567,19 @@ static bool cleanup_queued_task(struct iscsi_task *task) struct iscsi_conn *conn = task->conn; bool early_complete = false; - /* Bad target might have completed task while it was still running */ + /* + * We might have raced where we handled a R2T early and got a response + * but have not yet taken the task off the requeue list, then a TMF or + * recovery happened and so we can still see it here. + */ if (task->state == ISCSI_TASK_COMPLETED) early_complete = true; if (!list_empty(&task->running)) { list_del_init(&task->running); /* - * If it's on a list but still running, this could be from - * a bad target sending a rsp early, cleanup from a TMF, or - * session recovery. + * If it's on a list but still running this could be cleanup + * from a TMF or session recovery. */ if (task->state == ISCSI_TASK_RUNNING || task->state == ISCSI_TASK_COMPLETED) @@ -1485,7 +1488,7 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, } /* regular RX path uses back_lock */ spin_lock(&conn->session->back_lock); - if (rc && task->state == ISCSI_TASK_RUNNING) { + if (rc) { /* * get an extra ref that is released next time we access it * as conn->task above. -- cgit v1.2.3 From e1c6a7ec14290a0e371b09300685638f9009f2ab Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:45:55 -0500 Subject: scsi: iscsi: Remove iscsi_get_task back_lock requirement We currently require that the back_lock is held when calling the functions that manipulate the iscsi_task refcount. The only reason for this is to handle races where we are handling SCSI-ml EH callbacks and the cmd is completing at the same time the normal completion path is running, and we can't return from the EH callback until the driver has stopped accessing the cmd. Holding the back_lock while also accessing the task->state made it simple to check that a cmd is completing and also get/put a refcount at the same time, and at the time we were not as concerned about performance. The problem is that we don't want to take the back_lock from the xmit path for normal I/O since it causes contention with the completion path if the user has chosen to try and split those paths on different CPUs (in this case abusing the CPUs and ignoring caching improves perf for some uses). Begins to remove the back_lock requirement for iscsi_get/put_task by removing the requirement for the get path. Instead of always holding the back_lock we detect if something has done the last put and is about to call iscsi_free_task(). A subsequent commit will then allow iSCSI code to do the last put on a task and only grab the back_lock if the refcount is now zero and it's going to call iscsi_free_task(). Link: https://lore.kernel.org/r/20220616224557.115234-8-michael.christie@oracle.com Reviewed-by: Lee Duncan Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/scsi/be2iscsi/be_main.c | 19 ++++++++- drivers/scsi/libiscsi.c | 95 ++++++++++++++++++++++++++++------------- drivers/scsi/libiscsi_tcp.c | 6 ++- include/scsi/libiscsi.h | 2 +- 4 files changed, 89 insertions(+), 33 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/be2iscsi/be_main.c b/drivers/scsi/be2iscsi/be_main.c index 02026476c39c..50a577ac3bb4 100644 --- a/drivers/scsi/be2iscsi/be_main.c +++ b/drivers/scsi/be2iscsi/be_main.c @@ -231,6 +231,7 @@ static int beiscsi_eh_abort(struct scsi_cmnd *sc) cls_session = starget_to_session(scsi_target(sc->device)); session = cls_session->dd_data; +completion_check: /* check if we raced, task just got cleaned up under us */ spin_lock_bh(&session->back_lock); if (!abrt_task || !abrt_task->sc) { @@ -238,7 +239,13 @@ static int beiscsi_eh_abort(struct scsi_cmnd *sc) return SUCCESS; } /* get a task ref till FW processes the req for the ICD used */ - __iscsi_get_task(abrt_task); + if (!iscsi_get_task(abrt_task)) { + spin_unlock(&session->back_lock); + /* We are just about to call iscsi_free_task so wait for it. */ + udelay(5); + goto completion_check; + } + abrt_io_task = abrt_task->dd_data; conn = abrt_task->conn; beiscsi_conn = conn->dd_data; @@ -323,7 +330,15 @@ static int beiscsi_eh_device_reset(struct scsi_cmnd *sc) } /* get a task ref till FW processes the req for the ICD used */ - __iscsi_get_task(task); + if (!iscsi_get_task(task)) { + /* + * The task has completed in the driver and is + * completing in libiscsi. Just ignore it here. When we + * call iscsi_eh_device_reset, it will wait for us. + */ + continue; + } + io_task = task->dd_data; /* mark WRB invalid which have been not processed by FW yet */ if (is_chip_be2_be3r(phba)) { diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index fce9f9e5b00b..3a6b827496dd 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -83,6 +83,8 @@ MODULE_PARM_DESC(debug_libiscsi_eh, "%s " dbg_fmt, __func__, ##arg); \ } while (0); +#define ISCSI_CMD_COMPL_WAIT 5 + inline void iscsi_conn_queue_xmit(struct iscsi_conn *conn) { struct Scsi_Host *shost = conn->session->host; @@ -482,11 +484,11 @@ static void iscsi_free_task(struct iscsi_task *task) } } -void __iscsi_get_task(struct iscsi_task *task) +bool iscsi_get_task(struct iscsi_task *task) { - refcount_inc(&task->refcount); + return refcount_inc_not_zero(&task->refcount); } -EXPORT_SYMBOL_GPL(__iscsi_get_task); +EXPORT_SYMBOL_GPL(iscsi_get_task); void __iscsi_put_task(struct iscsi_task *task) { @@ -600,20 +602,17 @@ static bool cleanup_queued_task(struct iscsi_task *task) } /* - * session frwd lock must be held and if not called for a task that is still - * pending or from the xmit thread, then xmit thread must be suspended + * session back and frwd lock must be held and if not called for a task that + * is still pending or from the xmit thread, then xmit thread must be suspended */ -static void fail_scsi_task(struct iscsi_task *task, int err) +static void __fail_scsi_task(struct iscsi_task *task, int err) { struct iscsi_conn *conn = task->conn; struct scsi_cmnd *sc; int state; - spin_lock_bh(&conn->session->back_lock); - if (cleanup_queued_task(task)) { - spin_unlock_bh(&conn->session->back_lock); + if (cleanup_queued_task(task)) return; - } if (task->state == ISCSI_TASK_PENDING) { /* @@ -632,7 +631,15 @@ static void fail_scsi_task(struct iscsi_task *task, int err) sc->result = err << 16; scsi_set_resid(sc, scsi_bufflen(sc)); iscsi_complete_task(task, state); - spin_unlock_bh(&conn->session->back_lock); +} + +static void fail_scsi_task(struct iscsi_task *task, int err) +{ + struct iscsi_session *session = task->conn->session; + + spin_lock_bh(&session->back_lock); + __fail_scsi_task(task, err); + spin_unlock_bh(&session->back_lock); } static int iscsi_prep_mgmt_task(struct iscsi_conn *conn, @@ -1450,8 +1457,17 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, spin_lock_bh(&conn->session->back_lock); if (!conn->task) { - /* Take a ref so we can access it after xmit_task() */ - __iscsi_get_task(task); + /* + * Take a ref so we can access it after xmit_task(). + * + * This should never fail because the failure paths will have + * stopped the xmit thread. + */ + if (!iscsi_get_task(task)) { + spin_unlock_bh(&conn->session->back_lock); + WARN_ON_ONCE(1); + return 0; + } } else { /* Already have a ref from when we failed to send it last call */ conn->task = NULL; @@ -1493,7 +1509,7 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, * get an extra ref that is released next time we access it * as conn->task above. */ - __iscsi_get_task(task); + iscsi_get_task(task); conn->task = task; } @@ -1912,6 +1928,7 @@ static void fail_scsi_tasks(struct iscsi_conn *conn, u64 lun, int error) struct iscsi_task *task; int i; +restart_cmd_loop: spin_lock_bh(&session->back_lock); for (i = 0; i < session->cmds_max; i++) { task = session->cmds[i]; @@ -1920,22 +1937,25 @@ static void fail_scsi_tasks(struct iscsi_conn *conn, u64 lun, int error) if (lun != -1 && lun != task->sc->device->lun) continue; - - __iscsi_get_task(task); - spin_unlock_bh(&session->back_lock); + /* + * The cmd is completing but if this is called from an eh + * callout path then when we return scsi-ml owns the cmd. Wait + * for the completion path to finish freeing the cmd. + */ + if (!iscsi_get_task(task)) { + spin_unlock_bh(&session->back_lock); + spin_unlock_bh(&session->frwd_lock); + udelay(ISCSI_CMD_COMPL_WAIT); + spin_lock_bh(&session->frwd_lock); + goto restart_cmd_loop; + } ISCSI_DBG_SESSION(session, "failing sc %p itt 0x%x state %d\n", task->sc, task->itt, task->state); - fail_scsi_task(task, error); - - spin_unlock_bh(&session->frwd_lock); - iscsi_put_task(task); - spin_lock_bh(&session->frwd_lock); - - spin_lock_bh(&session->back_lock); + __fail_scsi_task(task, error); + __iscsi_put_task(task); } - spin_unlock_bh(&session->back_lock); } @@ -2040,7 +2060,16 @@ enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd *sc) spin_unlock(&session->back_lock); goto done; } - __iscsi_get_task(task); + if (!iscsi_get_task(task)) { + /* + * Racing with the completion path right now, so give it more + * time so that path can complete it like normal. + */ + rc = BLK_EH_RESET_TIMER; + task = NULL; + spin_unlock(&session->back_lock); + goto done; + } spin_unlock(&session->back_lock); if (session->state != ISCSI_STATE_LOGGED_IN) { @@ -2289,6 +2318,7 @@ int iscsi_eh_abort(struct scsi_cmnd *sc) ISCSI_DBG_EH(session, "aborting sc %p\n", sc); +completion_check: mutex_lock(&session->eh_mutex); spin_lock_bh(&session->frwd_lock); /* @@ -2328,13 +2358,20 @@ int iscsi_eh_abort(struct scsi_cmnd *sc) return SUCCESS; } + if (!iscsi_get_task(task)) { + spin_unlock(&session->back_lock); + spin_unlock_bh(&session->frwd_lock); + mutex_unlock(&session->eh_mutex); + /* We are just about to call iscsi_free_task so wait for it. */ + udelay(ISCSI_CMD_COMPL_WAIT); + goto completion_check; + } + + ISCSI_DBG_EH(session, "aborting [sc %p itt 0x%x]\n", sc, task->itt); conn = session->leadconn; iscsi_get_conn(conn->cls_conn); conn->eh_abort_cnt++; age = session->age; - - ISCSI_DBG_EH(session, "aborting [sc %p itt 0x%x]\n", sc, task->itt); - __iscsi_get_task(task); spin_unlock(&session->back_lock); if (task->state == ISCSI_TASK_PENDING) { diff --git a/drivers/scsi/libiscsi_tcp.c b/drivers/scsi/libiscsi_tcp.c index 883005757ddb..c182aa83f2c9 100644 --- a/drivers/scsi/libiscsi_tcp.c +++ b/drivers/scsi/libiscsi_tcp.c @@ -558,7 +558,11 @@ static int iscsi_tcp_r2t_rsp(struct iscsi_conn *conn, struct iscsi_hdr *hdr) return 0; } task->last_xfer = jiffies; - __iscsi_get_task(task); + if (!iscsi_get_task(task)) { + spin_unlock(&session->back_lock); + /* Let the path that got the early rsp complete it */ + return 0; + } tcp_conn = conn->dd_data; rhdr = (struct iscsi_r2t_rsp *)tcp_conn->in.hdr; diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index 5a10e5acfad2..ec15e820f62b 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -482,7 +482,7 @@ extern struct iscsi_task *iscsi_itt_to_task(struct iscsi_conn *, itt_t); extern void iscsi_requeue_task(struct iscsi_task *task); extern void iscsi_put_task(struct iscsi_task *task); extern void __iscsi_put_task(struct iscsi_task *task); -extern void __iscsi_get_task(struct iscsi_task *task); +extern bool iscsi_get_task(struct iscsi_task *task); extern void iscsi_complete_scsi_task(struct iscsi_task *task, uint32_t exp_cmdsn, uint32_t max_cmdsn); -- cgit v1.2.3 From 6d626150d6d1c14f2f3d4f907fd0d4381f2c0d35 Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:45:56 -0500 Subject: scsi: iscsi: Try to avoid taking back_lock in xmit path We need the back lock when freeing a task, so we hold it when calling __iscsi_put_task() from the completion path to make it easier and to avoid having to retake it in that path. For iscsi_put_task() we just grabbed it while also doing the decrement on the refcount but it's only really needed if the refcount is zero and we free the task. This modifies iscsi_put_task() to just take the lock when needed then has the xmit path use it. Normally we will then not take the back lock from the xmit path. It will only be rare cases where the network is so fast that we get a response right after we send the header/data. Link: https://lore.kernel.org/r/20220616224557.115234-9-michael.christie@oracle.com Reviewed-by: Lee Duncan Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/scsi/libiscsi.c | 30 ++++++++++++++---------------- 1 file changed, 14 insertions(+), 16 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 3a6b827496dd..163e914084e0 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -490,6 +490,12 @@ bool iscsi_get_task(struct iscsi_task *task) } EXPORT_SYMBOL_GPL(iscsi_get_task); +/** + * __iscsi_put_task - drop the refcount on a task + * @task: iscsi_task to drop the refcount on + * + * The back_lock must be held when calling in case it frees the task. + */ void __iscsi_put_task(struct iscsi_task *task) { if (refcount_dec_and_test(&task->refcount)) @@ -501,10 +507,11 @@ void iscsi_put_task(struct iscsi_task *task) { struct iscsi_session *session = task->conn->session; - /* regular RX path uses back_lock */ - spin_lock_bh(&session->back_lock); - __iscsi_put_task(task); - spin_unlock_bh(&session->back_lock); + if (refcount_dec_and_test(&task->refcount)) { + spin_lock_bh(&session->back_lock); + iscsi_free_task(task); + spin_unlock_bh(&session->back_lock); + } } EXPORT_SYMBOL_GPL(iscsi_put_task); @@ -1454,8 +1461,6 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, { int rc; - spin_lock_bh(&conn->session->back_lock); - if (!conn->task) { /* * Take a ref so we can access it after xmit_task(). @@ -1464,7 +1469,6 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, * stopped the xmit thread. */ if (!iscsi_get_task(task)) { - spin_unlock_bh(&conn->session->back_lock); WARN_ON_ONCE(1); return 0; } @@ -1478,7 +1482,7 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, * case a bad target sends a cmd rsp before we have handled the task. */ if (was_requeue) - __iscsi_put_task(task); + iscsi_put_task(task); /* * Do this after dropping the extra ref because if this was a requeue @@ -1490,10 +1494,8 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, * task and get woken up again. */ conn->task = task; - spin_unlock_bh(&conn->session->back_lock); return -ENODATA; } - spin_unlock_bh(&conn->session->back_lock); spin_unlock_bh(&conn->session->frwd_lock); rc = conn->session->tt->xmit_task(task); @@ -1501,10 +1503,7 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, if (!rc) { /* done with this task */ task->last_xfer = jiffies; - } - /* regular RX path uses back_lock */ - spin_lock(&conn->session->back_lock); - if (rc) { + } else { /* * get an extra ref that is released next time we access it * as conn->task above. @@ -1513,8 +1512,7 @@ static int iscsi_xmit_task(struct iscsi_conn *conn, struct iscsi_task *task, conn->task = task; } - __iscsi_put_task(task); - spin_unlock(&conn->session->back_lock); + iscsi_put_task(task); return rc; } -- cgit v1.2.3 From 6e637b723d8277cac4d347ea671ab798b70d8fc9 Mon Sep 17 00:00:00 2001 From: Mike Christie Date: Thu, 16 Jun 2022 17:45:57 -0500 Subject: scsi: libiscsi: Improve conn_send_pdu API The conn_send_pdu API is evil in that it returns a pointer to an iscsi_task, but that task might have been freed already so you can't touch it. This patch splits the task allocation and transmission, so functions like iscsi_send_nopout() can access the task before its sent and do whatever bookkeeping is needed before it is sent. Link: https://lore.kernel.org/r/20220616224557.115234-10-michael.christie@oracle.com Reviewed-by: Lee Duncan Signed-off-by: Mike Christie Signed-off-by: Martin K. Petersen --- drivers/scsi/libiscsi.c | 85 ++++++++++++++++++++++++++++++++++++------------- include/scsi/libiscsi.h | 3 -- 2 files changed, 62 insertions(+), 26 deletions(-) (limited to 'drivers/scsi/libiscsi.c') diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c index 163e914084e0..d95f4bcdeb2e 100644 --- a/drivers/scsi/libiscsi.c +++ b/drivers/scsi/libiscsi.c @@ -695,12 +695,18 @@ static int iscsi_prep_mgmt_task(struct iscsi_conn *conn, return 0; } +/** + * iscsi_alloc_mgmt_task - allocate and setup a mgmt task. + * @conn: iscsi conn that the task will be sent on. + * @hdr: iscsi pdu that will be sent. + * @data: buffer for data segment if needed. + * @data_size: length of data in bytes. + */ static struct iscsi_task * -__iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr, +iscsi_alloc_mgmt_task(struct iscsi_conn *conn, struct iscsi_hdr *hdr, char *data, uint32_t data_size) { struct iscsi_session *session = conn->session; - struct iscsi_host *ihost = shost_priv(session->host); uint8_t opcode = hdr->opcode & ISCSI_OPCODE_MASK; struct iscsi_task *task; itt_t itt; @@ -781,28 +787,57 @@ __iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr, task->conn->session->age); } - if (unlikely(READ_ONCE(conn->ping_task) == INVALID_SCSI_TASK)) - WRITE_ONCE(conn->ping_task, task); + return task; + +free_task: + iscsi_put_task(task); + return NULL; +} + +/** + * iscsi_send_mgmt_task - Send task created with iscsi_alloc_mgmt_task. + * @task: iscsi task to send. + * + * On failure this returns a non-zero error code, and the driver must free + * the task with iscsi_put_task; + */ +static int iscsi_send_mgmt_task(struct iscsi_task *task) +{ + struct iscsi_conn *conn = task->conn; + struct iscsi_session *session = conn->session; + struct iscsi_host *ihost = shost_priv(conn->session->host); + int rc = 0; if (!ihost->workq) { - if (iscsi_prep_mgmt_task(conn, task)) - goto free_task; + rc = iscsi_prep_mgmt_task(conn, task); + if (rc) + return rc; - if (session->tt->xmit_task(task)) - goto free_task; + rc = session->tt->xmit_task(task); + if (rc) + return rc; } else { list_add_tail(&task->running, &conn->mgmtqueue); iscsi_conn_queue_xmit(conn); } - return task; + return 0; +} -free_task: - /* regular RX path uses back_lock */ - spin_lock(&session->back_lock); - __iscsi_put_task(task); - spin_unlock(&session->back_lock); - return NULL; +static int __iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr, + char *data, uint32_t data_size) +{ + struct iscsi_task *task; + int rc; + + task = iscsi_alloc_mgmt_task(conn, hdr, data, data_size); + if (!task) + return -ENOMEM; + + rc = iscsi_send_mgmt_task(task); + if (rc) + iscsi_put_task(task); + return rc; } int iscsi_conn_send_pdu(struct iscsi_cls_conn *cls_conn, struct iscsi_hdr *hdr, @@ -813,7 +848,7 @@ int iscsi_conn_send_pdu(struct iscsi_cls_conn *cls_conn, struct iscsi_hdr *hdr, int err = 0; spin_lock_bh(&session->frwd_lock); - if (!__iscsi_conn_send_pdu(conn, hdr, data, data_size)) + if (__iscsi_conn_send_pdu(conn, hdr, data, data_size)) err = -EPERM; spin_unlock_bh(&session->frwd_lock); return err; @@ -986,7 +1021,6 @@ static int iscsi_send_nopout(struct iscsi_conn *conn, struct iscsi_nopin *rhdr) if (!rhdr) { if (READ_ONCE(conn->ping_task)) return -EINVAL; - WRITE_ONCE(conn->ping_task, INVALID_SCSI_TASK); } memset(&hdr, 0, sizeof(struct iscsi_nopout)); @@ -1000,10 +1034,18 @@ static int iscsi_send_nopout(struct iscsi_conn *conn, struct iscsi_nopin *rhdr) } else hdr.ttt = RESERVED_ITT; - task = __iscsi_conn_send_pdu(conn, (struct iscsi_hdr *)&hdr, NULL, 0); - if (!task) { + task = iscsi_alloc_mgmt_task(conn, (struct iscsi_hdr *)&hdr, NULL, 0); + if (!task) + return -ENOMEM; + + if (!rhdr) + WRITE_ONCE(conn->ping_task, task); + + if (iscsi_send_mgmt_task(task)) { if (!rhdr) WRITE_ONCE(conn->ping_task, NULL); + iscsi_put_task(task); + iscsi_conn_printk(KERN_ERR, conn, "Could not send nopout\n"); return -EIO; } else if (!rhdr) { @@ -1874,11 +1916,8 @@ static int iscsi_exec_task_mgmt_fn(struct iscsi_conn *conn, __must_hold(&session->frwd_lock) { struct iscsi_session *session = conn->session; - struct iscsi_task *task; - task = __iscsi_conn_send_pdu(conn, (struct iscsi_hdr *)hdr, - NULL, 0); - if (!task) { + if (__iscsi_conn_send_pdu(conn, (struct iscsi_hdr *)hdr, NULL, 0)) { spin_unlock_bh(&session->frwd_lock); iscsi_conn_printk(KERN_ERR, conn, "Could not send TMF.\n"); iscsi_conn_failure(conn, ISCSI_ERR_CONN_FAILED); diff --git a/include/scsi/libiscsi.h b/include/scsi/libiscsi.h index ec15e820f62b..654cc3918c94 100644 --- a/include/scsi/libiscsi.h +++ b/include/scsi/libiscsi.h @@ -135,9 +135,6 @@ struct iscsi_task { void *dd_data; /* driver/transport data */ }; -/* invalid scsi_task pointer */ -#define INVALID_SCSI_TASK (struct iscsi_task *)-1l - static inline int iscsi_task_has_unsol_data(struct iscsi_task *task) { return task->unsol_r2t.data_length > task->unsol_r2t.sent; -- cgit v1.2.3