iscsi fixes and cleanups
Mike Christie
maintainerv6 cover iscsi fixes and cleanups
The following patches made over Martin's 5.12 branches contain fixes for a cmd lifetime bug, software iscsi can_queue setup, and a couple of the lock cleanup patches Lee has already ackd.
V6:
- Remove task->sc check that is no longer needed because the helper function we use does the check for us.
V5:
- Fix up KERN_ERR/INFO use when detecting invalid max_cmds values from the user.
- Set iscsi_tcp can queue to max value it can support not including mgmt cmds since the driver itself is not limited and that is a libiscsi layer limit.
- Added the iscsi session class lock cleanup from the lock cleanup patchset since it was reviewed already and this is now a patchset for the next feature window.
V4:
- Add patch:
[PATCH 4/7] libiscsi: fix iscsi host workq destruction
to fix an issue where the user might only call iscsi_host_alloc then iscsi_host_free and that was leaving the iscsi workqueue running.
- Add check for if a driver were to set can_queue to ISCSI_MGMT_CMDS_MAX or less.
V3:
- Add some patches for issues found while testing.
- session queue depth was stuck at 128
- cmd window could not be detected when session was relogged in.
- Patch "libiscsi: drop taskqueuelock" had a bug where we did not disable bhs and during xmit thread suspension leaked the current task. V2:
- Take back_lock when looping over running cmds in iscsi_eh_cmd_timed_out in case those complete while we are accessing them.
Mike Christie
maintainerPATCH v6 1/9 libiscsi: fix iscsi_prep_scsi_cmd_pdu error handling
If iscsi_prep_scsi_cmd_pdu fails we try to add it back to the cmdqueue, but we leave it partially setup. We don't have functions that can undo the pdu and init task setup. We only have cleanup_task which can cleanup both parts. So this has us just fail the cmd and go through the standard cleanup routine and then have scsi-ml retry it like is done when it fails in the queuecommand path.
--- drivers/scsi/libiscsi.c | 11 +++-------- 1 file changed, 3 insertions(+), 8 deletions(-)
drivers/scsi/libiscsi.c +3 -8
@@ -1532,14 +1532,9 @@ static int iscsi_data_xmit(struct iscsi_conn *conn) } rc =iscsi_prep_scsi_cmd_pdu (conn->task);if (rc) { -if (rc == -ENOMEM || rc == -EACCES ) { -spin_lock_bh (&conn->taskqueuelock); -list_add_tail (&conn->task->running, - &conn->cmdqueue); - conn->task =NULL ; -spin_unlock_bh (&conn->taskqueuelock); - goto done; - }else +if (rc == -ENOMEM || rc == -EACCES ) +fail_scsi_task (conn->task,DID_IMM_RETRY ); +else fail_scsi_task (conn->task,DID_ABORT );spin_lock_bh (&conn->taskqueuelock);continue ;
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 2/9 libiscsi: drop taskqueuelock
In reply to Mike ChristieThe purpose of the taskqueuelock was to handle the issue where a bad target decides to send a R2T and before it's data has been sent decides to send a cmd response to complete the cmd. The following patches fix up the frwd/back locks so they are taken from the queue/xmit (frwd) and completion (back) paths again. To get there this patch removes the taskqueuelock which for iscsi xmit wq based drivers was taken in the queue, xmit and completion paths.
Instead of the lock, we just make sure we have a ref to the task when we queue a R2T, and then we always remove the task from the requeue list in the xmit path or the forced cleanup paths.
--- drivers/scsi/libiscsi.c | 178 +++++++++++++++++++++++------------- drivers/scsi/libiscsi_tcp.c | 86 +++++++++++------ include/scsi/libiscsi.h | 4 +- 3 files changed, 172 insertions(+), 96 deletions(-)
drivers/scsi/libiscsi.c +113 -65
@@ -523,16 +523,6 @@ static void iscsi_complete_task(struct iscsi_task *task, int state)WARN_ON_ONCE (task->state ==ISCSI_TASK_FREE ); task->state = state; -spin_lock_bh (&conn->taskqueuelock); -if (!list_empty (&task->running)) { -pr_debug_once ("%s while task on list" , __func__); -list_del_init (&task->running); - } -spin_unlock_bh (&conn->taskqueuelock); - -if (conn->task == task) - conn->task =NULL ; -if (READ_ONCE (conn->ping_task) == task)WRITE_ONCE (conn->ping_task,NULL );
@@ -564,9 +554,40 @@ void iscsi_complete_scsi_task(struct iscsi_task *task, }EXPORT_SYMBOL_GPL (iscsi_complete_scsi_task); +/* +* Must be called with back and frwd lock +*/ +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 */ +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 (task->state ==ISCSI_TASK_RUNNING || + task->state ==ISCSI_TASK_COMPLETED ) +__iscsi_put_task (task); + } + +if (conn->task == task) { + conn->task =NULL ; +__iscsi_put_task (task); + } + +return early_complete; +}/* -* session back_lock must be held and if not called for a task that is +* 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. */
@@ -585,6 +606,13 @@ static void fail_scsi_task(struct iscsi_task *task, int err)if (!sc)return ; +/* regular RX path uses back_lock */ +spin_lock_bh (&conn->session->back_lock); +if (cleanup_queued_task (task)) { +spin_unlock_bh (&conn->session->back_lock); +return ; + } +if (task->state ==ISCSI_TASK_PENDING ) {/* * cmd never made it to the xmit thread, so we should not count
@@ -600,9 +628,6 @@ static void fail_scsi_task(struct iscsi_task *task, int err) sc->result = err <<16 ;scsi_set_resid (sc,scsi_bufflen (sc)); - -/* regular RX path uses back_lock */ -spin_lock_bh (&conn->session->back_lock);iscsi_complete_task (task, state);spin_unlock_bh (&conn->session->back_lock); }
@@ -748,9 +773,7 @@ __iscsi_conn_send_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr,if (session->tt->xmit_task (task)) goto free_task; }else { -spin_lock_bh (&conn->taskqueuelock);list_add_tail (&task->running, &conn->mgmtqueue); -spin_unlock_bh (&conn->taskqueuelock);iscsi_conn_queue_work (conn); }
@@ -1411,31 +1434,61 @@ static int iscsi_check_cmdsn_window_closed(struct iscsi_conn *conn)return 0 ; } -static int iscsi_xmit_task (struct iscsi_conn *conn) +static int iscsi_xmit_task (struct iscsi_conn *conn,struct iscsi_task *task, +bool was_requeue) { -struct iscsi_task *task = conn->task;int rc; -if (test_bit (ISCSI_SUSPEND_BIT , &conn->suspend_tx)) -return -ENODATA ; -spin_lock_bh (&conn->session->back_lock); -if (conn->task ==NULL ) { + +if (!conn->task) { +/* Take a ref so we can access it after xmit_task() */ +__iscsi_get_task (task); + }else { +/* Already have a ref from when we failed to send it last call */ + conn->task =NULL ; + } + +/* +* If this was a requeue for a R2T we have an extra ref on the task in +* case a bad target sends a cmd rsp before we have handled the task. +*/ +if (was_requeue) +__iscsi_put_task (task); + +/* +* Do this after dropping the extra ref because if this was a requeue +* it's removed from that list and cleanup_queued_task would miss it. +*/ +if (test_bit (ISCSI_SUSPEND_BIT , &conn->suspend_tx)) { +/* +* Save the task and ref in case we weren't cleaning up this +* task and get woken up again. +*/ + conn->task = task;spin_unlock_bh (&conn->session->back_lock);return -ENODATA ; } -__iscsi_get_task (task);spin_unlock_bh (&conn->session->back_lock); +spin_unlock_bh (&conn->session->frwd_lock); rc = conn->session->tt->xmit_task (task);spin_lock_bh (&conn->session->frwd_lock);if (!rc) {/* done with this task */ task->last_xfer = jiffies; - conn->task =NULL ; }/* regular RX path uses back_lock */ spin_lock (&conn->session->back_lock); +if (rc && task->state ==ISCSI_TASK_RUNNING ) { +/* +* get an extra ref that is released next time we access it +* as conn->task above. +*/ +__iscsi_get_task (task); + conn->task = task; + } +__iscsi_put_task (task);spin_unlock (&conn->session->back_lock);return rc;
@@ -1445,9 +1498,7 @@ static int iscsi_xmit_task(struct iscsi_conn *conn)* iscsi_requeue_task - requeue task to run from session workqueue * @task: task to requeue * -* LLDs that need to run a task from the session workqueue should call -* this. The session frwd_lock must be held. This should only be called -* by software drivers. +* Callers must have taken a ref to the task that is going to be requeued. */ void iscsi_requeue_task (struct iscsi_task *task) {
@@ -1457,11 +1508,18 @@ void iscsi_requeue_task(struct iscsi_task *task)* this may be on the requeue list already if the xmit_task callout * is handling the r2ts while we are adding new ones */ -spin_lock_bh (&conn->taskqueuelock); -if (list_empty (&task->running)) +spin_lock_bh (&conn->session->frwd_lock); +if (list_empty (&task->running)) {list_add_tail (&task->running, &conn->requeue); -spin_unlock_bh (&conn->taskqueuelock); + }else { +/* +* Don't need the extra ref since it's already requeued and +* has a ref. +*/ +iscsi_put_task (task); + }iscsi_conn_queue_work (conn); +spin_unlock_bh (&conn->session->frwd_lock); }EXPORT_SYMBOL_GPL (iscsi_requeue_task);
@@ -1487,7 +1545,7 @@ static int iscsi_data_xmit(struct iscsi_conn *conn) }if (conn->task) { - rc =iscsi_xmit_task (conn); + rc =iscsi_xmit_task (conn, conn->task, false);if (rc) goto done; }
@@ -1497,49 +1555,41 @@ static int iscsi_data_xmit(struct iscsi_conn *conn)* only have one nop-out as a ping from us and targets should not * overflow us with nop-ins */ -spin_lock_bh (&conn->taskqueuelock); check_mgmt:while (!list_empty (&conn->mgmtqueue)) { - conn->task =list_entry (conn->mgmtqueue.next, - struct iscsi_task, running); -list_del_init (&conn->task->running); -spin_unlock_bh (&conn->taskqueuelock); -if (iscsi_prep_mgmt_task (conn, conn->task)) { + task = list_entry(conn->mgmtqueue.next, struct iscsi_task, + running); +list_del_init (&task->running); +if (iscsi_prep_mgmt_task (conn, task)) {/* regular RX path uses back_lock */ spin_lock_bh (&conn->session->back_lock); -__iscsi_put_task (conn->task); +__iscsi_put_task (task);spin_unlock_bh (&conn->session->back_lock); - conn->task =NULL ; -spin_lock_bh (&conn->taskqueuelock);continue ; } - rc =iscsi_xmit_task (conn); + rc =iscsi_xmit_task (conn, task, false);if (rc) goto done; -spin_lock_bh (&conn->taskqueuelock); }/* process pending command queue */ while (!list_empty (&conn->cmdqueue)) { - conn->task = list_entry(conn->cmdqueue.next, struct iscsi_task, - running); -list_del_init (&conn->task->running); -spin_unlock_bh (&conn->taskqueuelock); + task = list_entry(conn->cmdqueue.next, struct iscsi_task, + running); +list_del_init (&task->running);if (conn->session->state ==ISCSI_STATE_LOGGING_OUT ) { -fail_scsi_task (conn->task,DID_IMM_RETRY ); -spin_lock_bh (&conn->taskqueuelock); +fail_scsi_task (task,DID_IMM_RETRY );continue ; } - rc =iscsi_prep_scsi_cmd_pdu (conn->task); + rc =iscsi_prep_scsi_cmd_pdu (task);if (rc) {if (rc == -ENOMEM || rc == -EACCES ) -fail_scsi_task (conn->task,DID_IMM_RETRY ); +fail_scsi_task (task,DID_IMM_RETRY );else -fail_scsi_task (conn->task,DID_ABORT ); -spin_lock_bh (&conn->taskqueuelock); +fail_scsi_task (task,DID_ABORT );continue ; } - rc =iscsi_xmit_task (conn); + rc =iscsi_xmit_task (conn, task, false);if (rc) goto done;/*
@@ -1547,7 +1597,6 @@ static int iscsi_data_xmit(struct iscsi_conn *conn)* we need to check the mgmt queue for nops that need to * be sent to aviod starvation */ -spin_lock_bh (&conn->taskqueuelock);if (!list_empty (&conn->mgmtqueue)) goto check_mgmt; }
@@ -1561,21 +1610,17 @@ static int iscsi_data_xmit(struct iscsi_conn *conn) task = list_entry(conn->requeue.next, struct iscsi_task, running); +if (iscsi_check_tmf_restrictions (task,ISCSI_OP_SCSI_DATA_OUT ))break ; - conn->task = task; -list_del_init (&conn->task->running); - conn->task->state =ISCSI_TASK_RUNNING ; -spin_unlock_bh (&conn->taskqueuelock); - rc =iscsi_xmit_task (conn); +list_del_init (&task->running); + rc =iscsi_xmit_task (conn, task, true);if (rc) goto done; -spin_lock_bh (&conn->taskqueuelock);if (!list_empty (&conn->mgmtqueue)) goto check_mgmt; } -spin_unlock_bh (&conn->taskqueuelock);spin_unlock_bh (&conn->session->frwd_lock);return -ENODATA ;
@@ -1741,9 +1786,7 @@ int iscsi_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *sc) goto prepd_reject; } }else { -spin_lock_bh (&conn->taskqueuelock);list_add_tail (&task->running, &conn->cmdqueue); -spin_unlock_bh (&conn->taskqueuelock);iscsi_conn_queue_work (conn); }
@@ -2914,7 +2957,6 @@ iscsi_conn_setup(struct iscsi_cls_session *cls_session, int dd_size,INIT_LIST_HEAD (&conn->mgmtqueue);INIT_LIST_HEAD (&conn->cmdqueue);INIT_LIST_HEAD (&conn->requeue); -spin_lock_init (&conn->taskqueuelock);INIT_WORK (&conn->xmitwork, iscsi_xmitworker);/* allocate login_task used for the login/text sequences */
@@ -3080,10 +3122,16 @@ fail_mgmt_tasks(struct iscsi_session *session, struct iscsi_conn *conn)ISCSI_DBG_SESSION (conn->session,"failing mgmt itt 0x%x state %d\n" , task->itt, task->state); + +spin_lock_bh (&session->back_lock); +if (cleanup_queued_task (task)) { +spin_unlock_bh (&session->back_lock); +continue ; + } + state =ISCSI_TASK_ABRT_SESS_RECOV ;if (task->state ==ISCSI_TASK_PENDING ) state =ISCSI_TASK_COMPLETED ; -spin_lock_bh (&session->back_lock);iscsi_complete_task (task, state);spin_unlock_bh (&session->back_lock); }
drivers/scsi/libiscsi_tcp.c +57 -29
145 lines not on this page · raw message
include/scsi/libiscsi.h +2 -2
@@ -187,7 +187,7 @@ struct iscsi_conn {struct iscsi_task *task;/* xmit task in progress */ /* xmit */ -spinlock_t taskqueuelock;/* protects the next three lists */ +/* items must be added/deleted under frwd lock */ struct list_head mgmtqueue;/* mgmt (control) xmit queue */ struct list_head cmdqueue;/* data-path cmd queue */ struct list_head requeue;/* tasks needing another run */
@@ -332,7 +332,7 @@ struct iscsi_session {* cmdsn, queued_cmdsn * * session resources: * * - cmdpool kfifo_out , * -* - mgmtpool, */ +* - mgmtpool, queues */ spinlock_t back_lock;/* protects cmdsn_exp * * cmdsn_max, * * cmdpool kfifo_in */
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 3/9 libiscsi: fix iscsi_task use after free
In reply to Mike ChristieThe following bug was reported and debugged by [email protected]:
When testing kernel 4.18 version, NULL pointer dereference problem occurs
in iscsi_eh_cmd_timed_out function.
I think this bug in the upstream is still exists.
The analysis reasons are as follows:
1) For some reason, I/O command did not complete within the timeout period. The block layer timer works, call scsi_times_out() to handle I/O timeout logic. At the same time the command just completes.
2) scsi_times_out() call iscsi_eh_cmd_timed_out()
to processing timeout logic. although there is an NULL judgment
for the task, the task has not been released yet now.
3) iscsi_complete_task() call __iscsi_put_task(),
The task reference count reaches zero, the conditions for free task is met, then iscsi_free_task () free the task,
and let sc->SCp.ptr = NULL. After iscsi_eh_cmd_timed_out passes the task judgment check, there may be NULL dereference scenarios later.
CPU0 CPU3
|- scsi_times_out() |-iscsi_complete_task()
| |
|- iscsi_eh_cmd_timed_out() |-
__iscsi_put_task()
| |
|- task=sc->SCp.ptr, task is not NUL, check passed |-
iscsi_free_task(task)
| |
| |-> sc->SCp.ptr
= NULL
| |
|- task is NULL now, NULL pointer dereference |
| |
\|/ \|/Calltrace:
[380751.840862] BUG: unable to handle kernel NULL pointer dereference at 0000000000000138 [380751.843709] PGD 0 P4D 0 [380751.844770] Oops: 0000 [#1] SMP PTI [380751.846283] CPU: 0 PID: 403 Comm: kworker/0:1H Kdump: loaded Tainted: G [380751.851467] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996) [380751.856521] Workqueue: kblockd blk_mq_timeout_work [380751.858527] RIP: 0010:iscsi_eh_cmd_timed_out+0x15e/0x2e0 [libiscsi] [380751.861129] Code: 83 ea 01 48 8d 74 d0 08 48 8b 10 48 8b 4a 50 48 85 c9 74 2c 48 39 d5 74 [380751.868811] RSP: 0018:ffffc1e280a5fd58 EFLAGS: 00010246 [380751.870978] RAX: ffff9fd1e84e15e0 RBX: ffff9fd1e84e6dd0 RCX: 0000000116acc580 [380751.873791] RDX: ffff9fd1f97a9400 RSI: ffff9fd1e84e1800 RDI: ffff9fd1e4d6d420 [380751.876059] RBP: ffff9fd1e4d49000 R08: 0000000116acc580 R09: 0000000116acc580 [380751.878284] R10: 0000000000000000 R11: 0000000000000000 R12: ffff9fd1e6e931e8 [380751.880500] R13: ffff9fd1e84e6ee0 R14: 0000000000000010 R15: 0000000000000003 [380751.882687] FS: 0000000000000000(0000) GS:ffff9fd1fac00000(0000) knlGS:0000000000000000 [380751.885236] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [380751.887059] CR2: 0000000000000138 CR3: 000000011860a001 CR4: 00000000003606f0 [380751.889308] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [380751.891523] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [380751.893738] Call Trace: [380751.894639] scsi_times_out+0x60/0x1c0 [380751.895861] blk_mq_check_expired+0x144/0x200 [380751.897302] ? __switch_to_asm+0x35/0x70 [380751.898551] blk_mq_queue_tag_busy_iter+0x195/0x2e0 [380751.900091] ? __blk_mq_requeue_request+0x100/0x100 [380751.901611] ? __switch_to_asm+0x41/0x70 [380751.902853] ? __blk_mq_requeue_request+0x100/0x100 [380751.904398] blk_mq_timeout_work+0x54/0x130 [380751.905740] process_one_work+0x195/0x390 [380751.907228] worker_thread+0x30/0x390 [380751.908713] ? process_one_work+0x390/0x390 [380751.910350] kthread+0x10d/0x130 [380751.911470] ? kthread_flush_work_fn+0x10/0x10 [380751.913007] ret_from_fork+0x35/0x40
crash> dis -l iscsi_eh_cmd_timed_out+0x15exxxxx/drivers/scsi/libiscsi.c: 2062
1970 enum blk_eh_timer_return iscsi_eh_cmd_timed_out(struct scsi_cmnd *sc)
{
...
1984 spin_lock_bh(&session->frwd_lock);
1985 task = (struct iscsi_task *)sc->SCp.ptr;
1986 if (!task) {1987 /*
1988 * Raced with completion. Blk layer has taken
ownership
1989 * so let timeout code complete it now.
1990 */
1991 rc = BLK_EH_DONE;
1992 goto done;
1993 }
...
2052 for (i = 0; i < conn->session->cmds_max; i++) {
2053 running_task = conn->session->cmds[i];
2054 if (!running_task->sc || running_task == task ||
2055 running_task->state != ISCSI_TASK_RUNNING)
2056 continue;
2057
2058 /*
2059 * Only check if cmds started before this one have
made
2060 * progress, or this could never fail
2061 */
2062 if (time_after(running_task->sc->jiffies_at_alloc,
2063 task->sc->jiffies_at_alloc)) <---
2064 continue;
2065
...
}
carsh> struct scsi_cmnd ffff9fd1e6e931e8
struct scsi_cmnd {
...
SCp = {
ptr = 0x0, <--- iscsi_task
this_residual = 0,
... },
}
To prevent this, we take a ref to the cmd under the back (completion) lock so if the completion side were to call iscsi_complete_task on the task while the timer/eh paths are not holding the back_lock it will not be freed from under us.
Note that this requires the previous patch, "libiscsi: drop taskqueuelock" because bnx2i sleeps in its cleanup_task callout if the cmd is aborted. If the EH/timer and completion path are racing we don't know which path will do the last put. The previous patch moved the operations we needed to do under the forward lock to cleanup_queued_task. Once that has run we can drop the forward lock for the cmd and bnx2i no longer has to worry about if the EH, timer or completion path did the ast put and if the forward lock is held or not since it won't be.
--- drivers/scsi/bnx2i/bnx2i_iscsi.c | 2 - drivers/scsi/libiscsi.c | 71 ++++++++++++++++++++------------ 2 files changed, 45 insertions(+), 28 deletions(-)
drivers/scsi/bnx2i/bnx2i_iscsi.c +0 -2
@@ -1171,10 +1171,8 @@ static void bnx2i_cleanup_task(struct iscsi_task *task)bnx2i_send_cmd_cleanup_req (hba, task->dd_data);spin_unlock_bh (&conn->session->back_lock); -spin_unlock_bh (&conn->session->frwd_lock);wait_for_completion_timeout (&bnx2i_conn->cmd_cleanup_cmpl,msecs_to_jiffies (ISCSI_CMD_CLEANUP_TIMEOUT )); -spin_lock_bh (&conn->session->frwd_lock);spin_lock_bh (&conn->session->back_lock); }bnx2i_iscsi_unmap_sg_list (task->dd_data);
drivers/scsi/libiscsi.c +45 -26
169 lines not on this page · raw message
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 4/9 libiscsi: fix iscsi host workq destruction
In reply to Mike ChristieWe allocate the iscsi host workq in iscsi_host_alloc so iscsi_host_free should do the destruction. Drivers can then do their error/goto handling and call iscsi_host_free to clean up what has been allocated in iscsi_host_alloc.
--- drivers/scsi/libiscsi.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/scsi/libiscsi.c +3 -2
@@ -2738,8 +2738,6 @@ void iscsi_host_remove(struct Scsi_Host *shost)flush_signals (current);scsi_remove_host (shost); -if (ihost->workq) -destroy_workqueue (ihost->workq); }EXPORT_SYMBOL_GPL (iscsi_host_remove);
@@ -2747,6 +2745,9 @@ void iscsi_host_free(struct Scsi_Host *shost) {struct iscsi_host *ihost =shost_priv (shost); +if (ihost->workq) +destroy_workqueue (ihost->workq); +kfree (ihost->netdev);kfree (ihost->hwaddress);kfree (ihost->initiatorname);
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 5/9 libiscsi: add helper to calc max scsi cmds per session
In reply to Mike ChristieThis patch just breaks out the code that calculates the number of scsi cmds that will be used for a scsi session. It also adds a check that we don't go over the host's can_queue value.
--- drivers/scsi/libiscsi.c | 86 ++++++++++++++++++++++++++--------------- include/scsi/libiscsi.h | 2 + 2 files changed, 56 insertions(+), 32 deletions(-)
drivers/scsi/libiscsi.c +54 -32
@@ -2648,6 +2648,56 @@ void iscsi_pool_free(struct iscsi_pool *q) }EXPORT_SYMBOL_GPL (iscsi_pool_free); +int iscsi_host_get_max_scsi_cmds (struct Scsi_Host *shost, +uint16_t requested_cmds_max) +{ +int scsi_cmds, total_cmds = requested_cmds_max; + +check: +if (!total_cmds) + total_cmds =ISCSI_DEF_XMIT_CMDS_MAX ; +/* +* The iscsi layer needs some tasks for nop handling and tmfs, +* so the cmds_max must at least be greater than ISCSI_MGMT_CMDS_MAX +* + 1 command for scsi IO. +*/ +if (total_cmds <ISCSI_TOTAL_CMDS_MIN ) { +printk (KERN_ERR "iscsi: invalid max cmds of %d. Must be a power of two that is at least %d.\n" , + total_cmds,ISCSI_TOTAL_CMDS_MIN ); +return -EINVAL ; + } + +if (total_cmds >ISCSI_TOTAL_CMDS_MAX ) { +printk (KERN_INFO "iscsi: invalid max cmds of %d. Must be a power of 2 less than or equal to %d. Using %d.\n" , + requested_cmds_max,ISCSI_TOTAL_CMDS_MAX , +ISCSI_TOTAL_CMDS_MAX ); + total_cmds =ISCSI_TOTAL_CMDS_MAX ; + } + +if (!is_power_of_2 (total_cmds)) { + total_cmds =rounddown_pow_of_two (total_cmds); +if (total_cmds <ISCSI_TOTAL_CMDS_MIN ) { +printk (KERN_ERR "iscsi: invalid max cmds of %d. Must be a power of 2 greater than %d.\n" , requested_cmds_max,ISCSI_TOTAL_CMDS_MIN ); +return -EINVAL ; + } + +printk (KERN_INFO "iscsi: invalid max cmds %d. Must be a power of 2. Rounding max cmds down to %d.\n" , + requested_cmds_max, total_cmds); + } + + scsi_cmds = total_cmds -ISCSI_MGMT_CMDS_MAX ; +if (shost->can_queue && scsi_cmds > shost->can_queue) { + total_cmds = shost->can_queue; + +printk (KERN_INFO "iscsi: requested max cmds %u is higher than driver limit. Using driver limit %u\n" , + requested_cmds_max, shost->can_queue); + goto check; + } + +return scsi_cmds; +} +EXPORT_SYMBOL_GPL (iscsi_host_get_max_scsi_cmds); +/** * iscsi_host_add - add host to system * @shost: scsi host
@@ -2801,7 +2851,7 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost,struct iscsi_host *ihost =shost_priv (shost);struct iscsi_session *session;struct iscsi_cls_session *cls_session; -int cmd_i, scsi_cmds, total_cmds = cmds_max; +int cmd_i, scsi_cmds;unsigned long flags;spin_lock_irqsave (&ihost->lock, flags);
@@ -2812,37 +2862,9 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost, ihost->num_sessions++;spin_unlock_irqrestore (&ihost->lock, flags); -if (!total_cmds) - total_cmds =ISCSI_DEF_XMIT_CMDS_MAX ; -/* -* The iscsi layer needs some tasks for nop handling and tmfs, -* so the cmds_max must at least be greater than ISCSI_MGMT_CMDS_MAX -* + 1 command for scsi IO. -*/ -if (total_cmds <ISCSI_TOTAL_CMDS_MIN ) { -printk (KERN_ERR "iscsi: invalid can_queue of %d. can_queue " -"must be a power of two that is at least %d.\n" , - total_cmds,ISCSI_TOTAL_CMDS_MIN ); + scsi_cmds =iscsi_host_get_max_scsi_cmds (shost, cmds_max); +if (scsi_cmds <0 ) goto dec_session_count; - } - -if (total_cmds >ISCSI_TOTAL_CMDS_MAX ) { -printk (KERN_ERR "iscsi: invalid can_queue of %d. can_queue " -"must be a power of 2 less than or equal to %d.\n" , - cmds_max,ISCSI_TOTAL_CMDS_MAX ); - total_cmds =ISCSI_TOTAL_CMDS_MAX ; - } - -if (!is_power_of_2 (total_cmds)) { -printk (KERN_ERR "iscsi: invalid can_queue of %d. can_queue " -"must be a power of 2.\n" , total_cmds); - total_cmds =rounddown_pow_of_two (total_cmds); -if (total_cmds <ISCSI_TOTAL_CMDS_MIN ) - goto dec_session_count; -printk (KERN_INFO "iscsi: Rounding can_queue to %d.\n" , - total_cmds); - } - scsi_cmds = total_cmds -ISCSI_MGMT_CMDS_MAX ; cls_session = iscsi_alloc_session(shost, iscsit, sizeof(struct iscsi_session) +
@@ -2858,7 +2880,7 @@ iscsi_session_setup(struct iscsi_transport *iscsit, struct Scsi_Host *shost, session->lu_reset_timeout =15 ; session->abort_timeout =10 ; session->scsi_cmds_max = scsi_cmds; - session->cmds_max = total_cmds; + session->cmds_max = scsi_cmds +ISCSI_MGMT_CMDS_MAX ; session->queued_cmdsn = session->cmdsn = initial_cmdsn; session->exp_cmdsn = initial_cmdsn +1 ; session->max_cmdsn = initial_cmdsn +1 ;
include/scsi/libiscsi.h +2 -0
@@ -395,6 +395,8 @@ extern struct Scsi_Host *iscsi_host_alloc(struct scsi_host_template *sht,extern void iscsi_host_remove (struct Scsi_Host *shost);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, +uint16_t requested_cmds_max);/* * session management
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 6/9 iscsi_tcp: fix shost can_queue initialization
In reply to Mike ChristieWe are setting the shost's can_queue after we add the host which is too late, because scsi-ml will have allocated the tag set based on the can_queue value at that time. This patch has us use the iscsi_host_get_max_scsi_cmds helper to figure out the number of scsi cmds.
It also fixes up the template can_queue so it reflects the max scsi cmds we can support like how other drivers work.
--- drivers/scsi/iscsi_tcp.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
drivers/scsi/iscsi_tcp.c +7 -2
@@ -847,6 +847,7 @@ iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max,struct iscsi_session *session;struct iscsi_sw_tcp_host *tcp_sw_host;struct Scsi_Host *shost; +int rc;if (ep) {printk (KERN_ERR "iscsi_tcp: invalid ep %p.\n" , ep);
@@ -864,6 +865,11 @@ iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max, shost->max_channel =0 ; shost->max_cmd_len =SCSI_MAX_VARLEN_CDB_SIZE ; + rc =iscsi_host_get_max_scsi_cmds (shost, cmds_max); +if (rc <0 ) + goto free_host; + shost->can_queue = rc; +if (iscsi_host_add (shost,NULL )) goto free_host;
@@ -878,7 +884,6 @@ iscsi_sw_tcp_session_create(struct iscsi_endpoint *ep, uint16_t cmds_max, tcp_sw_host =iscsi_host_priv (shost); tcp_sw_host->session = session; - shost->can_queue = session->scsi_cmds_max;if (iscsi_tcp_r2tpool_alloc (session)) goto remove_session;return cls_session;
@@ -981,7 +986,7 @@ static struct scsi_host_template iscsi_sw_tcp_sht = { .name = "iSCSI Initiator over TCP/IP", .queuecommand = iscsi_queuecommand, .change_queue_depth = scsi_change_queue_depth, - .can_queue = ISCSI_DEF_XMIT_CMDS_MAX - 1, + .can_queue = ISCSI_TOTAL_CMDS_MAX, .sg_tablesize = 4096, .max_sectors = 0xFFFF, .cmd_per_lun = ISCSI_DEF_CMD_PER_LUN,
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 7/9 libiscsi: reset max/exp cmdsn during recovery
In reply to Mike ChristieIf we lose the session then relogin, but the new cmdsn window has shrunk (due to something like an admin changing a setting) we will have the old exp/max_cmdsn values and will never be able to update them. For example, max_cmdsn would be 64, but if on the target the user set the window to be smaller then the target could try to return the max_cmdsn as 32. We will see that new max_cmdsn in the rsp but because it's lower than the old max_cmdsn when the window was larger we will not update it.
So this patch has us reset the windown values during session cleanup so they can be updated after a new login.
--- drivers/scsi/libiscsi.c | 7 +++++++ 1 file changed, 7 insertions(+)
drivers/scsi/libiscsi.c +7 -0
@@ -3273,6 +3273,13 @@ int iscsi_conn_bind(struct iscsi_cls_session *cls_session, session->leadconn = conn;spin_unlock_bh (&session->frwd_lock); +/* +* The target could have reduced it's window size between logins, so +* we have to reset max/exp cmdsn so we can see the new values. +*/ +spin_lock_bh (&session->back_lock); + session->max_cmdsn = session->exp_cmdsn = session->cmdsn +1 ; +spin_unlock_bh (&session->back_lock);/* * Unblock xmitworker(), Login Phase will pass through. */
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 8/9 qla4xxx: use iscsi_is_session_online
In reply to Mike Christie__qla4xxx_is_chap_active just wants to know if a session is online and does not care about why it's not, so this has it use iscsi_is_session_online.
This is not a bug now, but the next patch changes the behavior of iscsi_session_chkready so this patch just prepares the driver for that change.
--- drivers/scsi/qla4xxx/ql4_os.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/scsi/qla4xxx/ql4_os.c +1 -1
@@ -841,7 +841,7 @@ static int __qla4xxx_is_chap_active(struct device *dev, void *data) sess = cls_session->dd_data; ddb_entry = sess->dd_data; -if (iscsi_session_chkready (cls_session)) +if (iscsi_is_session_online (cls_session)) goto exit_is_chap_active;if (ddb_entry->chap_tbl_idx == *chap_tbl_idx)
Signature · 1 line
2.25.1
Mike Christie
maintainerPATCH v6 9/9 iscsi class: drop session lock in iscsi_session_chkready
In reply to Mike ChristieThe session lock in iscsi_session_chkready is not needed because when we transition from logged into to another state we will block and/or remove the devices under the session, so no new IO will be sent to the drivers after the block/remove. IO that races with the block/removal is cleaned up by the drivers when it handles all outstanding IO, so this just added an extra lock in the main IO path. This patch removes the lock like other transport classes.
--- drivers/scsi/scsi_transport_iscsi.c | 3 --- 1 file changed, 3 deletions(-)
drivers/scsi/scsi_transport_iscsi.c +0 -3
@@ -1701,10 +1701,8 @@ static const char *iscsi_session_state_name(int state)int iscsi_session_chkready (struct iscsi_cls_session *session) { -unsigned long flags;int err; -spin_lock_irqsave (&session->lock, flags);switch (session->state) {case ISCSI_SESSION_LOGGED_IN : err =0 ;
@@ -1719,7 +1717,6 @@ int iscsi_session_chkready(struct iscsi_cls_session *session) err =DID_NO_CONNECT <<16 ;break ; } -spin_unlock_irqrestore (&session->lock, flags);return err; }EXPORT_SYMBOL_GPL (iscsi_session_chkready);
Signature · 1 line
2.25.1
Martin K. Petersen
AppliedMike,
Quoted 3 lines
The following patches made over Martin's 5.12 branches contain fixes for a cmd lifetime bug, software iscsi
can_queuesetup, and a couple of the lock cleanup patches Lee has already ackd.
Applied to 5.12/scsi-staging, thanks!
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering