| 1 |
From f809772d40dc7774cd047140bc0919c80ad521e7 Mon Sep 17 00:00:00 2001 |
| 2 |
From: James Smart <jsmart2021@gmail.com> |
| 3 |
Date: Thu, 27 Sep 2018 16:58:54 -0700 |
| 4 |
Subject: [PATCH 039/145] nvme: call nvme_complete_rq when nvmf_check_ready |
| 5 |
fails for mpath I/O |
| 6 |
|
| 7 |
[ Upstream commit 783f4a4408e1251d17f333ad56abac24dde988b9 ] |
| 8 |
|
| 9 |
When an io is rejected by nvmf_check_ready() due to validation of the |
| 10 |
controller state, the nvmf_fail_nonready_command() will normally return |
| 11 |
BLK_STS_RESOURCE to requeue and retry. However, if the controller is |
| 12 |
dying or the I/O is marked for NVMe multipath, the I/O is failed so that |
| 13 |
the controller can terminate or so that the io can be issued on a |
| 14 |
different path. Unfortunately, as this reject point is before the |
| 15 |
transport has accepted the command, blk-mq ends up completing the I/O |
| 16 |
and never calls nvme_complete_rq(), which is where multipath may preserve |
| 17 |
or re-route the I/O. The end result is, the device user ends up seeing an |
| 18 |
EIO error. |
| 19 |
|
| 20 |
Example: single path connectivity, controller is under load, and a reset |
| 21 |
is induced. An I/O is received: |
| 22 |
|
| 23 |
a) while the reset state has been set but the queues have yet to be |
| 24 |
stopped; or |
| 25 |
b) after queues are started (at end of reset) but before the reconnect |
| 26 |
has completed. |
| 27 |
|
| 28 |
The I/O finishes with an EIO status. |
| 29 |
|
| 30 |
This patch makes the following changes: |
| 31 |
|
| 32 |
- Adds the HOST_PATH_ERROR pathing status from TP4028 |
| 33 |
- Modifies the reject point such that it appears to queue successfully, |
| 34 |
but actually completes the io with the new pathing status and calls |
| 35 |
nvme_complete_rq(). |
| 36 |
- nvme_complete_rq() recognizes the new status, avoids resetting the |
| 37 |
controller (likely was already done in order to get this new status), |
| 38 |
and calls the multipather to clear the current path that errored. |
| 39 |
This allows the next command (retry or new command) to select a new |
| 40 |
path if there is one. |
| 41 |
|
| 42 |
Signed-off-by: James Smart <jsmart2021@gmail.com> |
| 43 |
Reviewed-by: Sagi Grimberg <sagi@grimberg.me> |
| 44 |
Signed-off-by: Christoph Hellwig <hch@lst.de> |
| 45 |
Signed-off-by: Sasha Levin <sashal@kernel.org> |
| 46 |
--- |
| 47 |
drivers/nvme/host/fabrics.c | 7 +++++-- |
| 48 |
drivers/nvme/host/multipath.c | 7 +++++++ |
| 49 |
include/linux/nvme.h | 1 + |
| 50 |
3 files changed, 13 insertions(+), 2 deletions(-) |
| 51 |
|
| 52 |
diff --git a/drivers/nvme/host/fabrics.c b/drivers/nvme/host/fabrics.c |
| 53 |
index 206d63cb1afc..bcd09d3a44da 100644 |
| 54 |
--- a/drivers/nvme/host/fabrics.c |
| 55 |
+++ b/drivers/nvme/host/fabrics.c |
| 56 |
@@ -552,8 +552,11 @@ blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl, |
| 57 |
ctrl->state != NVME_CTRL_DEAD && |
| 58 |
!blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH)) |
| 59 |
return BLK_STS_RESOURCE; |
| 60 |
- nvme_req(rq)->status = NVME_SC_ABORT_REQ; |
| 61 |
- return BLK_STS_IOERR; |
| 62 |
+ |
| 63 |
+ nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR; |
| 64 |
+ blk_mq_start_request(rq); |
| 65 |
+ nvme_complete_rq(rq); |
| 66 |
+ return BLK_STS_OK; |
| 67 |
} |
| 68 |
EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command); |
| 69 |
|
| 70 |
diff --git a/drivers/nvme/host/multipath.c b/drivers/nvme/host/multipath.c |
| 71 |
index 9fe3fff818b8..b71c9ad1bf45 100644 |
| 72 |
--- a/drivers/nvme/host/multipath.c |
| 73 |
+++ b/drivers/nvme/host/multipath.c |
| 74 |
@@ -77,6 +77,13 @@ void nvme_failover_req(struct request *req) |
| 75 |
queue_work(nvme_wq, &ns->ctrl->ana_work); |
| 76 |
} |
| 77 |
break; |
| 78 |
+ case NVME_SC_HOST_PATH_ERROR: |
| 79 |
+ /* |
| 80 |
+ * Temporary transport disruption in talking to the controller. |
| 81 |
+ * Try to send on a new path. |
| 82 |
+ */ |
| 83 |
+ nvme_mpath_clear_current_path(ns); |
| 84 |
+ break; |
| 85 |
default: |
| 86 |
/* |
| 87 |
* Reset the controller for any non-ANA error as we don't know |
| 88 |
diff --git a/include/linux/nvme.h b/include/linux/nvme.h |
| 89 |
index 68e91ef5494c..818dbe9331be 100644 |
| 90 |
--- a/include/linux/nvme.h |
| 91 |
+++ b/include/linux/nvme.h |
| 92 |
@@ -1241,6 +1241,7 @@ enum { |
| 93 |
NVME_SC_ANA_PERSISTENT_LOSS = 0x301, |
| 94 |
NVME_SC_ANA_INACCESSIBLE = 0x302, |
| 95 |
NVME_SC_ANA_TRANSITION = 0x303, |
| 96 |
+ NVME_SC_HOST_PATH_ERROR = 0x370, |
| 97 |
|
| 98 |
NVME_SC_DNR = 0x4000, |
| 99 |
}; |
| 100 |
-- |
| 101 |
2.19.1 |
| 102 |
|
Parent Directory
| 
Revision Log