[PATCH 4/5] nvmet: use atomic allocations when allocating fc requests

From: Johannes Thumshirn
Date: Tue May 15 2018 - 03:40:59 EST


fcloop_fcp_req() runs with the hctx_lock (a rcu_read_lock() locked
section) held, so memory allocations done in this context have to be
atomic.

This fixes the follwing lockdep complaint:
[ 9.753313] BUG: sleeping function called from invalid context at mm/slab.h:421
[ 9.754518] in_atomic(): 1, irqs_disabled(): 0, pid: 1420, name: nvme
[ 9.755613] 3 locks held by nvme/1420:
[ 9.756221] #0: (ptrval) (nvmf_dev_mutex){+.+.}, at: nvmf_dev_write+0x6a/0xb7d [nvme_fabrics]
[ 9.757575] #1: (ptrval) (nvmf_transports_rwsem){++++}, at: nvmf_dev_write+0x6e5/0xb7d [nvme_fabrics]
[ 9.759000] #2: (ptrval) (rcu_read_lock){....}, at: hctx_lock+0x56/0xd0
[ 9.760141] CPU: 2 PID: 1420 Comm: nvme Not tainted 4.17.0-rc5+ #883
[ 9.761078] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
[ 9.762624] Call Trace:
[ 9.763021] dump_stack+0x78/0xb3
[ 9.763505] ___might_sleep+0x227/0x250
[ 9.764115] kmem_cache_alloc_trace+0x1ae/0x270
[ 9.764793] fcloop_fcp_req+0x32/0x1a0 [nvme_fcloop]
[ 9.765561] nvme_fc_start_fcp_op.part.39+0x193/0x4c0 [nvme_fc]
[ 9.766480] blk_mq_dispatch_rq_list+0x7f/0x4a0
[ 9.767163] ? blk_mq_flush_busy_ctxs+0xa8/0xf0
[ 9.767871] blk_mq_sched_dispatch_requests+0x16e/0x170
[ 9.768644] __blk_mq_run_hw_queue+0x79/0xd0
[ 9.769294] __blk_mq_delay_run_hw_queue+0x11c/0x160
[ 9.770012] blk_mq_run_hw_queue+0x63/0xc0
[ 9.770667] blk_mq_sched_insert_request+0xb2/0x140
[ 9.771399] blk_execute_rq+0x64/0xc0
[ 9.771990] __nvme_submit_sync_cmd+0x63/0xd0 [nvme_core]
[ 9.772765] nvmf_connect_admin_queue+0x11e/0x190 [nvme_fabrics]
[ 9.773659] ? mark_held_locks+0x6b/0x90
[ 9.774798] nvme_fc_create_association+0x35b/0x970 [nvme_fc]
[ 9.775631] nvme_fc_create_ctrl+0x5d2/0x830 [nvme_fc]
[ 9.776423] nvmf_dev_write+0x92d/0xb7d [nvme_fabrics]
[ 9.777188] __vfs_write+0x21/0x130
[ 9.777725] ? selinux_file_permission+0xe9/0x140
[ 9.778417] ? security_file_permission+0x2f/0xb0
[ 9.779158] vfs_write+0xbd/0x1c0
[ 9.779644] ksys_write+0x40/0xa0
[ 9.780184] ? do_syscall_64+0xd/0x190
[ 9.780736] do_syscall_64+0x51/0x190
[ 9.781281] entry_SYSCALL_64_after_hwframe+0x49/0xbe

Signed-off-by: Johannes Thumshirn <jthumshirn@xxxxxxx>
---
drivers/nvme/target/fcloop.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/nvme/target/fcloop.c b/drivers/nvme/target/fcloop.c
index 34712def81b1..d2209c60f95f 100644
--- a/drivers/nvme/target/fcloop.c
+++ b/drivers/nvme/target/fcloop.c
@@ -509,7 +509,7 @@ fcloop_fcp_req(struct nvme_fc_local_port *localport,
if (!rport->targetport)
return -ECONNREFUSED;

- tfcp_req = kzalloc(sizeof(*tfcp_req), GFP_KERNEL);
+ tfcp_req = kzalloc(sizeof(*tfcp_req), GFP_ATOMIC);
if (!tfcp_req)
return -ENOMEM;

--
2.16.3