The __nvmf_check_ready() routine used to bounce all filesystem io if the
controller state isn't LIVE. However, a later patch changed the logic so
that it rejection ends up being based on the Q live check. The FC
transport has a slightly different sequence from rdma and tcp for
shutting down queues/marking them non-live. FC marks its queue non-live
after aborting all ios and waiting for their termination, leaving a
rather large window for filesystem io to continue to hit the transport.
Unfortunately this resulted in filesystem I/O or applications seeing I/O
errors.
Change the FC transport to mark the queues non-live at the first sign of
teardown for the association (when I/O is initially terminated).
Fixes: 73a5379937 ("nvme-fabrics: allow to queue requests for live queues")
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
queue_rq() in pci only checks if the dispatched queue (nvmeq) is ready,
e.g. not being suspended. Since nvme_alloc_admin_tags() in reset flow
restarts the admin queue, users are able to submit admin commands to a
controller before reset_work() completes. Commands submitted under this
condition may interfere with commands that performs identify, IO queue
setup in reset_work(), and may result in a hang described in the
following patch.
As seen in the fabrics, user commands are prevented from being executed
under inproper controller states. We may reuse this logic to maintain a
clear admin queue during reset_work().
Signed-off-by: Tao Chiu <taochiu@synology.com>
Signed-off-by: Cody Wong <codywong@synology.com>
Reviewed-by: Leon Chien <leonchien@synology.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
SGLs support is mandatory for NVMe/FC, make sure that the target is
aligned to the specification.
Signed-off-by: Max Gurtovoy <mgurtovoy@nvidia.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
All nvme transport drivers preallocate an nvme command for each request.
Assume to use that command for nvme_setup_cmd() instead of requiring
drivers pass a pointer to it. All nvme drivers must initialize the
generic nvme_request 'cmd' to point to the transport's preallocated
nvme_command.
The generic nvme_request cmd pointer had previously been used only as a
temporary copy for passthrough commands. Since it now points to the
command that gets dispatched, passthrough commands must directly set it
up prior to executing the request.
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The nvme_fc_rcv_ls_req() function has first argument as pointer to
remoteport named portprt, but in the documentation comment that is name
is used as remoteport. Fix that to get rid if the compilation warning.
drivers/nvme//host/fc.c:1724: warning: Function parameter or member 'portptr' not described in 'nvme_fc_rcv_ls_req'
drivers/nvme//host/fc.c:1724: warning: Excess function parameter 'remoteport' description in 'nvme_fc_rcv_ls_req'
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
This is a prep patch so that we can move the identify data structure
related code initialization from nvme_init_identify() into a helper.
Rename the function nvmet_init_identify() to nvmet_init_ctrl_finish().
Next patch will move the nvme_id_ctrl related initialization from newly
renamed function nvme_init_ctrl_finish() into the nvme_init_identify()
helper.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Fabrics drivers currently reserve two tags on the admin queue. But
given that the connect command is only run on a freshly created queue
or after all commands have been force aborted we only need to reserve
a single tag.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Recent patch to prevent calling __nvme_fc_abort_outstanding_ios in
interrupt context results in a possible race condition. A controller
reset results in errored io completions, which schedules error
work. The change of error work to a work element allows it to fire
after the ctrl state transition to NVME_CTRL_CONNECTING, causing
any outstanding io (used to initialize the controller) to fail and
cause problems for connect_work.
Add a state check to only schedule error work if not in the RESETTING
state.
Fixes: 19fce0470f ("nvme-fc: avoid calling _nvme_fc_abort_outstanding_ios from interrupt context")
Signed-off-by: Nigel Kirkland <nkirkland2304@gmail.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
When a command has been aborted we should return NVME_SC_HOST_ABORTED_CMD
to be consistent with the other transports.
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
nvme_fc_terminate_exchange() is being called when exchanges are
being deleted, and as such we should be setting the NVME_REQ_CANCELLED
flag to have identical behaviour on all transports.
Signed-off-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The only usage of these is to put their addresses in arrays of pointers
to const attribute_groups. Make them const to allow the compiler to put
them in read-only memory.
Signed-off-by: Rikard Falkeborn <rikard.falkeborn@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Recent patches changed calling sequences. nvme_fc_abort_outstanding_ios
used to be called from a timeout or work context. Now it is being called
in an io completion context, which can be an interrupt handler.
Unfortunately, the abort outstanding ios routine attempts to stop nvme
queues and nested routines that may try to sleep, which is in conflict
with the interrupt handler.
Correct replacing the direct call with a work element scheduling, and the
abort outstanding ios routine will be called in the work element.
Fixes: 95ced8a2c7 ("nvme-fc: eliminate terminate_io use by nvme_fc_error_recovery")
Signed-off-by: James Smart <james.smart@broadcom.com>
Reported-by: Daniel Wagner <dwagner@suse.de>
Tested-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
This is purely a clenaup patch, add prefix NVME to the ADMIN_TIMEOUT to
make consistent with NVME_IO_TIMEOUT.
Signed-off-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
__nvme_fc_terminate_io() is now called by only 1 place, in reset_work.
Consoldate and move the functionality of terminate_io into reset_work.
In reset_work, rather than calling the create_association directly,
schedule the connect work element to do its thing. After scheduling,
flush the connect work element to continue with semantic of not
returning until connect has been attempted at least once.
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
nvme_fc_error_recovery() special cases handling when in CONNECTING state
and calls __nvme_fc_terminate_io(). __nvme_fc_terminate_io() itself
special cases CONNECTING state and calls the routine to abort outstanding
ios.
Simplify the sequence by putting the call to abort outstanding I/Os
directly in nvme_fc_error_recovery.
Move the location of __nvme_fc_abort_outstanding_ios(), and
nvme_fc_terminate_exchange() which is called by it, to avoid adding
function prototypes for nvme_fc_error_recovery().
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
err_work was created to handle errors (mainly I/O timeouts) while in
CONNECTING state. The flag for err_work_active is also unneeded.
Remove err_work_active and err_work. The actions to abort I/Os are moved
inline to nvme_error_recovery().
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Whenever there are errors during CONNECTING, the driver recovers by
aborting all outstanding ios and counts on the io completion to fail them
and thus the connection/association they are on. However, the connection
failure depends on a failure state from the core routines. Not all
commands that are issued by the core routine are guaranteed to cause a
failure of the core routine. They may be treated as a failure status and
the status is then ignored.
As such, whenever the transport enters error_recovery while CONNECTING,
it will set a new flag indicating an association failed. The
create_association routine which creates and initializes the controller,
will monitor the state of the flag as well as the core routine error
status and ensure the association fails if there was an error.
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
We've had several complaints about a 10s reconnect delay (the default)
when there was an error while there is connectivity to a subsystem.
The max_reconnects and reconnect_delay are set in common code prior to
calling the transport to create the controller.
This change checks if the default reconnect delay is being used, and if
so, it adjusts it to a shorter period (2s) for the nvme-fc transport.
It does so by calculating the controller loss tmo window, changing the
value of the reconnect delay, and then recalculating the maximum number
of reconnect attempts allowed.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
On reconnect, the code currently does not freeze the controller before
possibly updating the number hw queues for the controller.
Add the freeze before updating the number of hw queues. Note: the queues
are already started and remain started through the reconnect.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The loop that backs out of hw io queue creation continues through index
0, which corresponds to the admin queue as well.
Fix the loop so it only proceeds through indexes 1..n which correspond to
I/O queues.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Currently, an I/O timeout unconditionally invokes
nvme_fc_error_recovery() which checks for LIVE or CONNECTING state. If
live, the routine resets the controller which initiates a reconnect -
which is valid. If CONNECTING, err_work is scheduled. Err_work then
calls the terminate_io routine, which also checks for CONNECTING and
noops any further action on outstanding I/O. The result is nothing
happened to the timed out io. As such, if the command was dropped on
the wire, it will never timeout / complete, and the connect process
will hang.
Change the behavior of the io timeout routine to unconditionally abort
the I/O. I/O completion handling will note that an io failed due to an
abort and will terminate the connection / association as needed. If the
abort was unable to happen, continue with a call to
nvme_fc_error_recovery(). To ensure something different happens in
nvme_fc_error_recovery() rework it so at it will abort all I/Os on the
association to force a failure.
As I/O aborts now may occur outside of delete_association, counting for
completion must be wary and only count those aborted during
delete_association when TERMIO is set on the controller.
Signed-off-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The lldd may have made calls to delete a remote port or local port and
the delete is in progress when the cli then attempts to create a new
controller. Currently, this proceeds without error although it can't be
very successful.
Fix this by validating that both the host port and remote port are
present when a new controller is to be created.
Signed-off-by: James Smart <james.smart@broadcom.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Cancel async event work in case async event has been queued up, and
nvme_fc_submit_async_event() runs after event has been freed.
Signed-off-by: David Milburn <dmilburn@redhat.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
nvme_end_request is a bit misnamed, as it wraps around the
blk_mq_complete_* API. It's semantics also are non-trivial, so give it
a more descriptive name and add a comment explaining the semantics.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
On an error exit path, a negative error code should be returned
instead of a positive return value.
Fixes: e399441de9 ("nvme-fabrics: Add host support for FC transport")
Cc: James Smart <jsmart2021@gmail.com>
Signed-off-by: Tianjia Zhang <tianjia.zhang@linux.alibaba.com>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently the FC transport is set max_hw_sectors based on the lldds
max sgl segment count. However, the block queue max segments is
set based on the controller's max_segments count, which the transport
does not set. As such, the lldd is receiving sgl lists that are
exceeding its max segment count.
Set the controller max segment count and derive max_hw_sectors from
the max segment count.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
A deadlock happens in the following scenario with multipath:
1) scan_work(nvme0) detects a new nsid while nvme0
is an optimized path to it, path nvme1 happens to be
inaccessible.
2) Before scan_work is complete nvme0 disconnect is initiated
nvme_delete_ctrl_sync() sets nvme0 state to NVME_CTRL_DELETING
3) scan_work(1) attempts to submit IO,
but nvme_path_is_optimized() observes nvme0 is not LIVE.
Since nvme1 is a possible path IO is requeued and scan_work hangs.
--
Workqueue: nvme-wq nvme_scan_work [nvme_core]
kernel: Call Trace:
kernel: __schedule+0x2b9/0x6c0
kernel: schedule+0x42/0xb0
kernel: io_schedule+0x16/0x40
kernel: do_read_cache_page+0x438/0x830
kernel: read_cache_page+0x12/0x20
kernel: read_dev_sector+0x27/0xc0
kernel: read_lba+0xc1/0x220
kernel: efi_partition+0x1e6/0x708
kernel: check_partition+0x154/0x244
kernel: rescan_partitions+0xae/0x280
kernel: __blkdev_get+0x40f/0x560
kernel: blkdev_get+0x3d/0x140
kernel: __device_add_disk+0x388/0x480
kernel: device_add_disk+0x13/0x20
kernel: nvme_mpath_set_live+0x119/0x140 [nvme_core]
kernel: nvme_update_ns_ana_state+0x5c/0x60 [nvme_core]
kernel: nvme_set_ns_ana_state+0x1e/0x30 [nvme_core]
kernel: nvme_parse_ana_log+0xa1/0x180 [nvme_core]
kernel: nvme_mpath_add_disk+0x47/0x90 [nvme_core]
kernel: nvme_validate_ns+0x396/0x940 [nvme_core]
kernel: nvme_scan_work+0x24f/0x380 [nvme_core]
kernel: process_one_work+0x1db/0x380
kernel: worker_thread+0x249/0x400
kernel: kthread+0x104/0x140
--
4) Delete also hangs in flush_work(ctrl->scan_work)
from nvme_remove_namespaces().
Similiarly a deadlock with ana_work may happen: if ana_work has started
and calls nvme_mpath_set_live and device_add_disk, it will
trigger I/O. When we trigger disconnect I/O will block because
our accessible (optimized) path is disconnecting, but the alternate
path is inaccessible, so I/O blocks. Then disconnect tries to flush
the ana_work and hangs.
[ 605.550896] Workqueue: nvme-wq nvme_ana_work [nvme_core]
[ 605.552087] Call Trace:
[ 605.552683] __schedule+0x2b9/0x6c0
[ 605.553507] schedule+0x42/0xb0
[ 605.554201] io_schedule+0x16/0x40
[ 605.555012] do_read_cache_page+0x438/0x830
[ 605.556925] read_cache_page+0x12/0x20
[ 605.557757] read_dev_sector+0x27/0xc0
[ 605.558587] amiga_partition+0x4d/0x4c5
[ 605.561278] check_partition+0x154/0x244
[ 605.562138] rescan_partitions+0xae/0x280
[ 605.563076] __blkdev_get+0x40f/0x560
[ 605.563830] blkdev_get+0x3d/0x140
[ 605.564500] __device_add_disk+0x388/0x480
[ 605.565316] device_add_disk+0x13/0x20
[ 605.566070] nvme_mpath_set_live+0x5e/0x130 [nvme_core]
[ 605.567114] nvme_update_ns_ana_state+0x2c/0x30 [nvme_core]
[ 605.568197] nvme_update_ana_state+0xca/0xe0 [nvme_core]
[ 605.569360] nvme_parse_ana_log+0xa1/0x180 [nvme_core]
[ 605.571385] nvme_read_ana_log+0x76/0x100 [nvme_core]
[ 605.572376] nvme_ana_work+0x15/0x20 [nvme_core]
[ 605.573330] process_one_work+0x1db/0x380
[ 605.574144] worker_thread+0x4d/0x400
[ 605.574896] kthread+0x104/0x140
[ 605.577205] ret_from_fork+0x35/0x40
[ 605.577955] INFO: task nvme:14044 blocked for more than 120 seconds.
[ 605.579239] Tainted: G OE 5.3.5-050305-generic #201910071830
[ 605.580712] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 605.582320] nvme D 0 14044 14043 0x00000000
[ 605.583424] Call Trace:
[ 605.583935] __schedule+0x2b9/0x6c0
[ 605.584625] schedule+0x42/0xb0
[ 605.585290] schedule_timeout+0x203/0x2f0
[ 605.588493] wait_for_completion+0xb1/0x120
[ 605.590066] __flush_work+0x123/0x1d0
[ 605.591758] __cancel_work_timer+0x10e/0x190
[ 605.593542] cancel_work_sync+0x10/0x20
[ 605.594347] nvme_mpath_stop+0x2f/0x40 [nvme_core]
[ 605.595328] nvme_stop_ctrl+0x12/0x50 [nvme_core]
[ 605.596262] nvme_do_delete_ctrl+0x3f/0x90 [nvme_core]
[ 605.597333] nvme_sysfs_delete+0x5c/0x70 [nvme_core]
[ 605.598320] dev_attr_store+0x17/0x30
Fix this by introducing a new state: NVME_CTRL_DELETE_NOIO, which will
indicate the phase of controller deletion where I/O cannot be allowed
to access the namespace. NVME_CTRL_DELETING still allows mpath I/O to
be issued to the bottom device, and only after we flush the ana_work
and scan_work (after nvme_stop_ctrl and nvme_prep_remove_namespaces)
we change the state to NVME_CTRL_DELETING_NOIO. Also we prevent ana_work
from re-firing by aborting early if we are not LIVE, so we should be safe
here.
In addition, change the transport drivers to follow the updated state
machine.
Fixes: 0d0b660f21 ("nvme: add ANA support")
Reported-by: Anton Eidelman <anton@lightbitslabs.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Use the new blk_mq_complete_request_remote helper to avoid an indirect
function call in the completion fast path.
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Asynchronous event notifications do not have an associated request.
When fcp_io() fails we unconditionally call nvme_cleanup_cmd() which
leads to a crash.
Fixes: 16686f3a6c ("nvme: move common call to nvme_cleanup_cmd to core layer")
Signed-off-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Himanshu Madhani <hmadhani2024@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The current codebase makes use of the zero-length array language
extension to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning
in case the flexible array does not occur last in the structure, which
will help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by
this change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
sizeof(flexible-array-member) triggers a warning because flexible array
members have incomplete type[1]. There are some instances of code in
which the sizeof operator is being incorrectly/erroneously applied to
zero-length arrays and the result is zero. Such instances may be hiding
some bugs. So, this work (flexible-array member conversions) will also
help to get completely rid of those sorts of issues.
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The nvme-fc devloss_tmo is computed as the min of either the
ctrl_loss_tmo (max_retries * reconnect_delay) or the remote port's
devloss_tmo. But what gets printed as the nvme-fc devloss_tmo in
nvme_fc_reconnect_or_delete() is always the remote port's devloss_tmo
value. So correct this by printing the min value instead.
Signed-off-by: Martin George <marting@netapp.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
When CONFIG_ARCH_NO_SG_CHAIN is set, op->sgl[0] cannot be dereferenced,
as gcc-10 now points out:
drivers/nvme/host/fc.c: In function 'nvme_fc_init_request':
drivers/nvme/host/fc.c:1774:29: warning: array subscript 0 is outside the bounds of an interior zero-length array 'struct scatterlist[0]' [-Wzero-length-bounds]
1774 | op->op.fcp_req.first_sgl = &op->sgl[0];
| ^~~~~~~~~~~
drivers/nvme/host/fc.c:98:21: note: while referencing 'sgl'
98 | struct scatterlist sgl[NVME_INLINE_SG_CNT];
| ^~~
I don't know if this is a legitimate warning or a false-positive.
If this is just a false alarm, the warning is easily suppressed
by interpreting the array as a pointer.
Fixes: b1ae1a2389 ("nvme-fc: Avoid preallocating big SGL for data")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The nvme-fc host transport did not support the reception of a
FC-NVME LS. Reception is necessary to implement full compliance
with FC-NVME-2.
Populate the LS receive handler, and specifically the handling
of a Disconnect Association LS. The response to the LS, if it
matched a controller, must be sent after the aborts for any
I/O on any connection have been sent.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Given that both host and target now generate and receive LS's create
a single table definition for LS names. Each tranport half will have
a local version of the table.
As Create Association LS is issued by both sides, and received by
both sides, create common routines to format the LS and to validate
the LS.
Convert the host side transport to use the new common Create
Association LS formatting routine.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Convert the assoc_active boolean flag to a bitop on the flags field.
The bit ops will provide atomicity.
To make this change, the flags field was converted to a long type,
which also affects the FCCTRL_TERMIO flag. Both FCCTRL_TERMIO and
now ASSOC_ACTIVE flags are set/cleared by bit operations.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Ensure that when allocations are done, and the lldd options indicate
no private data is needed, that private pointers will be set to NULL
(catches driver error that forgot to set private data size).
Slightly reorg the allocations so that private data follows allocations
for LS request/response buffers. Ensures better alignments for the buffers
as well as the private pointer.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Routines in the target will want to be used in the host as well.
Error definitions should now shared as both sides will process
requests and responses to requests.
Moved common declarations to new fc.h header kept in the host
subdirectory.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The current LLDD api has:
nvme-fc: contains api for transport to do LS requests (and aborts of
them). However, there is no interface for reception of LS's and sending
responses for them.
nvmet-fc: contains api for transport to do reception of LS's and sending
of responses for them. However, there is no interface for doing LS
requests.
Revise the api's so that both nvme-fc and nvmet-fc can send LS's, as well
as receiving LS's and sending their responses.
Change name of the rcv_ls_req struct to better reflect generic use as
a context to used to send an ls rsp. Specifically:
nvmefc_tgt_ls_req -> nvmefc_ls_rsp
nvmefc_tgt_ls_req.nvmet_fc_private -> nvmefc_ls_rsp.nvme_fc_private
Change nvmet_fc_rcv_ls_req() calling sequence to provide handle that
can be used by transport in later LS request sequences for an association.
nvme-fc nvmet_fc nvme_fcloop:
Revise to adapt to changed names in api header.
Change calling sequence to nvmet_fc_rcv_ls_req() for hosthandle.
Add stubs for new interfaces:
host/fc.c: nvme_fc_rcv_ls_req()
target/fc.c: nvmet_fc_invalidate_host()
lpfc:
Revise to adapt code to changed names in api header.
Change calling sequence to nvmet_fc_rcv_ls_req() for hosthandle.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The original patch was to resolve the lldd being able to be unloaded
while being used to talk to the boot device of the system. However, the
end result of the original patch is that any driver unload while a nvme
controller is live via the lldd is now being prohibited. Given the module
reference, the module teardown routine can't be called, thus there's no
way, other than manual actions to terminate the controllers.
Fixes: 863fbae929 ("nvme_fc: add module to ops template to allow module references")
Cc: <stable@vger.kernel.org> # v5.4+
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Put the ctrl reference count at nvme_uninit_ctrl as opposed to
nvme_init_ctrl which takes it. This decrease the reference count at the
core layer instead of decreasing it on each transport separately.
Also move the call of nvme_uninit_ctrl at PCI driver after calling to
nvme_release_prp_pools and nvme_dev_unmap, in order to put the reference
count after using the dev. This is safe because those functions use
nvme_dev which is freed only later at nvme_pci_free_ctrl.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
In case nvme_sysfs_delete() is called by the user before taking the ctrl
reference count, the ctrl may be freed during the creation and cause the
bug. Take the reference as soon as the controller is externally visible,
which is done by cdev_device_add() in nvme_init_ctrl(). Also take the
reference count at the core layer instead of taking it on each transport
separately.
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
If an error occurs on one of the ios used for creating an
association, the creating routine has error paths that are
invoked by the command failure and the error paths will free
up the controller resources created to that point.
But... the io was ultimately determined by an asynchronous
completion routine that detected the error and which
unconditionally invokes the error_recovery path which calls
delete_association. Delete association deletes all outstanding
io then tears down the controller resources. So the
create_association thread can be running in parallel with
the error_recovery thread. What was seen was the LLDD received
a call to delete a queue, causing the LLDD to do a free of a
resource, then the transport called the delete queue again
causing the driver to repeat the free call. The second free
routine corrupted the allocator. The transport shouldn't be
making the duplicate call, and the delete queue is just one
of the resources being freed.
To fix, it is realized that the create_association path is
completely serialized with one command at a time. So the
failed io completion will always be seen by the create_association
path and as of the failure, there are no ios to terminate and there
is no reason to be manipulating queue freeze states, etc.
The serialized condition stays true until the controller is
transitioned to the LIVE state. Thus the fix is to change the
error recovery path to check the controller state and only
invoke the teardown path if not already in the CONNECTING state.
Reviewed-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
In nvme-fc: it's possible to have connected active controllers
and as no references are taken on the LLDD, the LLDD can be
unloaded. The controller would enter a reconnect state and as
long as the LLDD resumed within the reconnect timeout, the
controller would resume. But if a namespace on the controller
is the root device, allowing the driver to unload can be problematic.
To reload the driver, it may require new io to the boot device,
and as it's no longer connected we get into a catch-22 that
eventually fails, and the system locks up.
Fix this issue by taking a module reference for every connected
controller (which is what the core layer did to the transport
module). Reference is cleared when the controller is removed.
Acked-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
nvme_fc_create_io_queues() preallocates a big buffer for the IO SGL based
on SG_CHUNK_SIZE.
Modern DMA engines are often capable of dealing with very big segments so
the SG_CHUNK_SIZE is often too big. SG_CHUNK_SIZE results in a static 4KB
SGL allocation per command.
If a controller has lots of deep queues, preallocation for the sg list can
consume substantial amounts of memory. For nvme-fc, nr_hw_queues can be
128 and each queue's depth 128. This means the resulting preallocation
for the data SGL is 128*128*4K = 64MB per controller.
Switch to runtime allocation for SGL for lists longer than 2 entries. This
is the approach used by NVMe PCI so it should be reasonable for NVMeOF as
well. Runtime SGL allocation has always been the case for the legacy I/O
path so this is nothing new.
Reviewed-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Israel Rukshin <israelr@mellanox.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
nvme_cleanup_cmd should be called for each call to nvme_setup_cmd
(symmetrical functions). Move the call for nvme_cleanup_cmd to the common
core layer and call it during nvme_complete_rq for the good flow. For
error flow, each transport will call nvme_cleanup_cmd independently. Also
take care of a special case of path failure, where we call
nvme_complete_rq without doing nvme_setup_cmd.
Signed-off-by: Max Gurtovoy <maxg@mellanox.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Code today only clears the association_id if a Disconnect LS is transmit.
Remove ambiguity and unconditionally clear the association_id if the
association has been terminated.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Change wording on a couple of messages to clarify what happened.
Signed-off-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Set the new category field in the FC-NVME CMND_IU based on queue number.
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Sync sources with revised structure and field names to correspond with
FC-NVME-2 header sync-up.
Tested interoperability with success:
- prior initiator with new target
- prior target with new initiator
- new on new
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
