In nvmet_sq_destroy we capture sq->ctrl early and if it is non-NULL we know that a ctrl was allocated (in the admin connect request handler) and we need to release pending AERs, clear ctrl->sqs and sq->ctrl (for nvme-loop primarily), and drop the final reference on the ctrl. However, a small window is possible where nvmet_sq_destroy starts (as a result of the client giving up and disconnecting) concurrently with the nvme admin connect cmd (which may be in an early stage). But *before* kill_and_confirm of sq->ref (i.e. the admin connect managed to get an sq live reference). In this case, sq->ctrl was allocated however after it was captured in a local variable in nvmet_sq_destroy. This prevented the final reference drop on the ctrl. Solve this by re-capturing the sq->ctrl after all inflight request has completed, where for sure sq->ctrl reference is final, and move forward based on that. This issue was observed in an environment with many hosts connecting multiple ctrls simoutanuosly, creating a delay in allocating a ctrl leading up to this race window. Reported-by: Alex Turin <alex@vastdata.com> Signed-off-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Keith Busch <kbusch@kernel.org>
The nvme pci driver synchronizes with all the namespace queues during a reset to ensure that there's no pending timeout work. Meanwhile the timeout work potentially iterates those same namespaces to freeze their queues. Each of those namespace iterations use the same read lock. If a write lock should somehow get between the synchronize and freeze steps, then forward progress is deadlocked. We had been relying on the nvme controller state machine to ensure the reset work wouldn't conflict with timeout work. That guarantee may be a bit fragile to rely on, so iterate the namespace lists without taking potentially circular locks, as reported by lockdep. Link: https://lore.kernel.org/all/20220930001943.zdbvolc3gkekfmcv@shindev/ Reported-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com> Tested-by: Shinichiro Kawasaki <shinichiro.kawasaki@wdc.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Keith Busch <kbusch@kernel.org>
bio_vec start offset may be relatively large particularly when large folio gets added to the bio. A bigger offset will result in avoiding the single-segment mapping optimization and end up using expensive mempool_alloc further. Rather than using absolute value, adjust bv_offset by NVME_CTRL_PAGE_SIZE while checking if segment can be fitted into one/two PRP entries. Suggested-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Kundan Kumar <kundan.kumar@samsung.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Keith Busch <kbusch@kernel.org>
sgs/sws are unused, so remove these from nvme_ns_head structure. Signed-off-by: Kanchan Joshi <joshi.k@samsung.com> Reviewed-by: Sagi Grimberg <sagi@grimberg.me> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Keith Busch <kbusch@kernel.org>
When disabling an nvmet namespace, there is a period where the
subsys->lock is released, as the ns disable waits for backend IO to
complete, and the ns percpu ref to be properly killed. The original
intent was to avoid taking the subsystem lock for a prolong period as
other processes may need to acquire it (for example new incoming
connections).
However, it opens up a window where another process may come in and
enable the ns, (re)intiailizing the ns percpu_ref, causing the disable
sequence to hang.
Solve this by taking the global nvmet_config_sem over the entire configfs
enable/disable sequence.
Fixes: a07b4970f464 ("nvmet: add a generic NVMe target")
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
There are io stats accounting that needs to be handled, so don't call
blk_mq_end_request() directly. Use the existing nvme_end_req() helper
that already handles everything.
Fixes: d4d957b53d91ee ("nvme-multipath: support io stats on the mpath device")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Batched completions were missing the io stats accounting and bio trace
events. Move the common code to a helper and call it from the batched
and non-batched functions.
Fixes: d4d957b53d91ee ("nvme-multipath: support io stats on the mpath device")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
In current native multipath design when a shared namespace is created, we loop through each possible numa-node, calculate the NUMA distance of that node from each nvme controller and then cache the optimal IO path for future reference while sending IO. The issue with this design is that we may refer to the NUMA distance table for an offline node which may not be populated at the time and so we may inadvertently end up finding and caching a non-optimal path for IO. Then latter when the corresponding numa-node becomes online and hence the NUMA distance table entry for that node is created, ideally we should re-calculate the multipath node distance for the newly added node however that doesn't happen unless we rescan/reset the controller. So essentially, we may keep using non-optimal IO path for a node which is made online after namespace is created. This patch helps fix this issue ensuring that when a shared namespace is created, we calculate the multipath node distance for each online numa-node instead of each possible numa-node. Then latter when a node becomes online and we receive any IO on that newly added node, we would calculate the multipath node distance for newly added node but this time NUMA distance table would have been already populated for newly added node. Hence we would be able to correctly calculate the multipath node distance and choose the optimal path for the IO. Signed-off-by: Nilay Shroff <nilay@linux.ibm.com> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Keith Busch <kbusch@kernel.org>
Pull NVMe updates and fixes from Keith: "nvme updates for Linux 6.10 - Fabrics connection retries (Daniel, Hannes) - Fabrics logging enhancements (Tokunori) - RDMA delete optimization (Sagi)" * tag 'nvme-6.10-2024-05-14' of git://git.infradead.org/nvme: nvme-rdma, nvme-tcp: include max reconnects for reconnect logging nvmet-rdma: Avoid o(n^2) loop in delete_ctrl nvme: do not retry authentication failures nvme-fabrics: short-circuit reconnect retries nvme: return kernel error codes for admin queue connect nvmet: return DHCHAP status codes from nvmet_setup_auth() nvmet: lock config semaphore when accessing DH-HMAC-CHAP key
Pull block updates from Jens Axboe: - Add a partscan attribute in sysfs, fixing an issue with systemd relying on an internal interface that went away. - Attempt #2 at making long running discards interruptible. The previous attempt went into 6.9, but we ended up mostly reverting it as it had issues. - Remove old ida_simple API in bcache - Support for zoned write plugging, greatly improving the performance on zoned devices. - Remove the old throttle low interface, which has been experimental since 2017 and never made it beyond that and isn't being used. - Remove page->index debugging checks in brd, as it hasn't caught anything and prepares us for removing in struct page. - MD pull request from Song - Don't schedule block workers on isolated CPUs * tag 'for-6.10/block-20240511' of git://git.kernel.dk/linux: (84 commits) blk-throttle: delay initialization until configuration blk-throttle: remove CONFIG_BLK_DEV_THROTTLING_LOW block: fix that util can be greater than 100% block: support to account io_ticks precisely block: add plug while submitting IO bcache: fix variable length array abuse in btree_iter bcache: Remove usage of the deprecated ida_simple_xx() API md: Revert "md: Fix overflow in is_mddev_idle" blk-lib: check for kill signal in ioctl BLKDISCARD block: add a bio_await_chain helper block: add a blk_alloc_discard_bio helper block: add a bio_chain_and_submit helper block: move discard checks into the ioctl handler block: remove the discard_granularity check in __blkdev_issue_discard block/ioctl: prefer different overflow check null_blk: Fix the WARNING: modpost: missing MODULE_DESCRIPTION() block: fix and simplify blkdevparts= cmdline parsing block: refine the EOF check in blkdev_iomap_begin block: add a partscan sysfs attribute for disks block: add a disk_has_partscan helper ...