kernel.git/drivers/nvme/host/core.c, branch linux-6.16.y

nvme: fix PI insert on write

2025-09-25T09:16:41Z

[ Upstream commit 7ac3c2889bc060c3f67cf44df0dbb093a835c176 ] I recently ran into an issue where the PI generated using the block layer integrity code differs from that from a kernel using the PRACT fallback when the block layer integrity code is disabled, and I tracked this down to us using PRACT incorrectly. The NVM Command Set Specification (section 5.33 in 1.2, similar in older versions) specifies the PRACT insert behavior as: Inserted protection information consists of the computed CRC for the protection information format (refer to section 5.3.1) in the Guard field, the LBAT field value in the Application Tag field, the LBST field value in the Storage Tag field, if defined, and the computed reference tag in the Logical Block Reference Tag. Where the computed reference tag is defined as following for type 1 and type 2 using the text below that is duplicated in the respective bullet points: the value of the computed reference tag for the first logical block of the command is the value contained in the Initial Logical Block Reference Tag (ILBRT) or Expected Initial Logical Block Reference Tag (EILBRT) field in the command, and the computed reference tag is incremented for each subsequent logical block. So we need to set ILBRT field, but we currently don't. Interestingly this works fine on my older type 1 formatted SSD, but Qemu trips up on this. We already set ILBRT for Write Same since commit aeb7bb061be5 ("nvme: set the PRACT bit when using Write Zeroes with T10 PI"). To ease this, move the PI type check into nvme_set_ref_tag. Reviewed-by: Martin K. Petersen Signed-off-by: Christoph Hellwig Signed-off-by: Keith Busch Signed-off-by: Sasha Levin

nvme: fix misaccounting of nvme-mpath inflight I/O

2025-07-15T07:49:13Z

Procedures for nvme-mpath IO accounting: 1) initialize nvme_request and clear flags; 2) set NVME_MPATH_IO_STATS and increase inflight counter when IO started; 3) check NVME_MPATH_IO_STATS and decrease inflight counter when IO is done; However, for the case nvme_fail_nonready_command(), both step 1) and 2) are skipped, and if old nvme_request set NVME_MPATH_IO_STATS and then request is reused, step 3) will still be executed, causing inflight I/O counter to be negative. Fix the problem by clearing nvme_request in nvme_fail_nonready_command(). Fixes: ea5e5f42cd2c ("nvme-fabrics: avoid double completions in nvmf_fail_nonready_command") Reported-by: Yi Zhang Closes: https://lore.kernel.org/all/CAHj4cs_+dauobyYyP805t33WMJVzOWj=7+51p4_j9rA63D9sog@mail.gmail.com/ Signed-off-by: Yu Kuai Signed-off-by: Christoph Hellwig

nvme: revert the cross-controller atomic write size validation

2025-07-15T07:49:13Z

This was originally added by commit 8695f060a029 ("nvme: all namespaces in a subsystem must adhere to a common atomic write size") to check the all controllers in a subsystem report the same atomic write size, but the check wasn't quite correct and caused problems for devices with multiple namespaces that report different LBA sizes. Commit f46d273449ba ("nvme: fix atomic write size validation") tried to fix this, but then caused problems for namespace rediscovery after a format with an LBA size change that changes the AWUPF value. This drops the validation and essentially reverts those two commits while keeping the cleanup that went in between the two. We'll need to figure out how to properly check for the mouse trap that nvme left us, but for now revert the check to keep devices working for users who couldn't care less about the atomic write feature. Fixes: 8695f060a029 ("nvme: all namespaces in a subsystem must adhere to a common atomic write size") Fixes: f46d273449ba ("nvme: fix atomic write size validation") Signed-off-by: Christoph Hellwig Reviewed-by: Alan Adamson Reviewed-by: Keith Busch Reviewed-by: John Garry Reviewed-by: Chaitanya Kulkarni Tested-by: Alan Adamson

nvme: fix endianness of command word prints in nvme_log_err_passthru()

2025-07-14T13:51:06Z

The command word members of struct nvme_common_command are __le32 type, so use helper le32_to_cpu() to read them properly. Fixes: 9f079dda1433 ("nvme: allow passthru cmd error logging") Signed-off-by: John Garry Reviewed-by: Alan Adamson Reviewed-by: Keith Busch Signed-off-by: Christoph Hellwig

nvme: fix inconsistent RCU list manipulation in nvme_ns_add_to_ctrl_list()

2025-07-14T13:50:32Z

When inserting a namespace into the controller's namespace list, the function uses list_add_rcu() when the namespace is inserted in the middle of the list, but falls back to a regular list_add() when adding at the head of the list. This inconsistency could lead to race conditions during concurrent access, as users might observe a partially updated list. Fix this by consistently using list_add_rcu() in both code paths to ensure proper RCU protection throughout the entire function. Fixes: be647e2c76b2 ("nvme: use srcu for iterating namespace list") Signed-off-by: Zheng Qixing Signed-off-by: Christoph Hellwig

nvme: correctly account for namespace head reference counter

2025-06-30T06:31:49Z

The blktests nvme/058 manifests an issue where the NVMe subsystem kobject entry remains stale in sysfs, causing a failure during subsequent NVMe module reloads[1]. Specifically, when attempting to register a new NVMe subsystem, the driver encounters a kobejct name collision because a stale kobject still exists. Though, please note that nvme/058 doesn't report any failure and test case passes and it's only during subsequent NVMe module reloads, the stale nvme sub- system kobject entry in sysfs causes the observed symptom[1]. This issue stems from an imbalance in the get/put usage of the namespace head (nshead) reference counter. The nshead holds a reference to the associated NVMe subsystem. If the nshead reference is not properly released, it prevents the cleanup of the subsystem's kobject, leaving nvme subsystem stale entry behind in sysfs. During the failure case, the last namespace path referencing a nshead is removed, but the nshead reference was not released. This occurs because the release logic currently only puts the nshead reference when its state is LIVE. However, in configurations where ANA (Asymmetric Namespace Access) is enabled, a namespace may be associated with an ANA state that is neither optimized nor non-optimized. In this case, the nshead may never transition to LIVE, and the corresponding nshead reference is then never dropped. In fact nvme/058 associates some of nvme namespaces to an inaccessible ANA state and with that nshead is created but it's state is not transitioned to LIVE. So the current logic would then causes nshead reference to be leaked for non-LIVE states. Another scenario, during namespace allocation, the driver first allocates a nshead and then issues an Identify Namespace command. If this command fails — which can happen in tests like nvme/058 that rapidly enables and disables namespaces — we must release the reference to the newly allocated nshead. However this reference release is currently missing in the failure, causing a nshead reference leak. To fix this, we now unconditionally release the nshead reference when the last nvme path referencing to the nshead is removed, regardless of the head’s state. Also during identify namespace failure case we now properly release the nshead reference. So this ensures proper cleanup of the nshead, and consequently, the NVMe subsystem and its associated kobject. This change prevents stale kobject entries from lingering in sysfs and eliminates the module reload failures observed just after running nvme/058. [1] https://lore.kernel.org/all/CAHj4cs8fOBS-eSjsd5LUBzy7faKXJtgLkCN+mDy_-ezCLLLq+Q@mail.gmail.com/ Reported-by: yi.zhang@redhat.com Closes: https://lore.kernel.org/all/CAHj4cs8fOBS-eSjsd5LUBzy7faKXJtgLkCN+mDy_-ezCLLLq+Q@mail.gmail.com/ Fixes: 62188639ec16 ("nvme-multipath: introduce delayed removal of the multipath head node") Tested-by: yi.zhang@redhat.com Signed-off-by: Nilay Shroff Reviewed-by: Hannes Reinecke Signed-off-by: Christoph Hellwig

nvme: Fix incorrect cdw15 value in passthru error logging

2025-06-30T06:31:45Z

Fix an error in nvme_log_err_passthru() where cdw14 was incorrectly printed twice instead of cdw15. This fix ensures accurate logging of the full passthrough command payload. Fixes: 9f079dda1433 ("nvme: allow passthru cmd error logging") Signed-off-by: Alok Tiwari Reviewed-by: Martin K. Petersen Signed-off-by: Christoph Hellwig

nvme: fix atomic write size validation

2025-06-26T11:04:37Z

Don't mix the namespace and controller values, and validate the per-controller limit when probing the controller. This avoid spurious failures for controllers with namespaces that have different namespaces with different logical block sizes, or report the per-namespace values only for some namespaces. It also fixes a missing queue_limits_cancel_update in an error path by removing that error path. Fixes: 8695f060a029 ("nvme: all namespaces in a subsystem must adhere to a common atomic write size") Reported-by: Yi Zhang Signed-off-by: Christoph Hellwig Reviewed-by: Luis Chamberlain Reviewed-by: John Garry Tested-by: Yi Zhang

nvme: refactor the atomic write unit detection

2025-06-26T11:04:37Z

Move all the code out of nvme_update_disk_info into the helper, and rename the helper to have a somewhat less clumsy name. Signed-off-by: Christoph Hellwig Reviewed-by: Luis Chamberlain Reviewed-by: John Garry

nvme: reset delayed remove_work after reconnect

2025-06-26T11:04:35Z

The remove_work will proceed with permanently disconnecting on the initial final path failure if the head shows no paths after the delay. If a new path connects while the remove_work is pending, and if that new path happens to disconnect before that remove_work executes, the delayed removal should reset based on the most recent path disconnect time, but queue_delayed_work() won't do anything if the work is already pending. Attempt to cancel the delayed work when a new path connects, and use mod_delayed_work() in case the remove_work remains pending anyway. Signed-off-by: Keith Busch Reviewed-by: Nilay Shroff Signed-off-by: Christoph Hellwig