[PATCH 6.1.y 0/8] nvme: correctly fix admin request_queue lifetime

[PATCH 6.1.y 0/8] nvme: correctly fix admin request_queue lifetime

[Heyne, Maximilian]

The initial attempt to backport upstream commit 03b3bcd319b3 ("nvme: fix
admin request_queue lifetime") was not correct leading to refcount
underflows and not even fixing the problem.

I've tested the reproduction steps from [1] (adding a delay to
nvme_submit_user_cmd and 'echo 1 | sudo tee
/sys/class/nvme/nvme0/delete_controller') on the nvme-tcp driver which
printed the KASAN UAF blurb.

Fixing the issue in the 6.1 series requires a few dependent patches.
This is mainly the upstream commit 2b3f056f72e5 ("blk-mq: move the call
to blk_put_queue out of blk_mq_destroy_queue") which allows to move the
blk_put_queue to a different location. While at it, I'm backporting the
whole patch series for completeness. However, the scsi and apple patches
are not strictly required and could be dropped from this series.

The backport of commit 03b3bcd319b3 ("nvme: fix admin
request_queue lifetime") needed a tweak to the nvme pci driver.

Furthermore, in this patch series I've also included a follow-up fixup
from upstream commit b84bb7bd913d ("nvme: fix admin queue leak on
controller reset"), again with an adaption to the nvme pci driver. This
issue could easily be reproduced by resetting the controller (no need to
run full blktests):

  echo 1 > /sys/class/nvme/nvme0/reset_controller

[1] https://lore.kernel.org/all/20251029210853.20768-1-cachen@purestorage.com/

Christoph Hellwig (5):
  blk-mq: move the call to blk_put_queue out of blk_mq_destroy_queue
  scsi: remove an extra queue reference
  nvme-pci: remove an extra queue reference
  nvme-apple: remove an extra queue reference
  nvme-pci: put the admin queue in nvme_dev_remove_admin

Keith Busch (1):
  nvme: fix admin request_queue lifetime

Maximilian Heyne (1):
  Revert "nvme: fix admin request_queue lifetime"

Ming Lei (1):
  nvme: fix admin queue leak on controller reset

 block/blk-mq.c            |  4 +---
 block/bsg-lib.c           |  2 ++
 drivers/nvme/host/apple.c |  8 --------
 drivers/nvme/host/core.c  | 16 ++++++++++++++--
 drivers/nvme/host/pci.c   | 14 +++++++-------
 drivers/scsi/scsi_scan.c  |  1 -
 drivers/ufs/core/ufshcd.c |  2 ++
 7 files changed, 26 insertions(+), 21 deletions(-)

-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 1/8] Revert "nvme: fix admin request_queue lifetime"

[Heyne, Maximilian]

This reverts commit ff037b5f47eeccc1636c03f84cd47db094eb73c9.

The backport of upstream commit 03b3bcd319b3 ("nvme: fix admin
request_queue lifetime") to 6.1 is broken in 2 ways. First of all it
doesn't actually fix the issue because blk_put_queue will still be
called as part of blk_mq_destroy_queue in nvme_remove_admin_tag_set
leading to the UAF.
Second, the backport leads to a refcount underflow when unbinding a pci
nvme device:

 refcount_t: underflow; use-after-free.
 WARNING: CPU: 2 PID: 1486 at lib/refcount.c:28 refcount_warn_saturate+0xba/0x110
 Modules linked in: bochs drm_vram_helper simpledrm skx_edac_common drm_shmem_helper drm_kms_helper kvm_intel cfbfillrect syscopyarea cfbimgblt sysfillrect sysimgblt fb_sys_fops cfbcopyarea drm_ttm_helper fb ttm kvm fbdev drm mousedev nls_ascii psmouse irqbypass nls_cp437 atkbd crc32_pclmul crc32c_intel libps2 vfat fat sunrpc virtio_net ata_piix vivaldi_fmap drm_panel_orientation_quirks libata backlight i2c_piix4 net_failover i8042 ghash_clmulni_intel failover serio i2c_core button sch_fq_codel
 CPU: 2 PID: 1486 Comm: bash Not tainted 6.1.167 #2
 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS edk2-20240813-306.amzn2 08/13/2024
 RIP: 0010:refcount_warn_saturate+0xba/0x110
 Code: 01 01 e8 89 79 ad ff 0f 0b e9 82 f4 7e 00 80 3d 73 03 cc 01 00 75 85 48 c7 c7 e0 5d 3b 8e c6 05 63 03 cc 01 01 e8 66 79 ad ff <0f> 0b c3 cc cc cc cc 80 3d 4e 03 cc 01 00 0f 85 5e ff ff ff 48 c7
 RSP: 0018:ffffd0cc011bfd18 EFLAGS: 00010286
 RAX: 0000000000000000 RBX: ffff8ada07b33210 RCX: 0000000000000027
 RDX: ffff8adb37d1f728 RSI: 0000000000000001 RDI: ffff8adb37d1f720
 RBP: ffff8ada07b33000 R08: 0000000000000000 R09: 00000000fffeffff
 R10: ffffd0cc011bfba8 R11: ffffffff8f1781a8 R12: ffffd0cc011bfd38
 R13: ffff8ada03080800 R14: ffff8ada07b33210 R15: ffff8ada07b33b10
 FS:  00007f50f6964740(0000) GS:ffff8adb37d00000(0000) knlGS:0000000000000000
 CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
 CR2: 000055cdb54e6ae0 CR3: 000000010224e001 CR4: 0000000000770ee0
 DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
 DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
 PKRU: 55555554
 Call Trace:
  <TASK>
  nvme_pci_free_ctrl+0x45/0x80
  nvme_free_ctrl+0x1aa/0x2b0
  device_release+0x34/0x90
  kobject_cleanup+0x3a/0x130
  pci_device_remove+0x3e/0xb0
  device_release_driver_internal+0x1aa/0x230
  unbind_store+0x11f/0x130
  kernfs_fop_write_iter+0x13a/0x1d0
  vfs_write+0x2a6/0x3b0
  ksys_write+0x5f/0xe0
  do_syscall_64+0x35/0x80
  entry_SYSCALL_64_after_hwframe+0x6e/0xd8
 RIP: 0033:0x7f50f66ff897
 Code: 0f 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff eb b7 0f 1f 00 f3 0f 1e fa 64 8b 04 25 18 00 00 00 85 c0 75 10 b8 01 00 00 00 0f 05 <48> 3d 00 f0 ff ff 77 51 c3 48 83 ec 28 48 89 54 24 18 48 89 74 24
 RSP: 002b:00007fffaef903d8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
 RAX: ffffffffffffffda RBX: 00007f50f67fd780 RCX: 00007f50f66ff897
 RDX: 000000000000000d RSI: 0000557f72ef6b90 RDI: 0000000000000001
 RBP: 000000000000000d R08: 0000000000000000 R09: 00007f50f67b2d20
 R10: 00007f50f67b2c20 R11: 0000000000000246 R12: 000000000000000d
 R13: 0000557f72ef6b90 R14: 000000000000000d R15: 00007f50f67f89c0
  </TASK>

The reason for this is that nvme_free_ctrl calls ->free_ctrl which
resolves to nvme_pci_free_ctrl in aforementioned case which also has a
blk_put_queue, so the admin queue is put twice. This is because on 6.1
we're missing the commit 96ef1be53663 ("nvme-pci: put the admin queue in
nvme_dev_remove_admin").

Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/nvme/host/core.c | 2 --
 1 file changed, 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 9df33b293ee3e..938af571dc13e 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -5180,8 +5180,6 @@ static void nvme_free_ctrl(struct device *dev)
 		container_of(dev, struct nvme_ctrl, ctrl_device);
 	struct nvme_subsystem *subsys = ctrl->subsys;
 
-	if (ctrl->admin_q)
-		blk_put_queue(ctrl->admin_q);
 	if (!subsys || ctrl->instance != subsys->instance)
 		ida_free(&nvme_instance_ida, ctrl->instance);
 
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 2/8] blk-mq: move the call to blk_put_queue out of blk_mq_destroy_queue

[Heyne, Maximilian]

From: Christoph Hellwig <hch@lst.de>

[ Upstream commit 2b3f056f72e56fa07df69b4705e0b46a6c08e77c ]

The fact that blk_mq_destroy_queue also drops a queue reference leads
to various places having to grab an extra reference.  Move the call to
blk_put_queue into the callers to allow removing the extra references.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-2-hch@lst.de
[axboe: fix fabrics_q vs admin_q conflict in nvme core.c]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Stable-dep-of: 03b3bcd319b3 ("nvme: fix admin request_queue lifetime")
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 block/blk-mq.c            |  4 +---
 block/bsg-lib.c           |  2 ++
 drivers/nvme/host/apple.c |  1 +
 drivers/nvme/host/core.c  | 10 ++++++++--
 drivers/nvme/host/pci.c   |  1 +
 drivers/scsi/scsi_sysfs.c |  1 +
 drivers/ufs/core/ufshcd.c |  2 ++
 7 files changed, 16 insertions(+), 5 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index a9697541d67f9..8b9e5ca398242 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -4194,9 +4194,6 @@ void blk_mq_destroy_queue(struct request_queue *q)
 	blk_sync_queue(q);
 	blk_mq_cancel_work_sync(q);
 	blk_mq_exit_queue(q);
-
-	/* @q is and will stay empty, shutdown and put */
-	blk_put_queue(q);
 }
 EXPORT_SYMBOL(blk_mq_destroy_queue);
 
@@ -4213,6 +4210,7 @@ struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
 	disk = __alloc_disk_node(q, set->numa_node, lkclass);
 	if (!disk) {
 		blk_mq_destroy_queue(q);
+		blk_put_queue(q);
 		return ERR_PTR(-ENOMEM);
 	}
 	set_bit(GD_OWNS_QUEUE, &disk->state);
diff --git a/block/bsg-lib.c b/block/bsg-lib.c
index d6f5dcdce748c..435c32373cd68 100644
--- a/block/bsg-lib.c
+++ b/block/bsg-lib.c
@@ -325,6 +325,7 @@ void bsg_remove_queue(struct request_queue *q)
 
 		bsg_unregister_queue(bset->bd);
 		blk_mq_destroy_queue(q);
+		blk_put_queue(q);
 		blk_mq_free_tag_set(&bset->tag_set);
 		kfree(bset);
 	}
@@ -400,6 +401,7 @@ struct request_queue *bsg_setup_queue(struct device *dev, const char *name,
 	return q;
 out_cleanup_queue:
 	blk_mq_destroy_queue(q);
+	blk_put_queue(q);
 out_queue:
 	blk_mq_free_tag_set(set);
 out_tag_set:
diff --git a/drivers/nvme/host/apple.c b/drivers/nvme/host/apple.c
index 262d2b60ac6dd..c5fc293c22123 100644
--- a/drivers/nvme/host/apple.c
+++ b/drivers/nvme/host/apple.c
@@ -1510,6 +1510,7 @@ static int apple_nvme_probe(struct platform_device *pdev)
 	if (!blk_get_queue(anv->ctrl.admin_q)) {
 		nvme_start_admin_queue(&anv->ctrl);
 		blk_mq_destroy_queue(anv->ctrl.admin_q);
+		blk_put_queue(anv->ctrl.admin_q);
 		anv->ctrl.admin_q = NULL;
 		ret = -ENODEV;
 		goto put_dev;
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 938af571dc13e..044e1a9c099b3 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -5031,6 +5031,7 @@ int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
 
 out_cleanup_admin_q:
 	blk_mq_destroy_queue(ctrl->admin_q);
+	blk_put_queue(ctrl->admin_q);
 out_free_tagset:
 	blk_mq_free_tag_set(set);
 	ctrl->admin_q = NULL;
@@ -5042,8 +5043,11 @@ EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
 {
 	blk_mq_destroy_queue(ctrl->admin_q);
-	if (ctrl->ops->flags & NVME_F_FABRICS)
+	blk_put_queue(ctrl->admin_q);
+	if (ctrl->ops->flags & NVME_F_FABRICS) {
 		blk_mq_destroy_queue(ctrl->fabrics_q);
+		blk_put_queue(ctrl->fabrics_q);
+	}
 	blk_mq_free_tag_set(ctrl->admin_tagset);
 }
 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
@@ -5099,8 +5103,10 @@ EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
 
 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
 {
-	if (ctrl->ops->flags & NVME_F_FABRICS)
+	if (ctrl->ops->flags & NVME_F_FABRICS) {
 		blk_mq_destroy_queue(ctrl->connect_q);
+		blk_put_queue(ctrl->connect_q);
+	}
 	blk_mq_free_tag_set(ctrl->tagset);
 }
 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 518f8c5012bdf..727585f580362 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1782,6 +1782,7 @@ static void nvme_dev_remove_admin(struct nvme_dev *dev)
 		 */
 		nvme_start_admin_queue(&dev->ctrl);
 		blk_mq_destroy_queue(dev->ctrl.admin_q);
+		blk_put_queue(dev->ctrl.admin_q);
 		blk_mq_free_tag_set(&dev->admin_tagset);
 	}
 }
diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
index 456b92c3a7811..af81b2ba0c9b3 100644
--- a/drivers/scsi/scsi_sysfs.c
+++ b/drivers/scsi/scsi_sysfs.c
@@ -1486,6 +1486,7 @@ void __scsi_remove_device(struct scsi_device *sdev)
 	mutex_unlock(&sdev->state_mutex);
 
 	blk_mq_destroy_queue(sdev->request_queue);
+	blk_put_queue(sdev->request_queue);
 	kref_put(&sdev->host->tagset_refcnt, scsi_mq_free_tags);
 	cancel_work_sync(&sdev->requeue_work);
 
diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c
index f72ba0b206437..a39ffc62d88a1 100644
--- a/drivers/ufs/core/ufshcd.c
+++ b/drivers/ufs/core/ufshcd.c
@@ -9651,6 +9651,7 @@ void ufshcd_remove(struct ufs_hba *hba)
 	ufshpb_remove(hba);
 	ufs_sysfs_remove_nodes(hba->dev);
 	blk_mq_destroy_queue(hba->tmf_queue);
+	blk_put_queue(hba->tmf_queue);
 	blk_mq_free_tag_set(&hba->tmf_tag_set);
 	scsi_remove_host(hba->host);
 	/* disable interrupts */
@@ -9953,6 +9954,7 @@ int ufshcd_init(struct ufs_hba *hba, void __iomem *mmio_base, unsigned int irq)
 
 free_tmf_queue:
 	blk_mq_destroy_queue(hba->tmf_queue);
+	blk_put_queue(hba->tmf_queue);
 free_tmf_tag_set:
 	blk_mq_free_tag_set(&hba->tmf_tag_set);
 out_remove_scsi_host:
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 3/8] scsi: remove an extra queue reference

[Heyne, Maximilian]

From: Christoph Hellwig <hch@lst.de>

[ Upstream commit dc917c361422388f0d39d3f0dc2bc5a188c01156 ]

Now that blk_mq_destroy_queue does not release the queue reference, there
is no need for a second queue reference to be held by the scsi_device.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/scsi/scsi_scan.c  | 1 -
 drivers/scsi/scsi_sysfs.c | 1 -
 2 files changed, 2 deletions(-)

diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index 696f178fb57d8..0eaec372f7560 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -344,7 +344,6 @@ static struct scsi_device *scsi_alloc_sdev(struct scsi_target *starget,
 	sdev->request_queue = q;
 	q->queuedata = sdev;
 	__scsi_init_queue(sdev->host, q);
-	WARN_ON_ONCE(!blk_get_queue(q));
 
 	depth = sdev->host->cmd_per_lun ?: 1;
 
diff --git a/drivers/scsi/scsi_sysfs.c b/drivers/scsi/scsi_sysfs.c
index af81b2ba0c9b3..456b92c3a7811 100644
--- a/drivers/scsi/scsi_sysfs.c
+++ b/drivers/scsi/scsi_sysfs.c
@@ -1486,7 +1486,6 @@ void __scsi_remove_device(struct scsi_device *sdev)
 	mutex_unlock(&sdev->state_mutex);
 
 	blk_mq_destroy_queue(sdev->request_queue);
-	blk_put_queue(sdev->request_queue);
 	kref_put(&sdev->host->tagset_refcnt, scsi_mq_free_tags);
 	cancel_work_sync(&sdev->requeue_work);
 
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 4/8] nvme-pci: remove an extra queue reference

[Heyne, Maximilian]

From: Christoph Hellwig <hch@lst.de>

[ Upstream commit 7dcebef90d35de13a326f765dd787538880566f9 ]

Now that blk_mq_destroy_queue does not release the queue reference, there
is no need for a second admin queue reference to be held by the nvme_dev.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Stable-dep-of: 03b3bcd319b3 ("nvme: fix admin request_queue lifetime")
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/nvme/host/pci.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 727585f580362..13c0098939ec0 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1782,7 +1782,6 @@ static void nvme_dev_remove_admin(struct nvme_dev *dev)
 		 */
 		nvme_start_admin_queue(&dev->ctrl);
 		blk_mq_destroy_queue(dev->ctrl.admin_q);
-		blk_put_queue(dev->ctrl.admin_q);
 		blk_mq_free_tag_set(&dev->admin_tagset);
 	}
 }
@@ -1811,11 +1810,6 @@ static int nvme_pci_alloc_admin_tag_set(struct nvme_dev *dev)
 		dev->ctrl.admin_q = NULL;
 		return -ENOMEM;
 	}
-	if (!blk_get_queue(dev->ctrl.admin_q)) {
-		nvme_dev_remove_admin(dev);
-		dev->ctrl.admin_q = NULL;
-		return -ENODEV;
-	}
 	return 0;
 }
 
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 5/8] nvme-apple: remove an extra queue reference

[Heyne, Maximilian]

From: Christoph Hellwig <hch@lst.de>

[ Upstream commit 941f7298c70c7668416e7845fa76eb72c07d966b ]

Now that blk_mq_destroy_queue does not release the queue reference, there
is no need for a second admin queue reference to be held by the
apple_nvme structure.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Sven Peter <sven@svenpeter.dev>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Link: https://lore.kernel.org/r/20221018135720.670094-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/nvme/host/apple.c | 9 ---------
 1 file changed, 9 deletions(-)

diff --git a/drivers/nvme/host/apple.c b/drivers/nvme/host/apple.c
index c5fc293c22123..c84ebfcfdeb88 100644
--- a/drivers/nvme/host/apple.c
+++ b/drivers/nvme/host/apple.c
@@ -1507,15 +1507,6 @@ static int apple_nvme_probe(struct platform_device *pdev)
 		goto put_dev;
 	}
 
-	if (!blk_get_queue(anv->ctrl.admin_q)) {
-		nvme_start_admin_queue(&anv->ctrl);
-		blk_mq_destroy_queue(anv->ctrl.admin_q);
-		blk_put_queue(anv->ctrl.admin_q);
-		anv->ctrl.admin_q = NULL;
-		ret = -ENODEV;
-		goto put_dev;
-	}
-
 	nvme_reset_ctrl(&anv->ctrl);
 	async_schedule(apple_nvme_async_probe, anv);
 
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 6/8] nvme-pci: put the admin queue in nvme_dev_remove_admin

[Heyne, Maximilian]

From: Christoph Hellwig <hch@lst.de>

[ Upstream commit 96ef1be53663a9343dffcf106e2f1b59da4b8799 ]

Once the controller is shutdown no one can access the admin queue.  Tear
it down in nvme_dev_remove_admin, which matches the flow in the other
drivers.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Tested-by Gerd Bayer <gbayer@linxu.ibm.com>
Stable-dep-of: 03b3bcd319b3 ("nvme: fix admin request_queue lifetime")
[ Context change due to missing commit 94cc781f69f4 ("nvme: move OPAL
  setup from PCIe to core")]
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/nvme/host/pci.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 13c0098939ec0..38732c0c28bbb 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1782,6 +1782,7 @@ static void nvme_dev_remove_admin(struct nvme_dev *dev)
 		 */
 		nvme_start_admin_queue(&dev->ctrl);
 		blk_mq_destroy_queue(dev->ctrl.admin_q);
+		blk_put_queue(dev->ctrl.admin_q);
 		blk_mq_free_tag_set(&dev->admin_tagset);
 	}
 }
@@ -2831,8 +2832,6 @@ static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl)
 
 	nvme_dbbuf_dma_free(dev);
 	nvme_free_tagset(dev);
-	if (dev->ctrl.admin_q)
-		blk_put_queue(dev->ctrl.admin_q);
 	free_opal_dev(dev->ctrl.opal_dev);
 	mempool_destroy(dev->iod_mempool);
 	put_device(dev->dev);
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 7/8] nvme: fix admin request_queue lifetime

[Heyne, Maximilian]

From: Keith Busch <kbusch@kernel.org>

[ Upstream commit 03b3bcd319b3ab5182bc9aaa0421351572c78ac0]

The namespaces can access the controller's admin request_queue, and
stale references on the namespaces may exist after tearing down the
controller. Ensure the admin request_queue is active by moving the
controller's 'put' to after all controller references have been released
to ensure no one is can access the request_queue. This fixes a reported
use-after-free bug:

  BUG: KASAN: slab-use-after-free in blk_queue_enter+0x41c/0x4a0
  Read of size 8 at addr ffff88c0a53819f8 by task nvme/3287
  CPU: 67 UID: 0 PID: 3287 Comm: nvme Tainted: G            E       6.13.2-ga1582f1a031e #15
  Tainted: [E]=UNSIGNED_MODULE
  Hardware name: Jabil /EGS 2S MB1, BIOS 1.00 06/18/2025
  Call Trace:
   <TASK>
   dump_stack_lvl+0x4f/0x60
   print_report+0xc4/0x620
   ? _raw_spin_lock_irqsave+0x70/0xb0
   ? _raw_read_unlock_irqrestore+0x30/0x30
   ? blk_queue_enter+0x41c/0x4a0
   kasan_report+0xab/0xe0
   ? blk_queue_enter+0x41c/0x4a0
   blk_queue_enter+0x41c/0x4a0
   ? __irq_work_queue_local+0x75/0x1d0
   ? blk_queue_start_drain+0x70/0x70
   ? irq_work_queue+0x18/0x20
   ? vprintk_emit.part.0+0x1cc/0x350
   ? wake_up_klogd_work_func+0x60/0x60
   blk_mq_alloc_request+0x2b7/0x6b0
   ? __blk_mq_alloc_requests+0x1060/0x1060
   ? __switch_to+0x5b7/0x1060
   nvme_submit_user_cmd+0xa9/0x330
   nvme_user_cmd.isra.0+0x240/0x3f0
   ? force_sigsegv+0xe0/0xe0
   ? nvme_user_cmd64+0x400/0x400
   ? vfs_fileattr_set+0x9b0/0x9b0
   ? cgroup_update_frozen_flag+0x24/0x1c0
   ? cgroup_leave_frozen+0x204/0x330
   ? nvme_ioctl+0x7c/0x2c0
   blkdev_ioctl+0x1a8/0x4d0
   ? blkdev_common_ioctl+0x1930/0x1930
   ? fdget+0x54/0x380
   __x64_sys_ioctl+0x129/0x190
   do_syscall_64+0x5b/0x160
   entry_SYSCALL_64_after_hwframe+0x4b/0x53
  RIP: 0033:0x7f765f703b0b
  Code: ff ff ff 85 c0 79 9b 49 c7 c4 ff ff ff ff 5b 5d 4c 89 e0 41 5c c3 66 0f 1f 84 00 00 00 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d dd 52 0f 00 f7 d8 64 89 01 48
  RSP: 002b:00007ffe2cefe808 EFLAGS: 00000202 ORIG_RAX: 0000000000000010
  RAX: ffffffffffffffda RBX: 00007ffe2cefe860 RCX: 00007f765f703b0b
  RDX: 00007ffe2cefe860 RSI: 00000000c0484e41 RDI: 0000000000000003
  RBP: 0000000000000000 R08: 0000000000000003 R09: 0000000000000000
  R10: 00007f765f611d50 R11: 0000000000000202 R12: 0000000000000003
  R13: 00000000c0484e41 R14: 0000000000000001 R15: 00007ffe2cefea60
   </TASK>

Reported-by: Casey Chen <cachen@purestorage.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
[ Because we're missing commit 0da7feaa5913 ("nvme-pci: use the tagset
  alloc/free helpers") we need to additionally remove the blk_put_queue
  from nvme_dev_remove_admin in pci.c to properly fix the UAF ]
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/nvme/host/core.c | 3 ++-
 drivers/nvme/host/pci.c  | 1 -
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 044e1a9c099b3..f17318f6c82b0 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -5043,7 +5043,6 @@ EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
 {
 	blk_mq_destroy_queue(ctrl->admin_q);
-	blk_put_queue(ctrl->admin_q);
 	if (ctrl->ops->flags & NVME_F_FABRICS) {
 		blk_mq_destroy_queue(ctrl->fabrics_q);
 		blk_put_queue(ctrl->fabrics_q);
@@ -5186,6 +5185,8 @@ static void nvme_free_ctrl(struct device *dev)
 		container_of(dev, struct nvme_ctrl, ctrl_device);
 	struct nvme_subsystem *subsys = ctrl->subsys;
 
+	if (ctrl->admin_q)
+		blk_put_queue(ctrl->admin_q);
 	if (!subsys || ctrl->instance != subsys->instance)
 		ida_free(&nvme_instance_ida, ctrl->instance);
 
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 38732c0c28bbb..e8b7b0004086c 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1782,7 +1782,6 @@ static void nvme_dev_remove_admin(struct nvme_dev *dev)
 		 */
 		nvme_start_admin_queue(&dev->ctrl);
 		blk_mq_destroy_queue(dev->ctrl.admin_q);
-		blk_put_queue(dev->ctrl.admin_q);
 		blk_mq_free_tag_set(&dev->admin_tagset);
 	}
 }
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

[PATCH 6.1.y 8/8] nvme: fix admin queue leak on controller reset

[Heyne, Maximilian]

From: Ming Lei <ming.lei@redhat.com>

[ Upstream commit b84bb7bd913d8ca2f976ee6faf4a174f91c02b8d ]

When nvme_alloc_admin_tag_set() is called during a controller reset,
a previous admin queue may still exist. Release it properly before
allocating a new one to avoid orphaning the old queue.

This fixes a regression introduced by commit 03b3bcd319b3 ("nvme: fix
admin request_queue lifetime").

Cc: Keith Busch <kbusch@kernel.org>
Fixes: 03b3bcd319b3 ("nvme: fix admin request_queue lifetime").
Reported-and-tested-by: Yi Zhang <yi.zhang@redhat.com>
Closes: https://lore.kernel.org/linux-block/CAHj4cs9wv3SdPo+N01Fw2SHBYDs9tj2M_e1-GdQOkRy=DsBB1w@mail.gmail.com/
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
[ Have to do analogous work in nvme_pci_alloc_admin_tag_set in pci.c due
  to missing upstream commit 0da7feaa5913 ("nvme-pci: use the tagset
  alloc/free helpers") ]
Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
---
 drivers/nvme/host/core.c | 7 +++++++
 drivers/nvme/host/pci.c  | 7 +++++++
 2 files changed, 14 insertions(+)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index f17318f6c82b0..09439fa7d083a 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -5012,6 +5012,13 @@ int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
 	if (ret)
 		return ret;
 
+	/*
+	 * If a previous admin queue exists (e.g., from before a reset),
+	 * put it now before allocating a new one to avoid orphaning it.
+	 */
+	if (ctrl->admin_q)
+		blk_put_queue(ctrl->admin_q);
+
 	ctrl->admin_q = blk_mq_init_queue(set);
 	if (IS_ERR(ctrl->admin_q)) {
 		ret = PTR_ERR(ctrl->admin_q);
diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index e8b7b0004086c..07ca1e1d920b8 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -1804,6 +1804,13 @@ static int nvme_pci_alloc_admin_tag_set(struct nvme_dev *dev)
 		return -ENOMEM;
 	dev->ctrl.admin_tagset = set;
 
+	/*
+	 * If a previous admin queue exists (e.g., from before a reset),
+	 * put it now before allocating a new one to avoid orphaning it.
+	 */
+	if (dev->ctrl.admin_q)
+		blk_put_queue(dev->ctrl.admin_q);
+
 	dev->ctrl.admin_q = blk_mq_init_queue(set);
 	if (IS_ERR(dev->ctrl.admin_q)) {
 		blk_mq_free_tag_set(set);
-- 
2.50.1




Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

Re: [PATCH 6.1.y 5/8] nvme-apple: remove an extra queue reference

[Fedor Pchelkin]

Hello,

"Heyne, Maximilian" <mheyne@amazon.de> wrote:
> From: Christoph Hellwig <hch@lst.de>
> 
> [ Upstream commit 941f7298c70c7668416e7845fa76eb72c07d966b ]
> 
> Now that blk_mq_destroy_queue does not release the queue reference, there
> is no need for a second admin queue reference to be held by the
> apple_nvme structure.

This patch is probably buggy in upstream.  It removes extra reference
->get, but doesn't remove the corresponding ->put which is located
inside apple_nvme_free_ctrl().

I'm reporting here currently just for the heads up - was looking at the
same nvme regression problem at 6.1.y, found this thread, and the
nvme-apple changes appeared suspicious.

nvme-apple patch is not required to fix the regression (this also holds
true for [PATCH 6.1.y 3/8] scsi: remove an extra queue reference).  Maybe
they shouldn't go to stable.

That said, the other part of the backport series FWIW looks good to me,
and I've also verified it resolves the 6.1.y regression.

Thanks.

> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
> Reviewed-by: Sven Peter <sven@svenpeter.dev>
> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
> Reviewed-by: Keith Busch <kbusch@kernel.org>
> Link: https://lore.kernel.org/r/20221018135720.670094-5-hch@lst.de
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
> ---
>  drivers/nvme/host/apple.c | 9 ---------
>  1 file changed, 9 deletions(-)
> 
> diff --git a/drivers/nvme/host/apple.c b/drivers/nvme/host/apple.c
> index c5fc293c22123..c84ebfcfdeb88 100644
> --- a/drivers/nvme/host/apple.c
> +++ b/drivers/nvme/host/apple.c
> @@ -1507,15 +1507,6 @@ static int apple_nvme_probe(struct platform_device *pdev)
>  		goto put_dev;
>  	}
>  
> -	if (!blk_get_queue(anv->ctrl.admin_q)) {
> -		nvme_start_admin_queue(&anv->ctrl);
> -		blk_mq_destroy_queue(anv->ctrl.admin_q);
> -		blk_put_queue(anv->ctrl.admin_q);
> -		anv->ctrl.admin_q = NULL;
> -		ret = -ENODEV;
> -		goto put_dev;
> -	}
> -
>  	nvme_reset_ctrl(&anv->ctrl);
>  	async_schedule(apple_nvme_async_probe, anv);
>  
> -- 
> 2.50.1

Re: [PATCH 6.1.y 5/8] nvme-apple: remove an extra queue reference

[Heyne, Maximilian]

On Wed, Apr 01, 2026 at 11:45:57PM +0300, Fedor Pchelkin wrote:
> Hello,
> 
> "Heyne, Maximilian" <mheyne@amazon.de> wrote:
> > From: Christoph Hellwig <hch@lst.de>
> > 
> > [ Upstream commit 941f7298c70c7668416e7845fa76eb72c07d966b ]
> > 
> > Now that blk_mq_destroy_queue does not release the queue reference, there
> > is no need for a second admin queue reference to be held by the
> > apple_nvme structure.
> 
> This patch is probably buggy in upstream.  It removes extra reference
> ->get, but doesn't remove the corresponding ->put which is located
> inside apple_nvme_free_ctrl().

Now I'm seeing this as well. Has the same problem as the pci driver in
6.1 where blk_put_queue is called from nvme_free_ctrl() and again from
apple_nvme_free_ctrl(). Thank you for catching this. I don't have the
hardware to test this.

Are you going to send a fix upstream? It's looks to be broken on master,
too.

> 
> I'm reporting here currently just for the heads up - was looking at the
> same nvme regression problem at 6.1.y, found this thread, and the
> nvme-apple changes appeared suspicious.
> 
> nvme-apple patch is not required to fix the regression (this also holds
> true for [PATCH 6.1.y 3/8] scsi: remove an extra queue reference).  Maybe
> they shouldn't go to stable.

I think, I'll send a v2 of the patch set without these 2 patches. It's
probably easier for Greg to apply.

> 
> That said, the other part of the backport series FWIW looks good to me,
> and I've also verified it resolves the 6.1.y regression.

You may leave a Tested-by if you want ;-)

> 
> Thanks.
> 
> > 
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
> > Reviewed-by: Sven Peter <sven@svenpeter.dev>
> > Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
> > Reviewed-by: Keith Busch <kbusch@kernel.org>
> > Link: https://lore.kernel.org/r/20221018135720.670094-5-hch@lst.de
> > Signed-off-by: Jens Axboe <axboe@kernel.dk>
> > Signed-off-by: Maximilian Heyne <mheyne@amazon.de>
> > ---
> >  drivers/nvme/host/apple.c | 9 ---------
> >  1 file changed, 9 deletions(-)
> > 
> > diff --git a/drivers/nvme/host/apple.c b/drivers/nvme/host/apple.c
> > index c5fc293c22123..c84ebfcfdeb88 100644
> > --- a/drivers/nvme/host/apple.c
> > +++ b/drivers/nvme/host/apple.c
> > @@ -1507,15 +1507,6 @@ static int apple_nvme_probe(struct platform_device *pdev)
> >  		goto put_dev;
> >  	}
> >  
> > -	if (!blk_get_queue(anv->ctrl.admin_q)) {
> > -		nvme_start_admin_queue(&anv->ctrl);
> > -		blk_mq_destroy_queue(anv->ctrl.admin_q);
> > -		blk_put_queue(anv->ctrl.admin_q);
> > -		anv->ctrl.admin_q = NULL;
> > -		ret = -ENODEV;
> > -		goto put_dev;
> > -	}
> > -
> >  	nvme_reset_ctrl(&anv->ctrl);
> >  	async_schedule(apple_nvme_async_probe, anv);
> >  
> > -- 
> > 2.50.1



Amazon Web Services Development Center Germany GmbH
Tamara-Danz-Str. 13
10243 Berlin
Geschaeftsfuehrung: Christof Hellmis, Andreas Stieger
Eingetragen am Amtsgericht Charlottenburg unter HRB 257764 B
Sitz: Berlin
Ust-ID: DE 365 538 597

Re: [PATCH 6.1.y 5/8] nvme-apple: remove an extra queue reference

[Fedor Pchelkin]

On Thu, 02. Apr 12:31, Heyne, Maximilian wrote:
> On Wed, Apr 01, 2026 at 11:45:57PM +0300, Fedor Pchelkin wrote:
> > Hello,
> > 
> > "Heyne, Maximilian" <mheyne@amazon.de> wrote:
> > > From: Christoph Hellwig <hch@lst.de>
> > > 
> > > [ Upstream commit 941f7298c70c7668416e7845fa76eb72c07d966b ]
> > > 
> > > Now that blk_mq_destroy_queue does not release the queue reference, there
> > > is no need for a second admin queue reference to be held by the
> > > apple_nvme structure.
> > 
> > This patch is probably buggy in upstream.  It removes extra reference
> > ->get, but doesn't remove the corresponding ->put which is located
> > inside apple_nvme_free_ctrl().
> 
> Now I'm seeing this as well. Has the same problem as the pci driver in
> 6.1 where blk_put_queue is called from nvme_free_ctrl() and again from
> apple_nvme_free_ctrl(). Thank you for catching this. I don't have the
> hardware to test this.
> 
> Are you going to send a fix upstream? It's looks to be broken on master,
> too.

I don't have the needed hardware either but will send the patch for review.

> > 
> > That said, the other part of the backport series FWIW looks good to me,
> > and I've also verified it resolves the 6.1.y regression.
> 
> You may leave a Tested-by if you want ;-)

I'll leave it for v2 then.