nvme-fc: fix double free of fabrics options when nvme_add_ctrl() fails
Subsystems (2)
PATCH v2 1/1 nvme-fc: fix double free of fabrics options when nvme_add_ctrl() fails
nvmf_create_ctrl() frees opts when ->create_ctrl() returns an error, so a transport must not free it on its own error paths. nvme_fc_ctrl_free() therefore only calls nvmf_free_options() while ctrl->ctrl.opts is still set, and nvme_fc_init_ctrl() clears that pointer before its last put.
It only does so on the fail_ctrl: path. When nvme_add_ctrl() fails, nvme_fc_init_ctrl() jumps to out_put_ctrl: instead, so nvme_fc_ctrl_free() still sees ctrl->ctrl.opts set and frees opts, and nvmf_create_ctrl() frees it again.
Reproduced with nvme-fcloop and failslab by failing the kvasprintf() in dev_set_name(), called from nvme_add_ctrl():
BUG: KASAN: slab-use-after-free in nvmf_free_options+0x30/0x190 nvmf_free_options+0x30/0x190 drivers/nvme/host/fabrics.c:1284 nvmf_create_ctrl drivers/nvme/host/fabrics.c:1374 [inline] Freed by task 5534: nvme_fc_ctrl_free drivers/nvme/host/fc.c:2374 [inline] nvme_fc_init_ctrl+0xe17/0x1450 drivers/nvme/host/fc.c:3605
Without KASAN, opts is freed twice.
nvme-tcp and nvme-rdma reach the same error path, but their free_ctrl only frees opts once the controller is on the global list, so they are not affected.
Move the clear down to out_put_ctrl:, which both error paths pass through. The same injection then returns -EIO without a report.
---
Changes since v1: move the existing clear from fail_ctrl: down toout_put_ctrl: rather than adding a second one (Keith). Re-tested by sweeping fail-nth 1..200 over the connect write with fcloop and failslab, which reaches both labels: unpatched reports at 17, this passes 200/200.
v1: https://lore.kernel.org/linux-nvme/[email protected]/
drivers/nvme/host/fc.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/nvme/host/fc.c +3 -2
@@ -3593,14 +3593,15 @@ nvme_fc_init_ctrl(struct device *dev, struct nvmf_ctrl_options *opts,cancel_work_sync (&ctrl->ctrl.reset_work);cancel_delayed_work_sync (&ctrl->connect_work); - ctrl->ctrl.opts =NULL ; -if (ctrl->ctrl.admin_tagset)nvme_remove_admin_tag_set (&ctrl->ctrl);/* initiate nvme ctrl ref counting teardown */ nvme_uninit_ctrl (&ctrl->ctrl); out_put_ctrl: +/* the caller frees opts when we return an error */ + ctrl->ctrl.opts =NULL ; +/* Remove core ctrl ref. */ nvme_put_ctrl (&ctrl->ctrl);
base-commit: 0ce37745d4bfbc493f718169c3974898ffec8ee7
Signature · 1 line
2.43.0
Hello Rihyeon,
Quoted 29 lines · Rihyeon Kim
nvmf_create_ctrl()frees opts when ->create_ctrl() returns an error, so a transport must not free it on its own error paths.nvme_fc_ctrl_free()therefore only callsnvmf_free_options()whilectrl->ctrl.optsis still set, andnvme_fc_init_ctrl()clears that pointer before its last put.It only does so on the
fail_ctrl: path. Whennvme_add_ctrl()fails,nvme_fc_init_ctrl()jumps toout_put_ctrl: instead, sonvme_fc_ctrl_free()still seesctrl->ctrl.optsset and frees opts, andnvmf_create_ctrl()frees it again.Reproduced with nvme-fcloop and failslab by failing the
kvasprintf()indev_set_name(), called fromnvme_add_ctrl():BUG: KASAN: slab-use-after-free in nvmf_free_options+0x30/0x190 nvmf_free_options+0x30/0x190 drivers/nvme/host/fabrics.c:1284 nvmf_create_ctrl drivers/nvme/host/fabrics.c:1374 [inline] Freed by task 5534: nvme_fc_ctrl_free drivers/nvme/host/fc.c:2374 [inline] nvme_fc_init_ctrl+0xe17/0x1450 drivers/nvme/host/fc.c:3605Without KASAN, opts is freed twice.
nvme-tcp and nvme-rdma reach the same error path, but their
free_ctrlonly frees opts once the controller is on the global list, so they are not affected.Move the clear down to
out_put_ctrl:, which both error paths pass through. The same injection then returns-EIOwithout a report.
Quoted 8 lines
Fixes 1a9e218195a5 ("nvme: split device add from initialization")Reported-by [email protected]Suggested-by Keith Busch <[email protected]>Assisted-by Claude:claude-opus-5Signed-off-by Rihyeon Kim <[email protected]>---
I created an alternative fix that makes fc behave the same way as tcp/rdma/loop, i.e. derive ownership from seeing if the ctrl is on the list or not:
https://lore.kernel.org/linux-nvme/[email protected]/T/#u
This has the advantage of avoiding the NULL pointer dereferences in nvme_auth_free() and when accessing the sysfs attributes during teardown, as reported by Sashiko, as ctrl->ctrl.opts now stays valid for the whole teardown.
Please review.
Kind regards,
Niklas