Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
RDMA/mlx5: Don't access ib_qp fields in internal destroy QP path
destroy_qp_common is called for flows where QP is already created by HW. While it is called from IB/core, the ibqp.* fields will be fully initialized, but it is not the case if this function is called during QP creation. Don't rely on ibqp fields as much as possible and initialize send_cq/recv_cq as temporal solution till all drivers will be converted to IB/core QP allocation scheme. refcount_t: underflow; use-after-free. WARNING: CPU: 1 PID: 5372 at lib/refcount.c:28 refcount_warn_saturate+0xfe/0x1a0 Kernel panic - not syncing: panic_on_warn set ... CPU: 1 PID: 5372 Comm: syz-executor.2 Not tainted 5.5.0-rc5 #2 Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.1-0-ga5cab58e9a3f-prebuilt.qemu.org 04/01/2014 Call Trace: mlx5_core_put_rsc+0x70/0x80 destroy_resource_common+0x8e/0xb0 mlx5_core_destroy_qp+0xaf/0x1d0 mlx5_ib_destroy_qp+0xeb0/0x1460 ib_destroy_qp_user+0x2d5/0x7d0 create_qp+0xed3/0x2130 ib_uverbs_create_qp+0x13e/0x190 ? ib_uverbs_ex_create_qp ib_uverbs_write+0xaa5/0xdf0 __vfs_write+0x7c/0x100 ksys_write+0xc8/0x200 do_syscall_64+0x9c/0x390 entry_SYSCALL_64_after_hwframe+0x44/0xa9 Fixes: 08d5397 ("RDMA/mlx5: Copy response to the user in one place") Link: https://lore.kernel.org/r/20200617130148.2846643-1-leon@kernel.org Signed-off-by: Leon Romanovsky <leonro@mellanox.com> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
- Loading branch information