[PATCH 0/2] scsi: bsg: fix io_uring passthrough issues
Yang Xiuwei <[email protected]> Mon, 20 Jul 2026 11:23:36 +0800
| Newsgroups | org.kernel.vger.io-uring,org.kernel.vger.linux-scsi |
|---|---|
| Message-ID | <[email protected]> |
Hi, Two fixes for the BSG io_uring passthrough path. Patch 1 addresses a TOCTOU where the shared SQE can change request_len between validation and copy_from_user(), overflowing scmd->cmnd. This is a v3 of Rahul Chandelkar's series, taking over after no follow-up to v2. Relative to v2, READ_ONCE is limited to fields used for validation and buffer mapping, as requested in review. Link: https://lore.kernel.org/r/[email protected] Link: https://lore.kernel.org/r/[email protected] Patch 2 uses GFP_KERNEL for user-buffer mapping. IO_URING_F_NONBLOCK still sets BLK_MQ_REQ_NOWAIT so request-tag allocation does not block; mapping itself may sleep and need not use GFP_NOWAIT. Changes for patch 1 since Rahul's v2: - READ_ONCE only request/request_len and dout/din xfer fields. - Pass stable map arguments into scsi_bsg_map_user_buffer(). Please review. Thanks, Yang Xiuwei Rahul Chandelkar (1): scsi: bsg: fix TOCTOU in io_uring passthrough command setup Yang Xiuwei (1): scsi: bsg: map io_uring user buffers with GFP_KERNEL drivers/scsi/scsi_bsg.c | 47 +++++++++++++++++++++++------------------ 1 file changed, 27 insertions(+), 20 deletions(-) -- 2.25.1