CVE-2026-43054

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> scsi: target: tcm_loop: Drain commands in target_reset handler<br /> <br /> tcm_loop_target_reset() violates the SCSI EH contract: it returns SUCCESS<br /> without draining any in-flight commands. The SCSI EH documentation<br /> (scsi_eh.rst) requires that when a reset handler returns SUCCESS the driver<br /> has made lower layers "forget about timed out scmds" and is ready for new<br /> commands. Every other SCSI LLD (virtio_scsi, mpt3sas, ipr, scsi_debug,<br /> mpi3mr) enforces this by draining or completing outstanding commands before<br /> returning SUCCESS.<br /> <br /> Because tcm_loop_target_reset() doesn&amp;#39;t drain, the SCSI EH reuses in-flight<br /> scsi_cmnd structures for recovery commands (e.g. TUR) while the target core<br /> still has async completion work queued for the old se_cmd. The memset in<br /> queuecommand zeroes se_lun and lun_ref_active, causing<br /> transport_lun_remove_cmd() to skip its percpu_ref_put(). The leaked LUN<br /> reference prevents transport_clear_lun_ref() from completing, hanging<br /> configfs LUN unlink forever in D-state:<br /> <br /> INFO: task rm:264 blocked for more than 122 seconds.<br /> rm D 0 264 258 0x00004000<br /> Call Trace:<br /> __schedule+0x3d0/0x8e0<br /> schedule+0x36/0xf0<br /> transport_clear_lun_ref+0x78/0x90 [target_core_mod]<br /> core_tpg_remove_lun+0x28/0xb0 [target_core_mod]<br /> target_fabric_port_unlink+0x50/0x60 [target_core_mod]<br /> configfs_unlink+0x156/0x1f0 [configfs]<br /> vfs_unlink+0x109/0x290<br /> do_unlinkat+0x1d5/0x2d0<br /> <br /> Fix this by making tcm_loop_target_reset() actually drain commands:<br /> <br /> 1. Issue TMR_LUN_RESET via tcm_loop_issue_tmr() to drain all commands that<br /> the target core knows about (those not yet CMD_T_COMPLETE).<br /> <br /> 2. Use blk_mq_tagset_busy_iter() to iterate all started requests and<br /> flush_work() on each se_cmd — this drains any deferred completion work<br /> for commands that already had CMD_T_COMPLETE set before the TMR (which<br /> the TMR skips via __target_check_io_state()). This is the same pattern<br /> used by mpi3mr, scsi_debug, and libsas to drain outstanding commands<br /> during reset.