CVE-2024-57804

Quoted from the kernel CNA's own scoring file, word for word. A highlighted metric carries the most severe value CVSS v3.1 defines for it, which is what pushes the score up. source

AV:LAttack vector, local

The vulnerable code is reached only through local sysfs attributes under /sys/class/sas_phy/ (phy error counters, enable, link rate) and the bsg device; there is no network-facing path into mpi3mr config-page processing.

AC:LAttack complexity, low

The attacker controls both sides of the race by concurrently reading the world-readable phy error-counter attributes across multiple phys/threads, each of which memsets and reuses the shared mrioc->cfg_page outside cfg_cmds.mutex; the commit itself notes the corruption occurs on operations "in rapid succession", with no attacker-uncontrollable precondition.

PR:LPrivileges required, low

The SAS transport exports invalid_dword_count, running_disparity_error_count, loss_of_dword_sync_count and phy_reset_problem_count with S_IRUGO (0444), and reading them invokes mpi3mr_transport_get_linkerrors() -> mpi3mr_cfg_get_sas_phy_pg1() -> mpi3mr_process_cfg_req(), so an unprivileged local user can drive the racy shared-buffer path and corrupt config requests issued by any other context, including a privileged WRITE_PERSISTENT.

UI:NUser interaction, none

The attacker's own concurrent sysfs reads race each other and race the driver's internal discovery/reset config-page reads; no victim action is required to trigger the corruption.

S:UScope, unchanged

The corrupted state is the HBA driver's DMA buffer and the controller configuration it manages, all within the kernel's own security authority; no VM, IOMMU, or sandbox boundary is crossed.

C:LConfidentiality impact, low

A thread's post-mutex read-out memcpy can return a different config page's contents (device SAS addresses/WWNs, enclosure identifiers, expander topology, driver page settings), which are then surfaced to unprivileged userspace through the world-readable phy sysfs attributes; the leak is bounded to the 1024-byte shared config buffer rather than arbitrary kernel memory.

I:HIntegrity impact, high

mpi3mr_cfg_set_sas_io_unit_pg1() performs WRITE_CURRENT and WRITE_PERSISTENT with a payload staged in the unlocked shared buffer, so a racing reader causes zeroed/foreign data to be DMA'd into the controller's non-volatile SAS IO Unit / SAS Expander configuration, and garbled reads poison the driver's SAS topology state (device handles, SAS addresses, port IDs) used for I/O routing — a direct and serious integrity loss.

A:HAvailability impact, high

A corrupted persistent SAS IO Unit Page 1 leaves PHYs disabled or with invalid link rates, so storage behind the HBA fails to come back and the misconfiguration survives reboot; additionally a botched config request drives mpi3mr_check_rh_fault_ioc(MPI3MR_RESET_FROM_CFG_REQ_TIMEOUT), diag-faulting and resetting the controller and potentially marking it unrecoverable, stalling all attached I/O.