Hi,
On 07/05/2025 18:10, Tavian Barnes wrote:
hda_generic_machine_select() appends -idisp to the tplg filename by allocating a new string with devm_kasprintf(), then stores the string right back into the global variable snd_soc_acpi_intel_hda_machines. When the module is unloaded, this memory is freed, resulting in a global variable pointing to freed memory. Reloading the module then triggers a use-after-free:
BUG: KFENCE: use-after-free read in string+0x48/0xe0
Use-after-free read at 0x00000000967e0109 (in kfence-#99): string+0x48/0xe0 vsnprintf+0x329/0x6e0 devm_kvasprintf+0x54/0xb0 devm_kasprintf+0x58/0x80 hda_machine_select.cold+0x198/0x17a2 [snd_sof_intel_hda_generic] sof_probe_work+0x7f/0x600 [snd_sof] process_one_work+0x17b/0x330 worker_thread+0x2ce/0x3f0 kthread+0xcf/0x100 ret_from_fork+0x31/0x50 ret_from_fork_asm+0x1a/0x30
kfence-#99: 0x00000000198a940f-0x00000000ace47d9d, size=64, cache=kmalloc-64
allocated by task 333 on cpu 8 at 17.798069s (130.453553s ago): devm_kmalloc+0x52/0x120 devm_kvasprintf+0x66/0xb0 devm_kasprintf+0x58/0x80 hda_machine_select.cold+0x198/0x17a2 [snd_sof_intel_hda_generic] sof_probe_work+0x7f/0x600 [snd_sof] process_one_work+0x17b/0x330 worker_thread+0x2ce/0x3f0 kthread+0xcf/0x100 ret_from_fork+0x31/0x50 ret_from_fork_asm+0x1a/0x30
freed by task 1543 on cpu 4 at 141.586686s (6.665010s ago): release_nodes+0x43/0xb0 devres_release_all+0x90/0xf0 device_unbind_cleanup+0xe/0x70 device_release_driver_internal+0x1c1/0x200 driver_detach+0x48/0x90 bus_remove_driver+0x6d/0xf0 pci_unregister_driver+0x42/0xb0 __do_sys_delete_module+0x1d1/0x310 do_syscall_64+0x82/0x190 entry_SYSCALL_64_after_hwframe+0x76/0x7e
Fix it by copying the machine with devm_kmemdup() before we modify it.
Fixes: 5458411d7594 ("ASoC: SOF: Intel: hda: refactoring topology name fixup for HDA mach") Suggested-by: Péter Ujfalusi" peter.ujfalusi@linux.intel.com Signed-off-by: Tavian Barnes tavianator@tavianator.com
v3: Copy the whole machine struct instead so we can safely modify it (instead of storing the name in pdata->tplg_filename, breaking other fixups). The problem with v2 was pointed out by Bard Liao, with the alternative fix suggested by Péter Ujfalusi.
sound/soc/sof/intel/hda.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
diff --git a/sound/soc/sof/intel/hda.c b/sound/soc/sof/intel/hda.c index b34e5fdf10f1..cbf96b342005 100644 --- a/sound/soc/sof/intel/hda.c +++ b/sound/soc/sof/intel/hda.c @@ -1049,7 +1049,13 @@ static void hda_generic_machine_select(struct snd_sof_dev *sdev, if (!*mach && codec_num <= 2) { bool tplg_fixup = false;
hda_mach = snd_soc_acpi_intel_hda_machines;
/* make a copy so we can modify it below */hda_mach = devm_kmemdup(sdev->dev,snd_soc_acpi_intel_hda_machines,sizeof(*hda_mach),GFP_KERNEL);
We need to copy 2x the size as the snd_soc_acpi_intel_hda_machines[] has two entries, the second is the sentinel (all 0).
if (!hda_mach)return; dev_info(bus->dev, "using HDA machine driver %s now\n", hda_mach->drv_name);