Re: bug report: using snd_BUG_ON() instead of WARN_ON()

[Clemens Ladisch <clemens@ladisch.de>]

Dan Carpenter wrote:
> Smatch found a couple places try use the return value for snd_BUG_ON() 
> without realizing it's always zero

snd_BUG_ON() returns the return value of WARN() which is the value of the
condition.

> or that it can be defined away entirely under certain configs.

This is the point of this debugging macro.

> sound/drivers/opl3/opl3_midi.c +652 snd_opl3_kill_voice(34) warn: buffer overflow 'opl3->voices' 18 <= 20
> 
> 	opl3_midi.c checks the range with snd_BUG_ON() and then adds 3 so 
> 	it possibly goes out of bounds.  I'm not sure the situation there.

A four-operator sound needs two voices with that offset.
opl3_get_voice() takes care of allocating appropriate voices for that,
but this case is not checked with snd_BUG_ON().  It would be possible to
add snd_BUG_ON(voice+3) into the if().

> sound/core/seq/seq_midi.c +403 snd_seq_midisynth_register_port(126) error: buffer overflow 'client->ports_per_device' 8 <= 8
> sound/core/seq/seq_midi.c +404 snd_seq_midisynth_register_port(127) error: buffer overflow 'client->ports' 8 <= 8

There is a snd_BUG_ON(device>=8) in line 291, so device can be at most 7.

> sound/core/info_oss.c +52 snd_oss_info_register(10) error: buffer overflow 'snd_sndstat_strings[num]' 6 <= 6
> sound/core/info_oss.c +52 snd_oss_info_register(10) error: buffer overflow 'snd_sndstat_strings' 32 <= 32
> sound/core/info_oss.c +63 snd_oss_info_register(21) error: buffer overflow 'snd_sndstat_strings[num]' 6 <= 6
> sound/core/info_oss.c +63 snd_oss_info_register(21) error: buffer overflow 'snd_sndstat_strings' 32 <= 32
> sound/pci/cs46xx/dsp_spos_scb_lib.c +1497 cs46xx_dsp_destroy_pcm_channel(28) error: buffer overflow 'ins->src_scb_slots' 14 <= 14

Same type of false positive; it looks as if >= in snd_BUG_ON() is not
handled correctly.


Regards,
Clemens