Re: [PATCH 4/4] ALSA: emu10k1: Cocci spatch "alloc_cast"
On Thu, 21 Sep 2017 19:39:15 +0200, Al Viro wrote: > > On Thu, Sep 21, 2017 at 08:33:46AM +0200, Thomas Meyer wrote: > > Remove casting the values returned by memory allocation functions like > > kmalloc, kzalloc, kmem_cache_alloc, kmem_cache_zalloc etc." > > Found by coccinelle spatch "api/alloc/alloc_cast.cocci" > > > > Signed-off-by: Thomas Meyer > > --- > > > > diff -u -p a/sound/pci/emu10k1/emufx.c b/sound/pci/emu10k1/emufx.c > > --- a/sound/pci/emu10k1/emufx.c > > +++ b/sound/pci/emu10k1/emufx.c > > @@ -1218,7 +1218,7 @@ static int _snd_emu10k1_audigy_init_efx( > > if (!icode) > > return err; > > > > - icode->gpr_map = (u_int32_t __user *) kcalloc(512 + 256 + 256 + 2 * > > 1024, > > + icode->gpr_map = kcalloc(512 + 256 + 256 + 2 * 1024, > > sizeof(u_int32_t), > > GFP_KERNEL); > > And _this_ is a wonderful example of the reasons why that kind of > patches is bloody bad. The code you've caught is very obviously smelly - > kcalloc() does *NOT* return a userland pointer. You are whitewashing it; > a major "something weird is going on here" sign is gone (something weird > in that case is hopefully a set_fs(KERNEL_DS) somewhere nearby, and it > is asking for careful review). Right, the code touched by the patch is a really tricky one. It uses the same struct that contains __user pointer which is supposed to be filled by ioctl, but in this case, the function allocates the kernel buffer and passes it with a special flag mentioning it being a kernel buffer later. thanks, Takashi
Re: [PATCH 4/4] ALSA: emu10k1: Cocci spatch "alloc_cast"
On Thu, Sep 21, 2017 at 08:33:46AM +0200, Thomas Meyer wrote: > Remove casting the values returned by memory allocation functions like > kmalloc, kzalloc, kmem_cache_alloc, kmem_cache_zalloc etc." > Found by coccinelle spatch "api/alloc/alloc_cast.cocci" > > Signed-off-by: Thomas Meyer > --- > > diff -u -p a/sound/pci/emu10k1/emufx.c b/sound/pci/emu10k1/emufx.c > --- a/sound/pci/emu10k1/emufx.c > +++ b/sound/pci/emu10k1/emufx.c > @@ -1218,7 +1218,7 @@ static int _snd_emu10k1_audigy_init_efx( > if (!icode) > return err; > > - icode->gpr_map = (u_int32_t __user *) kcalloc(512 + 256 + 256 + 2 * > 1024, > + icode->gpr_map = kcalloc(512 + 256 + 256 + 2 * 1024, > sizeof(u_int32_t), > GFP_KERNEL); And _this_ is a wonderful example of the reasons why that kind of patches is bloody bad. The code you've caught is very obviously smelly - kcalloc() does *NOT* return a userland pointer. You are whitewashing it; a major "something weird is going on here" sign is gone (something weird in that case is hopefully a set_fs(KERNEL_DS) somewhere nearby, and it is asking for careful review). FWIW, any patches of that sort anywhere near VFS are very much not welcome.
Re: [PATCH 4/4] ALSA: emu10k1: Cocci spatch "alloc_cast"
On Thu, 21 Sep 2017 08:33:46 +0200, Thomas Meyer wrote: > > Remove casting the values returned by memory allocation functions like > kmalloc, kzalloc, kmem_cache_alloc, kmem_cache_zalloc etc." > Found by coccinelle spatch "api/alloc/alloc_cast.cocci" > > Signed-off-by: Thomas Meyer No, this cast is still required because it has __user annotation. It's not about compiler warning but about sparse. thanks, Takashi
