Re: [PATCH] drivers/base/node.c: fix userspace break from using bin_attributes for cpumap and cpulist

[Phil Auld <pauld@redhat.com>]

On Wed, Jul 13, 2022 at 11:18:59AM +1200 Barry Song wrote:
> On Wed, Jul 13, 2022 at 9:58 AM Phil Auld <pauld@redhat.com> wrote:
> >
> > Using bin_attributes with a 0 size causes fstat and friends to return that 0 size.
> > This breaks userspace code that retrieves the size before reading the file. Rather
> > than reverting 75bd50fa841 ("drivers/base/node.c: use bin_attribute to break the size
> > limitation of cpumap ABI") let's put in a size value at compile time. Use direct
> > comparison and a worst-case maximum to ensure compile time constants. For cpulist the
> > max is on the order of NR_CPUS * (ceil(log10(NR_CPUS)) + 1) which for 8192 is 40960.
> > In order to get near that you'd need a system with every other CPU on one node or
> > something similar. e.g. (0,2,4,... 1024,1026...). We set it to a min of PAGE_SIZE
> > to retain the older behavior. For cpumap, PAGE_SIZE is plenty big.
> >
> > On an 80 cpu 4-node system (NR_CPUS == 8192)
> >
> > before:
> >
> > -r--r--r--. 1 root root 0 Jul 12 14:08 /sys/devices/system/node/node0/cpulist
> > -r--r--r--. 1 root root 0 Jul 11 17:25 /sys/devices/system/node/node0/cpumap
> 
> it is a fundamental problem of bin_attr, isn't it? when we don't know the
> exact size of an attribute, and this size might become more than one
> PAGE_SIZE, we use bin_attr to break the limitation. but the fact is that
> we really don't know or it is really hard to know the actual size of the
> attribute.
>

But it broke userspace applications. I figured rather than revert it maybe
we can find a max size to put in there and make it continue to work.

> >
> > after:
> >
> > -r--r--r--. 1 root root 40960 Jul 12 16:48 /sys/devices/system/node/node0/cpulist
> > -r--r--r--. 1 root root  4096 Jul 12 15:50 /sys/devices/system/node/node0/cpumap
> 
> if we finally set a size which might be improper, it seems we defeat the
> purpose we start to move to bin_attr?
> 
> >
> > Fixes: 75bd50fa841 ("drivers/base/node.c: use bin_attribute to break the size limitation of cpumap ABI")
> > Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > Cc: "Rafael J. Wysocki" <rafael@kernel.org>
> > Signed-off-by: Phil Auld <pauld@redhat.com>
> > ---
> >  drivers/base/node.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/base/node.c b/drivers/base/node.c
> > index 0ac6376ef7a1..291c69671f23 100644
> > --- a/drivers/base/node.c
> > +++ b/drivers/base/node.c
> > @@ -45,7 +45,7 @@ static inline ssize_t cpumap_read(struct file *file, struct kobject *kobj,
> >         return n;
> >  }
> >
> > -static BIN_ATTR_RO(cpumap, 0);
> > +static BIN_ATTR_RO(cpumap, PAGE_SIZE);
> 
> PAGE_SIZE is probably big enough, will we still calculate to get it rather than
> hard coding?

This one is actually wrong. I did not realize how big a NR_CPUS people were actually using.
It should be something like (NR_CPUS/4 + NR_CPUS/32). 

> 
> >
> >  static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
> >                                    struct bin_attribute *attr, char *buf,
> > @@ -66,7 +66,7 @@ static inline ssize_t cpulist_read(struct file *file, struct kobject *kobj,
> >         return n;
> >  }
> >
> > -static BIN_ATTR_RO(cpulist, 0);
> > +static BIN_ATTR_RO(cpulist, (((NR_CPUS * 5) > PAGE_SIZE) ? NR_CPUS *5 : PAGE_SIZE));
> 
> I am still not sure why it is NR_CPUS * 5. Is 5 bytes big enough to
> describe the number
> of cpu id? technically it seems not,  for example,  for cpuid=100000,
> we need at least 6
> bytes.

Sure. As I said in the comment I wanted to do NR_CPUS * (ceil(log10(NR_CPUS)) + 1) but doing
that math in the kernel was messy. So I used 5. Even that is probably way bigger than needed.
Are there really 100000 cpus on one node with discontiguous cpuids? "0-99999" is only, what,
9 characters?

We can put whatever number you want that is >= the size the read will return.

Thanks,
Phil

> 
> BTW, my silly question is that what if we set the size to MAXIMUM int?
> Will it fix
> the userspace fsstat?
> 
> >
> >  /**
> >   * struct node_access_nodes - Access class device to hold user visible
> > --
> > 2.31.1
> >
> 
> Thanks
> Barry
> 

--