2021-04-07 06:54:16

by Chuck Lever III

[permalink] [raw]
Subject: Re: [PATCH -next] NFSD: Use DEFINE_SPINLOCK() for spinlock



> On Apr 6, 2021, at 8:08 AM, Huang Guobin <[email protected]> wrote:
>
> From: Guobin Huang <[email protected]>
>
> spinlock can be initialized automatically with DEFINE_SPINLOCK()
> rather than explicitly calling spin_lock_init().
>
> Reported-by: Hulk Robot <[email protected]>
> Signed-off-by: Guobin Huang <[email protected]>

This same patch was sent last July, without the Reported-by:

https://lore.kernel.org/linux-nfs/[email protected]/T/#u

If I'm reading the code correctly, it appears that set_max_drc()
can be called more than once by user space API functions via
nfsd_create_serv().

It seems to me we want to guarantee that nfsd_drc_lock is
initialized exactly once, and using DEFINE_SPINLOCK() would
do just that.

Likewise, re-initializing nfsd_drc_mem_used is probably not good
to do either, but clobbering a spinlock like that might lead to
a crash.

Bruce, any further thoughts?


> ---
> fs/nfsd/nfssvc.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> index b2eef4112bc2..82ba034fa579 100644
> --- a/fs/nfsd/nfssvc.c
> +++ b/fs/nfsd/nfssvc.c
> @@ -84,7 +84,7 @@ DEFINE_MUTEX(nfsd_mutex);
> * version 4.1 DRC caches.
> * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
> */
> -spinlock_t nfsd_drc_lock;
> +DEFINE_SPINLOCK(nfsd_drc_lock);
> unsigned long nfsd_drc_max_mem;
> unsigned long nfsd_drc_mem_used;
>
> @@ -563,7 +563,6 @@ static void set_max_drc(void)
> nfsd_drc_max_mem = (nr_free_buffer_pages()
> >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
> nfsd_drc_mem_used = 0;
> - spin_lock_init(&nfsd_drc_lock);
> dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
> }
>
>

--
Chuck Lever




2021-04-07 06:59:09

by [email protected]

[permalink] [raw]
Subject: Re: [PATCH -next] NFSD: Use DEFINE_SPINLOCK() for spinlock

On Tue, Apr 06, 2021 at 03:46:34PM +0000, Chuck Lever III wrote:
>
>
> > On Apr 6, 2021, at 8:08 AM, Huang Guobin <[email protected]> wrote:
> >
> > From: Guobin Huang <[email protected]>
> >
> > spinlock can be initialized automatically with DEFINE_SPINLOCK()
> > rather than explicitly calling spin_lock_init().
> >
> > Reported-by: Hulk Robot <[email protected]>
> > Signed-off-by: Guobin Huang <[email protected]>
>
> This same patch was sent last July, without the Reported-by:
>
> https://lore.kernel.org/linux-nfs/[email protected]/T/#u
>
> If I'm reading the code correctly, it appears that set_max_drc()
> can be called more than once by user space API functions via
> nfsd_create_serv().
>
> It seems to me we want to guarantee that nfsd_drc_lock is
> initialized exactly once, and using DEFINE_SPINLOCK() would
> do just that.
>
> Likewise, re-initializing nfsd_drc_mem_used is probably not good
> to do either, but clobbering a spinlock like that might lead to
> a crash.
>
> Bruce, any further thoughts?

The other nfsd_drc_lock users run in nfsd threads, which shouldn't be
running yet at the point we call set_max_drc.

Reinitializing the lock each time is pretty weird, though, ACK to the
patch even if it's just cleanup.

--b.

>
>
> > ---
> > fs/nfsd/nfssvc.c | 3 +--
> > 1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> > index b2eef4112bc2..82ba034fa579 100644
> > --- a/fs/nfsd/nfssvc.c
> > +++ b/fs/nfsd/nfssvc.c
> > @@ -84,7 +84,7 @@ DEFINE_MUTEX(nfsd_mutex);
> > * version 4.1 DRC caches.
> > * nfsd_drc_pages_used tracks the current version 4.1 DRC memory usage.
> > */
> > -spinlock_t nfsd_drc_lock;
> > +DEFINE_SPINLOCK(nfsd_drc_lock);
> > unsigned long nfsd_drc_max_mem;
> > unsigned long nfsd_drc_mem_used;
> >
> > @@ -563,7 +563,6 @@ static void set_max_drc(void)
> > nfsd_drc_max_mem = (nr_free_buffer_pages()
> > >> NFSD_DRC_SIZE_SHIFT) * PAGE_SIZE;
> > nfsd_drc_mem_used = 0;
> > - spin_lock_init(&nfsd_drc_lock);
> > dprintk("%s nfsd_drc_max_mem %lu \n", __func__, nfsd_drc_max_mem);
> > }
> >
> >
>
> --
> Chuck Lever
>
>