On Fri, May 15, 2020 at 12:17:52AM +0800, Xiaoming Ni wrote:
> On 2020/5/14 14:05, Xiaoming Ni wrote:
> > On 2020/5/13 20:50, Luis Chamberlain wrote:
> > > On Wed, May 13, 2020 at 12:04:02PM +0800, Xiaoming Ni wrote:
> > > > On 2020/5/13 6:03, Luis Chamberlain wrote:
> > > > > On Tue, May 12, 2020 at 12:40:55PM -0500, Eric W. Biederman wrote:
> > > > > > Luis Chamberlain <mcgrof@xxxxxxxxxx> writes:
> > > > > >
> > > > > > > On Tue, May 12, 2020 at 06:52:35AM -0500, Eric W. Biederman wrote:
> > > > > > > > Luis Chamberlain <mcgrof@xxxxxxxxxx> writes:
> > > > > > > >
> > > > > > > > > +static struct ctl_table fs_base_table[] = {
> > > > > > > > > + {
> > > > > > > > > + .procname = "fs",
> > > > > > > > > + .mode = 0555,
> > > > > > > > > + .child = fs_table,
> > > > > > > > > + },
> > > > > > > > > + { }
> > > > > > > > > +};
> > > > > > > > ^^^^^^^^^^^^^^^^^^^^^^^^ You don't need this at all.
> > > > > > > > > > +static int __init fs_procsys_init(void)
> > > > > > > > > +{
> > > > > > > > > + struct ctl_table_header *hdr;
> > > > > > > > > +
> > > > > > > > > + hdr = register_sysctl_table(fs_base_table);
> > > > > > > > ^^^^^^^^^^^^^^^^^^^^^ Please use
> > > > > > > > register_sysctl instead.
> > > > > > > > AKA
> > > > > > > > hdr = register_sysctl("fs", fs_table);
> > > > > > >
> > > > > > > Ah, much cleaner thanks!
> > > > > >
> > > > > > It is my hope you we can get rid of register_sysctl_table one of these
> > > > > > days. It was the original interface but today it is just a
> > > > > > compatibility wrapper.
> > > > > >
> > > > > > I unfortunately ran out of steam last time before I
> > > > > > finished converting
> > > > > > everything over.
> > > > >
> > > > > Let's give it one more go. I'll start with the fs stuff.
> > > > >
> > > > > Luis
> > > > >
> > > > > .
> > > > >
> > > >
> > > > If we register each feature in its own feature code file using
> > > > register() to
> > > > register the sysctl interface. To avoid merge conflicts when different
> > > > features modify sysctl.c at the same time.
> > > > that is, try to Avoid mixing code with multiple features in the
> > > > same code
> > > > file.
> > > >
> > > > For example, the multiple file interfaces defined in sysctl.c by the
> > > > hung_task feature can be moved to hung_task.c.
> > > >
> > > > Perhaps later, without centralized sysctl.c ?
> > > > Is this better?
> > > >
> > > > Thanks
> > > > Xiaoming Ni
> > > >
> > > > ---
> > > > include/linux/sched/sysctl.h | 8 +----
> > > > kernel/hung_task.c | 78
> > > > +++++++++++++++++++++++++++++++++++++++++++-
> > > > kernel/sysctl.c | 50 ----------------------------
> > > > 3 files changed, 78 insertions(+), 58 deletions(-)
> > > >
> > > > diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
> > > > index d4f6215..bb4e0d3 100644
> > > > --- a/include/linux/sched/sysctl.h
> > > > +++ b/include/linux/sched/sysctl.h
> > > > @@ -7,14 +7,8 @@
> > > > struct ctl_table;
> > > >
> > > > #ifdef CONFIG_DETECT_HUNG_TASK
> > > > -extern int sysctl_hung_task_check_count;
> > > > -extern unsigned int sysctl_hung_task_panic;
> > > > +/* used for block/ */
> > > > extern unsigned long sysctl_hung_task_timeout_secs;
> > > > -extern unsigned long sysctl_hung_task_check_interval_secs;
> > > > -extern int sysctl_hung_task_warnings;
> > > > -extern int proc_dohung_task_timeout_secs(struct ctl_table *table, int
> > > > write,
> > > > - void __user *buffer,
> > > > - size_t *lenp, loff_t *ppos);
> > > > #else
> > > > /* Avoid need for ifdefs elsewhere in the code */
> > > > enum { sysctl_hung_task_timeout_secs = 0 };
> > > > diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> > > > index 14a625c..53589f2 100644
> > > > --- a/kernel/hung_task.c
> > > > +++ b/kernel/hung_task.c
> > > > @@ -20,10 +20,10 @@
> > > > #include <linux/utsname.h>
> > > > #include <linux/sched/signal.h>
> > > > #include <linux/sched/debug.h>
> > > > +#include <linux/kmemleak.h>
> > > > #include <linux/sched/sysctl.h>
> > > >
> > > > #include <trace/events/sched.h>
> > > > -
> > > > /*
> > > > * The number of tasks checked:
> > > > */
> > > > @@ -296,8 +296,84 @@ static int watchdog(void *dummy)
> > > > return 0;
> > > > }
> > > >
> > > > +/*
> > > > + * This is needed for proc_doulongvec_minmax of
> > > > sysctl_hung_task_timeout_secs
> > > > + * and hung_task_check_interval_secs
> > > > + */
> > > > +static unsigned long hung_task_timeout_max = (LONG_MAX / HZ);
> > >
> > > This is not generic so it can stay in this file.
> > >
> > > > +static int __maybe_unused neg_one = -1;
> > >
> > > This is generic so we can share it, I suggest we just rename this
> > > for now to sysctl_neg_one, export it to a symbol namespace,
> > > EXPORT_SYMBOL_NS_GPL(sysctl_neg_one, SYSCTL) and then import it with
> > > MODULE_IMPORT_NS(SYSCTL)
>
> When I made the patch, I found that only sysctl_writes_strict and
> hung_task_warnings use the neg_one variable, so is it necessary to merge and
> generate the SYSCTL_NEG_ONE variable?
Yes.
> In addition, the SYSCTL symbol namespace has not been created yet. Do I just
> need to add a new member -1 to the sysctl_vals array?
I had forgotten about our sysctl_vals, so disregard my request
to use EXPORT_SYMBOL_NS_GPL(sysctl_neg_one, SYSCTL) and using
MODULE_IMPORT_NS(SYSCTL). Since we are already using these and
have a prefix on the define we should be good.
> diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
> index b6f5d45..acae1fa 100644
> --- a/fs/proc/proc_sysctl.c
> +++ b/fs/proc/proc_sysctl.c
> @@ -23,7 +23,7 @@
> static const struct inode_operations proc_sys_dir_operations;
>
> /* shared constants to be used in various sysctls */
> -const int sysctl_vals[] = { 0, 1, INT_MAX };
> +const int sysctl_vals[] = { 0, 1, INT_MAX, -1 };
> EXPORT_SYMBOL(sysctl_vals);
>
> /* Support for permanently empty directories */
> diff --git a/include/linux/sysctl.h b/include/linux/sysctl.h
> index 02fa844..6d741d6 100644
> --- a/include/linux/sysctl.h
> +++ b/include/linux/sysctl.h
> @@ -41,6 +41,7 @@
> #define SYSCTL_ZERO ((void *)&sysctl_vals[0])
> #define SYSCTL_ONE ((void *)&sysctl_vals[1])
> #define SYSCTL_INT_MAX ((void *)&sysctl_vals[2])
> +#define SYSCTL_NEG_ONE ((void *)&sysctl_vals[3])
>
> extern const int sysctl_vals[];
This looks good.
Luis
