[v5,0/5] sysctl: encode the min/max values directly in the table entry

Message

Wen Yang Feb. 9, 2025, 3:58 p.m. UTC

Many modules use these additional static/global variables (such as
two_five_five, n_65535, ue_int_max, etc.) in the boundary checking of
sysctl, and they are read-only and never changed.

Eric points out: 

- One of the things that happens and that is worth acknowledging is there
- is code that wraps proc_doulongvec_minmax and proc_dointvec_minmax.
- Having the minmax information separate from the data pointer makes that
- wrapping easier.

- Further the min/max information is typically separate from other kinds
- of data.  So even when not wrapped it is nice just to take a quick
- glance and see what the minimums and maximums are.

- My original suggest was that we change struct ctl_table from:

- > /* A sysctl table is an array of struct ctl_table: */
- > struct ctl_table {
- > 	const char *procname;		/* Text ID for /proc/sys */
- > 	void *data;
- > 	int maxlen;
- > 	umode_t mode;
- > 	proc_handler *proc_handler;	/* Callback for text formatting */
- > 	struct ctl_table_poll *poll;
- > 	void *extra1;
- > 	void *extra2;
- > } __randomize_layout;

- to:

- > /* A sysctl table is an array of struct ctl_table: */
- > struct ctl_table {
- > 	const char *procname;		/* Text ID for /proc/sys */
- > 	void *data;
- > 	int maxlen;
- > 	umode_t mode;
- > 	proc_handler *proc_handler;	/* Callback for text formatting */
- > 	struct ctl_table_poll *poll;
- >       unsigned long min;
- >       unsigned long max;
- > } __randomize_layout;

- That is just replace extra1 and extra2 with min and max members.  The
- members don't have any reason to be pointers.  Without being pointers
- the min/max functions can just use long values to cap either ints or
- longs, and there is no room for error.  The integer promotion rules
- will ensure that even negative values can be stored in unsigned long
- min and max values successfully.  Plus it is all bog standard C
- so there is nothing special to learn.

- There are a bunch of fiddly little details to transition from where we
- are today.  The most straightforward way I can see of making the
- transition is to add the min and max members.  Come up with replacements
- for proc_doulongvec_minmax and proc_dointvec_minmax that read the new
- min and max members.  Update all of the users.  Update the few users
- that use extra1 or extra2 for something besides min and max.  Then
- remove extra1 and extra2.  At the end it is simpler and requires the
- same or a little less space.

This patch series achieves direct encoding min/max values in table entries
and still maintains compatibility with existing extra1/extra2 pointers.
Afterwards, we can remove these unnecessary static variables progressively and
also finally kill the shared const array.

v4: https://lore.kernel.org/all/[e-mail]/
v3: https://lore.kernel.org/all/[e-mail]/
v2: https://lore.kernel.org/all/[e-mail]/
v1: https://lore.kernel.org/all/[e-mail]/

and
https://lore.kernel.org/all/[e-mail]/

Wen Yang (5):
  sysctl: simplify the min/max boundary check
  sysctl: add helper functions to extract table->extra1/extra2
  sysctl: support encoding values directly in the table entry
  sysctl: add kunit test code to check the min/max encoding of sysctl
    table entries
  sysctl: delete six_hundred_forty_kb to save 4 bytes

 fs/proc/proc_sysctl.c  |  35 ++-
 include/linux/sysctl.h |  64 ++++-
 kernel/sysctl-test.c   | 581 +++++++++++++++++++++++++++++++++++++++++
 kernel/sysctl.c        | 211 ++++++---------
 4 files changed, 739 insertions(+), 152 deletions(-)