mbox series

[v5,-next,00/16] sysctl: move sysctls from vm_table into its own files

Message ID 20250111070751.2588654-1-yukaixiong@huawei.com (mailing list archive)
Headers show
Series sysctl: move sysctls from vm_table into its own files | expand

Message

yukaixiong Jan. 11, 2025, 7:07 a.m. UTC 
This patch series moves sysctls of vm_table in kernel/sysctl.c to
places where they actually belong, and do some related code clean-ups.
After this patch series, all sysctls in vm_table have been moved into its
own files, meanwhile, delete vm_table.

All the modifications of this patch series base on
linux-next(tags/next-20250110). To test this patch series, the code was
compiled with both the CONFIG_SYSCTL enabled and disabled on arm64 and
x86_64 architectures. After this patch series is applied, all files
under /proc/sys/vm can be read or written normally.

my test steps as below listed:

Step 1: Set CONFIG_SYSCTL to 'n' and compile the Linux kernel on the
arm64 architecture. The kernel compiles successfully without any errors
or warnings.

Step 2: Set CONFIG_SYSCTL to 'y' and compile the Linux kernel on the
arm64 architecture. The kernel compiles successfully without any errors
or warnings.

Step 3: Use QEMU to boot the kernel compiled in Step 2, then use the
Linux command |ls -ll /proc/sys/vm/| to check the permissions of all
sysctls under the |/proc/sys/vm| directory.
When compared with the kernel version that did not apply this series of
patches, there is no difference in the sysctl permissions under the
|/proc/sys/vm| directory between the two kernel versions.

Step 4: Repeat steps 1 to 3 on the x86_64 architecture.

Changes in v5:
 - add reviewed-by from Lorenzo Stoakes in patch6,8
 - Accept Brian Gerst's suggestion to reduce the preprocessor
   condition "#elif (defined(CONFIG_X86_32) && !defined(CONFIG_UML))"
   in patch13
 - fix the error discovered by Geert Uytterhoeven in patch14 in V4.
   Move the call to register_sysctl_init() into its own fs_initcall()
   as Geert Uytterhoeven's patch does.
 - take the advice of Joel Granados, separating patch14 in V4 into
   patch14 and patch15 in V5. patch14 in V5 just moves the vdso_enabled
   table. patch15 in V5 removes the vm_table.
 - modify the change log of patch14
 - clarify test steps.

Changes in v4:
 - due to my mistake, the previous version sent 15 patches twice.
   Please ignore that, as this version is the correct one.
 - change all "static struct ctl_table" type into
   "static const struct ctl_table" type in patch1~10,12,13,14
 - simplify result of rpcauth_cache_shrink_count() in patch11

Changes in v3:
 - change patch1~10, patch14 title suggested by Joel Granados
 - change sysctl_stat_interval to static type in patch1
 - add acked-by from Paul Moore in patch7
 - change dirtytime_expire_interval to static type in patch9
 - add acked-by from Anna Schumaker in patch11

Changes in v2:
 - fix sysctl_max_map_count undeclared issue in mm/nommu.c for patch6
 - update changelog for patch7/12, suggested by Kees/Paul
 - fix patch8, sorry for wrong changes and forget to built with NOMMU
 - add reviewed-by from Kees except patch8 since patch8 is wrong in v1
 - add reviewed-by from Jan Kara, Christian Brauner in patch12

Kaixiong Yu (16):
  mm: vmstat: move sysctls to mm/vmstat.c
  mm: filemap: move sysctl to mm/filemap.c
  mm: swap: move sysctl to mm/swap.c
  mm: vmscan: move vmscan sysctls to mm/vmscan.c
  mm: util: move sysctls to mm/util.c
  mm: mmap: move sysctl to mm/mmap.c
  security: min_addr: move sysctl to security/min_addr.c
  mm: nommu: move sysctl to mm/nommu.c
  fs: fs-writeback: move sysctl to fs/fs-writeback.c
  fs: drop_caches: move sysctl to fs/drop_caches.c
  sunrpc: simplify rpcauth_cache_shrink_count()
  fs: dcache: move the sysctl to fs/dcache.c
  x86: vdso: move the sysctl to arch/x86/entry/vdso/vdso32-setup.c
  sh: vdso: move the sysctl to arch/sh/kernel/vsyscall/vsyscall.c
  sysctl: remove the vm_table
  sysctl: remove unneeded include

 arch/sh/kernel/vsyscall/vsyscall.c |  20 +++
 arch/x86/entry/vdso/vdso32-setup.c |  16 ++-
 fs/dcache.c                        |  21 ++-
 fs/drop_caches.c                   |  23 ++-
 fs/fs-writeback.c                  |  30 ++--
 include/linux/dcache.h             |   7 +-
 include/linux/mm.h                 |  23 ---
 include/linux/mman.h               |   2 -
 include/linux/swap.h               |   9 --
 include/linux/vmstat.h             |  11 --
 include/linux/writeback.h          |   4 -
 kernel/sysctl.c                    | 221 -----------------------------
 mm/filemap.c                       |  18 ++-
 mm/internal.h                      |  10 ++
 mm/mmap.c                          |  54 +++++++
 mm/nommu.c                         |  15 +-
 mm/swap.c                          |  16 ++-
 mm/swap.h                          |   1 +
 mm/util.c                          |  67 +++++++--
 mm/vmscan.c                        |  23 +++
 mm/vmstat.c                        |  44 +++++-
 net/sunrpc/auth.c                  |   2 +-
 security/min_addr.c                |  11 ++
 23 files changed, 336 insertions(+), 312 deletions(-)

Comments

Joel Granados Jan. 14, 2025, 1:50 p.m. UTC | #1 
On Sat, Jan 11, 2025 at 03:07:35PM +0800, Kaixiong Yu wrote:
> This patch series moves sysctls of vm_table in kernel/sysctl.c to
> places where they actually belong, and do some related code clean-ups.
> After this patch series, all sysctls in vm_table have been moved into its
> own files, meanwhile, delete vm_table.
> 
> All the modifications of this patch series base on
> linux-next(tags/next-20250110). To test this patch series, the code was
> compiled with both the CONFIG_SYSCTL enabled and disabled on arm64 and
> x86_64 architectures. After this patch series is applied, all files
> under /proc/sys/vm can be read or written normally.

It is looking good! Here is how I think we should move it upstream:

1. These should queued in for 6.15 instead of the next merge window.
   It is too late in the current cycle and if we put it in now, it will
   not properly tested in linux-next.

2. I am putting this in sysctl-testing with the expectation of pushing this
   up for the 6.15 merge window. Please tell me if you want this to go
   through some other tree.

Thx for the contribution

Best
> 
> my test steps as below listed:
> 
> Step 1: Set CONFIG_SYSCTL to 'n' and compile the Linux kernel on the
> arm64 architecture. The kernel compiles successfully without any errors
> or warnings.
> 
...
>  mm/swap.c                          |  16 ++-
>  mm/swap.h                          |   1 +
>  mm/util.c                          |  67 +++++++--
>  mm/vmscan.c                        |  23 +++
>  mm/vmstat.c                        |  44 +++++-
>  net/sunrpc/auth.c                  |   2 +-
>  security/min_addr.c                |  11 ++
>  23 files changed, 336 insertions(+), 312 deletions(-)
> 
> -- 
> 2.34.1
>
yukaixiong Jan. 15, 2025, 1:53 a.m. UTC | #2 
On 2025/1/14 21:50, Joel Granados wrote:
> On Sat, Jan 11, 2025 at 03:07:35PM +0800, Kaixiong Yu wrote:
>> This patch series moves sysctls of vm_table in kernel/sysctl.c to
>> places where they actually belong, and do some related code clean-ups.
>> After this patch series, all sysctls in vm_table have been moved into its
>> own files, meanwhile, delete vm_table.
>>
>> All the modifications of this patch series base on
>> linux-next(tags/next-20250110). To test this patch series, the code was
>> compiled with both the CONFIG_SYSCTL enabled and disabled on arm64 and
>> x86_64 architectures. After this patch series is applied, all files
>> under /proc/sys/vm can be read or written normally.
> It is looking good! Here is how I think we should move it upstream:
>
> 1. These should queued in for 6.15 instead of the next merge window.
>     It is too late in the current cycle and if we put it in now, it will
>     not properly tested in linux-next.
>
> 2. I am putting this in sysctl-testing with the expectation of pushing this
>     up for the 6.15 merge window. Please tell me if you want this to go
>     through some other tree.
>
> Thx for the contribution
>
> Best

Thank you! I don't want this to go through some other tree.

Best ...
>> my test steps as below listed:
>>
>> Step 1: Set CONFIG_SYSCTL to 'n' and compile the Linux kernel on the
>> arm64 architecture. The kernel compiles successfully without any errors
>> or warnings.
>>
> ...
>>   mm/swap.c                          |  16 ++-
>>   mm/swap.h                          |   1 +
>>   mm/util.c                          |  67 +++++++--
>>   mm/vmscan.c                        |  23 +++
>>   mm/vmstat.c                        |  44 +++++-
>>   net/sunrpc/auth.c                  |   2 +-
>>   security/min_addr.c                |  11 ++
>>   23 files changed, 336 insertions(+), 312 deletions(-)
>>
>> -- 
>> 2.34.1
>>