Re: git: 7672cbef2c1e - main - pipes: reserve configured percentage of buffers zone to superuser

From: <tuexen_at_freebsd.org>
Date: Fri, 20 Sep 2024 13:04:33 UTC
After building a kernel including these changes, I see a lot of
negative pipecnt for uid = 0
Installing world also fails due to pipe problems.
Is this related to one of these changes? Any configuration change needed?

Best regards
Michael

> On 20. Sep 2024, at 08:46, Konstantin Belousov <kib@FreeBSD.org> wrote:
> 
> The branch main has been updated by kib:
> 
> URL: https://cgit.FreeBSD.org/src/commit/?id=7672cbef2c1e1267e42bb3aad6a6da9380f4347f
> 
> commit 7672cbef2c1e1267e42bb3aad6a6da9380f4347f
> Author:     Konstantin Belousov <kib@FreeBSD.org>
> AuthorDate: 2024-09-15 06:57:34 +0000
> Commit:     Konstantin Belousov <kib@FreeBSD.org>
> CommitDate: 2024-09-20 06:46:07 +0000
> 
>    pipes: reserve configured percentage of buffers zone to superuser
> 
>    Sponsored by:   The FreeBSD Foundation
>    MFC after:      1 week
>    Differential revision:  https://reviews.freebsd.org/D46619
> ---
> sys/kern/sys_pipe.c | 23 +++++++++++++++++++++--
> 1 file changed, 21 insertions(+), 2 deletions(-)
> 
> diff --git a/sys/kern/sys_pipe.c b/sys/kern/sys_pipe.c
> index 7ee2b5c76da3..68b57708d653 100644
> --- a/sys/kern/sys_pipe.c
> +++ b/sys/kern/sys_pipe.c
> @@ -103,6 +103,7 @@
> #include <sys/stat.h>
> #include <sys/malloc.h>
> #include <sys/poll.h>
> +#include <sys/priv.h>
> #include <sys/selinfo.h>
> #include <sys/signalvar.h>
> #include <sys/syscallsubr.h>
> @@ -206,6 +207,7 @@ static int pipeallocfail;
> static int piperesizefail;
> static int piperesizeallowed = 1;
> static long pipe_mindirect = PIPE_MINDIRECT;
> +static int pipebuf_reserv = 2;
> 
> SYSCTL_LONG(_kern_ipc, OID_AUTO, maxpipekva, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
>   &maxpipekva, 0, "Pipe KVA limit");
> @@ -219,6 +221,9 @@ SYSCTL_INT(_kern_ipc, OID_AUTO, piperesizefail, CTLFLAG_RD,
>  &piperesizefail, 0, "Pipe resize failures");
> SYSCTL_INT(_kern_ipc, OID_AUTO, piperesizeallowed, CTLFLAG_RW,
>  &piperesizeallowed, 0, "Pipe resizing allowed");
> +SYSCTL_INT(_kern_ipc, OID_AUTO, pipebuf_reserv, CTLFLAG_RW,
> +    &pipebuf_reserv, 0,
> +    "Superuser-reserved percentage of the pipe buffers space");
> 
> static void pipeinit(void *dummy __unused);
> static void pipeclose(struct pipe *cpipe);
> @@ -586,8 +591,22 @@ retry:
> return (ENOMEM);
> }
> 
> - error = vm_map_find(pipe_map, NULL, 0, (vm_offset_t *)&buffer, size, 0,
> -    VMFS_ANY_SPACE, VM_PROT_RW, VM_PROT_RW, 0);
> + vm_map_lock(pipe_map);
> + if (priv_check(curthread, PRIV_PIPEBUF) != 0 &&
> +    (vm_map_max(pipe_map) - vm_map_min(pipe_map)) *
> +    (100 - pipebuf_reserv) / 100 < pipe_map->size + size) {
> + vm_map_unlock(pipe_map);
> + if (cpipe->pipe_buffer.buffer == NULL &&
> +    size > SMALL_PIPE_SIZE) {
> + size = SMALL_PIPE_SIZE;
> + pipefragretry++;
> + goto retry;
> + }
> + return (ENOMEM);
> + }
> + error = vm_map_find_locked(pipe_map, NULL, 0, (vm_offset_t *)&buffer,
> +    size, 0, VMFS_ANY_SPACE, VM_PROT_RW, VM_PROT_RW, 0);
> + vm_map_unlock(pipe_map);
> if (error != KERN_SUCCESS) {
> chgpipecnt(cpipe->pipe_pair->pp_owner->cr_ruidinfo, -size, 0);
> if (cpipe->pipe_buffer.buffer == NULL &&