Re: [PATCH v2 bpf-next 05/21] bpf: Introduce bpf_sysctl_{get,set}_new_value helpers
From: Andrey Ignatov
Date: Thu Apr 04 2019 - 20:21:15 EST
Daniel Borkmann <daniel@xxxxxxxxxxxxx> [Thu, 2019-04-04 07:38 -0700]:
> On 03/26/2019 01:43 AM, Andrey Ignatov wrote:
> > Add helpers to work with new value being written to sysctl by user
> > space.
> >
> > bpf_sysctl_get_new_value() copies value being written to sysctl into
> > provided buffer.
> >
> > bpf_sysctl_set_new_value() overrides new value being written by user
> > space with a one from provided buffer. Buffer should contain string
> > representation of the value, similar to what can be seen in /proc/sys/.
> >
> > Both helpers can be used only on sysctl write.
> >
> > File position matters and can be managed by an interface that will be
> > introduced separately. E.g. if user space calls sys_write to a file in
> > /proc/sys/ at file position = X, where X > 0, then the value set by
> > bpf_sysctl_set_new_value() will be written starting from X. If program
> > wants to override whole value with specified buffer, file position has
> > to be set to zero.
> >
> > Documentation for the new helpers is provided in bpf.h UAPI.
> >
> > Signed-off-by: Andrey Ignatov <rdna@xxxxxx>
> > ---
> > fs/proc/proc_sysctl.c | 22 ++++++++---
> > include/linux/bpf-cgroup.h | 8 ++--
> > include/linux/filter.h | 3 ++
> > include/uapi/linux/bpf.h | 38 +++++++++++++++++-
> > kernel/bpf/cgroup.c | 81 +++++++++++++++++++++++++++++++++++++-
> > 5 files changed, 142 insertions(+), 10 deletions(-)
> >
> > diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
> > index 72f4a096c146..4d1ab22774f7 100644
> > --- a/fs/proc/proc_sysctl.c
> > +++ b/fs/proc/proc_sysctl.c
> > @@ -570,8 +570,8 @@ static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
> > struct inode *inode = file_inode(filp);
> > struct ctl_table_header *head = grab_header(inode);
> > struct ctl_table *table = PROC_I(inode)->sysctl_entry;
> > + void *new_buf = NULL;
> > ssize_t error;
> > - size_t res;
> >
> > if (IS_ERR(head))
> > return PTR_ERR(head);
> > @@ -589,15 +589,27 @@ static ssize_t proc_sys_call_handler(struct file *filp, void __user *buf,
> > if (!table->proc_handler)
> > goto out;
> >
> > - error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write);
> > + error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, &count,
> > + &new_buf);
> > if (error)
> > goto out;
> >
> > /* careful: calling conventions are nasty here */
> > - res = count;
> > - error = table->proc_handler(table, write, buf, &res, ppos);
> > + if (new_buf) {
> > + mm_segment_t old_fs;
> > +
> > + old_fs = get_fs();
> > + set_fs(KERNEL_DS);
> > + error = table->proc_handler(table, write, (void __user *)new_buf,
> > + &count, ppos);
> > + set_fs(old_fs);
>
> From quick glance on the set, the above stood out. Afaik, there is an ongoing
> effort by Al and other fs/core folks (as visible in the git log) to get rid of
> set_fs() calls in the tree with the goal of eliminating this interface /entirely/
> (more context on 'why' here: https://urldefense.proofpoint.com/v2/url?u=https-3A__lwn.net_Articles_722267_&d=DwICaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=3jAokpHyGuCuJ834j-tttQ&m=fmn6jd1czDvp5a6GeSw0zLMxU3VRcgm1ohqwAPOKf38&s=AfbvJ91arUzm328cKzcHXeeb104boAx8NJjsoIU6Lbk&e=). Is there a better
> way to achieve the above w/o needing it?
That's a good question. I've spent quite a lot of time looking for a
better way and the only one I'm aware of so far is to change
proc_handler signature, so that it accepts kernel 'buffer', and copying
between user and kernel happens outside of proc_handler.
But it would require changing all proc_handler implementations as well
so that they accept kernel 'buffer' and don't copy data from/to user by
themselves and there are just too many sysctl proc_handler
implementations:
% git grep -E '\.proc_handler\s+=\s+' | \
sed -Ee 's/^.*\.proc_handler\s+=\s+//' | sort -u | wc -l
179
% git grep -lE '\.proc_handler\s+=\s+' | wc -l
103
, i.e. it's huge refactoring that can be really hard to upstream.
Also I looked at the LWN article you mentioned and found this branch
that cleans up set_fs use cases:
http://git.infradead.org/users/hch/vfs.git/shortlog/refs/heads/setfs-elimination
I see it uses either similar approach, i.e. introduce separate (or use
available) function that accepts kernel buffer, then copy data from user
and pass it to such a function. Another approach, I see in the branch,
is to use iovec iterators (that's mentioned in the LWN article as well),
but that again would require changing all proc_handler implementations.
That's being said I don't know a better way to do this w/o huge
refactoring.
> > + kfree(new_buf);
> > + } else {
> > + error = table->proc_handler(table, write, buf, &count, ppos);
> > + }
> > +
> > if (!error)
> > - error = res;
> > + error = count;
> > out:
> > sysctl_head_finish(head);
> >
> > diff --git a/include/linux/bpf-cgroup.h b/include/linux/bpf-cgroup.h
> > index b1c45da20a26..1e97271f9a10 100644
> > --- a/include/linux/bpf-cgroup.h
> > +++ b/include/linux/bpf-cgroup.h
> > @@ -113,7 +113,8 @@ int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
> >
> > int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
> > struct ctl_table *table, int write,
> > - enum bpf_attach_type type);
> > + void __user *buf, size_t *pcount,
> > + void **new_buf, enum bpf_attach_type type);
> >
> > static inline enum bpf_cgroup_storage_type cgroup_storage_type(
> > struct bpf_map *map)
> > @@ -261,11 +262,12 @@ int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
> > })
> >
> >
> > -#define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write) \
> > +#define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, count, nbuf) \
> > ({ \
> > int __ret = 0; \
> > if (cgroup_bpf_enabled) \
> > __ret = __cgroup_bpf_run_filter_sysctl(head, table, write, \
> > + buf, count, nbuf, \
> > BPF_CGROUP_SYSCTL); \
> > __ret; \
> > })
> > @@ -338,7 +340,7 @@ static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
> > #define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
> > #define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
> > #define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type,major,minor,access) ({ 0; })
> > -#define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write) ({ 0; })
> > +#define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,nbuf) ({ 0; })
> >
> > #define for_each_cgroup_storage_type(stype) for (; false; )
> >
--
Andrey Ignatov