linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: xiujianfeng <xiujianfeng@huawei.com>
To: "Mickaël Salaün" <mic@digikod.net>,
	"Günther Noack" <gnoack3000@gmail.com>
Cc: <paul@paul-moore.com>, <jmorris@namei.org>, <serge@hallyn.com>,
	<shuah@kernel.org>, <corbet@lwn.net>,
	<linux-security-module@vger.kernel.org>,
	<linux-kernel@vger.kernel.org>, <linux-kselftest@vger.kernel.org>,
	<linux-doc@vger.kernel.org>
Subject: Re: [PATCH -next v2 3/6] landlock: add chmod and chown support
Date: Thu, 1 Sep 2022 21:06:02 +0800	[thread overview]
Message-ID: <2f286496-f4f8-76f7-2fb6-cc3dd5ffdeaa@huawei.com> (raw)
In-Reply-To: <de4620d2-3268-b3cc-71dd-acbbd204435e@digikod.net>

Hi,

在 2022/8/30 0:01, Mickaël Salaün 写道:
> 
> On 29/08/2022 03:17, xiujianfeng wrote:
>>
>> Hi,
>>
>> 在 2022/8/28 3:30, Günther Noack 写道:
>>> Hello!
>>>
>>> the mapping between Landlock rights to LSM hooks is now as follows in
>>> your patch set:
>>>
>>> * LANDLOCK_ACCESS_FS_CHMOD controls hook_path_chmod
>>> * LANDLOCK_ACCESS_FS_CHGRP controls hook_path_chown
>>>     (this hook can restrict both the chown(2) and chgrp(2) syscalls)
>>>
>>> Is this the desired mapping?
>>>
>>> The previous discussion I found on the topic was in
>>>
>>> [1] 
>>> https://lore.kernel.org/all/5873455f-fff9-618c-25b1-8b6a4ec94368@digikod.net/ 
>>>
>>> [2] 
>>> https://lore.kernel.org/all/b1d69dfa-6d93-2034-7854-e2bc4017d20e@schaufler-ca.com/ 
>>>
>>> [3] 
>>> https://lore.kernel.org/all/c369c45d-5aa8-3e39-c7d6-b08b165495fd@digikod.net/ 
>>>
>>>
>>> In my understanding the main arguments were the ones in [2] and [3].
>>>
>>> There were no further responses to [3], so I was under the impression
>>> that we were gravitating towards an approach where the
>>> file-metadata-modification operations were grouped more coarsely?
>>>
>>> For example with the approach suggested in [3], which would be to
>>> group the operations coarsely into (a) one Landlock right for
>>> modifying file metadata that is used in security contexts, and (b) one
>>> Landlock right for modifying metadata that was used in non-security
>>> contexts. That would mean that there would be:
>>>
>>> (a) LANDLOCK_ACCESS_FS_MODIFY_SECURITY_ATTRIBUTES to control the
>>> following operations:
>>>     * chmod(2)-variants through hook_path_chmod,
>>>     * chown(2)-variants and chgrp(2)-variants through hook_path_chown,
>>>     * setxattr(2)-variants and removexattr(2)-variants for extended
>>>       attributes that are not "user extended attributes" as described in
>>>       xattr(7) through hook_inode_setxattr and hook_inode_removexattr
>>>
>>> (b) LANDLOCK_ACCESS_FS_MODIFY_NON_SECURITY_ATTRIBUTES to control the
>>> following operations:
>>>     * utimes(2) and other operations for setting other non-security
>>>       sensitive attributes, probably through hook_inode_setattr(?)
>>>     * xattr modifications like above, but for the "user extended
>>>       attributes", though hook_inode_setxattr and hook_inode_removexattr
>>>
>>> In my mind, this would be a sensible grouping, and it would also help
>>> to decouple the userspace-exposed API from the underlying
>>> implementation, as Casey suggested to do in [2].
>>>
>>> Specifically for this patch set, if you want to use this grouping, you
>>> would only need to add one new Landlock right
>>> (LANDLOCK_ACCESS_FS_MODIFY_SECURITY_ATTRIBUTES) as described above
>>> under (a) (and maybe we can find a shorter name for it... :))?
>>>
>>> Did I miss any operations here that would be necessary to restrict?
>>>
>>> Would that make sense to you? Xiu, what is your opinion on how this
>>> should be grouped? Do you have use cases in mind where a more
>>> fine-grained grouping would be required?
>>
>> I apologize I may missed that discussion when I prepared v2:(
>>
>> Yes, agreed, this grouping is more sensible and resonnable. so in this
>> patchset only one right will be added, and I suppose the first commit
>> which expand access_mask_t to u32 can be droped.
>>
>>>
>>> —Günther
>>>
>>> P.S.: Regarding utimes: The hook_inode_setattr hook *also* gets called
>>> on a variety on attribute changes including file ownership, file size
>>> and file mode, so it might potentially interact with a bunch of other
>>> existing Landlock rights. Maybe that is not the right approach. In any
>>> case, it seems like it might require more thinking and it might be
>>> sensible to do that in a separate patch set IMHO.
>>
>> Thanks for you reminder, that seems it's more complicated to support
>> utimes, so I think we'd better not support it in this patchset.
> 
> The issue with this approach is that it makes it impossible to properly 
> group such access rights. Indeed, to avoid inconsistencies and much more 
> complexity, we cannot extend a Landlock access right once it is defined.
> 
> We also need to consider that file ownership and permissions have a 
> default (e.g. umask), which is also a way to set them. How to 
> consistently manage that? What if the application wants to protect its 
> files with chmod 0400?

what do you mean by this? do you mean that we should have a set of 
default permissions for files created by applications within the 
sandbox, so that it can update metadata of its own file.

> 
> About the naming, I think we can start with:
> - LANDLOCK_ACCESS_FS_READ_METADATA (read any file/dir metadata);
> - LANDLOCK_ACCESS_FS_WRITE_SAFE_METADATA: change file times, user xattr;

do you mean we should have permission controls on metadata level or 
operation level? e.g. should we allow update on user xattr but deny 
update on security xattr? or should we disallow update on any xattr?

> - LANDLOCK_ACCESS_FS_WRITE_UNSAFE_METADATA: interpreted by the kernel 
> (could change non-Landlock DAC or MAC, which could be considered as a 
> policy bypass; or other various xattr that might be interpreted by 
> filesystems), this should be denied most of the time.

do you mean FS_WRITE_UNSAFE_METADATA is security-related? and 
FS_WRITE_SAFE_METADATA is non-security-related?

> .

  reply	other threads:[~2022-09-01 13:11 UTC|newest]

Thread overview: 31+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-08-27 11:12 [PATCH -next v2 0/6] landlock: add chmod and chown support Xiu Jianfeng
2022-08-27 11:12 ` [PATCH -next v2 1/6] landlock: expand access_mask_t to u32 type Xiu Jianfeng
2022-08-27 11:12 ` [PATCH -next v2 2/6] landlock: abstract walk_to_visible_parent() helper Xiu Jianfeng
2022-08-30 11:22   ` Mickaël Salaün
2022-08-31 11:56     ` xiujianfeng
2022-08-27 11:12 ` [PATCH -next v2 3/6] landlock: add chmod and chown support Xiu Jianfeng
2022-08-27 19:30   ` Günther Noack
2022-08-29  1:17     ` xiujianfeng
2022-08-29 16:01       ` Mickaël Salaün
2022-09-01 13:06         ` xiujianfeng [this message]
2022-09-01 17:34           ` Mickaël Salaün
2022-10-29  8:33             ` xiujianfeng
2022-11-14 14:12               ` Mickaël Salaün
2022-11-18  9:03                 ` xiujianfeng
2022-11-18 12:32                   ` Mickaël Salaün
2022-11-21 13:48                     ` xiujianfeng
2022-08-29  6:30     ` xiujianfeng
2022-08-29  6:35   ` xiujianfeng
2022-08-27 11:12 ` [PATCH -next v2 4/6] landlock/selftests: add selftests for chmod and chown Xiu Jianfeng
2022-08-27 17:48   ` Günther Noack
2022-08-29  1:49     ` xiujianfeng
2022-08-27 11:12 ` [PATCH -next v2 5/6] landlock/samples: add chmod and chown support Xiu Jianfeng
2022-08-27 11:12 ` [PATCH -next v2 6/6] landlock: update chmod and chown support in document Xiu Jianfeng
2022-08-27 17:28   ` Günther Noack
2022-08-29  1:52     ` xiujianfeng
2022-08-30 11:22 ` [PATCH -next v2 0/6] landlock: add chmod and chown support Mickaël Salaün
2023-04-18 10:53 ` xiujianfeng
2023-04-20 17:40   ` Mickaël Salaün
2023-04-24  8:52     ` xiujianfeng
2023-04-26 13:58       ` Mickaël Salaün
2023-05-05  3:50         ` xiujianfeng

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=2f286496-f4f8-76f7-2fb6-cc3dd5ffdeaa@huawei.com \
    --to=xiujianfeng@huawei.com \
    --cc=corbet@lwn.net \
    --cc=gnoack3000@gmail.com \
    --cc=jmorris@namei.org \
    --cc=linux-doc@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-kselftest@vger.kernel.org \
    --cc=linux-security-module@vger.kernel.org \
    --cc=mic@digikod.net \
    --cc=paul@paul-moore.com \
    --cc=serge@hallyn.com \
    --cc=shuah@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).