linux-kernel - Re: [PATCH v4 2/3] mm: introduce MAP_VALIDATE a mechanism for adding new mmap flags

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAPcyv4hJ3VaCzE0tOtcSJPfMPDimH-_oeoUAha8MVJ6ZOQU8fw@mail.gmail.com>
Date:   Tue, 15 Aug 2017 09:24:24 -0700
From:   Dan Williams <dan.j.williams@...el.com>
To:     Jan Kara <jack@...e.cz>
Cc:     "Darrick J. Wong" <darrick.wong@...cle.com>,
        Arnd Bergmann <arnd@...db.de>,
        "linux-nvdimm@...ts.01.org" <linux-nvdimm@...ts.01.org>,
        Linux API <linux-api@...r.kernel.org>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        linux-xfs@...r.kernel.org, Linux MM <linux-mm@...ck.org>,
        Andy Lutomirski <luto@...nel.org>,
        linux-fsdevel <linux-fsdevel@...r.kernel.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Christoph Hellwig <hch@....de>
Subject: Re: [PATCH v4 2/3] mm: introduce MAP_VALIDATE a mechanism for adding
 new mmap flags

On Tue, Aug 15, 2017 at 5:27 AM, Jan Kara <jack@...e.cz> wrote:
> On Mon 14-08-17 23:12:16, Dan Williams wrote:
>> The mmap syscall suffers from the ABI anti-pattern of not validating
>> unknown flags. However, proposals like MAP_SYNC and MAP_DIRECT need a
>> mechanism to define new behavior that is known to fail on older kernels
>> without the feature. Use the fact that specifying MAP_SHARED and
>> MAP_PRIVATE at the same time is invalid as a cute hack to allow a new
>> set of validated flags to be introduced.
>>
>> This also introduces the ->fmmap() file operation that is ->mmap() plus
>> flags. Each ->fmmap() implementation must fail requests when a locally
>> unsupported flag is specified.
> ...
>> diff --git a/include/linux/fs.h b/include/linux/fs.h
>> index 1104e5df39ef..bbe755d0caee 100644
>> --- a/include/linux/fs.h
>> +++ b/include/linux/fs.h
>> @@ -1674,6 +1674,7 @@ struct file_operations {
>>       long (*unlocked_ioctl) (struct file *, unsigned int, unsigned long);
>>       long (*compat_ioctl) (struct file *, unsigned int, unsigned long);
>>       int (*mmap) (struct file *, struct vm_area_struct *);
>> +     int (*fmmap) (struct file *, struct vm_area_struct *, unsigned long);
>>       int (*open) (struct inode *, struct file *);
>>       int (*flush) (struct file *, fl_owner_t id);
>>       int (*release) (struct inode *, struct file *);
>> @@ -1748,6 +1749,12 @@ static inline int call_mmap(struct file *file, struct vm_area_struct *vma)
>>       return file->f_op->mmap(file, vma);
>>  }
>>
>> +static inline int call_fmmap(struct file *file, struct vm_area_struct *vma,
>> +             unsigned long flags)
>> +{
>> +     return file->f_op->fmmap(file, vma, flags);
>> +}
>> +
>
> Hum, I dislike a new file op for this when the only problem with ->mmap is
> that it misses 'flags' argument. I understand there are lots of ->mmap
> implementations out there and modifying prototype of them all is painful
> but is it so bad? Coccinelle patch for this should be rather easy...

Changing the prototype is relatively easy with Coccinelle, but we
still need the code in each ->mmap() implementation to validate a
local list of supported flags. How about adding a 'supported mmap
flags' field to 'struct file_operations' so that the validation code
can be made generic? I'll go with that since it's a bit less
surprising than a new operation type, and not as messy as teaching
every mmap implementation in the kernel to validate flags that they
will likely never care about.

> Also for MAP_SYNC I want the flag to be copied in VMA anyway so for that I
> don't need additional flags argument anyway. And I wonder how you want to
> make things work without VMA flag in case of MAP_DIRECT as well - VMAs can
> be split, partially unmapped etc. and so without VMA flag you are going to
> have hard time to detect whether there's any mapping left which blocks
> block mapping changes.

Outside of requiring a 64-bit arch, we're out of vm_flags. Also, the
core mm does not really care about MAP_DIRECT or MAP_SYNC so that's
why I added a new ->fs_flags field since these are more filesystem
properties than core mm.

The problem of tracking MAP_DIRECT over vma splits appears to already
be handled. __split_vma does:

        /* most fields are the same, copy all, and then fixup */
        *new = *vma;
...

        if (new->vm_ops && new->vm_ops->open)
                new->vm_ops->open(new);

In ->open() I'm checking if 'new' has MAP_DIRECT in ->fs_flags and
taking a reference against the S_IOMAP_SEALED flag.