linux-kernel - Re: [PATCH 1/3] mm/vmalloc: fix size check for remap_vmalloc_range

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: <dba7cb2c2882e034c8c99b09a432313a@suse.de>
Date:   Thu, 03 Jan 2019 20:27:26 +0100
From:   Roman Penyaev <rpenyaev@...e.de>
To:     Michal Hocko <mhocko@...nel.org>
Cc:     Andrew Morton <akpm@...ux-foundation.org>,
        Andrey Ryabinin <aryabinin@...tuozzo.com>,
        Joe Perches <joe@...ches.com>,
        "Luis R. Rodriguez" <mcgrof@...nel.org>, linux-mm@...ck.org,
        linux-kernel@...r.kernel.org, stable@...r.kernel.org
Subject: Re: [PATCH 1/3] mm/vmalloc: fix size check for
 remap_vmalloc_range_partial()

On 2019-01-03 16:13, Michal Hocko wrote:
> On Thu 03-01-19 15:59:52, Roman Penyaev wrote:
>> area->size can include adjacent guard page but get_vm_area_size()
>> returns actual size of the area.
>> 
>> This fixes possible kernel crash when userspace tries to map area
>> on 1 page bigger: size check passes but the following 
>> vmalloc_to_page()
>> returns NULL on last guard (non-existing) page.
> 
> Can this actually happen? I am not really familiar with all the callers
> of this API but VM_NO_GUARD is not really used wildly in the kernel.

Exactly, by default (VM_NO_GUARD is not set) each area has guard page,
thus the area->size will be bigger.  The bug is not reproduced if
VM_NO_GUARD is set.

> All I can see is kasan na arm64 which doesn't really seem to use it
> for vmalloc.
> 
> So is the problem real or this is a mere cleanup?

This is the real problem, try this hunk for any file descriptor which 
provides
mapping, or say modify epoll as example:

--------------------------------
diff --git a/fs/eventpoll.c b/fs/eventpoll.c

+static int ep_mmap(struct file *file, struct vm_area_struct *vma)
+{
+       void *mem;
+
+       mem = vmalloc_user(4096);
+       BUG_ON(!mem);
+       /* Do not care about mem leak */
+
+       return remap_vmalloc_range(vma, mem, 0);
+}
+
  /* File callbacks that implement the eventpoll file behaviour */
  static const struct file_operations eventpoll_fops = {
  #ifdef CONFIG_PROC_FS
         .show_fdinfo    = ep_show_fdinfo,
  #endif
+       .mmap           = ep_mmap,
         .release        = ep_eventpoll_release,
--------------------------------

and the following code from userspace, which maps 2 pages,
instead of 1:

--------------------------------
epfd = epoll_create1(0);
assert(epfd >= 0);

p = mmap(NULL, 2<<12, PROT_WRITE|PROT_READ, MAP_PRIVATE, epfd, 0);
assert(p != MAP_FAILED);
--------------------------------

You immediately get the following oops:

[   38.894571] BUG: unable to handle kernel NULL pointer dereference at 
0000000000000008
[   38.899048] #PF error: [normal kernel read fault]
[   38.901487] PGD 0 P4D 0
[   38.902801] Oops: 0000 [#1] PREEMPT SMP PTI
[   38.904984] CPU: 2 PID: 399 Comm: mmap-epoll Not tainted 4.20.0-1 
#238
[   38.914064] RIP: 0010:vm_insert_page+0x3b/0x1d0
[   38.941181] Call Trace:
[   38.941656]  remap_vmalloc_range_partial+0x8d/0xd0
[   38.942417]  mmap_region+0x3c7/0x630
[   38.942982]  do_mmap+0x38d/0x560
[   38.943479]  vm_mmap_pgoff+0x9a/0xf0
[   38.944028]  ksys_mmap_pgoff+0x18e/0x220
[   38.944554]  do_syscall_64+0x48/0xf0
[   38.945076]  entry_SYSCALL_64_after_hwframe+0x44/0xa9

--
Roman