[firefly] [PATCH] mm: fix overflow in alloc_vmap_area

[Daniel Baluta]

On Thu, Jun 20, 2013 at 4:20 PM, Ghennadi Procopciuc <unix140 at gmail.com> wrote:
> Inserting the following kernel module:
>
> <snip>
>
> static int simple_test_init(void)
> {
>         size_t i, j;
>         void *address;
>
>         for (i = 0 * MB; i<  60 * MB; i += 1 * MB) {
>                 for (j = i; j<  i + 1 * MB; j += KB) {
>                         address = vmalloc(j);
>                         vfree(address);
>                 }
>         }
>
>         return 0;
> }
>
> </snip>
>
> triggers BUG at mm/vmalloc.c:310 on a x86 machine:
>
> [   95.218283] Kernel BUG at c1126cdb [verbose debug info unavailable]
> [   95.218306] invalid opcode: 0000 [#1] SMP
> [   95.218324] Modules linked in: lkma_test(OF+)<snip lots of not tainted modules>
> [   95.218559] Pid: 2419, comm: insmod Tainted: GF          O 3.9.0+ #57 Hewlett-Packard HP Compaq 8200 Elite CMT PC/1494
> [   95.218597] EIP: 0060:[<c1126cdb>] EFLAGS: 00010207 CPU: 3
> [   95.218617] EIP is at __insert_vmap_area+0xfb/0x100
> [   95.218635] EAX: f85dc000 EBX: ef05cac0 ECX: f7be08c4 EDX: 00000007
> [   95.218657] ESI: f2ed044c EDI: c1a220b8 EBP: f027bd34 ESP: f027bd14
> [   95.218680]  DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068
> [   95.218699] CR0: 80050033 CR2: b5995118 CR3: 30364000 CR4: 000407f0
> [   95.218721] DR0: 00000000 DR1: 00000000 DR2: 00000000 DR3: 00000000
> [   95.218743] DR6: ffff0ff0 DR7: 00000400
> [   95.218758] Process insmod (pid: 2419, ti=f027a000 task=efe64010 task.ti=f027a000)
> [   95.218784] Stack:
> [   95.218792]  c177b964 fecfb000 f85e2000 00000000 f85dc000 ffbfe000 f02da1c0 0007f000
> [   95.218829]  f027bd7c c112802d c177b9b0 fecfb000 01305000 ffbfe000 00000001 ef05cac0
> [   95.218866]  00000000 00000000 f83fe000 01304fff f83fe000 ffffffff 01305000 ef05c500
> [   95.218903] Call Trace:
> [   95.218915]  [<f85e2000>] ? 0xf85e1fff
> [   95.218930]  [<f85dc000>] ? 0xf85dbfff
> [   95.218945]  [<c112802d>] alloc_vmap_area.isra.16+0x1bd/0x2f0
> [   95.218967]  [<c112867f>] __get_vm_area_node.isra.17+0x8f/0x130
> [   95.218988]  [<c1128d77>] __vmalloc_node_range+0x57/0x200
> [   95.219009]  [<f85f3075>] ? lkma_test_init+0x45/0x70 [lkma_test]
> [   95.219031]  [<c1128f82>] __vmalloc_node+0x62/0x70
> [   95.219049]  [<f85f3075>] ? lkma_test_init+0x45/0x70 [lkma_test]
> [   95.219071]  [<c1129058>] vmalloc+0x38/0x40
> [   95.219087]  [<f85f3075>] ? lkma_test_init+0x45/0x70 [lkma_test]
> [   95.219109]  [<f85f3075>] lkma_test_init+0x45/0x70 [lkma_test]
> [   95.219131]  [<f85f3030>] ? kzalloc+0x10/0x10 [lkma_test]
> [   95.219151]  [<c1001222>] do_one_initcall+0x112/0x160
> [   95.219171]  [<c15ca3cf>] ? set_section_ro_nx+0x54/0x59
> [   95.219190]  [<c1099b69>] load_module+0x1d79/0x2660
> [   95.219209]  [<c114721d>] ? create_object+0x19d/0x280
> [   95.219230]  [<c109a4c8>] sys_init_module+0x78/0xb0
> [   95.219250]  [<c15d9801>] sysenter_do_call+0x12/0x22
> [   95.219268] Code: 39 03 73 0c 8b 3f 89 f0 83 c7 08 e9 3d ff ff ff 8b 46 f4 39 43 04 76 13 8b 3f 89 f0 83 c7 04 e9 29 ff ff ff e8 fb 1b 4a 00 eb ab<0f>  0b 8d 76 00 55 89 e5 56 53 66 66 66 66 90 83 60 0c df 89 c6
> [   95.219415] EIP: [<c1126cdb>] __insert_vmap_area+0xfb/0x100 SS:ESP 0068:f027bd14
> [   95.228313] ---[ end trace e0a1efb2acb97c98 ]---
>
> A printk placed in __insert_vmap_area will show:
>
> [   95.218256] va->va_start=4275023872 tmp_va->va_end=4166918144 va->va_end=0 tmp_va->va_start=4166893568

You could also print addresses in hexa printk("va->va_start = %d [0x%x]....);

>
> and another one, before sum operation in alloc_vmap_area:
>
> [   95.218204] addr = 4275023872 size = 19943424 vend = 4290764800

Ditto.

>
> If after addition the result is smaller than one of the arguments,
> then an overflow occurred. In our case there is an obvious overflow.
>
> Don't know if this is the right solution, but the bug happens for me in
> 3.10-rc6 and 3.9.

Remove this from the commit message, add it at [XMARK]

> Signed-off-by: Ghennadi Procopciuc <unix140 at gmail.com>
> Cc: Daniel Baluta<dbaluta at ixiacom.com>
> ---

[XMARK] add comments that are not part of the commit here.

>  mm/vmalloc.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 0f751f2..6ea525c 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -421,12 +421,13 @@ nocache:
>         }
>
>  found:
> -       if (addr + size > vend)
> +       va->va_end = addr + size;
> +       if (va->va_end < addr || va->va_end < size || va->va_end > vend)
>                 goto overflow;
>
>         va->va_start = addr;
> -       va->va_end = addr + size;
>         va->flags = 0;
> +
>         __insert_vmap_area(va);
>         free_vmap_cache = &va->rb_node;
>         spin_unlock(&vmap_area_lock);
> --

Otherwise it looks good to me. Thanks Ghennadi!