lkml.org 
[lkml]   [2011]   [Oct]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v5 3.1.0-rc4-tip 18/26] uprobes: slot allocation.
* Oleg Nesterov <oleg@redhat.com> [2011-10-07 20:37:40]:

> On 09/20, Srikar Dronamraju wrote:
> >
> > - * valid_vma: Verify if the specified vma is an executable vma
> > + * valid_vma: Verify if the specified vma is an executable vma,
> > + * but not an XOL vma.
> > * - Return 1 if the specified virtual address is in an
> > - * executable vma.
> > + * executable vma, but not in an XOL vma.
> > */
> > static bool valid_vma(struct vm_area_struct *vma)
> > {
> > + struct uprobes_xol_area *area = vma->vm_mm->uprobes_xol_area;
> > +
> > if (!vma->vm_file)
> > return false;
> >
> > + if (area && (area->vaddr == vma->vm_start))
> > + return false;
>
> Could you explain why do we need this "but not an XOL vma" check?
> xol_vma->vm_file is always NULL, no?
>

Yes, xol_vma->vm_file is always NULL.
previously we used shmem_file_setup before we map the XOL area.
However we now use init_creds instead, so this should also change
accordingly. Will correct this.

> > +static struct uprobes_xol_area *xol_alloc_area(void)
> > +{
> > + struct uprobes_xol_area *area = NULL;
> > +
> > + area = kzalloc(sizeof(*area), GFP_KERNEL);
> > + if (unlikely(!area))
> > + return NULL;
> > +
> > + area->bitmap = kzalloc(BITS_TO_LONGS(UINSNS_PER_PAGE) * sizeof(long),
> > + GFP_KERNEL);
> > +
> > + if (!area->bitmap)
> > + goto fail;
> > +
> > + init_waitqueue_head(&area->wq);
> > + spin_lock_init(&area->slot_lock);
> > + if (!xol_add_vma(area) && !current->mm->uprobes_xol_area) {
> > + task_lock(current);
> > + if (!current->mm->uprobes_xol_area) {
> > + current->mm->uprobes_xol_area = area;
> > + task_unlock(current);
> > + return area;
> > + }
> > + task_unlock(current);
>
> But you can't rely on task_lock(), you can race with another thread
> with the same ->mm. I guess you need mmap_sem or xchg().

Agree,
I think its better to use cmpxchg instead of xchg(). Otherwise,
(using xchg), I would set area to new value, but the old area might be in
use already. So I cant unmap the old area.

If I use cmpxchg, I can free up the new area if previous area is non
NULL.

However setting uprobes_xol_area in xol_add_vma() where we already take
mmap_sem for write while maping the xol_area is the best option.

>
> > static int pre_ssout(struct uprobe *uprobe, struct pt_regs *regs,
> > unsigned long vaddr)
> > {
> > - /* TODO: Yet to be implemented */
> > + if (xol_get_insn_slot(uprobe, vaddr) && !pre_xol(uprobe, regs)) {
> > + set_instruction_pointer(regs, current->utask->xol_vaddr);
>
> set_instruction_pointer() looks unneded, pre_xol() has already changed
> regs->ip.
>

Agree.

--
Thanks and Regards
Srikar


\
 
 \ /
  Last update: 2011-10-09 14:09    [W:0.152 / U:6.152 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site