* Peter Zijlstra <peterz@xxxxxxxxxxxxx> [2011-09-27 13:37:15]:
> On Mon, 2011-09-26 at 21:14 +0530, Srikar Dronamraju wrote:
> >
> > > > +
> > > > +/*
> > > > + * Called from mmap_region.
> > > > + * called with mm->mmap_sem acquired.
> > > > + *
> > > > + * Return -ve no if we fail to insert probes and we cannot
> > > > + * bail-out.
> > > > + * Return 0 otherwise. i.e :
> > > > + * - successful insertion of probes
> > > > + * - (or) no possible probes to be inserted.
> > > > + * - (or) insertion of probes failed but we can bail-out.
> > > > + */
> > > > +int mmap_uprobe(struct vm_area_struct *vma)
> > > > +{
> > > > + struct list_head tmp_list;
> > > > + struct uprobe *uprobe, *u;
> > > > + struct inode *inode;
> > > > + int ret = 0;
> > > > +
> > > > + if (!valid_vma(vma))
> > > > + return ret; /* Bail-out */
> > > > +
> > > > + inode = igrab(vma->vm_file->f_mapping->host);
> > > > + if (!inode)
> > > > + return ret;
> > > > +
> > > > + INIT_LIST_HEAD(&tmp_list);
> > > > + mutex_lock(&uprobes_mmap_mutex);
> > > > + build_probe_list(inode, &tmp_list);
> > > > + list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
> > > > + loff_t vaddr;
> > > > +
> > > > + list_del(&uprobe->pending_list);
> > > > + if (!ret && uprobe->consumers) {
> > > > + vaddr = vma->vm_start + uprobe->offset;
> > > > + vaddr -= vma->vm_pgoff << PAGE_SHIFT;
> > > > + if (vaddr < vma->vm_start || vaddr >= vma->vm_end)
> > > > + continue;
> > > > + ret = install_breakpoint(vma->vm_mm, uprobe);
> > > > +
> > > > + if (ret && (ret == -ESRCH || ret == -EEXIST))
> > > > + ret = 0;
> > > > + }
> > > > + put_uprobe(uprobe);
> > > > + }
> > > > +
> > > > + mutex_unlock(&uprobes_mmap_mutex);
> > > > + iput(inode);
> > > > + return ret;
> > > > +}
> > > > +
> > > > +static void dec_mm_uprobes_count(struct vm_area_struct *vma,
> > > > + struct inode *inode)
> > > > +{
> > > > + struct uprobe *uprobe;
> > > > + struct rb_node *n;
> > > > + unsigned long flags;
> > > > +
> > > > + n = uprobes_tree.rb_node;
> > > > + spin_lock_irqsave(&uprobes_treelock, flags);
> > > > + uprobe = __find_uprobe(inode, 0, &n);
> > > > +
> > > > + /*
> > > > + * If indeed there is a probe for the inode and with offset zero,
> > > > + * then lets release its reference. (ref got thro __find_uprobe)
> > > > + */
> > > > + if (uprobe)
> > > > + put_uprobe(uprobe);
> > > > + for (; n; n = rb_next(n)) {
> > > > + loff_t vaddr;
> > > > +
> > > > + uprobe = rb_entry(n, struct uprobe, rb_node);
> > > > + if (uprobe->inode != inode)
> > > > + break;
> > > > + vaddr = vma->vm_start + uprobe->offset;
> > > > + vaddr -= vma->vm_pgoff << PAGE_SHIFT;
> > > > + if (vaddr < vma->vm_start || vaddr >= vma->vm_end)
> > > > + continue;
> > > > + atomic_dec(&vma->vm_mm->mm_uprobes_count);
> > > > + }
> > > > + spin_unlock_irqrestore(&uprobes_treelock, flags);
> > > > +}
> > > > +
> > > > +/*
> > > > + * Called in context of a munmap of a vma.
> > > > + */
> > > > +void munmap_uprobe(struct vm_area_struct *vma)
> > > > +{
> > > > + struct inode *inode;
> > > > +
> > > > + if (!valid_vma(vma))
> > > > + return; /* Bail-out */
> > > > +
> > > > + if (!atomic_read(&vma->vm_mm->mm_uprobes_count))
> > > > + return;
> > > > +
> > > > + inode = igrab(vma->vm_file->f_mapping->host);
> > > > + if (!inode)
> > > > + return;
> > > > +
> > > > + dec_mm_uprobes_count(vma, inode);
> > > > + iput(inode);
> > > > + return;
> > > > +}
> > >
> > > One has to wonder why mmap_uprobe() can be one function but
> > > munmap_uprobe() cannot.
> > >
> >
> > I didnt understand this comment, Can you please elaborate?
> > mmap_uprobe uses build_probe_list and munmap_uprobe uses
> > dec_mm_uprobes_count.
>
> Ah, I missed build_probe_list(), but I didn't see a reason for the
> existence of dec_mm_uprobe_count(), the name doesn't make sense and the
> content is 'small' enough to just put in munmap_uprobe.
>
> To me it looks similar to the list iteration you have in mmap_uprobe(),
> you didn't split that out into another function either.
Hmm, For me whats done in dec_mm_uprobe_count() is similar to whats
done in build_probe_list(), just that build_probe_list does a atomic_inc
+ list_add while dec_mm_uprobe_count() does a atomic_dec.
When I kept dec_mm_uprobe_count() inside munmap_uprobe(), I found most
of the code nested too deep. Hence carved it out as a separate function.
I open to suggestions for dec_mm_uprobe_count()
--
Thanks and Regards
Srikar
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>