On Wed, Jul 10, 2024 at 11:16:31AM +0200, Peter Zijlstra wrote:
> If it were an actual sequence count, I could make it work, but sadly,
> not. Also, vma_end_write() seems to be missing :-( If anything it could
> be used to lockdep annotate the thing.
>
> Mooo.. I need to stare more at this to see if perhaps it can be made to
> work, but so far, no joy :/
See, this is what I want, except I can't close the race against VMA
modification because of that crazy locking scheme :/
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -2146,11 +2146,58 @@ static int is_trap_at_addr(struct mm_str
return is_trap_insn(&opcode);
}
-static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
+#ifndef CONFIG_PER_VMA_LOCK
+static struct uprobe *__find_active_uprobe(unsigned long bp_vaddr)
+{
+ return NULL;
+}
+#else
+static struct uprobe *__find_active_uprobe(unsigned long bp_vaddr)
{
struct mm_struct *mm = current->mm;
struct uprobe *uprobe = NULL;
struct vm_area_struct *vma;
+ MA_STATE(mas, &mm->mm_mt, bp_vaddr, bp_vaddr);
+
+ guard(rcu)();
+
+again:
+ vma = mas_walk(&mas);
+ if (!vma)
+ return NULL;
+
+ /* vma_write_start() -- in progress */
+ if (READ_ONCE(vma->vm_lock_seq) == READ_ONCE(vma->vm_mm->mm_lock_seq))
+ return NULL;
+
+ /*
+ * Completely broken, because of the crazy vma locking scheme you
+ * cannot avoid the per-vma rwlock and doing so means you're racy
+ * against modifications.
+ *
+ * A simple actual seqcount would'be been cheaper and more usefull.
+ */
+
+ if (!valid_vma(vma, false))
+ return NULL;
+
+ struct inode = file_inode(vma->vm_file);
+ loff_t offset = vaddr_to_offset(vma, bp_vaddr);
+
+ // XXX: if (vma_seq_retry(...)) goto again;
+
+ return find_uprobe(inode, offset);
+}
+#endif
+
+static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
+{
+ struct uprobe *uprobe = __find_active_uprobe(bp_vaddr)
+ struct mm_struct *mm = current->mm;
+ struct vm_area_struct *vma;
+
+ if (uprobe)
+ return uprobe;
mmap_read_lock(mm);
vma = vma_lookup(mm, bp_vaddr);
