Re: [PATCH bpf-next 1/2] bpf: introduce helper bpf_find_vma

On Nov 3, 2021, at 4:54 PM, Song Liu [off-list ref] wrote:

quoted

On Nov 1, 2021, at 3:23 PM, Daniel Borkmann [off-list ref] wrote:

On 10/28/21 12:00 AM, Song Liu wrote:
[...]

quoted

/* integer value in 'imm' field of BPF_CALL instruction selects which helper

diff --git a/kernel/bpf/task_iter.c b/kernel/bpf/task_iter.c
index b48750bfba5aa..ad30f2e885356 100644
--- a/kernel/bpf/task_iter.c
+++ b/kernel/bpf/task_iter.c

@@ -8,6 +8,7 @@

#include <linux/fdtable.h>
#include <linux/filter.h>
#include <linux/btf_ids.h>
+#include <linux/irq_work.h>
  struct bpf_iter_seq_task_common {
	struct pid_namespace *ns;

@@ -21,6 +22,25 @@ struct bpf_iter_seq_task_info {

	u32 tid;
};
+/* irq_work to run mmap_read_unlock() */
+struct task_iter_irq_work {
+	struct irq_work irq_work;
+	struct mm_struct *mm;
+};
+
+static DEFINE_PER_CPU(struct task_iter_irq_work, mmap_unlock_work);
+
+static void do_mmap_read_unlock(struct irq_work *entry)
+{
+	struct task_iter_irq_work *work;
+
+	if (WARN_ON_ONCE(IS_ENABLED(CONFIG_PREEMPT_RT)))
+		return;
+
+	work = container_of(entry, struct task_iter_irq_work, irq_work);
+	mmap_read_unlock_non_owner(work->mm);
+}
+
static struct task_struct *task_seq_get_next(struct pid_namespace *ns,
					     u32 *tid,
					     bool skip_if_dup_files)

@@ -586,9 +606,89 @@ static struct bpf_iter_reg task_vma_reg_info = {

	.seq_info		= &task_vma_seq_info,
};
+BPF_CALL_5(bpf_find_vma, struct task_struct *, task, u64, start,
+	   bpf_callback_t, callback_fn, void *, callback_ctx, u64, flags)
+{
+	struct task_iter_irq_work *work = NULL;
+	struct mm_struct *mm = task->mm;

Won't this NULL deref if called with task argument as NULL?

Will fix. 

quoted

quoted

+	struct vm_area_struct *vma;
+	bool irq_work_busy = false;
+	int ret = -ENOENT;
+
+	if (flags)
+		return -EINVAL;
+
+	if (!mm)
+		return -ENOENT;
+
+	/*
+	 * Similar to stackmap with build_id support, we cannot simply do
+	 * mmap_read_unlock when the irq is disabled. Instead, we need do
+	 * the unlock in the irq_work.
+	 */
+	if (irqs_disabled()) {
+		if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
+			work = this_cpu_ptr(&mmap_unlock_work);
+			if (irq_work_is_busy(&work->irq_work)) {
+				/* cannot queue more mmap_unlock, abort. */
+				irq_work_busy = true;
+			}
+		} else {
+			/*
+			 * PREEMPT_RT does not allow to trylock mmap sem in
+			 * interrupt disabled context, abort.
+			 */
+			irq_work_busy = true;
+		}
+	}
+
+	if (irq_work_busy || !mmap_read_trylock(mm))
+		return -EBUSY;
+
+	vma = find_vma(mm, start);
+
+	if (vma && vma->vm_start <= start && vma->vm_end > start) {
+		callback_fn((u64)(long)task, (u64)(long)vma,
+			    (u64)(long)callback_ctx, 0, 0);
+		ret = 0;
+	}
+	if (!work) {
+		mmap_read_unlock(current->mm);
+	} else {
+		work->mm = current->mm;
+
+		/* The lock will be released once we're out of interrupt
+		 * context. Tell lockdep that we've released it now so
+		 * it doesn't complain that we forgot to release it.
+		 */
+		rwsem_release(&current->mm->mmap_lock.dep_map, _RET_IP_);
+		irq_work_queue(&work->irq_work);
+	}

Given this is pretty much the same logic around the vma retrieval, could this be
refactored/consolidated with stack map build id retrieval into a common function?

I thought about sharing the irq_work code amount the two. The problem was we need
to include irq_work.h in bpf.h. But on a second thought, maybe we should just 
move bpf_find_vma to stackmap.c? This will avoid including irq_work.h. I guess it 
is not too weird to have bpf_find_vma in stackmap.c.  

Actually, we can just add a local header for it in kernel/bpf. Adding bpf_find_vma
to stackmap.c means bpf_find_vma requires CONFIG_PERF_EVENTS. It is not a real 
issue in most systems, but may break some build tests. 

Song