Re: [PATCH v2 bpf-next 1/4] bpf: introduce task_vma bpf_iter

[Yonghong Song <yhs@fb.com>]

On 12/15/20 3:36 PM, Song Liu wrote:
> Introduce task_vma bpf_iter to print memory information of a process. It
> can be used to print customized information similar to /proc/<pid>/maps.
> 
> task_vma iterator releases mmap_lock before calling the BPF program.
> Therefore, we cannot pass vm_area_struct directly to the BPF program. A
> new __vm_area_struct is introduced to keep key information of a vma. On
> each iteration, task_vma gathers information in __vm_area_struct and
> passes it to the BPF program.
> 
> If the vma maps to a file, task_vma also holds a reference to the file
> while calling the BPF program.
> 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>   include/linux/bpf.h    |   2 +-
>   kernel/bpf/task_iter.c | 205 ++++++++++++++++++++++++++++++++++++++++-
>   2 files changed, 205 insertions(+), 2 deletions(-)
> 
> diff --git a/include/linux/bpf.h b/include/linux/bpf.h
> index 07cb5d15e7439..49dd1e29c8118 100644
> --- a/include/linux/bpf.h
> +++ b/include/linux/bpf.h
> @@ -1325,7 +1325,7 @@ enum bpf_iter_feature {
>   	BPF_ITER_RESCHED	= BIT(0),
>   };
>   
> -#define BPF_ITER_CTX_ARG_MAX 2
> +#define BPF_ITER_CTX_ARG_MAX 3
>   struct bpf_iter_reg {
>   	const char *target;
>   	bpf_iter_attach_target_t attach_target;
> diff --git a/kernel/bpf/task_iter.c b/kernel/bpf/task_iter.c
> index 0458a40edf10a..15a066b442f75 100644
> --- a/kernel/bpf/task_iter.c
> +++ b/kernel/bpf/task_iter.c
> @@ -304,9 +304,183 @@ static const struct seq_operations task_file_seq_ops = {
>   	.show	= task_file_seq_show,
>   };
>   
> +/*
> + * Key information from vm_area_struct. We need this because we cannot
> + * assume the vm_area_struct is still valid after each iteration.
> + */
> +struct __vm_area_struct {
> +	__u64 start;
> +	__u64 end;
> +	__u64 flags;
> +	__u64 pgoff;
> +};
> +
> +struct bpf_iter_seq_task_vma_info {
> +	/* The first field must be struct bpf_iter_seq_task_common.
> +	 * this is assumed by {init, fini}_seq_pidns() callback functions.
> +	 */
> +	struct bpf_iter_seq_task_common common;
> +	struct task_struct *task;
> +	struct __vm_area_struct vma;
> +	struct file *file;
> +	u32 tid;
> +};
> +
> +static struct __vm_area_struct *
> +task_vma_seq_get_next(struct bpf_iter_seq_task_vma_info *info)
> +{
> +	struct pid_namespace *ns = info->common.ns;
> +	struct task_struct *curr_task;
> +	struct vm_area_struct *vma;
> +	u32 curr_tid = info->tid;
> +	bool new_task = false;
> +
> +	/* If this function returns a non-NULL __vm_area_struct, it held
> +	 * a reference to the task_struct. If info->file is non-NULL, it
> +	 * also holds a reference to the file. If this function returns
> +	 * NULL, it does not hold any reference.
> +	 */
> +again:
> +	if (info->task) {
> +		curr_task = info->task;
> +	} else {
> +		curr_task = task_seq_get_next(ns, &curr_tid, true);
> +		if (!curr_task) {
> +			info->task = NULL;
> +			info->tid++;

Here, info->tid should be info->tid = curr_tid + 1.
For exmaple, suppose initial curr_tid = info->tid = 10, and the
above task_seq_get_next(...) returns NULL with curr_tid = 100
which means tid = 100 has been visited. So we would like
to set info->tid = 101 to avoid future potential redundant work.
Returning NULL here will signal end of iteration but user
space can still call read()...

> +			return NULL;
> +		}
> +
> +		if (curr_tid != info->tid) {
> +			info->tid = curr_tid;
> +			new_task = true;
> +		}
> +
> +		if (!curr_task->mm)
> +			goto next_task;
> +		info->task = curr_task;
> +	}
> +
> +	mmap_read_lock(curr_task->mm);
> +	if (new_task) {
> +		vma = curr_task->mm->mmap;
> +	} else {
> +		/* We drop the lock between each iteration, so it is
> +		 * necessary to use find_vma() to find the next vma. This
> +		 * is similar to the mechanism in show_smaps_rollup().
> +		 */
> +		vma = find_vma(curr_task->mm, info->vma.end - 1);
> +		/* same vma as previous iteration, use vma->next */
> +		if (vma && (vma->vm_start == info->vma.start))
> +			vma = vma->vm_next;

We may have some issues here if control is returned to user space
in the middle of iterations. For example,
    - seq_ops->next() sets info->vma properly (say corresponds to vma1 
of tid1)
    - control returns to user space
    - control backs to kernel and this is not a new task since
      tid is the same
    - but we skipped this vma for show().

I think the above skipping should be guarded. If the function
is called from seq_ops->next(), yes it can be skipped.
If the function is called from seq_ops->start(), it should not
be skipped.

Could you double check such a scenario with a smaller buffer
size for read() in user space?

> +	}
> +	if (!vma) {
> +		mmap_read_unlock(curr_task->mm);
> +		goto next_task;
> +	}
> +	info->task = curr_task;
> +	info->vma.start = vma->vm_start;
> +	info->vma.end = vma->vm_end;
> +	info->vma.pgoff = vma->vm_pgoff;
> +	info->vma.flags = vma->vm_flags;
> +	if (vma->vm_file)
> +		info->file = get_file(vma->vm_file);
> +	mmap_read_unlock(curr_task->mm);
> +	return &info->vma;
> +
> +next_task:
> +	put_task_struct(curr_task);
> +	info->task = NULL;
> +	curr_tid = ++(info->tid);
> +	goto again;
> +}
> +
> +static void *task_vma_seq_start(struct seq_file *seq, loff_t *pos)
> +{
> +	struct bpf_iter_seq_task_vma_info *info = seq->private;
> +	struct __vm_area_struct *vma;
> +
> +	vma = task_vma_seq_get_next(info);
> +	if (vma && *pos == 0)
> +		++*pos;
> +
> +	return vma;
> +}
> +
> +static void *task_vma_seq_next(struct seq_file *seq, void *v, loff_t *pos)
> +{
> +	struct bpf_iter_seq_task_vma_info *info = seq->private;
> +
> +	++*pos;
> +	if (info->file) {
> +		fput(info->file);
> +		info->file = NULL;
> +	}
> +	return task_vma_seq_get_next(info);
> +}
> +
> +struct bpf_iter__task_vma {
> +	__bpf_md_ptr(struct bpf_iter_meta *, meta);
> +	__bpf_md_ptr(struct task_struct *, task);
> +	__bpf_md_ptr(struct __vm_area_struct *, vma);
> +	__bpf_md_ptr(struct file *, file);
> +};
> +
> +DEFINE_BPF_ITER_FUNC(task_vma, struct bpf_iter_meta *meta,
> +		     struct task_struct *task, struct __vm_area_struct *vma,
> +		     struct file *file)
> +
> +static int __task_vma_seq_show(struct seq_file *seq, bool in_stop)
> +{
> +	struct bpf_iter_seq_task_vma_info *info = seq->private;
> +	struct bpf_iter__task_vma ctx;
> +	struct bpf_iter_meta meta;
> +	struct bpf_prog *prog;
> +
> +	meta.seq = seq;
> +	prog = bpf_iter_get_info(&meta, in_stop);
> +	if (!prog)
> +		return 0;
> +
> +	ctx.meta = &meta;
> +	ctx.task = info->task;
> +	ctx.vma = &info->vma;
> +	ctx.file = info->file;
> +	return bpf_iter_run_prog(prog, &ctx);
> +}
> +
> +static int task_vma_seq_show(struct seq_file *seq, void *v)
> +{
> +	return __task_vma_seq_show(seq, false);
> +}
> +
> +static void task_vma_seq_stop(struct seq_file *seq, void *v)
> +{
> +	struct bpf_iter_seq_task_vma_info *info = seq->private;
> +
> +	if (!v) {
> +		(void)__task_vma_seq_show(seq, true);
> +	} else {
> +		put_task_struct(info->task);
> +		if (info->file) {
> +			fput(info->file);
> +			info->file = NULL;
> +		}
> +		info->task = NULL;
> +	}
> +}
> +
[...]