Re: [PATCH v3 bpf-next 1/4] bpf: introduce task_vma bpf_iter

[Yonghong Song <yhs@fb.com>]

On 1/6/21 8:17 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.
> 
> Current /proc/<pid>/maps and /proc/<pid>/smaps provide information of
> vma's of a process. However, these information are not flexible enough to
> cover all use cases. For example, if a vma cover mixed 2MB pages and 4kB
> pages (x86_64), there is no easy way to tell which address ranges are
> backed by 2MB pages. task_vma solves the problem by enabling the user to
> generate customize information based on the vma (and vma->vm_mm,
> vma->vm_file, etc.).
> 
> To access the vma safely in the BPF program, task_vma iterator holds
> target mmap_lock while calling the BPF program. If the mmap_lock is
> contended, task_vma unlocks mmap_lock between iterations to unblock the
> writer(s). This lock contention avoidance mechanism is similar to the one
> used in show_smaps_rollup().
> 
> Signed-off-by: Song Liu <songliubraving@fb.com>
> ---
>   kernel/bpf/task_iter.c | 211 ++++++++++++++++++++++++++++++++++++++++-
>   1 file changed, 210 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/bpf/task_iter.c b/kernel/bpf/task_iter.c
> index 0458a40edf10a..a1d3c9d230f68 100644
> --- a/kernel/bpf/task_iter.c
> +++ b/kernel/bpf/task_iter.c
> @@ -304,9 +304,192 @@ static const struct seq_operations task_file_seq_ops = {
>   	.show	= task_file_seq_show,
>   };
>   
> +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;
> +	u32 tid;
> +	unsigned long prev_vm_start;
> +	unsigned long prev_vm_end;
> +};
> +
> +enum bpf_task_vma_iter_find_op {
> +	task_vma_iter_first_vma,   /* use mm->mmap */
> +	task_vma_iter_next_vma,    /* use curr_vma->vm_next */
> +	task_vma_iter_find_vma,    /* use find_vma() to find next vma */
> +};
> +
> +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;
> +	enum bpf_task_vma_iter_find_op op;
> +	struct vm_area_struct *curr_vma;
> +	struct task_struct *curr_task;
> +	u32 curr_tid = info->tid;
> +
> +	/* If this function returns a non-NULL vma, it holds a reference to
> +	 * the task_struct, and holds read lock on vma->mm->mmap_lock.
> +	 * If this function returns NULL, it does not hold any reference or
> +	 * lock.
> +	 */
> +again:

We can move label "again" to below right before "curr_task = 
task_seq_get_next()" since "goto again" always having info->task == NULL.

> +	if (info->task) {
> +		curr_task = info->task;
> +		curr_vma = info->vma;
> +		/* In case of lock contention, drop mmap_lock to unblock
> +		 * the writer.
> +		 */
> +		if (mmap_lock_is_contended(curr_task->mm)) {
> +			info->prev_vm_start = curr_vma->vm_start;
> +			info->prev_vm_end = curr_vma->vm_end;
> +			op = task_vma_iter_find_vma;
> +			mmap_read_unlock(curr_task->mm);
> +			if (mmap_read_lock_killable(curr_task->mm))

Could you explain when mmap_read_lock_killable() may fail?
What I try to find is if mmap_read_lock_killable() fails, we should
abort iterating or we may request user to try again (if user buffer
is not filled anything).

> +				goto finish;
> +		} else {
> +			op = task_vma_iter_next_vma;
> +		}
> +	} else {
> +		curr_task = task_seq_get_next(ns, &curr_tid, true);
> +		if (!curr_task) {
> +			info->tid = curr_tid + 1;
> +			goto finish;
> +		}
> +
> +		if (curr_tid != info->tid) {
> +			info->tid = curr_tid;
> +			op = task_vma_iter_first_vma;
> +		} else {
> +			op = task_vma_iter_find_vma;
> +		}
> +
> +		if (!curr_task->mm)
> +			goto next_task;
> +
> +		if (mmap_read_lock_killable(curr_task->mm))
> +			goto finish;
> +	}
> +
> +	switch (op) {
> +	case task_vma_iter_first_vma:
> +		curr_vma = curr_task->mm->mmap;
> +		break;
> +	case task_vma_iter_next_vma:
> +		curr_vma = curr_vma->vm_next;
> +		break;
> +	case task_vma_iter_find_vma:
> +		/* We dropped mmap_lock so it is necessary to use find_vma
> +		 * to find the next vma. This is similar to the  mechanism
> +		 * in show_smaps_rollup().
> +		 */
> +		curr_vma = find_vma(curr_task->mm, info->prev_vm_end - 1);
> +
> +		if (curr_vma && (curr_vma->vm_start == info->prev_vm_start))
> +			curr_vma = curr_vma->vm_next;
> +		break;
> +	}
> +	if (!curr_vma) {
> +		mmap_read_unlock(curr_task->mm);
> +		goto next_task;
> +	}
> +	info->task = curr_task;
> +	info->vma = curr_vma;
> +	return curr_vma;
> +
> +next_task:
> +	put_task_struct(curr_task);
> +	info->task = NULL;
> +	curr_tid++;
> +	goto again;
> +
> +finish:
> +	info->task = NULL;
> +	info->vma = NULL;
> +	return NULL;
> +}
> +
> +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;
> +	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);
> +};
> +
> +DEFINE_BPF_ITER_FUNC(task_vma, struct bpf_iter_meta *meta,
> +		     struct task_struct *task, struct vm_area_struct *vma)
> +
> +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;
> +	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 {
> +		info->prev_vm_start = info->vma->vm_start;
> +		info->prev_vm_end = info->vma->vm_end;

This change, together with task_vma_iter_find_vma action,
is not correct.

What we typically have is info->vma->vm_start/info->vma->vm_end
is the next show but since there is no room for that we will
delay until user space processed buffer and come back.
So here, it is not prev_vm_start/end as it has not been
displayed.

Maybe you can separate task_vma_iter_find_vma into two?
task_vma_iter_find_vma and task_vma_iter_find_vma_next.
(task_vma_iter_find_vma_next carriers the action with
what you have now for task_vma_iter_find_vma).

In the above task_vma_seq_get_next(), if info->task != NULL,
you will use task_vma_iter_find_vma_next (the one after previous
vm_start/end), if info->task == NULL, which means back from user
space, not going to next vma if there is an exact match.

> +		mmap_read_unlock(info->task->mm);
> +		put_task_struct(info->task);
> +		info->task = NULL;
> +	}
> +}
> +
> +static const struct seq_operations task_vma_seq_ops = {
> +	.start	= task_vma_seq_start,
> +	.next	= task_vma_seq_next,
> +	.stop	= task_vma_seq_stop,
> +	.show	= task_vma_seq_show,
> +};
> +
>   BTF_ID_LIST(btf_task_file_ids)
>   BTF_ID(struct, task_struct)
>   BTF_ID(struct, file)
> +BTF_ID(struct, vm_area_struct)
>   
>   static const struct bpf_iter_seq_info task_seq_info = {
>   	.seq_ops		= &task_seq_ops,
> @@ -346,6 +529,26 @@ static struct bpf_iter_reg task_file_reg_info = {
>   	.seq_info		= &task_file_seq_info,
>   };
>   
> +static const struct bpf_iter_seq_info task_vma_seq_info = {
> +	.seq_ops		= &task_vma_seq_ops,
> +	.init_seq_private	= init_seq_pidns,
> +	.fini_seq_private	= fini_seq_pidns,
> +	.seq_priv_size		= sizeof(struct bpf_iter_seq_task_vma_info),
> +};
> +
> +static struct bpf_iter_reg task_vma_reg_info = {
> +	.target			= "task_vma",
> +	.feature		= BPF_ITER_RESCHED,
> +	.ctx_arg_info_size	= 2,
> +	.ctx_arg_info		= {
> +		{ offsetof(struct bpf_iter__task_vma, task),
> +		  PTR_TO_BTF_ID_OR_NULL },
> +		{ offsetof(struct bpf_iter__task_vma, vma),
> +		  PTR_TO_BTF_ID_OR_NULL },
> +	},
> +	.seq_info		= &task_vma_seq_info,
> +};
> +
>   static int __init task_iter_init(void)
>   {
>   	int ret;
> @@ -357,6 +560,12 @@ static int __init task_iter_init(void)
>   
>   	task_file_reg_info.ctx_arg_info[0].btf_id = btf_task_file_ids[0];
>   	task_file_reg_info.ctx_arg_info[1].btf_id = btf_task_file_ids[1];
> -	return bpf_iter_reg_target(&task_file_reg_info);
> +	ret =  bpf_iter_reg_target(&task_file_reg_info);
> +	if (ret)
> +		return ret;
> +
> +	task_vma_reg_info.ctx_arg_info[0].btf_id = btf_task_file_ids[0];
> +	task_vma_reg_info.ctx_arg_info[1].btf_id = btf_task_file_ids[2];
> +	return bpf_iter_reg_target(&task_vma_reg_info);
>   }
>   late_initcall(task_iter_init);
>