From: Yonghong Song <yhs@fb.com>
To: Song Liu <songliubraving@fb.com>, <bpf@vger.kernel.org>,
<netdev@vger.kernel.org>
Cc: <ast@kernel.org>, <daniel@iogearbox.net>, <andrii@kernel.org>,
<john.fastabend@gmail.com>, <kpsingh@chromium.org>,
<kernel-team@fb.com>
Subject: Re: [PATCH v2 bpf-next 1/4] bpf: introduce task_vma bpf_iter
Date: Wed, 16 Dec 2020 09:36:03 -0800 [thread overview]
Message-ID: <a0892e60-7cc3-80ed-f4d3-004128cb6b8e@fb.com> (raw)
In-Reply-To: <20201215233702.3301881-2-songliubraving@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;
> + }
> +}
> +
[...]
next prev parent reply other threads:[~2020-12-16 17:37 UTC|newest]
Thread overview: 36+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-12-15 23:36 [PATCH v2 bpf-next 0/4] introduce bpf_iter for task_vma Song Liu
2020-12-15 23:36 ` [PATCH v2 bpf-next 1/4] bpf: introduce task_vma bpf_iter Song Liu
2020-12-16 17:36 ` Yonghong Song [this message]
2020-12-16 19:41 ` Song Liu
2020-12-17 0:34 ` Andrii Nakryiko
2020-12-17 1:51 ` Song Liu
2020-12-17 19:03 ` Alexei Starovoitov
2020-12-17 22:08 ` Song Liu
2020-12-18 2:34 ` Alexei Starovoitov
2020-12-18 3:15 ` Yonghong Song
2020-12-18 4:33 ` Song Liu
2020-12-18 5:23 ` Alexei Starovoitov
2020-12-18 16:38 ` Yonghong Song
2020-12-18 17:23 ` Song Liu
2021-01-05 1:46 ` Alexei Starovoitov
2021-01-05 5:47 ` Song Liu
2021-01-05 16:27 ` Alexei Starovoitov
2021-01-05 17:10 ` Song Liu
2021-01-05 17:27 ` Alexei Starovoitov
2021-01-05 19:38 ` Song Liu
2021-01-05 19:46 ` Alexei Starovoitov
2021-01-05 19:51 ` Song Liu
2020-12-15 23:37 ` [PATCH v2 bpf-next 2/4] bpf: allow bpf_d_path in sleepable bpf_iter program Song Liu
2020-12-16 17:41 ` Yonghong Song
2020-12-16 18:15 ` KP Singh
2020-12-16 18:31 ` KP Singh
2021-01-25 12:52 ` Jiri Olsa
2020-12-15 23:37 ` [PATCH v2 bpf-next 3/4] libbpf: introduce section "iter.s/" for " Song Liu
2020-12-16 17:42 ` Yonghong Song
2020-12-16 18:00 ` KP Singh
2020-12-15 23:37 ` [PATCH v2 bpf-next 4/4] selftests/bpf: add test for bpf_iter_task_vma Song Liu
2020-12-16 18:18 ` Yonghong Song
2020-12-16 23:23 ` Song Liu
2020-12-16 17:00 ` [PATCH v2 bpf-next 0/4] introduce bpf_iter for task_vma Yonghong Song
2020-12-16 17:35 ` Song Liu
2020-12-16 18:31 ` Yonghong Song
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=a0892e60-7cc3-80ed-f4d3-004128cb6b8e@fb.com \
--to=yhs@fb.com \
--cc=andrii@kernel.org \
--cc=ast@kernel.org \
--cc=bpf@vger.kernel.org \
--cc=daniel@iogearbox.net \
--cc=john.fastabend@gmail.com \
--cc=kernel-team@fb.com \
--cc=kpsingh@chromium.org \
--cc=netdev@vger.kernel.org \
--cc=songliubraving@fb.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).