Re: [bpf-next v2 1/2] bpf: Add bpf_copy_from_user_task_str kfunc

[Andrii Nakryiko <andrii.nakryiko@gmail.com>]

On Mon, Jan 6, 2025 at 6:12 PM Jordan Rome <linux@jordanrome.com> wrote:
>
> This new kfunc will be able to copy a string
> from another process's/task's address space.

nit: this is kernel code, task is unambiguous, so I'd drop the
"process" reference here

> This is similar to `bpf_copy_from_user_str`
> but accepts a `struct task_struct*` argument.
>
> This required adding an additional function
> in memory.c, namely `copy_str_from_process_vm`,
> which works similar to `access_process_vm`
> but utilizes the `strncpy_from_user` helper
> and only supports reading/copying and not writing.
>
> Signed-off-by: Jordan Rome <linux@jordanrome.com>
> ---
>  include/linux/mm.h   |   3 ++
>  kernel/bpf/helpers.c |  46 ++++++++++++++++++++
>  mm/memory.c          | 101 +++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 150 insertions(+)
>

please check kernel test bot's complains as well

> diff --git a/include/linux/mm.h b/include/linux/mm.h
> index c39c4945946c..52b304b20630 100644
> --- a/include/linux/mm.h
> +++ b/include/linux/mm.h
> @@ -2484,6 +2484,9 @@ extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
>  extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
>                 void *buf, int len, unsigned int gup_flags);
>
> +extern int copy_str_from_process_vm(struct task_struct *tsk, unsigned long addr,
> +               void *buf, int len, unsigned int gup_flags);

nit: curious what mm folks think about naming, I'd go with a slightly
less verbose naming: "copy_remote_vm_str" (copy vs access, _str suffix
for non fixed-sized semantics marking)

for the next revision, let's split out mm parts from helpers parts, I
don't think we lose much as this new internal API is self-contained,
and it will be easier for mm folks to review

> +
>  long get_user_pages_remote(struct mm_struct *mm,
>                            unsigned long start, unsigned long nr_pages,
>                            unsigned int gup_flags, struct page **pages,
> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> index cd5f9884d85b..45d41b7a9906 100644
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
> @@ -3072,6 +3072,51 @@ __bpf_kfunc void bpf_local_irq_restore(unsigned long *flags__irq_flag)
>         local_irq_restore(*flags__irq_flag);
>  }
>
> +/**
> + * bpf_copy_from_user_task_str() - Copy a string from an task's address space
> + * @dst:             Destination address, in kernel space.  This buffer must be
> + *                   at least @dst__sz bytes long.
> + * @dst__sz:         Maximum number of bytes to copy, includes the trailing NUL.
> + * @unsafe_ptr__ign: Source address in the task's address space.
> + * @tsk:             The task whose address space will be used
> + * @flags:           The only supported flag is BPF_F_PAD_ZEROS
> + *
> + * Copies a NULL-terminated string from a task's address space to BPF space.

there is no "BPF space", really... maybe "copies string into *dst* buffer"

> + * If user string is too long this will still ensure zero termination in the
> + * dst buffer unless buffer size is 0.
> + *
> + * If BPF_F_PAD_ZEROS flag is set, memset the tail of @dst to 0 on success and
> + * memset all of @dst on failure.
> + */
> +__bpf_kfunc int bpf_copy_from_user_task_str(void *dst, u32 dst__sz, const void __user *unsafe_ptr__ign, struct task_struct *tsk, u64 flags)

this looks like a long line, does it fit under 100 characters?

> +{
> +       int count = dst__sz - 1;
> +       int ret = 0;
> +
> +       if (unlikely(flags & ~BPF_F_PAD_ZEROS))
> +               return -EINVAL;
> +
> +       if (unlikely(!dst__sz))
> +               return 0;
> +
> +       ret = copy_str_from_process_vm(tsk, (unsigned long)unsafe_ptr__ign, dst, count, 0);
> +
> +       if (ret <= 0) {
> +               if (flags & BPF_F_PAD_ZEROS)
> +                       memset((char *)dst, 0, dst__sz);

nit: no need for (char *) cast? memset takes void *, I think

> +               return ret;

if ret == 0, is that an error? If so, `return ret ?: -EINVAL;` or
something along those lines?

pw-bot: cr

> +       }
> +
> +       if (ret < count) {
> +               if (flags & BPF_F_PAD_ZEROS)
> +                       memset((char *)dst + ret, 0, dst__sz - ret);
> +       } else {
> +               ((char *)dst)[count] = '\0';
> +       }
> +
> +       return ret + 1;
> +}
> +
>  __bpf_kfunc_end_defs();
>
>  BTF_KFUNCS_START(generic_btf_ids)
> @@ -3164,6 +3209,7 @@ BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
>  BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
>  BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
>  BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
> +BTF_ID_FLAGS(func, bpf_copy_from_user_task_str, KF_SLEEPABLE)
>  BTF_ID_FLAGS(func, bpf_get_kmem_cache)
>  BTF_ID_FLAGS(func, bpf_iter_kmem_cache_new, KF_ITER_NEW | KF_SLEEPABLE)
>  BTF_ID_FLAGS(func, bpf_iter_kmem_cache_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPABLE)
> diff --git a/mm/memory.c b/mm/memory.c
> index 75c2dfd04f72..514490bd7d6d 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -6673,6 +6673,75 @@ static int __access_remote_vm(struct mm_struct *mm, unsigned long addr,
>         return buf - old_buf;
>  }
>
> +/*
> + * Copy a string from another process's address space as given in mm.
> + * Don't return partial results. If there is any error return -EFAULT.

What does "don't return partial results" mean? What happens if we read
part of a string and then fail to read the rest?

> + */
> +static int __copy_str_from_remote_vm(struct mm_struct *mm, unsigned long addr,
> +                             void *buf, int len, unsigned int gup_flags)
> +{
> +       void *old_buf = buf;
> +       int err = 0;
> +
> +       if (mmap_read_lock_killable(mm))
> +               return -EFAULT;
> +
> +       /* Untag the address before looking up the VMA */
> +       addr = untagged_addr_remote(mm, addr);
> +
> +       /* Avoid triggering the temporary warning in __get_user_pages */
> +       if (!vma_lookup(mm, addr)) {
> +               mmap_read_unlock(mm);
> +               return -EFAULT;

maybe let's do (so that we do mmap_read_unlock in just one place)

err = -EFAULT;
goto err_out;

and then see below

> +       }
> +
> +       while (len) {
> +               int bytes, offset, retval;
> +               void *maddr;
> +               struct vm_area_struct *vma = NULL;
> +               struct page *page = get_user_page_vma_remote(mm, addr,
> +                                                            gup_flags, &vma);
> +

nit: I'd split page declaration and assignment and kept
get_user_page_vma_remote() invocation on a single line

> +               if (IS_ERR(page)) {
> +                       /*
> +                        * Treat as a total failure for now until we decide how
> +                        * to handle the CONFIG_HAVE_IOREMAP_PROT case and
> +                        * stack expansion.
> +                        */
> +                       err = -EFAULT;
> +                       break;
> +               }
> +
> +               bytes = len;
> +               offset = addr & (PAGE_SIZE - 1);
> +               if (bytes > PAGE_SIZE - offset)
> +                       bytes = PAGE_SIZE - offset;
> +
> +               maddr = kmap_local_page(page);
> +               retval = strncpy_from_user(buf, (const char __user *)addr, bytes);

you are not using maddr... that seems wrong (even if it works due to
how kmap_local_page is currently implemented)

> +               unmap_and_put_page(page, maddr);
> +
> +               if (retval < 0) {
> +                       err = retval;
> +                       break;
> +               }
> +
> +               len -= retval;
> +               buf += retval;
> +               addr += retval;
> +
> +               /* Found the end of the string */
> +               if (retval < bytes)
> +                       break;
> +       }

err_out: here

> +       mmap_read_unlock(mm);
> +
> +       if (err)
> +               return err;
> +
> +       return buf - old_buf;
> +}
> +
>  /**
>   * access_remote_vm - access another process' address space
>   * @mm:                the mm_struct of the target address space
> @@ -6714,6 +6783,38 @@ int access_process_vm(struct task_struct *tsk, unsigned long addr,
>  }
>  EXPORT_SYMBOL_GPL(access_process_vm);
>
> +/**
> + * copy_str_from_process_vm - copy a string from another process's address space.
> + * @tsk:       the task of the target address space
> + * @addr:      start address to access

access -> read from

> + * @buf:       source or destination buffer

for this api it's always the destination, right?

> + * @len:       number of bytes to transfer
> + * @gup_flags: flags modifying lookup behaviour
> + *
> + * The caller must hold a reference on @mm.
> + *
> + * Return: number of bytes copied from source to destination. If the string
> + * is shorter than @len then return the length of the string.

and if the string is longer than @len, then what happens? we should
either specify or drop the "if string is shorter bit" and make it
unambiguous whether terminating zero is included or not

> + * On any error, return -EFAULT.
> + */
> +int copy_str_from_process_vm(struct task_struct *tsk, unsigned long addr,
> +               void *buf, int len, unsigned int gup_flags)
> +{
> +       struct mm_struct *mm;
> +       int ret;
> +
> +       mm = get_task_mm(tsk);
> +       if (!mm)
> +               return -EFAULT;
> +
> +       ret = __copy_str_from_remote_vm(mm, addr, buf, len, gup_flags);
> +
> +       mmput(mm);
> +
> +       return ret;
> +}
> +EXPORT_SYMBOL_GPL(copy_str_from_process_vm);
> +
>  /*
>   * Print the name of a VMA.
>   */
> --
> 2.43.5
>