[Linaro-mm-sig] Re: [RFC v5 5/6] binder: Add flags to relinquish ownership of fds

On Thu, Apr 21, 2022 at 11:28 AM Carlos Llamas cmllamas@google.com wrote:

On Wed, Apr 20, 2022 at 11:52:23PM +0000, T.J. Mercier wrote:

...

From: Hridya Valsaraju hridya@google.com

This patch introduces flags BINDER_FD_FLAG_SENDER_NO_NEED, and BINDER_FDA_FLAG_SENDER_NO_NEED that a process sending an individual fd or fd array to another process over binder IPC can set to relinquish ownership of the fds being sent for memory accounting purposes. If the flag is found to be set during the fd or fd array translation and the fd is for a DMA-BUF, the buffer is uncharged from the sender's cgroup and charged to the receiving process's cgroup instead.

It is up to the sending process to ensure that it closes the fds regardless of whether the transfer failed or succeeded.

Most graphics shared memory allocations in Android are done by the graphics allocator HAL process. On requests from clients, the HAL process allocates memory and sends the fds to the clients over binder IPC. The graphics allocator HAL will not retain any references to the buffers. When the HAL sets *_FLAG_SENDER_NO_NEED for fd arrays holding DMA-BUF fds, or individual fd objects, the gpu cgroup controller will be able to correctly charge the buffers to the client processes instead of the graphics allocator HAL.

Since this is a new feature exposed to userspace, the kernel and userspace must be compatible for the accounting to work for transfers. In all cases the allocation and transport of DMA buffers via binder will succeed, but only when both the kernel supports, and userspace depends on this feature will the transfer accounting work. The possible scenarios are detailed below:

New binder driver features which require userspace coordination can be "advertised" by the kernel via binderfs. You can see an example of how oneway_spam_detection is exposed in commit fc470abf54b2 ("binderfs: add support for feature files"). This is just an option to consider if it makes things easier in userspace. Although it seems that for the second scenario (old kernel + new userpsace) the flags would just be ignored.

This is a cool idea. You're right that the flags would be ignored. Since this isn't a binder feature that can be toggled like BINDER_ENABLE_ONEWAY_SPAM_DETECTION, I think the presence of the GPU cgroup controller in the cgroup.controllers file would also tell us the same thing from userspace.

...

1. new kernel + old userspace

The kernel supports the feature but userspace does not use it. The old userspace won't mount the new cgroup controller, accounting is not performed, charge is not transferred.

2. old kernel + new userspace

The new cgroup controller is not supported by the kernel, accounting is not performed, charge is not transferred.

3. old kernel + old userspace

Same as #2

4. new kernel + new userspace

Cgroup is mounted, feature is supported and used.

Signed-off-by: Hridya Valsaraju hridya@google.com Signed-off-by: T.J. Mercier tjmercier@google.com

---

v5 changes Support both binder_fd_array_object and binder_fd_object. This is necessary because new versions of Android will use binder_fd_object instead of binder_fd_array_object, and we need to support both.

Use the new, simpler dma_buf_transfer_charge API.

v3 changes Remove android from title per Todd Kjos.

Use more common dual author commit message format per John Stultz.

Include details on behavior for all combinations of kernel/userspace versions in changelog (thanks Suren Baghdasaryan) per Greg Kroah-Hartman.

v2 changes Move dma-buf cgroup charge transfer from a dma_buf_op defined by every heap to a single dma-buf function for all heaps per Daniel Vetter and Christian König.

---

drivers/android/binder.c | 27 +++++++++++++++++++++++---- drivers/dma-buf/dma-buf.c | 4 ++-- include/linux/dma-buf.h | 2 +- include/uapi/linux/android/binder.h | 23 +++++++++++++++++++---- 4 files changed, 45 insertions(+), 11 deletions(-)

diff --git a/drivers/android/binder.c b/drivers/android/binder.c index 8351c5638880..b07d50fe1c80 100644 --- a/drivers/android/binder.c +++ b/drivers/android/binder.c @@ -42,6 +42,7 @@

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

+#include <linux/dma-buf.h> #include <linux/fdtable.h> #include <linux/file.h> #include <linux/freezer.h> @@ -2170,7 +2171,7 @@ static int binder_translate_handle(struct flat_binder_object *fp, return ret; }

-static int binder_translate_fd(u32 fd, binder_size_t fd_offset, +static int binder_translate_fd(u32 fd, binder_size_t fd_offset, __u32 flags, struct binder_transaction *t, struct binder_thread *thread, struct binder_transaction *in_reply_to) @@ -2208,6 +2209,23 @@ static int binder_translate_fd(u32 fd, binder_size_t fd_offset, goto err_security; }

• if (IS_ENABLED(CONFIG_CGROUP_GPU) && (flags & BINDER_FD_FLAG_SENDER_NO_NEED)) {
  

•         if (is_dma_buf_file(file)) {
  

•                 struct dma_buf *dmabuf = file->private_data;
  

•                 ret = dma_buf_transfer_charge(dmabuf, target_proc->tsk);
  

•                 if (ret)
  

•                         pr_warn("%d:%d Unable to transfer DMA-BUF fd charge to %d\n",
  

•                                 proc->pid, thread->pid, target_proc->pid);
  

If we fail to transfer the charge, it seems we continue with the fixup allocation and then propagate the error. Shouldn't the translation be aborted at this point instead? Or is this supposed to be handled?

I took the position that it was better to have incorrect accounting along with this log statement than potentially causing lots of crashes due to failed transactions. However if limiting gets added for the GPU cgroup, then we really should kill the transaction here. I'll go ahead and add that goto now.

...

•         } else {
  

nit: negating is_dma_buf_file() check eliminates the "else" here.

Thanks.

...

•                 binder_user_error(
  

•                         "%d:%d got transaction with SENDER_NO_NEED for non-dmabuf fd, %d\n",
  

•                         proc->pid, thread->pid, fd);
  

•                 ret = -EINVAL;
  

•                 goto err_noneed;
  

•         }
  

• }
  

• /*
   * Add fixup record for this transaction. The allocation
   * of the fd in the target needs to be done from a
  

@@ -2226,6 +2244,7 @@ static int binder_translate_fd(u32 fd, binder_size_t fd_offset, return ret;

err_alloc: +err_noneed: err_security: fput(file); err_fget: @@ -2528,7 +2547,7 @@ static int binder_translate_fd_array(struct list_head *pf_head,

          ret = copy_from_user(&fd, sender_ufda_base + sender_uoffset, sizeof(fd));
          if (!ret)

•                 ret = binder_translate_fd(fd, offset, t, thread,
  

•                 ret = binder_translate_fd(fd, offset, fda->flags, t, thread,
                                            in_reply_to);
          if (ret)
                  return ret > 0 ? -EINVAL : ret;
  

@@ -3179,8 +3198,8 @@ static void binder_transaction(struct binder_proc *proc, struct binder_fd_object *fp = to_binder_fd_object(hdr); binder_size_t fd_offset = object_offset + (uintptr_t)&fp->fd - (uintptr_t)fp;

•                 int ret = binder_translate_fd(fp->fd, fd_offset, t,
  

•                                               thread, in_reply_to);
  

•                 int ret = binder_translate_fd(fp->fd, fd_offset, fp->flags,
  

•                                               t, thread, in_reply_to);
  
                  fp->pad_binder = 0;
                  if (ret < 0 ||
  

diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index f3fb844925e2..36ed6cd4ddcc 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -31,7 +31,6 @@

#include "dma-buf-sysfs-stats.h"

-static inline int is_dma_buf_file(struct file *);

struct dma_buf_list { struct list_head head; @@ -400,10 +399,11 @@ static const struct file_operations dma_buf_fops = { /*

• is_dma_buf_file - Check if struct file* is associated with dma_buf

*/ -static inline int is_dma_buf_file(struct file *file) +int is_dma_buf_file(struct file *file) { return file->f_op == &dma_buf_fops; } +EXPORT_SYMBOL_NS_GPL(is_dma_buf_file, DMA_BUF);

static struct file *dma_buf_getfile(struct dma_buf *dmabuf, int flags) { diff --git a/include/linux/dma-buf.h b/include/linux/dma-buf.h index 438ad8577b76..2b9812758fee 100644 --- a/include/linux/dma-buf.h +++ b/include/linux/dma-buf.h @@ -614,7 +614,7 @@ dma_buf_attachment_is_dynamic(struct dma_buf_attachment *attach) { return !!attach->importer_ops; }

+int is_dma_buf_file(struct file *file); struct dma_buf_attachment *dma_buf_attach(struct dma_buf *dmabuf, struct device *dev); struct dma_buf_attachment * diff --git a/include/uapi/linux/android/binder.h b/include/uapi/linux/android/binder.h index 11157fae8a8e..b263cbb603ea 100644 --- a/include/uapi/linux/android/binder.h +++ b/include/uapi/linux/android/binder.h @@ -91,14 +91,14 @@ struct flat_binder_object { /**

• struct binder_fd_object - describes a filedescriptor to be fixed up.
• @hdr: common header structure

• • @pad_flags: padding to remain compatible with old userspace code

Does this mean we no longer need to keep the compatibility with the "old userspace code"? Maybe these old flags are all less than 0x2000?

This comes from before binder_fd{_array}_object existed as a distinct type from flat_binder_object. With this layout, it's possible to cast between flat_binder_object and the binder_fd{_array}_object types. I don't think there were ever any binder_fd{_array}_object specific flags before now, but yes the value of 0x2000 was chosen to be sure that the FLAT_BINDER_* flags do not conflict. I did try smaller values (0x02) but found that occasionally this bit was set when I was not expecting it to be.

https://lore.kernel.org/lkml/1486161652-2612-2-git-send-email-john.stultz@li...

...

• • @flags: One or more BINDER_FD_FLAG_* flags
  • @pad_binder: padding to remain compatible with old userspace code
  • @fd: file descriptor
  • @cookie: opaque data, used by user-space
  */

struct binder_fd_object { struct binder_object_header hdr;

• __u32                           pad_flags;
  

• __u32                           flags;
  union {
          binder_uintptr_t        pad_binder;
          __u32                   fd;
  

@@ -107,6 +107,17 @@ struct binder_fd_object { binder_uintptr_t cookie; };

+enum {

• /**
  

•  * @BINDER_FD_FLAG_SENDER_NO_NEED
  

•  *
  

•  * When set, the sender of a binder_fd_object wishes to relinquish ownership of the fd for
  

•  * memory accounting purposes. If the fd is for a DMA-BUF, the buffer is uncharged from the
  

•  * sender's cgroup and charged to the receiving process's cgroup instead.
  

•  */
  

• BINDER_FD_FLAG_SENDER_NO_NEED = 0x2000,
  

SENDER_NO_NEED wasn't straight-forward for me. Perhaps RELINQUISH or XFER_{OWNER|CHARGE|CGROUP} could be some other options to consider.

I'm happy to change this up. I like _XFER_CHARGE the best out of these.

...

+};

/* struct binder_buffer_object - object describing a userspace buffer

• @hdr: common header structure
• @flags: one or more BINDER_BUFFER_* flags

@@ -141,7 +152,7 @@ enum {

/* struct binder_fd_array_object - object describing an array of fds in a buffer

• @hdr: common header structure

• • @pad: padding to ensure correct alignment

• • flags: One or more BINDER_FDA_FLAG_* flags
  • @num_fds: number of file descriptors in the buffer
  • @parent: index in offset array to buffer holding the fd array
  • @parent_offset: start offset of fd array in the buffer

@@ -162,12 +173,16 @@ enum { */ struct binder_fd_array_object { struct binder_object_header hdr;

• __u32                           pad;
  

• __u32                           flags;
  binder_size_t                   num_fds;
  binder_size_t                   parent;
  binder_size_t                   parent_offset;
  

};

+enum {

• BINDER_FDA_FLAG_SENDER_NO_NEED = BINDER_FD_FLAG_SENDER_NO_NEED,
  

+};

/*

• On 64-bit platforms where user code may run in 32-bits the driver must
• translate the buffer (and local binder) addresses appropriately.

-- 2.36.0.rc0.470.gd361397f0d-goog

Other than included minor comments:

Reviewed-by: Carlos Llamas cmllamas@google.com

-- Carlos Llamas