ASoC: SOF: ipc-msg-injector: fix copy in sof_msg_inject_ipc4_dfs_write()
Subsystems (4)
PATCH 1/1 ASoC: SOF: ipc-msg-injector: fix copy in sof_msg_inject_ipc4_dfs_write()
The bug here is that when we copy the payload the value of *ppos should be zero but it is sizeof(ipc4_msg->header_u64) instead. That means that the buffer will be copied to the wrong location within the ipc4_msg->data_ptr buffer.
Really, in this context, it is simpler and more appropriate to use copy_from_user() instead of simple_write_to_buffer() so I have re-written the function.
---
From static analysis. Not tested. I believe this function is mostly used to write random junk to the device and see what breaks. So probably it works fine as-is.
sound/soc/sof/sof-client-ipc-msg-injector.c | 27 ++++++++------------- 1 file changed, 10 insertions(+), 17 deletions(-)
sound/soc/sof/sof-client-ipc-msg-injector.c +10 -17
@@ -181,7 +181,7 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,struct sof_client_dev *cdev = file->private_data;struct sof_msg_inject_priv *priv = cdev->data;struct sof_ipc4_msg *ipc4_msg = priv->tx_buffer; -ssize_t size; +size_t data_size;int ret;if (*ppos)
@@ -191,25 +191,18 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,return -EINVAL ;/* copy the header first */ - size =simple_write_to_buffer (&ipc4_msg->header_u64, -sizeof (ipc4_msg->header_u64), - ppos, buffer, count); -if (size <0 ) -return size; -if (size !=sizeof (ipc4_msg->header_u64)) +if (copy_from_user (&ipc4_msg->header_u64, buffer, +sizeof (ipc4_msg->header_u64)))return -EFAULT ; - count -= size; + data_size = count -sizeof (ipc4_msg->header_u64); +if (data_size > priv->max_msg_size) +return -EINVAL ;/* Copy the payload */ - size =simple_write_to_buffer (ipc4_msg->data_ptr, - priv->max_msg_size, ppos, buffer, - count); -if (size <0 ) -return size; -if (size != count) +if (copy_from_user (ipc4_msg->data_ptr, buffer, data_size))return -EFAULT ; - ipc4_msg->data_size = count; + ipc4_msg->data_size = data_size;/* Initialize the reply storage */ ipc4_msg = priv->rx_buffer;
@@ -221,9 +214,9 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,/* return the error code if test failed */ if (ret <0 ) - size = ret; + count = ret; -return size; +return count; };static int sof_msg_inject_dfs_release (struct inode *inode,struct file *file)
Signature · 1 line
2.35.1
Péter Ujfalusi
maintainerQuoted 48 lines · Dan Carpenter
The bug here is that when we copy the payload the value of *ppos should be zero but it is
sizeof(ipc4_msg->header_u64)instead. That means that the buffer will be copied to the wrong location within theipc4_msg->data_ptrbuffer.Really, in this context, it is simpler and more appropriate to use
copy_from_user()instead ofsimple_write_to_buffer()so I have re-written the function.Fixes 066c67624d8c ("ASoC: SOF: ipc-msg-injector: Add support for IPC4 messages")Signed-off-by Dan Carpenter <[email protected]>---
From static analysis. Not tested. I believe this function is mostly used to write random junk to the device and see what breaks. So probably it works fine as-is.sound/soc/sof/sof-client-ipc-msg-injector.c | 27 ++++++++------------- 1 file changed, 10 insertions(+), 17 deletions(-)sound/soc/sof/sof-client-ipc-msg-injector.c +4
-8@@ -181,7 +181,7 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,struct sof_client_dev *cdev = file->private_data;struct sof_msg_inject_priv *priv = cdev->data;struct sof_ipc4_msg *ipc4_msg = priv->tx_buffer; -ssize_t size; +size_t data_size;int ret;if (*ppos)@@ -191,25 +191,18 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,return -EINVAL ;/* copy the header first */ - size =simple_write_to_buffer (&ipc4_msg->header_u64, -sizeof (ipc4_msg->header_u64), - ppos, buffer, count); -if (size <0 ) -return size; -if (size !=sizeof (ipc4_msg->header_u64)) +if (copy_from_user (&ipc4_msg->header_u64, buffer, +sizeof (ipc4_msg->header_u64)))return -EFAULT ; - count -= size; + data_size = count -sizeof (ipc4_msg->header_u64);
sound/soc/sof/sof-client-ipc-msg-injector.c static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file, · line 202
Quoted 10 lines
sound/soc/sof/sof-client-ipc-msg-injector.c +3
-6+if (data_size > priv->max_msg_size) +return -EINVAL ;/* Copy the payload */ - size =simple_write_to_buffer (ipc4_msg->data_ptr, - priv->max_msg_size, ppos, buffer, - count); -if (size <0 ) -return size; -if (size != count) +if (copy_from_user (ipc4_msg->data_ptr, buffer, data_size))
I think this is still needs an update:
if (copy_from_user(ipc4_msg->data_ptr,
buffer + sizeof(ipc4_msg->header_u64), data_size))
To skip the already copied header.
Or without a rewrite the fix would be simple as:
/* Copy the payload */
size = simple_write_to_buffer(ipc4_msg->data_ptr, 0,
buffer + sizeof(ipc4_msg->header_u64), data_size), count);
Quoted 19 lines
sound/soc/sof/sof-client-ipc-msg-injector.c +3
-3return -EFAULT ; - ipc4_msg->data_size = count; + ipc4_msg->data_size = data_size;/* Initialize the reply storage */ ipc4_msg = priv->rx_buffer;@@ -221,9 +214,9 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,/* return the error code if test failed */ if (ret <0 ) - size = ret; + count = ret; -return size; +return count; };static int sof_msg_inject_dfs_release (struct inode *inode,struct file *file)
Signature · 1 line
Péter
Quoted 64 lines · Péter Ujfalusi
Quoted 48 lines · Dan Carpenter
The bug here is that when we copy the payload the value of *ppos should be zero but it is
sizeof(ipc4_msg->header_u64)instead. That means that the buffer will be copied to the wrong location within theipc4_msg->data_ptrbuffer.Really, in this context, it is simpler and more appropriate to use
copy_from_user()instead ofsimple_write_to_buffer()so I have re-written the function.Fixes 066c67624d8c ("ASoC: SOF: ipc-msg-injector: Add support for IPC4 messages")Signed-off-by Dan Carpenter <[email protected]>---
From static analysis. Not tested. I believe this function is mostly used to write random junk to the device and see what breaks. So probably it works fine as-is.sound/soc/sof/sof-client-ipc-msg-injector.c | 27 ++++++++------------- 1 file changed, 10 insertions(+), 17 deletions(-)sound/soc/sof/sof-client-ipc-msg-injector.c +4
-8@@ -181,7 +181,7 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,struct sof_client_dev *cdev = file->private_data;struct sof_msg_inject_priv *priv = cdev->data;struct sof_ipc4_msg *ipc4_msg = priv->tx_buffer; -ssize_t size; +size_t data_size;int ret;if (*ppos)@@ -191,25 +191,18 @@ static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file,return -EINVAL ;/* copy the header first */ - size =simple_write_to_buffer (&ipc4_msg->header_u64, -sizeof (ipc4_msg->header_u64), - ppos, buffer, count); -if (size <0 ) -return size; -if (size !=sizeof (ipc4_msg->header_u64)) +if (copy_from_user (&ipc4_msg->header_u64, buffer, +sizeof (ipc4_msg->header_u64)))return -EFAULT ; - count -= size; + data_size = count -sizeof (ipc4_msg->header_u64);sound/soc/sof/sof-client-ipc-msg-injector.c static ssize_t sof_msg_inject_ipc4_dfs_write(struct file *file, · line 202
Quoted 10 lines
sound/soc/sof/sof-client-ipc-msg-injector.c +3
-6+if (data_size > priv->max_msg_size) +return -EINVAL ;/* Copy the payload */ - size =simple_write_to_buffer (ipc4_msg->data_ptr, - priv->max_msg_size, ppos, buffer, - count); -if (size <0 ) -return size; -if (size != count) +if (copy_from_user (ipc4_msg->data_ptr, buffer, data_size))I think this is still needs an update:
if (copy_from_user(ipc4_msg->data_ptr,
Quoted 3 lines
buffer + sizeof(ipc4_msg->header_u64), data_size))To skip the already copied header.
Oh yeah. Thanks. Will do.
Quoted 6 lines
Or without a rewrite the fix would be simple as:
/* Copy the payload */
size = simple_write_to_buffer(ipc4_msg->data_ptr, 0,buffer + sizeof(ipc4_msg->header_u64), data_size), count);
That doesn't work. Simple_write_to_buffer takes a pointer and not a value. So it's pretty common to do:
llof_t dummy = 0; ... size = simple_write_to_buffer(ipc4_msg->data_ptr, &dummy, buffer + sizeof(ipc4_msg->header_u64), data_size), count);
But the simple_write_to_buffer() function is kind of bad. The only thing which saves us from a security nightmare is that it's mostly used in debugfs so it's root only.
The problem with simple_write_to_buffer is that it leads to uninitialized variable bugs if we don't zero out the buffer at the start of the function. It returns a positive success value if it is able to write even one byte somewhere in the buffer.
The start of the buffer can be uninitialized if *ppos is non-zero.
I think I have looked through every caller and I have not seen one where we care about *ppos. A bunch do the dummy = 0 thing. For the rest we could just add an "if (*ppos) return 0;" at the start of the function. These functions are writing a line. Just do it in one go. It's not like the rest of the write functions are set up to handle partial writes.
So in other words, you can find a code which will accept a non-zero *ppos and it's not going to crash, but the data from the user is invariably going to nonsense.
The end part of the buffer can be uninitialized. A lot of callers address this by adding a check:
size = simple_write_to_buffer(); if (size != buf_size) return size < 0 ? size : -EIO;
That ensures that *ppos is zero and that it was able to initialize the whole buffer. But if you want to fill a whole buffer then just use copy_from_user().
The only advantage of simple_write_to_buffer() is that it takes both the size of the buffer and the size of write into account. However it's probably better to just do it manually/explicitly:
if (*ppos) return 0; ... if (count > size_of_buf) count = size_of_buf; if (copy_from_user(buf, user, count)) return -EFAULT;
Looking at the sof_msg_inject_ipc4_dfs_write() function again, we check that:
size = simple_write_to_buffer( ... if (size != count) return -EFAULT;
This ensures that we wrote as much as expected, but it doesn't ensure that we filled the buffer. But since this function is just use to inject garbage data as part of fuzzing then re-using old data is fine.
regards,
dan carpenter