Skip to contentSkip to conversation

misc: fastrpc: Fix double free of 'buf' in error path

PATCH Superseded Sukrut Bellary 3 participants 14 messages May 18, 2023 – Jun 1, 2023

Subsystems (4)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 misc: fastrpc: Fix double free of 'buf' in error path

smatch warning:
drivers/misc/fastrpc.c:1926 fastrpc_req_mmap() error: double free of 'buf'

In fastrpc_req_mmap() error path, the fastrpc buffer is freed in fastrpc_req_munmap_impl() if unmap is successful.

But in the end, there is an unconditional call to fastrpc_buf_free(). So the above case triggers the double free of fastrpc buf.

Fix this by avoiding the call to the second fastrpc_buf_free() if fastrpc_req_munmap_impl() is successful.
'err' is not updated as it needs to retain the err returned by qcom_scm_assign_mem(), which is the starting point of this error path.

This is based on static analysis only. Compilation tested.

---
 drivers/misc/fastrpc.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/misc/fastrpc.c +4 -1
index f48466960f1b..1c3ab78f274f 100644
@@ -1921,7 +1921,10 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)
 	return 0;
 
 err_assign:
-	fastrpc_req_munmap_impl(fl, buf);
+	if (!fastrpc_req_munmap_impl(fl, buf)) {
+		/* buf is freed */
+		return err;
+	}
 err_invoke:
 	fastrpc_buf_free(buf);
 
Signature · 1 line
2.34.1
#1 lore raw
Quoted 9 lines · Sukrut Bellary

smatch warning:
drivers/misc/fastrpc.c:1926 fastrpc_req_mmap() error: double free of 'buf'

In fastrpc_req_mmap() error path, the fastrpc buffer is freed in fastrpc_req_munmap_impl() if unmap is successful.

But in the end, there is an unconditional call to fastrpc_buf_free(). So the above case triggers the double free of fastrpc buf.

Quoted 6 lines

Fix this by avoiding the call to the second fastrpc_buf_free() if fastrpc_req_munmap_impl() is successful.
'err' is not updated as it needs to retain the err returned by qcom_scm_assign_mem(), which is the starting point of this error path.

This is based on static analysis only. Compilation tested.

Please don't put this in the commit message. We want everyone reading the git log to believe everything is 100% rock solid. :P

We need a Fixes tag.

Fixes 72fa6f7820c4 ("misc: fastrpc: Rework fastrpc_req_munmap")

Let's add Abel to the CC list.

Quoted 4 lines

---

^^^
Put testing caveats here instead, where it will be removed from the git log.

Quoted 8 lines
 drivers/misc/fastrpc.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/misc/fastrpc.c +0 -0
index f48466960f1b..1c3ab78f274f 100644

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp) · line 1929

Quoted 10 lines
drivers/misc/fastrpc.c +4 -1
@@ -1921,7 +1921,10 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)
 	return 0;
 
 err_assign:
-	fastrpc_req_munmap_impl(fl, buf);
+	if (!fastrpc_req_munmap_impl(fl, buf)) {
+		/* buf is freed */
+		return err;
+	}
 err_invoke:
 	fastrpc_buf_free(buf);

This bug is real but the fix is not complete.

drivers/misc/fastrpc.c
1911 if (err) {
1912 dev_err(fl->sctx->dev, "Failed to assign memory phys 0x%llx size 0x%llx err %d",
1913 buf->phys, buf->size, err);
1914 goto err_assign;
1915 }
1916 }
1917
1918 spin_lock(&fl->lock);

  1919          list_add_tail(&buf->node, &fl->mmaps);
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

buf needs to be removed from the list before we free it, otherwise it leads to a use after free. The fastrpc_req_munmap_impl() function does that but here this function just calls fastrpc_buf_free().

  1920          spin_unlock(&fl->lock);
  1921  
  1922          if (copy_to_user((void __user *)argp, &req, sizeof(req))) {
  1923                  err = -EFAULT;
  1924                  goto err_assign;
  1925          }
  1926  
  1927          dev_dbg(dev, "mmap\t\tpt 0x%09lx OK [len 0x%08llx]\n",
  1928                  buf->raddr, buf->size);
  1929  
  1930          return 0;
  1931  
  1932  err_assign:
  1933          fastrpc_req_munmap_impl(fl, buf);
  1934  err_invoke:
  1935          fastrpc_buf_free(buf);
  1936  
  1937          return err;
  1938  }

regards,
dan carpenter

#2 lore raw
Quoted 17 lines · Dan Carpenter
Quoted 9 lines · Sukrut Bellary

smatch warning:
drivers/misc/fastrpc.c:1926 fastrpc_req_mmap() error: double free of 'buf'

In fastrpc_req_mmap() error path, the fastrpc buffer is freed in fastrpc_req_munmap_impl() if unmap is successful.

But in the end, there is an unconditional call to fastrpc_buf_free(). So the above case triggers the double free of fastrpc buf.

Quoted 6 lines

Fix this by avoiding the call to the second fastrpc_buf_free() if fastrpc_req_munmap_impl() is successful.
'err' is not updated as it needs to retain the err returned by qcom_scm_assign_mem(), which is the starting point of this error path.

This is based on static analysis only. Compilation tested.

Quoted 8 lines

Please don't put this in the commit message. We want everyone reading the git log to believe everything is 100% rock solid. :P

We need a Fixes tag.

Fixes 72fa6f7820c4 ("misc: fastrpc: Rework fastrpc_req_munmap")

Let's add Abel to the CC list.

Thank you for reviewing the patch.
I will add a Fixes tag and fix the commit message.

Quoted 8 lines
Quoted 4 lines

---

^^^
Put testing caveats here instead, where it will be removed from the git log.

Shall I add "This is based on static analysis only. Compilation tested" here
or
is it not required to mention this for all the fixes?
Can you please recommend what's is the preferred method I need to follow?

Quoted 47 lines
Quoted 8 lines
 drivers/misc/fastrpc.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
drivers/misc/fastrpc.c +0 -0
index f48466960f1b..1c3ab78f274f 100644

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp) · line 1929

Quoted 10 lines
drivers/misc/fastrpc.c +4 -1
@@ -1921,7 +1921,10 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)
 	return 0;
 
 err_assign:
-	fastrpc_req_munmap_impl(fl, buf);
+	if (!fastrpc_req_munmap_impl(fl, buf)) {
+		/* buf is freed */
+		return err;
+	}
 err_invoke:
 	fastrpc_buf_free(buf);

This bug is real but the fix is not complete.

drivers/misc/fastrpc.c
1911 if (err) {
1912 dev_err(fl->sctx->dev, "Failed to assign memory phys 0x%llx size 0x%llx err %d",
1913 buf->phys, buf->size, err);
1914 goto err_assign;
1915 }
1916 }
1917
1918 spin_lock(&fl->lock);

  1919          list_add_tail(&buf->node, &fl->mmaps);
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

buf needs to be removed from the list before we free it, otherwise it leads to a use after free. The fastrpc_req_munmap_impl() function does that but here this function just calls fastrpc_buf_free().

  1920          spin_unlock(&fl->lock);
  1921  
  1922          if (copy_to_user((void __user *)argp, &req, sizeof(req))) {
  1923                  err = -EFAULT;
  1924                  goto err_assign;
  1925          }
  1926  
  1927          dev_dbg(dev, "mmap\t\tpt 0x%09lx OK [len 0x%08llx]\n",
  1928                  buf->raddr, buf->size);
  1929  
  1930          return 0;
Quoted 9 lines
  1931  
  1932  err_assign:
  1933          fastrpc_req_munmap_impl(fl, buf);
  1934  err_invoke:
  1935          fastrpc_buf_free(buf);
  1936  
  1937          return err;
  1938  }

Nice catch!
I will address this in the next version.

Regards,
Sukrut Bellary

Quoted 2 lines

regards,
dan carpenter

#3 lore raw
Quoted 9 lines · Sukrut Bellary
Quoted 8 lines
Quoted 4 lines

---

^^^
Put testing caveats here instead, where it will be removed from the git log.

Quoted 5 lines

Shall I add "This is based on static analysis only. Compilation tested" here
or
is it not required to mention this for all the fixes?
Can you please recommend what's is the preferred method I need to follow?

You can't go wrong with always adding it. The untested part is useful to know. Also it's good to know how a bug is found so we can say if it's affecting user and so we can improve our processes going forward.

regards,
dan carpenter

#4 lore raw
Quoted 16 lines · Dan Carpenter
Quoted 9 lines · Sukrut Bellary
Quoted 8 lines
Quoted 4 lines

---

^^^
Put testing caveats here instead, where it will be removed from the git log.

Quoted 5 lines

Shall I add "This is based on static analysis only. Compilation tested" here
or
is it not required to mention this for all the fixes?
Can you please recommend what's is the preferred method I need to follow?

Quoted 4 lines

You can't go wrong with always adding it. The untested part is useful to know. Also it's good to know how a bug is found so we can say if it's affecting user and so we can improve our processes going forward.

Ok, that's clear. Thanks for the input.

Regards,
Sukrut Bellary

Quoted 3 lines

regards,
dan carpenter

#5 lore raw
In reply to Dan Carpenter
Quoted 43 lines · Dan Carpenter
Quoted 9 lines · Sukrut Bellary

smatch warning:
drivers/misc/fastrpc.c:1926 fastrpc_req_mmap() error: double free of 'buf'

In fastrpc_req_mmap() error path, the fastrpc buffer is freed in fastrpc_req_munmap_impl() if unmap is successful.

But in the end, there is an unconditional call to fastrpc_buf_free(). So the above case triggers the double free of fastrpc buf.

Quoted 6 lines

Fix this by avoiding the call to the second fastrpc_buf_free() if fastrpc_req_munmap_impl() is successful.
'err' is not updated as it needs to retain the err returned by qcom_scm_assign_mem(), which is the starting point of this error path.

This is based on static analysis only. Compilation tested.

Please don't put this in the commit message. We want everyone reading the git log to believe everything is 100% rock solid. :P

We need a Fixes tag.

Fixes 72fa6f7820c4 ("misc: fastrpc: Rework fastrpc_req_munmap")

Let's add Abel to the CC list.

Quoted 4 lines

---

^^^
Put testing caveats here instead, where it will be removed from the git log.

Quoted 10 lines
  drivers/misc/fastrpc.c | 5 ++++-
  1 file changed, 4 insertions(+), 1 deletion(-)
drivers/misc/fastrpc.c +0 -0
index f48466960f1b..1c3ab78f274f 100644
@@ -1921,7 +1921,10 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)
  	return 0;
 
Quoted 8 lines

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp) · line 1929

Quoted 8 lines
drivers/misc/fastrpc.c +4 -1
  err_assign:
-	fastrpc_req_munmap_impl(fl, buf);
+	if (!fastrpc_req_munmap_impl(fl, buf)) {
+		/* buf is freed */
+		return err;
+	}
  err_invoke:
  	fastrpc_buf_free(buf);

how about doing something like this:

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)
                                      &args[0]);
        if (err) {
                dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
        }
 
        /* update the buffer to be able to deallocate the memory on the 

DSP */
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)

        return 0;

err_assign:
- fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
- fastrpc_buf_free(buf);
-
- return err;
+ return fastrpc_req_munmap_impl(fl, buf);
}
----------------------->cut<---------------------------

Quoted 3 lines

This bug is real but the fix is not complete.

Yes, there is a danger of freeing the buf while its added to the list.

Above change should address that, in err cases fd close should take care of deleting list and freeing buf.

--srini

Quoted 37 lines

drivers/misc/fastrpc.c

   1911                  if (err) {
   1912                          dev_err(fl->sctx->dev, "Failed to assign memory phys 0x%llx size 0x%llx err %d",
   1913                                          buf->phys, buf->size, err);
   1914                          goto err_assign;
   1915                  }
   1916          }
   1917
   1918          spin_lock(&fl->lock);
   1919          list_add_tail(&buf->node, &fl->mmaps);
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

buf needs to be removed from the list before we free it, otherwise it leads to a use after free. The fastrpc_req_munmap_impl() function does that but here this function just calls fastrpc_buf_free().

   1920          spin_unlock(&fl->lock);
   1921
   1922          if (copy_to_user((void __user *)argp, &req, sizeof(req))) {
   1923                  err = -EFAULT;
   1924                  goto err_assign;
   1925          }
   1926
   1927          dev_dbg(dev, "mmap\t\tpt 0x%09lx OK [len 0x%08llx]\n",
   1928                  buf->raddr, buf->size);
   1929
   1930          return 0;
   1931
   1932  err_assign:
   1933          fastrpc_req_munmap_impl(fl, buf);
   1934  err_invoke:
   1935          fastrpc_buf_free(buf);
   1936
   1937          return err;
   1938  }

regards,
dan carpenter

#6 lore raw
Quoted 18 lines

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                      &args[0]);
        if (err) {
                dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
        }
 
        /* update the buffer to be able to deallocate the memory on the DSP

*/

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, · line 1935

Quoted 10 lines
drivers/misc/fastrpc.c +1 -5
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
        return 0;
 
 err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       return fastrpc_req_munmap_impl(fl, buf);

This will return success if copy_to_user() fails.

regards,
dan carpenter

#7 lore raw
Quoted 21 lines · Dan Carpenter
Quoted 21 lines

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                       &args[0]);
         if (err) {
                 dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
         }
 
         /* update the buffer to be able to deallocate the memory on the DSP

*/

drivers/misc/fastrpc.c +0 -0
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
         return 0;
 
Quoted 10 lines

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, · line 1935

Quoted 7 lines
drivers/misc/fastrpc.c +1 -5
  err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       return fastrpc_req_munmap_impl(fl, buf);

This will return success if copy_to_user() fails.

that is true, using return value of fastrpc_req_munmap_impl does not really make sense here we should just return err in either case to the user.

--srini

Quoted 3 lines

regards,
dan carpenter

#8 lore raw
Quoted 34 lines · Srinivas Kandagatla
Quoted 21 lines · Dan Carpenter
Quoted 21 lines

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                       &args[0]);
         if (err) {
                 dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
         }
 
         /* update the buffer to be able to deallocate the memory on the DSP

*/

drivers/misc/fastrpc.c +0 -0
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
         return 0;
 
Quoted 10 lines

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, · line 1935

Quoted 7 lines
drivers/misc/fastrpc.c +1 -5
  err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       return fastrpc_req_munmap_impl(fl, buf);

This will return success if copy_to_user() fails.

Quoted 3 lines

that is true, using return value of fastrpc_req_munmap_impl does not really make sense here we should just return err in either case to the user.

Thanks, Srini and Dan, for reviewing the patch and suggestions. I will add this in v2.

Regards,
Sukrut Bellary

Quoted 5 lines

--srini

Quoted 3 lines

regards,
dan carpenter

#9 lore raw
Quoted 35 lines · Sukrut Bellary
Quoted 34 lines · Srinivas Kandagatla
Quoted 31 lines · Dan Carpenter
Quoted 18 lines

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                       &args[0]);
         if (err) {
                 dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
         }
 
         /* update the buffer to be able to deallocate the memory on the DSP

*/

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, · line 1935

Quoted 10 lines
drivers/misc/fastrpc.c +1 -5
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
         return 0;
 
  err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       return fastrpc_req_munmap_impl(fl, buf);

This will return success if copy_to_user() fails.

Quoted 4 lines
Quoted 3 lines

that is true, using return value of fastrpc_req_munmap_impl does not really make sense here we should just return err in either case to the user.

I have one follow-up question before I send the v2 patch. With the following approach, I do see one issue.

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +0 -0
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,

char __user *argp)

                                      &args[0]);
        if (err) {
                dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
        }

        /* update the buffer to be able to deallocate the memory on the DSP */
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
char __user *argp)
        return 0;

err_assign:
- fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
- fastrpc_buf_free(buf);
-

-       return err;
+       fastrpc_req_munmap_impl(fl, buf);
+       return err;

}
----------------------->cut<---------------------------

In this, if qcom_scm_assign_mem() fails, the buf is not added to the list. But the call to fastrpc_req_munmap_impl() tries to delete the buf from the list.

To avoid this, we can use the following approach.
Here we first add the buf to the list and fastrpc_req_munmap_impl() would take care of deleting it in the failure path
Please let me know your review comments.

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +0 -0
index f48466960f1b..56751609f412 100644
@@ -1882,7 +1882,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,

char __user *argp)

                                      &args[0]);
        if (err) {
                dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
        }
 
        /* update the buffer to be able to deallocate the memory on the DSP */

@@ -1890,6 +1891,9 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)

        /* let the client know the address to use */
        req.vaddrout = rsp_msg.vaddr;
+       spin_lock(&fl->lock);
+       list_add_tail(&buf->node, &fl->mmaps);
+       spin_unlock(&fl->lock);
        /* Add memory to static PD pool, protection thru hypervisor */
        if (req.flags != ADSP_MMAP_REMOTE_HEAP_ADDR && fl->cctx->vmcount) {
@@ -1906,9 +1910,6 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
char __user *argp)
                }
        }
-       spin_lock(&fl->lock);
-       list_add_tail(&buf->node, &fl->mmaps);
-       spin_unlock(&fl->lock);
        if (copy_to_user((void __user *)argp, &req, sizeof(req))) {
                err = -EFAULT;

@@ -1922,10 +1923,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl, char __user *argp)

err_assign:

        fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
        return err;

+
}
----------------------->cut<---------------------------

Regards,
Sukrut Bellary

Quoted 11 lines

Thanks, Srini and Dan, for reviewing the patch and suggestions. I will add this in v2.

Regards,
Sukrut Bellary

Quoted 5 lines

--srini

Quoted 3 lines

regards,
dan carpenter

#10 lore raw
Quoted 75 lines · Sukrut Bellary
Quoted 35 lines · Sukrut Bellary
Quoted 34 lines · Srinivas Kandagatla
Quoted 31 lines · Dan Carpenter
Quoted 18 lines

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                       &args[0]);
         if (err) {
                 dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
         }
 
         /* update the buffer to be able to deallocate the memory on the DSP

*/

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, · line 1935

Quoted 10 lines
drivers/misc/fastrpc.c +1 -5
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
         return 0;
 
  err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       return fastrpc_req_munmap_impl(fl, buf);

This will return success if copy_to_user() fails.

Quoted 4 lines
Quoted 3 lines

that is true, using return value of fastrpc_req_munmap_impl does not really make sense here we should just return err in either case to the user.

I have one follow-up question before I send the v2 patch. With the following approach, I do see one issue.

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                      &args[0]);
        if (err) {
                dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
        }
 
        /* update the buffer to be able to deallocate the memory on the DSP */
drivers/misc/fastrpc.c +2 -6
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
        return 0;
 
 err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       fastrpc_req_munmap_impl(fl, buf);
+       return err;
 }
----------------------->cut<---------------------------
Quoted 5 lines

In this, if qcom_scm_assign_mem() fails, the buf is not added to the list. But the call to fastrpc_req_munmap_impl() tries to delete the buf from the list.

To avoid this, we can use the following approach.

The list is initialized in __fastrpc_buf_alloc().

	INIT_LIST_HEAD(&buf->node);

So calling list_del(&buf->node); is fine. It's not necessary but it's not harmful.

regards,
dan carpenter

#11 lore raw
Quoted 80 lines · Dan Carpenter
Quoted 79 lines · Sukrut Bellary
Quoted 35 lines · Sukrut Bellary
Quoted 34 lines · Srinivas Kandagatla
Quoted 31 lines · Dan Carpenter
Quoted 18 lines

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                       &args[0]);
         if (err) {
                 dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
         }
 
         /* update the buffer to be able to deallocate the memory on the DSP

*/

drivers/misc/fastrpc.c static int fastrpc_req_mmap(struct fastrpc_user *fl, · line 1935

Quoted 10 lines
drivers/misc/fastrpc.c +1 -5
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
         return 0;
 
  err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       return fastrpc_req_munmap_impl(fl, buf);

This will return success if copy_to_user() fails.

Quoted 4 lines
Quoted 3 lines

that is true, using return value of fastrpc_req_munmap_impl does not really make sense here we should just return err in either case to the user.

I have one follow-up question before I send the v2 patch. With the following approach, I do see one issue.

----------------------->cut<---------------------------

drivers/misc/fastrpc.c +2 -1
index f60bbf99485c..3fdd326e1ae8 100644
@@ -1891,7 +1891,8 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
                                      &args[0]);
        if (err) {
                dev_err(dev, "mmap error (len 0x%08llx)\n", buf->size);
-               goto err_invoke;
+               fastrpc_buf_free(buf);
+               return err;
        }
 
        /* update the buffer to be able to deallocate the memory on the DSP */
drivers/misc/fastrpc.c +2 -6
@@ -1930,11 +1931,7 @@ static int fastrpc_req_mmap(struct fastrpc_user *fl,
 char __user *argp)
        return 0;
 
 err_assign:
-       fastrpc_req_munmap_impl(fl, buf);
-err_invoke:
-       fastrpc_buf_free(buf);
-
-       return err;
+       fastrpc_req_munmap_impl(fl, buf);
+       return err;
 }
----------------------->cut<---------------------------

In this, if qcom_scm_assign_mem() fails, the buf is not added to the list. But the call to fastrpc_req_munmap_impl() tries to delete the buf from the list.

Quoted 9 lines
Quoted 1 line

To avoid this, we can use the following approach.

The list is initialized in __fastrpc_buf_alloc().

	INIT_LIST_HEAD(&buf->node);

So calling list_del(&buf->node); is fine. It's not necessary but it's not harmful.

OK, Thanks Dan. I will submit the v2 with 1st approach.

Regards,
Sukrut Bellary

Quoted 3 lines

regards,
dan carpenter

#12 lore raw
In reply to Dan Carpenter

This is unrelated but I was looking through the driver and I notice a bunch of code doing:

grep 'return ret ?' drivers/firmware/ -R

	return ret ? : res.result[0];

"ret" here is a kernel error code, and res.result[0] is a firmware error code. Mixing error codes is a dangerous thing. I was reviewing some of the callers and the firmware error code gets passed quite far back into the kernel to where we would only expect kernel error codes.

Presumably the firmware is returning positive error codes? To be honest, I am just guessing. It's better to convert custom error codes to kernel error codes as soon as possible. I am just guessing. Sukrut, do you think you could take a look? If the callers do not differentiate between negative kernel error codes and positive custom error codes then probably just do:

	if (res.result[0])
		ret = -EIO; // -EINVAL?
	return ret;

Also there are a couple places which do:

	return ret ? false : !!res.result[0];

Here true means success and false means failure. So the !! converts a firmware error code to true when it should be false so that's a bug. Quadruple negatives are confusing... It should be:

	if (ret || res.result[0])
		return false;
	return true;

regards,
dan carpenter

#13 lore raw
Quoted 19 lines · Dan Carpenter

This is unrelated but I was looking through the driver and I notice a bunch of code doing:

grep 'return ret ?' drivers/firmware/ -R

	return ret ? : res.result[0];

"ret" here is a kernel error code, and res.result[0] is a firmware error code. Mixing error codes is a dangerous thing. I was reviewing some of the callers and the firmware error code gets passed quite far back into the kernel to where we would only expect kernel error codes.

Presumably the firmware is returning positive error codes? To be honest, I am just guessing. It's better to convert custom error codes to kernel error codes as soon as possible. I am just guessing. Sukrut, do you think you could take a look? If the callers do not differentiate between negative kernel error codes and positive custom error codes then probably just do:

Quoted 4 lines
	if (res.result[0])
		ret = -EIO; // -EINVAL?
	return ret;

Thanks, Dan, for sharing your findings.
Yes, sure, I will take a look.

Regards,
Sukrut Bellary

Quoted 15 lines

Also there are a couple places which do:

	return ret ? false : !!res.result[0];

Here true means success and false means failure. So the !! converts a firmware error code to true when it should be false so that's a bug. Quadruple negatives are confusing... It should be:

	if (ret || res.result[0])
		return false;
	return true;

regards,
dan carpenter