Skip to contentSkip to conversation

do_sendfile ppos check ...

PATCH Herbert Poetzl 2 participants 4 messages Nov 3, 2005 – Nov 4, 2005 VFS

lore
Share & tools
lore
#0 lore raw

Hi Andrew!

friend of mine stumbled over the following issue:

do_sendfile() does an overflow check near the end, like this:

if (*ppos > max)
retval = -EOVERFLOW;

now both sys_sendfile and sys_sendfile64 do call do_sendfile() similar to this:

        if (offset) {
		...
                ret = do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
                return ret;
        }
	return do_sendfile(out_fd, in_fd, NULL, count, 0);

which passes ppos as NULL, which in turn leads to an oops ...

here is a patch (suggestion) to handle this properly, which also adjusts the max for sys_sendfile()
(let me know what you think!)

fs/read_write.c +0 -0
@@ -731,7 +731,8 @@ asmlinkage ssize_t sys_sendfile(int out_
		return ret;
	}

- return do_sendfile(out_fd, in_fd, NULL, count, 0);
+ pos = 0;
+ return do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
}

asmlinkage ssize_t sys_sendfile64(int out_fd, int in_fd, loff_t __user *offset, size_t count)
@@ -748,5 +749,6 @@ asmlinkage ssize_t sys_sendfile64(int ou

		return ret;
	}

- return do_sendfile(out_fd, in_fd, NULL, count, 0);
+ pos = 0;
+ return do_sendfile(out_fd, in_fd, &pos, count, 0);
}

#1 lore raw
Quoted 2 lines · Herbert Poetzl

which passes ppos as NULL, which in turn leads to an oops ...

But do_sendfile will set ppos to &in_file->f_pos if it's NULL. Why isn't that working?

Review asmlinkage ssize_t sys_sendfile(int out_ · line 736

Quoted 8 lines
(quoted diff) +2 -1
@@ -731,7 +731,8 @@ asmlinkage ssize_t sys_sendfile(int out_
               return ret;
       }
 
-       return do_sendfile(out_fd, in_fd, NULL, count, 0);
+       pos = 0;
+       return do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 }

The last argument is meant to be zero if you check the history.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#2 lore raw
Quoted 11 lines · Herbert Xu
Quoted 2 lines · Herbert Poetzl

which passes ppos as NULL, which in turn leads to an oops ...

But do_sendfile will set ppos to &in_file->f_pos if it's NULL. Why isn't that working?

Quoted 4 lines
(quoted diff) +0 -0
@@ -731,7 +731,8 @@ asmlinkage ssize_t sys_sendfile(int out_
               return ret;
       }
 
Quoted 6 lines

Review asmlinkage ssize_t sys_sendfile(int out_ · line 736

Quoted 4 lines
(quoted diff) +2 -1
-       return do_sendfile(out_fd, in_fd, NULL, count, 0);
+       pos = 0;
+       return do_sendfile(out_fd, in_fd, &pos, count, MAX_NON_LFS);
 }

The last argument is meant to be zero if you check the history.

hmm, why a different max than with offset?

currently investigating ... probably a removal of
the 'unnecessary' check (*ppos) would be a better
approach, something like:

fs/read_write.c 2005-11-03 17:28:50.000000000 +0100 renamed from fs/read_write.c 2005-10-28 23:59:02.000000000 +0200 +0 -3
@@ -719,9 +719,6 @@
       	current->syscr++;
       	current->syscw++;
 
-       if (*ppos > max)
-               retval = -EOVERFLOW;
-
 fput_out:
         	ds,   fput_light(out_file, fput_needed_out);
 fput_in:

thanks for the input,
Herbert

Quoted 7 lines

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#3 lore raw
Quoted 8 lines · Herbert Poetzl

currently investigating ... probably a removal of
the 'unnecessary' check (*ppos) would be a better
approach, something like:

fs/read_write.c 2005-11-03 17:28:50.000000000 +0100 line 724

Quoted 6 lines
fs/read_write.c 2005-11-03 17:28:50.000000000 +0100 renamed from fs/read_write.c 2005-10-28 23:59:02.000000000 +0200 +0 -3
@@ -719,9 +719,6 @@
       	current->syscr++;
       	current->syscw++;
 
-       if (*ppos > max)
-               retval = -EOVERFLOW;
-

This still doesn't make sense. If ppos came in as NULL, it should have become non-NULL long before it reaches this part of the function.

Look at the top of the do_sendfile function, it sets ppos if it is NULL.

Cheers,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt