Skip to contentSkip to conversation

fs/kcore: change copy_to_user to copy_in_user

RFC Superseded yalin wang 3 participants 4 messages Aug 20, 2015 – Aug 21, 2015

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

yalin wang

author
#0 lore raw

RFC 1/1 fs/kcore: change copy_to_user to copy_in_user

the copy_to_user() here expect can fix the fault on both kernel and user address, this is not true on other platforms except x86, change to user copy_in_user() so that can detect the page fault, work as expected.

---
 fs/proc/kcore.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
fs/proc/kcore.c +6 -2
index 92e6726..4f28deb 100644
@@ -515,8 +515,12 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
 		} else {
 			if (kern_addr_valid(start)) {
 				unsigned long n;
-
-				n = copy_to_user(buffer, (char *)start, tsz);
+				if ((start + tsz < tsz) ||
+						(start + tsz) > TASK_SIZE)
+					return -EFAULT;
+				set_fs(KERNEL_DS);
+				n = copy_in_user(buffer, (char *)start, tsz);
+				set_fs(USER_DS);
 				/*
 				 * We cannot distinguish between fault on source
 				 * and fault on destination. When this happens
Signature · 1 line
1.9.1
#1 lore raw
Quoted 4 lines · yalin wang

the copy_to_user() here expect can fix the fault on both kernel and user address, this is not true on other platforms except x86, change to user copy_in_user() so that can detect the page fault, work as expected.

Could you rephrase this into multiple sentences in comprehensible English? What is the expected behavior, what is the unexpected behavior and what can people do to trigger it?

Quoted 32 lines
---
 fs/proc/kcore.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)
fs/proc/kcore.c +6 -3
index 92e6726..4f28deb 100644
@@ -515,8 +515,12 @@ read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos)
                } else {
                        if (kern_addr_valid(start)) {
                                unsigned long n;
-
-                               n = copy_to_user(buffer, (char *)start, tsz);
+                               if ((start + tsz < tsz) ||
+                                               (start + tsz) > TASK_SIZE)
+                                       return -EFAULT;
+                               set_fs(KERNEL_DS);
+                               n = copy_in_user(buffer, (char *)start, tsz);
+                               set_fs(USER_DS);
                                /*
                                 * We cannot distinguish between fault on source
                                 * and fault on destination. When this happens
--

1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

author
#2 lore raw
Quoted 13 lines · Frans Klaver
Quoted 4 lines · yalin wang

the copy_to_user() here expect can fix the fault on both kernel and user address, this is not true on other platforms except x86, change to user copy_in_user() so that can detect the page fault, work as expected.

Could you rephrase this into multiple sentences in comprehensible English? What is the expected behavior, what is the unexpected behavior and what can people do to trigger it?

ok, i will send a V2 patch.

#3 lore raw
In reply to yalin wang

fs/proc/kcore.c read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos) · line 520

Quoted 5 lines · yalin wang
fs/proc/kcore.c +3 -2
-
-                               n = copy_to_user(buffer, (char *)start, tsz);
+                               if ((start + tsz < tsz) ||
+                                               (start + tsz) > TASK_SIZE)
+                                       return -EFAULT;

This is wrong. You apparently want to have

        if (!access_ok(start, tsz))
                return -EFAULT;

fs/proc/kcore.c read_kcore(struct file *file, char __user *buffer, size_t buflen, loff_t *fpos) · line 523

Quoted 3 lines
fs/proc/kcore.c +3 -0
+                               set_fs(KERNEL_DS);
+                               n = copy_in_user(buffer, (char *)start, tsz);
+                               set_fs(USER_DS);

.. and this is actually worse and even less portable than what we have now, in that it's actively wrong on platforms that may have a user address and a kernel address with the same value (ie they have explicitly separate kernel/user address spaces).

Now, that's admittedly unusual, but I think sparc32 actually can do that.

Anyway, I absolutely detest this patch. It replaces one piece of code that admittedly doesn't work on all architectures because kernel memory is accessed without testing, with another hack that happens to work on other architectures and is fragile and prone to be a security issue.

In other words, I think the end result is _worse_ than the current situation.

You probably want to use "probe_kernel_read()" and do it into a temporary buffer, and then just do the copy_to_user() from the temporary buffer. Sure, it's less efficient, but at least it's not actively wrong and a possible security problem in the long run.

Linus