Skip to contentSkip to conversation

x86/sgx: Fix add page bugs

PATCH Superseded Sean Christopherson 2 participants 8 messages Oct 10, 2019 – Oct 17, 2019

Subsystems (2)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover x86/sgx: Fix add page bugs

Three bug fixes in the add page flow, along with a related enhancement. Note, patch 4/4 is effectively a revert.

Sean Christopherson (4):
x86/sgx: Pass EADD the kernel's virtual address for the source page x86/sgx: Check the validity of the source page address for EADD x86/sgx: Fix EEXTEND error handling
x86/sgx: Drop mmap_sem before EEXTENDing an enclave page

 arch/x86/kernel/cpu/sgx/ioctl.c | 46 ++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 21 deletions(-)
Signature · 1 line
2.22.0
#1 lore raw

PATCH 1/4 x86/sgx: Pass EADD the kernel's virtual address for the source page

Use the kernel's virtual address to reference the source page when EADDing a page to the enclave. gup() "does not guarantee that the page exists in the user mappings", i.e. EADD can still fault and deadlock due to mmap_sem contention if it consumes the userspace address.

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +2 -4
index eeed919ed3e2..62965d003fda 100644
@@ -328,16 +328,14 @@ static int __sgx_encl_add_page(struct sgx_encl *encl,
 	if (ret < 1)
 		return ret;
 
-	__uaccess_begin();
-
 	pginfo.secs = (unsigned long)sgx_epc_addr(encl->secs.epc_page);
 	pginfo.addr = SGX_ENCL_PAGE_ADDR(encl_page);
 	pginfo.metadata = (unsigned long)secinfo;
-	pginfo.contents = (unsigned long)src;
+	pginfo.contents = (unsigned long)kmap_atomic(src_page);
 
 	ret = __eadd(&pginfo, sgx_epc_addr(epc_page));
 
-	__uaccess_end();
+	kunmap_atomic((void *)pginfo.contents);
 	put_page(src_page);
 
 	return ret ? -EFAULT : 0;
Signature · 1 line
2.22.0
#2 lore raw

PATCH 2/4 x86/sgx: Check the validity of the source page address for EADD

In reply to Sean Christopherson

Add an explicit access_ok() check on EADD's source page to avoid passing garbage to gup().

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 3 +++
 1 file changed, 3 insertions(+)
arch/x86/kernel/cpu/sgx/ioctl.c +3 -0
index 62965d003fda..61a2a7b6677e 100644
@@ -492,6 +492,9 @@ static long sgx_ioc_enclave_add_page(struct sgx_encl *encl, void __user *arg)
 	    !IS_ALIGNED(addp.src, PAGE_SIZE))
 		return -EINVAL;
 
+	if (!(access_ok(addp.src, PAGE_SIZE)))
+		return -EFAULT;
+
 	if (addp.addr < encl->base || addp.addr - encl->base >= encl->size)
 		return -EINVAL;
 
Signature · 1 line
2.22.0
#3 lore raw

PATCH 3/4 x86/sgx: Fix EEXTEND error handling

In reply to Sean Christopherson

Rework EEXTEND error handling to fix issues related to destroying the enclave in response to EEXTEND failure. At the time of EEXTEND, the page is already visibile in the sense that it has been added to the radix tree, and therefore will be processed by sgx_encl_destroy(). This means the "add" part needs to be fully completed prior to invoking sgx_encl_destroy() in order to avoid consuming half-baked state.

Move sgx_encl_destroy() to the call site of __sgx_encl_extend() so that it is somewhat more obvious why the add needs to complete before doing EEXTEND.

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 31 ++++++++++++++++---------------
 1 file changed, 16 insertions(+), 15 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +16 -15
index 61a2a7b6677e..fd4117f18564 100644
@@ -351,18 +351,14 @@ static int __sgx_encl_extend(struct sgx_encl *encl,
 	for_each_set_bit(i, &mrmask, 16) {
 		ret = __eextend(sgx_epc_addr(encl->secs.epc_page),
 				sgx_epc_addr(epc_page) + (i * 0x100));
-		if (ret)
-			goto err_out;
+		if (ret) {
+			if (encls_failed(ret))
+				ENCLS_WARN(ret, "EEXTEND");
+			return -EFAULT;
+		}
 	}
 
 	return 0;
-
-err_out:
-	if (encls_failed(ret))
-		ENCLS_WARN(ret, "EEXTEND");
-
-	sgx_encl_destroy(encl);
-	return -EFAULT;
 }
 
 static int sgx_encl_add_page(struct sgx_encl *encl,
@@ -415,19 +411,24 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 	if (ret)
 		goto err_out;
 
-	ret = __sgx_encl_extend(encl, epc_page, addp->mrmask);
-	if (ret)
-		goto err_out;
-
+	/*
+	 * Complete the "add" before doing the "extend" so that the "add"
+	 * isn't in a half-baked state in the extremely unlikely scenario the
+	 * the enclave will be destroyed in response to EEXTEND failure.
+	 */
 	encl_page->encl = encl;
 	encl_page->epc_page = epc_page;
 	encl->secs_child_cnt++;
 
-	sgx_mark_page_reclaimable(encl_page->epc_page);
+	ret = __sgx_encl_extend(encl, epc_page, addp->mrmask);
+	if (ret)
+		sgx_encl_destroy(encl);
+	else
+		sgx_mark_page_reclaimable(encl_page->epc_page);
 
 	mutex_unlock(&encl->lock);
 	up_read(&current->mm->mmap_sem);
-	return 0;
+	return ret;
 
 err_out:
 	radix_tree_delete(&encl_page->encl->page_tree,
Signature · 1 line
2.22.0
#4 lore raw

PATCH 4/4 x86/sgx: Drop mmap_sem before EEXTENDing an enclave page

In reply to Sean Christopherson

Drop mmap_sem, which needs to be held for read across EADD, prior to doing EEXTEND on the newly added page to avoid holding mmap_sem for an extended duration. EEXTEND doesn't access user pages and holding encl->lock without mmap_sem is perfectly ok, while EEXTEND is a _slow_ operation, to the point where it operates on 256-byte chunks instead of 4k pages to maintain a reasonable latency for a single instruction.

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +5 -3
index fd4117f18564..46f2769d16fe 100644
@@ -403,11 +403,15 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 	 */
 	ret = radix_tree_insert(&encl->page_tree, PFN_DOWN(encl_page->desc),
 				encl_page);
-	if (ret)
+	if (ret) {
+		up_read(&current->mm->mmap_sem);
 		goto err_out_unlock;
+	}
 
 	ret = __sgx_encl_add_page(encl, encl_page, epc_page, secinfo,
 				  addp->src);
+	up_read(&current->mm->mmap_sem);
+
 	if (ret)
 		goto err_out;
 
@@ -427,7 +431,6 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 		sgx_mark_page_reclaimable(encl_page->epc_page);
 
 	mutex_unlock(&encl->lock);
-	up_read(&current->mm->mmap_sem);
 	return ret;
 
 err_out:
@@ -437,7 +440,6 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 err_out_unlock:
 	sgx_encl_shrink(encl, va_page);
 	mutex_unlock(&encl->lock);
-	up_read(&current->mm->mmap_sem);
 
 err_out_free:
 	sgx_free_page(epc_page);
Signature · 1 line
2.22.0
#5 lore raw
In reply to Sean Christopherson
Quoted 11 lines · Sean Christopherson

Three bug fixes in the add page flow, along with a related enhancement. Note, patch 4/4 is effectively a revert.

Sean Christopherson (4):
x86/sgx: Pass EADD the kernel's virtual address for the source page x86/sgx: Check the validity of the source page address for EADD x86/sgx: Fix EEXTEND error handling
x86/sgx: Drop mmap_sem before EEXTENDing an enclave page

 arch/x86/kernel/cpu/sgx/ioctl.c | 46 ++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 21 deletions(-)

So, which one should be merged first. This or v2 of the other series.

Can you send a single "for-v23 fixes" series with everything needed just so that there is no room guess work? You can call v3 of the series that you first started.

You already know that I applied 1/9 and 2/9 so please don't include them.

/Jarkko

#6 lore raw
Quoted 10 lines · Jarkko Sakkinen
Quoted 9 lines · Sean Christopherson

Three bug fixes in the add page flow, along with a related enhancement. Note, patch 4/4 is effectively a revert.

Sean Christopherson (4):
x86/sgx: Pass EADD the kernel's virtual address for the source page x86/sgx: Check the validity of the source page address for EADD x86/sgx: Fix EEXTEND error handling
x86/sgx: Drop mmap_sem before EEXTENDing an enclave page

Quoted 4 lines
Quoted 2 lines
 arch/x86/kernel/cpu/sgx/ioctl.c | 46 ++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 21 deletions(-)

So, which one should be merged first. This or v2 of the other series.

Shouldn't matter.

Quoted 3 lines

Can you send a single "for-v23 fixes" series with everything needed just so that there is no room guess work? You can call v3 of the series that you first started.

Will do, I'll get it sent out today.

#7 lore raw
Quoted 16 lines · Sean Christopherson
Quoted 10 lines · Jarkko Sakkinen
Quoted 9 lines · Sean Christopherson

Three bug fixes in the add page flow, along with a related enhancement. Note, patch 4/4 is effectively a revert.

Sean Christopherson (4):
x86/sgx: Pass EADD the kernel's virtual address for the source page x86/sgx: Check the validity of the source page address for EADD x86/sgx: Fix EEXTEND error handling
x86/sgx: Drop mmap_sem before EEXTENDing an enclave page

Quoted 4 lines
Quoted 2 lines
 arch/x86/kernel/cpu/sgx/ioctl.c | 46 ++++++++++++++++++---------------
 1 file changed, 25 insertions(+), 21 deletions(-)

So, which one should be merged first. This or v2 of the other series.

Quoted 7 lines

Shouldn't matter.

Quoted 3 lines

Can you send a single "for-v23 fixes" series with everything needed just so that there is no room guess work? You can call v3 of the series that you first started.

Will do, I'll get it sent out today.

Thank you. I'll start to go through and merging (have the patch set now).

Rather than send any new versions lets just discuss if there is something that looks weird and I'll either pick the patch unmodified or do whatever mods needs depending on the resolution.

/Jarkko