Skip to contentSkip to conversation

x86/sgx: Use GFP_KERNEL for allocations

PATCH In review Jarkko Sakkinen 2 participants 14 messages Nov 4, 2019 – Nov 6, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v24 1/3 x86/sgx: Use GFP_KERNEL for allocations

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +2 -2
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);
 	if (!initp_page)
 		return -ENOMEM;
 
Signature · 1 line
2.20.1
#1 lore raw

PATCH v24 2/3 x86/sgx: Destroy enclave if EADD fails

__sgx_encl_add_page() can only fail in the case of EPCM conflict at least in non-artificial situations. Also, it consistent semantics in rollback is something to pursue for. Thus, destroy enclave when the EADD fails as we do when EEXTEND fails already.

In the cases it is sane to return -EIO. From this the caller can deduce the failure and knows that the enclave was destroyed. The previous -EFAULT could happen in numerous situations.

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +7 -6
index d53aee5a64c1..289af607f634 100644
@@ -338,7 +338,7 @@ static int __sgx_encl_add_page(struct sgx_encl *encl,
 	kunmap_atomic((void *)pginfo.contents);
 	put_page(src_page);
 
-	return ret ? -EFAULT : 0;
+	return ret ? -EIO : 0;
 }
 
 static int __sgx_encl_extend(struct sgx_encl *encl,
@@ -353,7 +353,7 @@ static int __sgx_encl_extend(struct sgx_encl *encl,
 		if (ret) {
 			if (encls_failed(ret))
 				ENCLS_WARN(ret, "EEXTEND");
-			return -EFAULT;
+			return -EIO;
 		}
 	}
 
@@ -413,8 +413,10 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 
 	ret = __sgx_encl_add_page(encl, encl_page, epc_page, secinfo,
 				  addp->src);
-	if (ret)
+	if (ret) {
+		sgx_encl_destroy(encl);
 		goto err_out;
+	}
 
 	/*
 	 * Complete the "add" before doing the "extend" so that the "add"
@@ -498,10 +500,9 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
  *
  * Return:
  *   0 on success,
- *   -EINVAL if any input param or the SECINFO contains invalid data,
  *   -EACCES if an executable source page is located in a noexec partition,
- *   -ENOMEM if any memory allocation, including EPC, fails,
- *   -ERESTARTSYS if a pending signal is recognized
+ *   -EIO if either ENCLS[EADD] or ENCLS[EEXTEND] fails
+ *   -errno otherwise
  */
 static long sgx_ioc_enclave_add_pages(struct sgx_encl *encl, void __user *arg)
 {
Signature · 1 line
2.20.1
#2 lore raw

Re: [PATCH for v24 2/3] x86/sgx: Destroy enclave if EADD fails

Quoted 2 lines · Jarkko Sakkinen

__sgx_encl_add_page() can only fail in the case of EPCM conflict at least in non-artificial situations.

Huh? EADD can fail for a variety of reasons. I can't think of a use case where userspace _won't_ kill the enclave in response to failure, but that doesn't justify killing the enclave, e.g. we don't kill the enclave in any other error path that is just as indicative of a userspace bug.

Quoted 1 line

Also, it consistent semantics in rollback is something to pursue for.

I don't follow this at all. How is it inconsistent to state that errors are handled gracefully unless they're unrecoverable?

Quoted 6 lines

Thus, destroy enclave when the EADD fails as we do when EEXTEND fails already.

In the cases it is sane to return -EIO. From this the caller can deduce the failure and knows that the enclave was destroyed. The previous -EFAULT could happen in numerous situations.

This should be a separate patch.

Quoted 40 lines
---
 arch/x86/kernel/cpu/sgx/ioctl.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +5 -3
index d53aee5a64c1..289af607f634 100644
@@ -338,7 +338,7 @@ static int __sgx_encl_add_page(struct sgx_encl *encl,
 	kunmap_atomic((void *)pginfo.contents);
 	put_page(src_page);
 
-	return ret ? -EFAULT : 0;
+	return ret ? -EIO : 0;
 }
 
 static int __sgx_encl_extend(struct sgx_encl *encl,
@@ -353,7 +353,7 @@ static int __sgx_encl_extend(struct sgx_encl *encl,
 		if (ret) {
 			if (encls_failed(ret))
 				ENCLS_WARN(ret, "EEXTEND");
-			return -EFAULT;
+			return -EIO;
 		}
 	}
 
@@ -413,8 +413,10 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 
 	ret = __sgx_encl_add_page(encl, encl_page, epc_page, secinfo,
 				  addp->src);
-	if (ret)
+	if (ret) {
+		sgx_encl_destroy(encl);
 		goto err_out;
+	}
 
 	/*
 	 * Complete the "add" before doing the "extend" so that the "add"

arch/x86/kernel/cpu/sgx/ioctl.c static int sgx_encl_add_page(struct sgx_encl *encl, · line 504

Quoted 7 lines
arch/x86/kernel/cpu/sgx/ioctl.c +0 -3
@@ -498,10 +500,9 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
  *
  * Return:
  *   0 on success,
- *   -EINVAL if any input param or the SECINFO contains invalid data,
  *   -EACCES if an executable source page is located in a noexec partition,
- *   -ENOMEM if any memory allocation, including EPC, fails,
- *   -ERESTARTSYS if a pending signal is recognized

Why are you removing the documentation for EINVAL, ENOMEM and ERESTARTSYS?

Quoted 8 lines
arch/x86/kernel/cpu/sgx/ioctl.c +2 -0
+ *   -EIO if either ENCLS[EADD] or ENCLS[EEXTEND] fails
+ *   -errno otherwise
  */
 static long sgx_ioc_enclave_add_pages(struct sgx_encl *encl, void __user *arg)
 {
Signature · 1 line
2.20.1
#3 lore raw

Re: [PATCH for v24 2/3] x86/sgx: Destroy enclave if EADD fails

Quoted 8 lines · Sean Christopherson
Quoted 2 lines · Jarkko Sakkinen

__sgx_encl_add_page() can only fail in the case of EPCM conflict at least in non-artificial situations.

Huh? EADD can fail for a variety of reasons. I can't think of a use case where userspace _won't_ kill the enclave in response to failure, but that doesn't justify killing the enclave, e.g. we don't kill the enclave in any other error path that is just as indicative of a userspace bug.

I think it does because it is the only sane metrics to take and it also makes the semantics more sound and coherent.

Quoted 4 lines
Quoted 1 line

Also, it consistent semantics in rollback is something to pursue for.

I don't follow this at all. How is it inconsistent to state that errors are handled gracefully unless they're unrecoverable?

Always when the user space gets -EIO it will know that enclave ceased to exist. That is very consistent.

Quoted 8 lines
Quoted 6 lines

Thus, destroy enclave when the EADD fails as we do when EEXTEND fails already.

In the cases it is sane to return -EIO. From this the caller can deduce the failure and knows that the enclave was destroyed. The previous -EFAULT could happen in numerous situations.

This should be a separate patch.

No it shouldn't because it is so closely connected to the semantics change.

/Jarkko

#4 lore raw

PATCH v24 3/3 x86/sgx: Remove a subordinate clause

In reply to Jarkko Sakkinen

The subordinate clause of last sentence of the sgx_ioc_enclave_pages() does not provide any insight not already provided. Thus, remove it. Also, using "i.e." (and "e.g.") in the documentation should be considered a bad practice because it leaves it open ended.

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +3 -5
index 289af607f634..87b2fb62825a 100644
@@ -468,11 +468,9 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
  * @encl:       pointer to an enclave instance (via ioctl() file pointer)
  * @arg:	a user pointer to a struct sgx_enclave_add_pages instance
  *
- * Add (EADD) one or more pages to an uninitialized enclave, and optionally
- * extend (EEXTEND) the measurement with the contents of the page. The range of
- * pages must be virtually contiguous. The SECINFO and measurement mask are
- * applied to all pages, i.e. pages with different properties must be added in
- * separate calls.
+ * Add one or more pages to an uninitialized enclave, and optionally extend the
+ * measurement with the contents of the page. The address range of pages must
+ * be contiguous. The SECINFO and measurement mask are applied to all pages.
  *
  * A SECINFO for a TCS is required to always contain zero permissions because
  * CPU silently zeros them. Allowing anything else would cause a mismatch in
Signature · 1 line
2.20.1
#5 lore raw

Re: [PATCH for v24 3/3] x86/sgx: Remove a subordinate clause

Quoted 2 lines · Jarkko Sakkinen

The subordinate clause of last sentence of the sgx_ioc_enclave_pages() does not provide any insight not already provided. Thus, remove it.

Isn't the whole point of the documentation to help the user understand *how* to use the API, not simply state exactly what the code does?

Quoted 2 lines

Also, using "i.e." (and "e.g.") in the documentation should be considered a bad practice because it leaves it open ended.

How does stating the practical effect of the semantics leave the docs open ended? The man pages for open[1], close[2], read[3], dlopen[4], etc... all provide examples (usually with "for example" verbiage) to call out common scenarios to help the reader understand the details, and close() also uses "i.e." to clarify.

[1] http://man7.org/linux/man-pages/man2/open.2.html
[2] http://man7.org/linux/man-pages/man2/close.2.html
[3] http://man7.org/linux/man-pages/man2/read.2.html
[4] http://man7.org/linux/man-pages/man3/dlopen.3.html

Quoted 10 lines
---
 arch/x86/kernel/cpu/sgx/ioctl.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +0 -0
index 289af607f634..87b2fb62825a 100644

arch/x86/kernel/cpu/sgx/ioctl.c static int sgx_encl_add_page(struct sgx_encl *encl, · line 472

Quoted 5 lines
arch/x86/kernel/cpu/sgx/ioctl.c +0 -2
@@ -468,11 +468,9 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
  * @encl:       pointer to an enclave instance (via ioctl() file pointer)
  * @arg:	a user pointer to a struct sgx_enclave_add_pages instance
  *
- * Add (EADD) one or more pages to an uninitialized enclave, and optionally
- * extend (EEXTEND) the measurement with the contents of the page. The range of

Why are you dropping the reference to the ENCLS functions? Yes, it's stating the obvious for those of us that have been buried in SGX for the last few years, but for newbies reading the docs I think there is value in explicitly connecting the actions to the ENCLS function.

arch/x86/kernel/cpu/sgx/ioctl.c static int sgx_encl_add_page(struct sgx_encl *encl, · line 473

Quoted 6 lines
arch/x86/kernel/cpu/sgx/ioctl.c +3 -3
- * pages must be virtually contiguous. The SECINFO and measurement mask are
- * applied to all pages, i.e. pages with different properties must be added in
- * separate calls.
+ * Add one or more pages to an uninitialized enclave, and optionally extend the
+ * measurement with the contents of the page. The address range of pages must
+ * be contiguous. The SECINFO and measurement mask are applied to all pages.

I like making it "address range", but I'd prefer to keep the "virtually" qualifier. Again, probably stating the obvious for many readers, but I don't think it's ever harmful and I like the clarification it provides without having to stop and consider the context.

Quoted 6 lines
arch/x86/kernel/cpu/sgx/ioctl.c +0 -0
  *
  * A SECINFO for a TCS is required to always contain zero permissions because
  * CPU silently zeros them. Allowing anything else would cause a mismatch in
Signature · 1 line
2.20.1
#6 lore raw

Re: [PATCH for v24 3/3] x86/sgx: Remove a subordinate clause

Quoted 6 lines · Sean Christopherson
Quoted 2 lines · Jarkko Sakkinen

The subordinate clause of last sentence of the sgx_ioc_enclave_pages() does not provide any insight not already provided. Thus, remove it.

Isn't the whole point of the documentation to help the user understand *how* to use the API, not simply state exactly what the code does?

For kdoc it should explain in clear and concise way

That clause does not provide any help for that. It just repeats with different vocabulary the exact same thing as was already said.

Quoted 8 lines
Quoted 2 lines

Also, using "i.e." (and "e.g.") in the documentation should be considered a bad practice because it leaves it open ended.

How does stating the practical effect of the semantics leave the docs open ended? The man pages for open[1], close[2], read[3], dlopen[4], etc... all provide examples (usually with "for example" verbiage) to call out common scenarios to help the reader understand the details, and close() also uses "i.e." to clarify.

I did not find the use of "i.e." or "e.g." from those. When they introduce an example it is done in a more structured way, not as a subordinate clause.

/Jarkko

#7 lore raw

Re: [PATCH for v24 3/3] x86/sgx: Remove a subordinate clause

Quoted 9 lines · Jarkko Sakkinen
Quoted 6 lines · Sean Christopherson
Quoted 2 lines · Jarkko Sakkinen

The subordinate clause of last sentence of the sgx_ioc_enclave_pages() does not provide any insight not already provided. Thus, remove it.

Isn't the whole point of the documentation to help the user understand *how* to use the API, not simply state exactly what the code does?

For kdoc it should explain in clear and concise way

... what the function does.

/Jarkko

#8 lore raw
In reply to Jarkko Sakkinen
Quoted 23 lines · Jarkko Sakkinen

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 

arch/x86/kernel/cpu/sgx/ioctl.c static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg) · line 677

Quoted 5 lines
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);

Would it make sense to use GFP_KERNEL_ACCOUNT? The accounting would be weird for the case where userspace is using a builder process, but even in that case it's not flat out wrong to account per-enclave memory allocations.

Quoted 6 lines
arch/x86/kernel/cpu/sgx/ioctl.c +0 -0
 	if (!initp_page)
 		return -ENOMEM;
 
Signature · 1 line
2.20.1
#9 lore raw
Quoted 27 lines · Sean Christopherson
Quoted 26 lines · Jarkko Sakkinen

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
Quoted 6 lines

arch/x86/kernel/cpu/sgx/ioctl.c static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg) · line 677

Quoted 2 lines
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);

Would it make sense to use GFP_KERNEL_ACCOUNT? The accounting would be weird for the case where userspace is using a builder process, but even in that case it's not flat out wrong to account per-enclave memory allocations.

I did not find a single call site that would use that for allocating memory for function-internal data.

/Jarkko

#10 lore raw
Quoted 35 lines · Jarkko Sakkinen
Quoted 27 lines · Sean Christopherson
Quoted 26 lines · Jarkko Sakkinen

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
Quoted 6 lines

arch/x86/kernel/cpu/sgx/ioctl.c static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg) · line 677

Quoted 2 lines
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);

Would it make sense to use GFP_KERNEL_ACCOUNT? The accounting would be weird for the case where userspace is using a builder process, but even in that case it's not flat out wrong to account per-enclave memory allocations.

Quoted 2 lines

I did not find a single call site that would use that for allocating memory for function-internal data.

Actually, the fact that the allocations are transient is an even better argument for accounting the memory, as the weirdness I was referring to doesn't exist for the builder concept.

But looking more closely, Documentation/core-api/memory-allocation.rst states:

* Untrusted allocations triggered from userspace should be a subject of kmem accounting and must have ``__GFP_ACCOUNT`` bit set. There is the handy ``GFP_KERNEL_ACCOUNT`` shortcut for ``GFP_KERNEL`` allocations that should be accounted.

That means all uses of GFP_KERNEL except in sgx_alloc_epc_section() should be converted to GFP_KERNEL_ACCOUNTED. As is, depending on fd limits[*], a single process can easily burn through multiple GBs of memory simply by opening /dev/sgx/enclave in a loop.

[*] AFAICT, systemd is upping the max number of open files to 1M on my
systems. I don't _think_ I changed a setting anywhere?

#11 lore raw
Quoted 51 lines · Sean Christopherson
Quoted 35 lines · Jarkko Sakkinen
Quoted 27 lines · Sean Christopherson
Quoted 26 lines · Jarkko Sakkinen

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
Quoted 6 lines

arch/x86/kernel/cpu/sgx/ioctl.c static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg) · line 677

Quoted 2 lines
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);

Would it make sense to use GFP_KERNEL_ACCOUNT? The accounting would be weird for the case where userspace is using a builder process, but even in that case it's not flat out wrong to account per-enclave memory allocations.

Quoted 2 lines

I did not find a single call site that would use that for allocating memory for function-internal data.

Actually, the fact that the allocations are transient is an even better argument for accounting the memory, as the weirdness I was referring to doesn't exist for the builder concept.

But looking more closely, Documentation/core-api/memory-allocation.rst states:

* Untrusted allocations triggered from userspace should be a subject of kmem accounting and must have ``__GFP_ACCOUNT`` bit set. There is the handy ``GFP_KERNEL_ACCOUNT`` shortcut for ``GFP_KERNEL`` allocations that should be accounted.

Quoted 4 lines

That means all uses of GFP_KERNEL except in sgx_alloc_epc_section() should be converted to GFP_KERNEL_ACCOUNTED. As is, depending on fd limits[*], a single process can easily burn through multiple GBs of memory simply by opening /dev/sgx/enclave in a loop.

What does the documentation mean by untrusted allocaton?

__GFP_ACCOUNT kernel and GFP_KERNEL_ACCOUNT are both quite alien flags to me as is kmemcg. Things that I know that exist but have never had to deal with them.

Looking at the kernel source code they rarely get used. Many drivers have process bound data structures but none of the drivers use these flags. I'm wondering why.

Why sgx_alloc_epc_section() is a use case given that it is something that allocates memory for the global EPC database?

Quoted 2 lines

[*] AFAICT, systemd is upping the max number of open files to 1M on my
systems. I don't _think_ I changed a setting anywhere?

/Jarkko

#12 lore raw
Quoted 63 lines · Jarkko Sakkinen
Quoted 51 lines · Sean Christopherson
Quoted 35 lines · Jarkko Sakkinen
Quoted 27 lines · Sean Christopherson
Quoted 26 lines · Jarkko Sakkinen

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
Quoted 6 lines

arch/x86/kernel/cpu/sgx/ioctl.c static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg) · line 677

Quoted 2 lines
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);

Would it make sense to use GFP_KERNEL_ACCOUNT? The accounting would be weird for the case where userspace is using a builder process, but even in that case it's not flat out wrong to account per-enclave memory allocations.

Quoted 2 lines

I did not find a single call site that would use that for allocating memory for function-internal data.

Actually, the fact that the allocations are transient is an even better argument for accounting the memory, as the weirdness I was referring to doesn't exist for the builder concept.

But looking more closely, Documentation/core-api/memory-allocation.rst states:

* Untrusted allocations triggered from userspace should be a subject of kmem accounting and must have ``__GFP_ACCOUNT`` bit set. There is the handy ``GFP_KERNEL_ACCOUNT`` shortcut for ``GFP_KERNEL`` allocations that should be accounted.

Quoted 4 lines

That means all uses of GFP_KERNEL except in sgx_alloc_epc_section() should be converted to GFP_KERNEL_ACCOUNTED. As is, depending on fd limits[*], a single process can easily burn through multiple GBs of memory simply by opening /dev/sgx/enclave in a loop.

What does the documentation mean by untrusted allocaton?

__GFP_ACCOUNT kernel and GFP_KERNEL_ACCOUNT are both quite alien flags to me as is kmemcg. Things that I know that exist but have never had to deal with them.

Quoted 9 lines

Looking at the kernel source code they rarely get used. Many drivers have process bound data structures but none of the drivers use these flags. I'm wondering why.

Why sgx_alloc_epc_section() is a use case given that it is something that allocates memory for the global EPC database?

Quoted 2 lines

[*] AFAICT, systemd is upping the max number of open files to 1M on my
systems. I don't _think_ I changed a setting anywhere?

Anyway, the tree is now updated:

arch/x86/kernel/cpu/sgx/ioctl.c +13 -5
index 5b82670bb79a..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -427,12 +427,20 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 
 	if (addp->flags & SGX_PAGE_MEASURE) {
 		ret = __sgx_encl_extend(encl, epc_page);
-		if (ret)
+
+		/*
+		 * Destroy the enclave if EEXTEND fails, EADD can't be undone.
+		 * Note, destroy() also frees the resources for the added page.
+		 */
+		if (ret) {
 			sgx_encl_destroy(encl);
-		else
-			sgx_mark_page_reclaimable(encl_page->epc_page);
+			goto out_unlock;
+		}
 	}
 
+	sgx_mark_page_reclaimable(encl_page->epc_page);
+
+out_unlock:
 	mutex_unlock(&encl->lock);
 	up_read(&current->mm->mmap_sem);
 	return ret;
@@ -666,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);
 	if (!initp_page)
 		return -ENOMEM;
 

Hope that all the updates will be fairly localized :-)

/Jarkko

#13 lore raw
Quoted 122 lines · Jarkko Sakkinen
Quoted 63 lines · Jarkko Sakkinen
Quoted 51 lines · Sean Christopherson
Quoted 35 lines · Jarkko Sakkinen
Quoted 27 lines · Sean Christopherson
Quoted 26 lines · Jarkko Sakkinen

The reasoning is the same as in

http://git.infradead.org/users/jjs/linux-tpmdd.git/commit/abd55954f91a3aacc1d260d2411cf776ec4d5fd2

---
 arch/x86/kernel/cpu/sgx/ioctl.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
index 5b28a9c0cb68..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -674,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
Quoted 6 lines

arch/x86/kernel/cpu/sgx/ioctl.c static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg) · line 677

Quoted 2 lines
arch/x86/kernel/cpu/sgx/ioctl.c +1 -1
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);

Would it make sense to use GFP_KERNEL_ACCOUNT? The accounting would be weird for the case where userspace is using a builder process, but even in that case it's not flat out wrong to account per-enclave memory allocations.

Quoted 2 lines

I did not find a single call site that would use that for allocating memory for function-internal data.

Actually, the fact that the allocations are transient is an even better argument for accounting the memory, as the weirdness I was referring to doesn't exist for the builder concept.

But looking more closely, Documentation/core-api/memory-allocation.rst states:

* Untrusted allocations triggered from userspace should be a subject of kmem accounting and must have ``__GFP_ACCOUNT`` bit set. There is the handy ``GFP_KERNEL_ACCOUNT`` shortcut for ``GFP_KERNEL`` allocations that should be accounted.

Quoted 4 lines

That means all uses of GFP_KERNEL except in sgx_alloc_epc_section() should be converted to GFP_KERNEL_ACCOUNTED. As is, depending on fd limits[*], a single process can easily burn through multiple GBs of memory simply by opening /dev/sgx/enclave in a loop.

What does the documentation mean by untrusted allocaton?

__GFP_ACCOUNT kernel and GFP_KERNEL_ACCOUNT are both quite alien flags to me as is kmemcg. Things that I know that exist but have never had to deal with them.

Quoted 9 lines

Looking at the kernel source code they rarely get used. Many drivers have process bound data structures but none of the drivers use these flags. I'm wondering why.

Why sgx_alloc_epc_section() is a use case given that it is something that allocates memory for the global EPC database?

Quoted 2 lines

[*] AFAICT, systemd is upping the max number of open files to 1M on my
systems. I don't _think_ I changed a setting anywhere?

Anyway, the tree is now updated:

arch/x86/kernel/cpu/sgx/ioctl.c +13 -5
index 5b82670bb79a..d53aee5a64c1 100644
@@ -259,7 +259,7 @@ static long sgx_ioc_enclave_create(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&ecreate, arg, sizeof(ecreate)))
 		return -EFAULT;
 
-	secs_page = alloc_page(GFP_HIGHUSER);
+	secs_page = alloc_page(GFP_KERNEL);
 	if (!secs_page)
 		return -ENOMEM;
 
@@ -427,12 +427,20 @@ static int sgx_encl_add_page(struct sgx_encl *encl,
 
 	if (addp->flags & SGX_PAGE_MEASURE) {
 		ret = __sgx_encl_extend(encl, epc_page);
-		if (ret)
+
+		/*
+		 * Destroy the enclave if EEXTEND fails, EADD can't be undone.
+		 * Note, destroy() also frees the resources for the added page.
+		 */
+		if (ret) {
 			sgx_encl_destroy(encl);
-		else
-			sgx_mark_page_reclaimable(encl_page->epc_page);
+			goto out_unlock;
+		}
 	}
 
+	sgx_mark_page_reclaimable(encl_page->epc_page);
+
+out_unlock:
 	mutex_unlock(&encl->lock);
 	up_read(&current->mm->mmap_sem);
 	return ret;
@@ -666,7 +674,7 @@ static long sgx_ioc_enclave_init(struct sgx_encl *encl, void __user *arg)
 	if (copy_from_user(&einit, arg, sizeof(einit)))
 		return -EFAULT;
 
-	initp_page = alloc_page(GFP_HIGHUSER);
+	initp_page = alloc_page(GFP_KERNEL);
 	if (!initp_page)
 		return -ENOMEM;
 
Quoted 1 line

Hope that all the updates will be fairly localized :-)

Also removed some patches on top that were pushed by accident (patches under review).

/Jarkko