Skip to contentSkip to conversation

avoid prepare_creds in faccessat when possible

PATCH In review Mateusz Guzik 2 participants 6 messages Mar 9, 2015 – Mar 13, 2015

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

cover avoid prepare_creds in faccessat when possible

Sometimes faccessat needs to modify current thread's credentials, but calls prepare_creds unconditionally.

However, typically resulting credentials are identical to original ones and in that case newcredentials are unnecessary. We can detect this before allocating anything.

This patch series adds a helper which allows comparing capability sets and modifies faccessat to use it.

Mateusz Guzik (2):
CAPABILITIES: add cap_isequal helper
fs: avoid unnecessary prepare_creds in faccessat

 fs/open.c                  | 53 ++++++++++++++++++++++++++++++----------------
 include/linux/capability.h | 10 +++++++++
 2 files changed, 45 insertions(+), 18 deletions(-)
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/2 CAPABILITIES: add cap_isequal helper

Can be used to determine whether two given sets have the same capabilities.

---
 include/linux/capability.h | 10 ++++++++++
 1 file changed, 10 insertions(+)
include/linux/capability.h +10 -0
index af9f0b9..2fcf941 100644
@@ -155,6 +155,16 @@ static inline int cap_isclear(const kernel_cap_t a)
 	return 1;
 }
 
+static inline int cap_isequal(const kernel_cap_t a, const kernel_cap_t b)
+{
+	unsigned __capi;
+	CAP_FOR_EACH_U32(__capi) {
+		if (a.cap[__capi] != b.cap[__capi])
+			return 0;
+	}
+	return 1;
+}
+
 /*
  * Check if "a" is a subset of "set".
  * return 1 if ALL of the capabilities in "a" are also in "set"
Signature · 1 line
1.8.3.1
#2 lore raw

Re: [PATCH 1/2] CAPABILITIES: add cap_isequal helper

Quoted 15 lines · Mateusz Guzik

Can be used to determine whether two given sets have the same capabilities.

---
 include/linux/capability.h | 10 ++++++++++
 1 file changed, 10 insertions(+)
include/linux/capability.h +0 -0
index af9f0b9..2fcf941 100644
@@ -155,6 +155,16 @@ static inline int cap_isclear(const kernel_cap_t a)
 	return 1;
 }

include/linux/capability.h static inline int cap_isclear(const kernel_cap_t a) · line 166

Quoted 10 lines
include/linux/capability.h +9 -0
 
+static inline int cap_isequal(const kernel_cap_t a, const kernel_cap_t b)
+{
+	unsigned __capi;
+	CAP_FOR_EACH_U32(__capi) {
+		if (a.cap[__capi] != b.cap[__capi])
+			return 0;
+	}
+	return 1;
+}

I realize it is currently only a two pass loop so probably not that big of a deal, but couldn't you accomplish the same with a memcmp()? I suppose the above implementation might be faster than those architectures which use the generic memcmp() implementation, but I wonder if the arch-specific memcmp() implementations would be faster.

Also, what is the main motivation for this patchset? Do you have a workload that is being hit hard by prepare_creds()?

Signature · 2 lines
paul moore
security @ redhat
#3 lore raw

Re: [PATCH 1/2] CAPABILITIES: add cap_isequal helper

Quoted 17 lines · Paul Moore
Quoted 16 lines · Mateusz Guzik

Can be used to determine whether two given sets have the same capabilities.

---
 include/linux/capability.h | 10 ++++++++++
 1 file changed, 10 insertions(+)
include/linux/capability.h +0 -0
index af9f0b9..2fcf941 100644
@@ -155,6 +155,16 @@ static inline int cap_isclear(const kernel_cap_t a)
 	return 1;
 }
 
Quoted 16 lines

include/linux/capability.h static inline int cap_isclear(const kernel_cap_t a) · line 166

Quoted 9 lines
include/linux/capability.h +9 -0
+static inline int cap_isequal(const kernel_cap_t a, const kernel_cap_t b)
+{
+	unsigned __capi;
+	CAP_FOR_EACH_U32(__capi) {
+		if (a.cap[__capi] != b.cap[__capi])
+			return 0;
+	}
+	return 1;
+}

I realize it is currently only a two pass loop so probably not that big of a deal, but couldn't you accomplish the same with a memcmp()? I suppose the above implementation might be faster than those architectures which use the generic memcmp() implementation, but I wonder if the arch-specific memcmp() implementations would be faster.

Well I did it this way for consistency with the rest of the file. Trying to use memcpy with only 2 elements to compare may be a dubious optimisation and would require providing additional macros for cap size.

As such, I would prefer to keep the loop as it is. This can be changed should caps ever grow.

Quoted 3 lines

Also, what is the main motivation for this patchset? Do you have a workload that is being hit hard by prepare_creds()?

It's just something I stumbled upon and decided to microoptimize (fwiw, faccessat is called quite often, but not enough for this change to be world-changing).

Given the triviality of the patch I figured it should be fine to do it.

Signature · 1 line
Mateusz Guzik
#4 lore raw

PATCH 2/2 fs: avoid unnecessary prepare_creds in faccessat

In reply to Mateusz Guzik

Sometimes faccessat needs to modify current thread's credentials, but calls prepare_creds unconditionally.

Take advantage of the fact that we can detect whether any modification to credentials is needed and in turn avoid unnecessary allocations.

---
 fs/open.c | 53 +++++++++++++++++++++++++++++++++++------------------
 1 file changed, 35 insertions(+), 18 deletions(-)
fs/open.c +35 -18
index 33f9cbf..166eb45 100644
@@ -330,8 +330,10 @@ SYSCALL_DEFINE4(fallocate, int, fd, int, mode, loff_t, offset, loff_t, len)
  */
 SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 {
-	const struct cred *old_cred;
-	struct cred *override_cred;
+	const struct cred *old_cred = current_cred();
+	struct cred *override_cred = NULL;
+	kernel_cap_t cap_effective;
+	int modify_cap_effective = 0;
 	struct path path;
 	struct inode *inode;
 	int res;
@@ -340,24 +342,37 @@ SYSCALL_DEFINE3(faccessat, int, dfd, const char __user *, filename, int, mode)
 	if (mode & ~S_IRWXO)	/* where's F_OK, X_OK, W_OK, R_OK? */
 		return -EINVAL;
 
-	override_cred = prepare_creds();
-	if (!override_cred)
-		return -ENOMEM;
-
-	override_cred->fsuid = override_cred->uid;
-	override_cred->fsgid = override_cred->gid;
-
 	if (!issecure(SECURE_NO_SETUID_FIXUP)) {
 		/* Clear the capabilities if we switch to a non-root user */
-		kuid_t root_uid = make_kuid(override_cred->user_ns, 0);
-		if (!uid_eq(override_cred->uid, root_uid))
-			cap_clear(override_cred->cap_effective);
-		else
-			override_cred->cap_effective =
-				override_cred->cap_permitted;
+		kuid_t root_uid = make_kuid(old_cred->user_ns, 0);
+		if (!uid_eq(old_cred->uid, root_uid)) {
+			if (!cap_isclear(old_cred->cap_effective)) {
+				cap_clear(cap_effective);
+				modify_cap_effective = 1;
+			}
+		} else {
+			if (!cap_isequal(old_cred->cap_effective,
+				old_cred->cap_permitted)) {
+				cap_effective = old_cred->cap_permitted;
+				modify_cap_effective = 1;
+			}
+		}
 	}
 
-	old_cred = override_creds(override_cred);
+	if (!uid_eq(old_cred->fsuid, old_cred->uid) ||
+	    !gid_eq(old_cred->fsgid, old_cred->gid) ||
+	    modify_cap_effective) {
+		override_cred = prepare_creds();
+		if (!override_cred)
+			return -ENOMEM;
+
+		override_cred->fsuid = override_cred->uid;
+		override_cred->fsgid = override_cred->gid;
+		if (modify_cap_effective)
+			override_cred->cap_effective = cap_effective;
+
+		override_creds(override_cred);
+	}
 retry:
 	res = user_path_at(dfd, filename, lookup_flags, &path);
 	if (res)
@@ -399,8 +414,10 @@ out_path_release:
 		goto retry;
 	}
 out:
-	revert_creds(old_cred);
-	put_cred(override_cred);
+	if (override_cred) {
+		revert_creds(old_cred);
+		put_cred(override_cred);
+	}
 	return res;
 }
 
Signature · 1 line
1.8.3.1
#5 lore raw
In reply to Mateusz Guzik
Quoted 7 lines · Mateusz Guzik

Sometimes faccessat needs to modify current thread's credentials, but calls prepare_creds unconditionally.

However, typically resulting credentials are identical to original ones and in that case newcredentials are unnecessary. We can detect this before allocating anything.

Quoted 7 lines

This patch series adds a helper which allows comparing capability sets and modifies faccessat to use it.

Mateusz Guzik (2):
CAPABILITIES: add cap_isequal helper
fs: avoid unnecessary prepare_creds in faccessat

Can I get some (N)ACKs on this one?

Thanks,

Signature · 1 line
Mateusz Guzik