[PATCH v2] keys: make keyring indexes independent of kernel addresses

[PATCH v2] keys: make keyring indexes independent of kernel addresses

[Kyle Zeng]

KEYCTL_READ returns the serial numbers in a keyring in associative-array
traversal order.  The first word of the index is a public, deterministic
hash of the type pointer, domain pointer and description.  An unprivileged
user can therefore insert keys with known descriptions, predict their
order for each possible KASLR slide, and recover the kernel image address.
On an x86-64 build with 16 MiB alignment, 45 user keys suffice to
distinguish the candidate slides.

Remove the addresses from the whole index, not just its first hash word.
Assign each key type and domain a nonzero 64-bit ID and hash those IDs and
the description with a once-initialized SipHash key.  Use the full IDs in
the later index chunks as well.  Truncating or hashing the IDs there would
lose the guarantee that distinct type/domain/description tuples have a
differing index bit.

Allocate IDs lazily since private key types need not be registered.  Cache
them in the index key so tree walks and copies of an existing index do not
need to dereference a type descriptor that may have been unregistered.
Keep the 32-bit chunk layout and diff_objects() bit offsets consistent,
and retain the separate root slot for keyrings used by recursive search.
The key comparison, permissions and userspace data format are unchanged;
only the address-dependent in-memory ordering is replaced.

Fixes: b2a4df200d57 ("KEYS: Expand the capacity of a keyring")
Assisted-by: Codex:gpt-6-astra
Signed-off-by: Kyle Zeng <kylebot@openai.com>
---
Changes in v2:
- Use reverse Christmas tree order for local variable declarations,
  preserving initializer dependencies. No functional changes.

 include/linux/key-type.h |    1 
 include/linux/key.h      |    5 ++
 security/keys/keyring.c  |  150 +++++++++++++++++++++++++---------------------
 3 files changed, 86 insertions(+), 70 deletions(-)

diff --git a/include/linux/key-type.h b/include/linux/key-type.h
index bb97bd3e5af4..c4f329b6ae3f 100644
--- a/include/linux/key-type.h
+++ b/include/linux/key-type.h
@@ -161,6 +161,7 @@ struct key_type {
 				     const void *in, const void *in2);
 
 	/* internal fields */
+	atomic64_t		index_id;	/* Opaque keyring index ID */
 	struct list_head	link;		/* link in types list */
 	struct lock_class_key	lock_class;	/* key->sem lock class */
 } __randomize_layout;
diff --git a/include/linux/key.h b/include/linux/key.h
index 81b8f05c6898..e40b5267658f 100644
--- a/include/linux/key.h
+++ b/include/linux/key.h
@@ -107,6 +107,7 @@ struct keyring_name;
 
 struct key_tag {
 	struct rcu_head		rcu;
+	atomic64_t		index_id;	/* Opaque keyring index ID */
 	refcount_t		usage;
 	bool			removed;	/* T when subject removed */
 };
@@ -129,6 +130,8 @@ struct keyring_index_key {
 	struct key_type		*type;
 	struct key_tag		*domain_tag;	/* Domain of operation */
 	const char		*description;
+	u64			type_id;	/* Cached index IDs */
+	u64			domain_id;
 };
 
 union key_payload {
@@ -251,6 +254,8 @@ struct key {
 			struct key_type	*type;		/* type of key */
 			struct key_tag	*domain_tag;	/* Domain of operation */
 			char		*description;
+			u64		type_id;
+			u64		domain_id;
 		};
 	};
 
diff --git a/security/keys/keyring.c b/security/keys/keyring.c
index 15bf4af8f282..6910ddf00457 100644
--- a/security/keys/keyring.c
+++ b/security/keys/keyring.c
@@ -14,6 +14,9 @@
 #include <linux/err.h>
 #include <linux/user_namespace.h>
 #include <linux/nsproxy.h>
+#include <linux/once.h>
+#include <linux/random.h>
+#include <linux/siphash.h>
 #include <keys/keyring-type.h>
 #include <keys/user-type.h>
 #include <linux/assoc_array_priv.h>
@@ -147,49 +150,46 @@ static int keyring_instantiate(struct key *keyring,
 }
 
 /*
- * Multiply 64-bits by 32-bits to 96-bits and fold back to 64-bit.  Ideally we'd
- * fold the carry back too, but that requires inline asm.
+ * Give key types and domains address-independent identities.  Zero means that
+ * an ID has not yet been allocated.  As with inode sequence numbers used by
+ * futexes, this relies on the 64-bit counter not wrapping in a machine's
+ * lifetime.
  */
-static u64 mult_64x32_and_fold(u64 x, u32 y)
+static u64 key_index_id(atomic64_t *id)
 {
-	u64 hi = (u64)(u32)(x >> 32) * y;
-	u64 lo = (u64)(u32)(x) * y;
-	return lo + ((u64)(u32)hi << 32) + (u32)(hi >> 32);
+	static atomic64_t next_id;
+	u64 old, new;
+
+	old = atomic64_read(id);
+	if (likely(old))
+		return old;
+
+	do {
+		new = atomic64_inc_return(&next_id);
+	} while (WARN_ON_ONCE(!new));
+
+	old = 0;
+	if (!atomic64_try_cmpxchg_relaxed(id, &old, new))
+		return old;
+	return new;
 }
 
 /*
- * Hash a key type and description.
+ * Hash a key type, domain and description.  KEYCTL_READ exposes the tree's
+ * traversal order, so no part of the index may contain kernel addresses.
  */
 static void hash_key_type_and_desc(struct keyring_index_key *index_key)
 {
 	const unsigned level_shift = ASSOC_ARRAY_LEVEL_STEP;
 	const unsigned long fan_mask = ASSOC_ARRAY_FAN_MASK;
+	static siphash_key_t hash_key __read_mostly;
-	const char *description = index_key->description;
-	unsigned long hash, type;
-	u32 piece;
+	unsigned long hash;
 	u64 acc;
-	int n, desc_len = index_key->desc_len;
-
-	type = (unsigned long)index_key->type;
-	acc = mult_64x32_and_fold(type, desc_len + 13);
-	acc = mult_64x32_and_fold(acc, 9207);
-	piece = (unsigned long)index_key->domain_tag;
-	acc = mult_64x32_and_fold(acc, piece);
-	acc = mult_64x32_and_fold(acc, 9207);
-
-	for (;;) {
-		n = desc_len;
-		if (n <= 0)
-			break;
-		if (n > 4)
-			n = 4;
-		piece = 0;
-		memcpy(&piece, description, n);
-		description += n;
-		desc_len -= n;
-		acc = mult_64x32_and_fold(acc, piece);
-		acc = mult_64x32_and_fold(acc, 9207);
-	}
+
+	get_random_once(&hash_key, sizeof(hash_key));
+	acc = siphash(index_key->description, index_key->desc_len, &hash_key);
+	acc = siphash_3u64(index_key->type_id, index_key->domain_id, acc,
+			   &hash_key);
 
 	/* Fold the hash down to 32 bits if need be. */
 	hash = acc;
@@ -209,7 +209,7 @@ static void hash_key_type_and_desc(struct keyring_index_key *index_key)
 
 /*
  * Finalise an index key to include a part of the description actually in the
- * index key, to set the domain tag and to calculate the hash.
+ * index key, to set the domain tag and index IDs, and to calculate the hash.
  */
 void key_set_index_key(struct keyring_index_key *index_key)
 {
@@ -225,6 +225,14 @@ void key_set_index_key(struct keyring_index_key *index_key)
 			index_key->domain_tag = &default_domain_tag;
 	}
 
+	/* Preserve the IDs when copying an existing key's index.  In
+	 * particular, its type may already have been unregistered.
+	 */
+	if (!index_key->type_id)
+		index_key->type_id = key_index_id(&index_key->type->index_id);
+	if (!index_key->domain_id)
+		index_key->domain_id = key_index_id(&index_key->domain_tag->index_id);
+
 	hash_key_type_and_desc(index_key);
 }
 
@@ -263,43 +271,50 @@ void key_remove_domain(struct key_tag *domain_tag)
 /*
  * Build the next index key chunk.
  *
+ * The hash and inline description are followed by the 64-bit type and domain
+ * IDs, low word first, and then by the rest of the description.
  * We return it one word-sized chunk at a time.
  */
 static unsigned long keyring_get_key_chunk(const void *data, int level)
 {
+	const unsigned int id_chunks = sizeof(u64) / sizeof(unsigned long);
 	const struct keyring_index_key *index_key = data;
+	int desc_len = index_key->desc_len;
 	unsigned long chunk = 0;
-	const u8 *d;
-	int desc_len = index_key->desc_len, n = sizeof(chunk);
+	int n = sizeof(chunk);
 	unsigned int offset;
+	const u8 *d;
 
 	level /= ASSOC_ARRAY_KEY_CHUNK_SIZE;
 	switch (level) {
 	case 0:
 		return index_key->hash;
 	case 1:
 		return index_key->x;
-	case 2:
-		return (unsigned long)index_key->type;
-	case 3:
-		return (unsigned long)index_key->domain_tag;
-	default:
-		level -= 4;
-		offset = sizeof(index_key->desc) + level * sizeof(long);
-		if (desc_len <= offset)
-			return 0;
-
-		d = index_key->description + offset;
-		desc_len -= offset;
-		if (desc_len > n)
-			desc_len = n;
-		d += desc_len;
-		do {
-			chunk <<= 8;
-			chunk |= *--d;
-		} while (--desc_len > 0);
-		return chunk;
 	}
+
+	level -= 2;
+	if (level < id_chunks)
+		return index_key->type_id >> (level * ASSOC_ARRAY_KEY_CHUNK_SIZE);
+	level -= id_chunks;
+	if (level < id_chunks)
+		return index_key->domain_id >> (level * ASSOC_ARRAY_KEY_CHUNK_SIZE);
+	level -= id_chunks;
+
+	offset = sizeof(index_key->desc) + level * sizeof(long);
+	if (desc_len <= offset)
+		return 0;
+
+	d = index_key->description + offset;
+	desc_len -= offset;
+	if (desc_len > n)
+		desc_len = n;
+	d += desc_len;
+	do {
+		chunk <<= 8;
+		chunk |= *--d;
+	} while (--desc_len > 0);
+	return chunk;
 }
 
 static unsigned long keyring_get_object_key_chunk(const void *object, int level)
@@ -330,6 +345,7 @@ static int keyring_diff_objects(const void *object, const void *data)
 	const struct keyring_index_key *a = &key_a->index_key;
 	const struct keyring_index_key *b = data;
 	unsigned long seg_a, seg_b;
 	int level, i;
+	u64 diff;
 
 	level = 0;
@@ -339,28 +355,22 @@ static int keyring_diff_objects(const void *object, const void *data)
 		goto differ;
 	level += ASSOC_ARRAY_KEY_CHUNK_SIZE / 8;
 
-	/* The number of bits contributed by the hash is controlled by a
-	 * constant in the assoc_array headers.  Everything else thereafter we
-	 * can deal with as being machine word-size dependent.
-	 */
+	/* The inline description contributes one machine-word-sized chunk. */
 	seg_a = a->x;
 	seg_b = b->x;
 	if ((seg_a ^ seg_b) != 0)
 		goto differ;
 	level += sizeof(unsigned long);
 
-	/* The next bit may not work on big endian */
-	seg_a = (unsigned long)a->type;
-	seg_b = (unsigned long)b->type;
-	if ((seg_a ^ seg_b) != 0)
-		goto differ;
-	level += sizeof(unsigned long);
+	diff = a->type_id ^ b->type_id;
+	if (diff)
+		return level * 8 + __ffs64(diff);
+	level += sizeof(u64);
 
-	seg_a = (unsigned long)a->domain_tag;
-	seg_b = (unsigned long)b->domain_tag;
-	if ((seg_a ^ seg_b) != 0)
-		goto differ;
-	level += sizeof(unsigned long);
+	diff = a->domain_id ^ b->domain_id;
+	if (diff)
+		return level * 8 + __ffs64(diff);
+	level += sizeof(u64);
 
 	i = sizeof(a->desc);
 	if (a->desc_len <= i)
-- 
2.53.0

Re: [PATCH v2] keys: make keyring indexes independent of kernel addresses

[Jarkko Sakkinen]

On Tue, Oct 13, 2026 at 04:12:51PM -0700, Kyle Zeng wrote:
> KEYCTL_READ returns the serial numbers in a keyring in associative-array
> traversal order.  The first word of the index is a public, deterministic
> hash of the type pointer, domain pointer and description.  An unprivileged
> user can therefore insert keys with known descriptions, predict their
> order for each possible KASLR slide, and recover the kernel image address.
> On an x86-64 build with 16 MiB alignment, 45 user keys suffice to
> distinguish the candidate slides.
> 
> Remove the addresses from the whole index, not just its first hash word.
> Assign each key type and domain a nonzero 64-bit ID and hash those IDs and
> the description with a once-initialized SipHash key.  Use the full IDs in
> the later index chunks as well.  Truncating or hashing the IDs there would
> lose the guarantee that distinct type/domain/description tuples have a
> differing index bit.
> 
> Allocate IDs lazily since private key types need not be registered.  Cache
> them in the index key so tree walks and copies of an existing index do not
> need to dereference a type descriptor that may have been unregistered.
> Keep the 32-bit chunk layout and diff_objects() bit offsets consistent,
> and retain the separate root slot for keyrings used by recursive search.
> The key comparison, permissions and userspace data format are unchanged;
> only the address-dependent in-memory ordering is replaced.
> 
> Fixes: b2a4df200d57 ("KEYS: Expand the capacity of a keyring")
> Assisted-by: Codex:gpt-6-astra
> Signed-off-by: Kyle Zeng <kylebot@openai.com>


Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>

I'll apply this.

And thanks for addressing that and sorry for nitpicking. Looks good to
me and thanks for addressing the bug. This is kind of AI assisted commit
to works for me; and I believe that checks are balances are in place.

Not everyone gets this, and the problem is that they also in effect
reduce the number of point of failures from two to one due
ignorance-by-a-choice.

A person submitting an AI-assisted patch is also a reviewer, and
ignoring this increases the probability of bad getting into mainline
quite literally.

I think some people might not really understand this at all but is
just that it is much less likely that two people get a patch wrong
and at the same time, than just one. That's it ;-)

This is not an AI/LLM problem. It is a misconduct by people quite
well scoped really...

Sorry for the rant just wanted to write this down and get this out
of my chest. Absolutely nothing to do with this patch.

Br, Jarkko

Re: [PATCH v2] keys: make keyring indexes independent of kernel addresses

[Jarkko Sakkinen]

On Wed, Oct 14, 2026 at 03:43:06PM +0300, Jarkko Sakkinen wrote:
> On Tue, Oct 13, 2026 at 04:12:51PM -0700, Kyle Zeng wrote:
> > KEYCTL_READ returns the serial numbers in a keyring in associative-array
> > traversal order.  The first word of the index is a public, deterministic
> > hash of the type pointer, domain pointer and description.  An unprivileged
> > user can therefore insert keys with known descriptions, predict their
> > order for each possible KASLR slide, and recover the kernel image address.
> > On an x86-64 build with 16 MiB alignment, 45 user keys suffice to
> > distinguish the candidate slides.
> > 
> > Remove the addresses from the whole index, not just its first hash word.
> > Assign each key type and domain a nonzero 64-bit ID and hash those IDs and
> > the description with a once-initialized SipHash key.  Use the full IDs in
> > the later index chunks as well.  Truncating or hashing the IDs there would
> > lose the guarantee that distinct type/domain/description tuples have a
> > differing index bit.
> > 
> > Allocate IDs lazily since private key types need not be registered.  Cache
> > them in the index key so tree walks and copies of an existing index do not
> > need to dereference a type descriptor that may have been unregistered.
> > Keep the 32-bit chunk layout and diff_objects() bit offsets consistent,
> > and retain the separate root slot for keyrings used by recursive search.
> > The key comparison, permissions and userspace data format are unchanged;
> > only the address-dependent in-memory ordering is replaced.
> > 
> > Fixes: b2a4df200d57 ("KEYS: Expand the capacity of a keyring")
> > Assisted-by: Codex:gpt-6-astra
> > Signed-off-by: Kyle Zeng <kylebot@openai.com>
> 
> 
> Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>
> 
> I'll apply this.

Now I did it, and your patch is now in my master branch (not yet -next,
will get there). Thank you.

It took a while because kunit test for assoc-array took almost the whole
day [1] (mainly figuring out test data that is not too much pain
to recap later on that replaced lottery I originally had).

[1] https://lore.kernel.org/keyrings/20261014160144.1835860-1-jarkko@kernel.org/T/#u

Br, Jarkko