All of lore.kernel.org
 help / color / mirror / Atom feed
From: Petr Pavlu <petr.pavlu@suse.com>
To: dhowells@redhat.com, jarkko@kernel.org
Cc: keyrings@vger.kernel.org, linux-kernel@vger.kernel.org,
	Petr Pavlu <petr.pavlu@suse.com>
Subject: [PATCH] keys: Fix linking a duplicate key to a keyring's assoc_array
Date: Thu, 23 Mar 2023 14:04:12 +0100	[thread overview]
Message-ID: <20230323130412.32097-1-petr.pavlu@suse.com> (raw)

When making a DNS query inside the kernel using dns_query(), the request
code can in rare cases end up creating a duplicate index key in the
assoc_array of the destination keyring. It is eventually found by
a BUG_ON() check in the assoc_array implementation and results in
a crash.

Example report:
[2158499.700025] kernel BUG at ../lib/assoc_array.c:652!
[2158499.700039] invalid opcode: 0000 [#1] SMP PTI
[2158499.700065] CPU: 3 PID: 31985 Comm: kworker/3:1 Kdump: loaded Not tainted 5.3.18-150300.59.90-default #1 SLE15-SP3
[2158499.700096] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 11/12/2020
[2158499.700351] Workqueue: cifsiod cifs_resolve_server [cifs]
[2158499.700380] RIP: 0010:assoc_array_insert+0x85f/0xa40
[2158499.700401] Code: ff 74 2b 48 8b 3b 49 8b 45 18 4c 89 e6 48 83 e7 fe e8 95 ec 74 00 3b 45 88 7d db 85 c0 79 d4 0f 0b 0f 0b 0f 0b e8 41 f2 be ff <0f> 0b 0f 0b 81 7d 88 ff ff ff 7f 4c 89 eb 4c 8b ad 58 ff ff ff 0f
[2158499.700448] RSP: 0018:ffffc0bd6187faf0 EFLAGS: 00010282
[2158499.700470] RAX: ffff9f1ea7da2fe8 RBX: ffff9f1ea7da2fc1 RCX: 0000000000000005
[2158499.700492] RDX: 0000000000000000 RSI: 0000000000000005 RDI: 0000000000000000
[2158499.700515] RBP: ffffc0bd6187fbb0 R08: ffff9f185faf1100 R09: 0000000000000000
[2158499.700538] R10: ffff9f1ea7da2cc0 R11: 000000005ed8cec8 R12: ffffc0bd6187fc28
[2158499.700561] R13: ffff9f15feb8d000 R14: ffff9f1ea7da2fc0 R15: ffff9f168dc0d740
[2158499.700585] FS:  0000000000000000(0000) GS:ffff9f185fac0000(0000) knlGS:0000000000000000
[2158499.700610] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[2158499.700630] CR2: 00007fdd94fca238 CR3: 0000000809d8c006 CR4: 00000000003706e0
[2158499.700702] Call Trace:
[2158499.700741]  ? key_alloc+0x447/0x4b0
[2158499.700768]  ? __key_link_begin+0x43/0xa0
[2158499.700790]  __key_link_begin+0x43/0xa0
[2158499.700814]  request_key_and_link+0x2c7/0x730
[2158499.700847]  ? dns_resolver_read+0x20/0x20 [dns_resolver]
[2158499.700873]  ? key_default_cmp+0x20/0x20
[2158499.700898]  request_key_tag+0x43/0xa0
[2158499.700926]  dns_query+0x114/0x2ca [dns_resolver]
[2158499.701127]  dns_resolve_server_name_to_ip+0x194/0x310 [cifs]
[2158499.701164]  ? scnprintf+0x49/0x90
[2158499.701190]  ? __switch_to_asm+0x40/0x70
[2158499.701211]  ? __switch_to_asm+0x34/0x70
[2158499.701405]  reconn_set_ipaddr_from_hostname+0x81/0x2a0 [cifs]
[2158499.701603]  cifs_resolve_server+0x4b/0xd0 [cifs]
[2158499.701632]  process_one_work+0x1f8/0x3e0
[2158499.701658]  worker_thread+0x2d/0x3f0
[2158499.701682]  ? process_one_work+0x3e0/0x3e0
[2158499.701703]  kthread+0x10d/0x130
[2158499.701723]  ? kthread_park+0xb0/0xb0
[2158499.701746]  ret_from_fork+0x1f/0x40

The situation occurs as follows:
* Some kernel facility invokes dns_query() to resolve a hostname, for
  example, "abcdef". The function registers its global DNS resolver
  cache as current->cred.thread_keyring and passes the query to
  request_key_net() -> request_key_tag() -> request_key_and_link().
* Function request_key_and_link() creates a keyring_search_context
  object. Its match_data.cmp method gets set via a call to
  type->match_preparse() (resolves to dns_resolver_match_preparse()) to
  dns_resolver_cmp().
* Function request_key_and_link() continues and invokes
  search_process_keyrings_rcu() which returns that a given key was not
  found. The control is then passed to request_key_and_link() ->
  construct_alloc_key().
* Concurrently to that, a second task similarly makes a DNS query for
  "abcdef." and its result gets inserted into the DNS resolver cache.
* Back on the first task, function construct_alloc_key() first runs
  __key_link_begin() to determine an assoc_array_edit operation to
  insert a new key. Index keys in the array are compared exactly as-is,
  using keyring_compare_object(). The operation finds that "abcdef" is
  not yet present in the destination keyring.
* Function construct_alloc_key() continues and checks if a given key is
  already present on some keyring by again calling
  search_process_keyrings_rcu(). This search is done using
  dns_resolver_cmp() and "abcdef" gets matched with now present key
  "abcdef.".
* The found key is linked on the destination keyring by calling
  __key_link() and using the previously calculated assoc_array_edit
  operation. This inserts the "abcdef." key in the array but creates
  a duplicity because the same index key is already present.

Fix the problem by postponing __key_link_begin() in
construct_alloc_key() until an actual key which should be linked into
the destination keyring is determined.

Signed-off-by: Petr Pavlu <petr.pavlu@suse.com>
---
 security/keys/request_key.c | 35 ++++++++++++++++++++++++-----------
 1 file changed, 24 insertions(+), 11 deletions(-)

diff --git a/security/keys/request_key.c b/security/keys/request_key.c
index 2da4404276f0..04eb7e4cedad 100644
--- a/security/keys/request_key.c
+++ b/security/keys/request_key.c
@@ -398,17 +398,21 @@ static int construct_alloc_key(struct keyring_search_context *ctx,
 	set_bit(KEY_FLAG_USER_CONSTRUCT, &key->flags);
 
 	if (dest_keyring) {
-		ret = __key_link_lock(dest_keyring, &ctx->index_key);
+		ret = __key_link_lock(dest_keyring, &key->index_key);
 		if (ret < 0)
 			goto link_lock_failed;
-		ret = __key_link_begin(dest_keyring, &ctx->index_key, &edit);
-		if (ret < 0)
-			goto link_prealloc_failed;
 	}
 
-	/* attach the key to the destination keyring under lock, but we do need
+	/*
+	 * Attach the key to the destination keyring under lock, but we do need
 	 * to do another check just in case someone beat us to it whilst we
-	 * waited for locks */
+	 * waited for locks.
+	 *
+	 * The caller might specify a comparison function which looks for keys
+	 * that do not exactly match but are still equivalent from the caller's
+	 * perspective. The __key_link_begin() operation must be done only after
+	 * an actual key is determined.
+	 */
 	mutex_lock(&key_construction_mutex);
 
 	rcu_read_lock();
@@ -417,12 +421,16 @@ static int construct_alloc_key(struct keyring_search_context *ctx,
 	if (!IS_ERR(key_ref))
 		goto key_already_present;
 
-	if (dest_keyring)
+	if (dest_keyring) {
+		ret = __key_link_begin(dest_keyring, &key->index_key, &edit);
+		if (ret < 0)
+			goto link_alloc_failed;
 		__key_link(dest_keyring, key, &edit);
+	}
 
 	mutex_unlock(&key_construction_mutex);
 	if (dest_keyring)
-		__key_link_end(dest_keyring, &ctx->index_key, edit);
+		__key_link_end(dest_keyring, &key->index_key, edit);
 	mutex_unlock(&user->cons_lock);
 	*_key = key;
 	kleave(" = 0 [%d]", key_serial(key));
@@ -435,10 +443,13 @@ static int construct_alloc_key(struct keyring_search_context *ctx,
 	mutex_unlock(&key_construction_mutex);
 	key = key_ref_to_ptr(key_ref);
 	if (dest_keyring) {
+		ret = __key_link_begin(dest_keyring, &key->index_key, &edit);
+		if (ret < 0)
+			goto link_alloc_failed_unlocked;
 		ret = __key_link_check_live_key(dest_keyring, key);
 		if (ret == 0)
 			__key_link(dest_keyring, key, &edit);
-		__key_link_end(dest_keyring, &ctx->index_key, edit);
+		__key_link_end(dest_keyring, &key->index_key, edit);
 		if (ret < 0)
 			goto link_check_failed;
 	}
@@ -453,8 +464,10 @@ static int construct_alloc_key(struct keyring_search_context *ctx,
 	kleave(" = %d [linkcheck]", ret);
 	return ret;
 
-link_prealloc_failed:
-	__key_link_end(dest_keyring, &ctx->index_key, edit);
+link_alloc_failed:
+	mutex_unlock(&key_construction_mutex);
+link_alloc_failed_unlocked:
+	__key_link_end(dest_keyring, &key->index_key, edit);
 link_lock_failed:
 	mutex_unlock(&user->cons_lock);
 	key_put(key);
-- 
2.35.3


             reply	other threads:[~2023-03-23 13:04 UTC|newest]

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-03-23 13:04 Petr Pavlu [this message]
2023-03-30  0:13 ` [PATCH] keys: Fix linking a duplicate key to a keyring's assoc_array Jarkko Sakkinen
2023-06-08  9:55   ` Petr Pavlu
2023-06-08 13:18     ` Jarkko Sakkinen
2023-06-08 13:28       ` Jarkko Sakkinen
2023-04-21  2:39 ` joeyli
2023-06-08 14:12 ` David Howells
2023-06-08 19:03   ` Jarkko Sakkinen
2023-06-09  9:49   ` Petr Pavlu

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20230323130412.32097-1-petr.pavlu@suse.com \
    --to=petr.pavlu@suse.com \
    --cc=dhowells@redhat.com \
    --cc=jarkko@kernel.org \
    --cc=keyrings@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.