linux-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Greg Price <price@MIT.EDU>
To: "Theodore Ts'o" <tytso@MIT.EDU>
Cc: linux-kernel@vger.kernel.org, Jiri Kosina <jkosina@suse.cz>
Subject: [PATCH 10/11] random: pull 'min' check in accounting to inside lockless update
Date: Thu, 7 Nov 2013 18:59:32 -0500	[thread overview]
Message-ID: <20131107235932.GM16018@ringworld.MIT.EDU> (raw)
In-Reply-To: <cover.1383868558.git.price@mit.edu>

Since 902c098a3 ("random: use lockless techniques in the interrupt
path") we have protected entropy_count only with cmpxchg, not the
lock.  In 10b3a32d2 ("random: fix accounting race condition") we
put a cmpxchg-retry loop around most of the logic in account(),
but the enforcement of the 'min' parameter stayed outside.

Under concurrent accesses to /dev/urandom this can suffer a race
that defeats our "catastrophic reseed" measures so that our
reseeding isn't effective.  If accesses to /dev/urandom and
/dev/random are interleaved in the wrong pattern, we could go
indefinitely long without a successful catastrophic reseed of
/dev/urandom, even while consuming an indefinite amount of entropy
in ineffective smaller reseeds.

NB that with or without this bug, if /dev/random is simply
accessed constantly, we can go indefinitely long without any
reseed of /dev/urandom.  So the effect of the bug is not that big.

The race comes when two tasks are in account() on input_pool
and access ->entropy_count in the following order:

             task A                           task B
   if (r->entropy_count / 8
               < min + reserved)
                                   ACCESS_ONCE(r->entropy_count)
                                   cmpxchg
   ACCESS_ONCE(r->entropy_count)
   cmpxchg

where B causes r->entropy_count / 8 to fall below A's
min + reserved but above reserved.  Task A will cheerfully
take r->entropy_count / 8 - reserved bytes from the pool,
even though this is less than min.

Move the "min" check inside the ACCESS_ONCE/cmpxchg loop to
prevent the race.

Cc: Jiri Kosina <jkosina@suse.cz>
Cc: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Price <price@mit.edu>
---
 drivers/char/random.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/drivers/char/random.c b/drivers/char/random.c
index 1bf6bf8..87d3728 100644
--- a/drivers/char/random.c
+++ b/drivers/char/random.c
@@ -842,18 +842,17 @@ static size_t account(struct entropy_store *r, size_t nbytes, int min,
 {
 	unsigned long flags;
 	int wakeup_write = 0;
+	int entropy_count, orig;
 
 	BUG_ON(r->entropy_count > r->poolinfo->POOLBITS);
 	DEBUG_ENT("trying to extract %zu bits from %s\n",
 		  nbytes * 8, r->name);
 
-	/* Can we pull enough? */
-	if (r->entropy_count / 8 < min + reserved) {
+retry:
+	entropy_count = orig = ACCESS_ONCE(r->entropy_count);
+	if (entropy_count / 8 < min + reserved) {
 		nbytes = 0;
 	} else {
-		int entropy_count, orig;
-retry:
-		entropy_count = orig = ACCESS_ONCE(r->entropy_count);
 		/* If limited, never pull more than available */
 		if (r->limit)
 			nbytes = min_t(size_t, nbytes, entropy_count/8 - reserved);
-- 
1.8.3.2


  parent reply	other threads:[~2013-11-08  0:04 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2013-11-07 23:57 [PATCH 00/11] random: code cleanups Greg Price
2013-11-07 23:57 ` [PATCH 01/11] random: fix typos / spelling errors in comments Greg Price
2013-11-07 23:58 ` [PATCH 02/11] random: fix comment on proc_do_uuid Greg Price
2013-11-07 23:58 ` [PATCH 03/11] random: fix description of get_random_bytes Greg Price
2013-11-07 23:58 ` [PATCH 04/11] random: simplify loop in random_read Greg Price
2013-11-07 23:58 ` [PATCH 05/11] random: declare trickle_count unsigned Greg Price
2013-11-07 23:58 ` [PATCH 06/11] random: fix comment on "account" Greg Price
2013-11-07 23:58 ` [PATCH 07/11] random: simplify accounting code slightly Greg Price
2013-11-07 23:59 ` [PATCH 08/11] random: simplify accounting logic Greg Price
2013-11-07 23:59 ` [PATCH 09/11] random: forget lock in lockless accounting Greg Price
2013-11-07 23:59 ` Greg Price [this message]
2013-11-07 23:59 ` [PATCH 11/11] random: simplify accounting code Greg Price
2013-11-12  4:24 ` [PATCH 00/11] random: code cleanups Theodore Ts'o
2013-11-12 22:40   ` Greg Price
2013-11-13  3:32     ` Theodore Ts'o
2013-11-13  4:02       ` H. Peter Anvin
2013-11-13  4:37         ` Greg Price
2013-11-13  4:51           ` H. Peter Anvin
2013-11-13  6:06             ` Greg Price
2013-11-13  4:23       ` Greg Price
2013-11-13  6:08         ` Theodore Ts'o
2013-11-13  6:28           ` Greg Price

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=20131107235932.GM16018@ringworld.MIT.EDU \
    --to=price@mit.edu \
    --cc=jkosina@suse.cz \
    --cc=linux-kernel@vger.kernel.org \
    --cc=tytso@MIT.EDU \
    /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 a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).