kernel/mm-pmd_read_atomic-fix-32bit-PAE-pmd-walk-vs-pmd_populate-SMP-race-condition.patch

273 lines
12 KiB
Diff

Path: news.gmane.org!not-for-mail
From: Andrea Arcangeli <aarcange@redhat.com>
Newsgroups: gmane.linux.kernel.mm
Subject: [PATCH] mm: pmd_read_atomic: fix 32bit PAE pmd walk vs pmd_populate SMP race condition
Date: Thu, 24 May 2012 01:39:01 +0200
Lines: 208
Approved: news@gmane.org
Message-ID: <1337816341-30743-1-git-send-email-aarcange@redhat.com>
References: <20120518230028.GF32479@redhat.com>
NNTP-Posting-Host: plane.gmane.org
X-Trace: dough.gmane.org 1337816354 18906 80.91.229.3 (23 May 2012 23:39:14 GMT)
X-Complaints-To: usenet@dough.gmane.org
NNTP-Posting-Date: Wed, 23 May 2012 23:39:14 +0000 (UTC)
Cc: Andrew Morton <akpm@linux-foundation.org>, Mel Gorman <mgorman@suse.de>,
Hugh Dickins <hughd@google.com>, Larry Woodman <lwoodman@redhat.com>,
Petr Matousek <pmatouse@redhat.com>,
Ulrich Obergfell <uobergfe@redhat.com>, Rik van Riel <riel@redhat.com>
To: linux-mm@kvack.org
Original-X-From: owner-linux-mm@kvack.org Thu May 24 01:39:12 2012
Return-path: <owner-linux-mm@kvack.org>
Envelope-to: glkm-linux-mm-2@m.gmane.org
Original-Received: from kanga.kvack.org ([205.233.56.17])
by plane.gmane.org with esmtp (Exim 4.69)
(envelope-from <owner-linux-mm@kvack.org>)
id 1SXL94-0002ub-3P
for glkm-linux-mm-2@m.gmane.org; Thu, 24 May 2012 01:39:10 +0200
Original-Received: by kanga.kvack.org (Postfix)
id 1684A6B0083; Wed, 23 May 2012 19:39:09 -0400 (EDT)
Delivered-To: linux-mm-outgoing@kvack.org
Original-Received: by kanga.kvack.org (Postfix, from userid 40)
id 080DD6B0092; Wed, 23 May 2012 19:39:08 -0400 (EDT)
X-Original-To: int-list-linux-mm@kvack.org
Delivered-To: int-list-linux-mm@kvack.org
Original-Received: by kanga.kvack.org (Postfix, from userid 63042)
id C84046B00E7; Wed, 23 May 2012 19:39:08 -0400 (EDT)
X-Original-To: linux-mm@kvack.org
Delivered-To: linux-mm@kvack.org
Original-Received: from psmtp.com (na3sys010amx119.postini.com [74.125.245.119])
by kanga.kvack.org (Postfix) with SMTP id 0B2DC6B0083
for <linux-mm@kvack.org>; Wed, 23 May 2012 19:39:07 -0400 (EDT)
Original-Received: from mx1.redhat.com ([209.132.183.28]) (using TLSv1) by na3sys010amx119.postini.com ([74.125.244.10]) with SMTP;
Wed, 23 May 2012 18:39:08 CDT
Original-Received: from int-mx12.intmail.prod.int.phx2.redhat.com (int-mx12.intmail.prod.int.phx2.redhat.com [10.5.11.25])
by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id q4NNd3dP002492
(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK);
Wed, 23 May 2012 19:39:03 -0400
Original-Received: from random.random (ovpn-113-72.phx2.redhat.com [10.3.113.72])
by int-mx12.intmail.prod.int.phx2.redhat.com (8.14.4/8.14.4) with ESMTP id q4NNd1P7012233;
Wed, 23 May 2012 19:39:02 -0400
In-Reply-To: <20120518230028.GF32479@redhat.com>
X-Scanned-By: MIMEDefang 2.68 on 10.5.11.25
X-pstn-neptune: 0/0/0.00/0
X-pstn-levels: (S:99.90000/99.90000 CV:99.9000 FC:95.5390 LC:95.5390 R:95.9108 P:95.9108 M:97.0282 C:98.6951 )
X-pstn-dkim: 0 skipped:not-enabled
X-pstn-settings: 3 (1.0000:1.0000) s cv gt3 gt2 gt1 r p m c
X-pstn-addresses: from <aarcange@redhat.com> [db-null]
X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.2
Original-Sender: owner-linux-mm@kvack.org
Precedence: bulk
X-Loop: owner-majordomo@kvack.org
List-ID: <linux-mm.kvack.org>
Xref: news.gmane.org gmane.linux.kernel.mm:78936
Archived-At: <http://permalink.gmane.org/gmane.linux.kernel.mm/78936>
When holding the mmap_sem for reading, pmd_offset_map_lock should only
run on a pmd_t that has been read atomically from the pmdp
pointer, otherwise we may read only half of it leading to this crash.
PID: 11679 TASK: f06e8000 CPU: 3 COMMAND: "do_race_2_panic"
#0 [f06a9dd8] crash_kexec at c049b5ec
#1 [f06a9e2c] oops_end at c083d1c2
#2 [f06a9e40] no_context at c0433ded
#3 [f06a9e64] bad_area_nosemaphore at c043401a
#4 [f06a9e6c] __do_page_fault at c0434493
#5 [f06a9eec] do_page_fault at c083eb45
#6 [f06a9f04] error_code (via page_fault) at c083c5d5
EAX: 01fb470c EBX: fff35000 ECX: 00000003 EDX: 00000100 EBP:
00000000
DS: 007b ESI: 9e201000 ES: 007b EDI: 01fb4700 GS: 00e0
CS: 0060 EIP: c083bc14 ERR: ffffffff EFLAGS: 00010246
#7 [f06a9f38] _spin_lock at c083bc14
#8 [f06a9f44] sys_mincore at c0507b7d
#9 [f06a9fb0] system_call at c083becd
start len
EAX: ffffffda EBX: 9e200000 ECX: 00001000 EDX: 6228537f
DS: 007b ESI: 00000000 ES: 007b EDI: 003d0f00
SS: 007b ESP: 62285354 EBP: 62285388 GS: 0033
CS: 0073 EIP: 00291416 ERR: 000000da EFLAGS: 00000286
This should be a longstanding bug affecting x86 32bit PAE without
THP. Only archs with 64bit large pmd_t and 32bit unsigned long should
be affected.
With THP enabled the barrier() in
pmd_none_or_trans_huge_or_clear_bad() would partly hide the bug when
the pmd transition from none to stable, by forcing a re-read of the
*pmd in pmd_offset_map_lock, but when THP is enabled a new set of
problem arises by the fact could then transition freely in any of the
none, pmd_trans_huge or pmd_trans_stable states. So making the barrier
in pmd_none_or_trans_huge_or_clear_bad() unconditional isn't good idea
and it would be a flakey solution.
This should be fully fixed by introducing a pmd_read_atomic that reads
the pmd in order with THP disabled, or by reading the pmd atomically
with cmpxchg8b with THP enabled.
Luckily this new race condition only triggers in the places that must
already be covered by pmd_none_or_trans_huge_or_clear_bad() so the fix
is localized there but this bug is not related to THP.
NOTE: this can trigger on x86 32bit systems with PAE enabled with more
than 4G of ram, otherwise the high part of the pmd will never risk to
be truncated because it would be zero at all times, in turn so hiding
the SMP race.
This bug was discovered and fully debugged by Ulrich, quote:
----
[..]
pmd_none_or_trans_huge_or_clear_bad() loads the content of edx and
eax.
496 static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t
*pmd)
497 {
498 /* depend on compiler for an atomic pmd read */
499 pmd_t pmdval = *pmd;
// edi = pmd pointer
0xc0507a74 <sys_mincore+548>: mov 0x8(%esp),%edi
...
// edx = PTE page table high address
0xc0507a84 <sys_mincore+564>: mov 0x4(%edi),%edx
...
// eax = PTE page table low address
0xc0507a8e <sys_mincore+574>: mov (%edi),%eax
[..]
Please note that the PMD is not read atomically. These are two "mov"
instructions where the high order bits of the PMD entry are fetched
first. Hence, the above machine code is prone to the following race.
- The PMD entry {high|low} is 0x0000000000000000.
The "mov" at 0xc0507a84 loads 0x00000000 into edx.
- A page fault (on another CPU) sneaks in between the two "mov"
instructions and instantiates the PMD.
- The PMD entry {high|low} is now 0x00000003fda38067.
The "mov" at 0xc0507a8e loads 0xfda38067 into eax.
----
Reported-by: Ulrich Obergfell <uobergfe@redhat.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
---
arch/x86/include/asm/pgtable-3level.h | 50 +++++++++++++++++++++++++++++++++
include/asm-generic/pgtable.h | 22 +++++++++++++-
2 files changed, 70 insertions(+), 2 deletions(-)
diff --git a/arch/x86/include/asm/pgtable-3level.h b/arch/x86/include/asm/pgtable-3level.h
index effff47..43876f1 100644
--- a/arch/x86/include/asm/pgtable-3level.h
+++ b/arch/x86/include/asm/pgtable-3level.h
@@ -31,6 +31,56 @@ static inline void native_set_pte(pte_t *ptep, pte_t pte)
ptep->pte_low = pte.pte_low;
}
+#define pmd_read_atomic pmd_read_atomic
+/*
+ * pte_offset_map_lock on 32bit PAE kernels was reading the pmd_t with
+ * a "*pmdp" dereference done by gcc. Problem is, in certain places
+ * where pte_offset_map_lock is called, concurrent page faults are
+ * allowed, if the mmap_sem is hold for reading. An example is mincore
+ * vs page faults vs MADV_DONTNEED. On the page fault side
+ * pmd_populate rightfully does a set_64bit, but if we're reading the
+ * pmd_t with a "*pmdp" on the mincore side, a SMP race can happen
+ * because gcc will not read the 64bit of the pmd atomically. To fix
+ * this all places running pmd_offset_map_lock() while holding the
+ * mmap_sem in read mode, shall read the pmdp pointer using this
+ * function to know if the pmd is null nor not, and in turn to know if
+ * they can run pmd_offset_map_lock or pmd_trans_huge or other pmd
+ * operations.
+ *
+ * Without THP if the mmap_sem is hold for reading, the
+ * pmd can only transition from null to not null while pmd_read_atomic runs.
+ * So there's no need of literally reading it atomically.
+ *
+ * With THP if the mmap_sem is hold for reading, the pmd can become
+ * THP or null or point to a pte (and in turn become "stable") at any
+ * time under pmd_read_atomic, so it's mandatory to read it atomically
+ * with cmpxchg8b.
+ */
+#ifndef CONFIG_TRANSPARENT_HUGEPAGE
+static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
+{
+ pmdval_t ret;
+ u32 *tmp = (u32 *)pmdp;
+
+ ret = (pmdval_t) (*tmp);
+ if (ret) {
+ /*
+ * If the low part is null, we must not read the high part
+ * or we can end up with a partial pmd.
+ */
+ smp_rmb();
+ ret |= ((pmdval_t)*(tmp + 1)) << 32;
+ }
+
+ return (pmd_t) { ret };
+}
+#else /* CONFIG_TRANSPARENT_HUGEPAGE */
+static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
+{
+ return (pmd_t) { atomic64_read((atomic64_t *)pmdp) };
+}
+#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+
static inline void native_set_pte_atomic(pte_t *ptep, pte_t pte)
{
set_64bit((unsigned long long *)(ptep), native_pte_val(pte));
diff --git a/include/asm-generic/pgtable.h b/include/asm-generic/pgtable.h
index 125c54e..fa596d9 100644
--- a/include/asm-generic/pgtable.h
+++ b/include/asm-generic/pgtable.h
@@ -446,6 +446,18 @@ static inline int pmd_write(pmd_t pmd)
#endif /* __HAVE_ARCH_PMD_WRITE */
#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
+#ifndef pmd_read_atomic
+static inline pmd_t pmd_read_atomic(pmd_t *pmdp)
+{
+ /*
+ * Depend on compiler for an atomic pmd read. NOTE: this is
+ * only going to work, if the pmdval_t isn't larger than
+ * an unsigned long.
+ */
+ return *pmdp;
+}
+#endif
+
/*
* This function is meant to be used by sites walking pagetables with
* the mmap_sem hold in read mode to protect against MADV_DONTNEED and
@@ -459,11 +471,17 @@ static inline int pmd_write(pmd_t pmd)
* undefined so behaving like if the pmd was none is safe (because it
* can return none anyway). The compiler level barrier() is critically
* important to compute the two checks atomically on the same pmdval.
+ *
+ * For 32bit kernels with a 64bit large pmd_t this automatically takes
+ * care of reading the pmd atomically to avoid SMP race conditions
+ * against pmd_populate() when the mmap_sem is hold for reading by the
+ * caller (a special atomic read not done by "gcc" as in the generic
+ * version above, is also needed when THP is disabled because the page
+ * fault can populate the pmd from under us).
*/
static inline int pmd_none_or_trans_huge_or_clear_bad(pmd_t *pmd)
{
- /* depend on compiler for an atomic pmd read */
- pmd_t pmdval = *pmd;
+ pmd_t pmdval = pmd_read_atomic(pmd);
/*
* The barrier will stabilize the pmdval in a register or on
* the stack so that it will stop changing under the code.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>