hugetlb/hugemmap: add new testcase hugemmap06.c

Description of Problem:

        There is a race condition if we map a same file on different processes.
        Region tracking is protected by mmap_sem and hugetlb_instantiation_mutex.
        When we do mmap, we don't grab a hugetlb_instantiation_mutex, but only
        mmap_sem (exclusively).  This doesn't prevent other tasks from modifying
        the region structure, so it can be modified by two processes concurrently.

        Testcase hugemmap06.c is the trigger to cause system crash:
        crash> bt -s
        PID: 4492   TASK: ffff88033e437520  CPU: 2   COMMAND: "hugemmap06"
         #0 [ffff88033dbb3960] machine_kexec+395 at ffffffff8103d1ab
         #1 [ffff88033dbb39c0] crash_kexec+114 at ffffffff810cc4f2
         #2 [ffff88033dbb3a90] oops_end+192 at ffffffff8153c840
         #3 [ffff88033dbb3ac0] die+91 at ffffffff81010f5b
         #4 [ffff88033dbb3af0] do_general_protection+338 at ffffffff8153c332
         #5 [ffff88033dbb3b20] general_protection+37 at ffffffff8153bb05
            [exception RIP: list_del+40]
            RIP: ffffffff812a3598  RSP: ffff88033dbb3bd8  RFLAGS: 00010292
            RAX: dead000000100100  RBX: ffff88013cf37340  RCX: 0000000000002dc2
            RDX: dead000000200200  RSI: 0000000000000046  RDI: 0000000000000009
            RBP: ffff88033dbb3be8   R8: 0000000000015598   R9: 0000000000000000
            R10: 000000000000000f  R11: 0000000000000009  R12: 000000000000000a
            R13: ffff88033d64b9e8  R14: ffff88033e5b9720  R15: ffff88013cf37340
            ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0000
         #6 [ffff88033dbb3bf0] region_add+154 at ffffffff811698da
         #7 [ffff88033dbb3c40] alloc_huge_page+669 at ffffffff8116a61d
         #8 [ffff88033dbb3ce0] hugetlb_fault+1083 at ffffffff8116b9bb
         #9 [ffff88033dbb3d90] handle_mm_fault+917 at ffffffff81153295
        #10 [ffff88033dbb3e00] __do_page_fault+326 at ffffffff8104f156
        #11 [ffff88033dbb3f20] do_page_fault+62 at ffffffff8153e78e
        #12 [ffff88033dbb3f50] page_fault+37 at ffffffff8153bb35
            RIP: 00000000004027c6  RSP: 00007f7cadef9e80  RFLAGS: 00010297
            RAX: 000000005a49238f  RBX: 00007ffcb2d19320  RCX: 000000357498e084
            RDX: 000000357498e0b0  RSI: 00007f7cadef9e5c  RDI: 000000357498e4e0
            RBP: 0000000000000008   R8: 000000357498e0a0   R9: 000000357498e100
            R10: 00007f7cadefa9d0  R11: 0000000000000206  R12: 0000000000000007
            R13: 0000000000000002  R14: 0000000000000003  R15: 00002aaaac000000
            ORIG_RAX: ffffffffffffffff  CS: 0033  SS: 002b

The fix are all these below commits:
f522c3ac00(mm, hugetlb: change variable name reservations to resv) 9119a41e90(mm, hugetlb: unify region structure handling) 7b24d8616b(mm, hugetlb: fix race in region tracking) 1406ec9ba6(mm, hugetlb: improve, cleanup resv_map parameters)

Signed-off-by Li Wang <[email protected]>

---

Notes:
v3 --> v4
- declare sz type and assign value at the same line
- change 'void *addr' to 'char * addr' in the mp struct
- take use of 'mmap_sz->addr + ...' in a clean way
- remove the useless pthread_join
- fix a tpyo 'mremap'

 runtest/hugetlb                                    |   1 +
 testcases/kernel/mem/.gitignore                    |   1 +
 testcases/kernel/mem/hugetlb/hugemmap/Makefile     |   2 +
 testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c | 175 +++++++++++++++++++++
 4 files changed, 179 insertions(+)
 create mode 100644 testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c

runtest/hugetlb +1 -0

index 2e9f215..ac24513 100644
@@ -2,6 +2,7 @@ hugemmap01 hugemmap01
 hugemmap02 hugemmap02
 hugemmap04 hugemmap04
 hugemmap05 hugemmap05
+hugemmap06 hugemmap06
 hugemmap05_1 hugemmap05 -m
 hugemmap05_2 hugemmap05 -s
 hugemmap05_3 hugemmap05 -s -m

testcases/kernel/mem/.gitignore +1 -0

index 4702377..ac8e6d8 100644
@@ -3,6 +3,7 @@
 /hugetlb/hugemmap/hugemmap02
 /hugetlb/hugemmap/hugemmap04
 /hugetlb/hugemmap/hugemmap05
+/hugetlb/hugemmap/hugemmap06
 /hugetlb/hugeshmat/hugeshmat01
 /hugetlb/hugeshmat/hugeshmat02
 /hugetlb/hugeshmat/hugeshmat03

testcases/kernel/mem/hugetlb/hugemmap/Makefile +2 -0

index f51f6b9..2473aca 100644
@@ -25,3 +25,5 @@ top_srcdir		?= ../../../../..
 include $(top_srcdir)/include/mk/testcases.mk
 include $(abs_srcdir)/../Makefile.inc
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
+
+hugemmap06: CFLAGS+=-pthread

testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c new file +65 -0

new file mode 100644
index 0000000..e52b21d
@@ -0,0 +1,175 @@
+/*
+ *  Copyright (c) 2015 Red Hat, Inc.
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program. If not, see <http://www.gnu.org/licenses/>.
+ */
+
+/*
+ * DESCRIPTION
+ *
+ *   There is a race condition if we map a same file on different processes.
+ *   Region tracking is protected by mmap_sem and hugetlb_instantiation_mutex.
+ *   When we do mmap, we don't grab a hugetlb_instantiation_mutex, but only
+ *   mmap_sem (exclusively).  This doesn't prevent other tasks from modifying
+ *   the region structure, so it can be modified by two processes concurrently.
+ *
+ *   This bug was fixed on stable kernel by commits:
+ *       f522c3ac00(mm, hugetlb: change variable name reservations to resv)
+ *       9119a41e90(mm, hugetlb: unify region structure handling)
+ *       7b24d8616b(mm, hugetlb: fix race in region tracking)
+ *       1406ec9ba6(mm, hugetlb: improve, cleanup resv_map parameters)
+ *
+ * AUTHOR:
+ *    Herton R. Krzesinski <[email protected]>
+ *    Li Wang <[email protected]>
+ */
+
+#define _GNU_SOURCE
+#include <errno.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "test.h"
+#include "mem.h"
+#include "hugetlb.h"
+
+char *TCID = "hugemmap06";
+int TST_TOTAL = 5;
+
+static long hpage_size;
+static long hugepages;
+
+struct mp {
+	char *addr;
+	int sz;
+};
+
+#define ARSZ 50
+
+void setup(void)
+{

testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c new file +100 -0

+
+	set_sys_tune("nr_hugepages", hugepages, 1);
+
+	TEST_PAUSE;
+}
+
+void cleanup(void)
+{
+	set_sys_tune("nr_hugepages", orig_hugepages, 0);
+}
+
+void *thr(void *arg)
+{
+	struct mp *mmap_sz = arg;
+	int i, lim, a, b, c;
+
+	srand(time(NULL));
+	lim = rand() % 10;
+	for (i = 0; i < lim; i++) {
+		a = rand() % mmap_sz->sz;
+		for (c = 0; c <= a; c++) {
+			b = rand() % mmap_sz->sz;
+			*(mmap_sz->addr + b * hpage_size) = rand();
+		}
+	}
+	return NULL;
+}
+
+void do_mmap(void)
+{
+	int i, sz = ARSZ + 1;
+	void *addr, *new_addr;
+	struct mp mmap_sz[ARSZ];
+	pthread_t tid[ARSZ];
+
+	addr = mmap(NULL, sz * hpage_size,
+			PROT_READ | PROT_WRITE,
+			MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB,
+			-1, 0);
+
+	if (addr == MAP_FAILED) {
+		if (errno == ENOMEM) {
+			tst_brkm(TCONF, cleanup,
+				"Cannot allocate hugepage, memory too fragmented?");
+		}
+
+		tst_brkm(TBROK | TERRNO, cleanup, "Cannot allocate hugepage");
+	}
+
+	for (i = 0; i < ARSZ; ++i, --sz) {
+		mmap_sz[i].sz = sz;
+		mmap_sz[i].addr = addr;
+
+		TEST(pthread_create(&tid[i], NULL, thr, &mmap_sz[i]));
+		if (TEST_RETURN)
+			tst_brkm(TBROK | TRERRNO, cleanup,
+					"pthread_create failed");
+
+		new_addr = mmap(addr, (sz - 1) * hpage_size,
+				PROT_READ | PROT_WRITE,
+				MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_FIXED,
+				-1, 0);
+
+		if (new_addr == MAP_FAILED)
+			tst_brkm(TFAIL | TERRNO, cleanup, "mmap failed");
+
+		addr = new_addr;
+	}
+
+	for (i = 0; i < ARSZ; ++i) {
+		TEST(pthread_join(tid[i], NULL));
+		if (TEST_RETURN)
+			tst_brkm(TBROK | TRERRNO, cleanup,
+					"pthread_join failed");
+	}
+
+	if (munmap(addr, sz * hpage_size) == -1)
+		tst_brkm(TFAIL | TERRNO, cleanup, "huge munmap failed");
+}
+
+int main(int ac, char **av)
+{
+	int lc, i;
+
+	tst_parse_opts(ac, av, NULL, NULL);
+
+	setup();
+
+	for (lc = 0; TEST_LOOPING(lc); lc++) {
+		tst_count = 0;
+
+		for (i = 0; i < TST_TOTAL; i++)
+			do_mmap();
+
+		tst_resm(TPASS, "No regression found.");
+	}
+
+	cleanup();
+	tst_exit();
+}

...

Quoted 27 lines

testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c new file +27 -0

+#define _GNU_SOURCE
+#include <errno.h>
+#include <pthread.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "test.h"
+#include "mem.h"
+#include "hugetlb.h"
+
+char *TCID = "hugemmap06";
+int TST_TOTAL = 5;
+
+static long hpage_size;
+static long hugepages;
+
+struct mp {
+     char *addr;
+     int sz;
+};
+
+#define ARSZ 50
+
+void setup(void)

testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c

Quoted 10 lines

testcases/kernel/mem/hugetlb/hugemmap/hugemmap06.c new file +10 -0

+{
+     tst_require_root();
+     check_hugepage();
+
+     hpage_size = read_meminfo("Hugepagesize:") * 1024;
+     orig_hugepages = get_sys_tune("nr_hugepages");
+
+     hugepages = (ARSZ + 1) * TST_TOTAL;
+
+     if (hugepages * read_meminfo("Hugepagesize:") >

read_meminfo("MemTotal:"))

Quoted 1 line

+ tst_brkm(TCONF, NULL, "System RAM is not enough to test.");

Just a general note:

What happens on architectures where no hugepages are available? Maybe you should add a check and return TCONF() in those cases?

I think this is missing for the other existing hugepage testcases as well...

Helge

Quoted 4 lines

I think this is missing for the other existing hugepage testcases as well..

Hi,

On Mon, Dec 21, 2015 at 9:36 PM, Helge Deller <[email protected] <mailto:[email protected]>> wrote:

    ...
    > +#define _GNU_SOURCE
    > +#include <errno.h>
    > +#include <pthread.h>
    > +#include <stdio.h>
    > +#include <stdlib.h>
    > +#include <sys/mman.h>
    > +#include <sys/types.h>
    > +#include <unistd.h>
    > +
    > +#include "test.h"
    > +#include "mem.h"
    > +#include "hugetlb.h"
    > +
    > +char *TCID = "hugemmap06";
    > +int TST_TOTAL = 5;
    > +
    > +static long hpage_size;
    > +static long hugepages;
    > +
    > +struct mp {
    > +     char *addr;
    > +     int sz;
    > +};
    > +
    > +#define ARSZ 50
    > +
    > +void setup(void)
    > +{
    > +     tst_require_root();
    > +     check_hugepage();
    > +
    > +     hpage_size = read_meminfo("Hugepagesize:") * 1024;
    > +     orig_hugepages = get_sys_tune("nr_hugepages");
    > +
    > +     hugepages = (ARSZ + 1) * TST_TOTAL;
    > +
    > +     if (hugepages * read_meminfo("Hugepagesize:") > read_meminfo("MemTotal:"))
    > +             tst_brkm(TCONF, NULL, "System RAM is not enough to test.");


    Just a general note:

    What happens on architectures where no hugepages are available?
    Maybe you should add a check and return TCONF() in those cases?

thanks for you kind note, the function above ?check_hugepage()? do that work. you could find details in the include file: /hugetlb/lib/hugetlb.h

From: "Li Wang" <[email protected]>
To: [email protected]
Cc: [email protected]
Sent: Monday, 21 December, 2015 12:16:53 PM
Subject: [PATCH v4] hugetlb/hugemmap: add new testcase hugemmap06.c