generic: test extending sub-block AIO writes for races
authorEric Sandeen <sandeen@sandeen.net>
Wed, 14 Oct 2015 03:19:31 +0000 (14:19 +1100)
committerDave Chinner <david@fromorbit.com>
Wed, 14 Oct 2015 03:19:31 +0000 (14:19 +1100)
This tests Brian Foster's fix for xfs:

   xfs: always drain dio before extending aio write submission

It launches four adjacent 1k IOs past EOF, then reads back
to see if we have 4k worth of the data we wrote, or something else -
possibly zeros from sub-block zeroing and eof racing.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
src/aio-dio-regress/aio-dio-eof-race.c [new file with mode: 0644]
tests/generic/114 [new file with mode: 0755]
tests/generic/114.out [new file with mode: 0644]
tests/generic/group

diff --git a/src/aio-dio-regress/aio-dio-eof-race.c b/src/aio-dio-regress/aio-dio-eof-race.c
new file mode 100644 (file)
index 0000000..79fd928
--- /dev/null
@@ -0,0 +1,173 @@
+/*
+ * Launch 4 sub-block AIOs past EOF and ensure that we don't see
+ * corruption from racing sub-block zeroing when they're complete.
+ *
+ * Copyright (C) 2015 Red Hat, Inc. All Rights reserved.
+ *
+ * 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 2 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, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA.
+ */
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <ctype.h>
+
+#include <libaio.h>
+
+/* Sized to allow 4 x 512 AIOs */
+#define BUF_SIZE       2048
+#define IO_PATTERN     0xab
+
+void
+dump_buffer(
+       void    *buf,
+       off64_t offset,
+       ssize_t len)
+{
+       int     i, j;
+       char    *p;
+       int     new;
+
+       for (i = 0, p = (char *)buf; i < len; i += 16) {
+               char    *s = p;
+
+               if (i && !memcmp(p, p - 16, 16)) {
+                       new = 0;
+               } else {
+                       if (i)
+                               printf("*\n");
+                       new = 1;
+               }
+
+               if (!new) {
+                       p += 16;
+                       continue;
+               }
+
+               printf("%08llx  ", (unsigned long long)offset + i);
+               for (j = 0; j < 16 && i + j < len; j++, p++)
+                       printf("%02x ", *p);
+               printf(" ");
+               for (j = 0; j < 16 && i + j < len; j++, s++) {
+                       if (isalnum((int)*s))
+                               printf("%c", *s);
+                       else
+                               printf(".");
+               }
+               printf("\n");
+
+       }
+       printf("%08llx\n", (unsigned long long)offset + i);
+}
+
+int main(int argc, char *argv[])
+{
+       struct io_context *ctx = NULL;
+       struct io_event evs[4];
+       struct iocb iocb1, iocb2, iocb3, iocb4;
+       struct iocb *iocbs[] = { &iocb1, &iocb2, &iocb3, &iocb4 };
+       void *buf;
+       struct stat statbuf;
+       char cmp_buf[BUF_SIZE];
+       int fd, err = 0;
+       off_t eof;
+
+       fd = open(argv[1], O_DIRECT | O_CREAT | O_TRUNC | O_RDWR, 0600);
+       if (fd == -1) {
+               perror("open");
+               return 1;
+       }
+
+       err = posix_memalign(&buf, getpagesize(), BUF_SIZE);
+       if (err) {
+               fprintf(stderr, "error %s during %s\n",
+                       strerror(err),
+                       "posix_memalign");
+               return 1;
+       }
+       memset(cmp_buf, IO_PATTERN, BUF_SIZE);
+
+       err = io_setup(4, &ctx);
+       if (err) {
+               fprintf(stderr, "error %s during %s\n",
+                       strerror(err),
+                       "io_setup");
+               return 1;
+       }
+
+       eof = 0;
+
+       /* Keep extending until 8MB (fairly arbitrary) */
+       while (eof < 8 * 1024 * 1024) {
+               memset(buf, IO_PATTERN, BUF_SIZE);
+               fstat(fd, &statbuf);
+               eof = statbuf.st_size;
+
+               /*
+                * Split the buffer into multiple I/Os to send a mix of block
+                * aligned/unaligned writes as well as writes that start beyond
+                * the current EOF. This stresses things like inode size
+                * management and stale block zeroing for races and can lead to
+                * data corruption when not handled properly.
+                */
+               io_prep_pwrite(&iocb1, fd, buf, BUF_SIZE/4, eof + 0*BUF_SIZE/4);
+               io_prep_pwrite(&iocb2, fd, buf, BUF_SIZE/4, eof + 1*BUF_SIZE/4);
+               io_prep_pwrite(&iocb3, fd, buf, BUF_SIZE/4, eof + 2*BUF_SIZE/4);
+               io_prep_pwrite(&iocb4, fd, buf, BUF_SIZE/4, eof + 3*BUF_SIZE/4);
+
+               err = io_submit(ctx, 4, iocbs);
+               if (err != 4) {
+                       fprintf(stderr, "error %s during %s\n",
+                               strerror(err),
+                               "io_submit");
+                       return 1;
+               }
+
+               err = io_getevents(ctx, 4, 4, evs, NULL);
+               if (err != 4) {
+                       fprintf(stderr, "error %s during %s\n",
+                               strerror(err),
+                               "io_getevents");
+                       return 1;
+               }
+
+               /*
+                * And then read it back.
+                *
+                * Using pread to keep it simple, but AIO has the same effect.
+                * eof is the prior eof; we just wrote BUF_SIZE more.
+                */
+               if (pread(fd, buf, BUF_SIZE, eof) != BUF_SIZE) {
+                       perror("pread");
+                       return 1;
+               }
+
+               /*
+                * We launched 4 AIOs which, stitched together, should write
+                * a seamless BUF_SIZE worth of IO_PATTERN to the last block.
+                */
+               if (memcmp(buf, cmp_buf, BUF_SIZE)) {
+                       printf("corruption while extending from %ld\n", eof);
+                       dump_buffer(buf, 0, BUF_SIZE);
+                       return 1;
+               }
+       }
+
+       printf("Success, all done.\n");
+       return 0;
+}
diff --git a/tests/generic/114 b/tests/generic/114
new file mode 100755 (executable)
index 0000000..8082a0d
--- /dev/null
@@ -0,0 +1,65 @@
+#! /bin/bash
+# FS QA Test No. 114
+#
+# Test races while extending past EOF via sub-block AIO writes
+#
+#-----------------------------------------------------------------------
+# Copyright (c) 2015 Red Hat, Inc.  All Rights Reserved.
+#
+# 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.
+#
+# This program is distributed in the hope that it would 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, write the Free Software Foundation,
+# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+#-----------------------------------------------------------------------
+#
+
+seq=`basename $0`
+seqres=$RESULT_DIR/$seq
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=/tmp/$$
+status=1       # failure is the default!
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+_cleanup()
+{
+    cd /
+    rm -f $TEST_DIR/tst-aio-dio-eof-race
+}
+
+# get standard environment, filters and checks
+. ./common/rc
+. ./common/filter
+
+_supported_fs generic
+_supported_os Linux
+
+_require_test
+_require_sparse_files
+_require_aiodio aio-dio-eof-race
+
+# Test does 512 byte DIO, so make sure that'll work
+logical_block_size=`_min_dio_alignment $TEST_DEV`
+
+if [ "$logical_block_size" -gt "512" ]; then
+       _notrun "device block size: $logical_block_size greater than 512"
+fi
+
+# We don't mind 512-byte fs blocks; the IOs won't be sub-block,
+# but the test should still pass, even if it doesn't stress the code
+# we're targeting.
+
+# Note, this test does several extending loops internally
+$AIO_TEST $TEST_DIR/tst-aio-dio-eof-race
+
+status=$?
+exit
diff --git a/tests/generic/114.out b/tests/generic/114.out
new file mode 100644 (file)
index 0000000..4c57dc9
--- /dev/null
@@ -0,0 +1,2 @@
+QA output created by 114
+Success, all done.
index 4ae256fc54f915d65af0aaf72a16a9aba09decb9..1dd42698ee1a9d07383056fac3685d4f5dc9d361 100644 (file)
 109 auto metadata dir
 112 rw aio auto quick
 113 rw aio auto quick
+114 rw aio auto quick
 117 attr auto quick
 120 other auto quick
 123 perms auto quick