Merge "ioshark: A Repeatable Application Workload Based Storage Benchmark."

11 files changed, 2449 insertions, 0 deletions

diff --git a/ioshark/Android.mk b/ioshark/Android.mk
new file mode 100644
index 00000000..a2e4816d
--- /dev/null
+++ b/ioshark/Android.mk
@@ -0,0 +1,30 @@
+# Copyright (C) 2016 The Android Open Source Project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+LOCAL_PATH:= $(call my-dir)
+
+include $(CLEAR_VARS)
+LOCAL_SRC_FILES := ioshark_bench.c ioshark_bench_subr.c ioshark_bench_mmap.c
+LOCAL_CFLAGS := -g -O2 -Wall  -Werror
+LOCAL_MODULE := ioshark_bench
+include $(BUILD_EXECUTABLE)
+
+include $(CLEAR_VARS)
+LOCAL_SRC_FILES := compile_ioshark.c compile_ioshark_subr.c
+LOCAL_CFLAGS := -g -O2 -Wall -Werror -D_GNU_SOURCE
+LOCAL_MODULE := compile_ioshark
+include $(BUILD_HOST_EXECUTABLE)
+
+
+
diff --git a/ioshark/README b/ioshark/README
new file mode 100644
index 00000000..3afbe76e
--- /dev/null
+++ b/ioshark/README
@@ -0,0 +1,26 @@
+IOshark is a repeatable application workload storage benchmark. You
+can find more documentation on IOshark at :
+https://docs.google.com/a/google.com/document/d/1Bhq7iNPVc_JzwRrkmZqcPjMvWgpHX0r3Ncq-ZsRNOBA/edit?usp=sharing
+
+The short summary of what IOshark is : IOshark has 2 components, one
+is a strace compiler that takes straces fed into it and compiles this
+into bytecodes (stored in *.wl files). The compiler runs on a Linux
+host. The second component (which runs on the device) is the tester
+that takes as input the bytecode files (*.wl files) and executes them
+on the device.
+
+How to Run :
+----------
+- First collect straces and compile these into bytecodes. The wrapper
+script provided (collect-straces.sh) collects straces, ships them to
+the host where the script runs, compiles and packages up the bytecode
+files into a wl.tar file.
+- Ship the wl.tar file and the iostark_bench binaries to the target
+device (on /data/local/tmp say). Explode the tarfile.
+- Run the tester. "ioshark_bench *.wl" runs the test with default
+options. Supported ioshark_bench options :
+-d : Preserve the delays between successive filesystem syscalls as
+seen in the original straces.
+-n <N> : Run for N iterations
+-t <N> : Limit to N threads. By default (without this option), IOshark
+will launch as many threads as there are input files, so 1 thread/file.
\ No newline at end of file
diff --git a/ioshark/collect-straces.sh b/ioshark/collect-straces.sh
new file mode 100755
index 00000000..4577e736
--- /dev/null
+++ b/ioshark/collect-straces.sh
@@ -0,0 +1,94 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#!/bin/sh
+
+# When signal is received, the stracer will get killed
+# Call this (just to make sure anyway)
+kill_strace() {
+    ps_line=`ps -ef | grep strace | grep adb `
+    if [ $? == 0 ]; then
+        echo Killing `echo $ps_line | awk '{s = ""; for (i=8; i <= NF ; i++) s = s \
+$i " "; print s}' `
+        kill `echo $ps_line | awk '{print $2}' `
+    fi
+}
+
+catch_sigint()
+{
+    echo "signal INT received, killing streaming trace capture"
+    kill_strace
+}
+
+compile_tracefiles()
+{
+    for i in trace.*
+    do
+	if [ $i != trace.begin ] && [ $i != trace.tar ];
+	then
+	    egrep '\/system\/|\/data\/|\/vendor\/' $i > bar
+# parse out /sys/devices/system/...
+	    egrep -v '\/sys\/devices\/system\/' bar > bar0
+	    mv bar0 bar
+	    fgrep -v '= -1'	bar > foo
+	    mv foo bar
+	    if [ -s bar ]
+	    then
+		echo parsing $i
+		compile_ioshark bar $i.wl
+		rm -f bar
+	    else
+		rm -f $i bar
+	    fi
+	fi
+    done
+}
+
+# main() starts here
+
+adb root && adb wait-for-device
+
+adb shell 'ps -ef' | grep zygote > zygote_pids
+fgrep -v grep zygote_pids > bar
+mv bar zygote_pids
+pid1=`grep -w zygote zygote_pids | awk '{print $2}' `
+pid2=`grep -w zygote64 zygote_pids | awk '{print $2}' `
+rm -f zygote_pids
+
+trap 'catch_sigint' INT
+
+echo "^C this script once you finish running your test"
+
+adb shell "date +%s > /data/local/tmp/trace.begin ; strace -p $pid1,$pid2 -o /data/local/tmp/trace -q -qq -f -ff -y -ttt -e trace=mmap2,read,write,pread64,pwrite64,fsync,fdatasync,openat,close,lseek,_llseek"
+
+# Remove any remnant tracefiles first
+rm -f trace.*
+
+# Get the tracefiles from the device
+adb shell 'cd /data/local/tmp ; tar cvf trace.tar trace.*'
+adb pull /data/local/tmp/trace.tar
+
+# Extract the tracefiles from the device
+rm -f *.wl
+tar xf trace.tar
+
+# Compile the tracefiles
+compile_tracefiles
+
+# tar up the .wl files just created
+tar zcf wl.tgz *.wl
+
+
diff --git a/ioshark/compile_ioshark.c b/ioshark/compile_ioshark.c
new file mode 100644
index 00000000..609d2bc2
--- /dev/null
+++ b/ioshark/compile_ioshark.c
@@ -0,0 +1,578 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <stdio.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include <linux/fs.h>
+#include <fcntl.h>
+#include <ctype.h>
+#include "ioshark.h"
+#include "compile_ioshark.h"
+
+char *progname;
+
+char in_buf[2048];
+
+struct flags_map_s {
+	char *flag_str;
+	int flag;
+};
+
+#define ARRAY_SIZE(a)	(sizeof(a) / sizeof(a[0]))
+
+struct flags_map_s open_flags_map[] = {
+	{ "O_RDONLY", O_RDONLY },
+	{ "O_WRONLY", O_WRONLY },
+	{ "O_RDWR", O_RDWR },
+	{ "O_CREAT", O_CREAT },
+	{ "O_SYNC", O_SYNC },
+	{ "O_TRUNC", O_TRUNC },
+	{ "O_EXCL", O_EXCL },
+	{ "O_APPEND", O_APPEND },
+	{ "O_NOATIME", O_NOATIME },
+	{ "O_ASYNC", O_ASYNC },
+	{ "O_CLOEXEC", O_CLOEXEC },
+	{ "O_DIRECT", O_DIRECT },
+	{ "O_DIRECTORY", O_DIRECTORY },
+	{ "O_LARGEFILE", O_LARGEFILE },
+	{ "O_NOCTTY", O_NOCTTY },
+	{ "O_NOFOLLOW", O_NOFOLLOW },
+	{ "O_NONBLOCK", O_NONBLOCK },
+	{ "O_NDELAY", O_NDELAY },
+	{ "O_PATH", O_PATH }
+};
+
+struct flags_map_s lseek_action_map[] = {
+	{ "SEEK_SET", SEEK_SET },
+	{ "SEEK_CUR", SEEK_CUR },
+	{ "SEEK_END", SEEK_END }
+};
+
+struct flags_map_s fileop_map[] = {
+	{ "lseek", IOSHARK_LSEEK },
+	{ "_llseek", IOSHARK_LLSEEK },
+	{ "pread64", IOSHARK_PREAD64 },
+	{ "pwrite64", IOSHARK_PWRITE64 },
+	{ "read", IOSHARK_READ },
+	{ "write", IOSHARK_WRITE },
+	{ "mmap", IOSHARK_MMAP },
+	{ "mmap2", IOSHARK_MMAP2 },
+	{ "openat", IOSHARK_OPEN },
+	{ "fsync", IOSHARK_FSYNC },
+	{ "fdatasync", IOSHARK_FDATASYNC },
+	{ "close", IOSHARK_CLOSE }
+};
+
+struct in_mem_file_op {
+	struct ioshark_file_operation disk_file_op;
+	struct in_mem_file_op *next;
+};
+
+struct in_mem_file_op *in_mem_file_op_head = NULL, *in_mem_file_op_tail = NULL;
+
+void usage(void)
+{
+	fprintf(stderr, "%s in_file out_file\n", progname);
+}
+
+u_int64_t
+get_delta_ts(char *buf, struct timeval *start_tv)
+{
+	struct timeval op_tv, tv_res;
+
+	sscanf(buf, "%lu.%lu", &op_tv.tv_sec, &op_tv.tv_usec);
+	timersub(&op_tv, start_tv, &tv_res);
+	return (tv_res.tv_usec + (tv_res.tv_sec * 1000000));
+}
+
+void
+get_pathname(char *buf, char *pathname, enum file_op file_op)
+{
+	char *s, *s2, save;
+
+	if (file_op == IOSHARK_OPEN)
+		s = strchr(buf, '"');
+	else
+		s = strchr(buf, '<');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 1;
+	if (file_op == IOSHARK_OPEN)
+		s2 = strchr(s, '"');
+	else
+		s2 = strchr(s, '>');
+	if (s2 == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	save = *s2;
+	*s2 = '\0';
+	strcpy(pathname, s);
+	*s2 = save;
+}
+
+void
+get_syscall(char *buf, char *syscall)
+{
+	char *s, *s2;
+
+	s = strchr(buf, ' ');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 1;
+	s2 = strchr(s, '(');
+	if (s2 == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	*s2 = '\0';
+	strcpy(syscall, s);
+	*s2 = '(';
+}
+
+void
+get_mmap_offset_len_prot(char *buf, int *prot, off_t *offset, u_int64_t *len)
+{
+	char *s, *s1;
+	int i;
+	char protstr[128];
+
+	s = strchr(buf, ',');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 2;
+	sscanf(s, "%lu", len);
+	s1 = strchr(s, ',');
+	if (s1 == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s1 += 2;
+	s = strchr(s1, ',');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	*s = '\0';
+	strcpy(protstr, s1);
+	*prot = 0;
+	if (strstr(protstr, "PROT_READ"))
+		*prot |= IOSHARK_PROT_READ;
+	if (strstr(protstr, "PROT_WRITE"))
+		*prot |= IOSHARK_PROT_WRITE;
+	*s = ',';
+	s += 2;
+	for (i = 0 ; i < 2 ; i++) {
+		s = strchr(s, ',');
+		if (s == NULL) {
+			fprintf(stderr, "%s: Malformed line: %s\n",
+				__func__, buf);
+			exit(EXIT_FAILURE);
+		}
+		s += 2;
+	}
+	sscanf(s, "%lx", offset);
+}
+
+void
+get_lseek_offset_action(char *buf, enum file_op op,
+			off_t *offset, char *action)
+{
+	char *s, *s2;
+
+	s = strchr(buf, ',');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 2;
+	sscanf(s, "%lu", offset);
+	s = strchr(s, ',');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 2;
+	if (op == IOSHARK_LLSEEK) {
+		s = strchr(s, ',');
+		if (s == NULL) {
+			fprintf(stderr, "%s: Malformed line: %s\n",
+				__func__, buf);
+			exit(EXIT_FAILURE);
+		}
+		s += 2;
+	}
+	s2 = strchr(s, ')');
+	if (s2 == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	*s2 = '\0';
+	strcpy(action, s);
+	*s2 = ')';
+}
+
+void
+get_rw_len(char *buf,
+	   u_int64_t *len)
+{
+	char *s_len;
+
+	s_len = strrchr(buf, ',');
+	if (s_len == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	sscanf(s_len + 2, "%lu", len);
+}
+
+void
+get_prw64_offset_len(char *buf,
+		     off_t *offset,
+		     u_int64_t *len)
+{
+	char *s_offset, *s_len;
+
+	s_offset = strrchr(buf, ',');
+	if (s_offset == NULL) {
+		fprintf(stderr, "%s: Malformed line 1: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	*s_offset = '\0';
+	s_len = strrchr(buf, ',');
+	if (s_len == NULL) {
+		fprintf(stderr, "%s: Malformed line 2: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	*s_offset = ',';
+	sscanf(s_len + 2, "%lu", len);
+	sscanf(s_offset + 2, "%lu", offset);
+}
+
+void
+get_openat_flags_mode(char *buf, char *flags, mode_t *mode)
+{
+	char *s, *s2, lookfor;
+
+	s = strchr(buf, ',');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 2;
+	s = strchr(s, ',');
+	if (s == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	s += 2;
+	if (strstr(s, "O_CREAT") == NULL)
+		lookfor = ')';
+	else
+		lookfor = ',';
+	s2 = strchr(s, lookfor);
+	if (s2 == NULL) {
+		fprintf(stderr, "%s: Malformed line: %s\n",
+			__func__, buf);
+		exit(EXIT_FAILURE);
+	}
+	*s2 = '\0';
+	strcpy(flags, s);
+	*s2 = lookfor;
+	if (strstr(s, "O_CREAT") != NULL) {
+		s = s2 + 2;
+		s2 = strchr(s, ')');
+		if (s2 == NULL) {
+			fprintf(stderr, "%s: Malformed line: %s\n",
+				__func__, buf);
+			exit(EXIT_FAILURE);
+		}
+		*s2 = '\0';
+		sscanf(s, "%o", mode);
+		*s2 = ')';
+	}
+}
+
+int
+lookup_map(char *s, struct flags_map_s *flags_map, int maplen)
+{
+	int found = 0, flag = 0;
+	int i;
+
+	while (isspace(*s))
+		s++;
+	for (i = 0 ; i < maplen ; i++) {
+		if (strcmp(flags_map[i].flag_str, s) == 0) {
+			flag = flags_map[i].flag;
+			found = 1;
+			break;
+		}
+	}
+	if (found == 0) {
+		fprintf(stderr, "%s: Unknown syscall %s\n",
+			__func__, s);
+		exit(1);
+	}
+	return flag;
+}
+
+int
+map_open_flags(char *s)
+{
+	int flags = 0;
+	char *s1;
+
+	while ((s1 = strchr(s, '|'))) {
+		*s1 = '\0';
+		flags |= lookup_map(s, open_flags_map,
+				    ARRAY_SIZE(open_flags_map));
+		*s1 = '|';
+		s = s1 + 1;
+	}
+	/* Last option */
+	flags |= lookup_map(s, open_flags_map,
+			    ARRAY_SIZE(open_flags_map));
+	return flags;
+}
+
+int
+map_lseek_action(char *s)
+{
+	int flags = 0;
+	char *s1;
+
+	while ((s1 = strchr(s, '|'))) {
+		*s1 = '\0';
+		flags |= lookup_map(s, lseek_action_map,
+				    ARRAY_SIZE(lseek_action_map));
+		*s1 = '|';
+		s = s1 + 1;
+	}
+	/* Last option */
+	flags |= lookup_map(s, lseek_action_map,
+			    ARRAY_SIZE(lseek_action_map));
+	return flags;
+}
+
+enum file_op
+map_syscall(char *syscall)
+{
+	return lookup_map(syscall, fileop_map,
+			  ARRAY_SIZE(fileop_map));
+}
+
+void
+get_start_time(struct timeval *tv)
+{
+	FILE *fp;
+
+	tv->tv_usec = 0;
+	fp = fopen("trace.begin", "r");
+	if (fp == NULL) {
+		fprintf(stderr, "%s Can't open trace.begin\n",
+			__func__);
+		exit(EXIT_FAILURE);
+	}
+	if (fscanf(fp, "%lu", &tv->tv_sec) != 1) {
+		fprintf(stderr, "%s Can't read seconds from trace.begin\n",
+			__func__);
+		exit(EXIT_FAILURE);
+	}
+}
+
+/*
+ * For each tracefile, we first create in-memory structures, then once
+ * we've processed each tracefile completely, we write out the in-memory
+ * structures and free them.
+ */
+int main(int argc, char **argv)
+{
+	FILE *fp;
+	char path[512];
+	char syscall[512];
+	char lseek_action_str[512];
+	char *s;
+	char open_flags_str[64];
+	void *db_node;
+	int num_io_operations = 0;
+	struct ioshark_header header;
+	struct ioshark_file_operation *disk_file_op;
+	struct in_mem_file_op *in_mem_fop;
+	struct stat st;
+	char *infile, *outfile;
+	struct timeval start_time;
+
+	progname = argv[0];
+	if (argc != 3) {
+		usage();
+		exit(EXIT_FAILURE);
+	}
+	infile = argv[1];
+	outfile = argv[2];
+	if (stat(infile, &st) < 0) {
+		fprintf(stderr, "%s Can't stat %s\n",
+			progname, infile);
+		exit(EXIT_FAILURE);
+	}
+	if (st.st_size == 0) {
+		fprintf(stderr, "%s Empty file %s\n",
+			progname, infile);
+		exit(EXIT_FAILURE);
+	}
+	get_start_time(&start_time);
+	fp = fopen(infile, "r");
+	if (fp == NULL) {
+		fprintf(stderr, "%s Can't open %s\n",
+			progname, infile);
+		exit(EXIT_FAILURE);
+	}
+	while (fgets(in_buf, 2048, fp)) {
+		s = in_buf;
+		while (isspace(*s))
+			s++;
+		in_mem_fop = malloc(sizeof(struct in_mem_file_op));
+		disk_file_op = &in_mem_fop->disk_file_op;
+		disk_file_op->delta_us = get_delta_ts(s, &start_time);
+		get_syscall(s, syscall);
+		disk_file_op->file_op = map_syscall(syscall);
+		get_pathname(s, path, disk_file_op->file_op);
+		db_node = files_db_add(path);
+		disk_file_op->fileno = files_db_get_fileno(db_node);
+		switch (disk_file_op->file_op) {
+		case IOSHARK_LLSEEK:
+		case IOSHARK_LSEEK:
+			get_lseek_offset_action(s,
+					disk_file_op->file_op,
+					&disk_file_op->lseek_offset,
+					lseek_action_str);
+			disk_file_op->lseek_action =
+				map_lseek_action(lseek_action_str);
+			if (disk_file_op->lseek_action == SEEK_SET)
+				files_db_update_size(db_node,
+						     disk_file_op->lseek_offset);
+			break;
+		case IOSHARK_PREAD64:
+		case IOSHARK_PWRITE64:
+			get_prw64_offset_len(s,
+					     &disk_file_op->prw_offset,
+					     &disk_file_op->prw_len);
+			files_db_update_size(db_node,
+					     disk_file_op->prw_offset +
+					     disk_file_op->prw_len);
+			break;
+		case IOSHARK_READ:
+		case IOSHARK_WRITE:
+			get_rw_len(s, &disk_file_op->rw_len);
+			files_db_add_to_size(db_node,
+					     disk_file_op->rw_len);
+			break;
+		case IOSHARK_MMAP:
+		case IOSHARK_MMAP2:
+			get_mmap_offset_len_prot(s,
+				    &disk_file_op->mmap_prot,
+				    &disk_file_op->mmap_offset,
+				    &disk_file_op->mmap_len);
+			files_db_update_size(db_node,
+				     disk_file_op->mmap_offset +
+				     disk_file_op->mmap_len);
+			break;
+		case IOSHARK_OPEN:
+			disk_file_op->open_mode = 0;
+			get_openat_flags_mode(s, open_flags_str,
+				      &disk_file_op->open_mode);
+			disk_file_op->open_flags =
+				map_open_flags(open_flags_str);
+			break;
+		case IOSHARK_FSYNC:
+		case IOSHARK_FDATASYNC:
+			break;
+		case IOSHARK_CLOSE:
+			break;
+		default:
+			break;
+		}
+		/* Chain at the end */
+		num_io_operations++;
+		in_mem_fop->next = NULL;
+		if (in_mem_file_op_head == NULL) {
+			in_mem_file_op_tail = in_mem_fop;
+			in_mem_file_op_head = in_mem_fop;
+		} else {
+			in_mem_file_op_tail->next = in_mem_fop;
+			in_mem_file_op_tail = in_mem_fop;
+		}
+	}
+	fclose(fp);
+	/*
+	 * Now we can write everything out to the output tracefile.
+	 */
+	fp = fopen(outfile, "w+");
+	if (fp == NULL) {
+		fprintf(stderr, "%s Can't open trace.outfile\n",
+			progname);
+		exit(EXIT_FAILURE);
+	}
+	header.num_io_operations = num_io_operations;
+	header.num_files = files_db_get_total_obj();
+	if (fwrite(&header, sizeof(struct ioshark_header), 1, fp) != 1) {
+		fprintf(stderr, "%s Write error trace.outfile\n",
+			progname);
+		exit(EXIT_FAILURE);
+	}
+	files_db_write_objects(fp);
+	while (in_mem_file_op_head != NULL) {
+		struct in_mem_file_op *temp;
+
+		disk_file_op = &in_mem_file_op_head->disk_file_op;
+		if (fwrite(disk_file_op,
+			   sizeof(struct ioshark_file_operation), 1, fp) != 1) {
+			fprintf(stderr, "%s Write error trace.outfile\n",
+				progname);
+			exit(EXIT_FAILURE);
+		}
+		temp = in_mem_file_op_head;
+		in_mem_file_op_head = in_mem_file_op_head->next;
+		free(temp);
+	}
+}
diff --git a/ioshark/compile_ioshark.h b/ioshark/compile_ioshark.h
new file mode 100644
index 00000000..13095bf6
--- /dev/null
+++ b/ioshark/compile_ioshark.h
@@ -0,0 +1,78 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#define FILE_DB_HASHSIZE	8192
+
+struct files_db_s {
+	char *filename;
+	int fileno;
+	struct files_db_s *next;
+	size_t	size;
+};
+
+/* Lifted from Wikipedia Jenkins Hash function page */
+static inline u_int32_t
+jenkins_one_at_a_time_hash(char *key, size_t len)
+{
+	u_int32_t hash, i;
+
+	for(hash = i = 0; i < len; ++i) {
+		hash += key[i];
+		hash += (hash << 10);
+		hash ^= (hash >> 6);
+	}
+	hash += (hash << 3);
+	hash ^= (hash >> 11);
+	hash += (hash << 15);
+	return hash;
+}
+
+static inline void
+files_db_update_size(void *node, u_int64_t new_size)
+{
+	struct files_db_s *db_node = (struct files_db_s *)node;
+
+	if (db_node->size < new_size)
+		db_node->size = new_size;
+}
+
+static inline void
+files_db_add_to_size(void *node, u_int64_t size_incr)
+{
+	((struct files_db_s *)node)->size += size_incr;
+}
+
+static inline int
+files_db_get_fileno(void *node)
+{
+	return (((struct files_db_s *)node)->fileno);
+}
+
+static inline char *
+files_db_get_filename(void *node)
+{
+	return (((struct files_db_s *)node)->filename);
+}
+
+
+void *files_db_create_handle(void);
+void files_db_write_objects(FILE *fp);
+void *files_db_add(char *filename);
+void *files_db_lookup(char *filename);
+int files_db_get_total_obj(void);
+
+
+
diff --git a/ioshark/compile_ioshark_subr.c b/ioshark/compile_ioshark_subr.c
new file mode 100644
index 00000000..e2b6b891
--- /dev/null
+++ b/ioshark/compile_ioshark_subr.c
@@ -0,0 +1,102 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <fcntl.h>
+#include <stdio.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include "ioshark.h"
+#include "compile_ioshark.h"
+
+extern char *progname;
+
+static struct files_db_s *files_db_buckets[FILE_DB_HASHSIZE];
+static int current_fileno = 1;
+static int num_objects = 0;
+
+void
+files_db_write_objects(FILE *fp)
+{
+	int i;
+	struct ioshark_file_state st;
+
+	for (i = 0 ; i < FILE_DB_HASHSIZE ; i++) {
+		struct files_db_s *db_node, *s;
+
+		db_node = files_db_buckets[i];
+		while (db_node != NULL) {
+			st.fileno = db_node->fileno;
+			st.size = db_node->size;
+			if (fwrite(&st, sizeof(st), 1, fp) != 1) {
+				fprintf(stderr, "%s Write error trace.outfile\n",
+					progname);
+				exit(EXIT_FAILURE);
+			}
+			s = db_node;
+			db_node = db_node->next;
+			free(s->filename);
+			free(s);
+		}
+	}
+}
+
+void *files_db_lookup(char *pathname)
+{
+	u_int32_t hash;
+	struct files_db_s *db_node;
+
+	hash = jenkins_one_at_a_time_hash(pathname, strlen(pathname));
+	hash %= FILE_DB_HASHSIZE;
+	db_node = files_db_buckets[hash];
+	while (db_node != NULL) {
+		if (strcmp(db_node->filename, pathname) == 0)
+			break;
+		db_node = db_node->next;
+	}
+	return db_node;
+}
+
+void *files_db_add(char *filename)
+{
+	u_int32_t hash;
+	struct files_db_s *db_node;
+
+	if ((db_node = files_db_lookup(filename)))
+		return db_node;
+	hash = jenkins_one_at_a_time_hash(filename, strlen(filename));
+	hash %= FILE_DB_HASHSIZE;
+	db_node = malloc(sizeof(struct files_db_s));
+	db_node->filename = strdup(filename);
+	db_node->fileno = current_fileno++;
+	db_node->next = files_db_buckets[hash];
+	db_node->size = 0;
+	files_db_buckets[hash] = db_node;
+	num_objects++;
+	return db_node;
+}
+
+int
+files_db_get_total_obj(void)
+{
+	return num_objects;
+}
+
+
+
diff --git a/ioshark/ioshark.h b/ioshark/ioshark.h
new file mode 100644
index 00000000..f0a715f7
--- /dev/null
+++ b/ioshark/ioshark.h
@@ -0,0 +1,108 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/*
+ * Format of the parsed workload files.
+ * 1) Header
+ * 2) Table of the entries, each entry describes 1 file
+ * 3) Table of IO operations to perform on the files
+ */
+
+/*
+ * The parsed workload file starts off with the header, which
+ * contains the count of the total # of files that are operated on.
+ * and the total number of IO operations.
+ */
+struct ioshark_header {
+	int	num_files;
+	int	num_io_operations;
+};
+
+/*
+ * After the header, we have a table of #files entries. Each entry
+ * in this table describes 1 file, indexed by fileno and with the
+ * specified size.
+ * Before the tests starts, these files are pre-created.
+ */
+struct ioshark_file_state {
+	int	fileno;	/* 1..num_files, with files name ioshark.<fileno> */
+	size_t	size;
+};
+
+enum file_op {
+	IOSHARK_LSEEK = 0,
+	IOSHARK_LLSEEK,
+	IOSHARK_PREAD64,
+	IOSHARK_PWRITE64,
+	IOSHARK_READ,
+	IOSHARK_WRITE,
+	IOSHARK_MMAP,
+	IOSHARK_MMAP2,
+	IOSHARK_OPEN,
+	IOSHARK_FSYNC,
+	IOSHARK_FDATASYNC,
+	IOSHARK_CLOSE,
+	IOSHARK_MAPPED_PREAD,
+	IOSHARK_MAPPED_PWRITE,
+	IOSHARK_MAX_FILE_OP
+};
+
+/* mmap prot flags */
+#define IOSHARK_PROT_READ	0x1
+#define IOSHARK_PROT_WRITE	0x2
+
+/*
+ * Next we have the table of IO operatiosn to perform. Each
+ * IO operation is described by this entry.
+ */
+struct ioshark_file_operation {
+	/* delta us between previous file op and this */
+	u_int64_t		delta_us;
+	enum file_op		file_op;
+	int			fileno;
+	union {
+		struct lseek_args {
+#define lseek_offset	u.lseek_a.offset
+#define lseek_action	u.lseek_a.action
+			off_t	offset;
+			int action;
+		} lseek_a;
+		struct prw_args {
+#define prw_offset	u.prw_a.offset
+#define prw_len		u.prw_a.len
+			off_t	offset;
+			size_t	len;
+		} prw_a;
+#define rw_len		u.rw_a.len
+		struct rw_args {
+			size_t	len;
+		} rw_a;
+#define mmap_offset	u.mmap_a.offset
+#define mmap_len	u.mmap_a.len
+#define mmap_prot	u.mmap_a.prot
+		struct mmap_args {
+			off_t	offset;
+			size_t	len;
+			int	prot;
+	} mmap_a;
+#define open_flags	u.open_a.flags
+#define open_mode	u.open_a.mode
+		struct open_args {
+			int	flags;
+			mode_t	mode;
+		} open_a;
+	} u;
+};
diff --git a/ioshark/ioshark_bench.c b/ioshark/ioshark_bench.c
new file mode 100644
index 00000000..9bbfcae1
--- /dev/null
+++ b/ioshark/ioshark_bench.c
@@ -0,0 +1,745 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <stdio.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <stdlib.h>
+#include <string.h>
+#include <pthread.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include <linux/fs.h>
+#include <fcntl.h>
+#include <string.h>
+#include <assert.h>
+#include <pthread.h>
+#include <sys/statfs.h>
+#include "ioshark.h"
+#define IOSHARK_MAIN
+#include "ioshark_bench.h"
+
+#define UNUSED_PARAM(X)		((void)X)
+
+char *progname;
+
+#define MAX_INPUT_FILES		1024
+#define MAX_THREADS		1024
+
+struct thread_state_s {
+	char *filename;
+	FILE *fp;
+	int num_files;
+	void *db_handle;
+};
+
+struct thread_state_s thread_state[MAX_INPUT_FILES];
+int num_input_files = 0;
+int next_input_file;
+
+pthread_t tid[MAX_THREADS];
+
+/*
+ * Global options
+ */
+int do_delay = 0;
+
+#if 0
+static long gettid()
+{
+        return syscall(__NR_gettid);
+}
+#endif
+
+void usage()
+{
+	fprintf(stderr, "%s [-d preserve_delays] [-n num_iterations] [-t num_threads] <list of parsed input files>\n",
+		progname);
+	exit(EXIT_FAILURE);
+}
+
+pthread_mutex_t time_mutex = PTHREAD_MUTEX_INITIALIZER;
+pthread_mutex_t stats_mutex = PTHREAD_MUTEX_INITIALIZER;
+pthread_mutex_t work_mutex = PTHREAD_MUTEX_INITIALIZER;
+struct timeval aggregate_file_create_time;
+struct timeval aggregate_file_remove_time;
+struct timeval aggregate_IO_time;
+struct timeval aggregate_delay_time;
+
+u_int64_t aggr_op_counts[IOSHARK_MAX_FILE_OP];
+struct rw_bytes_s aggr_io_rw_bytes;
+struct rw_bytes_s aggr_create_rw_bytes;
+
+static void
+update_time(struct timeval *aggr_time,
+	    struct timeval *delta_time)
+{
+	struct timeval tmp;
+
+	pthread_mutex_lock(&time_mutex);
+	timeradd(aggr_time, delta_time, &tmp);
+	*aggr_time = tmp;
+	pthread_mutex_unlock(&time_mutex);
+}
+
+static void
+update_op_counts(u_int64_t *op_counts)
+{
+	int i;
+
+	pthread_mutex_lock(&stats_mutex);
+	for (i = IOSHARK_LSEEK ; i < IOSHARK_MAX_FILE_OP ; i++)
+		aggr_op_counts[i] += op_counts[i];
+	pthread_mutex_unlock(&stats_mutex);
+}
+
+static void
+update_byte_counts(struct rw_bytes_s *dest, struct rw_bytes_s *delta)
+{
+	pthread_mutex_lock(&stats_mutex);
+	dest->bytes_read += delta->bytes_read;
+	dest->bytes_written += delta->bytes_written;
+	pthread_mutex_unlock(&stats_mutex);
+}
+
+static int work_next_file;
+static int work_num_files;
+
+void
+init_work(int next_file, int num_files)
+{
+	pthread_mutex_lock(&work_mutex);
+	work_next_file = next_file;
+	work_num_files = work_next_file + num_files;
+	pthread_mutex_unlock(&work_mutex);
+}
+
+/* Dole out the next file to work on to the thread */
+static struct thread_state_s *
+get_work()
+{
+	struct thread_state_s *work = NULL;
+
+	pthread_mutex_lock(&work_mutex);
+	if (work_next_file < work_num_files)
+		work = &thread_state[work_next_file++];
+	pthread_mutex_unlock(&work_mutex);
+	return work;
+}
+
+static void
+create_files(struct thread_state_s *state)
+{
+	struct timeval file_create_time;
+	int i;
+	struct ioshark_file_state file_state;
+	char path[512];
+	void *db_node;
+	struct rw_bytes_s rw_bytes;
+
+	timerclear(&file_create_time);
+	memset(&rw_bytes, 0, sizeof(struct rw_bytes_s));
+	for (i = 0 ; i < state->num_files ; i++) {
+		if (fread(&file_state, sizeof(struct ioshark_file_state),
+			  1, state->fp) != 1) {
+			fprintf(stderr, "%s read error tracefile\n",
+				progname);
+			exit(EXIT_FAILURE);
+		}
+		sprintf(path, "file.%d.%d",
+			(int)(state - thread_state),
+			file_state.fileno);
+		create_file(path, file_state.size,
+			    &file_create_time, &rw_bytes);
+		db_node = files_db_add_byfileno(state->db_handle,
+						file_state.fileno);
+		files_db_update_size(db_node, file_state.size);
+		files_db_update_filename(db_node, path);
+	}
+	update_byte_counts(&aggr_create_rw_bytes, &rw_bytes);
+	update_time(&aggregate_file_create_time, &file_create_time);
+}
+
+static void
+do_one_io(void *db_node,
+	  struct ioshark_file_operation *file_op,
+	  struct timeval *io_time, u_int64_t *op_counts,
+	  struct rw_bytes_s *rw_bytes,
+	  char **bufp, int *buflen)
+{
+	struct timeval start;
+
+	assert(file_op->file_op < IOSHARK_MAX_FILE_OP);
+	op_counts[file_op->file_op]++;
+	switch (file_op->file_op) {
+	int ret;
+	char *p;
+	int fd;
+
+	case IOSHARK_LSEEK:
+	case IOSHARK_LLSEEK:
+	(void)gettimeofday(&start, (struct timezone *)NULL);
+		ret = lseek(files_db_get_fd(db_node),
+			    file_op->lseek_offset,
+			    file_op->lseek_action);
+		update_delta_time(&start, io_time);
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: lseek(%s %lu %d) returned error %d\n",
+				progname, files_db_get_filename(db_node),
+				file_op->lseek_offset,
+				file_op->lseek_action, errno);
+			exit(EXIT_FAILURE);
+		}
+		break;
+	case IOSHARK_PREAD64:
+		p = get_buf(bufp, buflen, file_op->prw_len, 0);
+		(void)gettimeofday(&start,
+				   (struct timezone *)NULL);
+		ret = pread(files_db_get_fd(db_node), p,
+			    file_op->prw_len, file_op->prw_offset);
+		update_delta_time(&start, io_time);
+		rw_bytes->bytes_read += file_op->prw_len;
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: pread(%s %lu %lu) error %d\n",
+				progname,
+				files_db_get_filename(db_node),
+				file_op->prw_len,
+				file_op->prw_offset, errno);
+			exit(EXIT_FAILURE);
+		}
+		break;
+	case IOSHARK_PWRITE64:
+		p = get_buf(bufp, buflen, file_op->prw_len, 1);
+		(void)gettimeofday(&start,
+				   (struct timezone *)NULL);
+		ret = pwrite(files_db_get_fd(db_node), p,
+			     file_op->prw_len, file_op->prw_offset);
+		update_delta_time(&start, io_time);
+		rw_bytes->bytes_written += file_op->prw_len;
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: pwrite(%s %lu %lu) error %d\n",
+				progname,
+				files_db_get_filename(db_node),
+				file_op->prw_len,
+				file_op->prw_offset, errno);
+			exit(EXIT_FAILURE);
+		}
+		break;
+	case IOSHARK_READ:
+		p = get_buf(bufp, buflen, file_op->rw_len, 0);
+		(void)gettimeofday(&start,
+				   (struct timezone *)NULL);
+		ret = read(files_db_get_fd(db_node), p,
+			   file_op->rw_len);
+		update_delta_time(&start, io_time);
+		rw_bytes->bytes_read += file_op->rw_len;
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: read(%s %lu) error %d\n",
+				progname,
+				files_db_get_filename(db_node),
+				file_op->rw_len,
+				errno);
+			exit(EXIT_FAILURE);
+		}
+		break;
+	case IOSHARK_WRITE:
+		p = get_buf(bufp, buflen, file_op->rw_len, 1);
+		(void)gettimeofday(&start,
+				   (struct timezone *)NULL);
+		ret = write(files_db_get_fd(db_node), p,
+			    file_op->rw_len);
+		update_delta_time(&start, io_time);
+		rw_bytes->bytes_written += file_op->rw_len;
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: write(%s %lu) error %d\n",
+				progname,
+				files_db_get_filename(db_node),
+				file_op->rw_len,
+				errno);
+			exit(EXIT_FAILURE);
+		}
+		break;
+	case IOSHARK_MMAP:
+	case IOSHARK_MMAP2:
+		ioshark_handle_mmap(db_node, file_op,
+				    bufp, buflen, op_counts,
+				    rw_bytes, io_time);
+		break;
+	case IOSHARK_OPEN:
+		if (file_op->open_flags & O_CREAT) {
+			(void)gettimeofday(&start,
+					   (struct timezone *)NULL);
+			fd = open(files_db_get_filename(db_node),
+				  file_op->open_flags,
+				  file_op->open_mode);
+			if (fd < 0) {
+				/*
+				 * EEXIST error acceptable, others are fatal.
+				 * Although we failed to O_CREAT the file (O_EXCL)
+				 * We will force an open of the file before any
+				 * IO.
+				 */
+				if (errno == EEXIST) {
+					update_delta_time(&start, io_time);
+					return;
+				} else {
+					fprintf(stderr,
+						"%s: O_CREAT open(%s %x %o) error %d\n",
+						progname,
+						files_db_get_filename(db_node),
+						file_op->open_flags,
+						file_op->open_mode, errno);
+					exit(EXIT_FAILURE);
+				}
+			} else
+				update_delta_time(&start, io_time);
+		} else {
+			(void)gettimeofday(&start,
+					   (struct timezone *)NULL);
+			fd = open(files_db_get_filename(db_node),
+				  file_op->open_flags);
+			update_delta_time(&start, io_time);
+			if (fd < 0) {
+				if (file_op->open_flags & O_DIRECTORY) {
+					/* O_DIRECTORY open()s should fail */
+					return;
+				} else {
+					fprintf(stderr,
+						"%s: open(%s %x) error %d\n",
+						progname,
+						files_db_get_filename(db_node),
+						file_op->open_flags,
+						errno);
+					exit(EXIT_FAILURE);
+				}
+			}
+		}
+		files_db_close_fd(db_node);
+		files_db_update_fd(db_node, fd);
+		break;
+	case IOSHARK_FSYNC:
+	case IOSHARK_FDATASYNC:
+		if (file_op->file_op == IOSHARK_FSYNC) {
+			(void)gettimeofday(&start,
+					   (struct timezone *)NULL);
+			ret = fsync(files_db_get_fd(db_node));
+			update_delta_time(&start, io_time);
+			if (ret < 0) {
+				fprintf(stderr,
+					"%s: fsync(%s) error %d\n",
+					progname,
+					files_db_get_filename(db_node),
+					errno);
+				exit(EXIT_FAILURE);
+			}
+		} else {
+			(void)gettimeofday(&start,
+					   (struct timezone *)NULL);
+			ret = fdatasync(files_db_get_fd(db_node));
+			update_delta_time(&start, io_time);
+			if (ret < 0) {
+				fprintf(stderr,
+					"%s: fdatasync(%s) error %d\n",
+					progname,
+					files_db_get_filename(db_node),
+					errno);
+				exit(EXIT_FAILURE);
+			}
+		}
+		break;
+	case IOSHARK_CLOSE:
+		(void)gettimeofday(&start,
+				   (struct timezone *)NULL);
+		ret = close(files_db_get_fd(db_node));
+		update_delta_time(&start, io_time);
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: close(%s) error %d\n",
+				progname,
+				files_db_get_filename(db_node), errno);
+			exit(EXIT_FAILURE);
+		}
+		files_db_update_fd(db_node, -1);
+		break;
+	default:
+		fprintf(stderr, "%s: unknown FILE_OP %d\n",
+			progname, file_op->file_op);
+		exit(EXIT_FAILURE);
+		break;
+	}
+}
+
+static void
+do_io(struct thread_state_s *state)
+{
+	void *db_node;
+	struct ioshark_header header;
+	struct ioshark_file_operation file_op;
+	u_int64_t prev_delay = 0;
+	int fd;
+	int i;
+	char *buf = NULL;
+	int buflen = 0;
+	struct timeval total_io_time;
+	struct timeval total_delay_time;
+	struct timeval start;
+	u_int64_t op_counts[IOSHARK_MAX_FILE_OP];
+	struct rw_bytes_s rw_bytes;
+
+	rewind(state->fp);
+	if (fread(&header, sizeof(struct ioshark_header), 1, state->fp) != 1) {
+		fprintf(stderr, "%s read error %s\n",
+			progname, state->filename);
+		exit(EXIT_FAILURE);
+	}
+	/*
+	 * First open and pre-create all the files. Indexed by fileno.
+	 */
+	timerclear(&total_io_time);
+	timerclear(&total_delay_time);
+	memset(&rw_bytes, 0, sizeof(struct rw_bytes_s));
+	memset(op_counts, 0, sizeof(op_counts));
+	fseek(state->fp,
+	      sizeof(struct ioshark_header) +
+	      header.num_files * sizeof(struct ioshark_file_state),
+	      SEEK_SET);
+	/*
+	 * Loop over all the IOs, and launch each
+	 */
+	for (i = 0 ; i < header.num_io_operations ; i++) {
+		if (fread(&file_op, sizeof(struct ioshark_file_operation),
+			  1, state->fp) != 1) {
+			fprintf(stderr, "%s read error trace.outfile\n",
+				progname);
+			exit(EXIT_FAILURE);
+		}
+		if (do_delay) {
+			(void)gettimeofday(&start, (struct timezone *)NULL);
+			usleep(file_op.delta_us - prev_delay);
+			update_delta_time(&start, &total_delay_time);
+			prev_delay = file_op.delta_us;
+		}
+		db_node = files_db_lookup_byfileno(state->db_handle,
+						   file_op.fileno);
+		if (db_node == NULL) {
+			fprintf(stderr,
+				"%s Can't lookup fileno %d, fatal error\n",
+				progname, file_op.fileno);
+			exit(EXIT_FAILURE);
+		}
+		if (file_op.file_op != IOSHARK_OPEN &&
+		    files_db_get_fd(db_node) == -1) {
+			/*
+			 * This is a hack to workaround the fact that we did not
+			 * see an open() for this file until now. open() the file
+			 * O_RDWR, so that we can perform the IO.
+			 */
+			fd = open(files_db_get_filename(db_node), O_RDWR);
+			if (fd < 0) {
+				fprintf(stderr, "%s: open(%s O_RDWR) error %d\n",
+					progname, files_db_get_filename(db_node),
+					errno);
+				exit(EXIT_FAILURE);
+			}
+			files_db_update_fd(db_node, fd);
+		}
+		do_one_io(db_node, &file_op, &total_io_time,
+			  op_counts, &rw_bytes, &buf, &buflen);
+	}
+	(void)gettimeofday(&start, (struct timezone *)NULL);
+	files_db_fsync_discard_files(state->db_handle);
+	files_db_close_files(state->db_handle);
+	update_delta_time(&start, &total_io_time);
+	update_time(&aggregate_IO_time, &total_io_time);
+	update_time(&aggregate_delay_time, &total_delay_time);
+	update_op_counts(op_counts);
+	update_byte_counts(&aggr_io_rw_bytes, &rw_bytes);
+}
+
+void *
+io_thread(void *unused)
+{
+	struct thread_state_s *state;
+
+	UNUSED_PARAM(unused);
+	srand(gettid());
+	while ((state = get_work()))
+		do_io(state);
+	pthread_exit(NULL);
+        return(NULL);
+}
+
+static void
+do_create(struct thread_state_s *state)
+{
+	struct ioshark_header header;
+
+	if (fread(&header, sizeof(struct ioshark_header), 1, state->fp) != 1) {
+		fprintf(stderr, "%s read error %s\n",
+			progname, state->filename);
+		exit(EXIT_FAILURE);
+	}
+	state->num_files = header.num_files;
+	state->db_handle = files_db_create_handle();
+	create_files(state);
+}
+
+void *
+create_files_thread(void *unused)
+{
+	struct thread_state_s *state;
+
+	UNUSED_PARAM(unused);
+	while ((state = get_work()))
+		do_create(state);
+	pthread_exit(NULL);
+	return(NULL);
+}
+
+int
+get_start_end(int *start_ix)
+{
+	int i, j, ret_numfiles;
+	u_int64_t free_fs_bytes;
+	char *infile;
+	FILE *fp;
+	struct ioshark_header header;
+	struct ioshark_file_state file_state;
+	struct statfs fsstat;
+	static int fssize_clamp_next_index = 0;
+
+	if (fssize_clamp_next_index == num_input_files)
+		return 0;
+	if (statfs("/data/local/tmp", &fsstat) < 0) {
+		fprintf(stderr, "%s: Can't statfs /data/local/tmp\n",
+			progname);
+		exit(EXIT_FAILURE);
+	}
+	free_fs_bytes = (fsstat.f_bavail * fsstat.f_bsize) * 9 /10;
+	for (i = fssize_clamp_next_index; i < num_input_files; i++) {
+		infile = thread_state[i].filename;
+		fp = fopen(infile, "r");
+		if (fp == NULL) {
+			fprintf(stderr, "%s: Can't open %s\n",
+				progname, infile);
+			exit(EXIT_FAILURE);
+		}
+		if (fread(&header, sizeof(struct ioshark_header),
+			  1, fp) != 1) {
+			fprintf(stderr, "%s read error %s\n",
+				progname, infile);
+			exit(EXIT_FAILURE);
+		}
+		for (j = 0 ; j < header.num_files ; j++) {
+			if (fread(&file_state, sizeof(struct ioshark_file_state),
+				  1, fp) != 1) {
+				fprintf(stderr, "%s read error tracefile\n",
+					progname);
+				exit(EXIT_FAILURE);
+			}
+			if (file_state.size > free_fs_bytes) {
+				fclose(fp);
+				printf("Reducing number of input files from %d down to %d\n",
+				       num_input_files,
+				       i - fssize_clamp_next_index);
+				goto out;
+			}
+			free_fs_bytes -= file_state.size;
+		}
+		fclose(fp);
+	}
+out:
+	*start_ix = fssize_clamp_next_index;
+	ret_numfiles = i - fssize_clamp_next_index;
+	fssize_clamp_next_index = i;
+	return ret_numfiles;
+}
+
+int
+ioshark_pthread_create(pthread_t *tidp, void *(*start_routine)(void *))
+{
+	pthread_attr_t attr;
+
+	pthread_attr_init(&attr);
+	pthread_attr_setscope(&attr, PTHREAD_SCOPE_SYSTEM);
+	pthread_attr_setstacksize(&attr, (size_t)(1024*1024));
+	return pthread_create(tidp, &attr, start_routine, (void *)NULL);
+}
+
+void
+wait_for_threads(int num_threads)
+{
+	int i;
+
+	for (i = 0; i < num_threads; i++) {
+		pthread_join(tid[i], NULL);
+		tid[i] = 0;
+	}
+}
+
+int
+main(int argc, char **argv)
+{
+	int i;
+	FILE *fp;
+	struct stat st;
+	char *infile;
+	int num_threads = 0;
+	int num_iterations = 1;
+	int c;
+	int num_files, start_file;
+	struct thread_state_s *state;
+
+	progname = argv[0];
+        while ((c = getopt(argc, argv, "dn:t:")) != EOF) {
+                switch (c) {
+                case 'd':
+			do_delay = 1;
+			break;
+                case 'n':
+			num_iterations = atoi(optarg);
+			break;
+                case 't':
+			num_threads = atoi(optarg);
+			break;
+ 	        default:
+                        usage();
+		}
+	}
+
+	if (num_threads > MAX_THREADS)
+		usage();
+
+	if (optind == argc)
+                usage();
+
+	for (i = optind; i < argc; i++) {
+		infile = argv[i];
+		if (stat(infile, &st) < 0) {
+			fprintf(stderr, "%s: Can't stat %s\n",
+				progname, infile);
+			exit(EXIT_FAILURE);
+			continue;
+		}
+		if (st.st_size == 0) {
+			fprintf(stderr, "%s: Empty file %s\n",
+				progname, infile);
+			continue;
+		}
+		fp = fopen(infile, "r");
+		if (fp == NULL) {
+			fprintf(stderr, "%s: Can't open %s\n",
+				progname, infile);
+			continue;
+		}
+		thread_state[num_input_files].filename = infile;
+		thread_state[num_input_files].fp = fp;
+		num_input_files++;
+	}
+
+	if (num_input_files == 0) {
+		exit(EXIT_SUCCESS);
+	}
+	printf("Total Input Files = %d\n", num_input_files);
+	printf("Num Iterations = %d\n", num_iterations);
+
+	timerclear(&aggregate_file_create_time);
+	timerclear(&aggregate_file_remove_time);
+	timerclear(&aggregate_IO_time);
+
+	/*
+	 * We pre-create the files that we need once and then we
+	 * loop around N times doing IOs on the pre-created files.
+	 *
+	 * get_start_end() breaks up the total work here to make sure
+	 * that all the files we need to pre-create fit into the
+	 * available space in /data/local/tmp (hardcoded for now).
+	 *
+	 * If it won't fit, then we do several sweeps.
+	 */
+	while ((num_files = get_start_end(&start_file))) {
+		/* Create files once */
+		printf("Doing Pre-creation of Files\n");
+		if (num_threads == 0 || num_threads > num_files)
+			num_threads = num_files;
+		(void)system("echo 3 > /proc/sys/vm/drop_caches");
+		init_work(start_file, num_files);
+		for (i = 0; i < num_threads; i++) {
+			if (ioshark_pthread_create(&(tid[i]),
+						   create_files_thread)) {
+				fprintf(stderr,
+					"%s: Can't create creator thread %d\n",
+					progname, i);
+				exit(EXIT_FAILURE);
+			}
+		}
+		wait_for_threads(num_threads);
+
+		/* Do the IOs N times */
+		for (i = 0 ; i < num_iterations ; i++) {
+			(void)system("echo 3 > /proc/sys/vm/drop_caches");
+			printf("Starting Test. Iteration %d\n", i);
+			init_work(start_file, num_files);
+			for (c = 0; c < num_threads; c++) {
+				if (ioshark_pthread_create(&(tid[c]),
+							   io_thread)) {
+					fprintf(stderr,
+						"%s: Can't create thread %d\n",
+						progname, c);
+					exit(EXIT_FAILURE);
+				}
+			}
+			wait_for_threads(num_threads);
+		}
+		/*
+		 * We are done with the N iterations of IO.
+		 * Destroy the files we pre-created.
+		 */
+		init_work(start_file, num_files);
+		while ((state = get_work())) {
+			struct timeval start;
+
+			(void)gettimeofday(&start, (struct timezone *)NULL);
+			files_db_unlink_files(state->db_handle);
+			update_delta_time(&start, &aggregate_file_remove_time);
+			files_db_free_memory(state->db_handle);
+		}
+	}
+	printf("Total Creation time = %lu.%lu (msecs.usecs)\n",
+	       get_msecs(&aggregate_file_create_time),
+	       get_usecs(&aggregate_file_create_time));
+	printf("Total Remove time = %lu.%lu (msecs.usecs)\n",
+	       get_msecs(&aggregate_file_remove_time),
+	       get_usecs(&aggregate_file_remove_time));
+	if (do_delay)
+		printf("Total delay time = %lu.%lu (msecs.usecs)\n",
+		       get_msecs(&aggregate_delay_time),
+		       get_usecs(&aggregate_delay_time));
+	printf("Total IO time = %lu.%lu (msecs.usecs)\n",
+	       get_msecs(&aggregate_IO_time),
+	       get_usecs(&aggregate_IO_time));
+	print_bytes("Upfront File Creation bytes",
+		    &aggr_create_rw_bytes);
+	print_bytes("IO bytes", &aggr_io_rw_bytes);
+	print_op_stats(aggr_op_counts);
+}
diff --git a/ioshark/ioshark_bench.h b/ioshark/ioshark_bench.h
new file mode 100644
index 00000000..6ce9322f
--- /dev/null
+++ b/ioshark/ioshark_bench.h
@@ -0,0 +1,151 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifdef IOSHARK_MAIN
+const char *IO_op[] = {
+	"LSEEK",
+	"LLSEEK",
+	"PREAD64",
+	"PWRITE64",
+	"READ",
+	"WRITE",
+	"MMAP",
+	"MMAP2",
+	"OPEN",
+	"FSYNC",
+	"FDATASYNC",
+	"CLOSE",
+	"MAPPED_PREAD",
+	"MAPPED_PWRITE",
+	"MAX_FILE_OP"
+};
+#endif
+
+#define MAX(A, B)	((A) > (B) ? (A) : (B))
+#define MIN(A, B)	((A) < (B) ? (A) : (B))
+
+#define MINBUFLEN	(16*1024)
+
+#define FILE_DB_HASHSIZE	8192
+
+struct files_db_s {
+	char *filename;
+	int fileno;
+	size_t	size;
+	int fd;
+	int debug_open_flags;
+	struct files_db_s *next;
+};
+
+struct files_db_handle {
+	struct files_db_s *files_db_buckets[FILE_DB_HASHSIZE];
+};
+
+struct IO_operation_s {
+	char *IO_op;
+};
+
+struct rw_bytes_s {
+	u_int64_t bytes_read;
+	u_int64_t bytes_written;
+};
+
+static inline void
+files_db_update_size(void *node, u_int64_t new_size)
+{
+	struct files_db_s *db_node = (struct files_db_s *)node;
+
+	if (db_node->size < new_size)
+		db_node->size = new_size;
+}
+
+static inline void
+files_db_update_filename(void *node, char *filename)
+{
+	((struct files_db_s *)node)->filename = strdup(filename);
+}
+
+static inline int
+files_db_get_fileno(void *node)
+{
+	return (((struct files_db_s *)node)->fileno);
+}
+
+static inline int
+files_db_get_fd(void *node)
+{
+	return (((struct files_db_s *)node)->fd);
+}
+
+static inline char *
+files_db_get_filename(void *node)
+{
+	return (((struct files_db_s *)node)->filename);
+}
+
+static inline u_int64_t
+get_msecs(struct timeval *tv)
+{
+	return ((tv->tv_sec * 1000) + (tv->tv_usec / 1000));
+}
+
+static inline u_int64_t
+get_usecs(struct timeval *tv)
+{
+	return (tv->tv_usec % 1000);
+}
+
+static inline void
+update_delta_time(struct timeval *start,
+		  struct timeval *destination)
+{
+	struct timeval res, finish;
+
+	(void)gettimeofday(&finish, (struct timezone *)NULL);
+	timersub(&finish, start, &res);
+	timeradd(destination, &res, &finish);
+	*destination = finish;
+}
+
+void *files_db_create_handle(void);
+void *files_db_lookup_byfileno(void *handle, int fileno);
+void *files_db_add_byfileno(void *handle, int fileno);
+void files_db_update_fd(void *node, int fd);
+void files_db_unlink_files(void *db_handle);
+void files_db_close_files(void *handle);
+void files_db_close_fd(void *node);
+void files_db_free_memory(void *handle);
+void create_file(char *path, size_t size,
+		 struct timeval *total_time,
+		 struct rw_bytes_s *rw_bytes);
+char *get_buf(char **buf, int *buflen, int len, int do_fill);
+void files_db_fsync_discard_files(void *handle);
+void print_op_stats(u_int64_t *op_counts);
+void print_bytes(char *desc, struct rw_bytes_s *rw_bytes);
+void ioshark_handle_mmap(void *db_node,
+			 struct ioshark_file_operation *file_op,
+			 char **bufp, int *buflen, u_int64_t *op_counts,
+			 struct rw_bytes_s *rw_bytes, struct timeval *io_time);
+
+
+
+
+
+
+
+
+
+
diff --git a/ioshark/ioshark_bench_mmap.c b/ioshark/ioshark_bench_mmap.c
new file mode 100644
index 00000000..7ee7c41b
--- /dev/null
+++ b/ioshark/ioshark_bench_mmap.c
@@ -0,0 +1,213 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <stdio.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <sys/syscall.h>
+#include <stdlib.h>
+#include <string.h>
+#include <pthread.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include <linux/fs.h>
+#include <fcntl.h>
+#include <string.h>
+#include <assert.h>
+#include <pthread.h>
+#include "ioshark.h"
+#include "ioshark_bench.h"
+
+/*
+ * The purpose of this code is to convert mmap() calls into
+ * a mix of (semi)-random reads and writes.
+ * PROT_READ => 4KB/8KB/16KB random reads.
+ * PROT_WRITE => adds 4KB random writes.
+ */
+
+extern char *progname;
+
+#define IOSHARK_MAX_MMAP_IOLEN	(16*1024)
+
+#define MMAP_ENTS		16
+
+struct mmap_io_ent_tab_s {
+	off_t offset;
+	size_t len;
+};
+
+struct mmap_io_ent_s {
+	int				num_entries;
+	struct mmap_io_ent_tab_s	table[MMAP_ENTS + 1];
+	size_t				resid;
+};
+
+static void
+setup_mmap_io_state(struct mmap_io_ent_s *mio,
+		    size_t total_len, off_t offset)
+{
+	int slice;
+
+	memset(mio, 0, sizeof(struct mmap_io_ent_s));
+	mio->resid = total_len;
+	slice = MAX(IOSHARK_MAX_MMAP_IOLEN,
+		    total_len / MMAP_ENTS);
+	while (total_len > 0) {
+		assert(mio->num_entries < MMAP_ENTS + 1);
+		mio->table[mio->num_entries].offset = offset;
+		mio->table[mio->num_entries].len =
+			MIN((u_int64_t)total_len, (u_int64_t)slice);
+		total_len -= mio->table[mio->num_entries].len;
+		offset += mio->table[mio->num_entries].len;
+		mio->num_entries++;
+	}
+}
+
+static size_t
+mmap_getnext_off_len(struct mmap_io_ent_s *mio,
+		     off_t *offset)
+{
+	int i;
+	int find_rand_index[MMAP_ENTS + 1];
+	int rand_index_len = 0;
+	size_t iolength;
+
+	if (mio->resid == 0)
+		return 0;
+	/* Pick a slot with residual length > 0 at random first */
+	for (i = 0 ; i < MMAP_ENTS + 1 ; i++) {
+		if (mio->table[i].len > 0)
+			find_rand_index[rand_index_len++] = i;
+	}
+	i = find_rand_index[rand() % rand_index_len];
+	/* Randomize iolength 4K-16K */
+	iolength = ((rand() % 4) + 1) * 4096;
+	iolength = MIN(mio->table[i].len, iolength);
+	*offset = mio->table[i].offset;
+	mio->table[i].offset += iolength;
+	mio->table[i].len -= iolength;
+	mio->resid -= iolength;
+	return iolength;
+}
+
+static void
+mmap_do_io(void *db_node, int prot, off_t offset, size_t len,
+	   char **bufp, int *buflen, u_int64_t *op_counts,
+	   struct rw_bytes_s *rw_bytes, struct timeval *io_time)
+{
+	struct timeval start;
+	char *p;
+	int ret;
+
+	if (!(prot & IOSHARK_PROT_WRITE)) {
+		/* Only preads */
+		p = get_buf(bufp, buflen, len, 0);
+		(void)gettimeofday(&start,
+				   (struct timezone *)NULL);
+		ret = pread(files_db_get_fd(db_node),
+			    p, len, offset);
+		update_delta_time(&start, io_time);
+		rw_bytes->bytes_read += len;
+		if (ret < 0) {
+			fprintf(stderr,
+				"%s: mapped pread(%s %lu %lu) error fd=%d %s\n",
+				progname, files_db_get_filename(db_node),
+				len, offset, files_db_get_fd(db_node),
+				strerror(errno));
+			exit(EXIT_FAILURE);
+		}
+		op_counts[IOSHARK_MAPPED_PREAD]++;
+	} else {
+		/* 50-50 R/W */
+		if ((rand() % 2) == 1) {
+			p = get_buf(bufp, buflen, len, 1);
+			(void)gettimeofday(&start,
+					   (struct timezone *)NULL);
+			ret = pwrite(files_db_get_fd(db_node),
+				     p, len, offset);
+			update_delta_time(&start, io_time);
+			rw_bytes->bytes_written += len;
+			if (ret < 0) {
+#if 0
+				fprintf(stderr,
+					"%s: mapped pwrite failed, file unwriteable ? open_flags=%x\n",
+					progname,
+					fcntl(files_db_get_fd(db_node), F_GETFL));
+				exit(EXIT_FAILURE);
+#endif
+			} else
+				op_counts[IOSHARK_MAPPED_PWRITE]++;
+		} else {
+			p = get_buf(bufp, buflen, len, 0);
+			(void)gettimeofday(&start,
+					   (struct timezone *)NULL);
+			ret = pread(files_db_get_fd(db_node),
+				    p, len, offset);
+			update_delta_time(&start, io_time);
+			rw_bytes->bytes_read += len;
+			if (ret < 0) {
+				fprintf(stderr,
+				"%s: mapped pread(%s %lu %lu) error fd=%d %s\n",
+					progname, files_db_get_filename(db_node),
+					len,
+					offset, files_db_get_fd(db_node),
+					strerror(errno));
+				exit(EXIT_FAILURE);
+			}
+			op_counts[IOSHARK_MAPPED_PREAD]++;
+		}
+	}
+}
+
+void
+ioshark_handle_mmap(void *db_node,
+		    struct ioshark_file_operation *file_op,
+		    char **bufp, int *buflen, u_int64_t *op_counts,
+		    struct rw_bytes_s *rw_bytes, struct timeval *io_time)
+{
+	off_t offset = file_op->mmap_offset;
+	size_t len = file_op->mmap_len;
+	int prot = file_op->mmap_prot;
+	struct mmap_io_ent_s mio;
+	struct stat statbuf;
+
+	if (fstat(files_db_get_fd(db_node), &statbuf) < 0) {
+		fprintf(stderr,
+			"%s: fstat failure %s\n",
+			__func__, strerror(errno));
+		exit(EXIT_FAILURE);
+	}
+	/*
+	 * The size of the file better accomodate offset + len
+	 * Else there is an issue with pre-creation
+	 */
+	assert(offset + len <= statbuf.st_size);
+	if (len <= IOSHARK_MAX_MMAP_IOLEN) {
+		mmap_do_io(db_node, prot, offset, len,
+			   bufp, buflen, op_counts,
+			   rw_bytes, io_time);
+		return;
+	}
+	setup_mmap_io_state(&mio, len, offset);
+	assert(mio.num_entries > 0);
+	while ((len = mmap_getnext_off_len(&mio, &offset))) {
+		assert((offset + len) <=
+		       (file_op->mmap_offset + file_op->mmap_len));
+		mmap_do_io(db_node, prot, offset, len, bufp, buflen,
+			   op_counts, rw_bytes, io_time);
+	}
+}
diff --git a/ioshark/ioshark_bench_subr.c b/ioshark/ioshark_bench_subr.c
new file mode 100644
index 00000000..ff273cce
--- /dev/null
+++ b/ioshark/ioshark_bench_subr.c
@@ -0,0 +1,324 @@
+/*
+ * Copyright (C) 2016 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <stdio.h>
+#include <sys/time.h>
+#include <sys/types.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <string.h>
+#include <sys/stat.h>
+#include <sys/errno.h>
+#include <linux/fs.h>
+#include <fcntl.h>
+#include <string.h>
+#include <assert.h>
+#include "ioshark.h"
+#include "ioshark_bench.h"
+
+extern char *progname;
+
+void *
+files_db_create_handle(void)
+{
+	struct files_db_handle *h;
+	int i;
+
+	h = malloc(sizeof(struct files_db_handle));
+	for (i = 0 ; i < FILE_DB_HASHSIZE ; i++)
+		h->files_db_buckets[i] = NULL;
+	return h;
+}
+
+void *files_db_lookup_byfileno(void *handle, int fileno)
+{
+	u_int32_t	hash;
+	struct files_db_handle *h = (struct files_db_handle *)handle;
+	struct files_db_s *db_node;
+
+	hash = fileno % FILE_DB_HASHSIZE;
+	db_node = h->files_db_buckets[hash];
+	while (db_node != NULL) {
+		if (db_node->fileno == fileno)
+			break;
+		db_node = db_node->next;
+	}
+	return db_node;
+}
+
+void *files_db_add_byfileno(void *handle, int fileno)
+{
+	u_int32_t	hash = fileno % FILE_DB_HASHSIZE;
+	struct files_db_handle *h = (struct files_db_handle *)handle;
+	struct files_db_s *db_node;
+
+	db_node = (struct files_db_s *)
+		files_db_lookup_byfileno(handle, fileno);
+	if (db_node == NULL) {
+		db_node = malloc(sizeof(struct files_db_s));
+		db_node->fileno = fileno;
+		db_node->filename = NULL;
+		db_node->size = 0;
+		db_node->fd = -1;
+		db_node->next = h->files_db_buckets[hash];
+		h->files_db_buckets[hash] = db_node;
+	} else {
+		fprintf(stderr,
+			"%s: Node to be added already exists fileno = %d\n\n",
+			__func__, fileno);
+		exit(EXIT_FAILURE);
+	}
+	return db_node;
+}
+
+void
+files_db_fsync_discard_files(void *handle)
+{
+	struct files_db_handle *h = (struct files_db_handle *)handle;
+	struct files_db_s *db_node;
+	int i;
+
+	for (i = 0 ; i < FILE_DB_HASHSIZE ; i++) {
+		db_node = h->files_db_buckets[i];
+		while (db_node != NULL) {
+			int do_close = 0;
+
+			if (db_node->fd == -1) {
+				int fd;
+
+				/*
+				 * File was closed, let's open it so we can
+				 * fsync and fadvise(DONTNEED) it.
+				 */
+				do_close = 1;
+				fd = open(files_db_get_filename(db_node),
+					  O_RDWR);
+				if (fd < 0) {
+					fprintf(stderr,
+						"%s: open(%s O_RDWR) error %d\n",
+						progname, db_node->filename,
+						errno);
+					exit(EXIT_FAILURE);
+				}
+				db_node->fd = fd;
+			}
+			if (fsync(db_node->fd) < 0) {
+				fprintf(stderr, "%s: Cannot fsync %s\n",
+					__func__, db_node->filename);
+				exit(1);
+			}
+			if (posix_fadvise(db_node->fd, 0, 0,
+					  POSIX_FADV_DONTNEED) < 0) {
+				fprintf(stderr,
+					"%s: Cannot fadvise(DONTNEED) %s\n",
+					__func__, db_node->filename);
+				exit(1);
+			}
+			if (do_close) {
+				close(db_node->fd);
+				db_node->fd = -1;
+			}
+			db_node = db_node->next;
+		}
+	}
+}
+
+void
+files_db_update_fd(void *node, int fd)
+{
+	struct files_db_s *db_node = (struct files_db_s *)node;
+
+	db_node->fd = fd;
+}
+
+void
+files_db_close_fd(void *node)
+{
+	struct files_db_s *db_node = (struct files_db_s *)node;
+
+	if (db_node->fd != -1)
+		close(db_node->fd);
+	db_node->fd = -1;
+}
+
+void
+files_db_close_files(void *handle)
+{
+	struct files_db_handle *h = (struct files_db_handle *)handle;
+	struct files_db_s *db_node;
+	int i;
+
+	for (i = 0 ; i < FILE_DB_HASHSIZE ; i++) {
+		db_node = h->files_db_buckets[i];
+		while (db_node != NULL) {
+			if ((db_node->fd != -1) && close(db_node->fd) < 0) {
+				fprintf(stderr, "%s: Cannot close %s\n",
+					__func__, db_node->filename);
+				exit(1);
+			}
+			db_node->fd = -1;
+			db_node = db_node->next;
+		}
+	}
+}
+
+void
+files_db_unlink_files(void *handle)
+{
+	struct files_db_handle *h = (struct files_db_handle *)handle;
+	struct files_db_s *db_node;
+	int i;
+
+	for (i = 0 ; i < FILE_DB_HASHSIZE ; i++) {
+		db_node = h->files_db_buckets[i];
+		while (db_node != NULL) {
+			if ((db_node->fd != -1) && close(db_node->fd) < 0) {
+				fprintf(stderr, "%s: Cannot close %s\n",
+					__func__, db_node->filename);
+				exit(1);
+			}
+			db_node->fd = -1;
+			if (unlink(db_node->filename) < 0) {
+				fprintf(stderr, "%s: Cannot unlink %s:%s\n",
+					__func__, db_node->filename, strerror(errno));
+				exit(1);
+			}
+			db_node = db_node->next;
+		}
+	}
+}
+
+void
+files_db_free_memory(void *handle)
+{
+	struct files_db_handle *h = (struct files_db_handle *)handle;
+	struct files_db_s *db_node, *tmp;
+	int i;
+
+	for (i = 0 ; i < FILE_DB_HASHSIZE ; i++) {
+		db_node = h->files_db_buckets[i];
+		while (db_node != NULL) {
+			tmp = db_node;
+			db_node = db_node->next;
+			free(tmp->filename);
+			free(tmp);
+		}
+	}
+	free(h);
+}
+
+char *
+get_buf(char **buf, int *buflen, int len, int do_fill)
+{
+	if (len == 0 && *buf == NULL) {
+		/*
+		 * If we ever get a zero len
+		 * request, start with MINBUFLEN
+		 */
+		if (*buf == NULL)
+			len = MINBUFLEN / 2;
+	}
+	if (*buflen < len) {
+		*buflen = MAX(MINBUFLEN, len * 2);
+		if (*buf)
+			free(*buf);
+		*buf = malloc(*buflen);
+	}
+	if (do_fill) {
+		u_int32_t *s;
+		int count;
+
+		s = (u_int32_t *)*buf;
+		count = *buflen / sizeof(u_int32_t);
+		while (count > 0) {
+			*s++ = rand();
+			count--;
+		}
+	}
+	assert(*buf != NULL);
+	return *buf;
+}
+
+void
+create_file(char *path, size_t size, struct timeval *total_time,
+	    struct rw_bytes_s *rw_bytes)
+{
+	int fd, n;
+	char *buf = NULL;
+	int buflen = 0;
+        struct timeval start;
+
+	(void)gettimeofday(&start, (struct timezone *)NULL);
+	fd = open(path, O_WRONLY|O_CREAT|O_TRUNC, 0644);
+	update_delta_time(&start, total_time);
+	if (fd < 0) {
+		fprintf(stderr, "%s Cannot create file %s, error = %d\n",
+			progname, path, errno);
+		exit(EXIT_FAILURE);
+	}
+	while (size > 0) {
+		n = MIN(size, MINBUFLEN);
+		buf = get_buf(&buf, &buflen, n, 1);
+		(void)gettimeofday(&start, (struct timezone *)NULL);
+		if (write(fd, buf, n) < n) {
+			fprintf(stderr,
+				"%s Cannot write file %s, error = %d\n",
+				progname, path, errno);
+			exit(EXIT_FAILURE);
+		}
+		rw_bytes->bytes_written += n;
+		update_delta_time(&start, total_time);
+		size -= n;
+	}
+	(void)gettimeofday(&start, (struct timezone *)NULL);
+	if (fsync(fd) < 0) {
+		fprintf(stderr, "%s Cannot fsync file %s, error = %d\n",
+			progname, path, errno);
+		exit(EXIT_FAILURE);
+	}
+	if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED) < 0) {
+		fprintf(stderr,
+			"%s Cannot fadvise(DONTNEED) file %s, error = %d\n",
+			progname, path, errno);
+		exit(EXIT_FAILURE);
+	}
+	close(fd);
+	update_delta_time(&start, total_time);
+}
+
+void
+print_op_stats(u_int64_t *op_counts)
+{
+	int i;
+	extern char *IO_op[];
+
+	printf("IO Operation counts :\n");
+	for (i = IOSHARK_LSEEK ; i < IOSHARK_MAX_FILE_OP ; i++) {
+		printf("%s: %lu\n",
+		       IO_op[i], op_counts[i]);
+	}
+}
+
+void
+print_bytes(char *desc, struct rw_bytes_s *rw_bytes)
+{
+	printf("%s: Reads = %dMB, Writes = %dMB\n",
+	       desc,
+	       (int)(rw_bytes->bytes_read / (1024 * 1024)),
+	       (int)(rw_bytes->bytes_written / (1024 * 1024)));
+}
+