I've done some work to merge some of the existing fsx.c mods into a
single version. Over & above the version that is in the LTP, I've
included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
segfault handling, and per-write fsync.
The twist for the O_DIRECT feature is that it will randomly open file
descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
support fsx will be testing concurrent buffered and O_DIRECT and mmap
IO on the same file.
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
I've done some work to merge some of the existing fsx.c mods into a
single version. Over & above the version that is in the LTP, I've
included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
segfault handling, and per-write fsync.
The twist for the O_DIRECT feature is that it will randomly open file
descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
support (by specifying multiple pathnames for the output file) fsx will
be testing buffered and O_DIRECT and mmap IO on the same file.
Signed-off-by: Andreas Dilger <[email protected]>
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
I've done some work to merge some of the existing fsx.c mods into a
single version. Over & above the version that is in the LTP, I've
included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
segfault handling, and per-write fsync.
The twist for the O_DIRECT feature is that it will randomly open file
descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
support (by specifying multiple pathnames for the output file) fsx will
be testing buffered and O_DIRECT and mmap IO on the same file.
Updated patch to have proper fallocate() handling in case glibc doesn't
have this, for non-x86 architectures from MingMing Cao, based on code
used by DB2.
Signed-off-by: Andreas Dilger <[email protected]>
/*
* Copyright (C) 1991, NeXT Computer, Inc. All Rights Reserverd.
* Copyright (c) 1998-2001 Apple Computer, Inc. All rights reserved.
*
* @APPLE_LICENSE_HEADER_START@
*
* The contents of this file constitute Original Code as defined in and
* are subject to the Apple Public Source License Version 1.1 (the
* "License"). You may not use this file except in compliance with the
* License. Please obtain a copy of the License at
* http://www.apple.com/publicsource and read it before using this file.
*
* This Original Code and all software distributed under the License are
* distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
* EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
* INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
* License for the specific language governing rights and limitations
* under the License.
*
* @APPLE_LICENSE_HEADER_END@
*
* File: fsx.c
* Author: Avadis Tevanian, Jr.
*
* File system exerciser.
*
* Rewrite and enhancements 1998-2001 Conrad Minshall -- [email protected]
*
* Various features from Joe Sokol, Pat Dirks, and Clark Warner.
*
* Small changes to work under Linux -- [email protected]
*
* Sundry porting patches from Guy Harris 12/2001
*
* Checks for mmap last-page zero fill.
*
* Add multi-file testing feature -- Zach Brown <[email protected]>
*
* $FreeBSD: src/tools/regression/fsx/fsx.c,v 1.2 2003/04/23 23:42:23 jkh Exp $
* $DragonFly: src/test/stress/fsx/fsx.c,v 1.2 2005/05/02 19:31:56 dillon Exp $
*
*/
#include <sys/types.h>
#include <sys/stat.h>
#if defined(_UWIN) || defined(__linux__)
# include <sys/param.h>
# include <limits.h>
# include <time.h>
# include <strings.h>
# include <sys/time.h>
#endif
#include <fcntl.h>
#include <sys/mman.h>
#ifndef MAP_FILE
# define MAP_FILE 0
#endif
#include <limits.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <stdarg.h>
#include <errno.h>
#include <setjmp.h>
/*
* A log entry is an operation and a bunch of arguments.
*/
struct log_entry {
int operation;
int args[3];
struct timeval tv;
};
#define LOGSIZE 100000
struct log_entry oplog[LOGSIZE]; /* the log */
int logptr = 0; /* current position in log */
int logcount = 0; /* total ops */
int jmpbuf_good;
jmp_buf jmpbuf;
/*
* Define operations
*/
#define OP_READ 1
#define OP_WRITE 2
#define OP_TRUNCATE 3
#define OP_CLOSEOPEN 4
#define OP_MAPREAD 5
#define OP_MAPWRITE 6
#define OP_SKIPPED 7
int page_size;
int page_mask;
char *original_buf; /* a pointer to the original data */
char *good_buf; /* a pointer to the correct data */
char *temp_buf; /* a pointer to the current data */
char *fname; /* name of our test file */
char logfile[1024]; /* name of our log file */
char goodfile[1024]; /* name of our test file */
off_t file_size = 0;
off_t biggest = 0;
char state[256];
unsigned long testcalls = 0; /* calls to function "test" */
unsigned long simulatedopcount = 0; /* -b flag */
int closeprob = 0; /* -c flag */
int debug = 0; /* -d flag */
unsigned long debugstart = 0; /* -D flag */
int do_fsync = 0; /* -f flag */
unsigned long maxfilelen = 256 * 1024; /* -l flag */
int sizechecks = 1; /* -n flag disables them */
int maxoplen = 64 * 1024; /* -o flag */
int quiet = 0; /* -q flag */
unsigned long progressinterval = 0; /* -p flag */
int readbdy = 1; /* -r flag */
int style = 0; /* -s flag */
int truncbdy = 1; /* -t flag */
int writebdy = 1; /* -w flag */
long monitorstart = -1; /* -m flag */
long monitorend = -1; /* -m flag */
int lite = 0; /* -L flag */
long numops = -1; /* -N flag */
int randomoplen = 1; /* -O flag disables it */
int seed = 1; /* -S flag */
int mapped_writes = 1; /* -W flag disables */
int mapped_reads = 1; /* -R flag disables it */
int prealloc = 0; /* -x [0|1|2] run with prealloc */
int o_direct = 0; /* -Z flag */
int fsxgoodfd = 0;
FILE *fsxlogf = NULL;
int badoff = -1;
void
vwarnc(code, fmt, ap)
int code;
const char *fmt;
va_list ap;
{
fprintf(stderr, "fsx: ");
if (fmt != NULL) {
vfprintf(stderr, fmt, ap);
fprintf(stderr, ": ");
}
fprintf(stderr, "%s\n", strerror(code));
}
void
warn(const char * fmt, ...)
{
va_list ap;
va_start(ap, fmt);
vwarnc(errno, fmt, ap);
va_end(ap);
}
static void *round_up(void *ptr, unsigned long align)
{
unsigned long ret = (unsigned long)ptr;
ret = ((ret + align - 1) & ~(align - 1));
return (void *)ret;
}
void
__attribute__((format(printf, 1, 2)))
prt(char *fmt, ...)
{
va_list args;
va_start(args, fmt);
vfprintf(stdout, fmt, args);
va_end(args);
if (fsxlogf) {
va_start(args, fmt);
vfprintf(fsxlogf, fmt, args);
va_end(args);
}
}
void
prterr(char *prefix)
{
prt("%s%s%s\n", prefix, prefix ? ": " : "", strerror(errno));
}
void
log4(int operation, int arg0, int arg1, int arg2, struct timeval *tv)
{
struct log_entry *le;
le = &oplog[logptr];
le->tv = *tv;
le->operation = operation;
le->args[0] = arg0;
le->args[1] = arg1;
le->args[2] = arg2;
logptr++;
logcount++;
if (logptr >= LOGSIZE)
logptr = 0;
}
void
logdump(void)
{
int i, count, down;
struct log_entry *lp;
prt("LOG DUMP (%d total operations):\n", logcount);
if (logcount < LOGSIZE) {
i = 0;
count = logcount;
} else {
i = logptr;
count = LOGSIZE;
}
for ( ; count > 0; count--) {
int opnum;
opnum = i+1 + (logcount/LOGSIZE)*LOGSIZE;
lp = &oplog[i];
prt("%d: %lu.%06lu ", opnum,
lp->tv.tv_sec, lp->tv.tv_usec);
switch (lp->operation) {
case OP_MAPREAD:
prt("MAPREAD 0x%x thru 0x%x (0x%x bytes)",
lp->args[0], lp->args[0] + lp->args[1] - 1,
lp->args[1]);
if (badoff >= lp->args[0] &&
badoff < lp->args[0] + lp->args[1])
prt("\t***RRRR***");
break;
case OP_MAPWRITE:
prt("MAPWRITE 0x%x thru 0x%x (0x%x bytes)",
lp->args[0], lp->args[0] + lp->args[1] - 1,
lp->args[1]);
if (badoff >= lp->args[0] &&
badoff < lp->args[0] + lp->args[1])
prt("\t******WWWW");
break;
case OP_READ:
case OP_READ + O_DIRECT:
prt("READ%s 0x%x thru 0x%x (0x%x bytes)",
lp->operation & O_DIRECT ? "_OD" : " ",
lp->args[0], lp->args[0] + lp->args[1] - 1,
lp->args[1]);
if (badoff >= lp->args[0] &&
badoff < lp->args[0] + lp->args[1])
prt("\t***RRRR***");
break;
case OP_WRITE:
case OP_WRITE + O_DIRECT:
prt("WRITE%s 0x%x thru 0x%x (0x%x bytes)",
lp->operation & O_DIRECT ? "_OD" : " ",
lp->args[0], lp->args[0] + lp->args[1] - 1,
lp->args[1]);
if (lp->args[0] > lp->args[2])
prt(" HOLE");
else if (lp->args[0] + lp->args[1] > lp->args[2])
prt(" EXTEND");
if ((badoff >= lp->args[0] || badoff >=lp->args[2]) &&
badoff < lp->args[0] + lp->args[1])
prt("\t***WWWW");
break;
case OP_TRUNCATE:
down = lp->args[0] < lp->args[1];
prt("TRUNCATE %s\tfrom 0x%x to 0x%x",
down ? "DOWN" : "UP", lp->args[1], lp->args[0]);
if (badoff >= lp->args[!down] &&
badoff < lp->args[!!down])
prt("\t******WWWW");
break;
case OP_CLOSEOPEN:
case OP_CLOSEOPEN + O_DIRECT:
prt("CLOSE/OPEN%s",
lp->operation & O_DIRECT ? "_OD" : " ");
break;
case OP_SKIPPED:
prt("SKIPPED (no operation)");
break;
default:
prt("BOGUS LOG ENTRY (operation code = %d)!",
lp->operation);
}
prt("\n");
i++;
if (i == LOGSIZE)
i = 0;
}
}
void
save_buffer(char *buffer, off_t bufferlength, int fd)
{
off_t ret;
ssize_t byteswritten;
if (fd <= 0 || bufferlength == 0)
return;
if (bufferlength > INT_MAX) {
prt("fsx flaw: overflow in save_buffer\n");
exit(67);
}
if (lite) {
off_t size_by_seek = lseek(fd, (off_t)0, SEEK_END);
if (size_by_seek == (off_t)-1)
prterr("save_buffer: lseek eof");
else if (bufferlength > size_by_seek) {
warn("save_buffer: .fsxgood file too short... will"
"save 0x%llx bytes instead of 0x%llx\n",
(unsigned long long)size_by_seek,
(unsigned long long)bufferlength);
bufferlength = size_by_seek;
}
}
ret = lseek(fd, (off_t)0, SEEK_SET);
if (ret == (off_t)-1)
prterr("save_buffer: lseek 0");
byteswritten = write(fd, buffer, (size_t)bufferlength);
if (byteswritten != bufferlength) {
if (byteswritten == -1)
prterr("save_buffer write");
else
warn("save_buffer: short write, 0x%x bytes instead"
"of 0x%llx\n",
(unsigned)byteswritten,
(unsigned long long)bufferlength);
}
}
void
report_failure(int status)
{
logdump();
if (fsxgoodfd) {
if (good_buf) {
save_buffer(good_buf, file_size, fsxgoodfd);
prt("Correct content saved for comparison\n");
prt("(maybe hexdump \"%s\" vs \"%s\")\n",
fname, goodfile);
}
close(fsxgoodfd);
}
exit(status);
}
#define short_at(cp) ((unsigned short)((*((unsigned char *)(cp)) << 8) | \
*(((unsigned char *)(cp)) + 1)))
void
check_buffers(unsigned offset, unsigned size)
{
unsigned char c, t;
unsigned i = 0;
unsigned n = 0;
unsigned op = 0;
unsigned bad = 0;
if (memcmp(good_buf + offset, temp_buf, size) != 0) {
prt("READ BAD DATA: offset = 0x%x, size = 0x%x\n",
offset, size);
prt("OFFSET\tGOOD\tBAD\tRANGE\n");
while (size > 0) {
c = good_buf[offset];
t = temp_buf[i];
if (c != t) {
if (n == 0) {
bad = short_at(&temp_buf[i]);
prt("%#07x\t%#06x\t%#06x", offset,
short_at(&good_buf[offset]), bad);
op = temp_buf[offset & 1 ? i+1 : i];
}
n++;
badoff = offset;
}
offset++;
i++;
size--;
}
if (n) {
prt("\t%#7x\n", n);
if (bad)
prt("operation# (mod 256) for the bad data"
"may be %u\n", ((unsigned)op & 0xff));
else
prt("operation# (mod 256) for the bad data"
"unknown, check HOLE and EXTEND ops\n");
} else
prt("????????????????\n");
report_failure(110);
}
}
struct test_file {
char *path;
int fd;
int o_direct;
} *test_files = NULL;
int num_test_files = 0;
enum fd_iteration_policy {
FD_SINGLE,
FD_ROTATE,
FD_RANDOM,
};
int fd_policy = FD_RANDOM;
int fd_last = 0;
struct test_file *
get_tf(void)
{
unsigned index = 0;
switch (fd_policy) {
case FD_ROTATE:
index = fd_last++;
break;
case FD_RANDOM:
index = random();
break;
case FD_SINGLE:
index = 0;
break;
default:
prt("unknown policy");
exit(1);
break;
}
return &test_files[ index % num_test_files ];
}
void
assign_fd_policy(char *policy)
{
if (!strcmp(policy, "random"))
fd_policy = FD_RANDOM;
else if (!strcmp(policy, "rotate"))
fd_policy = FD_ROTATE;
else {
prt("unknown -I policy: '%s'\n", policy);
exit(1);
}
}
int
get_fd(void)
{
struct test_file *tf = get_tf();
return tf->fd;
}
static const char *my_basename(const char *path)
{
char *c = strrchr(path, '/');
return c ? c++ : path;
}
int do_fallocate(int fd, int mode, loff_t offset, loff_t length)
{
#ifdef FALLOC_FL_KEEP_SIZE
return fallocate(fd, mode, offset, length);
#else
#define FALLOC_FL_KEEP_SIZE 0x01
#if defined(__x86_64__)
#define __NR_fallocate 285
#elif defined(__i386__)
#define __NR_fallocate 324
#elif defined(__powerpc__)
#define __NR_fallocate 309
#elif defined(__s390__)
#define __NR_fallocate 314
#else
errno = ENOSYS;
return -1;
#endif
#if defined(__x86_64__) || defined(__powerpc64__) || defined(__s390x__)
/* 64-bit: pass the 64-bit offset and length arguments directly */
return syscall(__NR_fallocate, fd, mode, offset, length);
#elif defined(__i386__)
/* 32-bit x86: offset and length are split into high- and low- words
* order is offsetLo, offsetHi, lengthLo, lengthHi */
return syscall(__NR_fallocate, fd, mode,
(unsigned int)offset, (unsigned int)(offset >> 32),
(unsigned int)length, (unsigned int)(length >> 32));
#elif defined(__powerpc__)
/* 32-bit PPC: offset and length are split into high- and low- words
* order is offsetHi, offsetLo, lengthHi, lengthLo */
return syscall(__NR_fallocate, fd, mode,
(unsigned int)(offset >> 32), (unsigned int)offset,
(unsigned int)(length >> 32), (unsigned int)length);
#elif defined(__s390__)
/* 31-bit s390: length is split into high- and low- words, offset not
* order is offset, lengthHi, lengthLo */
return syscall(__NR_fallocate, fd, mode, offset,
(unsigned int)(length >> 32), (unsigned int)length);
#endif
#endif /* FALLOC_FL_KEEP_SIZE */
}
void
open_test_files(char **argv, int argc)
{
struct test_file *tf;
int i;
num_test_files = argc;
if (num_test_files == 1)
fd_policy = FD_SINGLE;
test_files = calloc(num_test_files, sizeof(*test_files));
if (test_files == NULL) {
prterr("reallocating space for test files");
exit(1);
}
for (i = 0, tf = test_files; i < num_test_files; i++, tf++) {
tf->path = argv[i];
tf->o_direct = (o_direct && (random() % o_direct == 0)) ?
O_DIRECT : 0;
tf->fd = open(tf->path, O_RDWR | (lite ? 0 : O_CREAT|O_TRUNC) |
tf->o_direct,
0666);
if (tf->fd < 0) {
prterr(tf->path);
exit(91);
}
if (prealloc == 1) {
if (do_fallocate(tf->fd, 0, 0, maxfilelen) < 0) {
(void)ftruncate(tf->fd, maxfilelen);
prt("%s: fallocate(0, %lu): %s\n",
tf->path, maxfilelen, strerror(errno));
//exit(89);
}
} else if (prealloc == 2) {
if (do_fallocate(tf->fd, FALLOC_FL_KEEP_SIZE, 0,
maxfilelen) < 0){
prt("%s: fallocate(KEEP_SIZE, 0, %lu): %s\n",
tf->path, maxfilelen, strerror(errno));
//exit(89);
}
}
}
if (quiet || fd_policy == FD_SINGLE)
return;
for (i = 0, tf = test_files; i < num_test_files; i++, tf++)
prt("fd %d: %s\n", i, tf->path);
}
void
close_test_files(void)
{
int i;
struct test_file *tf;
for (i = 0, tf = test_files; i < num_test_files; i++, tf++) {
if (close(tf->fd)) {
prterr("close");
report_failure(99);
}
}
}
void
check_size(void)
{
struct stat statbuf;
off_t size_by_seek;
int fd = get_fd();
if (fstat(fd, &statbuf)) {
prterr("check_size: fstat");
statbuf.st_size = -1;
}
size_by_seek = lseek(fd, (off_t)0, SEEK_END);
if (file_size != statbuf.st_size || file_size != size_by_seek) {
prt("Size error: expected 0x%llx stat 0x%llx seek 0x%llx\n",
(unsigned long long)file_size,
(unsigned long long)statbuf.st_size,
(unsigned long long)size_by_seek);
report_failure(120);
}
}
void
check_trunc_hack(void)
{
struct stat statbuf;
int fd = get_fd();
ftruncate(fd, (off_t)0);
ftruncate(fd, (off_t)100000);
if (fstat(fd, &statbuf)) {
prterr("trunc_hack: fstat");
statbuf.st_size = -1;
}
if (statbuf.st_size != (off_t)100000) {
prt("no extend on truncate! not posix!\n");
exit(130);
}
ftruncate(fd, 0);
}
static char *tf_buf = NULL;
static int max_tf_len = 0;
void
alloc_tf_buf(void)
{
char dummy = '\0';
int highest = num_test_files - 1;
int len;
len = snprintf(&dummy, 0, "%u ", highest);
if (len < 1) {
prterr("finding max tf_buf");
exit(1);
}
len++;
tf_buf = malloc(len);
if (tf_buf == NULL) {
prterr("allocating tf_buf");
exit(1);
}
max_tf_len = snprintf(tf_buf, len, "%u ", highest);
if (max_tf_len < 1) {
prterr("fiding max_tv_len\n");
exit(1);
}
if (max_tf_len != len - 1) {
warn("snprintf() gave %d instead of %d?\n",
max_tf_len, len - 1);
exit(1);
}
}
char *
fill_tf_buf(struct test_file *tf)
{
if (tf_buf == NULL)
alloc_tf_buf();
sprintf(tf_buf,"%lu ", (unsigned long)(tf - test_files));
return tf_buf;
}
void
output_line(struct test_file *tf, int op, unsigned offset,
unsigned size, struct timeval *tv)
{
char *tf_num = "";
char *ops[] = {
[OP_READ] = "read",
[OP_WRITE] = "write",
[OP_TRUNCATE] = "trunc from",
[OP_MAPREAD] = "mapread",
[OP_MAPWRITE] = "mapwrite",
[OP_READ + O_DIRECT] = "read_OD",
[OP_WRITE + O_DIRECT] = "write_OD",
};
if (fd_policy != FD_SINGLE)
tf_num = fill_tf_buf(tf);
/* W. */
if (!(!quiet && ((progressinterval &&
testcalls % progressinterval == 0) ||
(debug &&
(monitorstart == -1 ||
(offset + size > monitorstart &&
(monitorend == -1 || offset <= monitorend)))))))
return;
prt("%06lu %lu.%06lu %.*s%-10s %#08x %s %#08x\t(0x%05x bytes)\n",
testcalls, tv->tv_sec, tv->tv_usec, max_tf_len,
tf_num, ops[op],
offset, op == OP_TRUNCATE ? " to " : "thru",
offset + size - 1, (int)size < 0 ? -(int)size : size);
}
void output_debug(unsigned offset, unsigned size, const char *what)
{
if (!quiet && (debug > 1 &&
(monitorstart == -1 ||
(offset + size >= monitorstart &&
(monitorend == -1 || offset <= monitorend))))) {
struct timeval t;
gettimeofday(&t, NULL);
prt(" %lu.%06lu %s\n", t.tv_sec, t.tv_usec, what);
}
}
void
doread(unsigned offset, unsigned size)
{
struct timeval t;
off_t ret;
unsigned iret;
struct test_file *tf = get_tf();
int fd = tf->fd;
if (tf->o_direct) {
offset -= offset % (readbdy == 1 ? page_size : readbdy);
size += page_size - size % (readbdy == 1 ? page_size : readbdy);
} else {
offset -= offset % readbdy;
}
gettimeofday(&t, NULL);
if (size == 0) {
if (!quiet && testcalls > simulatedopcount && !tf->o_direct)
prt("skipping zero size read\n");
log4(OP_SKIPPED, OP_READ, offset, size, &t);
return;
}
if (size + offset > file_size) {
if (!quiet && testcalls > simulatedopcount)
prt("skipping seek/read past end of file\n");
log4(OP_SKIPPED, OP_READ, offset, size, &t);
return;
}
log4(OP_READ + tf->o_direct, offset, size, 0, &t);
if (testcalls <= simulatedopcount)
return;
output_line(tf, OP_READ + tf->o_direct, offset, size, &t);
ret = lseek(fd, (off_t)offset, SEEK_SET);
if (ret == (off_t)-1) {
prterr("doread: lseek");
report_failure(140);
}
iret = read(fd, temp_buf, size);
output_debug(offset, size, "read done");
if (iret != size) {
if (iret == -1)
prterr("doread: read");
else
prt("short read: 0x%x bytes instead of 0x%x\n",
iret, size);
report_failure(141);
}
check_buffers(offset, size);
}
void
check_eofpage(char *s, unsigned offset, char *p, int size)
{
long last_page, should_be_zero;
if (offset + size <= (file_size & ~page_mask))
return;
/*
* we landed in the last page of the file
* test to make sure the VM system provided 0's
* beyond the true end of the file mapping
* (as required by mmap def in 1996 posix 1003.1)
*/
last_page = ((long)p + (offset & page_mask) + size) & ~page_mask;
for (should_be_zero = last_page + (file_size & page_mask);
should_be_zero < last_page + page_size;
should_be_zero++)
if (*(char *)should_be_zero) {
prt("non-zero mmap past EOF %#llx page @ %#lx is %#x\n",
(long long)file_size -1, should_be_zero & page_mask,
short_at(should_be_zero));
report_failure(205);
}
}
void
domapread(unsigned offset, unsigned size)
{
struct timeval t;
unsigned pg_offset;
unsigned map_size;
char *p;
struct test_file *tf = get_tf();
int fd = tf->fd;
if (tf->o_direct) {
doread(offset, size);
return;
}
offset -= offset % readbdy;
gettimeofday(&t, NULL);
if (size == 0) {
if (!quiet && testcalls > simulatedopcount)
prt("skipping zero size read\n");
log4(OP_SKIPPED, OP_MAPREAD, offset, size, &t);
return;
}
if (size + offset > file_size) {
if (!quiet && testcalls > simulatedopcount)
prt("skipping seek/read past end of file\n");
log4(OP_SKIPPED, OP_MAPREAD, offset, size, &t);
return;
}
log4(OP_MAPREAD, offset, size, 0, &t);
if (testcalls <= simulatedopcount)
return;
output_line(tf, OP_MAPREAD, offset, size, &t);
pg_offset = offset & page_mask;
map_size = pg_offset + size;
if ((p = mmap(0, map_size, PROT_READ, MAP_FILE | MAP_SHARED, fd,
(off_t)(offset - pg_offset))) == MAP_FAILED) {
prterr("domapread: mmap");
report_failure(190);
}
output_debug(offset, size, "mmap done");
if (setjmp(jmpbuf) == 0) {
jmpbuf_good = 1;
memcpy(temp_buf, p + pg_offset, size);
check_eofpage("Read", offset, p, size);
jmpbuf_good = 0;
} else {
report_failure(1901);
}
output_debug(offset, size, "memcpy done");
if (munmap(p, map_size) != 0) {
prterr("domapread: munmap");
report_failure(191);
}
output_debug(offset, size, "munmap done");
check_buffers(offset, size);
}
void
gendata(char *original_buf, char *good_buf, unsigned offset, unsigned size)
{
while (size--) {
good_buf[offset] = testcalls % 256;
if (offset % 2)
good_buf[offset] += original_buf[offset];
offset++;
}
}
void
dowrite(unsigned offset, unsigned size)
{
struct timeval t;
off_t ret;
unsigned iret;
struct test_file *tf = get_tf();
int fd = tf->fd;
if (tf->o_direct) {
offset -= offset % (writebdy == 1 ? page_size : writebdy);
size += page_size - (size % (writebdy == 1 ? page_size : writebdy));
} else {
offset -= offset % writebdy;
}
gettimeofday(&t, NULL);
if (size == 0) {
if (!quiet && testcalls > simulatedopcount && !tf->o_direct)
prt("skipping zero size write\n");
log4(OP_SKIPPED, OP_WRITE, offset, size, &t);
return;
}
log4(OP_WRITE + tf->o_direct, offset, size, file_size, &t);
gendata(original_buf, good_buf, offset, size);
if (file_size < offset + size) {
if (file_size < offset)
memset(good_buf + file_size, '\0', offset - file_size);
file_size = offset + size;
if (lite) {
warn("Lite file size bug in fsx!");
report_failure(149);
}
}
if (testcalls <= simulatedopcount)
return;
output_line(tf, OP_WRITE + tf->o_direct, offset, size, &t);
ret = lseek(fd, (off_t)offset, SEEK_SET);
if (ret == (off_t)-1) {
prterr("dowrite: lseek");
report_failure(150);
}
iret = write(fd, good_buf + offset, size);
output_debug(offset, size, "write done");
if (iret != size) {
if (iret == -1)
prterr("dowrite: write");
else
prt("short write: 0x%x bytes instead of 0x%x\n",
iret, size);
report_failure(151);
}
if (do_fsync) {
if (fsync(fd)) {
prt("fsync() failed: %s\n", strerror(errno));
report_failure(152);
}
output_debug(offset, size, "fsync done");
}
}
void
domapwrite(unsigned offset, unsigned size)
{
struct timeval t;
unsigned pg_offset;
unsigned map_size;
off_t cur_filesize;
char *p;
struct test_file *tf = get_tf();
int fd = tf->fd;
if (tf->o_direct) {
dowrite(offset, size);
return;
}
offset -= offset % writebdy;
gettimeofday(&t, NULL);
if (size == 0) {
if (!quiet && testcalls > simulatedopcount)
prt("skipping zero size write\n");
log4(OP_SKIPPED, OP_MAPWRITE, offset, size, &t);
return;
}
cur_filesize = file_size;
log4(OP_MAPWRITE, offset, size, 0, &t);
gendata(original_buf, good_buf, offset, size);
if (file_size < offset + size) {
if (file_size < offset)
memset(good_buf + file_size, '\0', offset - file_size);
file_size = offset + size;
if (lite) {
warn("Lite file size bug in fsx!");
report_failure(200);
}
}
if (testcalls <= simulatedopcount)
return;
output_line(tf, OP_MAPWRITE, offset, size, &t);
if (file_size > cur_filesize) {
if (ftruncate(fd, file_size) == -1) {
prterr("domapwrite: ftruncate");
exit(201);
}
output_debug(offset, size, "truncate done");
}
pg_offset = offset & page_mask;
map_size = pg_offset + size;
if ((p = mmap(0, map_size, PROT_READ | PROT_WRITE, MAP_FILE|MAP_SHARED,
fd, (off_t)(offset - pg_offset))) == MAP_FAILED) {
prterr("domapwrite: mmap");
report_failure(202);
}
output_debug(offset, map_size, "mmap done");
if (setjmp(jmpbuf) == 0) {
jmpbuf_good = 1;
memcpy(p + pg_offset, good_buf + offset, size);
if (msync(p, map_size, 0) != 0) {
prterr("domapwrite: msync");
report_failure(203);
}
check_eofpage("Write", offset, p, size);
jmpbuf_good = 0;
} else {
report_failure(2021);
}
output_debug(offset, size, "memcpy done");
if (msync(p, map_size, 0) != 0) {
prterr("domapwrite: msync");
report_failure(203);
}
output_debug(offset, map_size, "msync done");
if (munmap(p, map_size) != 0) {
prterr("domapwrite: munmap");
report_failure(204);
}
output_debug(offset, map_size, "munmap done");
}
void
dotruncate(unsigned size)
{
struct timeval t;
int oldsize = file_size;
struct test_file *tf = get_tf();
int fd = tf->fd;
size -= size % truncbdy;
gettimeofday(&t, NULL);
if (size > biggest) {
biggest = size;
if (!quiet && testcalls > simulatedopcount)
prt("truncating to largest ever: 0x%x\n", size);
}
log4(OP_TRUNCATE, size, (unsigned)file_size, 0, &t);
if (size > file_size)
memset(good_buf + file_size, '\0', size - file_size);
file_size = size;
if (testcalls <= simulatedopcount)
return;
output_line(tf, OP_TRUNCATE, oldsize, size - oldsize, &t);
if (ftruncate(fd, (off_t)size) == -1) {
prt("ftruncate1: %x\n", size);
prterr("dotruncate: ftruncate");
report_failure(160);
}
output_debug(size, 0, "truncate done");
}
void
writefileimage()
{
ssize_t iret;
int fd = get_fd();
if (lseek(fd, (off_t)0, SEEK_SET) == (off_t)-1) {
prterr("writefileimage: lseek");
report_failure(171);
}
iret = write(fd, good_buf, file_size);
if ((off_t)iret != file_size) {
if (iret == -1)
prterr("writefileimage: write");
else
prt("short write: 0x%lx bytes instead of 0x%llx\n",
(unsigned long)iret,
(unsigned long long)file_size);
report_failure(172);
}
if (lite ? 0 : ftruncate(fd, file_size) == -1) {
prt("ftruncate2: %llx\n", (unsigned long long)file_size);
prterr("writefileimage: ftruncate");
report_failure(173);
}
}
void
docloseopen(void)
{
struct timeval t;
struct test_file *tf = get_tf();
int direct = (o_direct && (random() % o_direct == 0)) ? O_DIRECT : 0;
char *tf_num = "";
if (testcalls <= simulatedopcount)
return;
gettimeofday(&t, NULL);
log4(OP_CLOSEOPEN + direct, file_size, (unsigned)file_size, 0, &t);
if (fd_policy != FD_SINGLE)
tf_num = fill_tf_buf(tf);
if (debug)
prt("%06lu %lu.%06lu %sclose/open%s\n", testcalls, t.tv_sec,
t.tv_usec, tf_num, direct ? "(O_DIRECT)" : "");
if (close(tf->fd)) {
prterr(tf->o_direct ?
"docloseopen: close(O_DIRECT)" : "docloseopen: close");
report_failure(180);
}
output_debug(monitorstart, 0, "close done");
tf->o_direct = direct;
tf->fd = open(tf->path, O_RDWR | tf->o_direct, 0);
if (tf->fd < 0) {
prterr(tf->o_direct ?
"docloseopen: open(O_DIRECT)" : "docloseopen: open");
report_failure(181);
}
output_debug(monitorstart, 0,
tf->o_direct ? "open done" : "open(O_DIRECT) done");
}
void
test(void)
{
unsigned long offset;
unsigned long size = maxoplen;
unsigned long rv = random();
unsigned long op = rv % (3 + !lite + mapped_writes);
/* turn off the map read if necessary */
if (op == 2 && !mapped_reads)
op = 0;
if (simulatedopcount > 0 && testcalls == simulatedopcount)
writefileimage();
testcalls++;
if (debugstart > 0 && testcalls >= debugstart)
debug = 1;
if (!quiet && testcalls < simulatedopcount && testcalls % 100000 == 0)
prt("%lu...\n", testcalls);
/*
* READ: op = 0
* WRITE: op = 1
* MAPREAD: op = 2
* TRUNCATE: op = 3
* MAPWRITE: op = 3 or 4
*/
if (lite ? 0 : op == 3 && (style & 1) == 0) /* vanilla truncate? */
dotruncate(random() % maxfilelen);
else {
if (randomoplen)
size = random() % (maxoplen+1);
if (lite ? 0 : op == 3)
dotruncate(size);
else {
offset = random();
if (op == 1 || op == (lite ? 3 : 4)) {
offset %= maxfilelen;
if (offset + size > maxfilelen)
size = maxfilelen - offset;
if (op != 1)
domapwrite(offset, size);
else
dowrite(offset, size);
} else {
if (file_size)
offset %= file_size;
else
offset = 0;
if (offset + size > file_size)
size = file_size - offset;
if (op != 0)
domapread(offset, size);
else
doread(offset, size);
}
}
}
if (sizechecks && testcalls > simulatedopcount)
check_size();
if (closeprob && (rv >> 3) < (1 << 28) / closeprob)
docloseopen();
}
void
segv(int sig)
{
if (jmpbuf_good) {
jmpbuf_good = 0;
longjmp(jmpbuf, 1);
}
report_failure(9999);
}
void
cleanup(sig)
int sig;
{
if (sig)
prt("signal %d\n", sig);
prt("testcalls = %lu\n", testcalls);
exit(sig);
}
void
usage(void)
{
fprintf(stdout, "usage: fsx [-dfnqLOW] [-b opnum] [-c Prob] [-l flen]\n"
"\t\t[-m start:end] [-o oplen] [-p progressinterval] [-r readbdy] [-s style]\n"
"\t\t[-t truncbdy] [-w writebdy] [-D startingop] [-N numops] [-P dirpath]\n"
"\t\t[-S seed] [-Z [prob]] [ -I random|rotate ] fname [more paths to fname..]\n"
" -b opnum: beginning operation number (default 1)\n"
" -c P: 1 in P chance of file close+open at each op (default infinity)\n"
" -d: debug output for all operations [-d -d = more debugging]\n"
/* OSX: -d duration: number of hours for the tool to run\n" */
/* OSX: -e: tests using an extended attribute rather than a file\n" */
" -f: fsync after every write operation\n"
/* OSX: -f forkname: test the named fork of fname\n" */
/* OSX: -g logpath: path for .fsxlog file\n" */
/* OSX: -h: write 0s instead of creating holes (i.e. sparse file)\n" */
/* OSX: -i: interactive mode, hit return before performing each operation\n" */
" -l flen: the upper bound on file size (default 262144)\n"
/* OSX: -l logpath: path for XILog file\n" */
" -m startop:endop: monitor (print debug output) specified byte rang"
"(default 0:infinity)\n"
" -n: no verifications of file size\n"
" -o oplen: the upper bound on operation size (default 65536)\n"
" -p progressinterval: debug output at specified operation interval\n"
" -q: quieter operation\n"
" -r readbdy: %u would make reads page aligned (default 1)\n"
" -s style: 1 gives smaller truncates (default 0)\n"
" -t truncbdy: %u would make truncates page aligned (default 1)\n"
/* OSX: -v: debug output for all operations\n" */
" -w writebdy: %u would make writes page aligned (default 1)\n"
" -x[1|2]: preallocate file space (2 does no size update)\n"
/* OSX: -x: write output in XML (XILOG)\n" */
/* OSX: -y: call fsync before closing the file\n" */
/* AKPM:-A: Use the AIO system calls\n" */
/* OSX: -C mix cached and un-cached read/write ops\n" */
" -D startingop: debug output starting at specified operation\n"
/* OSX: -F flen: the upper bound on file size (default 262144)\n" */
/* OSX: -G logsize: #entries in oplog (default 1024)\n" */
" -I {rotate|random}: When multiple paths to the file are given,\n"
" each operation uses a different path. Iterate through them in\n"
" order with 'rotate' or chose then at 'random'. (defaults random)\n"
/* OSX: -I: start interactive mode since operation opnum\n" */
" -L: fsxLite - no file creations & no file size changes\n"
/* OSX: -M: slow motion mode, wait 1 second before each op\n" */
" -N numops: total # operations to do (default infinity)\n"
" -O: use oplen (see -o flag) for every op (default random)\n"
" -P: save .fsxlog and .fsxgood files in dirpath (default ./)\n"
" -R: read() system calls only (mapped reads disabled)\n"
" -S seed: for random # generator (default 1) 0 gets timestamp\n"
/* OSX: -T datasize: size of atomic data element writes [1,2,4] (default 4)\n" */
" -W: mapped write operations DISabled\n"
" -Z[P]: O_DIRECT file IO [1 in P chance for each open] (default off)\n"
" fname: this filename is REQUIRED (no default)\n",
page_size, page_size, page_size);
exit(90);
}
int
getnum(char *s, char **e)
{
int ret = -1;
*e = (char *) 0;
ret = strtol(s, e, 0);
if (*e)
switch (**e) {
case 'b':
case 'B':
ret *= 512;
*e = *e + 1;
break;
case 'k':
case 'K':
ret *= 1024;
*e = *e + 1;
break;
case 'm':
case 'M':
ret *= 1024*1024;
*e = *e + 1;
break;
case 'w':
case 'W':
ret *= 4;
*e = *e + 1;
break;
}
return (ret);
}
int
main(int argc, char **argv)
{
int i, style, ch;
char *orig_good_buf, *orig_temp_buf;
char *endp;
int dirpath = 0;
goodfile[0] = 0;
logfile[0] = 0;
page_size = getpagesize();
page_mask = page_size - 1;
setvbuf(stdout, (char *)0, _IOLBF, 0); /* line buffered stdout */
while ((ch = getopt(argc, argv,
"b:c:dfl:m:no:p:qr:s:t:w:x::D:I:LN:OP:RS:WZ::"))
!= EOF)
switch (ch) {
case 'b':
simulatedopcount = getnum(optarg, &endp);
if (!quiet)
fprintf(stdout, "Will begin at operation %ld\n",
simulatedopcount);
if (simulatedopcount == 0)
usage();
simulatedopcount -= 1;
break;
case 'c':
closeprob = getnum(optarg, &endp);
if (!quiet)
fprintf(stdout,
"Chance of close/open is 1 in %d\n",
closeprob);
if (closeprob <= 0)
usage();
break;
case 'd':
debug++;
break;
case 'f':
do_fsync = 1;
break;
case 'l':
maxfilelen = getnum(optarg, &endp);
if (maxfilelen <= 0)
usage();
break;
case 'm':
monitorstart = getnum(optarg, &endp);
if (monitorstart < 0)
usage();
if (!endp || *endp++ != ':')
usage();
monitorend = getnum(endp, &endp);
if (monitorend < 0)
usage();
if (monitorend == 0)
monitorend = -1; /* aka infinity */
debug = 1;
case 'n':
sizechecks = 0;
break;
case 'o':
maxoplen = getnum(optarg, &endp);
if (maxoplen <= 0)
usage();
break;
case 'p':
progressinterval = getnum(optarg, &endp);
if (progressinterval < 0)
usage();
break;
case 'q':
quiet = 1;
break;
case 'r':
readbdy = getnum(optarg, &endp);
if (readbdy <= 0)
usage();
break;
case 's':
style = getnum(optarg, &endp);
if (style < 0 || style > 1)
usage();
break;
case 't':
truncbdy = getnum(optarg, &endp);
if (truncbdy <= 0)
usage();
break;
case 'w':
writebdy = getnum(optarg, &endp);
if (writebdy <= 0)
usage();
break;
case 'x':
if (optarg == NULL || (prealloc = getnum(optarg, &endp)) <= 0)
prealloc = 1;
break;
case 'D':
debugstart = getnum(optarg, &endp);
if (debugstart < 1)
usage();
break;
case 'I':
assign_fd_policy(optarg);
break;
case 'L':
lite = 1;
break;
case 'N':
numops = getnum(optarg, &endp);
if (numops < 0)
usage();
break;
case 'O':
randomoplen = 0;
break;
case 'P':
strncpy(goodfile, optarg, sizeof(goodfile));
strcat(goodfile, "/");
strncpy(logfile, optarg, sizeof(logfile));
strcat(logfile, "/");
dirpath = 1;
break;
case 'R':
mapped_reads = 0;
break;
case 'S':
seed = getnum(optarg, &endp);
if (seed == 0)
seed = time(0) % 10000;
if (!quiet)
fprintf(stdout, "Seed set to %d\n", seed);
if (seed < 0)
usage();
break;
case 'W':
mapped_writes = 0;
if (!quiet)
fprintf(stdout, "mapped writes DISABLED\n");
break;
case 'Z':
if (optarg == NULL || (o_direct = getnum(optarg, &endp)) == 0)
o_direct = 1;
break;
default:
usage();
/* NOTREACHED */
}
argc -= optind;
argv += optind;
if (argc < 1)
usage();
fname = argv[0];
signal(SIGHUP, cleanup);
signal(SIGINT, cleanup);
signal(SIGPIPE, cleanup);
signal(SIGALRM, cleanup);
signal(SIGTERM, cleanup);
signal(SIGXCPU, cleanup);
signal(SIGXFSZ, cleanup);
signal(SIGVTALRM, cleanup);
signal(SIGUSR1, cleanup);
signal(SIGUSR2, cleanup);
signal(SIGSEGV, segv);
initstate(seed, state, 256);
setstate(state);
open_test_files(argv, argc);
strncat(goodfile, dirpath ? my_basename(fname) : fname, 256);
strcat (goodfile, ".fsxgood");
fsxgoodfd = open(goodfile, O_RDWR|O_CREAT|O_TRUNC, 0666);
if (fsxgoodfd < 0) {
prterr(goodfile);
exit(92);
}
strncat(logfile, dirpath ? my_basename(fname) : fname, 256);
strcat (logfile, ".fsxlog");
fsxlogf = fopen(logfile, "w");
if (fsxlogf == NULL) {
prterr(logfile);
exit(93);
}
if (lite) {
off_t ret;
int fd = get_fd();
file_size = maxfilelen = lseek(fd, (off_t)0, SEEK_END);
if (file_size == (off_t)-1) {
prterr(fname);
warn("main: lseek eof");
exit(94);
}
ret = lseek(fd, (off_t)0, SEEK_SET);
if (ret == (off_t)-1) {
prterr(fname);
warn("main: lseek 0");
exit(95);
}
}
original_buf = (char *) malloc(maxfilelen);
if (original_buf == NULL)
exit(96);
for (i = 0; i < maxfilelen; i++)
original_buf[i] = random() % 256;
orig_good_buf = malloc(maxfilelen + page_size);
if (orig_good_buf == NULL)
exit(97);
good_buf = round_up(orig_good_buf, page_size);
memset(good_buf, '\0', maxfilelen);
orig_temp_buf = malloc(maxoplen + page_size);
if (orig_temp_buf == NULL)
exit(99);
temp_buf = round_up(orig_temp_buf, page_size);
memset(temp_buf, '\0', maxoplen);
if (lite) { /* zero entire existing file */
ssize_t written;
int fd = get_fd();
written = write(fd, good_buf, (size_t)maxfilelen);
if (written != maxfilelen) {
if (written == -1) {
prterr(fname);
warn("main: error on write");
} else
warn("main: short write, 0x%x bytes instead"
"of 0x%x\n",
(unsigned)written, maxfilelen);
exit(98);
}
} else {
check_trunc_hack();
}
while (numops == -1 || numops--)
test();
close_test_files();
prt("All operations completed A-OK!\n");
if (tf_buf)
free(tf_buf);
free(original_buf);
free(orig_good_buf);
free(orig_temp_buf);
return 0;
}
Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.
Andreas Dilger wrote:
> I've done some work to merge some of the existing fsx.c mods into a
> single version. Over & above the version that is in the LTP, I've
> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
> segfault handling, and per-write fsync.
>
> The twist for the O_DIRECT feature is that it will randomly open file
> descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
> support (by specifying multiple pathnames for the output file) fsx will
> be testing buffered and O_DIRECT and mmap IO on the same file.
>
> Updated patch to have proper fallocate() handling in case glibc doesn't
> have this, for non-x86 architectures from MingMing Cao, based on code
> used by DB2.
>
> Signed-off-by: Andreas Dilger <[email protected]>
>
And here's a patch on top of that to implement random fallocate
calls during the runs rather than just one at the beginning.
This was a fairly quick patch merge from one I did for xfstests'
version of fsx, but it seems to run properly.
Signed-off-by: Eric Sandeen <[email protected]>
--- fsx.c.orig 2009-09-29 21:40:31.045330451 -0500
+++ fsx.c 2009-09-29 22:30:25.670330430 -0500
@@ -37,11 +37,12 @@
*
* Add multi-file testing feature -- Zach Brown <[email protected]>
*
+ * Add random preallocation calls - Eric Sandeen <[email protected]>
+ *
* $FreeBSD: src/tools/regression/fsx/fsx.c,v 1.2 2003/04/23 23:42:23 jkh Exp $
* $DragonFly: src/test/stress/fsx/fsx.c,v 1.2 2005/05/02 19:31:56 dillon Exp $
*
*/
-
#include <sys/types.h>
#include <sys/stat.h>
#if defined(_UWIN) || defined(__linux__)
@@ -95,6 +96,7 @@
#define OP_MAPREAD 5
#define OP_MAPWRITE 6
#define OP_SKIPPED 7
+#define OP_FALLOCATE 8
int page_size;
int page_mask;
@@ -127,6 +129,7 @@
int writebdy = 1; /* -w flag */
long monitorstart = -1; /* -m flag */
long monitorend = -1; /* -m flag */
+int fallocate_calls = 1; /* -F flag disables */
int lite = 0; /* -L flag */
long numops = -1; /* -N flag */
int randomoplen = 1; /* -O flag disables it */
@@ -219,6 +222,7 @@
{
int i, count, down;
struct log_entry *lp;
+ char *falloc_type[3] = {"PAST_EOF", "EXTENDING", "INTERIOR"};
prt("LOG DUMP (%d total operations):\n", logcount);
if (logcount < LOGSIZE) {
@@ -290,6 +294,14 @@
prt("CLOSE/OPEN%s",
lp->operation & O_DIRECT ? "_OD" : " ");
break;
+ case OP_FALLOCATE:
+ /* 0: offset 1: length 2: where alloced */
+ prt("FALLOCATE %s\tfrom 0x%x to 0x%x",
+ falloc_type[lp->args[2]], lp->args[0], lp->args[0] + lp->args[1]);
+ if (badoff >= lp->args[0] &&
+ badoff < lp->args[0] + lp->args[1])
+ prt("\t******FFFF");
+ break;
case OP_SKIPPED:
prt("SKIPPED (no operation)");
break;
@@ -687,6 +699,7 @@
[OP_MAPWRITE] = "mapwrite",
[OP_READ + O_DIRECT] = "read_OD",
[OP_WRITE + O_DIRECT] = "write_OD",
+ [OP_FALLOCATE] = "fallocate",
};
if (fd_policy != FD_SINGLE)
@@ -1068,6 +1081,59 @@
output_debug(size, 0, "truncate done");
}
+/* fallocate is basically a no-op unless extending, then a lot like a truncate */
+void
+dofallocate(unsigned offset, unsigned length)
+{
+ struct timeval t;
+ unsigned end_offset;
+ int keep_size;
+ struct test_file *tf = get_tf();
+ int fd = tf->fd;
+
+ gettimeofday(&t, NULL);
+ if (length == 0) {
+ if (!quiet && testcalls > simulatedopcount)
+ prt("skipping zero length fallocate\n");
+ log4(OP_SKIPPED, OP_FALLOCATE, offset, length, &t);
+ return;
+ }
+
+ keep_size = random() % 2;
+
+ end_offset = keep_size ? 0 : offset + length;
+
+ if (end_offset > biggest) {
+ biggest = end_offset;
+ if (!quiet && testcalls > simulatedopcount)
+ prt("fallocating to largest ever: 0x%x\n", end_offset);
+ }
+
+ /*
+ * last arg:
+ * 1: allocate past EOF
+ * 2: extending prealloc
+ * 3: interior prealloc
+ */
+ log4(OP_FALLOCATE, offset, length, (end_offset > file_size) ? (keep_size ? 1 : 2) : 3, &t);
+
+ if (end_offset > file_size) {
+ memset(good_buf + file_size, '\0', end_offset - file_size);
+ file_size = end_offset;
+ }
+
+ if (testcalls <= simulatedopcount)
+ return;
+
+ output_line(tf, OP_FALLOCATE, offset, length, &t);
+
+ if (do_fallocate(fd, keep_size ? FALLOC_FL_KEEP_SIZE : 0, (loff_t)offset, (loff_t)length) == -1) {
+ prt("fallocate: %x to %x\n", offset, length);
+ prterr("dofallocate: fallocate");
+ report_failure(161);
+ }
+ output_debug(offset, length, "fallocate done");
+}
void
writefileimage()
@@ -1141,7 +1207,7 @@
unsigned long offset;
unsigned long size = maxoplen;
unsigned long rv = random();
- unsigned long op = rv % (3 + !lite + mapped_writes);
+ unsigned long op = rv % (3 + !lite + mapped_writes + fallocate_calls);
/* turn off the map read if necessary */
@@ -1160,22 +1226,32 @@
prt("%lu...\n", testcalls);
/*
- * READ: op = 0
- * WRITE: op = 1
- * MAPREAD: op = 2
- * TRUNCATE: op = 3
- * MAPWRITE: op = 3 or 4
+ * lite !lite
+ * READ: op = 0 0
+ * WRITE: op = 1 1
+ * MAPREAD: op = 2 2
+ * TRUNCATE: op = - 3
+ * MAPWRITE: op = 3 4
+ * FALLOCATE: op = - 5
*/
if (lite ? 0 : op == 3 && (style & 1) == 0) /* vanilla truncate? */
dotruncate(random() % maxfilelen);
else {
if (randomoplen)
size = random() % (maxoplen+1);
+ /* truncate */
if (lite ? 0 : op == 3)
dotruncate(size);
else {
offset = random();
- if (op == 1 || op == (lite ? 3 : 4)) {
+ /* fallocate */
+ if (op == 5) {
+ offset %= maxfilelen;
+ if (offset + size > maxfilelen)
+ size = maxfilelen - offset;
+ dofallocate(offset, size);
+ /* write / mapwrite */
+ } else if (op == 1 || op == (lite ? 3 : 4)) {
offset %= maxfilelen;
if (offset + size > maxfilelen)
size = maxfilelen - offset;
@@ -1183,6 +1259,7 @@
domapwrite(offset, size);
else
dowrite(offset, size);
+ /* read / mapread */
} else {
if (file_size)
offset %= file_size;
@@ -1227,7 +1304,7 @@
void
usage(void)
{
- fprintf(stdout, "usage: fsx [-dfnqLOW] [-b opnum] [-c Prob] [-l flen]\n"
+ fprintf(stdout, "usage: fsx [-dfnqFLOW] [-b opnum] [-c Prob] [-l flen]\n"
"\t\t[-m start:end] [-o oplen] [-p progressinterval] [-r readbdy] [-s style]\n"
"\t\t[-t truncbdy] [-w writebdy] [-D startingop] [-N numops] [-P dirpath]\n"
"\t\t[-S seed] [-Z [prob]] [ -I random|rotate ] fname [more paths to fname..]\n"
@@ -1266,6 +1343,7 @@
" each operation uses a different path. Iterate through them in\n"
" order with 'rotate' or chose then at 'random'. (defaults random)\n"
/* OSX: -I: start interactive mode since operation opnum\n" */
+" -F: Do not use fallocate (preallocation) calls\n"
" -L: fsxLite - no file creations & no file size changes\n"
/* OSX: -M: slow motion mode, wait 1 second before each op\n" */
" -N numops: total # operations to do (default infinity)\n"
@@ -1332,7 +1410,7 @@
setvbuf(stdout, (char *)0, _IOLBF, 0); /* line buffered stdout */
while ((ch = getopt(argc, argv,
- "b:c:dfl:m:no:p:qr:s:t:w:x::D:I:LN:OP:RS:WZ::"))
+ "b:c:dfl:m:no:p:qr:s:t:w:x::D:FI:LN:OP:RS:WZ::"))
!= EOF)
switch (ch) {
case 'b':
@@ -1421,6 +1499,9 @@
if (debugstart < 1)
usage();
break;
+ case 'F':
+ fallocate_calls = 0;
+ break;
case 'I':
assign_fd_policy(optarg);
break;
@@ -1557,6 +1638,15 @@
check_trunc_hack();
}
+ if (!lite && fallocate_calls) {
+ int fd = get_fd();
+ if (do_fallocate(fd, 0, 0, 1) && errno == EOPNOTSUPP) {
+ warn("main: filesystem does not support fallocate, disabling");
+ fallocate_calls = 0;
+ } else
+ ftruncate(fd, 0);
+ }
+
while (numops == -1 || numops--)
test();
On Tue, 29 Sep 2009 16:39:38 -0600 Andreas Dilger wrote:
> I've done some work to merge some of the existing fsx.c mods into a
> single version. Over & above the version that is in the LTP, I've
> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
> segfault handling, and per-write fsync.
It sure would be Good if fsx-linux had a fixed, known home, instead of
having to look at Andrew's ext3-tools, Dave's codemonkey files,
LTP, linux-fsdevel mailing list, etc.
Is there a linux-fs-tests package somewhere? (like where the xfstests
are being merged into)
thanks,
---
~Randy
Randy Dunlap wrote:
> On Tue, 29 Sep 2009 16:39:38 -0600 Andreas Dilger wrote:
>
>> I've done some work to merge some of the existing fsx.c mods into a
>> single version. Over & above the version that is in the LTP, I've
>> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
>> segfault handling, and per-write fsync.
>
> It sure would be Good if fsx-linux had a fixed, known home, instead of
> having to look at Andrew's ext3-tools, Dave's codemonkey files,
> LTP, linux-fsdevel mailing list, etc.
>
> Is there a linux-fs-tests package somewhere? (like where the xfstests
> are being merged into)
>
>
> thanks,
> ---
> ~Randy
Well, I think Andreas' idea was to make LTP the official home for now
which is why I was going to send it here ...
But I don't think xfstests is being merged into anything at the moment;
if anything, we're working to make it a more generic suite of tests (it
can run about 50 tests on generic posix filesystems by now).
-Eric
On Wed, Sep 30, 2009 at 10:32:37AM -0500, Eric Sandeen wrote:
> Randy Dunlap wrote:
> > On Tue, 29 Sep 2009 16:39:38 -0600 Andreas Dilger wrote:
> >
> >> I've done some work to merge some of the existing fsx.c mods into a
> >> single version. Over & above the version that is in the LTP, I've
> >> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
> >> segfault handling, and per-write fsync.
> >
> > It sure would be Good if fsx-linux had a fixed, known home, instead of
> > having to look at Andrew's ext3-tools, Dave's codemonkey files,
> > LTP, linux-fsdevel mailing list, etc.
> >
> > Is there a linux-fs-tests package somewhere? (like where the xfstests
> > are being merged into)
>
> Well, I think Andreas' idea was to make LTP the official home for now
> which is why I was going to send it here ...
>
> But I don't think xfstests is being merged into anything at the moment;
> if anything, we're working to make it a more generic suite of tests (it
> can run about 50 tests on generic posix filesystems by now).
I'd love to see a tests/ directory shipped with the kernel.
It's been talked about for about several years and still hasn't happened.
Is it just because no-one has stepped up to do it ?
I would be happy to manage a git tree for marshalling stuff on its way to Linus
if that's all that's holding this idea back.
Of course if there are other objections..
Dave
Dave Jones wrote:
> On Wed, Sep 30, 2009 at 10:32:37AM -0500, Eric Sandeen wrote:
> > Randy Dunlap wrote:
> > > On Tue, 29 Sep 2009 16:39:38 -0600 Andreas Dilger wrote:
> > >
> > >> I've done some work to merge some of the existing fsx.c mods into a
> > >> single version. Over & above the version that is in the LTP, I've
> > >> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
> > >> segfault handling, and per-write fsync.
> > >
> > > It sure would be Good if fsx-linux had a fixed, known home, instead of
> > > having to look at Andrew's ext3-tools, Dave's codemonkey files,
> > > LTP, linux-fsdevel mailing list, etc.
> > >
> > > Is there a linux-fs-tests package somewhere? (like where the xfstests
> > > are being merged into)
> >
> > Well, I think Andreas' idea was to make LTP the official home for now
> > which is why I was going to send it here ...
> >
> > But I don't think xfstests is being merged into anything at the moment;
> > if anything, we're working to make it a more generic suite of tests (it
> > can run about 50 tests on generic posix filesystems by now).
>
> I'd love to see a tests/ directory shipped with the kernel.
> It's been talked about for about several years and still hasn't happened.
>
> Is it just because no-one has stepped up to do it ?
> I would be happy to manage a git tree for marshalling stuff on its way to Linus
> if that's all that's holding this idea back.
>
> Of course if there are other objections..
>
> Dave
>
Perhaps for simple, small things like fsx.c that might make sense, but a
full-blown test suite could add significantly to the size of the tarball
I'd think (especially by the time you start including things like
including pre-corrupted images, perhaps)...
-Eric
On Tue, 2009-09-29 at 16:39 -0600, Andreas Dilger wrote:
> I've done some work to merge some of the existing fsx.c mods into a
> single version. Over & above the version that is in the LTP, I've
> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
> segfault handling, and per-write fsync.
>
> The twist for the O_DIRECT feature is that it will randomly open file
> descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
> support (by specifying multiple pathnames for the output file) fsx will
> be testing buffered and O_DIRECT and mmap IO on the same file.
>
> Updated patch to have proper fallocate() handling in case glibc doesn't
> have this, for non-x86 architectures from MingMing Cao, based on code
> used by DB2.
>
> Signed-off-by: Andreas Dilger <[email protected]>
On top of this, a little update to add O_DIRECT define.
Signed-off-by: Mingming Cao <[email protected]>
--- fsx.c.orig 2009-10-02 14:22:12.299565348 -0700
+++ fsx.c 2009-10-01 16:36:17.271593794 -0700
@@ -78,6 +78,10 @@
#define LOGSIZE 100000
+#ifndef O_DIRECT
+#define O_DIRECT 040000 /* direct disk access hint */
+#endif
+
struct log_entry oplog[LOGSIZE]; /* the log */
int logptr = 0; /* current position in log */
int logcount = 0; /* total ops */
Mingming wrote:
> On Tue, 2009-09-29 at 16:39 -0600, Andreas Dilger wrote:
>> I've done some work to merge some of the existing fsx.c mods into a
>> single version. Over & above the version that is in the LTP, I've
>> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
>> segfault handling, and per-write fsync.
>>
>> The twist for the O_DIRECT feature is that it will randomly open file
>> descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
>> support (by specifying multiple pathnames for the output file) fsx will
>> be testing buffered and O_DIRECT and mmap IO on the same file.
>>
>> Updated patch to have proper fallocate() handling in case glibc doesn't
>> have this, for non-x86 architectures from MingMing Cao, based on code
>> used by DB2.
>>
>> Signed-off-by: Andreas Dilger <[email protected]>
>
> On top of this, a little update to add O_DIRECT define.
>
> Signed-off-by: Mingming Cao <[email protected]>
>
> --- fsx.c.orig 2009-10-02 14:22:12.299565348 -0700
> +++ fsx.c 2009-10-01 16:36:17.271593794 -0700
> @@ -78,6 +78,10 @@
>
> #define LOGSIZE 100000
>
> +#ifndef O_DIRECT
> +#define O_DIRECT 040000 /* direct disk access hint */
> +#endif
> +
> struct log_entry oplog[LOGSIZE]; /* the log */
> int logptr = 0; /* current position in log */
> int logcount = 0; /* total ops */
>
>
I think that
#define _GNU_SOURCE
will pull in the glibc headers' definition rather than doing our own....
-Eric
On Fri, 2009-10-02 at 17:40 -0500, Eric Sandeen wrote:
> Mingming wrote:
> > On Tue, 2009-09-29 at 16:39 -0600, Andreas Dilger wrote:
> >> I've done some work to merge some of the existing fsx.c mods into a
> >> single version. Over & above the version that is in the LTP, I've
> >> included AKPM's O_DIRECT fixes (with a twist), the BSD mmap page and
> >> segfault handling, and per-write fsync.
> >>
> >> The twist for the O_DIRECT feature is that it will randomly open file
> >> descriptors with O_DIRECT, and if you use the Lustre-inspired multi-fd
> >> support (by specifying multiple pathnames for the output file) fsx will
> >> be testing buffered and O_DIRECT and mmap IO on the same file.
> >>
> >> Updated patch to have proper fallocate() handling in case glibc doesn't
> >> have this, for non-x86 architectures from MingMing Cao, based on code
> >> used by DB2.
> >>
> >> Signed-off-by: Andreas Dilger <[email protected]>
> >
> > On top of this, a little update to add O_DIRECT define.
> >
> > Signed-off-by: Mingming Cao <[email protected]>
> >
> > --- fsx.c.orig 2009-10-02 14:22:12.299565348 -0700
> > +++ fsx.c 2009-10-01 16:36:17.271593794 -0700
> > @@ -78,6 +78,10 @@
> >
> > #define LOGSIZE 100000
> >
> > +#ifndef O_DIRECT
> > +#define O_DIRECT 040000 /* direct disk access hint */
> > +#endif
> > +
> > struct log_entry oplog[LOGSIZE]; /* the log */
> > int logptr = 0; /* current position in log */
> > int logcount = 0; /* total ops */
> >
> >
>
> I think that
>
> #define _GNU_SOURCE
>
> will pull in the glibc headers' definition rather than doing our own....
>
Oh? that's sounds good...
I tried to replace above with
#define _GNU_SOURCE 1
gcc complained O_DIRECT is still not defined. what did I missing?
Mingming
> -Eric
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
Mingming wrote:
> > I think that
> >
> > #define _GNU_SOURCE
> >
> > will pull in the glibc headers' definition rather than doing our own....
> >
>
> Oh? that's sounds good...
>
> I tried to replace above with
> #define _GNU_SOURCE 1
>
> gcc complained O_DIRECT is still not defined. what did I missing?
>
,----
| #define _GNU_SOURCE 1
|
| #include <fcntl.h>
`----
should do it.
Nick
Nick Dokos wrote:
> Mingming wrote:
>>> I think that
>>>
>>> #define _GNU_SOURCE
>>>
>>> will pull in the glibc headers' definition rather than doing our own....
>>>
>> Oh? that's sounds good...
>>
>> I tried to replace above with
>> #define _GNU_SOURCE 1
>>
>> gcc complained O_DIRECT is still not defined. what did I missing?
>>
>
>
> ,----
> | #define _GNU_SOURCE 1
> |
> | #include <fcntl.h>
> `----
>
> should do it.
>
> Nick
yep, works for me as long as I define it ahead of any #includes (not
sure just fcntl is enough, but anyway; just put it at the top).
-Eric
--- fsx.c.orig 2009-10-02 22:40:15.816142278 -0500
+++ fsx.c 2009-10-02 22:40:26.886145268 -0500
@@ -42,6 +42,8 @@
*
*/
+#define _GNU_SOURCE
+
#include <sys/types.h>
#include <sys/stat.h>
#if defined(_UWIN) || defined(__linux__)