4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, Intel Corporation.
34 * This file is part of Lustre, http://www.lustre.org/
35 * Lustre is a trademark of Sun Microsystems, Inc.
37 * lustre/utils/llverdev.c
39 * Large Block Device Verification Tool.
40 * This program is used to test whether the block device is correctly
41 * handling IO beyond 2TB boundary.
42 * This tool have two working modes
46 * In full mode, the program writes a test pattern on the entire disk.
47 * The test pattern (device offset and timestamp) is written at the
48 * beginning of each 4kB block. When the whole device is full the read
49 * operation is performed to verify that the test pattern is correct.
51 * In partial mode, the program writes data at the critical locations
52 * of the device such as start of the device, before and after multiple of 1GB
53 * offset and at the end.
55 * A chunk buffer with default size of 1MB is used to write and read test
65 #ifndef _LARGEFILE64_SOURCE
66 #define _LARGEFILE64_SOURCE
68 #ifndef _FILE_OFFSET_BITS
69 #define _FILE_OFFSET_BITS 64
84 #include <sys/types.h>
86 #include <sys/ioctl.h>
87 #include <sys/mount.h>
89 #include <gnu/stubs.h>
91 #ifdef HAVE_EXT2FS_EXT2FS_H
92 # include <ext2fs/ext2fs.h>
95 #define ONE_MB (1024 * 1024)
96 #define ONE_GB (1024 * 1024 * 1024)
97 #define HALF_MB (ONE_MB / 2)
99 #define HALF_KB (ONE_KB / 2)
100 #define BLOCKSIZE 4096
102 /* Structure for writting test pattern */
104 unsigned long long bd_offset;
105 unsigned long long bd_time;
106 unsigned long long bd_inode;
108 static char *progname; /* name by which this program was run. */
109 static unsigned verbose = 1; /* prints offset in kB, operation rate */
110 static int readoption; /* run test in read-only (verify) mode */
111 static int writeoption; /* run test in write_only mode */
112 const char *devname; /* name of device to be tested. */
113 static unsigned full = 1; /* flag to full check */
114 static int error_count; /* number of IO errors hit during run */
115 static int isatty_flag;
117 static struct option const longopts[] =
119 { "chunksize", required_argument, 0, 'c' },
120 { "force", no_argument, 0, 'f' },
121 { "help", no_argument, 0, 'h' },
122 { "offset", required_argument, 0, 'o' },
123 { "partial", required_argument, 0, 'p' },
124 { "quiet", required_argument, 0, 'q' },
125 { "read", no_argument, 0, 'r' },
126 { "timestamp", required_argument, 0, 't' },
127 { "verbose", no_argument, 0, 'v' },
128 { "write", no_argument, 0, 'w' },
129 { "long", no_argument, 0, 'l' },
130 { "full", no_argument, 0, 'l' },
135 * Usage: displays help information, whenever user supply --help option in
136 * command or enters incorrect command line.
138 void usage(int status)
141 printf("\nUsage: %s [OPTION]... <device-name> ...\n",
143 printf("Block device verification tool.\n"
144 "\t-t {seconds}, --timestamp, "
145 "set test time (default=current time())\n"
146 "\t-o {offset}, --offset, "
147 "offset in kB of start of test, default=0\n"
148 "\t-r, --read, run in verify mode\n"
149 "\t-w, --write, run in test-pattern mode, default=rw\n"
152 "\t-l, --long, --full check of device\n"
153 "\t-p, --partial, for partial check (1GB steps)\n"
154 "\t-c {bytes}, --chunksize, IO size, default=1048576\n"
155 "\t-f, --force, force test to run without confirmation\n"
156 "\t-h, --help, display this help and exit\n");
162 * Open_dev: Opens device in specified mode and returns fd.
164 static int open_dev(const char *devname, int mode)
167 #ifdef HAVE_EXT2FS_EXT2FS_H
169 char mountpt[80] = "";
171 if (ext2fs_check_mount_point(devname, &mount_flags, mountpt,
173 fprintf(stderr, "%s: ext2fs_check_mount_point failed:%s",
174 progname, strerror(errno));
177 if (mount_flags & EXT2_MF_MOUNTED){
178 fprintf(stderr, "%s: %s is already mounted\n", progname,
183 fd = open(devname, mode | O_EXCL | O_LARGEFILE);
185 fprintf(stderr, "%s: Open failed: %s",progname,strerror(errno));
191 #ifdef HAVE_BLKID_BLKID_H
192 #include <blkid/blkid.h>
195 * sizeof_dev: Returns size of device in bytes
197 static loff_t sizeof_dev(int fd)
201 #ifdef HAVE_BLKID_BLKID_H
202 numbytes = blkid_get_dev_size(fd);
204 fprintf(stderr, "%s: blkid_get_dev_size(%s) failed",
210 # if defined BLKGETSIZE64 /* in sys/mount.h */
211 if (ioctl(fd, BLKGETSIZE64, &numbytes) >= 0)
214 # if defined BLKGETSIZE /* in sys/mount.h */
216 unsigned long sectors;
218 if (ioctl(fd, BLKGETSIZE, §ors) >= 0) {
219 numbytes = (loff_t)sectors << 9;
227 if (fstat(fd, &statbuf) == 0 && S_ISREG(statbuf.st_mode)) {
228 numbytes = statbuf.st_size;
232 fprintf(stderr, "%s: unable to determine size of %s\n",
239 printf("%s: %s is %llu bytes (%g GB) in size\n",
241 (unsigned long long)numbytes, (double)numbytes / ONE_GB);
247 * Verify_chunk: Verifies test pattern in each 4kB (BLOCKSIZE) is correct.
248 * Returns 0 if test offset and timestamp is correct otherwise 1.
250 int verify_chunk(char *chunk_buf, const size_t chunksize,
251 unsigned long long chunk_off, const unsigned long long time_st,
252 const unsigned long long inode_st, const char *file)
254 struct block_data *bd;
257 for (chunk_end = chunk_buf + chunksize - sizeof(*bd);
258 (char *)chunk_buf < chunk_end;
259 chunk_buf += BLOCKSIZE, chunk_off += BLOCKSIZE) {
260 bd = (struct block_data *)chunk_buf;
261 if ((bd->bd_offset == chunk_off) && (bd->bd_time == time_st) &&
262 (bd->bd_inode == inode_st))
265 fprintf(stderr, "\n%s: verify %s failed offset/timestamp/inode "
266 "%llu/%llu/%llu: found %llu/%llu/%llu instead\n",
267 progname, file, chunk_off, time_st, inode_st,
268 bd->bd_offset, bd->bd_time, bd->bd_inode);
276 * fill_chunk: Fills the chunk with current or user specified timestamp
277 * and offset. The test pattern is filled at the beginning of
278 * each 4kB(BLOCKSIZE) blocks in chunk_buf.
280 void fill_chunk(char *chunk_buf, size_t chunksize, loff_t chunk_off,
281 const time_t time_st, const ino_t inode_st)
283 struct block_data *bd;
286 for (chunk_end = chunk_buf + chunksize - sizeof(*bd);
287 (char *)chunk_buf < chunk_end;
288 chunk_buf += BLOCKSIZE, chunk_off += BLOCKSIZE) {
289 bd = (struct block_data *)chunk_buf;
290 bd->bd_offset = chunk_off;
291 bd->bd_time = time_st;
292 bd->bd_inode = inode_st;
296 void show_rate(char *op, unsigned long long offset, unsigned long long *count)
309 printf("%s offset: %14llukB %5g MB/s ", op,
310 offset / ONE_KB, (double)(*count) /ONE_MB /diff);
323 * Write a chunk to disk, handling errors, interrupted writes, etc.
325 * If there is an IO error hit during the write, it is possible that
326 * this will just show up as a short write, and a subsequent write
327 * will return the actual error. We want to continue in the face of
328 * minor media errors so that we can validate the whole device if
329 * possible, but if there are many errors we don't want to loop forever.
331 * The error count will be returned upon exit to ensure that the
332 * media errors are detected even if nobody is looking at the output.
334 * Returns 0 on success, or -ve errno on failure.
336 size_t write_retry(int fd, const char *chunk_buf, size_t nrequested,
337 unsigned long long offset, const char *file)
342 nwritten = write(fd, chunk_buf, nrequested);
344 if (errno != ENOSPC) {
345 fprintf(stderr, "\n%s: write %s@%llu+%zi failed: %s\n",
346 progname, file, offset, nrequested,
348 if (error_count++ < 100)
353 if (nwritten < nrequested) {
354 fprintf(stderr, "\n%s: write %s@%llu+%zi short: %ld written\n",
355 progname, file, offset, nrequested, nwritten);
357 chunk_buf += nwritten;
358 nrequested -= nwritten;
366 * write_chunks: write the chunk_buf on the device. The number of write
367 * operations are based on the parameters write_end, offset, and chunksize.
369 * Returns 0 on success, or -ve error number on failure.
371 int write_chunks(int fd, unsigned long long offset,unsigned long long write_end,
372 char *chunk_buf, size_t chunksize, const time_t time_st,
373 const ino_t inode_st, const char *file)
375 unsigned long long stride, count = 0;
377 stride = full ? chunksize : (ONE_GB - chunksize);
378 for (offset = offset & ~(chunksize - 1); offset < write_end;
382 if (lseek64(fd, offset, SEEK_SET) == -1) {
383 fprintf(stderr, "\n%s: lseek64(%s+%llu) failed: %s\n",
384 progname, file, offset, strerror(errno));
387 if (offset + chunksize > write_end)
388 chunksize = write_end - offset;
389 if (!full && offset > chunksize) {
390 fill_chunk(chunk_buf, chunksize, offset, time_st,
392 ret = write_retry(fd, chunk_buf, chunksize,
398 if (offset + chunksize > write_end)
399 chunksize = write_end - offset;
401 fill_chunk(chunk_buf, chunksize, offset, time_st, inode_st);
402 ret = write_retry(fd, chunk_buf, chunksize, offset, file);
408 show_rate("write", offset, &count);
412 show_rate("write", offset, &count);
413 printf("\nwrite complete\n");
415 if (fsync(fd) == -1) {
416 fprintf(stderr, "%s: fsync failed: %s\n", progname,
424 * read_chunk: reads the chunk_buf from the device. The number of read
425 * operations are based on the parameters read_end, offset, and chunksize.
427 int read_chunks(int fd, unsigned long long offset, unsigned long long read_end,
428 char *chunk_buf, size_t chunksize, const time_t time_st,
429 const ino_t inode_st, const char *file)
431 unsigned long long stride, count = 0;
433 if (ioctl(fd, BLKFLSBUF, 0) < 0 && verbose)
434 fprintf(stderr, "%s: ioctl BLKFLSBUF failed: %s (ignoring)\n",
435 progname, strerror(errno));
437 stride = full ? chunksize : (ONE_GB - chunksize);
438 for (offset = offset & ~(chunksize - 1); offset < read_end;
442 if (lseek64(fd, offset, SEEK_SET) == -1) {
443 fprintf(stderr, "\n%s: lseek64(%llu) failed: %s\n",
444 progname, offset, strerror(errno));
447 if (offset + chunksize > read_end)
448 chunksize = read_end - offset;
450 if (!full && offset > chunksize) {
451 nread = read(fd, chunk_buf, chunksize);
453 fprintf(stderr,"\n%s: read %s@%llu+%zi failed: "
454 "%s\n", progname, file, offset,
455 chunksize, strerror(errno));
459 if (nread < chunksize) {
460 fprintf(stderr, "\n%s: read %s@%llu+%zi short: "
461 "%zi read\n", progname, file, offset,
465 if (verify_chunk(chunk_buf, nread, offset, time_st,
466 inode_st, file) != 0)
471 /* Need to reset position after read error */
472 if (nread < chunksize &&
473 lseek64(fd, offset, SEEK_SET) == -1) {
475 "\n%s: lseek64(%s@%llu) failed: %s\n",
476 progname, file, offset,strerror(errno));
479 if (offset + chunksize >= read_end)
480 chunksize = read_end - offset;
483 nread = read(fd, chunk_buf, chunksize);
485 fprintf(stderr, "\n%s: read failed: %s\n", progname,
490 if (nread < chunksize) {
491 fprintf(stderr, "\n%s: read %s@%llu+%zi short: "
492 "%zi read\n", progname, file, offset,
497 if (verify_chunk(chunk_buf, nread, offset, time_st,
498 inode_st, file) != 0)
503 show_rate("read", offset, &count);
506 show_rate("read", offset, &count);
507 printf("\nread complete\n");
512 int main(int argc, char **argv)
514 time_t time_st = 0; /* Default timestamp */
515 long long offset = 0, offset_orig; /* offset in kB */
516 size_t chunksize = ONE_MB; /* IO chunk size */
517 char *chunk_buf = NULL;
518 unsigned int force = 0; /* run test run without confirmation*/
519 unsigned long long dev_size = 0;
521 int mode = O_RDWR; /* mode which device should be opened */
525 progname = strrchr(argv[0], '/') == NULL ?
526 argv[0] : strrchr(argv[0], '/') + 1;
527 while ((c = getopt_long(argc, argv, "c:fhlo:pqrt:vw", longopts,
531 chunksize = (strtoul(optarg, NULL, 0) * ONE_MB);
533 fprintf(stderr, "%s: chunk size value should be"
534 "nonzero and multiple of 1MB\n",
546 offset = strtoull(optarg, NULL, 0) * ONE_KB;
559 time_st = (time_t)strtoul(optarg, NULL, 0);
574 offset_orig = offset;
575 devname = argv[optind];
577 fprintf(stderr, "%s: device name not given\n", progname);
582 if (readoption && writeoption)
584 if (!readoption && !writeoption) {
589 if (!force && writeoption) {
590 printf("%s: permanently overwrite all data on %s (yes/no)? ",
592 if (scanf("%3s", yesno) == EOF && ferror(stdin)) {
593 perror("reading from stdin");
596 if (!(strcasecmp("yes", yesno) || strcasecmp("y", yesno))) {
597 printf("Not continuing due to '%s' response", yesno);
602 if (!writeoption && time_st == 0) {
603 fprintf(stderr, "%s: must give timestamp for read-only test\n",
608 fd = open_dev(devname, mode);
609 dev_size = sizeof_dev(fd);
611 fprintf(stderr, "%s: cannot test on device size < 1MB\n",
617 if (dev_size < (offset * 2)) {
618 fprintf(stderr, "%s: device size %llu < offset %llu\n",
619 progname, dev_size, offset);
624 (void)time(&time_st);
626 isatty_flag = isatty(STDOUT_FILENO);
629 printf("Timestamp: %lu\n", time_st);
631 chunk_buf = (char *)calloc(chunksize, 1);
632 if (chunk_buf == NULL) {
633 fprintf(stderr, "%s: memory allocation failed for chunk_buf\n",
639 c = write_chunks(fd, offset, dev_size, chunk_buf, chunksize,
640 time_st, 0, devname);
641 if (c < 0 && c != -ENOSPC) {
645 if (!full) { /* end of device aligned to a block */
646 offset = ((dev_size - chunksize + BLOCKSIZE - 1) &
648 c = write_chunks(fd, offset, dev_size, chunk_buf,
649 chunksize, time_st, 0, devname);
650 if (c < 0 && c != -ENOSPC) {
655 offset = offset_orig;
658 if (read_chunks(fd, offset, dev_size, chunk_buf, chunksize,
659 time_st, 0, devname)) {
663 if (!full) { /* end of device aligned to a block */
664 offset = ((dev_size - chunksize + BLOCKSIZE - 1) &
666 if (read_chunks(fd, offset, dev_size, chunk_buf,
667 chunksize, time_st, 0, devname)) {
673 printf("\n%s: data verified successfully\n", progname);