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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, Intel Corporation.
30 * This file is part of Lustre, http://www.lustre.org/
31 * Lustre is a trademark of Sun Microsystems, Inc.
33 * lustre/utils/llverdev.c
35 * Large Block Device Verification Tool.
36 * This program is used to test whether the block device is correctly
37 * handling IO beyond 2TB boundary.
38 * This tool have two working modes
42 * In full mode, the program writes a test pattern on the entire disk.
43 * The test pattern (device offset and timestamp) is written at the
44 * beginning of each 4kB block. When the whole device is full the read
45 * operation is performed to verify that the test pattern is correct.
47 * In partial mode, the program writes data at the critical locations
48 * of the device such as start of the device, before and after multiple of 1GB
49 * offset and at the end.
51 * A chunk buffer with default size of 1MB is used to write and read test
61 #ifndef _LARGEFILE64_SOURCE
62 #define _LARGEFILE64_SOURCE
64 #ifndef _FILE_OFFSET_BITS
65 #define _FILE_OFFSET_BITS 64
80 #include <sys/types.h>
82 #include <sys/ioctl.h>
83 #include <sys/mount.h>
85 #include <gnu/stubs.h>
87 #ifdef HAVE_EXT2FS_EXT2FS_H
88 # include <ext2fs/ext2fs.h>
91 #define ONE_MB (1024 * 1024)
92 #define ONE_GB (1024 * 1024 * 1024)
93 #define HALF_MB (ONE_MB / 2)
95 #define HALF_KB (ONE_KB / 2)
96 #define BLOCKSIZE 4096
98 /* Structure for writting test pattern */
100 unsigned long long bd_offset;
101 unsigned long long bd_time;
102 unsigned long long bd_inode;
104 static char *progname; /* name by which this program was run. */
105 static unsigned verbose = 1; /* prints offset in kB, operation rate */
106 static int readoption; /* run test in read-only (verify) mode */
107 static int writeoption; /* run test in write_only mode */
108 const char *devname; /* name of device to be tested. */
109 static unsigned full = 1; /* flag to full check */
110 static int error_count; /* number of IO errors hit during run */
111 static int isatty_flag;
113 static struct option const longopts[] =
115 { "chunksize", required_argument, 0, 'c' },
116 { "force", no_argument, 0, 'f' },
117 { "help", no_argument, 0, 'h' },
118 { "offset", required_argument, 0, 'o' },
119 { "partial", required_argument, 0, 'p' },
120 { "quiet", required_argument, 0, 'q' },
121 { "read", no_argument, 0, 'r' },
122 { "timestamp", required_argument, 0, 't' },
123 { "verbose", no_argument, 0, 'v' },
124 { "write", no_argument, 0, 'w' },
125 { "long", no_argument, 0, 'l' },
126 { "full", no_argument, 0, 'l' },
131 * Usage: displays help information, whenever user supply --help option in
132 * command or enters incorrect command line.
134 void usage(int status)
137 printf("\nUsage: %s [OPTION]... <device-name> ...\n",
139 printf("Block device verification tool.\n"
140 "\t-t {seconds}, --timestamp, "
141 "set test time (default=current time())\n"
142 "\t-o {offset}, --offset, "
143 "offset in kB of start of test, default=0\n"
144 "\t-r, --read, run in verify mode\n"
145 "\t-w, --write, run in test-pattern mode, default=rw\n"
148 "\t-l, --long, --full check of device\n"
149 "\t-p, --partial, for partial check (1GB steps)\n"
150 "\t-c {bytes}, --chunksize, IO size, default=1048576\n"
151 "\t-f, --force, force test to run without confirmation\n"
152 "\t-h, --help, display this help and exit\n");
158 * Open_dev: Opens device in specified mode and returns fd.
160 static int open_dev(const char *devname, int mode)
163 #ifdef HAVE_EXT2FS_EXT2FS_H
165 char mountpt[80] = "";
167 if (ext2fs_check_mount_point(devname, &mount_flags, mountpt,
169 fprintf(stderr, "%s: ext2fs_check_mount_point failed:%s",
170 progname, strerror(errno));
173 if (mount_flags & EXT2_MF_MOUNTED){
174 fprintf(stderr, "%s: %s is already mounted\n", progname,
179 fd = open(devname, mode | O_EXCL | O_LARGEFILE);
181 fprintf(stderr, "%s: Open failed: %s",progname,strerror(errno));
187 #ifdef HAVE_BLKID_BLKID_H
188 #include <blkid/blkid.h>
191 * sizeof_dev: Returns size of device in bytes
193 static loff_t sizeof_dev(int fd)
197 #ifdef HAVE_BLKID_BLKID_H
198 numbytes = blkid_get_dev_size(fd);
200 fprintf(stderr, "%s: blkid_get_dev_size(%s) failed",
206 # if defined BLKGETSIZE64 /* in sys/mount.h */
207 if (ioctl(fd, BLKGETSIZE64, &numbytes) >= 0)
210 # if defined BLKGETSIZE /* in sys/mount.h */
212 unsigned long sectors;
214 if (ioctl(fd, BLKGETSIZE, §ors) >= 0) {
215 numbytes = (loff_t)sectors << 9;
223 if (fstat(fd, &statbuf) == 0 && S_ISREG(statbuf.st_mode)) {
224 numbytes = statbuf.st_size;
228 fprintf(stderr, "%s: unable to determine size of %s\n",
235 printf("%s: %s is %llu bytes (%g GB) in size\n",
237 (unsigned long long)numbytes, (double)numbytes / ONE_GB);
243 * Verify_chunk: Verifies test pattern in each 4kB (BLOCKSIZE) is correct.
244 * Returns 0 if test offset and timestamp is correct otherwise 1.
246 int verify_chunk(char *chunk_buf, const size_t chunksize,
247 unsigned long long chunk_off, const unsigned long long time_st,
248 const unsigned long long inode_st, const char *file)
250 struct block_data *bd;
253 for (chunk_end = chunk_buf + chunksize - sizeof(*bd);
254 (char *)chunk_buf < chunk_end;
255 chunk_buf += BLOCKSIZE, chunk_off += BLOCKSIZE) {
256 bd = (struct block_data *)chunk_buf;
257 if ((bd->bd_offset == chunk_off) && (bd->bd_time == time_st) &&
258 (bd->bd_inode == inode_st))
261 fprintf(stderr, "\n%s: verify %s failed offset/timestamp/inode "
262 "%llu/%llu/%llu: found %llu/%llu/%llu instead\n",
263 progname, file, chunk_off, time_st, inode_st,
264 bd->bd_offset, bd->bd_time, bd->bd_inode);
272 * fill_chunk: Fills the chunk with current or user specified timestamp
273 * and offset. The test pattern is filled at the beginning of
274 * each 4kB(BLOCKSIZE) blocks in chunk_buf.
276 void fill_chunk(char *chunk_buf, size_t chunksize, loff_t chunk_off,
277 const time_t time_st, const ino_t inode_st)
279 struct block_data *bd;
282 for (chunk_end = chunk_buf + chunksize - sizeof(*bd);
283 (char *)chunk_buf < chunk_end;
284 chunk_buf += BLOCKSIZE, chunk_off += BLOCKSIZE) {
285 bd = (struct block_data *)chunk_buf;
286 bd->bd_offset = chunk_off;
287 bd->bd_time = time_st;
288 bd->bd_inode = inode_st;
292 void show_rate(char *op, unsigned long long offset, unsigned long long *count)
305 printf("%s offset: %14llukB %5g MB/s ", op,
306 offset / ONE_KB, (double)(*count) /ONE_MB /diff);
319 * Write a chunk to disk, handling errors, interrupted writes, etc.
321 * If there is an IO error hit during the write, it is possible that
322 * this will just show up as a short write, and a subsequent write
323 * will return the actual error. We want to continue in the face of
324 * minor media errors so that we can validate the whole device if
325 * possible, but if there are many errors we don't want to loop forever.
327 * The error count will be returned upon exit to ensure that the
328 * media errors are detected even if nobody is looking at the output.
330 * Returns 0 on success, or -ve errno on failure.
332 size_t write_retry(int fd, const char *chunk_buf, size_t nrequested,
333 unsigned long long offset, const char *file)
338 nwritten = write(fd, chunk_buf, nrequested);
340 if (errno != ENOSPC) {
341 fprintf(stderr, "\n%s: write %s@%llu+%zi failed: %s\n",
342 progname, file, offset, nrequested,
344 if (error_count++ < 100)
349 if (nwritten < nrequested) {
350 fprintf(stderr, "\n%s: write %s@%llu+%zi short: %ld written\n",
351 progname, file, offset, nrequested, nwritten);
353 chunk_buf += nwritten;
354 nrequested -= nwritten;
362 * write_chunks: write the chunk_buf on the device. The number of write
363 * operations are based on the parameters write_end, offset, and chunksize.
365 * Returns 0 on success, or -ve error number on failure.
367 int write_chunks(int fd, unsigned long long offset,unsigned long long write_end,
368 char *chunk_buf, size_t chunksize, const time_t time_st,
369 const ino_t inode_st, const char *file)
371 unsigned long long stride, count = 0;
373 stride = full ? chunksize : (ONE_GB - chunksize);
374 for (offset = offset & ~(chunksize - 1); offset < write_end;
378 if (lseek64(fd, offset, SEEK_SET) == -1) {
379 fprintf(stderr, "\n%s: lseek64(%s+%llu) failed: %s\n",
380 progname, file, offset, strerror(errno));
383 if (offset + chunksize > write_end)
384 chunksize = write_end - offset;
385 if (!full && offset > chunksize) {
386 fill_chunk(chunk_buf, chunksize, offset, time_st,
388 ret = write_retry(fd, chunk_buf, chunksize,
394 if (offset + chunksize > write_end)
395 chunksize = write_end - offset;
397 fill_chunk(chunk_buf, chunksize, offset, time_st, inode_st);
398 ret = write_retry(fd, chunk_buf, chunksize, offset, file);
404 show_rate("write", offset, &count);
408 show_rate("write", offset, &count);
409 printf("\nwrite complete\n");
411 if (fsync(fd) == -1) {
412 fprintf(stderr, "%s: fsync failed: %s\n", progname,
420 * read_chunk: reads the chunk_buf from the device. The number of read
421 * operations are based on the parameters read_end, offset, and chunksize.
423 int read_chunks(int fd, unsigned long long offset, unsigned long long read_end,
424 char *chunk_buf, size_t chunksize, const time_t time_st,
425 const ino_t inode_st, const char *file)
427 unsigned long long stride, count = 0;
429 if (ioctl(fd, BLKFLSBUF, 0) < 0 && verbose)
430 fprintf(stderr, "%s: ioctl BLKFLSBUF failed: %s (ignoring)\n",
431 progname, strerror(errno));
433 stride = full ? chunksize : (ONE_GB - chunksize);
434 for (offset = offset & ~(chunksize - 1); offset < read_end;
438 if (lseek64(fd, offset, SEEK_SET) == -1) {
439 fprintf(stderr, "\n%s: lseek64(%llu) failed: %s\n",
440 progname, offset, strerror(errno));
443 if (offset + chunksize > read_end)
444 chunksize = read_end - offset;
446 if (!full && offset > chunksize) {
447 nread = read(fd, chunk_buf, chunksize);
449 fprintf(stderr,"\n%s: read %s@%llu+%zi failed: "
450 "%s\n", progname, file, offset,
451 chunksize, strerror(errno));
455 if (nread < chunksize) {
456 fprintf(stderr, "\n%s: read %s@%llu+%zi short: "
457 "%zi read\n", progname, file, offset,
461 if (verify_chunk(chunk_buf, nread, offset, time_st,
462 inode_st, file) != 0)
467 /* Need to reset position after read error */
468 if (nread < chunksize &&
469 lseek64(fd, offset, SEEK_SET) == -1) {
471 "\n%s: lseek64(%s@%llu) failed: %s\n",
472 progname, file, offset,strerror(errno));
475 if (offset + chunksize >= read_end)
476 chunksize = read_end - offset;
479 nread = read(fd, chunk_buf, chunksize);
481 fprintf(stderr, "\n%s: read failed: %s\n", progname,
486 if (nread < chunksize) {
487 fprintf(stderr, "\n%s: read %s@%llu+%zi short: "
488 "%zi read\n", progname, file, offset,
493 if (verify_chunk(chunk_buf, nread, offset, time_st,
494 inode_st, file) != 0)
499 show_rate("read", offset, &count);
502 show_rate("read", offset, &count);
503 printf("\nread complete\n");
508 int main(int argc, char **argv)
510 time_t time_st = 0; /* Default timestamp */
511 long long offset = 0, offset_orig; /* offset in kB */
512 size_t chunksize = ONE_MB; /* IO chunk size */
513 char *chunk_buf = NULL;
514 unsigned int force = 0; /* run test run without confirmation*/
515 unsigned long long dev_size = 0;
517 int mode = O_RDWR; /* mode which device should be opened */
521 progname = strrchr(argv[0], '/') == NULL ?
522 argv[0] : strrchr(argv[0], '/') + 1;
523 while ((c = getopt_long(argc, argv, "c:fhlo:pqrt:vw", longopts,
527 chunksize = (strtoul(optarg, NULL, 0) * ONE_MB);
529 fprintf(stderr, "%s: chunk size value should be"
530 "nonzero and multiple of 1MB\n",
542 offset = strtoull(optarg, NULL, 0) * ONE_KB;
555 time_st = (time_t)strtoul(optarg, NULL, 0);
570 offset_orig = offset;
571 devname = argv[optind];
573 fprintf(stderr, "%s: device name not given\n", progname);
578 if (readoption && writeoption)
580 if (!readoption && !writeoption) {
585 if (!force && writeoption) {
586 printf("%s: permanently overwrite all data on %s (yes/no)? ",
588 if (scanf("%3s", yesno) == EOF && ferror(stdin)) {
589 perror("reading from stdin");
592 if (!(strcasecmp("yes", yesno) || strcasecmp("y", yesno))) {
593 printf("Not continuing due to '%s' response", yesno);
598 if (!writeoption && time_st == 0) {
599 fprintf(stderr, "%s: must give timestamp for read-only test\n",
604 fd = open_dev(devname, mode);
605 dev_size = sizeof_dev(fd);
607 fprintf(stderr, "%s: cannot test on device size < 1MB\n",
613 if (dev_size < (offset * 2)) {
614 fprintf(stderr, "%s: device size %llu < offset %llu\n",
615 progname, dev_size, offset);
620 (void)time(&time_st);
622 isatty_flag = isatty(STDOUT_FILENO);
625 printf("Timestamp: %lu\n", time_st);
627 chunk_buf = (char *)calloc(chunksize, 1);
628 if (chunk_buf == NULL) {
629 fprintf(stderr, "%s: memory allocation failed for chunk_buf\n",
635 c = write_chunks(fd, offset, dev_size, chunk_buf, chunksize,
636 time_st, 0, devname);
637 if (c < 0 && c != -ENOSPC) {
641 if (!full) { /* end of device aligned to a block */
642 offset = ((dev_size - chunksize + BLOCKSIZE - 1) &
644 c = write_chunks(fd, offset, dev_size, chunk_buf,
645 chunksize, time_st, 0, devname);
646 if (c < 0 && c != -ENOSPC) {
651 offset = offset_orig;
654 if (read_chunks(fd, offset, dev_size, chunk_buf, chunksize,
655 time_st, 0, devname)) {
659 if (!full) { /* end of device aligned to a block */
660 offset = ((dev_size - chunksize + BLOCKSIZE - 1) &
662 if (read_chunks(fd, offset, dev_size, chunk_buf,
663 chunksize, time_st, 0, devname)) {
669 printf("\n%s: data verified successfully\n", progname);