18 #include "crc/crc32c.h"
21 * Check if mmap/mmaphuge has a :/foo/bar/file at the end. If so, return that.
23 static char *get_opt_postfix(const char *str)
25 char *p = strstr(str, ":");
31 strip_blank_front(&p);
36 static int converthexchartoint(char a)
56 static int bs_cmp(const void *p1, const void *p2)
58 const struct bssplit *bsp1 = p1;
59 const struct bssplit *bsp2 = p2;
61 return bsp1->perc < bsp2->perc;
64 static int bssplit_ddir(struct thread_data *td, int ddir, char *str)
66 struct bssplit *bssplit;
67 unsigned int i, perc, perc_missing;
68 unsigned int max_bs, min_bs;
72 td->o.bssplit_nr[ddir] = 4;
73 bssplit = malloc(4 * sizeof(struct bssplit));
78 while ((fname = strsep(&str, ":")) != NULL) {
85 * grow struct buffer, if needed
87 if (i == td->o.bssplit_nr[ddir]) {
88 td->o.bssplit_nr[ddir] <<= 1;
89 bssplit = realloc(bssplit, td->o.bssplit_nr[ddir]
90 * sizeof(struct bssplit));
93 perc_str = strstr(fname, "/");
97 perc = atoi(perc_str);
105 if (str_to_decimal(fname, &val, 1, td)) {
106 log_err("fio: bssplit conversion failed\n");
117 bssplit[i].perc = perc;
121 td->o.bssplit_nr[ddir] = i;
124 * Now check if the percentages add up, and how much is missing
126 perc = perc_missing = 0;
127 for (i = 0; i < td->o.bssplit_nr[ddir]; i++) {
128 struct bssplit *bsp = &bssplit[i];
130 if (bsp->perc == (unsigned char) -1)
137 log_err("fio: bssplit percentages add to more than 100%%\n");
142 * If values didn't have a percentage set, divide the remains between
146 for (i = 0; i < td->o.bssplit_nr[ddir]; i++) {
147 struct bssplit *bsp = &bssplit[i];
149 if (bsp->perc == (unsigned char) -1)
150 bsp->perc = (100 - perc) / perc_missing;
154 td->o.min_bs[ddir] = min_bs;
155 td->o.max_bs[ddir] = max_bs;
158 * now sort based on percentages, for ease of lookup
160 qsort(bssplit, td->o.bssplit_nr[ddir], sizeof(struct bssplit), bs_cmp);
161 td->o.bssplit[ddir] = bssplit;
166 static int str_bssplit_cb(void *data, const char *input)
168 struct thread_data *td = data;
169 char *str, *p, *odir;
172 p = str = strdup(input);
174 strip_blank_front(&str);
175 strip_blank_end(str);
177 odir = strchr(str, ',');
179 ret = bssplit_ddir(td, DDIR_WRITE, odir + 1);
182 ret = bssplit_ddir(td, DDIR_READ, str);
189 ret = bssplit_ddir(td, DDIR_READ, str);
191 ret = bssplit_ddir(td, DDIR_WRITE, op);
200 static int str_rw_cb(void *data, const char *str)
202 struct thread_data *td = data;
203 char *nr = get_opt_postfix(str);
205 td->o.ddir_seq_nr = 1;
207 td->o.ddir_seq_nr = atoi(nr);
214 static int str_mem_cb(void *data, const char *mem)
216 struct thread_data *td = data;
218 if (td->o.mem_type == MEM_MMAPHUGE || td->o.mem_type == MEM_MMAP) {
219 td->mmapfile = get_opt_postfix(mem);
220 if (td->o.mem_type == MEM_MMAPHUGE && !td->mmapfile) {
221 log_err("fio: mmaphuge:/path/to/file\n");
229 static int str_verify_cb(void *data, const char *mem)
231 struct thread_data *td = data;
233 if (td->o.verify != VERIFY_CRC32C_INTEL)
236 if (!crc32c_intel_works()) {
237 log_info("fio: System does not support hw accelerated crc32c. Falling back to sw crc32c.\n");
238 td->o.verify = VERIFY_CRC32C;
244 static int fio_clock_source_cb(void *data, const char *str)
246 struct thread_data *td = data;
248 fio_clock_source = td->o.clocksource;
253 static int str_lockmem_cb(void fio_unused *data, unsigned long long *val)
259 static int str_rwmix_read_cb(void *data, unsigned long long *val)
261 struct thread_data *td = data;
263 td->o.rwmix[DDIR_READ] = *val;
264 td->o.rwmix[DDIR_WRITE] = 100 - *val;
268 static int str_rwmix_write_cb(void *data, unsigned long long *val)
270 struct thread_data *td = data;
272 td->o.rwmix[DDIR_WRITE] = *val;
273 td->o.rwmix[DDIR_READ] = 100 - *val;
277 #ifdef FIO_HAVE_IOPRIO
278 static int str_prioclass_cb(void *data, unsigned long long *val)
280 struct thread_data *td = data;
284 * mask off old class bits, str_prio_cb() may have set a default class
286 mask = (1 << IOPRIO_CLASS_SHIFT) - 1;
289 td->ioprio |= *val << IOPRIO_CLASS_SHIFT;
294 static int str_prio_cb(void *data, unsigned long long *val)
296 struct thread_data *td = data;
301 * If no class is set, assume BE
303 if ((td->ioprio >> IOPRIO_CLASS_SHIFT) == 0)
304 td->ioprio |= IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT;
311 static int str_exitall_cb(void)
313 exitall_on_terminate = 1;
317 #ifdef FIO_HAVE_CPU_AFFINITY
318 static int str_cpumask_cb(void *data, unsigned long long *val)
320 struct thread_data *td = data;
325 ret = fio_cpuset_init(&td->o.cpumask);
327 log_err("fio: cpuset_init failed\n");
328 td_verror(td, ret, "fio_cpuset_init");
332 max_cpu = sysconf(_SC_NPROCESSORS_ONLN);
334 for (i = 0; i < sizeof(int) * 8; i++) {
335 if ((1 << i) & *val) {
337 log_err("fio: CPU %d too large (max=%ld)\n", i,
341 dprint(FD_PARSE, "set cpu allowed %d\n", i);
342 fio_cpu_set(&td->o.cpumask, i);
346 td->o.cpumask_set = 1;
350 static int set_cpus_allowed(struct thread_data *td, os_cpu_mask_t *mask,
357 ret = fio_cpuset_init(mask);
359 log_err("fio: cpuset_init failed\n");
360 td_verror(td, ret, "fio_cpuset_init");
364 p = str = strdup(input);
366 strip_blank_front(&str);
367 strip_blank_end(str);
369 max_cpu = sysconf(_SC_NPROCESSORS_ONLN);
371 while ((cpu = strsep(&str, ",")) != NULL) {
380 while ((cpu2 = strsep(&str2, "-")) != NULL) {
390 while (icpu <= icpu2) {
391 if (icpu >= FIO_MAX_CPUS) {
392 log_err("fio: your OS only supports up to"
393 " %d CPUs\n", (int) FIO_MAX_CPUS);
397 if (icpu > max_cpu) {
398 log_err("fio: CPU %d too large (max=%ld)\n",
404 dprint(FD_PARSE, "set cpu allowed %d\n", icpu);
405 fio_cpu_set(mask, icpu);
414 td->o.cpumask_set = 1;
418 static int str_cpus_allowed_cb(void *data, const char *input)
420 struct thread_data *td = data;
423 ret = set_cpus_allowed(td, &td->o.cpumask, input);
425 td->o.cpumask_set = 1;
430 static int str_verify_cpus_allowed_cb(void *data, const char *input)
432 struct thread_data *td = data;
435 ret = set_cpus_allowed(td, &td->o.verify_cpumask, input);
437 td->o.verify_cpumask_set = 1;
444 static int str_verify_trim_cb(void *data, unsigned long long *val)
446 struct thread_data *td = data;
448 td->o.trim_percentage = *val;
453 static int str_fst_cb(void *data, const char *str)
455 struct thread_data *td = data;
456 char *nr = get_opt_postfix(str);
458 td->file_service_nr = 1;
460 td->file_service_nr = atoi(nr);
467 #ifdef FIO_HAVE_SYNC_FILE_RANGE
468 static int str_sfr_cb(void *data, const char *str)
470 struct thread_data *td = data;
471 char *nr = get_opt_postfix(str);
473 td->sync_file_range_nr = 1;
475 td->sync_file_range_nr = atoi(nr);
483 static int check_dir(struct thread_data *td, char *fname)
485 char file[PATH_MAX], *dir;
488 if (td->o.directory) {
489 strcpy(file, td->o.directory);
494 sprintf(file + elen, "%s", fname);
501 * We can't do this on FIO_DISKLESSIO engines. The engine isn't loaded
502 * yet, so we can't do this check right here...
504 if (lstat(dir, &sb) < 0) {
507 log_err("fio: %s is not a directory\n", dir);
508 td_verror(td, ret, "lstat");
512 if (!S_ISDIR(sb.st_mode)) {
513 log_err("fio: %s is not a directory\n", dir);
523 * Return next file in the string. Files are separated with ':'. If the ':'
524 * is escaped with a '\', then that ':' is part of the filename and does not
525 * indicate a new file.
527 static char *get_next_file_name(char **ptr)
532 if (!str || !strlen(str))
538 * No colon, we are done
540 p = strchr(str, ':');
547 * We got a colon, but it's the first character. Skip and
555 if (*(p - 1) != '\\') {
561 memmove(p - 1, p, strlen(p) + 1);
568 static int str_filename_cb(void *data, const char *input)
570 struct thread_data *td = data;
571 char *fname, *str, *p;
573 p = str = strdup(input);
575 strip_blank_front(&str);
576 strip_blank_end(str);
578 if (!td->files_index)
581 while ((fname = get_next_file_name(&str)) != NULL) {
584 if (check_dir(td, fname)) {
596 static int str_directory_cb(void *data, const char fio_unused *str)
598 struct thread_data *td = data;
601 if (lstat(td->o.directory, &sb) < 0) {
604 log_err("fio: %s is not a directory\n", td->o.directory);
605 td_verror(td, ret, "lstat");
608 if (!S_ISDIR(sb.st_mode)) {
609 log_err("fio: %s is not a directory\n", td->o.directory);
616 static int str_opendir_cb(void *data, const char fio_unused *str)
618 struct thread_data *td = data;
620 if (!td->files_index)
623 return add_dir_files(td, td->o.opendir);
626 static int str_verify_offset_cb(void *data, unsigned long long *off)
628 struct thread_data *td = data;
630 if (*off && *off < sizeof(struct verify_header)) {
631 log_err("fio: verify_offset too small\n");
635 td->o.verify_offset = *off;
639 static int str_verify_pattern_cb(void *data, const char *input)
641 struct thread_data *td = data;
643 int i = 0, j = 0, len, k, base = 10;
646 loc1 = strstr(input, "0x");
647 loc2 = strstr(input, "0X");
650 off = strtol(input, NULL, base);
651 if (off != LONG_MAX || errno != ERANGE) {
653 td->o.verify_pattern[i] = off & 0xff;
662 j = loc1 - input + 2;
664 j = loc2 - input + 2;
667 if (len - j < MAX_PATTERN_SIZE * 2) {
669 off = converthexchartoint(input[k--]);
671 off += (converthexchartoint(input[k--])
673 td->o.verify_pattern[i++] = (char) off;
677 td->o.verify_pattern_bytes = i;
681 static int str_lockfile_cb(void *data, const char *str)
683 struct thread_data *td = data;
684 char *nr = get_opt_postfix(str);
686 td->o.lockfile_batch = 1;
688 td->o.lockfile_batch = atoi(nr);
695 static int str_write_bw_log_cb(void *data, const char *str)
697 struct thread_data *td = data;
700 td->o.bw_log_file = strdup(str);
702 td->o.write_bw_log = 1;
706 static int str_write_lat_log_cb(void *data, const char *str)
708 struct thread_data *td = data;
711 td->o.lat_log_file = strdup(str);
713 td->o.write_lat_log = 1;
717 static int str_gtod_reduce_cb(void *data, int *il)
719 struct thread_data *td = data;
722 td->o.disable_lat = !!val;
723 td->o.disable_clat = !!val;
724 td->o.disable_slat = !!val;
725 td->o.disable_bw = !!val;
727 td->tv_cache_mask = 63;
732 static int str_gtod_cpu_cb(void *data, long long *il)
734 struct thread_data *td = data;
737 td->o.gtod_cpu = val;
738 td->o.gtod_offload = 1;
742 static int rw_verify(struct fio_option *o, void *data)
744 struct thread_data *td = data;
746 if (read_only && td_write(td)) {
747 log_err("fio: job <%s> has write bit set, but fio is in"
748 " read-only mode\n", td->o.name);
755 static int gtod_cpu_verify(struct fio_option *o, void *data)
757 #ifndef FIO_HAVE_CPU_AFFINITY
758 struct thread_data *td = data;
760 if (td->o.gtod_cpu) {
761 log_err("fio: platform must support CPU affinity for"
762 "gettimeofday() offloading\n");
770 static int kb_base_verify(struct fio_option *o, void *data)
772 struct thread_data *td = data;
774 if (td->o.kb_base != 1024 && td->o.kb_base != 1000) {
775 log_err("fio: kb_base set to nonsensical value: %u\n",
783 #define __stringify_1(x) #x
784 #define __stringify(x) __stringify_1(x)
787 * Map of job/command line options
789 static struct fio_option options[FIO_MAX_OPTS] = {
791 .name = "description",
792 .type = FIO_OPT_STR_STORE,
793 .off1 = td_var_offset(description),
794 .help = "Text job description",
798 .type = FIO_OPT_STR_STORE,
799 .off1 = td_var_offset(name),
800 .help = "Name of this job",
804 .type = FIO_OPT_STR_STORE,
805 .off1 = td_var_offset(directory),
806 .cb = str_directory_cb,
807 .help = "Directory to store files in",
811 .type = FIO_OPT_STR_STORE,
812 .off1 = td_var_offset(filename),
813 .cb = str_filename_cb,
814 .prio = -1, /* must come after "directory" */
815 .help = "File(s) to use for the workload",
820 .off1 = td_var_offset(kb_base),
821 .verify = kb_base_verify,
824 .help = "How many bytes per KB for reporting (1000 or 1024)",
829 .cb = str_lockfile_cb,
830 .off1 = td_var_offset(file_lock_mode),
831 .help = "Lock file when doing IO to it",
832 .parent = "filename",
836 .oval = FILE_LOCK_NONE,
837 .help = "No file locking",
839 { .ival = "exclusive",
840 .oval = FILE_LOCK_EXCLUSIVE,
841 .help = "Exclusive file lock",
845 .oval = FILE_LOCK_READWRITE,
846 .help = "Read vs write lock",
852 .type = FIO_OPT_STR_STORE,
853 .off1 = td_var_offset(opendir),
854 .cb = str_opendir_cb,
855 .help = "Recursively add files from this directory and down",
859 .alias = "readwrite",
862 .off1 = td_var_offset(td_ddir),
863 .help = "IO direction",
868 .oval = TD_DDIR_READ,
869 .help = "Sequential read",
872 .oval = TD_DDIR_WRITE,
873 .help = "Sequential write",
875 { .ival = "randread",
876 .oval = TD_DDIR_RANDREAD,
877 .help = "Random read",
879 { .ival = "randwrite",
880 .oval = TD_DDIR_RANDWRITE,
881 .help = "Random write",
885 .help = "Sequential read and write mix",
888 .oval = TD_DDIR_RANDRW,
889 .help = "Random read and write mix"
894 .name = "rw_sequencer",
896 .off1 = td_var_offset(rw_seq),
897 .help = "IO offset generator modifier",
900 { .ival = "sequential",
902 .help = "Generate sequential offsets",
904 { .ival = "identical",
905 .oval = RW_SEQ_IDENT,
906 .help = "Generate identical offsets",
913 .type = FIO_OPT_STR_STORE,
914 .off1 = td_var_offset(ioengine),
915 .help = "IO engine to use",
919 .help = "Use read/write",
922 .help = "Use pread/pwrite",
925 .help = "Use readv/writev",
927 #ifdef FIO_HAVE_LIBAIO
929 .help = "Linux native asynchronous IO",
932 #ifdef FIO_HAVE_POSIXAIO
933 { .ival = "posixaio",
934 .help = "POSIX asynchronous IO",
937 #ifdef FIO_HAVE_SOLARISAIO
938 { .ival = "solarisaio",
939 .help = "Solaris native asynchronous IO",
943 .help = "Memory mapped IO",
945 #ifdef FIO_HAVE_SPLICE
947 .help = "splice/vmsplice based IO",
949 { .ival = "netsplice",
950 .help = "splice/vmsplice to/from the network",
955 .help = "SCSI generic v3 IO",
959 .help = "Testing engine (no data transfer)",
962 .help = "Network IO",
964 #ifdef FIO_HAVE_SYSLET
965 { .ival = "syslet-rw",
966 .help = "syslet enabled async pread/pwrite IO",
970 .help = "CPU cycler burner engine",
972 #ifdef FIO_HAVE_GUASI
974 .help = "GUASI IO engine",
977 { .ival = "external",
978 .help = "Load external engine (append name)",
985 .off1 = td_var_offset(iodepth),
986 .help = "Amount of IO buffers to keep in flight",
991 .name = "iodepth_batch",
992 .alias = "iodepth_batch_submit",
994 .off1 = td_var_offset(iodepth_batch),
995 .help = "Number of IO buffers to submit in one go",
1001 .name = "iodepth_batch_complete",
1002 .type = FIO_OPT_INT,
1003 .off1 = td_var_offset(iodepth_batch_complete),
1004 .help = "Number of IO buffers to retrieve in one go",
1005 .parent = "iodepth",
1010 .name = "iodepth_low",
1011 .type = FIO_OPT_INT,
1012 .off1 = td_var_offset(iodepth_low),
1013 .help = "Low water mark for queuing depth",
1014 .parent = "iodepth",
1018 .type = FIO_OPT_STR_VAL,
1019 .off1 = td_var_offset(size),
1021 .help = "Total size of device or files",
1024 .name = "fill_device",
1025 .type = FIO_OPT_BOOL,
1026 .off1 = td_var_offset(fill_device),
1027 .help = "Write until an ENOSPC error occurs",
1032 .type = FIO_OPT_STR_VAL,
1033 .off1 = td_var_offset(file_size_low),
1034 .off2 = td_var_offset(file_size_high),
1036 .help = "Size of individual files",
1040 .alias = "fileoffset",
1041 .type = FIO_OPT_STR_VAL,
1042 .off1 = td_var_offset(start_offset),
1043 .help = "Start IO from this offset",
1048 .alias = "blocksize",
1049 .type = FIO_OPT_INT,
1050 .off1 = td_var_offset(bs[DDIR_READ]),
1051 .off2 = td_var_offset(bs[DDIR_WRITE]),
1053 .help = "Block size unit",
1059 .alias = "blockalign",
1060 .type = FIO_OPT_INT,
1061 .off1 = td_var_offset(ba[DDIR_READ]),
1062 .off2 = td_var_offset(ba[DDIR_WRITE]),
1064 .help = "IO block offset alignment",
1069 .alias = "blocksize_range",
1070 .type = FIO_OPT_RANGE,
1071 .off1 = td_var_offset(min_bs[DDIR_READ]),
1072 .off2 = td_var_offset(max_bs[DDIR_READ]),
1073 .off3 = td_var_offset(min_bs[DDIR_WRITE]),
1074 .off4 = td_var_offset(max_bs[DDIR_WRITE]),
1076 .help = "Set block size range (in more detail than bs)",
1081 .type = FIO_OPT_STR,
1082 .cb = str_bssplit_cb,
1083 .help = "Set a specific mix of block sizes",
1087 .name = "bs_unaligned",
1088 .alias = "blocksize_unaligned",
1089 .type = FIO_OPT_STR_SET,
1090 .off1 = td_var_offset(bs_unaligned),
1091 .help = "Don't sector align IO buffer sizes",
1095 .name = "randrepeat",
1096 .type = FIO_OPT_BOOL,
1097 .off1 = td_var_offset(rand_repeatable),
1098 .help = "Use repeatable random IO pattern",
1103 .name = "norandommap",
1104 .type = FIO_OPT_STR_SET,
1105 .off1 = td_var_offset(norandommap),
1106 .help = "Accept potential duplicate random blocks",
1110 .name = "softrandommap",
1111 .type = FIO_OPT_BOOL,
1112 .off1 = td_var_offset(softrandommap),
1113 .help = "Set norandommap if randommap allocation fails",
1114 .parent = "norandommap",
1119 .type = FIO_OPT_INT,
1120 .off1 = td_var_offset(nr_files),
1121 .help = "Split job workload between this number of files",
1125 .name = "openfiles",
1126 .type = FIO_OPT_INT,
1127 .off1 = td_var_offset(open_files),
1128 .help = "Number of files to keep open at the same time",
1131 .name = "file_service_type",
1132 .type = FIO_OPT_STR,
1134 .off1 = td_var_offset(file_service_type),
1135 .help = "How to select which file to service next",
1136 .def = "roundrobin",
1139 .oval = FIO_FSERVICE_RANDOM,
1140 .help = "Choose a file at random",
1142 { .ival = "roundrobin",
1143 .oval = FIO_FSERVICE_RR,
1144 .help = "Round robin select files",
1146 { .ival = "sequential",
1147 .oval = FIO_FSERVICE_SEQ,
1148 .help = "Finish one file before moving to the next",
1151 .parent = "nrfiles",
1153 #ifdef FIO_HAVE_FALLOCATE
1155 .name = "fallocate",
1156 .type = FIO_OPT_BOOL,
1157 .off1 = td_var_offset(fallocate),
1158 .help = "Use fallocate() when laying out files",
1163 .name = "fadvise_hint",
1164 .type = FIO_OPT_BOOL,
1165 .off1 = td_var_offset(fadvise_hint),
1166 .help = "Use fadvise() to advise the kernel on IO pattern",
1171 .type = FIO_OPT_INT,
1172 .off1 = td_var_offset(fsync_blocks),
1173 .help = "Issue fsync for writes every given number of blocks",
1177 .name = "fdatasync",
1178 .type = FIO_OPT_INT,
1179 .off1 = td_var_offset(fdatasync_blocks),
1180 .help = "Issue fdatasync for writes every given number of blocks",
1183 #ifdef FIO_HAVE_SYNC_FILE_RANGE
1185 .name = "sync_file_range",
1187 { .ival = "wait_before",
1188 .oval = SYNC_FILE_RANGE_WAIT_BEFORE,
1189 .help = "SYNC_FILE_RANGE_WAIT_BEFORE",
1193 .oval = SYNC_FILE_RANGE_WRITE,
1194 .help = "SYNC_FILE_RANGE_WRITE",
1198 .ival = "wait_after",
1199 .oval = SYNC_FILE_RANGE_WAIT_AFTER,
1200 .help = "SYNC_FILE_RANGE_WAIT_AFTER",
1204 .type = FIO_OPT_STR_MULTI,
1206 .off1 = td_var_offset(sync_file_range),
1207 .help = "Use sync_file_range()",
1212 .type = FIO_OPT_BOOL,
1213 .off1 = td_var_offset(odirect),
1214 .help = "Use O_DIRECT IO (negates buffered)",
1219 .type = FIO_OPT_BOOL,
1220 .off1 = td_var_offset(odirect),
1222 .help = "Use buffered IO (negates direct)",
1226 .name = "overwrite",
1227 .type = FIO_OPT_BOOL,
1228 .off1 = td_var_offset(overwrite),
1229 .help = "When writing, set whether to overwrite current data",
1234 .type = FIO_OPT_INT,
1235 .off1 = td_var_offset(loops),
1236 .help = "Number of times to run the job",
1241 .type = FIO_OPT_INT,
1242 .off1 = td_var_offset(numjobs),
1243 .help = "Duplicate this job this many times",
1247 .name = "startdelay",
1248 .type = FIO_OPT_STR_VAL_TIME,
1249 .off1 = td_var_offset(start_delay),
1250 .help = "Only start job when this period has passed",
1256 .type = FIO_OPT_STR_VAL_TIME,
1257 .off1 = td_var_offset(timeout),
1258 .help = "Stop workload when this amount of time has passed",
1262 .name = "time_based",
1263 .type = FIO_OPT_STR_SET,
1264 .off1 = td_var_offset(time_based),
1265 .help = "Keep running until runtime/timeout is met",
1268 .name = "ramp_time",
1269 .type = FIO_OPT_STR_VAL_TIME,
1270 .off1 = td_var_offset(ramp_time),
1271 .help = "Ramp up time before measuring performance",
1274 .name = "clocksource",
1275 .type = FIO_OPT_STR,
1276 .cb = fio_clock_source_cb,
1277 .off1 = td_var_offset(clocksource),
1278 .help = "What type of timing source to use",
1280 { .ival = "gettimeofday",
1282 .help = "Use gettimeofday(2) for timing",
1284 { .ival = "clock_gettime",
1285 .oval = CS_CGETTIME,
1286 .help = "Use clock_gettime(2) for timing",
1288 #ifdef ARCH_HAVE_CPU_CLOCK
1290 .oval = CS_CPUCLOCK,
1291 .help = "Use CPU private clock",
1299 .type = FIO_OPT_STR,
1301 .off1 = td_var_offset(mem_type),
1302 .help = "Backing type for IO buffers",
1307 .help = "Use malloc(3) for IO buffers",
1311 .help = "Use shared memory segments for IO buffers",
1313 #ifdef FIO_HAVE_HUGETLB
1314 { .ival = "shmhuge",
1315 .oval = MEM_SHMHUGE,
1316 .help = "Like shm, but use huge pages",
1321 .help = "Use mmap(2) (file or anon) for IO buffers",
1323 #ifdef FIO_HAVE_HUGETLB
1324 { .ival = "mmaphuge",
1325 .oval = MEM_MMAPHUGE,
1326 .help = "Like mmap, but use huge pages",
1332 .name = "iomem_align",
1333 .alias = "mem_align",
1334 .type = FIO_OPT_INT,
1335 .off1 = td_var_offset(mem_align),
1337 .help = "IO memory buffer offset alignment",
1343 .type = FIO_OPT_STR,
1344 .off1 = td_var_offset(verify),
1345 .help = "Verify data written",
1346 .cb = str_verify_cb,
1350 .oval = VERIFY_NONE,
1351 .help = "Don't do IO verification",
1355 .help = "Use md5 checksums for verification",
1358 .oval = VERIFY_CRC64,
1359 .help = "Use crc64 checksums for verification",
1362 .oval = VERIFY_CRC32,
1363 .help = "Use crc32 checksums for verification",
1365 { .ival = "crc32c-intel",
1366 .oval = VERIFY_CRC32C_INTEL,
1367 .help = "Use hw crc32c checksums for verification",
1370 .oval = VERIFY_CRC32C,
1371 .help = "Use crc32c checksums for verification",
1374 .oval = VERIFY_CRC16,
1375 .help = "Use crc16 checksums for verification",
1378 .oval = VERIFY_CRC7,
1379 .help = "Use crc7 checksums for verification",
1382 .oval = VERIFY_SHA1,
1383 .help = "Use sha1 checksums for verification",
1386 .oval = VERIFY_SHA256,
1387 .help = "Use sha256 checksums for verification",
1390 .oval = VERIFY_SHA512,
1391 .help = "Use sha512 checksums for verification",
1394 .oval = VERIFY_META,
1395 .help = "Use io information",
1399 .oval = VERIFY_NULL,
1400 .help = "Pretend to verify",
1405 .name = "do_verify",
1406 .type = FIO_OPT_BOOL,
1407 .off1 = td_var_offset(do_verify),
1408 .help = "Run verification stage after write",
1413 .name = "verifysort",
1414 .type = FIO_OPT_BOOL,
1415 .off1 = td_var_offset(verifysort),
1416 .help = "Sort written verify blocks for read back",
1421 .name = "verify_interval",
1422 .type = FIO_OPT_INT,
1423 .off1 = td_var_offset(verify_interval),
1424 .minval = 2 * sizeof(struct verify_header),
1425 .help = "Store verify buffer header every N bytes",
1429 .name = "verify_offset",
1430 .type = FIO_OPT_INT,
1431 .help = "Offset verify header location by N bytes",
1433 .cb = str_verify_offset_cb,
1437 .name = "verify_pattern",
1438 .type = FIO_OPT_STR,
1439 .cb = str_verify_pattern_cb,
1440 .help = "Fill pattern for IO buffers",
1444 .name = "verify_fatal",
1445 .type = FIO_OPT_BOOL,
1446 .off1 = td_var_offset(verify_fatal),
1448 .help = "Exit on a single verify failure, don't continue",
1452 .name = "verify_async",
1453 .type = FIO_OPT_INT,
1454 .off1 = td_var_offset(verify_async),
1456 .help = "Number of async verifier threads to use",
1460 .name = "verify_backlog",
1461 .type = FIO_OPT_STR_VAL,
1462 .off1 = td_var_offset(verify_backlog),
1463 .help = "Verify after this number of blocks are written",
1467 .name = "verify_backlog_batch",
1468 .type = FIO_OPT_INT,
1469 .off1 = td_var_offset(verify_batch),
1470 .help = "Verify this number of IO blocks",
1473 #ifdef FIO_HAVE_CPU_AFFINITY
1475 .name = "verify_async_cpus",
1476 .type = FIO_OPT_STR,
1477 .cb = str_verify_cpus_allowed_cb,
1478 .help = "Set CPUs allowed for async verify threads",
1479 .parent = "verify_async",
1482 #ifdef FIO_HAVE_TRIM
1484 .name = "trim_percentage",
1485 .type = FIO_OPT_INT,
1486 .cb = str_verify_trim_cb,
1488 .help = "Number of verify blocks to discard/trim",
1493 .name = "trim_verify_zero",
1494 .type = FIO_OPT_INT,
1495 .help = "Verify that trim/discarded blocks are returned as zeroes",
1496 .off1 = td_var_offset(trim_zero),
1497 .parent = "trim_percentage",
1501 .name = "trim_backlog",
1502 .type = FIO_OPT_STR_VAL,
1503 .off1 = td_var_offset(trim_backlog),
1504 .help = "Trim after this number of blocks are written",
1505 .parent = "trim_percentage",
1508 .name = "trim_backlog_batch",
1509 .type = FIO_OPT_INT,
1510 .off1 = td_var_offset(trim_batch),
1511 .help = "Trim this number of IO blocks",
1512 .parent = "trim_percentage",
1516 .name = "write_iolog",
1517 .type = FIO_OPT_STR_STORE,
1518 .off1 = td_var_offset(write_iolog_file),
1519 .help = "Store IO pattern to file",
1522 .name = "read_iolog",
1523 .type = FIO_OPT_STR_STORE,
1524 .off1 = td_var_offset(read_iolog_file),
1525 .help = "Playback IO pattern from file",
1528 .name = "replay_no_stall",
1529 .type = FIO_OPT_INT,
1530 .off1 = td_var_offset(no_stall),
1532 .parent = "read_iolog",
1533 .help = "Playback IO pattern file as fast as possible without stalls",
1536 .name = "replay_redirect",
1537 .type = FIO_OPT_STR_STORE,
1538 .off1 = td_var_offset(replay_redirect),
1539 .parent = "read_iolog",
1540 .help = "Replay all I/O onto this device, regardless of trace device",
1543 .name = "exec_prerun",
1544 .type = FIO_OPT_STR_STORE,
1545 .off1 = td_var_offset(exec_prerun),
1546 .help = "Execute this file prior to running job",
1549 .name = "exec_postrun",
1550 .type = FIO_OPT_STR_STORE,
1551 .off1 = td_var_offset(exec_postrun),
1552 .help = "Execute this file after running job",
1554 #ifdef FIO_HAVE_IOSCHED_SWITCH
1556 .name = "ioscheduler",
1557 .type = FIO_OPT_STR_STORE,
1558 .off1 = td_var_offset(ioscheduler),
1559 .help = "Use this IO scheduler on the backing device",
1564 .type = FIO_OPT_STR_VAL,
1565 .off1 = td_var_offset(zone_size),
1566 .help = "Give size of an IO zone",
1571 .type = FIO_OPT_STR_VAL,
1572 .off1 = td_var_offset(zone_skip),
1573 .help = "Space between IO zones",
1578 .type = FIO_OPT_STR_VAL,
1579 .cb = str_lockmem_cb,
1580 .help = "Lock down this amount of memory",
1584 .name = "rwmixread",
1585 .type = FIO_OPT_INT,
1586 .cb = str_rwmix_read_cb,
1588 .help = "Percentage of mixed workload that is reads",
1592 .name = "rwmixwrite",
1593 .type = FIO_OPT_INT,
1594 .cb = str_rwmix_write_cb,
1596 .help = "Percentage of mixed workload that is writes",
1600 .name = "rwmixcycle",
1601 .type = FIO_OPT_DEPRECATED,
1605 .type = FIO_OPT_INT,
1606 .off1 = td_var_offset(nice),
1607 .help = "Set job CPU nice value",
1612 #ifdef FIO_HAVE_IOPRIO
1615 .type = FIO_OPT_INT,
1617 .help = "Set job IO priority value",
1622 .name = "prioclass",
1623 .type = FIO_OPT_INT,
1624 .cb = str_prioclass_cb,
1625 .help = "Set job IO priority class",
1631 .name = "thinktime",
1632 .type = FIO_OPT_INT,
1633 .off1 = td_var_offset(thinktime),
1634 .help = "Idle time between IO buffers (usec)",
1638 .name = "thinktime_spin",
1639 .type = FIO_OPT_INT,
1640 .off1 = td_var_offset(thinktime_spin),
1641 .help = "Start think time by spinning this amount (usec)",
1643 .parent = "thinktime",
1646 .name = "thinktime_blocks",
1647 .type = FIO_OPT_INT,
1648 .off1 = td_var_offset(thinktime_blocks),
1649 .help = "IO buffer period between 'thinktime'",
1651 .parent = "thinktime",
1655 .type = FIO_OPT_INT,
1656 .off1 = td_var_offset(rate[0]),
1657 .off2 = td_var_offset(rate[1]),
1658 .help = "Set bandwidth rate",
1662 .type = FIO_OPT_INT,
1663 .off1 = td_var_offset(ratemin[0]),
1664 .off2 = td_var_offset(ratemin[1]),
1665 .help = "Job must meet this rate or it will be shutdown",
1669 .name = "rate_iops",
1670 .type = FIO_OPT_INT,
1671 .off1 = td_var_offset(rate_iops[0]),
1672 .off2 = td_var_offset(rate_iops[1]),
1673 .help = "Limit IO used to this number of IO operations/sec",
1676 .name = "rate_iops_min",
1677 .type = FIO_OPT_INT,
1678 .off1 = td_var_offset(rate_iops_min[0]),
1679 .off2 = td_var_offset(rate_iops_min[1]),
1680 .help = "Job must meet this rate or it will be shutdown",
1681 .parent = "rate_iops",
1684 .name = "ratecycle",
1685 .type = FIO_OPT_INT,
1686 .off1 = td_var_offset(ratecycle),
1687 .help = "Window average for rate limits (msec)",
1692 .name = "invalidate",
1693 .type = FIO_OPT_BOOL,
1694 .off1 = td_var_offset(invalidate_cache),
1695 .help = "Invalidate buffer/page cache prior to running job",
1700 .type = FIO_OPT_BOOL,
1701 .off1 = td_var_offset(sync_io),
1702 .help = "Use O_SYNC for buffered writes",
1704 .parent = "buffered",
1707 .name = "bwavgtime",
1708 .type = FIO_OPT_INT,
1709 .off1 = td_var_offset(bw_avg_time),
1710 .help = "Time window over which to calculate bandwidth"
1715 .name = "create_serialize",
1716 .type = FIO_OPT_BOOL,
1717 .off1 = td_var_offset(create_serialize),
1718 .help = "Serialize creating of job files",
1722 .name = "create_fsync",
1723 .type = FIO_OPT_BOOL,
1724 .off1 = td_var_offset(create_fsync),
1725 .help = "Fsync file after creation",
1729 .name = "create_on_open",
1730 .type = FIO_OPT_BOOL,
1731 .off1 = td_var_offset(create_on_open),
1732 .help = "Create files when they are opened for IO",
1737 .type = FIO_OPT_BOOL,
1738 .off1 = td_var_offset(pre_read),
1739 .help = "Preread files before starting official testing",
1744 .type = FIO_OPT_INT,
1745 .off1 = td_var_offset(cpuload),
1746 .help = "Use this percentage of CPU",
1749 .name = "cpuchunks",
1750 .type = FIO_OPT_INT,
1751 .off1 = td_var_offset(cpucycle),
1752 .help = "Length of the CPU burn cycles (usecs)",
1754 .parent = "cpuload",
1756 #ifdef FIO_HAVE_CPU_AFFINITY
1759 .type = FIO_OPT_INT,
1760 .cb = str_cpumask_cb,
1761 .help = "CPU affinity mask",
1764 .name = "cpus_allowed",
1765 .type = FIO_OPT_STR,
1766 .cb = str_cpus_allowed_cb,
1767 .help = "Set CPUs allowed",
1771 .name = "end_fsync",
1772 .type = FIO_OPT_BOOL,
1773 .off1 = td_var_offset(end_fsync),
1774 .help = "Include fsync at the end of job",
1778 .name = "fsync_on_close",
1779 .type = FIO_OPT_BOOL,
1780 .off1 = td_var_offset(fsync_on_close),
1781 .help = "fsync files on close",
1786 .type = FIO_OPT_BOOL,
1787 .off1 = td_var_offset(unlink),
1788 .help = "Unlink created files after job has completed",
1793 .type = FIO_OPT_STR_SET,
1794 .cb = str_exitall_cb,
1795 .help = "Terminate all jobs when one exits",
1798 .name = "stonewall",
1799 .type = FIO_OPT_STR_SET,
1800 .off1 = td_var_offset(stonewall),
1801 .help = "Insert a hard barrier between this job and previous",
1804 .name = "new_group",
1805 .type = FIO_OPT_STR_SET,
1806 .off1 = td_var_offset(new_group),
1807 .help = "Mark the start of a new group (for reporting)",
1811 .type = FIO_OPT_STR_SET,
1812 .off1 = td_var_offset(use_thread),
1813 .help = "Use threads instead of forks",
1816 .name = "write_bw_log",
1817 .type = FIO_OPT_STR,
1818 .off1 = td_var_offset(write_bw_log),
1819 .cb = str_write_bw_log_cb,
1820 .help = "Write log of bandwidth during run",
1823 .name = "write_lat_log",
1824 .type = FIO_OPT_STR,
1825 .off1 = td_var_offset(write_lat_log),
1826 .cb = str_write_lat_log_cb,
1827 .help = "Write log of latency during run",
1830 .name = "hugepage-size",
1831 .type = FIO_OPT_INT,
1832 .off1 = td_var_offset(hugepage_size),
1833 .help = "When using hugepages, specify size of each page",
1834 .def = __stringify(FIO_HUGE_PAGE),
1837 .name = "group_reporting",
1838 .type = FIO_OPT_STR_SET,
1839 .off1 = td_var_offset(group_reporting),
1840 .help = "Do reporting on a per-group basis",
1843 .name = "zero_buffers",
1844 .type = FIO_OPT_STR_SET,
1845 .off1 = td_var_offset(zero_buffers),
1846 .help = "Init IO buffers to all zeroes",
1849 .name = "refill_buffers",
1850 .type = FIO_OPT_STR_SET,
1851 .off1 = td_var_offset(refill_buffers),
1852 .help = "Refill IO buffers on every IO submit",
1854 #ifdef FIO_HAVE_DISK_UTIL
1856 .name = "disk_util",
1857 .type = FIO_OPT_BOOL,
1858 .off1 = td_var_offset(do_disk_util),
1859 .help = "Log disk utilization statistics",
1864 .name = "gtod_reduce",
1865 .type = FIO_OPT_BOOL,
1866 .help = "Greatly reduce number of gettimeofday() calls",
1867 .cb = str_gtod_reduce_cb,
1871 .name = "disable_lat",
1872 .type = FIO_OPT_BOOL,
1873 .off1 = td_var_offset(disable_lat),
1874 .help = "Disable latency numbers",
1875 .parent = "gtod_reduce",
1879 .name = "disable_clat",
1880 .type = FIO_OPT_BOOL,
1881 .off1 = td_var_offset(disable_clat),
1882 .help = "Disable completion latency numbers",
1883 .parent = "gtod_reduce",
1887 .name = "disable_slat",
1888 .type = FIO_OPT_BOOL,
1889 .off1 = td_var_offset(disable_slat),
1890 .help = "Disable submissionn latency numbers",
1891 .parent = "gtod_reduce",
1895 .name = "disable_bw_measurement",
1896 .type = FIO_OPT_BOOL,
1897 .off1 = td_var_offset(disable_bw),
1898 .help = "Disable bandwidth logging",
1899 .parent = "gtod_reduce",
1904 .type = FIO_OPT_INT,
1905 .cb = str_gtod_cpu_cb,
1906 .help = "Setup dedicated gettimeofday() thread on this CPU",
1907 .verify = gtod_cpu_verify,
1910 .name = "continue_on_error",
1911 .type = FIO_OPT_BOOL,
1912 .off1 = td_var_offset(continue_on_error),
1913 .help = "Continue on non-fatal errors during I/O",
1918 .type = FIO_OPT_STR_STORE,
1919 .off1 = td_var_offset(profile),
1920 .help = "Select a specific builtin performance test",
1924 .type = FIO_OPT_STR_STORE,
1925 .off1 = td_var_offset(cgroup),
1926 .help = "Add job to cgroup of this name",
1929 .name = "cgroup_weight",
1930 .type = FIO_OPT_INT,
1931 .off1 = td_var_offset(cgroup_weight),
1932 .help = "Use given weight for cgroup",
1937 .name = "cgroup_nodelete",
1938 .type = FIO_OPT_BOOL,
1939 .off1 = td_var_offset(cgroup_nodelete),
1940 .help = "Do not delete cgroups after job completion",
1945 .type = FIO_OPT_INT,
1946 .off1 = td_var_offset(uid),
1947 .help = "Run job with this user ID",
1951 .type = FIO_OPT_INT,
1952 .off1 = td_var_offset(gid),
1953 .help = "Run job with this group ID",
1960 static void add_to_lopt(struct option *lopt, struct fio_option *o,
1963 lopt->name = (char *) name;
1964 lopt->val = FIO_GETOPT_JOB;
1965 if (o->type == FIO_OPT_STR_SET)
1966 lopt->has_arg = no_argument;
1968 lopt->has_arg = required_argument;
1971 void fio_options_dup_and_init(struct option *long_options)
1973 struct fio_option *o;
1976 options_init(options);
1979 while (long_options[i].name)
1984 add_to_lopt(&long_options[i], o, o->name);
1987 add_to_lopt(&long_options[i], o, o->alias);
1992 assert(i < FIO_NR_OPTIONS);
1996 struct fio_keyword {
2002 static struct fio_keyword fio_keywords[] = {
2004 .word = "$pagesize",
2005 .desc = "Page size in the system",
2008 .word = "$mb_memory",
2009 .desc = "Megabytes of memory online",
2013 .desc = "Number of CPUs online in the system",
2020 void fio_keywords_init(void)
2022 unsigned long long mb_memory;
2026 sprintf(buf, "%lu", page_size);
2027 fio_keywords[0].replace = strdup(buf);
2029 mb_memory = os_phys_mem() / page_size;
2030 sprintf(buf, "%llu", mb_memory);
2031 fio_keywords[1].replace = strdup(buf);
2033 l = sysconf(_SC_NPROCESSORS_ONLN);
2034 sprintf(buf, "%lu", l);
2035 fio_keywords[2].replace = strdup(buf);
2040 static char *bc_calc(char *str)
2042 char *buf, *tmp, opt[80];
2047 * No math, just return string
2049 if (!strchr(str, '+') && !strchr(str, '-') && !strchr(str, '*') &&
2054 * Split option from value, we only need to calculate the value
2056 tmp = strchr(str, '=');
2061 memset(opt, 0, sizeof(opt));
2062 strncpy(opt, str, tmp - str);
2066 sprintf(buf, "which %s > /dev/null", BC_APP);
2068 log_err("fio: bc is needed for performing math\n");
2073 sprintf(buf, "echo %s | %s", tmp, BC_APP);
2074 f = popen(buf, "r");
2080 ret = fread(buf, 1, 128, f);
2086 buf[ret - 1] = '\0';
2095 * Look for reserved variable names and replace them with real values
2097 static char *fio_keyword_replace(char *opt)
2102 for (i = 0; fio_keywords[i].word != NULL; i++) {
2103 struct fio_keyword *kw = &fio_keywords[i];
2105 while ((s = strstr(opt, kw->word)) != NULL) {
2106 char *new = malloc(strlen(opt) + 1);
2112 * Copy part of the string before the keyword and
2113 * sprintf() the replacement after it.
2115 memcpy(new, opt, olen);
2116 len = sprintf(new + olen, "%s", kw->replace);
2119 * If there's more in the original string, copy that
2122 opt += strlen(kw->word) + olen;
2124 memcpy(new + olen + len, opt, opt - o_org - 1);
2127 * replace opt and free the old opt
2133 * Check for potential math and invoke bc, if possible
2142 int fio_options_parse(struct thread_data *td, char **opts, int num_opts)
2146 sort_options(opts, options, num_opts);
2148 for (ret = 0, i = 0; i < num_opts; i++) {
2149 opts[i] = fio_keyword_replace(opts[i]);
2150 ret |= parse_option(opts[i], options, td);
2156 int fio_cmd_option_parse(struct thread_data *td, const char *opt, char *val)
2158 return parse_cmd_option(opt, val, options, td);
2161 void fio_fill_default_options(struct thread_data *td)
2163 fill_default_options(td, options);
2166 int fio_show_option_help(const char *opt)
2168 return show_cmd_help(options, opt);
2171 static void __options_mem(struct thread_data *td, int alloc)
2173 struct thread_options *o = &td->o;
2174 struct fio_option *opt;
2178 for (i = 0, opt = &options[0]; opt->name; i++, opt = &options[i]) {
2179 if (opt->type != FIO_OPT_STR_STORE)
2182 ptr = (void *) o + opt->off1;
2185 *ptr = strdup(*ptr);
2195 * dupe FIO_OPT_STR_STORE options
2197 void options_mem_dupe(struct thread_data *td)
2199 __options_mem(td, 1);
2202 void options_mem_free(struct thread_data fio_unused *td)
2205 __options_mem(td, 0);
2209 unsigned int fio_get_kb_base(void *data)
2211 struct thread_data *td = data;
2212 unsigned int kb_base = 0;
2215 kb_base = td->o.kb_base;
2222 int add_option(struct fio_option *o)
2224 struct fio_option *__o;
2233 memcpy(&options[opt_index], o, sizeof(*o));
2237 void invalidate_profile_options(const char *prof_name)
2239 struct fio_option *o;
2243 if (o->prof_name && !strcmp(o->prof_name, prof_name)) {
2244 o->type = FIO_OPT_INVALID;
2245 o->prof_name = NULL;
2251 void add_opt_posval(const char *optname, const char *ival, const char *help)
2253 struct fio_option *o;
2256 o = find_option(options, optname);
2260 for (i = 0; i < PARSE_MAX_VP; i++) {
2261 if (o->posval[i].ival)
2264 o->posval[i].ival = ival;
2265 o->posval[i].help = help;
2270 void del_opt_posval(const char *optname, const char *ival)
2272 struct fio_option *o;
2275 o = find_option(options, optname);
2279 for (i = 0; i < PARSE_MAX_VP; i++) {
2280 if (!o->posval[i].ival)
2282 if (strcmp(o->posval[i].ival, ival))
2285 o->posval[i].ival = NULL;
2286 o->posval[i].help = NULL;