[PATCH] Improve io logging
[fio.git] / init.c
1 /*
2  * This file contains job initialization and setup functions.
3  */
4 #include <stdio.h>
5 #include <stdlib.h>
6 #include <unistd.h>
7 #include <fcntl.h>
8 #include <ctype.h>
9 #include <string.h>
10 #include <errno.h>
11 #include <sys/ipc.h>
12 #include <sys/shm.h>
13 #include <sys/types.h>
14 #include <sys/stat.h>
15
16 #include "fio.h"
17 #include "parse.h"
18
19 /*
20  * The default options
21  */
22 #define DEF_BS                  (4096)
23 #define DEF_TIMEOUT             (0)
24 #define DEF_RATE_CYCLE          (1000)
25 #define DEF_ODIRECT             (1)
26 #define DEF_IO_ENGINE           (FIO_SYNCIO)
27 #define DEF_IO_ENGINE_NAME      "sync"
28 #define DEF_SEQUENTIAL          (1)
29 #define DEF_RAND_REPEAT         (1)
30 #define DEF_OVERWRITE           (1)
31 #define DEF_INVALIDATE          (1)
32 #define DEF_SYNCIO              (0)
33 #define DEF_RANDSEED            (0xb1899bedUL)
34 #define DEF_BWAVGTIME           (500)
35 #define DEF_CREATE_SER          (1)
36 #define DEF_CREATE_FSYNC        (1)
37 #define DEF_LOOPS               (1)
38 #define DEF_VERIFY              (0)
39 #define DEF_STONEWALL           (0)
40 #define DEF_NUMJOBS             (1)
41 #define DEF_USE_THREAD          (0)
42 #define DEF_FILE_SIZE           (1024 * 1024 * 1024UL)
43 #define DEF_ZONE_SIZE           (0)
44 #define DEF_ZONE_SKIP           (0)
45 #define DEF_RWMIX_CYCLE         (500)
46 #define DEF_RWMIX_READ          (50)
47 #define DEF_NICE                (0)
48 #define DEF_NR_FILES            (1)
49 #define DEF_UNLINK              (0)
50 #define DEF_WRITE_BW_LOG        (0)
51 #define DEF_WRITE_LAT_LOG       (0)
52
53 #define td_var_offset(var)      ((size_t) &((struct thread_data *)0)->var)
54
55 static int str_rw_cb(void *, char *);
56 static int str_ioengine_cb(void *, char *);
57 static int str_mem_cb(void *, char *);
58 static int str_verify_cb(void *, char *);
59 static int str_lockmem_cb(void *, unsigned long *);
60 static int str_prio_cb(void *, unsigned int *);
61 static int str_prioclass_cb(void *, unsigned int *);
62 static int str_exitall_cb(void);
63 static int str_cpumask_cb(void *, unsigned int *);
64
65 /*
66  * Map of job/command line options
67  */
68 static struct fio_option options[] = {
69         {
70                 .name   = "name",
71                 .type   = FIO_OPT_STR_STORE,
72                 .off1   = td_var_offset(name),
73         },
74         {
75                 .name   = "directory",
76                 .type   = FIO_OPT_STR_STORE,
77                 .off1   = td_var_offset(directory),
78         },
79         {
80                 .name   = "filename",
81                 .type   = FIO_OPT_STR_STORE,
82                 .off1   = td_var_offset(filename),
83         },
84         {
85                 .name   = "rw",
86                 .type   = FIO_OPT_STR,
87                 .cb     = str_rw_cb,
88         },
89         {
90                 .name   = "ioengine",
91                 .type   = FIO_OPT_STR,
92                 .cb     = str_ioengine_cb,
93         },
94         {
95                 .name   = "mem",
96                 .type   = FIO_OPT_STR,
97                 .cb     = str_mem_cb,
98         },
99         {
100                 .name   = "verify",
101                 .type   = FIO_OPT_STR,
102                 .cb     = str_verify_cb,
103         },
104         {
105                 .name   = "write_iolog",
106                 .type   = FIO_OPT_STR_STORE,
107                 .off1   = td_var_offset(write_iolog_file),
108         },
109         {
110                 .name   = "read_iolog",
111                 .type   = FIO_OPT_STR_STORE,
112                 .off1   = td_var_offset(read_iolog_file),
113         },
114         {
115                 .name   = "exec_prerun",
116                 .type   = FIO_OPT_STR_STORE,
117                 .off1   = td_var_offset(exec_prerun),
118         },
119         {
120                 .name   = "exec_postrun",
121                 .type   = FIO_OPT_STR_STORE,
122                 .off1   = td_var_offset(exec_postrun),
123         },
124 #ifdef FIO_HAVE_IOSCHED_SWITCH
125         {
126                 .name   = "ioscheduler",
127                 .type   = FIO_OPT_STR_STORE,
128                 .off1   = td_var_offset(ioscheduler),
129         },
130 #endif
131         {
132                 .name   = "size",
133                 .type   = FIO_OPT_STR_VAL,
134                 .off1   = td_var_offset(total_file_size),
135         },
136         {
137                 .name   = "bs",
138                 .type   = FIO_OPT_STR_VAL,
139                 .off1   = td_var_offset(bs),
140         },
141         {
142                 .name   = "offset",
143                 .type   = FIO_OPT_STR_VAL,
144                 .off1   = td_var_offset(start_offset),
145         },
146         {
147                 .name   = "zonesize",
148                 .type   = FIO_OPT_STR_VAL,
149                 .off1   = td_var_offset(zone_size),
150         },
151         {
152                 .name   = "zoneskip",
153                 .type   = FIO_OPT_STR_VAL,
154                 .off1   = td_var_offset(zone_skip),
155         },
156         {
157                 .name   = "lockmem",
158                 .type   = FIO_OPT_STR_VAL,
159                 .cb     = str_lockmem_cb,
160         },
161         {
162                 .name   = "bsrange",
163                 .type   = FIO_OPT_RANGE,
164                 .off1   = td_var_offset(min_bs),
165                 .off2   = td_var_offset(max_bs),
166         },
167         {
168                 .name   = "nrfiles",
169                 .type   = FIO_OPT_INT,
170                 .off1   = td_var_offset(nr_files),
171         },
172         {
173                 .name   = "iodepth",
174                 .type   = FIO_OPT_INT,
175                 .off1   = td_var_offset(iodepth),
176         },
177         {
178                 .name   = "fsync",
179                 .type   = FIO_OPT_INT,
180                 .off1   = td_var_offset(fsync_blocks),
181         },
182         {
183                 .name   = "rwmixcycle",
184                 .type   = FIO_OPT_INT,
185                 .off1   = td_var_offset(rwmixcycle),
186         },
187         {
188                 .name   = "rwmixread",
189                 .type   = FIO_OPT_INT,
190                 .off1   = td_var_offset(rwmixread),
191                 .max_val= 100,
192         },
193         {
194                 .name   = "rwmixwrite",
195                 .type   = FIO_OPT_INT,
196                 .off1   = td_var_offset(rwmixwrite),
197                 .max_val= 100,
198         },
199         {
200                 .name   = "nice",
201                 .type   = FIO_OPT_INT,
202                 .off1   = td_var_offset(nice),
203         },
204 #ifdef FIO_HAVE_IOPRIO
205         {
206                 .name   = "prio",
207                 .type   = FIO_OPT_INT,
208                 .cb     = str_prio_cb,
209         },
210         {
211                 .name   = "prioclass",
212                 .type   = FIO_OPT_INT,
213                 .cb     = str_prioclass_cb,
214         },
215 #endif
216         {
217                 .name   = "thinktime",
218                 .type   = FIO_OPT_INT,
219                 .off1   = td_var_offset(thinktime)
220         },
221         {
222                 .name   = "rate",
223                 .type   = FIO_OPT_INT,
224                 .off1   = td_var_offset(rate)
225         },
226         {
227                 .name   = "ratemin",
228                 .type   = FIO_OPT_INT,
229                 .off1   = td_var_offset(ratemin)
230         },
231         {
232                 .name   = "ratecycle",
233                 .type   = FIO_OPT_INT,
234                 .off1   = td_var_offset(ratecycle)
235         },
236         {
237                 .name   = "startdelay",
238                 .type   = FIO_OPT_INT,
239                 .off1   = td_var_offset(start_delay)
240         },
241         {
242                 .name   = "timeout",
243                 .type   = FIO_OPT_STR_VAL_TIME,
244                 .off1   = td_var_offset(timeout)
245         },
246         {
247                 .name   = "invalidate",
248                 .type   = FIO_OPT_INT,
249                 .off1   = td_var_offset(invalidate_cache)
250         },
251         {
252                 .name   = "sync",
253                 .type   = FIO_OPT_INT,
254                 .off1   = td_var_offset(sync_io)
255         },
256         {
257                 .name   = "bwavgtime",
258                 .type   = FIO_OPT_INT,
259                 .off1   = td_var_offset(bw_avg_time)
260         },
261         {
262                 .name   = "create_serialize",
263                 .type   = FIO_OPT_INT,
264                 .off1   = td_var_offset(create_serialize)
265         },
266         {
267                 .name   = "create_fsync",
268                 .type   = FIO_OPT_INT,
269                 .off1   = td_var_offset(create_fsync)
270         },
271         {
272                 .name   = "loops",
273                 .type   = FIO_OPT_INT,
274                 .off1   = td_var_offset(loops)
275         },
276         {
277                 .name   = "numjobs",
278                 .type   = FIO_OPT_INT,
279                 .off1   = td_var_offset(numjobs)
280         },
281         {
282                 .name   = "cpuload",
283                 .type   = FIO_OPT_INT,
284                 .off1   = td_var_offset(cpuload)
285         },
286         {
287                 .name   = "cpuchunks",
288                 .type   = FIO_OPT_INT,
289                 .off1   = td_var_offset(cpucycle)
290         },
291         {
292                 .name   = "direct",
293                 .type   = FIO_OPT_INT,
294                 .off1   = td_var_offset(odirect)
295         },
296         {
297                 .name   = "overwrite",
298                 .type   = FIO_OPT_INT,
299                 .off1   = td_var_offset(overwrite)
300         },
301 #ifdef FIO_HAVE_CPU_AFFINITY
302         {
303                 .name   = "cpumask",
304                 .type   = FIO_OPT_INT,
305                 .cb     = str_cpumask_cb,
306         },
307 #endif
308         {
309                 .name   = "end_fsync",
310                 .type   = FIO_OPT_INT,
311                 .off1   = td_var_offset(end_fsync)
312         },
313         {
314                 .name   = "unlink",
315                 .type   = FIO_OPT_STR_SET,
316                 .off1   = td_var_offset(unlink),
317         },
318         {
319                 .name   = "exitall",
320                 .type   = FIO_OPT_STR_SET,
321                 .cb     = str_exitall_cb,
322         },
323         {
324                 .name   = "stonewall",
325                 .type   = FIO_OPT_STR_SET,
326                 .off1   = td_var_offset(stonewall),
327         },
328         {
329                 .name   = "thread",
330                 .type   = FIO_OPT_STR_SET,
331                 .off1   = td_var_offset(thread),
332         },
333         {
334                 .name   = "write_bw_log",
335                 .type   = FIO_OPT_STR_SET,
336                 .off1   = td_var_offset(write_bw_log),
337         },
338         {
339                 .name   = "write_lat_log",
340                 .type   = FIO_OPT_STR_SET,
341                 .off1   = td_var_offset(write_lat_log),
342         },
343         {
344                 .name = NULL,
345         },
346 };
347
348 static int def_timeout = DEF_TIMEOUT;
349
350 static char fio_version_string[] = "fio 1.5";
351
352 static char **ini_file;
353 static int max_jobs = MAX_JOBS;
354
355 struct thread_data def_thread;
356 struct thread_data *threads = NULL;
357
358 int rate_quit = 0;
359 int exitall_on_terminate = 0;
360 int terse_output = 0;
361 unsigned long long mlock_size = 0;
362 FILE *f_out = NULL;
363 FILE *f_err = NULL;
364
365 static int write_lat_log = DEF_WRITE_LAT_LOG;
366 static int write_bw_log = DEF_WRITE_BW_LOG;
367
368 /*
369  * Return a free job structure.
370  */
371 static struct thread_data *get_new_job(int global, struct thread_data *parent)
372 {
373         struct thread_data *td;
374
375         if (global)
376                 return &def_thread;
377         if (thread_number >= max_jobs)
378                 return NULL;
379
380         td = &threads[thread_number++];
381         *td = *parent;
382         td->name[0] = '\0';
383
384         td->thread_number = thread_number;
385         return td;
386 }
387
388 static void put_job(struct thread_data *td)
389 {
390         memset(&threads[td->thread_number - 1], 0, sizeof(*td));
391         thread_number--;
392 }
393
394 /*
395  * Lazy way of fixing up options that depend on each other. We could also
396  * define option callback handlers, but this is easier.
397  */
398 static void fixup_options(struct thread_data *td)
399 {
400         if (!td->min_bs)
401                 td->min_bs = td->bs;
402         if (!td->max_bs)
403                 td->max_bs = td->bs;
404
405         if (!td->rwmixread && td->rwmixwrite)
406                 td->rwmixread = 100 - td->rwmixwrite;
407
408         if (td->write_iolog_file && td->read_iolog_file) {
409                 log_err("fio: read iolog overrides write_iolog\n");
410                 free(td->write_iolog_file);
411                 td->write_iolog_file = NULL;
412         }
413 }
414
415 /*
416  * Adds a job to the list of things todo. Sanitizes the various options
417  * to make sure we don't have conflicts, and initializes various
418  * members of td.
419  */
420 static int add_job(struct thread_data *td, const char *jobname, int job_add_num)
421 {
422         char *ddir_str[] = { "read", "write", "randread", "randwrite",
423                              "rw", NULL, "randrw" };
424         struct stat sb;
425         int numjobs, ddir, i;
426         struct fio_file *f;
427
428 #ifndef FIO_HAVE_LIBAIO
429         if (td->io_engine == FIO_LIBAIO) {
430                 log_err("Linux libaio not available\n");
431                 return 1;
432         }
433 #endif
434 #ifndef FIO_HAVE_POSIXAIO
435         if (td->io_engine == FIO_POSIXAIO) {
436                 log_err("posix aio not available\n");
437                 return 1;
438         }
439 #endif
440
441         fixup_options(td);
442
443         /*
444          * the def_thread is just for options, it's not a real job
445          */
446         if (td == &def_thread)
447                 return 0;
448
449         /*
450          * Set default io engine, if none set
451          */
452         if (!td->io_ops) {
453                 td->io_ops = load_ioengine(td, DEF_IO_ENGINE_NAME);
454                 if (!td->io_ops) {
455                         log_err("default engine %s not there?\n", DEF_IO_ENGINE_NAME);
456                         return 1;
457                 }
458         }
459
460         if (td->io_ops->flags & FIO_SYNCIO)
461                 td->iodepth = 1;
462         else {
463                 if (!td->iodepth)
464                         td->iodepth = td->nr_files;
465         }
466
467         /*
468          * only really works for sequential io for now, and with 1 file
469          */
470         if (td->zone_size && !td->sequential && td->nr_files == 1)
471                 td->zone_size = 0;
472
473         /*
474          * Reads can do overwrites, we always need to pre-create the file
475          */
476         if (td_read(td) || td_rw(td))
477                 td->overwrite = 1;
478
479         td->filetype = FIO_TYPE_FILE;
480         if (!stat(jobname, &sb)) {
481                 if (S_ISBLK(sb.st_mode))
482                         td->filetype = FIO_TYPE_BD;
483                 else if (S_ISCHR(sb.st_mode))
484                         td->filetype = FIO_TYPE_CHAR;
485         }
486
487         if (td->odirect)
488                 td->io_ops->flags |= FIO_RAWIO;
489
490         if (td->filename)
491                 td->nr_uniq_files = 1;
492         else
493                 td->nr_uniq_files = td->nr_files;
494
495         if (td->filetype == FIO_TYPE_FILE || td->filename) {
496                 char tmp[PATH_MAX];
497                 int len = 0;
498                 int i;
499
500                 if (td->directory && td->directory[0] != '\0')
501                         sprintf(tmp, "%s/", td->directory);
502
503                 td->files = malloc(sizeof(struct fio_file) * td->nr_files);
504
505                 for_each_file(td, f, i) {
506                         memset(f, 0, sizeof(*f));
507                         f->fd = -1;
508
509                         if (td->filename)
510                                 sprintf(tmp + len, "%s", td->filename);
511                         else
512                                 sprintf(tmp + len, "%s.%d.%d", jobname, td->thread_number, i);
513                         f->file_name = strdup(tmp);
514                 }
515         } else {
516                 td->nr_files = 1;
517                 td->files = malloc(sizeof(struct fio_file));
518                 f = &td->files[0];
519
520                 memset(f, 0, sizeof(*f));
521                 f->fd = -1;
522                 f->file_name = strdup(jobname);
523         }
524
525         for_each_file(td, f, i) {
526                 f->file_size = td->total_file_size / td->nr_files;
527                 f->file_offset = td->start_offset;
528         }
529                 
530         fio_sem_init(&td->mutex, 0);
531
532         td->clat_stat[0].min_val = td->clat_stat[1].min_val = ULONG_MAX;
533         td->slat_stat[0].min_val = td->slat_stat[1].min_val = ULONG_MAX;
534         td->bw_stat[0].min_val = td->bw_stat[1].min_val = ULONG_MAX;
535
536         if (td->min_bs == -1U)
537                 td->min_bs = td->bs;
538         if (td->max_bs == -1U)
539                 td->max_bs = td->bs;
540         if (td_read(td) && !td_rw(td))
541                 td->verify = 0;
542
543         if (td->stonewall && td->thread_number > 1)
544                 groupid++;
545
546         td->groupid = groupid;
547
548         if (setup_rate(td))
549                 goto err;
550
551         if (td->write_lat_log) {
552                 setup_log(&td->slat_log);
553                 setup_log(&td->clat_log);
554         }
555         if (td->write_bw_log)
556                 setup_log(&td->bw_log);
557
558         if (td->name[0] == '\0')
559                 snprintf(td->name, sizeof(td->name)-1, "client%d", td->thread_number);
560
561         ddir = td->ddir + (!td->sequential << 1) + (td->iomix << 2);
562
563         if (!terse_output) {
564                 if (!job_add_num) {
565                         if (td->io_ops->flags & FIO_CPUIO)
566                                 fprintf(f_out, "%s: ioengine=cpu, cpuload=%u, cpucycle=%u\n", td->name, td->cpuload, td->cpucycle);
567                         else
568                                 fprintf(f_out, "%s: (g=%d): rw=%s, odir=%d, bs=%d-%d, rate=%d, ioengine=%s, iodepth=%d\n", td->name, td->groupid, ddir_str[ddir], td->odirect, td->min_bs, td->max_bs, td->rate, td->io_ops->name, td->iodepth);
569                 } else if (job_add_num == 1)
570                         fprintf(f_out, "...\n");
571         }
572
573         /*
574          * recurse add identical jobs, clear numjobs and stonewall options
575          * as they don't apply to sub-jobs
576          */
577         numjobs = td->numjobs;
578         while (--numjobs) {
579                 struct thread_data *td_new = get_new_job(0, td);
580
581                 if (!td_new)
582                         goto err;
583
584                 td_new->numjobs = 1;
585                 td_new->stonewall = 0;
586                 job_add_num = numjobs - 1;
587
588                 if (add_job(td_new, jobname, job_add_num))
589                         goto err;
590         }
591         return 0;
592 err:
593         put_job(td);
594         return -1;
595 }
596
597 /*
598  * Initialize the various random states we need (random io, block size ranges,
599  * read/write mix, etc).
600  */
601 int init_random_state(struct thread_data *td)
602 {
603         unsigned long seeds[4];
604         int fd, num_maps, blocks, i;
605         struct fio_file *f;
606
607         fd = open("/dev/urandom", O_RDONLY);
608         if (fd == -1) {
609                 td_verror(td, errno);
610                 return 1;
611         }
612
613         if (read(fd, seeds, sizeof(seeds)) < (int) sizeof(seeds)) {
614                 td_verror(td, EIO);
615                 close(fd);
616                 return 1;
617         }
618
619         close(fd);
620
621         os_random_seed(seeds[0], &td->bsrange_state);
622         os_random_seed(seeds[1], &td->verify_state);
623         os_random_seed(seeds[2], &td->rwmix_state);
624
625         if (td->sequential)
626                 return 0;
627
628         if (td->rand_repeatable)
629                 seeds[3] = DEF_RANDSEED;
630
631         for_each_file(td, f, i) {
632                 blocks = (f->file_size + td->min_bs - 1) / td->min_bs;
633                 num_maps = blocks / BLOCKS_PER_MAP;
634                 f->file_map = malloc(num_maps * sizeof(long));
635                 f->num_maps = num_maps;
636                 memset(f->file_map, 0, num_maps * sizeof(long));
637         }
638
639         os_random_seed(seeds[3], &td->random_state);
640         return 0;
641 }
642
643 static void fill_cpu_mask(os_cpu_mask_t cpumask, int cpu)
644 {
645 #ifdef FIO_HAVE_CPU_AFFINITY
646         unsigned int i;
647
648         CPU_ZERO(&cpumask);
649
650         for (i = 0; i < sizeof(int) * 8; i++) {
651                 if ((1 << i) & cpu)
652                         CPU_SET(i, &cpumask);
653         }
654 #endif
655 }
656
657 static int is_empty_or_comment(char *line)
658 {
659         unsigned int i;
660
661         for (i = 0; i < strlen(line); i++) {
662                 if (line[i] == ';')
663                         return 1;
664                 if (!isspace(line[i]) && !iscntrl(line[i]))
665                         return 0;
666         }
667
668         return 1;
669 }
670
671 static int str_rw_cb(void *data, char *mem)
672 {
673         struct thread_data *td = data;
674
675         if (!strncmp(mem, "read", 4) || !strncmp(mem, "0", 1)) {
676                 td->ddir = DDIR_READ;
677                 td->sequential = 1;
678                 return 0;
679         } else if (!strncmp(mem, "randread", 8)) {
680                 td->ddir = DDIR_READ;
681                 td->sequential = 0;
682                 return 0;
683         } else if (!strncmp(mem, "write", 5) || !strncmp(mem, "1", 1)) {
684                 td->ddir = DDIR_WRITE;
685                 td->sequential = 1;
686                 return 0;
687         } else if (!strncmp(mem, "randwrite", 9)) {
688                 td->ddir = DDIR_WRITE;
689                 td->sequential = 0;
690                 return 0;
691         } else if (!strncmp(mem, "rw", 2)) {
692                 td->ddir = 0;
693                 td->iomix = 1;
694                 td->sequential = 1;
695                 return 0;
696         } else if (!strncmp(mem, "randrw", 6)) {
697                 td->ddir = 0;
698                 td->iomix = 1;
699                 td->sequential = 0;
700                 return 0;
701         }
702
703         log_err("fio: data direction: read, write, randread, randwrite, rw, randrw\n");
704         return 1;
705 }
706
707 static int str_verify_cb(void *data, char *mem)
708 {
709         struct thread_data *td = data;
710
711         if (!strncmp(mem, "0", 1)) {
712                 td->verify = VERIFY_NONE;
713                 return 0;
714         } else if (!strncmp(mem, "md5", 3) || !strncmp(mem, "1", 1)) {
715                 td->verify = VERIFY_MD5;
716                 return 0;
717         } else if (!strncmp(mem, "crc32", 5)) {
718                 td->verify = VERIFY_CRC32;
719                 return 0;
720         }
721
722         log_err("fio: verify types: md5, crc32\n");
723         return 1;
724 }
725
726 static int str_mem_cb(void *data, char *mem)
727 {
728         struct thread_data *td = data;
729
730         if (!strncmp(mem, "malloc", 6)) {
731                 td->mem_type = MEM_MALLOC;
732                 return 0;
733         } else if (!strncmp(mem, "shm", 3)) {
734                 td->mem_type = MEM_SHM;
735                 return 0;
736         } else if (!strncmp(mem, "mmap", 4)) {
737                 td->mem_type = MEM_MMAP;
738                 return 0;
739         }
740
741         log_err("fio: mem type: malloc, shm, mmap\n");
742         return 1;
743 }
744
745 static int str_ioengine_cb(void *data, char *str)
746 {
747         struct thread_data *td = data;
748
749         td->io_ops = load_ioengine(td, str);
750         if (td->io_ops)
751                 return 0;
752
753         log_err("fio: ioengine: { linuxaio, aio, libaio }, posixaio, sync, mmap, sgio, splice, cpu\n");
754         return 1;
755 }
756
757 static int str_lockmem_cb(void fio_unused *data, unsigned long *val)
758 {
759         mlock_size = *val;
760         return 0;
761 }
762
763 static int str_prioclass_cb(void *data, unsigned int *val)
764 {
765         struct thread_data *td = data;
766
767         td->ioprio |= *val << IOPRIO_CLASS_SHIFT;
768         return 0;
769 }
770
771 static int str_prio_cb(void *data, unsigned int *val)
772 {
773         struct thread_data *td = data;
774
775         td->ioprio |= *val;
776         return 0;
777 }
778
779 static int str_exitall_cb(void)
780 {
781         exitall_on_terminate = 1;
782         return 0;
783 }
784
785 static int str_cpumask_cb(void *data, unsigned int *val)
786 {
787         struct thread_data *td = data;
788
789         fill_cpu_mask(td->cpumask, *val);
790         return 0;
791 }
792
793 /*
794  * This is our [ini] type file parser.
795  */
796 int parse_jobs_ini(char *file, int stonewall_flag)
797 {
798         unsigned int global;
799         struct thread_data *td;
800         char *string, *name, *tmpbuf;
801         fpos_t off;
802         FILE *f;
803         char *p;
804         int ret = 0, stonewall;
805
806         f = fopen(file, "r");
807         if (!f) {
808                 perror("fopen job file");
809                 return 1;
810         }
811
812         string = malloc(4096);
813         name = malloc(256);
814         tmpbuf = malloc(4096);
815
816         stonewall = stonewall_flag;
817         while ((p = fgets(string, 4096, f)) != NULL) {
818                 if (ret)
819                         break;
820                 if (is_empty_or_comment(p))
821                         continue;
822                 if (sscanf(p, "[%s]", name) != 1)
823                         continue;
824
825                 global = !strncmp(name, "global", 6);
826
827                 name[strlen(name) - 1] = '\0';
828
829                 td = get_new_job(global, &def_thread);
830                 if (!td) {
831                         ret = 1;
832                         break;
833                 }
834
835                 /*
836                  * Seperate multiple job files by a stonewall
837                  */
838                 if (!global && stonewall) {
839                         td->stonewall = stonewall;
840                         stonewall = 0;
841                 }
842
843                 fgetpos(f, &off);
844                 while ((p = fgets(string, 4096, f)) != NULL) {
845                         if (is_empty_or_comment(p))
846                                 continue;
847                         if (strstr(p, "["))
848                                 break;
849
850                         strip_blank_front(&p);
851                         strip_blank_end(p);
852
853                         fgetpos(f, &off);
854
855                         /*
856                          * Don't break here, continue parsing options so we
857                          * dump all the bad ones. Makes trial/error fixups
858                          * easier on the user.
859                          */
860                         ret = parse_option(p, options, td);
861                 }
862
863                 if (!ret) {
864                         fsetpos(f, &off);
865                         ret = add_job(td, name, 0);
866                 }
867                 if (ret)
868                         break;
869         }
870
871         free(string);
872         free(name);
873         free(tmpbuf);
874         fclose(f);
875         return ret;
876 }
877
878 static int fill_def_thread(void)
879 {
880         memset(&def_thread, 0, sizeof(def_thread));
881
882         if (fio_getaffinity(getpid(), &def_thread.cpumask) == -1) {
883                 perror("sched_getaffinity");
884                 return 1;
885         }
886
887         /*
888          * fill globals
889          */
890         def_thread.ddir = DDIR_READ;
891         def_thread.iomix = 0;
892         def_thread.bs = DEF_BS;
893         def_thread.min_bs = -1;
894         def_thread.max_bs = -1;
895         def_thread.odirect = DEF_ODIRECT;
896         def_thread.ratecycle = DEF_RATE_CYCLE;
897         def_thread.sequential = DEF_SEQUENTIAL;
898         def_thread.timeout = def_timeout;
899         def_thread.overwrite = DEF_OVERWRITE;
900         def_thread.invalidate_cache = DEF_INVALIDATE;
901         def_thread.sync_io = DEF_SYNCIO;
902         def_thread.mem_type = MEM_MALLOC;
903         def_thread.bw_avg_time = DEF_BWAVGTIME;
904         def_thread.create_serialize = DEF_CREATE_SER;
905         def_thread.create_fsync = DEF_CREATE_FSYNC;
906         def_thread.loops = DEF_LOOPS;
907         def_thread.verify = DEF_VERIFY;
908         def_thread.stonewall = DEF_STONEWALL;
909         def_thread.numjobs = DEF_NUMJOBS;
910         def_thread.use_thread = DEF_USE_THREAD;
911         def_thread.rwmixcycle = DEF_RWMIX_CYCLE;
912         def_thread.rwmixread = DEF_RWMIX_READ;
913         def_thread.nice = DEF_NICE;
914         def_thread.rand_repeatable = DEF_RAND_REPEAT;
915         def_thread.nr_files = DEF_NR_FILES;
916         def_thread.unlink = DEF_UNLINK;
917         def_thread.write_bw_log = write_bw_log;
918         def_thread.write_lat_log = write_lat_log;
919 #ifdef FIO_HAVE_DISK_UTIL
920         def_thread.do_disk_util = 1;
921 #endif
922
923         return 0;
924 }
925
926 static void usage(void)
927 {
928         printf("%s\n", fio_version_string);
929         printf("\t-o Write output to file\n");
930         printf("\t-t Runtime in seconds\n");
931         printf("\t-l Generate per-job latency logs\n");
932         printf("\t-w Generate per-job bandwidth logs\n");
933         printf("\t-m Minimal (terse) output\n");
934         printf("\t-v Print version info and exit\n");
935 }
936
937 static int parse_cmd_line(int argc, char *argv[])
938 {
939         int c, idx = 1, ini_idx = 0;
940
941         while ((c = getopt(argc, argv, "t:o:lwvhm")) != EOF) {
942                 switch (c) {
943                         case 't':
944                                 def_timeout = atoi(optarg);
945                                 idx = optind;
946                                 break;
947                         case 'l':
948                                 write_lat_log = 1;
949                                 idx = optind;
950                                 break;
951                         case 'w':
952                                 write_bw_log = 1;
953                                 idx = optind;
954                                 break;
955                         case 'o':
956                                 f_out = fopen(optarg, "w+");
957                                 if (!f_out) {
958                                         perror("fopen output");
959                                         exit(1);
960                                 }
961                                 f_err = f_out;
962                                 idx = optind;
963                                 break;
964                         case 'm':
965                                 terse_output = 1;
966                                 idx = optind;
967                                 break;
968                         case 'h':
969                                 usage();
970                                 exit(0);
971                         case 'v':
972                                 printf("%s\n", fio_version_string);
973                                 exit(0);
974                 }
975         }
976
977         while (idx < argc) {
978                 ini_idx++;
979                 ini_file = realloc(ini_file, ini_idx * sizeof(char *));
980                 ini_file[ini_idx - 1] = strdup(argv[idx]);
981                 idx++;
982         }
983
984         if (!f_out) {
985                 f_out = stdout;
986                 f_err = stderr;
987         }
988
989         return ini_idx;
990 }
991
992 static void free_shm(void)
993 {
994         struct shmid_ds sbuf;
995
996         if (threads) {
997                 shmdt((void *) threads);
998                 threads = NULL;
999                 shmctl(shm_id, IPC_RMID, &sbuf);
1000         }
1001 }
1002
1003 /*
1004  * The thread area is shared between the main process and the job
1005  * threads/processes. So setup a shared memory segment that will hold
1006  * all the job info.
1007  */
1008 static int setup_thread_area(void)
1009 {
1010         /*
1011          * 1024 is too much on some machines, scale max_jobs if
1012          * we get a failure that looks like too large a shm segment
1013          */
1014         do {
1015                 size_t size = max_jobs * sizeof(struct thread_data);
1016
1017                 shm_id = shmget(0, size, IPC_CREAT | 0600);
1018                 if (shm_id != -1)
1019                         break;
1020                 if (errno != EINVAL) {
1021                         perror("shmget");
1022                         break;
1023                 }
1024
1025                 max_jobs >>= 1;
1026         } while (max_jobs);
1027
1028         if (shm_id == -1)
1029                 return 1;
1030
1031         threads = shmat(shm_id, NULL, 0);
1032         if (threads == (void *) -1) {
1033                 perror("shmat");
1034                 return 1;
1035         }
1036
1037         atexit(free_shm);
1038         return 0;
1039 }
1040
1041 int parse_options(int argc, char *argv[])
1042 {
1043         int job_files, i;
1044
1045         if (setup_thread_area())
1046                 return 1;
1047         if (fill_def_thread())
1048                 return 1;
1049
1050         job_files = parse_cmd_line(argc, argv);
1051         if (!job_files) {
1052                 log_err("Need job file(s)\n");
1053                 usage();
1054                 return 1;
1055         }
1056
1057         for (i = 0; i < job_files; i++) {
1058                 if (fill_def_thread())
1059                         return 1;
1060                 if (parse_jobs_ini(ini_file[i], i))
1061                         return 1;
1062                 free(ini_file[i]);
1063         }
1064
1065         free(ini_file);
1066         return 0;
1067 }