backend: respect return value of init_io_u_buffers
[fio.git] / iolog.c
CommitLineData
ac9b9101
JA
1/*
2 * Code related to writing an iolog of what a thread is doing, and to
3 * later read that back and replay
4 */
5#include <stdio.h>
6#include <stdlib.h>
ac9b9101 7#include <assert.h>
b26317c9
JA
8#include <sys/types.h>
9#include <sys/stat.h>
10#include <unistd.h>
aee2ab67
JA
11#ifdef CONFIG_ZLIB
12#include <zlib.h>
13#endif
14
ac9b9101
JA
15#include "flist.h"
16#include "fio.h"
ac9b9101 17#include "trim.h"
243bfe19 18#include "filelock.h"
a47591e4 19#include "smalloc.h"
a3e59412 20#include "blktrace.h"
ae626d4e 21#include "pshared.h"
b5aba537 22#include "lib/roundup.h"
ac9b9101 23
2f8f4821
AK
24#include <netinet/in.h>
25#include <netinet/tcp.h>
26#include <arpa/inet.h>
27#include <sys/stat.h>
28#include <sys/socket.h>
29#include <sys/un.h>
30
7e419452
JA
31static int iolog_flush(struct io_log *log);
32
ac9b9101 33static const char iolog_ver2[] = "fio version 2 iolog";
315bbf01 34static const char iolog_ver3[] = "fio version 3 iolog";
ac9b9101
JA
35
36void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
37{
38 flist_add_tail(&ipo->list, &td->io_log_list);
39 td->total_io_size += ipo->len;
40}
41
1f440ece 42void log_io_u(const struct thread_data *td, const struct io_u *io_u)
ac9b9101 43{
e8cf24e5
MG
44 struct timespec now;
45
ac9b9101
JA
46 if (!td->o.write_iolog_file)
47 return;
48
e8cf24e5 49 fio_gettime(&now, NULL);
6d01ac19
JA
50 fprintf(td->iolog_f, "%llu %s %s %llu %llu\n",
51 (unsigned long long) utime_since_now(&td->io_log_start_time),
52 io_u->file->file_name, io_ddir_name(io_u->ddir), io_u->offset,
53 io_u->buflen);
e8cf24e5 54
ac9b9101
JA
55}
56
57void log_file(struct thread_data *td, struct fio_file *f,
58 enum file_log_act what)
59{
60 const char *act[] = { "add", "open", "close" };
e8cf24e5 61 struct timespec now;
ac9b9101
JA
62
63 assert(what < 3);
64
65 if (!td->o.write_iolog_file)
66 return;
67
68
69 /*
70 * this happens on the pre-open/close done before the job starts
71 */
72 if (!td->iolog_f)
73 return;
74
e8cf24e5 75 fio_gettime(&now, NULL);
6d01ac19
JA
76 fprintf(td->iolog_f, "%llu %s %s\n",
77 (unsigned long long) utime_since_now(&td->io_log_start_time),
78 f->file_name, act[what]);
ac9b9101
JA
79}
80
81static void iolog_delay(struct thread_data *td, unsigned long delay)
82{
df8fb609 83 uint64_t usec = utime_since_now(&td->last_issue);
f4b49c55 84 unsigned long orig_delay = delay;
df8fb609 85 uint64_t this_delay;
8b6a404c 86 struct timespec ts;
ac9b9101 87
df8fb609
JA
88 if (delay < td->time_offset) {
89 td->time_offset = 0;
90 return;
91 }
92
93 delay -= td->time_offset;
ac9b9101
JA
94 if (delay < usec)
95 return;
96
97 delay -= usec;
98
8b6a404c 99 fio_gettime(&ts, NULL);
30b18672
JA
100 while (delay && !td->terminate) {
101 this_delay = delay;
102 if (this_delay > 500000)
103 this_delay = 500000;
104
105 usec_sleep(td, this_delay);
106 delay -= this_delay;
107 }
df8fb609 108
8b6a404c 109 usec = utime_since_now(&ts);
f4b49c55
JA
110 if (usec > orig_delay)
111 td->time_offset = usec - orig_delay;
df8fb609
JA
112 else
113 td->time_offset = 0;
ac9b9101
JA
114}
115
116static int ipo_special(struct thread_data *td, struct io_piece *ipo)
117{
118 struct fio_file *f;
119 int ret;
120
121 /*
122 * Not a special ipo
123 */
124 if (ipo->ddir != DDIR_INVAL)
125 return 0;
126
127 f = td->files[ipo->fileno];
128
315bbf01
MG
129 if (ipo->delay)
130 iolog_delay(td, ipo->delay);
131 if (fio_fill_issue_time(td))
132 fio_gettime(&td->last_issue, NULL);
ac9b9101
JA
133 switch (ipo->file_action) {
134 case FIO_LOG_OPEN_FILE:
96d663ad
SW
135 if (td->o.replay_redirect && fio_file_open(f)) {
136 dprint(FD_FILE, "iolog: ignoring re-open of file %s\n",
137 f->file_name);
138 break;
139 }
ac9b9101
JA
140 ret = td_io_open_file(td, f);
141 if (!ret)
142 break;
143 td_verror(td, ret, "iolog open file");
144 return -1;
145 case FIO_LOG_CLOSE_FILE:
146 td_io_close_file(td, f);
147 break;
148 case FIO_LOG_UNLINK_FILE:
38ef9c90 149 td_io_unlink_file(td, f);
ac9b9101 150 break;
315bbf01
MG
151 case FIO_LOG_ADD_FILE:
152 /*
153 * Nothing to do
154 */
155 break;
ac9b9101
JA
156 default:
157 log_err("fio: bad file action %d\n", ipo->file_action);
158 break;
159 }
160
161 return 1;
162}
163
315bbf01
MG
164static bool read_iolog(struct thread_data *td);
165
166unsigned long long delay_since_ttime(const struct thread_data *td,
167 unsigned long long time)
168{
169 double tmp;
170 double scale;
171 const unsigned long long *last_ttime = &td->io_log_last_ttime;
172
173 if (!*last_ttime || td->o.no_stall || time < *last_ttime)
174 return 0;
175 else if (td->o.replay_time_scale == 100)
176 return time - *last_ttime;
177
178
179 scale = (double) 100.0 / (double) td->o.replay_time_scale;
180 tmp = time - *last_ttime;
181 return tmp * scale;
182}
98e7161c 183
ac9b9101
JA
184int read_iolog_get(struct thread_data *td, struct io_u *io_u)
185{
186 struct io_piece *ipo;
187 unsigned long elapsed;
3c3ed070 188
ac9b9101
JA
189 while (!flist_empty(&td->io_log_list)) {
190 int ret;
b61fdc8a 191
10f74940 192 if (td->o.read_iolog_chunked) {
98e7161c 193 if (td->io_log_checkmark == td->io_log_current) {
10f74940
LS
194 if (td->io_log_blktrace) {
195 if (!read_blktrace(td))
196 return 1;
197 } else {
315bbf01 198 if (!read_iolog(td))
10f74940
LS
199 return 1;
200 }
98e7161c
AK
201 }
202 td->io_log_current--;
203 }
9342d5f8 204 ipo = flist_first_entry(&td->io_log_list, struct io_piece, list);
ac9b9101
JA
205 flist_del(&ipo->list);
206 remove_trim_entry(td, ipo);
207
208 ret = ipo_special(td, ipo);
209 if (ret < 0) {
210 free(ipo);
211 break;
212 } else if (ret > 0) {
213 free(ipo);
214 continue;
215 }
216
217 io_u->ddir = ipo->ddir;
218 if (ipo->ddir != DDIR_WAIT) {
219 io_u->offset = ipo->offset;
4fff54cc 220 io_u->verify_offset = ipo->offset;
ac9b9101
JA
221 io_u->buflen = ipo->len;
222 io_u->file = td->files[ipo->fileno];
223 get_file(io_u->file);
5fff9543 224 dprint(FD_IO, "iolog: get %llu/%llu/%s\n", io_u->offset,
ac9b9101
JA
225 io_u->buflen, io_u->file->file_name);
226 if (ipo->delay)
227 iolog_delay(td, ipo->delay);
228 } else {
229 elapsed = mtime_since_genesis();
230 if (ipo->delay > elapsed)
231 usec_sleep(td, (ipo->delay - elapsed) * 1000);
ac9b9101
JA
232 }
233
234 free(ipo);
3c3ed070 235
ac9b9101
JA
236 if (io_u->ddir != DDIR_WAIT)
237 return 0;
238 }
239
240 td->done = 1;
241 return 1;
242}
243
244void prune_io_piece_log(struct thread_data *td)
245{
246 struct io_piece *ipo;
e96c0125 247 struct fio_rb_node *n;
ac9b9101
JA
248
249 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
250 ipo = rb_entry(n, struct io_piece, rb_node);
251 rb_erase(n, &td->io_hist_tree);
252 remove_trim_entry(td, ipo);
253 td->io_hist_len--;
254 free(ipo);
255 }
256
257 while (!flist_empty(&td->io_hist_list)) {
4e1020f8 258 ipo = flist_first_entry(&td->io_hist_list, struct io_piece, list);
ac9b9101
JA
259 flist_del(&ipo->list);
260 remove_trim_entry(td, ipo);
261 td->io_hist_len--;
262 free(ipo);
263 }
264}
265
266/*
267 * log a successful write, so we can unwind the log for verify
268 */
269void log_io_piece(struct thread_data *td, struct io_u *io_u)
270{
e96c0125 271 struct fio_rb_node **p, *parent;
ac9b9101
JA
272 struct io_piece *ipo, *__ipo;
273
8812d7f2 274 ipo = calloc(1, sizeof(struct io_piece));
ac9b9101
JA
275 init_ipo(ipo);
276 ipo->file = io_u->file;
277 ipo->offset = io_u->offset;
278 ipo->len = io_u->buflen;
da0a7bd2 279 ipo->numberio = io_u->numberio;
f9401285
JA
280 ipo->flags = IP_F_IN_FLIGHT;
281
282 io_u->ipo = ipo;
ac9b9101
JA
283
284 if (io_u_should_trim(td, io_u)) {
285 flist_add_tail(&ipo->trim_list, &td->trim_list);
286 td->trim_entries++;
287 }
288
289 /*
e43f210a 290 * Only sort writes if we don't have a random map in which case we need
4a06cb41
SW
291 * to check for duplicate blocks and drop the old one, which we rely on
292 * the rb insert/lookup for handling.
ac9b9101 293 */
f31feaa2 294 if (file_randommap(td, ipo->file)) {
ac9b9101
JA
295 INIT_FLIST_HEAD(&ipo->list);
296 flist_add_tail(&ipo->list, &td->io_hist_list);
297 ipo->flags |= IP_F_ONLIST;
298 td->io_hist_len++;
299 return;
300 }
301
302 RB_CLEAR_NODE(&ipo->rb_node);
303
304 /*
305 * Sort the entry into the verification list
306 */
307restart:
308 p = &td->io_hist_tree.rb_node;
309 parent = NULL;
310 while (*p) {
3b1ff869 311 int overlap = 0;
ac9b9101
JA
312 parent = *p;
313
314 __ipo = rb_entry(parent, struct io_piece, rb_node);
315 if (ipo->file < __ipo->file)
316 p = &(*p)->rb_left;
317 else if (ipo->file > __ipo->file)
318 p = &(*p)->rb_right;
3b1ff869 319 else if (ipo->offset < __ipo->offset) {
ac9b9101 320 p = &(*p)->rb_left;
3b1ff869
JE
321 overlap = ipo->offset + ipo->len > __ipo->offset;
322 }
323 else if (ipo->offset > __ipo->offset) {
ac9b9101 324 p = &(*p)->rb_right;
3b1ff869
JE
325 overlap = __ipo->offset + __ipo->len > ipo->offset;
326 }
327 else
328 overlap = 1;
329
330 if (overlap) {
7abe6a27 331 dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu\n",
885ac623
JA
332 __ipo->offset, __ipo->len,
333 ipo->offset, ipo->len);
ac9b9101
JA
334 td->io_hist_len--;
335 rb_erase(parent, &td->io_hist_tree);
336 remove_trim_entry(td, __ipo);
e8b74655
SW
337 if (!(__ipo->flags & IP_F_IN_FLIGHT))
338 free(__ipo);
ac9b9101
JA
339 goto restart;
340 }
341 }
342
343 rb_link_node(&ipo->rb_node, parent, p);
344 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
345 ipo->flags |= IP_F_ONRB;
346 td->io_hist_len++;
347}
348
890b6656
JA
349void unlog_io_piece(struct thread_data *td, struct io_u *io_u)
350{
351 struct io_piece *ipo = io_u->ipo;
352
66347cfa
DE
353 if (td->ts.nr_block_infos) {
354 uint32_t *info = io_u_block_info(td, io_u);
355 if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) {
356 if (io_u->ddir == DDIR_TRIM)
357 *info = BLOCK_INFO_SET_STATE(*info,
358 BLOCK_STATE_TRIM_FAILURE);
359 else if (io_u->ddir == DDIR_WRITE)
360 *info = BLOCK_INFO_SET_STATE(*info,
361 BLOCK_STATE_WRITE_FAILURE);
362 }
363 }
364
890b6656
JA
365 if (!ipo)
366 return;
367
368 if (ipo->flags & IP_F_ONRB)
369 rb_erase(&ipo->rb_node, &td->io_hist_tree);
370 else if (ipo->flags & IP_F_ONLIST)
371 flist_del(&ipo->list);
372
373 free(ipo);
374 io_u->ipo = NULL;
375 td->io_hist_len--;
376}
377
94a78d52 378void trim_io_piece(const struct io_u *io_u)
890b6656
JA
379{
380 struct io_piece *ipo = io_u->ipo;
381
382 if (!ipo)
383 return;
384
385 ipo->len = io_u->xfer_buflen - io_u->resid;
386}
387
ac9b9101
JA
388void write_iolog_close(struct thread_data *td)
389{
d087d28f
DLM
390 if (!td->iolog_f)
391 return;
392
ac9b9101
JA
393 fflush(td->iolog_f);
394 fclose(td->iolog_f);
395 free(td->iolog_buf);
396 td->iolog_f = NULL;
397 td->iolog_buf = NULL;
398}
399
a21ed2b5 400int64_t iolog_items_to_fetch(struct thread_data *td)
1d73ff2a
JA
401{
402 struct timespec now;
403 uint64_t elapsed;
404 uint64_t for_1s;
405 int64_t items_to_fetch;
406
407 if (!td->io_log_highmark)
408 return 10;
409
410
411 fio_gettime(&now, NULL);
412 elapsed = ntime_since(&td->io_log_highmark_time, &now);
413 if (elapsed) {
414 for_1s = (td->io_log_highmark - td->io_log_current) * 1000000000 / elapsed;
415 items_to_fetch = for_1s - td->io_log_current;
416 if (items_to_fetch < 0)
417 items_to_fetch = 0;
418 } else
419 items_to_fetch = 0;
420
421 td->io_log_highmark = td->io_log_current + items_to_fetch;
422 td->io_log_checkmark = (td->io_log_highmark + 1) / 2;
423 fio_gettime(&td->io_log_highmark_time, NULL);
424
425 return items_to_fetch;
426}
427
315bbf01
MG
428#define io_act(_td, _r) (((_td)->io_log_version == 3 && (r) == 5) || \
429 ((_td)->io_log_version == 2 && (r) == 4))
430#define file_act(_td, _r) (((_td)->io_log_version == 3 && (r) == 3) || \
431 ((_td)->io_log_version == 2 && (r) == 2))
432
ac9b9101 433/*
315bbf01 434 * Read version 2 and 3 iolog data. It is enhanced to include per-file logging,
ac9b9101
JA
435 * syncs, etc.
436 */
315bbf01 437static bool read_iolog(struct thread_data *td)
ac9b9101
JA
438{
439 unsigned long long offset;
440 unsigned int bytes;
315bbf01 441 unsigned long long delay = 0;
ac9b9101 442 int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */
dac34079 443 char *rfname, *fname, *act;
ac9b9101
JA
444 char *str, *p;
445 enum fio_ddir rw;
71e6e5a2 446 bool realloc = false;
98e7161c 447 int64_t items_to_fetch = 0;
78c0d7a0 448 int syncs;
ac9b9101 449
98e7161c 450 if (td->o.read_iolog_chunked) {
1d73ff2a
JA
451 items_to_fetch = iolog_items_to_fetch(td);
452 if (!items_to_fetch)
453 return true;
98e7161c 454 }
1d73ff2a 455
ac9b9101
JA
456 /*
457 * Read in the read iolog and store it, reuse the infrastructure
458 * for doing verifications.
459 */
460 str = malloc(4096);
dac34079 461 rfname = fname = malloc(256+16);
ac9b9101
JA
462 act = malloc(256+16);
463
78c0d7a0 464 syncs = reads = writes = waits = 0;
98e7161c 465 while ((p = fgets(str, 4096, td->io_log_rfile)) != NULL) {
ac9b9101
JA
466 struct io_piece *ipo;
467 int r;
315bbf01 468 unsigned long long ttime;
ac9b9101 469
315bbf01
MG
470 if (td->io_log_version == 3) {
471 r = sscanf(p, "%llu %256s %256s %llu %u", &ttime, rfname, act,
472 &offset, &bytes);
473 delay = delay_since_ttime(td, ttime);
474 td->io_log_last_ttime = ttime;
475 /*
476 * "wait" is not allowed with version 3
477 */
478 if (!strcmp(act, "wait")) {
479 log_err("iolog: ignoring wait command with"
480 " version 3 for file %s\n", fname);
481 continue;
482 }
483 } else /* version 2 */
484 r = sscanf(p, "%256s %256s %llu %u", rfname, act, &offset, &bytes);
dac34079
JA
485
486 if (td->o.replay_redirect)
487 fname = td->o.replay_redirect;
488
315bbf01 489 if (io_act(td, r)) {
ac9b9101
JA
490 /*
491 * Check action first
492 */
493 if (!strcmp(act, "wait"))
494 rw = DDIR_WAIT;
495 else if (!strcmp(act, "read"))
496 rw = DDIR_READ;
497 else if (!strcmp(act, "write"))
498 rw = DDIR_WRITE;
499 else if (!strcmp(act, "sync"))
500 rw = DDIR_SYNC;
501 else if (!strcmp(act, "datasync"))
502 rw = DDIR_DATASYNC;
503 else if (!strcmp(act, "trim"))
504 rw = DDIR_TRIM;
505 else {
506 log_err("fio: bad iolog file action: %s\n",
507 act);
508 continue;
509 }
033ace1e 510 fileno = get_fileno(td, fname);
315bbf01 511 } else if (file_act(td, r)) {
ac9b9101
JA
512 rw = DDIR_INVAL;
513 if (!strcmp(act, "add")) {
96d663ad
SW
514 if (td->o.replay_redirect &&
515 get_fileno(td, fname) != -1) {
516 dprint(FD_FILE, "iolog: ignoring"
517 " re-add of file %s\n", fname);
518 } else {
c70c7f58 519 fileno = add_file(td, fname, td->subjob_number, 1);
96d663ad
SW
520 file_action = FIO_LOG_ADD_FILE;
521 }
ac9b9101
JA
522 } else if (!strcmp(act, "open")) {
523 fileno = get_fileno(td, fname);
524 file_action = FIO_LOG_OPEN_FILE;
525 } else if (!strcmp(act, "close")) {
526 fileno = get_fileno(td, fname);
527 file_action = FIO_LOG_CLOSE_FILE;
528 } else {
529 log_err("fio: bad iolog file action: %s\n",
530 act);
531 continue;
532 }
533 } else {
315bbf01 534 log_err("bad iolog%d: %s\n", td->io_log_version, p);
ac9b9101
JA
535 continue;
536 }
537
538 if (rw == DDIR_READ)
539 reads++;
540 else if (rw == DDIR_WRITE) {
541 /*
542 * Don't add a write for ro mode
543 */
544 if (read_only)
545 continue;
546 writes++;
547 } else if (rw == DDIR_WAIT) {
74dff443
JA
548 if (td->o.no_stall)
549 continue;
ac9b9101
JA
550 waits++;
551 } else if (rw == DDIR_INVAL) {
78c0d7a0
JA
552 } else if (ddir_sync(rw)) {
553 syncs++;
554 } else {
ac9b9101
JA
555 log_err("bad ddir: %d\n", rw);
556 continue;
557 }
558
559 /*
560 * Make note of file
561 */
8812d7f2 562 ipo = calloc(1, sizeof(*ipo));
ac9b9101
JA
563 init_ipo(ipo);
564 ipo->ddir = rw;
315bbf01
MG
565 if (td->io_log_version == 3)
566 ipo->delay = delay;
ac9b9101
JA
567 if (rw == DDIR_WAIT) {
568 ipo->delay = offset;
569 } else {
a79f17bf
SW
570 if (td->o.replay_scale)
571 ipo->offset = offset / td->o.replay_scale;
572 else
573 ipo->offset = offset;
574 ipo_bytes_align(td->o.replay_align, ipo);
575
ac9b9101 576 ipo->len = bytes;
71e6e5a2
AK
577 if (rw != DDIR_INVAL && bytes > td->o.max_bs[rw]) {
578 realloc = true;
ac9b9101 579 td->o.max_bs[rw] = bytes;
71e6e5a2 580 }
ac9b9101
JA
581 ipo->fileno = fileno;
582 ipo->file_action = file_action;
70afff59 583 td->o.size += bytes;
ac9b9101 584 }
3c3ed070 585
ac9b9101 586 queue_io_piece(td, ipo);
98e7161c
AK
587
588 if (td->o.read_iolog_chunked) {
589 td->io_log_current++;
590 items_to_fetch--;
591 if (items_to_fetch == 0)
592 break;
593 }
ac9b9101
JA
594 }
595
596 free(str);
597 free(act);
dac34079 598 free(rfname);
ac9b9101 599
98e7161c
AK
600 if (td->o.read_iolog_chunked) {
601 td->io_log_highmark = td->io_log_current;
602 td->io_log_checkmark = (td->io_log_highmark + 1) / 2;
603 fio_gettime(&td->io_log_highmark_time, NULL);
604 }
605
ac9b9101
JA
606 if (writes && read_only) {
607 log_err("fio: <%s> skips replay of %d writes due to"
608 " read-only\n", td->o.name, writes);
609 writes = 0;
610 }
78c0d7a0
JA
611 if (syncs)
612 td->flags |= TD_F_SYNCS;
ac9b9101 613
98e7161c
AK
614 if (td->o.read_iolog_chunked) {
615 if (td->io_log_current == 0) {
616 return false;
617 }
618 td->o.td_ddir = TD_DDIR_RW;
71e6e5a2
AK
619 if (realloc && td->orig_buffer)
620 {
621 io_u_quiesce(td);
622 free_io_mem(td);
6d8fe6e8
SK
623 if (init_io_u_buffers(td))
624 return false;
71e6e5a2 625 }
98e7161c
AK
626 return true;
627 }
628
ac9b9101 629 if (!reads && !writes && !waits)
b153f94a 630 return false;
ac9b9101
JA
631 else if (reads && !writes)
632 td->o.td_ddir = TD_DDIR_READ;
633 else if (!reads && writes)
634 td->o.td_ddir = TD_DDIR_WRITE;
635 else
636 td->o.td_ddir = TD_DDIR_RW;
637
b153f94a 638 return true;
ac9b9101
JA
639}
640
2f8f4821
AK
641static bool is_socket(const char *path)
642{
643 struct stat buf;
8ea203d3
JA
644 int r;
645
646 r = stat(path, &buf);
2f8f4821
AK
647 if (r == -1)
648 return false;
649
650 return S_ISSOCK(buf.st_mode);
651}
652
653static int open_socket(const char *path)
654{
2f8f4821 655 struct sockaddr_un addr;
8ea203d3
JA
656 int ret, fd;
657
658 fd = socket(AF_UNIX, SOCK_STREAM, 0);
2f8f4821
AK
659 if (fd < 0)
660 return fd;
8ea203d3 661
2f8f4821 662 addr.sun_family = AF_UNIX;
c7d23f4b 663 if (snprintf(addr.sun_path, sizeof(addr.sun_path), "%s", path) >=
8ea203d3 664 sizeof(addr.sun_path)) {
c7d23f4b
BVA
665 log_err("%s: path name %s is too long for a Unix socket\n",
666 __func__, path);
8ea203d3
JA
667 }
668
669 ret = connect(fd, (const struct sockaddr *)&addr, strlen(path) + sizeof(addr.sun_family));
670 if (!ret)
2f8f4821 671 return fd;
8ea203d3
JA
672
673 close(fd);
2f8f4821
AK
674 return -1;
675}
676
ac9b9101
JA
677/*
678 * open iolog, check version, and call appropriate parser
679 */
22d7fdf3 680static bool init_iolog_read(struct thread_data *td, char *fname)
ac9b9101 681{
22d7fdf3 682 char buffer[256], *p;
2f8f4821 683 FILE *f = NULL;
8ea203d3 684
c70c7f58
AK
685 dprint(FD_IO, "iolog: name=%s\n", fname);
686
687 if (is_socket(fname)) {
8ea203d3
JA
688 int fd;
689
690 fd = open_socket(fname);
691 if (fd >= 0)
2f8f4821 692 f = fdopen(fd, "r");
d19c04d1 693 } else if (!strcmp(fname, "-")) {
694 f = stdin;
2f8f4821 695 } else
c70c7f58 696 f = fopen(fname, "r");
8ea203d3 697
ac9b9101
JA
698 if (!f) {
699 perror("fopen read iolog");
b153f94a 700 return false;
ac9b9101
JA
701 }
702
703 p = fgets(buffer, sizeof(buffer), f);
704 if (!p) {
705 td_verror(td, errno, "iolog read");
706 log_err("fio: unable to read iolog\n");
707 fclose(f);
b153f94a 708 return false;
ac9b9101 709 }
8ea203d3 710
ac9b9101 711 /*
315bbf01 712 * versions 2 and 3 of the iolog store a specific string as the
ac9b9101
JA
713 * first line, check for that
714 */
315bbf01
MG
715 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
716 td->io_log_version = 2;
717 else if (!strncmp(iolog_ver3, buffer, strlen(iolog_ver3)))
718 td->io_log_version = 3;
719 else {
720 log_err("fio: iolog version 1 is no longer supported\n");
721 fclose(f);
722 return false;
ac9b9101
JA
723 }
724
315bbf01
MG
725 free_release_files(td);
726 td->io_log_rfile = f;
727 return read_iolog(td);
ac9b9101
JA
728}
729
730/*
731 * Set up a log for storing io patterns.
732 */
b153f94a 733static bool init_iolog_write(struct thread_data *td)
ac9b9101
JA
734{
735 struct fio_file *ff;
736 FILE *f;
737 unsigned int i;
738
739 f = fopen(td->o.write_iolog_file, "a");
740 if (!f) {
741 perror("fopen write iolog");
b153f94a 742 return false;
ac9b9101
JA
743 }
744
745 /*
746 * That's it for writing, setup a log buffer and we're done.
747 */
748 td->iolog_f = f;
749 td->iolog_buf = malloc(8192);
750 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
e8cf24e5 751 fio_gettime(&td->io_log_start_time, NULL);
ac9b9101
JA
752
753 /*
754 * write our version line
755 */
e8cf24e5 756 if (fprintf(f, "%s\n", iolog_ver3) < 0) {
ac9b9101 757 perror("iolog init\n");
b153f94a 758 return false;
ac9b9101
JA
759 }
760
761 /*
762 * add all known files
763 */
764 for_each_file(td, ff, i)
765 log_file(td, ff, FIO_LOG_ADD_FILE);
766
b153f94a 767 return true;
ac9b9101
JA
768}
769
b153f94a 770bool init_iolog(struct thread_data *td)
ac9b9101 771{
b153f94a 772 bool ret;
ac9b9101
JA
773
774 if (td->o.read_iolog_file) {
d95b34a6 775 int need_swap;
22d7fdf3 776 char * fname = get_name_by_idx(td->o.read_iolog_file, td->subjob_number);
d95b34a6 777
ac9b9101
JA
778 /*
779 * Check if it's a blktrace file and load that if possible.
780 * Otherwise assume it's a normal log file and load that.
781 */
b61fdc8a
JA
782 if (is_blktrace(fname, &need_swap)) {
783 td->io_log_blktrace = 1;
10f74940 784 ret = init_blktrace_read(td, fname, need_swap);
b61fdc8a
JA
785 } else {
786 td->io_log_blktrace = 0;
22d7fdf3 787 ret = init_iolog_read(td, fname);
b61fdc8a 788 }
3a3e5c6e 789 free(fname);
ac9b9101
JA
790 } else if (td->o.write_iolog_file)
791 ret = init_iolog_write(td);
b153f94a 792 else
43864deb 793 ret = true;
ac9b9101 794
b153f94a 795 if (!ret)
f01b34ae
JA
796 td_verror(td, EINVAL, "failed initializing iolog");
797
ac9b9101
JA
798 return ret;
799}
800
aee2ab67
JA
801void setup_log(struct io_log **log, struct log_params *p,
802 const char *filename)
ac9b9101 803{
e75cc8f3 804 struct io_log *l;
65a4d15c
KC
805 int i;
806 struct io_u_plat_entry *entry;
807 struct flist_head *list;
ac9b9101 808
7e419452
JA
809 l = scalloc(1, sizeof(*l));
810 INIT_FLIST_HEAD(&l->io_logs);
aee2ab67
JA
811 l->log_type = p->log_type;
812 l->log_offset = p->log_offset;
03ec570f 813 l->log_prio = p->log_prio;
aee2ab67 814 l->log_gz = p->log_gz;
b26317c9 815 l->log_gz_store = p->log_gz_store;
aee2ab67 816 l->avg_msec = p->avg_msec;
1e613c9c
KC
817 l->hist_msec = p->hist_msec;
818 l->hist_coarseness = p->hist_coarseness;
cb7e0ace 819 l->filename = strdup(filename);
aee2ab67
JA
820 l->td = p->td;
821
65a4d15c
KC
822 /* Initialize histogram lists for each r/w direction,
823 * with initial io_u_plat of all zeros:
824 */
825 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
d730bc58 826 list = &l->hist_window[i].list;
65a4d15c
KC
827 INIT_FLIST_HEAD(list);
828 entry = calloc(1, sizeof(struct io_u_plat_entry));
829 flist_add(&entry->list, list);
830 }
831
1fed2080 832 if (l->td && l->td->o.io_submit_mode != IO_MODE_OFFLOAD) {
b5aba537 833 unsigned int def_samples = DEF_LOG_ENTRIES;
a43f4461 834 struct io_logs *__p;
1fed2080 835
a43f4461 836 __p = calloc(1, sizeof(*l->pending));
b5aba537
JA
837 if (l->td->o.iodepth > DEF_LOG_ENTRIES)
838 def_samples = roundup_pow2(l->td->o.iodepth);
839 __p->max_samples = def_samples;
a43f4461
JA
840 __p->log = calloc(__p->max_samples, log_entry_sz(l));
841 l->pending = __p;
1fed2080
JA
842 }
843
b26317c9 844 if (l->log_offset)
49e98daa 845 l->log_ddir_mask = LOG_OFFSET_SAMPLE_BIT;
03ec570f
DLM
846 if (l->log_prio)
847 l->log_ddir_mask |= LOG_PRIO_SAMPLE_BIT;
b26317c9 848
aee2ab67
JA
849 INIT_FLIST_HEAD(&l->chunk_list);
850
851 if (l->log_gz && !p->td)
852 l->log_gz = 0;
19417eda 853 else if (l->log_gz || l->log_gz_store) {
34febb23 854 mutex_init_pshared(&l->chunk_lock);
858bce70 855 mutex_init_pshared(&l->deferred_free_lock);
aee2ab67
JA
856 p->td->flags |= TD_F_COMPRESS_LOG;
857 }
858
ac9b9101
JA
859 *log = l;
860}
861
2e802282
JA
862#ifdef CONFIG_SETVBUF
863static void *set_file_buffer(FILE *f)
864{
865 size_t size = 1048576;
866 void *buf;
867
868 buf = malloc(size);
869 setvbuf(f, buf, _IOFBF, size);
870 return buf;
871}
872
873static void clear_file_buffer(void *buf)
874{
875 free(buf);
876}
877#else
878static void *set_file_buffer(FILE *f)
879{
880 return NULL;
881}
882
883static void clear_file_buffer(void *buf)
884{
885}
886#endif
887
518dac09 888void free_log(struct io_log *log)
cb7e0ace 889{
7e419452
JA
890 while (!flist_empty(&log->io_logs)) {
891 struct io_logs *cur_log;
892
893 cur_log = flist_first_entry(&log->io_logs, struct io_logs, list);
894 flist_del_init(&cur_log->list);
895 free(cur_log->log);
2b7625e2 896 sfree(cur_log);
7e419452
JA
897 }
898
1fed2080
JA
899 if (log->pending) {
900 free(log->pending->log);
901 free(log->pending);
902 log->pending = NULL;
903 }
904
905 free(log->pending);
cb7e0ace 906 free(log->filename);
a47591e4 907 sfree(log);
cb7e0ace
JA
908}
909
6cc0e5aa
AL
910uint64_t hist_sum(int j, int stride, uint64_t *io_u_plat,
911 uint64_t *io_u_plat_last)
548df930 912{
6cc0e5aa 913 uint64_t sum;
f2c972cc 914 int k;
548df930 915
868f6f03
KC
916 if (io_u_plat_last) {
917 for (k = sum = 0; k < stride; k++)
918 sum += io_u_plat[j + k] - io_u_plat_last[j + k];
919 } else {
920 for (k = sum = 0; k < stride; k++)
921 sum += io_u_plat[j + k];
922 }
548df930 923
1e613c9c
KC
924 return sum;
925}
926
a89ba4b1
JA
927static void flush_hist_samples(FILE *f, int hist_coarseness, void *samples,
928 uint64_t sample_size)
1e613c9c
KC
929{
930 struct io_sample *s;
931 int log_offset;
932 uint64_t i, j, nr_samples;
65a4d15c 933 struct io_u_plat_entry *entry, *entry_before;
6cc0e5aa
AL
934 uint64_t *io_u_plat;
935 uint64_t *io_u_plat_before;
1e613c9c
KC
936
937 int stride = 1 << hist_coarseness;
938
939 if (!sample_size)
940 return;
941
942 s = __get_sample(samples, 0, 0);
943 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
944
945 nr_samples = sample_size / __log_entry_sz(log_offset);
946
947 for (i = 0; i < nr_samples; i++) {
948 s = __get_sample(samples, log_offset, i);
d730bc58 949
af2fde19 950 entry = s->data.plat_entry;
65a4d15c 951 io_u_plat = entry->io_u_plat;
d730bc58 952
65a4d15c
KC
953 entry_before = flist_first_entry(&entry->list, struct io_u_plat_entry, list);
954 io_u_plat_before = entry_before->io_u_plat;
d730bc58 955
5fff9543
JF
956 fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time,
957 io_sample_ddir(s), (unsigned long long) s->bs);
1e613c9c 958 for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) {
6cc0e5aa
AL
959 fprintf(f, "%llu, ", (unsigned long long)
960 hist_sum(j, stride, io_u_plat, io_u_plat_before));
1e613c9c 961 }
6cc0e5aa 962 fprintf(f, "%llu\n", (unsigned long long)
65a4d15c 963 hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat,
d730bc58
JA
964 io_u_plat_before));
965
65a4d15c
KC
966 flist_del(&entry_before->list);
967 free(entry_before);
1e613c9c
KC
968 }
969}
970
bead01d7 971void flush_samples(FILE *f, void *samples, uint64_t sample_size)
ac9b9101 972{
b26317c9 973 struct io_sample *s;
03ec570f 974 int log_offset, log_prio;
b26317c9 975 uint64_t i, nr_samples;
03ec570f
DLM
976 unsigned int prio_val;
977 const char *fmt;
b26317c9
JA
978
979 if (!sample_size)
980 return;
981
982 s = __get_sample(samples, 0, 0);
49e98daa 983 log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0;
03ec570f
DLM
984 log_prio = (s->__ddir & LOG_PRIO_SAMPLE_BIT) != 0;
985
986 if (log_offset) {
987 if (log_prio)
988 fmt = "%lu, %" PRId64 ", %u, %llu, %llu, 0x%04x\n";
989 else
990 fmt = "%lu, %" PRId64 ", %u, %llu, %llu, %u\n";
991 } else {
992 if (log_prio)
993 fmt = "%lu, %" PRId64 ", %u, %llu, 0x%04x\n";
994 else
995 fmt = "%lu, %" PRId64 ", %u, %llu, %u\n";
996 }
b26317c9
JA
997
998 nr_samples = sample_size / __log_entry_sz(log_offset);
ac9b9101 999
aee2ab67 1000 for (i = 0; i < nr_samples; i++) {
b26317c9 1001 s = __get_sample(samples, log_offset, i);
ac9b9101 1002
03ec570f
DLM
1003 if (log_prio)
1004 prio_val = s->priority;
1005 else
1006 prio_val = ioprio_value_is_class_rt(s->priority);
1007
aee2ab67 1008 if (!log_offset) {
03ec570f
DLM
1009 fprintf(f, fmt,
1010 (unsigned long) s->time,
1011 s->data.val,
1012 io_sample_ddir(s), (unsigned long long) s->bs,
1013 prio_val);
ae588852
JA
1014 } else {
1015 struct io_sample_offset *so = (void *) s;
1016
03ec570f
DLM
1017 fprintf(f, fmt,
1018 (unsigned long) s->time,
1019 s->data.val,
1020 io_sample_ddir(s), (unsigned long long) s->bs,
1021 (unsigned long long) so->offset,
1022 prio_val);
ae588852 1023 }
ac9b9101 1024 }
aee2ab67
JA
1025}
1026
1027#ifdef CONFIG_ZLIB
97eabb2a
JA
1028
1029struct iolog_flush_data {
155f2f02 1030 struct workqueue_work work;
97eabb2a
JA
1031 struct io_log *log;
1032 void *samples;
7e419452
JA
1033 uint32_t nr_samples;
1034 bool free;
97eabb2a
JA
1035};
1036
97eabb2a
JA
1037#define GZ_CHUNK 131072
1038
1039static struct iolog_compress *get_new_chunk(unsigned int seq)
1040{
1041 struct iolog_compress *c;
1042
1043 c = malloc(sizeof(*c));
1044 INIT_FLIST_HEAD(&c->list);
1045 c->buf = malloc(GZ_CHUNK);
1046 c->len = 0;
1047 c->seq = seq;
97eabb2a
JA
1048 return c;
1049}
1050
1051static void free_chunk(struct iolog_compress *ic)
1052{
c07826cd
JA
1053 free(ic->buf);
1054 free(ic);
97eabb2a
JA
1055}
1056
b26317c9 1057static int z_stream_init(z_stream *stream, int gz_hdr)
aee2ab67 1058{
b26317c9
JA
1059 int wbits = 15;
1060
acb2a69b 1061 memset(stream, 0, sizeof(*stream));
aee2ab67
JA
1062 stream->zalloc = Z_NULL;
1063 stream->zfree = Z_NULL;
1064 stream->opaque = Z_NULL;
1065 stream->next_in = Z_NULL;
1066
1a8e7458
JA
1067 /*
1068 * zlib magic - add 32 for auto-detection of gz header or not,
1069 * if we decide to store files in a gzip friendly format.
1070 */
b26317c9
JA
1071 if (gz_hdr)
1072 wbits += 32;
1073
1074 if (inflateInit2(stream, wbits) != Z_OK)
aee2ab67
JA
1075 return 1;
1076
1077 return 0;
1078}
1079
1a8e7458 1080struct inflate_chunk_iter {
aee2ab67 1081 unsigned int seq;
dad95da1 1082 int err;
aee2ab67
JA
1083 void *buf;
1084 size_t buf_size;
1085 size_t buf_used;
1086 size_t chunk_sz;
1087};
1088
b26317c9 1089static void finish_chunk(z_stream *stream, FILE *f,
1a8e7458 1090 struct inflate_chunk_iter *iter)
aee2ab67 1091{
aee2ab67
JA
1092 int ret;
1093
1094 ret = inflateEnd(stream);
1095 if (ret != Z_OK)
50f940f1
JA
1096 log_err("fio: failed to end log inflation seq %d (%d)\n",
1097 iter->seq, ret);
aee2ab67 1098
b26317c9 1099 flush_samples(f, iter->buf, iter->buf_used);
aee2ab67
JA
1100 free(iter->buf);
1101 iter->buf = NULL;
1102 iter->buf_size = iter->buf_used = 0;
1103}
1104
1a8e7458
JA
1105/*
1106 * Iterative chunk inflation. Handles cases where we cross into a new
1107 * sequence, doing flush finish of previous chunk if needed.
1108 */
1109static size_t inflate_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f,
1110 z_stream *stream, struct inflate_chunk_iter *iter)
aee2ab67 1111{
0c56718d
JA
1112 size_t ret;
1113
0f989d2e 1114 dprint(FD_COMPRESS, "inflate chunk size=%lu, seq=%u\n",
0c56718d
JA
1115 (unsigned long) ic->len, ic->seq);
1116
aee2ab67
JA
1117 if (ic->seq != iter->seq) {
1118 if (iter->seq)
b26317c9 1119 finish_chunk(stream, f, iter);
aee2ab67 1120
b26317c9 1121 z_stream_init(stream, gz_hdr);
aee2ab67
JA
1122 iter->seq = ic->seq;
1123 }
1124
1125 stream->avail_in = ic->len;
1126 stream->next_in = ic->buf;
1127
1128 if (!iter->buf_size) {
1129 iter->buf_size = iter->chunk_sz;
1130 iter->buf = malloc(iter->buf_size);
1131 }
1132
1133 while (stream->avail_in) {
b26317c9 1134 size_t this_out = iter->buf_size - iter->buf_used;
aee2ab67
JA
1135 int err;
1136
b26317c9 1137 stream->avail_out = this_out;
aee2ab67
JA
1138 stream->next_out = iter->buf + iter->buf_used;
1139
1140 err = inflate(stream, Z_NO_FLUSH);
1141 if (err < 0) {
1142 log_err("fio: failed inflating log: %d\n", err);
dad95da1 1143 iter->err = err;
aee2ab67
JA
1144 break;
1145 }
1146
b26317c9
JA
1147 iter->buf_used += this_out - stream->avail_out;
1148
1149 if (!stream->avail_out) {
1150 iter->buf_size += iter->chunk_sz;
1151 iter->buf = realloc(iter->buf, iter->buf_size);
1152 continue;
1153 }
1154
1155 if (err == Z_STREAM_END)
1156 break;
aee2ab67
JA
1157 }
1158
0c56718d
JA
1159 ret = (void *) stream->next_in - ic->buf;
1160
d0d0cf0a 1161 dprint(FD_COMPRESS, "inflated to size=%lu\n", (unsigned long) iter->buf_size);
0c56718d
JA
1162
1163 return ret;
aee2ab67
JA
1164}
1165
1a8e7458
JA
1166/*
1167 * Inflate stored compressed chunks, or write them directly to the log
1168 * file if so instructed.
1169 */
dad95da1 1170static int inflate_gz_chunks(struct io_log *log, FILE *f)
aee2ab67 1171{
1a8e7458 1172 struct inflate_chunk_iter iter = { .chunk_sz = log->log_gz, };
aee2ab67
JA
1173 z_stream stream;
1174
1175 while (!flist_empty(&log->chunk_list)) {
1176 struct iolog_compress *ic;
1177
9342d5f8 1178 ic = flist_first_entry(&log->chunk_list, struct iolog_compress, list);
aee2ab67 1179 flist_del(&ic->list);
b26317c9 1180
1a8e7458
JA
1181 if (log->log_gz_store) {
1182 size_t ret;
1183
0c56718d
JA
1184 dprint(FD_COMPRESS, "log write chunk size=%lu, "
1185 "seq=%u\n", (unsigned long) ic->len, ic->seq);
1186
1a8e7458 1187 ret = fwrite(ic->buf, ic->len, 1, f);
dad95da1
JA
1188 if (ret != 1 || ferror(f)) {
1189 iter.err = errno;
1a8e7458 1190 log_err("fio: error writing compressed log\n");
dad95da1 1191 }
1a8e7458
JA
1192 } else
1193 inflate_chunk(ic, log->log_gz_store, f, &stream, &iter);
c07826cd
JA
1194
1195 free_chunk(ic);
aee2ab67
JA
1196 }
1197
1198 if (iter.seq) {
b26317c9 1199 finish_chunk(&stream, f, &iter);
aee2ab67
JA
1200 free(iter.buf);
1201 }
dad95da1
JA
1202
1203 return iter.err;
aee2ab67
JA
1204}
1205
1a8e7458
JA
1206/*
1207 * Open compressed log file and decompress the stored chunks and
1208 * write them to stdout. The chunks are stored sequentially in the
1209 * file, so we iterate over them and do them one-by-one.
1210 */
b26317c9
JA
1211int iolog_file_inflate(const char *file)
1212{
1a8e7458 1213 struct inflate_chunk_iter iter = { .chunk_sz = 64 * 1024 * 1024, };
b26317c9
JA
1214 struct iolog_compress ic;
1215 z_stream stream;
1216 struct stat sb;
2b8b8f0d 1217 size_t ret;
f302710c
JA
1218 size_t total;
1219 void *buf;
b26317c9
JA
1220 FILE *f;
1221
a7c386ed 1222 f = fopen(file, "rb");
b26317c9
JA
1223 if (!f) {
1224 perror("fopen");
1225 return 1;
1226 }
1227
fd771971
JA
1228 if (stat(file, &sb) < 0) {
1229 fclose(f);
1230 perror("stat");
1231 return 1;
1232 }
1233
f302710c 1234 ic.buf = buf = malloc(sb.st_size);
b26317c9 1235 ic.len = sb.st_size;
b26317c9
JA
1236 ic.seq = 1;
1237
1238 ret = fread(ic.buf, ic.len, 1, f);
2b8b8f0d 1239 if (ret == 0 && ferror(f)) {
b26317c9
JA
1240 perror("fread");
1241 fclose(f);
bb1116fe 1242 free(buf);
b26317c9 1243 return 1;
2b8b8f0d 1244 } else if (ferror(f) || (!feof(f) && ret != 1)) {
b26317c9
JA
1245 log_err("fio: short read on reading log\n");
1246 fclose(f);
bb1116fe 1247 free(buf);
b26317c9
JA
1248 return 1;
1249 }
1250
1251 fclose(f);
1252
1a8e7458
JA
1253 /*
1254 * Each chunk will return Z_STREAM_END. We don't know how many
1255 * chunks are in the file, so we just keep looping and incrementing
1256 * the sequence number until we have consumed the whole compressed
1257 * file.
1258 */
f302710c
JA
1259 total = ic.len;
1260 do {
8a1db9a1 1261 size_t iret;
f302710c 1262
8a1db9a1
JA
1263 iret = inflate_chunk(&ic, 1, stdout, &stream, &iter);
1264 total -= iret;
f302710c
JA
1265 if (!total)
1266 break;
dad95da1
JA
1267 if (iter.err)
1268 break;
f302710c
JA
1269
1270 ic.seq++;
8a1db9a1
JA
1271 ic.len -= iret;
1272 ic.buf += iret;
f302710c 1273 } while (1);
b26317c9
JA
1274
1275 if (iter.seq) {
1276 finish_chunk(&stream, stdout, &iter);
1277 free(iter.buf);
1278 }
1279
f302710c 1280 free(buf);
dad95da1 1281 return iter.err;
b26317c9
JA
1282}
1283
aee2ab67
JA
1284#else
1285
dad95da1 1286static int inflate_gz_chunks(struct io_log *log, FILE *f)
aee2ab67 1287{
b7364e21 1288 return 0;
aee2ab67
JA
1289}
1290
a68c66b2
JA
1291int iolog_file_inflate(const char *file)
1292{
1293 log_err("fio: log inflation not possible without zlib\n");
1294 return 1;
1295}
1296
aee2ab67
JA
1297#endif
1298
60a25727 1299void flush_log(struct io_log *log, bool do_append)
aee2ab67
JA
1300{
1301 void *buf;
1302 FILE *f;
1303
a7c386ed 1304 /*
1305 * If log_gz_store is true, we are writing a binary file.
1306 * Set the mode appropriately (on all platforms) to avoid issues
1307 * on windows (line-ending conversions, etc.)
1308 */
3a5db920 1309 if (!do_append)
a7c386ed 1310 if (log->log_gz_store)
1311 f = fopen(log->filename, "wb");
1312 else
1313 f = fopen(log->filename, "w");
3a5db920 1314 else
a7c386ed 1315 if (log->log_gz_store)
1316 f = fopen(log->filename, "ab");
1317 else
1318 f = fopen(log->filename, "a");
aee2ab67
JA
1319 if (!f) {
1320 perror("fopen log");
1321 return;
1322 }
1323
1324 buf = set_file_buffer(f);
1325
1a8e7458 1326 inflate_gz_chunks(log, f);
aee2ab67 1327
7e419452
JA
1328 while (!flist_empty(&log->io_logs)) {
1329 struct io_logs *cur_log;
1330
1331 cur_log = flist_first_entry(&log->io_logs, struct io_logs, list);
1332 flist_del_init(&cur_log->list);
1e613c9c 1333
a9179eeb 1334 if (log->td && log == log->td->clat_hist_log)
1e613c9c 1335 flush_hist_samples(f, log->hist_coarseness, cur_log->log,
868f6f03 1336 log_sample_sz(log, cur_log));
1e613c9c 1337 else
868f6f03 1338 flush_samples(f, cur_log->log, log_sample_sz(log, cur_log));
1e613c9c 1339
90d2d53f 1340 sfree(cur_log);
7e419452 1341 }
ac9b9101
JA
1342
1343 fclose(f);
2e802282 1344 clear_file_buffer(buf);
ac9b9101
JA
1345}
1346
cb7e0ace 1347static int finish_log(struct thread_data *td, struct io_log *log, int trylock)
ac9b9101 1348{
155f2f02 1349 if (td->flags & TD_F_COMPRESS_LOG)
7e419452 1350 iolog_flush(log);
aee2ab67 1351
243bfe19 1352 if (trylock) {
cb7e0ace 1353 if (fio_trylock_file(log->filename))
243bfe19
JA
1354 return 1;
1355 } else
cb7e0ace 1356 fio_lock_file(log->filename);
243bfe19 1357
0cba0f91 1358 if (td->client_type == FIO_CLIENT_TYPE_GUI || is_backend)
cb7e0ace 1359 fio_send_iolog(td, log, log->filename);
aee2ab67 1360 else
3a5db920 1361 flush_log(log, !td->o.per_job_logs);
243bfe19 1362
cb7e0ace 1363 fio_unlock_file(log->filename);
518dac09 1364 free_log(log);
243bfe19 1365 return 0;
ac9b9101
JA
1366}
1367
0cba0f91
JA
1368size_t log_chunk_sizes(struct io_log *log)
1369{
1370 struct flist_head *entry;
1371 size_t ret;
1372
1373 if (flist_empty(&log->chunk_list))
1374 return 0;
1375
1376 ret = 0;
1377 pthread_mutex_lock(&log->chunk_lock);
1378 flist_for_each(entry, &log->chunk_list) {
1379 struct iolog_compress *c;
1380
1381 c = flist_entry(entry, struct iolog_compress, list);
1382 ret += c->len;
1383 }
1384 pthread_mutex_unlock(&log->chunk_lock);
1385 return ret;
1386}
1387
aee2ab67
JA
1388#ifdef CONFIG_ZLIB
1389
858bce70
JA
1390static void iolog_put_deferred(struct io_log *log, void *ptr)
1391{
1392 if (!ptr)
1393 return;
1394
1395 pthread_mutex_lock(&log->deferred_free_lock);
1396 if (log->deferred < IOLOG_MAX_DEFER) {
1397 log->deferred_items[log->deferred] = ptr;
1398 log->deferred++;
e2446166 1399 } else if (!fio_did_warn(FIO_WARN_IOLOG_DROP))
858bce70 1400 log_err("fio: had to drop log entry free\n");
858bce70
JA
1401 pthread_mutex_unlock(&log->deferred_free_lock);
1402}
1403
1404static void iolog_free_deferred(struct io_log *log)
1405{
1406 int i;
1407
1408 if (!log->deferred)
1409 return;
1410
1411 pthread_mutex_lock(&log->deferred_free_lock);
1412
1413 for (i = 0; i < log->deferred; i++) {
1414 free(log->deferred_items[i]);
1415 log->deferred_items[i] = NULL;
1416 }
1417
1418 log->deferred = 0;
1419 pthread_mutex_unlock(&log->deferred_free_lock);
1420}
1421
78e93aa6 1422static int gz_work(struct iolog_flush_data *data)
c143980a 1423{
040238ec 1424 struct iolog_compress *c = NULL;
aee2ab67
JA
1425 struct flist_head list;
1426 unsigned int seq;
1427 z_stream stream;
1428 size_t total = 0;
d73ac887 1429 int ret;
aee2ab67
JA
1430
1431 INIT_FLIST_HEAD(&list);
1432
7e419452 1433 memset(&stream, 0, sizeof(stream));
aee2ab67
JA
1434 stream.zalloc = Z_NULL;
1435 stream.zfree = Z_NULL;
1436 stream.opaque = Z_NULL;
1437
b26317c9 1438 ret = deflateInit(&stream, Z_DEFAULT_COMPRESSION);
b26317c9 1439 if (ret != Z_OK) {
aee2ab67 1440 log_err("fio: failed to init gz stream\n");
152ea035 1441 goto err;
aee2ab67
JA
1442 }
1443
1444 seq = ++data->log->chunk_seq;
b26317c9 1445
aee2ab67
JA
1446 stream.next_in = (void *) data->samples;
1447 stream.avail_in = data->nr_samples * log_entry_sz(data->log);
1448
d0d0cf0a
JA
1449 dprint(FD_COMPRESS, "deflate input size=%lu, seq=%u, log=%s\n",
1450 (unsigned long) stream.avail_in, seq,
1451 data->log->filename);
aee2ab67 1452 do {
040238ec 1453 if (c)
0caccfa7
JA
1454 dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq,
1455 (unsigned long) c->len);
aee2ab67
JA
1456 c = get_new_chunk(seq);
1457 stream.avail_out = GZ_CHUNK;
1458 stream.next_out = c->buf;
1459 ret = deflate(&stream, Z_NO_FLUSH);
1460 if (ret < 0) {
1461 log_err("fio: deflate log (%d)\n", ret);
dad95da1
JA
1462 free_chunk(c);
1463 goto err;
aee2ab67
JA
1464 }
1465
1466 c->len = GZ_CHUNK - stream.avail_out;
1467 flist_add_tail(&c->list, &list);
1468 total += c->len;
1469 } while (stream.avail_in);
1470
1471 stream.next_out = c->buf + c->len;
1472 stream.avail_out = GZ_CHUNK - c->len;
1473
1474 ret = deflate(&stream, Z_FINISH);
6fa3ad51
JA
1475 if (ret < 0) {
1476 /*
1477 * Z_BUF_ERROR is special, it just means we need more
1478 * output space. We'll handle that below. Treat any other
1479 * error as fatal.
1480 */
1481 if (ret != Z_BUF_ERROR) {
1482 log_err("fio: deflate log (%d)\n", ret);
1483 flist_del(&c->list);
1484 free_chunk(c);
1485 goto err;
1486 }
1487 }
1488
1489 total -= c->len;
1490 c->len = GZ_CHUNK - stream.avail_out;
1491 total += c->len;
0caccfa7 1492 dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, (unsigned long) c->len);
040238ec 1493
6fa3ad51 1494 if (ret != Z_STREAM_END) {
aee2ab67
JA
1495 do {
1496 c = get_new_chunk(seq);
1497 stream.avail_out = GZ_CHUNK;
1498 stream.next_out = c->buf;
1499 ret = deflate(&stream, Z_FINISH);
1500 c->len = GZ_CHUNK - stream.avail_out;
0c56718d 1501 total += c->len;
aee2ab67 1502 flist_add_tail(&c->list, &list);
0caccfa7
JA
1503 dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq,
1504 (unsigned long) c->len);
aee2ab67
JA
1505 } while (ret != Z_STREAM_END);
1506 }
1507
0c56718d
JA
1508 dprint(FD_COMPRESS, "deflated to size=%lu\n", (unsigned long) total);
1509
aee2ab67
JA
1510 ret = deflateEnd(&stream);
1511 if (ret != Z_OK)
1512 log_err("fio: deflateEnd %d\n", ret);
1513
858bce70 1514 iolog_put_deferred(data->log, data->samples);
aee2ab67
JA
1515
1516 if (!flist_empty(&list)) {
1517 pthread_mutex_lock(&data->log->chunk_lock);
1518 flist_splice_tail(&list, &data->log->chunk_list);
1519 pthread_mutex_unlock(&data->log->chunk_lock);
1520 }
1521
dad95da1
JA
1522 ret = 0;
1523done:
7e419452 1524 if (data->free)
dab41b1c 1525 sfree(data);
dad95da1
JA
1526 return ret;
1527err:
1528 while (!flist_empty(&list)) {
1529 c = flist_first_entry(list.next, struct iolog_compress, list);
1530 flist_del(&c->list);
1531 free_chunk(c);
1532 }
1533 ret = 1;
1534 goto done;
aee2ab67
JA
1535}
1536
78e93aa6
JA
1537/*
1538 * Invoked from our compress helper thread, when logging would have exceeded
1539 * the specified memory limitation. Compresses the previously stored
1540 * entries.
1541 */
1542static int gz_work_async(struct submit_worker *sw, struct workqueue_work *work)
1543{
1544 return gz_work(container_of(work, struct iolog_flush_data, work));
1545}
1546
c08f9fe2
JA
1547static int gz_init_worker(struct submit_worker *sw)
1548{
1549 struct thread_data *td = sw->wq->td;
1550
1551 if (!fio_option_is_set(&td->o, log_gz_cpumask))
1552 return 0;
1553
1554 if (fio_setaffinity(gettid(), td->o.log_gz_cpumask) == -1) {
1555 log_err("gz: failed to set CPU affinity\n");
1556 return 1;
1557 }
1558
1559 return 0;
1560}
1561
155f2f02 1562static struct workqueue_ops log_compress_wq_ops = {
78e93aa6 1563 .fn = gz_work_async,
c08f9fe2 1564 .init_worker_fn = gz_init_worker,
24660963 1565 .nice = 1,
155f2f02
JA
1566};
1567
24660963 1568int iolog_compress_init(struct thread_data *td, struct sk_out *sk_out)
155f2f02
JA
1569{
1570 if (!(td->flags & TD_F_COMPRESS_LOG))
1571 return 0;
1572
24660963 1573 workqueue_init(td, &td->log_compress_wq, &log_compress_wq_ops, 1, sk_out);
155f2f02
JA
1574 return 0;
1575}
1576
1577void iolog_compress_exit(struct thread_data *td)
1578{
1579 if (!(td->flags & TD_F_COMPRESS_LOG))
1580 return;
1581
1582 workqueue_exit(&td->log_compress_wq);
1583}
1584
1a8e7458 1585/*
949ae6dc
JA
1586 * Queue work item to compress the existing log entries. We reset the
1587 * current log to a small size, and reference the existing log in the
1588 * data that we queue for compression. Once compression has been done,
03900b0b 1589 * this old log is freed. Will not return until the log compression
1590 * has completed, and will flush all previous logs too
1a8e7458 1591 */
7e419452 1592static int iolog_flush(struct io_log *log)
aee2ab67 1593{
aee2ab67 1594 struct iolog_flush_data *data;
aee2ab67 1595
03900b0b 1596 workqueue_flush(&log->td->log_compress_wq);
aee2ab67
JA
1597 data = malloc(sizeof(*data));
1598 if (!data)
1599 return 1;
1600
1601 data->log = log;
7e419452 1602 data->free = false;
aee2ab67 1603
7e419452
JA
1604 while (!flist_empty(&log->io_logs)) {
1605 struct io_logs *cur_log;
949ae6dc 1606
7e419452
JA
1607 cur_log = flist_first_entry(&log->io_logs, struct io_logs, list);
1608 flist_del_init(&cur_log->list);
1609
1610 data->samples = cur_log->log;
1611 data->nr_samples = cur_log->nr_samples;
1612
2b7625e2 1613 sfree(cur_log);
aee2ab67 1614
78e93aa6 1615 gz_work(data);
7e419452 1616 }
aee2ab67 1617
7e419452 1618 free(data);
aee2ab67
JA
1619 return 0;
1620}
1621
7e419452
JA
1622int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log)
1623{
1624 struct iolog_flush_data *data;
1625
dab41b1c 1626 data = smalloc(sizeof(*data));
7e419452
JA
1627 if (!data)
1628 return 1;
1629
1630 data->log = log;
1631
1632 data->samples = cur_log->log;
1633 data->nr_samples = cur_log->nr_samples;
1634 data->free = true;
1635
1636 cur_log->nr_samples = cur_log->max_samples = 0;
1637 cur_log->log = NULL;
1638
1639 workqueue_enqueue(&log->td->log_compress_wq, &data->work);
858bce70
JA
1640
1641 iolog_free_deferred(log);
1642
7e419452
JA
1643 return 0;
1644}
aee2ab67
JA
1645#else
1646
7e419452
JA
1647static int iolog_flush(struct io_log *log)
1648{
1649 return 1;
1650}
1651
1652int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log)
aee2ab67
JA
1653{
1654 return 1;
1655}
1656
24660963 1657int iolog_compress_init(struct thread_data *td, struct sk_out *sk_out)
52618a2a
JA
1658{
1659 return 0;
1660}
1661
1662void iolog_compress_exit(struct thread_data *td)
1663{
1664}
1665
aee2ab67
JA
1666#endif
1667
7e419452
JA
1668struct io_logs *iolog_cur_log(struct io_log *log)
1669{
1670 if (flist_empty(&log->io_logs))
1671 return NULL;
1672
1673 return flist_last_entry(&log->io_logs, struct io_logs, list);
1674}
1675
1676uint64_t iolog_nr_samples(struct io_log *iolog)
1677{
1678 struct flist_head *entry;
1679 uint64_t ret = 0;
1680
1681 flist_for_each(entry, &iolog->io_logs) {
1682 struct io_logs *cur_log;
1683
1684 cur_log = flist_entry(entry, struct io_logs, list);
1685 ret += cur_log->nr_samples;
1686 }
1687
1688 return ret;
1689}
1690
1c725e70 1691static int __write_log(struct thread_data *td, struct io_log *log, int try)
905e3d4f 1692{
1c725e70
JA
1693 if (log)
1694 return finish_log(td, log, try);
905e3d4f 1695
1c725e70
JA
1696 return 0;
1697}
905e3d4f 1698
a47591e4 1699static int write_iops_log(struct thread_data *td, int try, bool unit_log)
1c725e70 1700{
a47591e4
JA
1701 int ret;
1702
1703 if (per_unit_log(td->iops_log) != unit_log)
1704 return 0;
1705
1706 ret = __write_log(td, td->iops_log, try);
1707 if (!ret)
1708 td->iops_log = NULL;
1709
1710 return ret;
905e3d4f
JA
1711}
1712
a47591e4 1713static int write_slat_log(struct thread_data *td, int try, bool unit_log)
905e3d4f 1714{
a47591e4
JA
1715 int ret;
1716
1717 if (!unit_log)
1718 return 0;
1719
1720 ret = __write_log(td, td->slat_log, try);
1721 if (!ret)
1722 td->slat_log = NULL;
1723
1724 return ret;
905e3d4f
JA
1725}
1726
a47591e4 1727static int write_clat_log(struct thread_data *td, int try, bool unit_log)
905e3d4f 1728{
a47591e4
JA
1729 int ret;
1730
1731 if (!unit_log)
1732 return 0;
1733
1734 ret = __write_log(td, td->clat_log, try);
1735 if (!ret)
1736 td->clat_log = NULL;
1737
1738 return ret;
905e3d4f
JA
1739}
1740
1e613c9c
KC
1741static int write_clat_hist_log(struct thread_data *td, int try, bool unit_log)
1742{
1743 int ret;
1744
1745 if (!unit_log)
1746 return 0;
1747
1748 ret = __write_log(td, td->clat_hist_log, try);
1749 if (!ret)
1750 td->clat_hist_log = NULL;
1751
1752 return ret;
1753}
1754
a47591e4 1755static int write_lat_log(struct thread_data *td, int try, bool unit_log)
905e3d4f 1756{
a47591e4
JA
1757 int ret;
1758
1759 if (!unit_log)
1760 return 0;
1761
1762 ret = __write_log(td, td->lat_log, try);
1763 if (!ret)
1764 td->lat_log = NULL;
1765
1766 return ret;
905e3d4f
JA
1767}
1768
a47591e4 1769static int write_bandw_log(struct thread_data *td, int try, bool unit_log)
905e3d4f 1770{
a47591e4
JA
1771 int ret;
1772
1773 if (per_unit_log(td->bw_log) != unit_log)
1774 return 0;
1775
1776 ret = __write_log(td, td->bw_log, try);
1777 if (!ret)
1778 td->bw_log = NULL;
1779
1780 return ret;
905e3d4f
JA
1781}
1782
1783enum {
1784 BW_LOG_MASK = 1,
1785 LAT_LOG_MASK = 2,
1786 SLAT_LOG_MASK = 4,
1787 CLAT_LOG_MASK = 8,
1788 IOPS_LOG_MASK = 16,
1e613c9c 1789 CLAT_HIST_LOG_MASK = 32,
905e3d4f 1790
1e613c9c 1791 ALL_LOG_NR = 6,
905e3d4f
JA
1792};
1793
1794struct log_type {
1795 unsigned int mask;
a47591e4 1796 int (*fn)(struct thread_data *, int, bool);
905e3d4f
JA
1797};
1798
1799static struct log_type log_types[] = {
1800 {
1801 .mask = BW_LOG_MASK,
1802 .fn = write_bandw_log,
1803 },
1804 {
1805 .mask = LAT_LOG_MASK,
1806 .fn = write_lat_log,
1807 },
1808 {
1809 .mask = SLAT_LOG_MASK,
1810 .fn = write_slat_log,
1811 },
1812 {
1813 .mask = CLAT_LOG_MASK,
1814 .fn = write_clat_log,
1815 },
1816 {
1817 .mask = IOPS_LOG_MASK,
1818 .fn = write_iops_log,
1819 },
1e613c9c
KC
1820 {
1821 .mask = CLAT_HIST_LOG_MASK,
1822 .fn = write_clat_hist_log,
1823 }
905e3d4f
JA
1824};
1825
a47591e4 1826void td_writeout_logs(struct thread_data *td, bool unit_logs)
905e3d4f 1827{
ea5409f9 1828 unsigned int log_mask = 0;
905e3d4f
JA
1829 unsigned int log_left = ALL_LOG_NR;
1830 int old_state, i;
1831
1832 old_state = td_bump_runstate(td, TD_FINISHING);
1833
a47591e4 1834 finalize_logs(td, unit_logs);
905e3d4f
JA
1835
1836 while (log_left) {
1837 int prev_log_left = log_left;
1838
1839 for (i = 0; i < ALL_LOG_NR && log_left; i++) {
1840 struct log_type *lt = &log_types[i];
1841 int ret;
1842
ea5409f9 1843 if (!(log_mask & lt->mask)) {
a47591e4 1844 ret = lt->fn(td, log_left != 1, unit_logs);
905e3d4f
JA
1845 if (!ret) {
1846 log_left--;
ea5409f9 1847 log_mask |= lt->mask;
905e3d4f
JA
1848 }
1849 }
1850 }
1851
1852 if (prev_log_left == log_left)
1853 usleep(5000);
1854 }
1855
1856 td_restore_runstate(td, old_state);
1857}
a47591e4
JA
1858
1859void fio_writeout_logs(bool unit_logs)
1860{
1861 struct thread_data *td;
1862 int i;
1863
1864 for_each_td(td, i)
1865 td_writeout_logs(td, unit_logs);
1866}