Commit | Line | Data |
---|---|---|
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" |
ac9b9101 | 22 | |
2f8f4821 AK |
23 | #include <netinet/in.h> |
24 | #include <netinet/tcp.h> | |
25 | #include <arpa/inet.h> | |
26 | #include <sys/stat.h> | |
27 | #include <sys/socket.h> | |
28 | #include <sys/un.h> | |
29 | ||
7e419452 JA |
30 | static int iolog_flush(struct io_log *log); |
31 | ||
ac9b9101 JA |
32 | static const char iolog_ver2[] = "fio version 2 iolog"; |
33 | ||
34 | void queue_io_piece(struct thread_data *td, struct io_piece *ipo) | |
35 | { | |
36 | flist_add_tail(&ipo->list, &td->io_log_list); | |
37 | td->total_io_size += ipo->len; | |
38 | } | |
39 | ||
1f440ece | 40 | void log_io_u(const struct thread_data *td, const struct io_u *io_u) |
ac9b9101 | 41 | { |
ac9b9101 JA |
42 | if (!td->o.write_iolog_file) |
43 | return; | |
44 | ||
5fff9543 | 45 | fprintf(td->iolog_f, "%s %s %llu %llu\n", io_u->file->file_name, |
173375c2 JA |
46 | io_ddir_name(io_u->ddir), |
47 | io_u->offset, io_u->buflen); | |
ac9b9101 JA |
48 | } |
49 | ||
50 | void log_file(struct thread_data *td, struct fio_file *f, | |
51 | enum file_log_act what) | |
52 | { | |
53 | const char *act[] = { "add", "open", "close" }; | |
54 | ||
55 | assert(what < 3); | |
56 | ||
57 | if (!td->o.write_iolog_file) | |
58 | return; | |
59 | ||
60 | ||
61 | /* | |
62 | * this happens on the pre-open/close done before the job starts | |
63 | */ | |
64 | if (!td->iolog_f) | |
65 | return; | |
66 | ||
67 | fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]); | |
68 | } | |
69 | ||
70 | static void iolog_delay(struct thread_data *td, unsigned long delay) | |
71 | { | |
df8fb609 | 72 | uint64_t usec = utime_since_now(&td->last_issue); |
f4b49c55 | 73 | unsigned long orig_delay = delay; |
df8fb609 | 74 | uint64_t this_delay; |
8b6a404c | 75 | struct timespec ts; |
ac9b9101 | 76 | |
df8fb609 JA |
77 | if (delay < td->time_offset) { |
78 | td->time_offset = 0; | |
79 | return; | |
80 | } | |
81 | ||
82 | delay -= td->time_offset; | |
ac9b9101 JA |
83 | if (delay < usec) |
84 | return; | |
85 | ||
86 | delay -= usec; | |
87 | ||
8b6a404c | 88 | fio_gettime(&ts, NULL); |
30b18672 JA |
89 | while (delay && !td->terminate) { |
90 | this_delay = delay; | |
91 | if (this_delay > 500000) | |
92 | this_delay = 500000; | |
93 | ||
94 | usec_sleep(td, this_delay); | |
95 | delay -= this_delay; | |
96 | } | |
df8fb609 | 97 | |
8b6a404c | 98 | usec = utime_since_now(&ts); |
f4b49c55 JA |
99 | if (usec > orig_delay) |
100 | td->time_offset = usec - orig_delay; | |
df8fb609 JA |
101 | else |
102 | td->time_offset = 0; | |
ac9b9101 JA |
103 | } |
104 | ||
105 | static int ipo_special(struct thread_data *td, struct io_piece *ipo) | |
106 | { | |
107 | struct fio_file *f; | |
108 | int ret; | |
109 | ||
110 | /* | |
111 | * Not a special ipo | |
112 | */ | |
113 | if (ipo->ddir != DDIR_INVAL) | |
114 | return 0; | |
115 | ||
116 | f = td->files[ipo->fileno]; | |
117 | ||
118 | switch (ipo->file_action) { | |
119 | case FIO_LOG_OPEN_FILE: | |
96d663ad SW |
120 | if (td->o.replay_redirect && fio_file_open(f)) { |
121 | dprint(FD_FILE, "iolog: ignoring re-open of file %s\n", | |
122 | f->file_name); | |
123 | break; | |
124 | } | |
ac9b9101 JA |
125 | ret = td_io_open_file(td, f); |
126 | if (!ret) | |
127 | break; | |
128 | td_verror(td, ret, "iolog open file"); | |
129 | return -1; | |
130 | case FIO_LOG_CLOSE_FILE: | |
131 | td_io_close_file(td, f); | |
132 | break; | |
133 | case FIO_LOG_UNLINK_FILE: | |
38ef9c90 | 134 | td_io_unlink_file(td, f); |
ac9b9101 JA |
135 | break; |
136 | default: | |
137 | log_err("fio: bad file action %d\n", ipo->file_action); | |
138 | break; | |
139 | } | |
140 | ||
141 | return 1; | |
142 | } | |
143 | ||
144 | int read_iolog_get(struct thread_data *td, struct io_u *io_u) | |
145 | { | |
146 | struct io_piece *ipo; | |
147 | unsigned long elapsed; | |
3c3ed070 | 148 | |
ac9b9101 JA |
149 | while (!flist_empty(&td->io_log_list)) { |
150 | int ret; | |
151 | ||
9342d5f8 | 152 | ipo = flist_first_entry(&td->io_log_list, struct io_piece, list); |
ac9b9101 JA |
153 | flist_del(&ipo->list); |
154 | remove_trim_entry(td, ipo); | |
155 | ||
156 | ret = ipo_special(td, ipo); | |
157 | if (ret < 0) { | |
158 | free(ipo); | |
159 | break; | |
160 | } else if (ret > 0) { | |
161 | free(ipo); | |
162 | continue; | |
163 | } | |
164 | ||
165 | io_u->ddir = ipo->ddir; | |
166 | if (ipo->ddir != DDIR_WAIT) { | |
167 | io_u->offset = ipo->offset; | |
168 | io_u->buflen = ipo->len; | |
169 | io_u->file = td->files[ipo->fileno]; | |
170 | get_file(io_u->file); | |
5fff9543 | 171 | dprint(FD_IO, "iolog: get %llu/%llu/%s\n", io_u->offset, |
ac9b9101 JA |
172 | io_u->buflen, io_u->file->file_name); |
173 | if (ipo->delay) | |
174 | iolog_delay(td, ipo->delay); | |
175 | } else { | |
176 | elapsed = mtime_since_genesis(); | |
177 | if (ipo->delay > elapsed) | |
178 | usec_sleep(td, (ipo->delay - elapsed) * 1000); | |
ac9b9101 JA |
179 | } |
180 | ||
181 | free(ipo); | |
3c3ed070 | 182 | |
ac9b9101 JA |
183 | if (io_u->ddir != DDIR_WAIT) |
184 | return 0; | |
185 | } | |
186 | ||
187 | td->done = 1; | |
188 | return 1; | |
189 | } | |
190 | ||
191 | void prune_io_piece_log(struct thread_data *td) | |
192 | { | |
193 | struct io_piece *ipo; | |
e96c0125 | 194 | struct fio_rb_node *n; |
ac9b9101 JA |
195 | |
196 | while ((n = rb_first(&td->io_hist_tree)) != NULL) { | |
197 | ipo = rb_entry(n, struct io_piece, rb_node); | |
198 | rb_erase(n, &td->io_hist_tree); | |
199 | remove_trim_entry(td, ipo); | |
200 | td->io_hist_len--; | |
201 | free(ipo); | |
202 | } | |
203 | ||
204 | while (!flist_empty(&td->io_hist_list)) { | |
4e1020f8 | 205 | ipo = flist_first_entry(&td->io_hist_list, struct io_piece, list); |
ac9b9101 JA |
206 | flist_del(&ipo->list); |
207 | remove_trim_entry(td, ipo); | |
208 | td->io_hist_len--; | |
209 | free(ipo); | |
210 | } | |
211 | } | |
212 | ||
213 | /* | |
214 | * log a successful write, so we can unwind the log for verify | |
215 | */ | |
216 | void log_io_piece(struct thread_data *td, struct io_u *io_u) | |
217 | { | |
e96c0125 | 218 | struct fio_rb_node **p, *parent; |
ac9b9101 JA |
219 | struct io_piece *ipo, *__ipo; |
220 | ||
8812d7f2 | 221 | ipo = calloc(1, sizeof(struct io_piece)); |
ac9b9101 JA |
222 | init_ipo(ipo); |
223 | ipo->file = io_u->file; | |
224 | ipo->offset = io_u->offset; | |
225 | ipo->len = io_u->buflen; | |
da0a7bd2 | 226 | ipo->numberio = io_u->numberio; |
f9401285 JA |
227 | ipo->flags = IP_F_IN_FLIGHT; |
228 | ||
229 | io_u->ipo = ipo; | |
ac9b9101 JA |
230 | |
231 | if (io_u_should_trim(td, io_u)) { | |
232 | flist_add_tail(&ipo->trim_list, &td->trim_list); | |
233 | td->trim_entries++; | |
234 | } | |
235 | ||
236 | /* | |
e43f210a | 237 | * Only sort writes if we don't have a random map in which case we need |
4a06cb41 SW |
238 | * to check for duplicate blocks and drop the old one, which we rely on |
239 | * the rb insert/lookup for handling. | |
ac9b9101 | 240 | */ |
f31feaa2 | 241 | if (file_randommap(td, ipo->file)) { |
ac9b9101 JA |
242 | INIT_FLIST_HEAD(&ipo->list); |
243 | flist_add_tail(&ipo->list, &td->io_hist_list); | |
244 | ipo->flags |= IP_F_ONLIST; | |
245 | td->io_hist_len++; | |
246 | return; | |
247 | } | |
248 | ||
249 | RB_CLEAR_NODE(&ipo->rb_node); | |
250 | ||
251 | /* | |
252 | * Sort the entry into the verification list | |
253 | */ | |
254 | restart: | |
255 | p = &td->io_hist_tree.rb_node; | |
256 | parent = NULL; | |
257 | while (*p) { | |
3b1ff869 | 258 | int overlap = 0; |
ac9b9101 JA |
259 | parent = *p; |
260 | ||
261 | __ipo = rb_entry(parent, struct io_piece, rb_node); | |
262 | if (ipo->file < __ipo->file) | |
263 | p = &(*p)->rb_left; | |
264 | else if (ipo->file > __ipo->file) | |
265 | p = &(*p)->rb_right; | |
3b1ff869 | 266 | else if (ipo->offset < __ipo->offset) { |
ac9b9101 | 267 | p = &(*p)->rb_left; |
3b1ff869 JE |
268 | overlap = ipo->offset + ipo->len > __ipo->offset; |
269 | } | |
270 | else if (ipo->offset > __ipo->offset) { | |
ac9b9101 | 271 | p = &(*p)->rb_right; |
3b1ff869 JE |
272 | overlap = __ipo->offset + __ipo->len > ipo->offset; |
273 | } | |
274 | else | |
275 | overlap = 1; | |
276 | ||
277 | if (overlap) { | |
7abe6a27 | 278 | dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu\n", |
885ac623 JA |
279 | __ipo->offset, __ipo->len, |
280 | ipo->offset, ipo->len); | |
ac9b9101 JA |
281 | td->io_hist_len--; |
282 | rb_erase(parent, &td->io_hist_tree); | |
283 | remove_trim_entry(td, __ipo); | |
e8b74655 SW |
284 | if (!(__ipo->flags & IP_F_IN_FLIGHT)) |
285 | free(__ipo); | |
ac9b9101 JA |
286 | goto restart; |
287 | } | |
288 | } | |
289 | ||
290 | rb_link_node(&ipo->rb_node, parent, p); | |
291 | rb_insert_color(&ipo->rb_node, &td->io_hist_tree); | |
292 | ipo->flags |= IP_F_ONRB; | |
293 | td->io_hist_len++; | |
294 | } | |
295 | ||
890b6656 JA |
296 | void unlog_io_piece(struct thread_data *td, struct io_u *io_u) |
297 | { | |
298 | struct io_piece *ipo = io_u->ipo; | |
299 | ||
66347cfa DE |
300 | if (td->ts.nr_block_infos) { |
301 | uint32_t *info = io_u_block_info(td, io_u); | |
302 | if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) { | |
303 | if (io_u->ddir == DDIR_TRIM) | |
304 | *info = BLOCK_INFO_SET_STATE(*info, | |
305 | BLOCK_STATE_TRIM_FAILURE); | |
306 | else if (io_u->ddir == DDIR_WRITE) | |
307 | *info = BLOCK_INFO_SET_STATE(*info, | |
308 | BLOCK_STATE_WRITE_FAILURE); | |
309 | } | |
310 | } | |
311 | ||
890b6656 JA |
312 | if (!ipo) |
313 | return; | |
314 | ||
315 | if (ipo->flags & IP_F_ONRB) | |
316 | rb_erase(&ipo->rb_node, &td->io_hist_tree); | |
317 | else if (ipo->flags & IP_F_ONLIST) | |
318 | flist_del(&ipo->list); | |
319 | ||
320 | free(ipo); | |
321 | io_u->ipo = NULL; | |
322 | td->io_hist_len--; | |
323 | } | |
324 | ||
94a78d52 | 325 | void trim_io_piece(const struct io_u *io_u) |
890b6656 JA |
326 | { |
327 | struct io_piece *ipo = io_u->ipo; | |
328 | ||
329 | if (!ipo) | |
330 | return; | |
331 | ||
332 | ipo->len = io_u->xfer_buflen - io_u->resid; | |
333 | } | |
334 | ||
ac9b9101 JA |
335 | void write_iolog_close(struct thread_data *td) |
336 | { | |
337 | fflush(td->iolog_f); | |
338 | fclose(td->iolog_f); | |
339 | free(td->iolog_buf); | |
340 | td->iolog_f = NULL; | |
341 | td->iolog_buf = NULL; | |
342 | } | |
343 | ||
344 | /* | |
345 | * Read version 2 iolog data. It is enhanced to include per-file logging, | |
346 | * syncs, etc. | |
347 | */ | |
b153f94a | 348 | static bool read_iolog2(struct thread_data *td, FILE *f) |
ac9b9101 JA |
349 | { |
350 | unsigned long long offset; | |
351 | unsigned int bytes; | |
352 | int reads, writes, waits, fileno = 0, file_action = 0; /* stupid gcc */ | |
dac34079 | 353 | char *rfname, *fname, *act; |
ac9b9101 JA |
354 | char *str, *p; |
355 | enum fio_ddir rw; | |
356 | ||
357 | free_release_files(td); | |
358 | ||
359 | /* | |
360 | * Read in the read iolog and store it, reuse the infrastructure | |
361 | * for doing verifications. | |
362 | */ | |
363 | str = malloc(4096); | |
dac34079 | 364 | rfname = fname = malloc(256+16); |
ac9b9101 JA |
365 | act = malloc(256+16); |
366 | ||
367 | reads = writes = waits = 0; | |
368 | while ((p = fgets(str, 4096, f)) != NULL) { | |
369 | struct io_piece *ipo; | |
370 | int r; | |
371 | ||
dac34079 | 372 | r = sscanf(p, "%256s %256s %llu %u", rfname, act, &offset, |
ac9b9101 | 373 | &bytes); |
dac34079 JA |
374 | |
375 | if (td->o.replay_redirect) | |
376 | fname = td->o.replay_redirect; | |
377 | ||
ac9b9101 JA |
378 | if (r == 4) { |
379 | /* | |
380 | * Check action first | |
381 | */ | |
382 | if (!strcmp(act, "wait")) | |
383 | rw = DDIR_WAIT; | |
384 | else if (!strcmp(act, "read")) | |
385 | rw = DDIR_READ; | |
386 | else if (!strcmp(act, "write")) | |
387 | rw = DDIR_WRITE; | |
388 | else if (!strcmp(act, "sync")) | |
389 | rw = DDIR_SYNC; | |
390 | else if (!strcmp(act, "datasync")) | |
391 | rw = DDIR_DATASYNC; | |
392 | else if (!strcmp(act, "trim")) | |
393 | rw = DDIR_TRIM; | |
394 | else { | |
395 | log_err("fio: bad iolog file action: %s\n", | |
396 | act); | |
397 | continue; | |
398 | } | |
033ace1e | 399 | fileno = get_fileno(td, fname); |
ac9b9101 JA |
400 | } else if (r == 2) { |
401 | rw = DDIR_INVAL; | |
402 | if (!strcmp(act, "add")) { | |
96d663ad SW |
403 | if (td->o.replay_redirect && |
404 | get_fileno(td, fname) != -1) { | |
405 | dprint(FD_FILE, "iolog: ignoring" | |
406 | " re-add of file %s\n", fname); | |
407 | } else { | |
408 | fileno = add_file(td, fname, 0, 1); | |
409 | file_action = FIO_LOG_ADD_FILE; | |
410 | } | |
ac9b9101 JA |
411 | continue; |
412 | } else if (!strcmp(act, "open")) { | |
413 | fileno = get_fileno(td, fname); | |
414 | file_action = FIO_LOG_OPEN_FILE; | |
415 | } else if (!strcmp(act, "close")) { | |
416 | fileno = get_fileno(td, fname); | |
417 | file_action = FIO_LOG_CLOSE_FILE; | |
418 | } else { | |
419 | log_err("fio: bad iolog file action: %s\n", | |
420 | act); | |
421 | continue; | |
422 | } | |
423 | } else { | |
c5839011 | 424 | log_err("bad iolog2: %s\n", p); |
ac9b9101 JA |
425 | continue; |
426 | } | |
427 | ||
428 | if (rw == DDIR_READ) | |
429 | reads++; | |
430 | else if (rw == DDIR_WRITE) { | |
431 | /* | |
432 | * Don't add a write for ro mode | |
433 | */ | |
434 | if (read_only) | |
435 | continue; | |
436 | writes++; | |
437 | } else if (rw == DDIR_WAIT) { | |
74dff443 JA |
438 | if (td->o.no_stall) |
439 | continue; | |
ac9b9101 JA |
440 | waits++; |
441 | } else if (rw == DDIR_INVAL) { | |
442 | } else if (!ddir_sync(rw)) { | |
443 | log_err("bad ddir: %d\n", rw); | |
444 | continue; | |
445 | } | |
446 | ||
447 | /* | |
448 | * Make note of file | |
449 | */ | |
8812d7f2 | 450 | ipo = calloc(1, sizeof(*ipo)); |
ac9b9101 JA |
451 | init_ipo(ipo); |
452 | ipo->ddir = rw; | |
453 | if (rw == DDIR_WAIT) { | |
454 | ipo->delay = offset; | |
455 | } else { | |
a79f17bf SW |
456 | if (td->o.replay_scale) |
457 | ipo->offset = offset / td->o.replay_scale; | |
458 | else | |
459 | ipo->offset = offset; | |
460 | ipo_bytes_align(td->o.replay_align, ipo); | |
461 | ||
ac9b9101 | 462 | ipo->len = bytes; |
42793d94 | 463 | if (rw != DDIR_INVAL && bytes > td->o.max_bs[rw]) |
ac9b9101 JA |
464 | td->o.max_bs[rw] = bytes; |
465 | ipo->fileno = fileno; | |
466 | ipo->file_action = file_action; | |
70afff59 | 467 | td->o.size += bytes; |
ac9b9101 | 468 | } |
3c3ed070 | 469 | |
ac9b9101 JA |
470 | queue_io_piece(td, ipo); |
471 | } | |
472 | ||
473 | free(str); | |
474 | free(act); | |
dac34079 | 475 | free(rfname); |
ac9b9101 JA |
476 | |
477 | if (writes && read_only) { | |
478 | log_err("fio: <%s> skips replay of %d writes due to" | |
479 | " read-only\n", td->o.name, writes); | |
480 | writes = 0; | |
481 | } | |
482 | ||
483 | if (!reads && !writes && !waits) | |
b153f94a | 484 | return false; |
ac9b9101 JA |
485 | else if (reads && !writes) |
486 | td->o.td_ddir = TD_DDIR_READ; | |
487 | else if (!reads && writes) | |
488 | td->o.td_ddir = TD_DDIR_WRITE; | |
489 | else | |
490 | td->o.td_ddir = TD_DDIR_RW; | |
491 | ||
b153f94a | 492 | return true; |
ac9b9101 JA |
493 | } |
494 | ||
2f8f4821 AK |
495 | static bool is_socket(const char *path) |
496 | { | |
497 | struct stat buf; | |
498 | int r = stat(path, &buf); | |
499 | if (r == -1) | |
500 | return false; | |
501 | ||
502 | return S_ISSOCK(buf.st_mode); | |
503 | } | |
504 | ||
505 | static int open_socket(const char *path) | |
506 | { | |
507 | int fd = socket(AF_UNIX, SOCK_STREAM, 0); | |
508 | struct sockaddr_un addr; | |
509 | if (fd < 0) | |
510 | return fd; | |
511 | addr.sun_family = AF_UNIX; | |
512 | strncpy(addr.sun_path, path, sizeof(addr.sun_path)); | |
513 | if (connect(fd, (const struct sockaddr *)&addr, strlen(path) + sizeof(addr.sun_family)) == 0) | |
514 | return fd; | |
515 | else | |
516 | close(fd); | |
517 | return -1; | |
518 | } | |
519 | ||
ac9b9101 JA |
520 | /* |
521 | * open iolog, check version, and call appropriate parser | |
522 | */ | |
b153f94a | 523 | static bool init_iolog_read(struct thread_data *td) |
ac9b9101 JA |
524 | { |
525 | char buffer[256], *p; | |
2f8f4821 | 526 | FILE *f = NULL; |
b153f94a | 527 | bool ret; |
2f8f4821 AK |
528 | if (is_socket(td->o.read_iolog_file)) { |
529 | int fd = open_socket(td->o.read_iolog_file); | |
530 | if (fd >= 0) { | |
531 | f = fdopen(fd, "r"); | |
532 | } | |
533 | } else | |
534 | f = fopen(td->o.read_iolog_file, "r"); | |
ac9b9101 JA |
535 | if (!f) { |
536 | perror("fopen read iolog"); | |
b153f94a | 537 | return false; |
ac9b9101 JA |
538 | } |
539 | ||
540 | p = fgets(buffer, sizeof(buffer), f); | |
541 | if (!p) { | |
542 | td_verror(td, errno, "iolog read"); | |
543 | log_err("fio: unable to read iolog\n"); | |
544 | fclose(f); | |
b153f94a | 545 | return false; |
ac9b9101 JA |
546 | } |
547 | ||
548 | /* | |
549 | * version 2 of the iolog stores a specific string as the | |
550 | * first line, check for that | |
551 | */ | |
552 | if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2))) | |
553 | ret = read_iolog2(td, f); | |
554 | else { | |
555 | log_err("fio: iolog version 1 is no longer supported\n"); | |
b153f94a | 556 | ret = false; |
ac9b9101 JA |
557 | } |
558 | ||
559 | fclose(f); | |
560 | return ret; | |
561 | } | |
562 | ||
563 | /* | |
564 | * Set up a log for storing io patterns. | |
565 | */ | |
b153f94a | 566 | static bool init_iolog_write(struct thread_data *td) |
ac9b9101 JA |
567 | { |
568 | struct fio_file *ff; | |
569 | FILE *f; | |
570 | unsigned int i; | |
571 | ||
572 | f = fopen(td->o.write_iolog_file, "a"); | |
573 | if (!f) { | |
574 | perror("fopen write iolog"); | |
b153f94a | 575 | return false; |
ac9b9101 JA |
576 | } |
577 | ||
578 | /* | |
579 | * That's it for writing, setup a log buffer and we're done. | |
580 | */ | |
581 | td->iolog_f = f; | |
582 | td->iolog_buf = malloc(8192); | |
583 | setvbuf(f, td->iolog_buf, _IOFBF, 8192); | |
584 | ||
585 | /* | |
586 | * write our version line | |
587 | */ | |
588 | if (fprintf(f, "%s\n", iolog_ver2) < 0) { | |
589 | perror("iolog init\n"); | |
b153f94a | 590 | return false; |
ac9b9101 JA |
591 | } |
592 | ||
593 | /* | |
594 | * add all known files | |
595 | */ | |
596 | for_each_file(td, ff, i) | |
597 | log_file(td, ff, FIO_LOG_ADD_FILE); | |
598 | ||
b153f94a | 599 | return true; |
ac9b9101 JA |
600 | } |
601 | ||
b153f94a | 602 | bool init_iolog(struct thread_data *td) |
ac9b9101 | 603 | { |
b153f94a | 604 | bool ret; |
ac9b9101 JA |
605 | |
606 | if (td->o.read_iolog_file) { | |
d95b34a6 JA |
607 | int need_swap; |
608 | ||
ac9b9101 JA |
609 | /* |
610 | * Check if it's a blktrace file and load that if possible. | |
611 | * Otherwise assume it's a normal log file and load that. | |
612 | */ | |
d95b34a6 JA |
613 | if (is_blktrace(td->o.read_iolog_file, &need_swap)) |
614 | ret = load_blktrace(td, td->o.read_iolog_file, need_swap); | |
ac9b9101 JA |
615 | else |
616 | ret = init_iolog_read(td); | |
617 | } else if (td->o.write_iolog_file) | |
618 | ret = init_iolog_write(td); | |
b153f94a | 619 | else |
43864deb | 620 | ret = true; |
ac9b9101 | 621 | |
b153f94a | 622 | if (!ret) |
f01b34ae JA |
623 | td_verror(td, EINVAL, "failed initializing iolog"); |
624 | ||
ac9b9101 JA |
625 | return ret; |
626 | } | |
627 | ||
aee2ab67 JA |
628 | void setup_log(struct io_log **log, struct log_params *p, |
629 | const char *filename) | |
ac9b9101 | 630 | { |
e75cc8f3 | 631 | struct io_log *l; |
65a4d15c KC |
632 | int i; |
633 | struct io_u_plat_entry *entry; | |
634 | struct flist_head *list; | |
ac9b9101 | 635 | |
7e419452 JA |
636 | l = scalloc(1, sizeof(*l)); |
637 | INIT_FLIST_HEAD(&l->io_logs); | |
aee2ab67 JA |
638 | l->log_type = p->log_type; |
639 | l->log_offset = p->log_offset; | |
640 | l->log_gz = p->log_gz; | |
b26317c9 | 641 | l->log_gz_store = p->log_gz_store; |
aee2ab67 | 642 | l->avg_msec = p->avg_msec; |
1e613c9c KC |
643 | l->hist_msec = p->hist_msec; |
644 | l->hist_coarseness = p->hist_coarseness; | |
cb7e0ace | 645 | l->filename = strdup(filename); |
aee2ab67 JA |
646 | l->td = p->td; |
647 | ||
65a4d15c KC |
648 | /* Initialize histogram lists for each r/w direction, |
649 | * with initial io_u_plat of all zeros: | |
650 | */ | |
651 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { | |
d730bc58 | 652 | list = &l->hist_window[i].list; |
65a4d15c KC |
653 | INIT_FLIST_HEAD(list); |
654 | entry = calloc(1, sizeof(struct io_u_plat_entry)); | |
655 | flist_add(&entry->list, list); | |
656 | } | |
657 | ||
1fed2080 | 658 | if (l->td && l->td->o.io_submit_mode != IO_MODE_OFFLOAD) { |
a43f4461 | 659 | struct io_logs *__p; |
1fed2080 | 660 | |
a43f4461 JA |
661 | __p = calloc(1, sizeof(*l->pending)); |
662 | __p->max_samples = DEF_LOG_ENTRIES; | |
663 | __p->log = calloc(__p->max_samples, log_entry_sz(l)); | |
664 | l->pending = __p; | |
1fed2080 JA |
665 | } |
666 | ||
b26317c9 | 667 | if (l->log_offset) |
49e98daa | 668 | l->log_ddir_mask = LOG_OFFSET_SAMPLE_BIT; |
b26317c9 | 669 | |
aee2ab67 JA |
670 | INIT_FLIST_HEAD(&l->chunk_list); |
671 | ||
672 | if (l->log_gz && !p->td) | |
673 | l->log_gz = 0; | |
19417eda | 674 | else if (l->log_gz || l->log_gz_store) { |
34febb23 | 675 | mutex_init_pshared(&l->chunk_lock); |
858bce70 | 676 | mutex_init_pshared(&l->deferred_free_lock); |
aee2ab67 JA |
677 | p->td->flags |= TD_F_COMPRESS_LOG; |
678 | } | |
679 | ||
ac9b9101 JA |
680 | *log = l; |
681 | } | |
682 | ||
2e802282 JA |
683 | #ifdef CONFIG_SETVBUF |
684 | static void *set_file_buffer(FILE *f) | |
685 | { | |
686 | size_t size = 1048576; | |
687 | void *buf; | |
688 | ||
689 | buf = malloc(size); | |
690 | setvbuf(f, buf, _IOFBF, size); | |
691 | return buf; | |
692 | } | |
693 | ||
694 | static void clear_file_buffer(void *buf) | |
695 | { | |
696 | free(buf); | |
697 | } | |
698 | #else | |
699 | static void *set_file_buffer(FILE *f) | |
700 | { | |
701 | return NULL; | |
702 | } | |
703 | ||
704 | static void clear_file_buffer(void *buf) | |
705 | { | |
706 | } | |
707 | #endif | |
708 | ||
518dac09 | 709 | void free_log(struct io_log *log) |
cb7e0ace | 710 | { |
7e419452 JA |
711 | while (!flist_empty(&log->io_logs)) { |
712 | struct io_logs *cur_log; | |
713 | ||
714 | cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); | |
715 | flist_del_init(&cur_log->list); | |
716 | free(cur_log->log); | |
2b7625e2 | 717 | sfree(cur_log); |
7e419452 JA |
718 | } |
719 | ||
1fed2080 JA |
720 | if (log->pending) { |
721 | free(log->pending->log); | |
722 | free(log->pending); | |
723 | log->pending = NULL; | |
724 | } | |
725 | ||
726 | free(log->pending); | |
cb7e0ace | 727 | free(log->filename); |
a47591e4 | 728 | sfree(log); |
cb7e0ace JA |
729 | } |
730 | ||
6cc0e5aa AL |
731 | uint64_t hist_sum(int j, int stride, uint64_t *io_u_plat, |
732 | uint64_t *io_u_plat_last) | |
548df930 | 733 | { |
6cc0e5aa | 734 | uint64_t sum; |
f2c972cc | 735 | int k; |
548df930 | 736 | |
868f6f03 KC |
737 | if (io_u_plat_last) { |
738 | for (k = sum = 0; k < stride; k++) | |
739 | sum += io_u_plat[j + k] - io_u_plat_last[j + k]; | |
740 | } else { | |
741 | for (k = sum = 0; k < stride; k++) | |
742 | sum += io_u_plat[j + k]; | |
743 | } | |
548df930 | 744 | |
1e613c9c KC |
745 | return sum; |
746 | } | |
747 | ||
a89ba4b1 JA |
748 | static void flush_hist_samples(FILE *f, int hist_coarseness, void *samples, |
749 | uint64_t sample_size) | |
1e613c9c KC |
750 | { |
751 | struct io_sample *s; | |
752 | int log_offset; | |
753 | uint64_t i, j, nr_samples; | |
65a4d15c | 754 | struct io_u_plat_entry *entry, *entry_before; |
6cc0e5aa AL |
755 | uint64_t *io_u_plat; |
756 | uint64_t *io_u_plat_before; | |
1e613c9c KC |
757 | |
758 | int stride = 1 << hist_coarseness; | |
759 | ||
760 | if (!sample_size) | |
761 | return; | |
762 | ||
763 | s = __get_sample(samples, 0, 0); | |
764 | log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0; | |
765 | ||
766 | nr_samples = sample_size / __log_entry_sz(log_offset); | |
767 | ||
768 | for (i = 0; i < nr_samples; i++) { | |
769 | s = __get_sample(samples, log_offset, i); | |
d730bc58 | 770 | |
af2fde19 | 771 | entry = s->data.plat_entry; |
65a4d15c | 772 | io_u_plat = entry->io_u_plat; |
d730bc58 | 773 | |
65a4d15c KC |
774 | entry_before = flist_first_entry(&entry->list, struct io_u_plat_entry, list); |
775 | io_u_plat_before = entry_before->io_u_plat; | |
d730bc58 | 776 | |
5fff9543 JF |
777 | fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time, |
778 | io_sample_ddir(s), (unsigned long long) s->bs); | |
1e613c9c | 779 | for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) { |
6cc0e5aa AL |
780 | fprintf(f, "%llu, ", (unsigned long long) |
781 | hist_sum(j, stride, io_u_plat, io_u_plat_before)); | |
1e613c9c | 782 | } |
6cc0e5aa | 783 | fprintf(f, "%llu\n", (unsigned long long) |
65a4d15c | 784 | hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, |
d730bc58 JA |
785 | io_u_plat_before)); |
786 | ||
65a4d15c KC |
787 | flist_del(&entry_before->list); |
788 | free(entry_before); | |
1e613c9c KC |
789 | } |
790 | } | |
791 | ||
bead01d7 | 792 | void flush_samples(FILE *f, void *samples, uint64_t sample_size) |
ac9b9101 | 793 | { |
b26317c9 JA |
794 | struct io_sample *s; |
795 | int log_offset; | |
796 | uint64_t i, nr_samples; | |
797 | ||
798 | if (!sample_size) | |
799 | return; | |
800 | ||
801 | s = __get_sample(samples, 0, 0); | |
49e98daa | 802 | log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0; |
b26317c9 JA |
803 | |
804 | nr_samples = sample_size / __log_entry_sz(log_offset); | |
ac9b9101 | 805 | |
aee2ab67 | 806 | for (i = 0; i < nr_samples; i++) { |
b26317c9 | 807 | s = __get_sample(samples, log_offset, i); |
ac9b9101 | 808 | |
aee2ab67 | 809 | if (!log_offset) { |
5fff9543 | 810 | fprintf(f, "%lu, %" PRId64 ", %u, %llu\n", |
ae588852 | 811 | (unsigned long) s->time, |
af2fde19 | 812 | s->data.val, |
5fff9543 | 813 | io_sample_ddir(s), (unsigned long long) s->bs); |
ae588852 JA |
814 | } else { |
815 | struct io_sample_offset *so = (void *) s; | |
816 | ||
5fff9543 | 817 | fprintf(f, "%lu, %" PRId64 ", %u, %llu, %llu\n", |
ae588852 | 818 | (unsigned long) s->time, |
af2fde19 | 819 | s->data.val, |
5fff9543 | 820 | io_sample_ddir(s), (unsigned long long) s->bs, |
ae588852 JA |
821 | (unsigned long long) so->offset); |
822 | } | |
ac9b9101 | 823 | } |
aee2ab67 JA |
824 | } |
825 | ||
826 | #ifdef CONFIG_ZLIB | |
97eabb2a JA |
827 | |
828 | struct iolog_flush_data { | |
155f2f02 | 829 | struct workqueue_work work; |
97eabb2a JA |
830 | struct io_log *log; |
831 | void *samples; | |
7e419452 JA |
832 | uint32_t nr_samples; |
833 | bool free; | |
97eabb2a JA |
834 | }; |
835 | ||
97eabb2a JA |
836 | #define GZ_CHUNK 131072 |
837 | ||
838 | static struct iolog_compress *get_new_chunk(unsigned int seq) | |
839 | { | |
840 | struct iolog_compress *c; | |
841 | ||
842 | c = malloc(sizeof(*c)); | |
843 | INIT_FLIST_HEAD(&c->list); | |
844 | c->buf = malloc(GZ_CHUNK); | |
845 | c->len = 0; | |
846 | c->seq = seq; | |
97eabb2a JA |
847 | return c; |
848 | } | |
849 | ||
850 | static void free_chunk(struct iolog_compress *ic) | |
851 | { | |
c07826cd JA |
852 | free(ic->buf); |
853 | free(ic); | |
97eabb2a JA |
854 | } |
855 | ||
b26317c9 | 856 | static int z_stream_init(z_stream *stream, int gz_hdr) |
aee2ab67 | 857 | { |
b26317c9 JA |
858 | int wbits = 15; |
859 | ||
acb2a69b | 860 | memset(stream, 0, sizeof(*stream)); |
aee2ab67 JA |
861 | stream->zalloc = Z_NULL; |
862 | stream->zfree = Z_NULL; | |
863 | stream->opaque = Z_NULL; | |
864 | stream->next_in = Z_NULL; | |
865 | ||
1a8e7458 JA |
866 | /* |
867 | * zlib magic - add 32 for auto-detection of gz header or not, | |
868 | * if we decide to store files in a gzip friendly format. | |
869 | */ | |
b26317c9 JA |
870 | if (gz_hdr) |
871 | wbits += 32; | |
872 | ||
873 | if (inflateInit2(stream, wbits) != Z_OK) | |
aee2ab67 JA |
874 | return 1; |
875 | ||
876 | return 0; | |
877 | } | |
878 | ||
1a8e7458 | 879 | struct inflate_chunk_iter { |
aee2ab67 | 880 | unsigned int seq; |
dad95da1 | 881 | int err; |
aee2ab67 JA |
882 | void *buf; |
883 | size_t buf_size; | |
884 | size_t buf_used; | |
885 | size_t chunk_sz; | |
886 | }; | |
887 | ||
b26317c9 | 888 | static void finish_chunk(z_stream *stream, FILE *f, |
1a8e7458 | 889 | struct inflate_chunk_iter *iter) |
aee2ab67 | 890 | { |
aee2ab67 JA |
891 | int ret; |
892 | ||
893 | ret = inflateEnd(stream); | |
894 | if (ret != Z_OK) | |
50f940f1 JA |
895 | log_err("fio: failed to end log inflation seq %d (%d)\n", |
896 | iter->seq, ret); | |
aee2ab67 | 897 | |
b26317c9 | 898 | flush_samples(f, iter->buf, iter->buf_used); |
aee2ab67 JA |
899 | free(iter->buf); |
900 | iter->buf = NULL; | |
901 | iter->buf_size = iter->buf_used = 0; | |
902 | } | |
903 | ||
1a8e7458 JA |
904 | /* |
905 | * Iterative chunk inflation. Handles cases where we cross into a new | |
906 | * sequence, doing flush finish of previous chunk if needed. | |
907 | */ | |
908 | static size_t inflate_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f, | |
909 | z_stream *stream, struct inflate_chunk_iter *iter) | |
aee2ab67 | 910 | { |
0c56718d JA |
911 | size_t ret; |
912 | ||
0f989d2e | 913 | dprint(FD_COMPRESS, "inflate chunk size=%lu, seq=%u\n", |
0c56718d JA |
914 | (unsigned long) ic->len, ic->seq); |
915 | ||
aee2ab67 JA |
916 | if (ic->seq != iter->seq) { |
917 | if (iter->seq) | |
b26317c9 | 918 | finish_chunk(stream, f, iter); |
aee2ab67 | 919 | |
b26317c9 | 920 | z_stream_init(stream, gz_hdr); |
aee2ab67 JA |
921 | iter->seq = ic->seq; |
922 | } | |
923 | ||
924 | stream->avail_in = ic->len; | |
925 | stream->next_in = ic->buf; | |
926 | ||
927 | if (!iter->buf_size) { | |
928 | iter->buf_size = iter->chunk_sz; | |
929 | iter->buf = malloc(iter->buf_size); | |
930 | } | |
931 | ||
932 | while (stream->avail_in) { | |
b26317c9 | 933 | size_t this_out = iter->buf_size - iter->buf_used; |
aee2ab67 JA |
934 | int err; |
935 | ||
b26317c9 | 936 | stream->avail_out = this_out; |
aee2ab67 JA |
937 | stream->next_out = iter->buf + iter->buf_used; |
938 | ||
939 | err = inflate(stream, Z_NO_FLUSH); | |
940 | if (err < 0) { | |
941 | log_err("fio: failed inflating log: %d\n", err); | |
dad95da1 | 942 | iter->err = err; |
aee2ab67 JA |
943 | break; |
944 | } | |
945 | ||
b26317c9 JA |
946 | iter->buf_used += this_out - stream->avail_out; |
947 | ||
948 | if (!stream->avail_out) { | |
949 | iter->buf_size += iter->chunk_sz; | |
950 | iter->buf = realloc(iter->buf, iter->buf_size); | |
951 | continue; | |
952 | } | |
953 | ||
954 | if (err == Z_STREAM_END) | |
955 | break; | |
aee2ab67 JA |
956 | } |
957 | ||
0c56718d JA |
958 | ret = (void *) stream->next_in - ic->buf; |
959 | ||
d0d0cf0a | 960 | dprint(FD_COMPRESS, "inflated to size=%lu\n", (unsigned long) iter->buf_size); |
0c56718d JA |
961 | |
962 | return ret; | |
aee2ab67 JA |
963 | } |
964 | ||
1a8e7458 JA |
965 | /* |
966 | * Inflate stored compressed chunks, or write them directly to the log | |
967 | * file if so instructed. | |
968 | */ | |
dad95da1 | 969 | static int inflate_gz_chunks(struct io_log *log, FILE *f) |
aee2ab67 | 970 | { |
1a8e7458 | 971 | struct inflate_chunk_iter iter = { .chunk_sz = log->log_gz, }; |
aee2ab67 JA |
972 | z_stream stream; |
973 | ||
974 | while (!flist_empty(&log->chunk_list)) { | |
975 | struct iolog_compress *ic; | |
976 | ||
9342d5f8 | 977 | ic = flist_first_entry(&log->chunk_list, struct iolog_compress, list); |
aee2ab67 | 978 | flist_del(&ic->list); |
b26317c9 | 979 | |
1a8e7458 JA |
980 | if (log->log_gz_store) { |
981 | size_t ret; | |
982 | ||
0c56718d JA |
983 | dprint(FD_COMPRESS, "log write chunk size=%lu, " |
984 | "seq=%u\n", (unsigned long) ic->len, ic->seq); | |
985 | ||
1a8e7458 | 986 | ret = fwrite(ic->buf, ic->len, 1, f); |
dad95da1 JA |
987 | if (ret != 1 || ferror(f)) { |
988 | iter.err = errno; | |
1a8e7458 | 989 | log_err("fio: error writing compressed log\n"); |
dad95da1 | 990 | } |
1a8e7458 JA |
991 | } else |
992 | inflate_chunk(ic, log->log_gz_store, f, &stream, &iter); | |
c07826cd JA |
993 | |
994 | free_chunk(ic); | |
aee2ab67 JA |
995 | } |
996 | ||
997 | if (iter.seq) { | |
b26317c9 | 998 | finish_chunk(&stream, f, &iter); |
aee2ab67 JA |
999 | free(iter.buf); |
1000 | } | |
dad95da1 JA |
1001 | |
1002 | return iter.err; | |
aee2ab67 JA |
1003 | } |
1004 | ||
1a8e7458 JA |
1005 | /* |
1006 | * Open compressed log file and decompress the stored chunks and | |
1007 | * write them to stdout. The chunks are stored sequentially in the | |
1008 | * file, so we iterate over them and do them one-by-one. | |
1009 | */ | |
b26317c9 JA |
1010 | int iolog_file_inflate(const char *file) |
1011 | { | |
1a8e7458 | 1012 | struct inflate_chunk_iter iter = { .chunk_sz = 64 * 1024 * 1024, }; |
b26317c9 JA |
1013 | struct iolog_compress ic; |
1014 | z_stream stream; | |
1015 | struct stat sb; | |
2b8b8f0d | 1016 | size_t ret; |
f302710c JA |
1017 | size_t total; |
1018 | void *buf; | |
b26317c9 JA |
1019 | FILE *f; |
1020 | ||
b26317c9 JA |
1021 | f = fopen(file, "r"); |
1022 | if (!f) { | |
1023 | perror("fopen"); | |
1024 | return 1; | |
1025 | } | |
1026 | ||
fd771971 JA |
1027 | if (stat(file, &sb) < 0) { |
1028 | fclose(f); | |
1029 | perror("stat"); | |
1030 | return 1; | |
1031 | } | |
1032 | ||
f302710c | 1033 | ic.buf = buf = malloc(sb.st_size); |
b26317c9 | 1034 | ic.len = sb.st_size; |
b26317c9 JA |
1035 | ic.seq = 1; |
1036 | ||
1037 | ret = fread(ic.buf, ic.len, 1, f); | |
2b8b8f0d | 1038 | if (ret == 0 && ferror(f)) { |
b26317c9 JA |
1039 | perror("fread"); |
1040 | fclose(f); | |
bb1116fe | 1041 | free(buf); |
b26317c9 | 1042 | return 1; |
2b8b8f0d | 1043 | } else if (ferror(f) || (!feof(f) && ret != 1)) { |
b26317c9 JA |
1044 | log_err("fio: short read on reading log\n"); |
1045 | fclose(f); | |
bb1116fe | 1046 | free(buf); |
b26317c9 JA |
1047 | return 1; |
1048 | } | |
1049 | ||
1050 | fclose(f); | |
1051 | ||
1a8e7458 JA |
1052 | /* |
1053 | * Each chunk will return Z_STREAM_END. We don't know how many | |
1054 | * chunks are in the file, so we just keep looping and incrementing | |
1055 | * the sequence number until we have consumed the whole compressed | |
1056 | * file. | |
1057 | */ | |
f302710c JA |
1058 | total = ic.len; |
1059 | do { | |
8a1db9a1 | 1060 | size_t iret; |
f302710c | 1061 | |
8a1db9a1 JA |
1062 | iret = inflate_chunk(&ic, 1, stdout, &stream, &iter); |
1063 | total -= iret; | |
f302710c JA |
1064 | if (!total) |
1065 | break; | |
dad95da1 JA |
1066 | if (iter.err) |
1067 | break; | |
f302710c JA |
1068 | |
1069 | ic.seq++; | |
8a1db9a1 JA |
1070 | ic.len -= iret; |
1071 | ic.buf += iret; | |
f302710c | 1072 | } while (1); |
b26317c9 JA |
1073 | |
1074 | if (iter.seq) { | |
1075 | finish_chunk(&stream, stdout, &iter); | |
1076 | free(iter.buf); | |
1077 | } | |
1078 | ||
f302710c | 1079 | free(buf); |
dad95da1 | 1080 | return iter.err; |
b26317c9 JA |
1081 | } |
1082 | ||
aee2ab67 JA |
1083 | #else |
1084 | ||
dad95da1 | 1085 | static int inflate_gz_chunks(struct io_log *log, FILE *f) |
aee2ab67 | 1086 | { |
b7364e21 | 1087 | return 0; |
aee2ab67 JA |
1088 | } |
1089 | ||
a68c66b2 JA |
1090 | int iolog_file_inflate(const char *file) |
1091 | { | |
1092 | log_err("fio: log inflation not possible without zlib\n"); | |
1093 | return 1; | |
1094 | } | |
1095 | ||
aee2ab67 JA |
1096 | #endif |
1097 | ||
60a25727 | 1098 | void flush_log(struct io_log *log, bool do_append) |
aee2ab67 JA |
1099 | { |
1100 | void *buf; | |
1101 | FILE *f; | |
1102 | ||
3a5db920 JA |
1103 | if (!do_append) |
1104 | f = fopen(log->filename, "w"); | |
1105 | else | |
1106 | f = fopen(log->filename, "a"); | |
aee2ab67 JA |
1107 | if (!f) { |
1108 | perror("fopen log"); | |
1109 | return; | |
1110 | } | |
1111 | ||
1112 | buf = set_file_buffer(f); | |
1113 | ||
1a8e7458 | 1114 | inflate_gz_chunks(log, f); |
aee2ab67 | 1115 | |
7e419452 JA |
1116 | while (!flist_empty(&log->io_logs)) { |
1117 | struct io_logs *cur_log; | |
1118 | ||
1119 | cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); | |
1120 | flist_del_init(&cur_log->list); | |
1e613c9c | 1121 | |
a9179eeb | 1122 | if (log->td && log == log->td->clat_hist_log) |
1e613c9c | 1123 | flush_hist_samples(f, log->hist_coarseness, cur_log->log, |
868f6f03 | 1124 | log_sample_sz(log, cur_log)); |
1e613c9c | 1125 | else |
868f6f03 | 1126 | flush_samples(f, cur_log->log, log_sample_sz(log, cur_log)); |
1e613c9c | 1127 | |
90d2d53f | 1128 | sfree(cur_log); |
7e419452 | 1129 | } |
ac9b9101 JA |
1130 | |
1131 | fclose(f); | |
2e802282 | 1132 | clear_file_buffer(buf); |
ac9b9101 JA |
1133 | } |
1134 | ||
cb7e0ace | 1135 | static int finish_log(struct thread_data *td, struct io_log *log, int trylock) |
ac9b9101 | 1136 | { |
155f2f02 | 1137 | if (td->flags & TD_F_COMPRESS_LOG) |
7e419452 | 1138 | iolog_flush(log); |
aee2ab67 | 1139 | |
243bfe19 | 1140 | if (trylock) { |
cb7e0ace | 1141 | if (fio_trylock_file(log->filename)) |
243bfe19 JA |
1142 | return 1; |
1143 | } else | |
cb7e0ace | 1144 | fio_lock_file(log->filename); |
243bfe19 | 1145 | |
0cba0f91 | 1146 | if (td->client_type == FIO_CLIENT_TYPE_GUI || is_backend) |
cb7e0ace | 1147 | fio_send_iolog(td, log, log->filename); |
aee2ab67 | 1148 | else |
3a5db920 | 1149 | flush_log(log, !td->o.per_job_logs); |
243bfe19 | 1150 | |
cb7e0ace | 1151 | fio_unlock_file(log->filename); |
518dac09 | 1152 | free_log(log); |
243bfe19 | 1153 | return 0; |
ac9b9101 JA |
1154 | } |
1155 | ||
0cba0f91 JA |
1156 | size_t log_chunk_sizes(struct io_log *log) |
1157 | { | |
1158 | struct flist_head *entry; | |
1159 | size_t ret; | |
1160 | ||
1161 | if (flist_empty(&log->chunk_list)) | |
1162 | return 0; | |
1163 | ||
1164 | ret = 0; | |
1165 | pthread_mutex_lock(&log->chunk_lock); | |
1166 | flist_for_each(entry, &log->chunk_list) { | |
1167 | struct iolog_compress *c; | |
1168 | ||
1169 | c = flist_entry(entry, struct iolog_compress, list); | |
1170 | ret += c->len; | |
1171 | } | |
1172 | pthread_mutex_unlock(&log->chunk_lock); | |
1173 | return ret; | |
1174 | } | |
1175 | ||
aee2ab67 JA |
1176 | #ifdef CONFIG_ZLIB |
1177 | ||
858bce70 JA |
1178 | static void iolog_put_deferred(struct io_log *log, void *ptr) |
1179 | { | |
1180 | if (!ptr) | |
1181 | return; | |
1182 | ||
1183 | pthread_mutex_lock(&log->deferred_free_lock); | |
1184 | if (log->deferred < IOLOG_MAX_DEFER) { | |
1185 | log->deferred_items[log->deferred] = ptr; | |
1186 | log->deferred++; | |
e2446166 | 1187 | } else if (!fio_did_warn(FIO_WARN_IOLOG_DROP)) |
858bce70 | 1188 | log_err("fio: had to drop log entry free\n"); |
858bce70 JA |
1189 | pthread_mutex_unlock(&log->deferred_free_lock); |
1190 | } | |
1191 | ||
1192 | static void iolog_free_deferred(struct io_log *log) | |
1193 | { | |
1194 | int i; | |
1195 | ||
1196 | if (!log->deferred) | |
1197 | return; | |
1198 | ||
1199 | pthread_mutex_lock(&log->deferred_free_lock); | |
1200 | ||
1201 | for (i = 0; i < log->deferred; i++) { | |
1202 | free(log->deferred_items[i]); | |
1203 | log->deferred_items[i] = NULL; | |
1204 | } | |
1205 | ||
1206 | log->deferred = 0; | |
1207 | pthread_mutex_unlock(&log->deferred_free_lock); | |
1208 | } | |
1209 | ||
78e93aa6 | 1210 | static int gz_work(struct iolog_flush_data *data) |
c143980a | 1211 | { |
040238ec | 1212 | struct iolog_compress *c = NULL; |
aee2ab67 JA |
1213 | struct flist_head list; |
1214 | unsigned int seq; | |
1215 | z_stream stream; | |
1216 | size_t total = 0; | |
d73ac887 | 1217 | int ret; |
aee2ab67 JA |
1218 | |
1219 | INIT_FLIST_HEAD(&list); | |
1220 | ||
7e419452 | 1221 | memset(&stream, 0, sizeof(stream)); |
aee2ab67 JA |
1222 | stream.zalloc = Z_NULL; |
1223 | stream.zfree = Z_NULL; | |
1224 | stream.opaque = Z_NULL; | |
1225 | ||
b26317c9 | 1226 | ret = deflateInit(&stream, Z_DEFAULT_COMPRESSION); |
b26317c9 | 1227 | if (ret != Z_OK) { |
aee2ab67 | 1228 | log_err("fio: failed to init gz stream\n"); |
152ea035 | 1229 | goto err; |
aee2ab67 JA |
1230 | } |
1231 | ||
1232 | seq = ++data->log->chunk_seq; | |
b26317c9 | 1233 | |
aee2ab67 JA |
1234 | stream.next_in = (void *) data->samples; |
1235 | stream.avail_in = data->nr_samples * log_entry_sz(data->log); | |
1236 | ||
d0d0cf0a JA |
1237 | dprint(FD_COMPRESS, "deflate input size=%lu, seq=%u, log=%s\n", |
1238 | (unsigned long) stream.avail_in, seq, | |
1239 | data->log->filename); | |
aee2ab67 | 1240 | do { |
040238ec | 1241 | if (c) |
0caccfa7 JA |
1242 | dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, |
1243 | (unsigned long) c->len); | |
aee2ab67 JA |
1244 | c = get_new_chunk(seq); |
1245 | stream.avail_out = GZ_CHUNK; | |
1246 | stream.next_out = c->buf; | |
1247 | ret = deflate(&stream, Z_NO_FLUSH); | |
1248 | if (ret < 0) { | |
1249 | log_err("fio: deflate log (%d)\n", ret); | |
dad95da1 JA |
1250 | free_chunk(c); |
1251 | goto err; | |
aee2ab67 JA |
1252 | } |
1253 | ||
1254 | c->len = GZ_CHUNK - stream.avail_out; | |
1255 | flist_add_tail(&c->list, &list); | |
1256 | total += c->len; | |
1257 | } while (stream.avail_in); | |
1258 | ||
1259 | stream.next_out = c->buf + c->len; | |
1260 | stream.avail_out = GZ_CHUNK - c->len; | |
1261 | ||
1262 | ret = deflate(&stream, Z_FINISH); | |
6fa3ad51 JA |
1263 | if (ret < 0) { |
1264 | /* | |
1265 | * Z_BUF_ERROR is special, it just means we need more | |
1266 | * output space. We'll handle that below. Treat any other | |
1267 | * error as fatal. | |
1268 | */ | |
1269 | if (ret != Z_BUF_ERROR) { | |
1270 | log_err("fio: deflate log (%d)\n", ret); | |
1271 | flist_del(&c->list); | |
1272 | free_chunk(c); | |
1273 | goto err; | |
1274 | } | |
1275 | } | |
1276 | ||
1277 | total -= c->len; | |
1278 | c->len = GZ_CHUNK - stream.avail_out; | |
1279 | total += c->len; | |
0caccfa7 | 1280 | dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, (unsigned long) c->len); |
040238ec | 1281 | |
6fa3ad51 | 1282 | if (ret != Z_STREAM_END) { |
aee2ab67 JA |
1283 | do { |
1284 | c = get_new_chunk(seq); | |
1285 | stream.avail_out = GZ_CHUNK; | |
1286 | stream.next_out = c->buf; | |
1287 | ret = deflate(&stream, Z_FINISH); | |
1288 | c->len = GZ_CHUNK - stream.avail_out; | |
0c56718d | 1289 | total += c->len; |
aee2ab67 | 1290 | flist_add_tail(&c->list, &list); |
0caccfa7 JA |
1291 | dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, |
1292 | (unsigned long) c->len); | |
aee2ab67 JA |
1293 | } while (ret != Z_STREAM_END); |
1294 | } | |
1295 | ||
0c56718d JA |
1296 | dprint(FD_COMPRESS, "deflated to size=%lu\n", (unsigned long) total); |
1297 | ||
aee2ab67 JA |
1298 | ret = deflateEnd(&stream); |
1299 | if (ret != Z_OK) | |
1300 | log_err("fio: deflateEnd %d\n", ret); | |
1301 | ||
858bce70 | 1302 | iolog_put_deferred(data->log, data->samples); |
aee2ab67 JA |
1303 | |
1304 | if (!flist_empty(&list)) { | |
1305 | pthread_mutex_lock(&data->log->chunk_lock); | |
1306 | flist_splice_tail(&list, &data->log->chunk_list); | |
1307 | pthread_mutex_unlock(&data->log->chunk_lock); | |
1308 | } | |
1309 | ||
dad95da1 JA |
1310 | ret = 0; |
1311 | done: | |
7e419452 | 1312 | if (data->free) |
dab41b1c | 1313 | sfree(data); |
dad95da1 JA |
1314 | return ret; |
1315 | err: | |
1316 | while (!flist_empty(&list)) { | |
1317 | c = flist_first_entry(list.next, struct iolog_compress, list); | |
1318 | flist_del(&c->list); | |
1319 | free_chunk(c); | |
1320 | } | |
1321 | ret = 1; | |
1322 | goto done; | |
aee2ab67 JA |
1323 | } |
1324 | ||
78e93aa6 JA |
1325 | /* |
1326 | * Invoked from our compress helper thread, when logging would have exceeded | |
1327 | * the specified memory limitation. Compresses the previously stored | |
1328 | * entries. | |
1329 | */ | |
1330 | static int gz_work_async(struct submit_worker *sw, struct workqueue_work *work) | |
1331 | { | |
1332 | return gz_work(container_of(work, struct iolog_flush_data, work)); | |
1333 | } | |
1334 | ||
c08f9fe2 JA |
1335 | static int gz_init_worker(struct submit_worker *sw) |
1336 | { | |
1337 | struct thread_data *td = sw->wq->td; | |
1338 | ||
1339 | if (!fio_option_is_set(&td->o, log_gz_cpumask)) | |
1340 | return 0; | |
1341 | ||
1342 | if (fio_setaffinity(gettid(), td->o.log_gz_cpumask) == -1) { | |
1343 | log_err("gz: failed to set CPU affinity\n"); | |
1344 | return 1; | |
1345 | } | |
1346 | ||
1347 | return 0; | |
1348 | } | |
1349 | ||
155f2f02 | 1350 | static struct workqueue_ops log_compress_wq_ops = { |
78e93aa6 | 1351 | .fn = gz_work_async, |
c08f9fe2 | 1352 | .init_worker_fn = gz_init_worker, |
24660963 | 1353 | .nice = 1, |
155f2f02 JA |
1354 | }; |
1355 | ||
24660963 | 1356 | int iolog_compress_init(struct thread_data *td, struct sk_out *sk_out) |
155f2f02 JA |
1357 | { |
1358 | if (!(td->flags & TD_F_COMPRESS_LOG)) | |
1359 | return 0; | |
1360 | ||
24660963 | 1361 | workqueue_init(td, &td->log_compress_wq, &log_compress_wq_ops, 1, sk_out); |
155f2f02 JA |
1362 | return 0; |
1363 | } | |
1364 | ||
1365 | void iolog_compress_exit(struct thread_data *td) | |
1366 | { | |
1367 | if (!(td->flags & TD_F_COMPRESS_LOG)) | |
1368 | return; | |
1369 | ||
1370 | workqueue_exit(&td->log_compress_wq); | |
1371 | } | |
1372 | ||
1a8e7458 | 1373 | /* |
949ae6dc JA |
1374 | * Queue work item to compress the existing log entries. We reset the |
1375 | * current log to a small size, and reference the existing log in the | |
1376 | * data that we queue for compression. Once compression has been done, | |
7e419452 JA |
1377 | * this old log is freed. If called with finish == true, will not return |
1378 | * until the log compression has completed, and will flush all previous | |
1379 | * logs too | |
1a8e7458 | 1380 | */ |
7e419452 | 1381 | static int iolog_flush(struct io_log *log) |
aee2ab67 | 1382 | { |
aee2ab67 | 1383 | struct iolog_flush_data *data; |
aee2ab67 JA |
1384 | |
1385 | data = malloc(sizeof(*data)); | |
1386 | if (!data) | |
1387 | return 1; | |
1388 | ||
1389 | data->log = log; | |
7e419452 | 1390 | data->free = false; |
aee2ab67 | 1391 | |
7e419452 JA |
1392 | while (!flist_empty(&log->io_logs)) { |
1393 | struct io_logs *cur_log; | |
949ae6dc | 1394 | |
7e419452 JA |
1395 | cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); |
1396 | flist_del_init(&cur_log->list); | |
1397 | ||
1398 | data->samples = cur_log->log; | |
1399 | data->nr_samples = cur_log->nr_samples; | |
1400 | ||
2b7625e2 | 1401 | sfree(cur_log); |
aee2ab67 | 1402 | |
78e93aa6 | 1403 | gz_work(data); |
7e419452 | 1404 | } |
aee2ab67 | 1405 | |
7e419452 | 1406 | free(data); |
aee2ab67 JA |
1407 | return 0; |
1408 | } | |
1409 | ||
7e419452 JA |
1410 | int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log) |
1411 | { | |
1412 | struct iolog_flush_data *data; | |
1413 | ||
dab41b1c | 1414 | data = smalloc(sizeof(*data)); |
7e419452 JA |
1415 | if (!data) |
1416 | return 1; | |
1417 | ||
1418 | data->log = log; | |
1419 | ||
1420 | data->samples = cur_log->log; | |
1421 | data->nr_samples = cur_log->nr_samples; | |
1422 | data->free = true; | |
1423 | ||
1424 | cur_log->nr_samples = cur_log->max_samples = 0; | |
1425 | cur_log->log = NULL; | |
1426 | ||
1427 | workqueue_enqueue(&log->td->log_compress_wq, &data->work); | |
858bce70 JA |
1428 | |
1429 | iolog_free_deferred(log); | |
1430 | ||
7e419452 JA |
1431 | return 0; |
1432 | } | |
aee2ab67 JA |
1433 | #else |
1434 | ||
7e419452 JA |
1435 | static int iolog_flush(struct io_log *log) |
1436 | { | |
1437 | return 1; | |
1438 | } | |
1439 | ||
1440 | int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log) | |
aee2ab67 JA |
1441 | { |
1442 | return 1; | |
1443 | } | |
1444 | ||
24660963 | 1445 | int iolog_compress_init(struct thread_data *td, struct sk_out *sk_out) |
52618a2a JA |
1446 | { |
1447 | return 0; | |
1448 | } | |
1449 | ||
1450 | void iolog_compress_exit(struct thread_data *td) | |
1451 | { | |
1452 | } | |
1453 | ||
aee2ab67 JA |
1454 | #endif |
1455 | ||
7e419452 JA |
1456 | struct io_logs *iolog_cur_log(struct io_log *log) |
1457 | { | |
1458 | if (flist_empty(&log->io_logs)) | |
1459 | return NULL; | |
1460 | ||
1461 | return flist_last_entry(&log->io_logs, struct io_logs, list); | |
1462 | } | |
1463 | ||
1464 | uint64_t iolog_nr_samples(struct io_log *iolog) | |
1465 | { | |
1466 | struct flist_head *entry; | |
1467 | uint64_t ret = 0; | |
1468 | ||
1469 | flist_for_each(entry, &iolog->io_logs) { | |
1470 | struct io_logs *cur_log; | |
1471 | ||
1472 | cur_log = flist_entry(entry, struct io_logs, list); | |
1473 | ret += cur_log->nr_samples; | |
1474 | } | |
1475 | ||
1476 | return ret; | |
1477 | } | |
1478 | ||
1c725e70 | 1479 | static int __write_log(struct thread_data *td, struct io_log *log, int try) |
905e3d4f | 1480 | { |
1c725e70 JA |
1481 | if (log) |
1482 | return finish_log(td, log, try); | |
905e3d4f | 1483 | |
1c725e70 JA |
1484 | return 0; |
1485 | } | |
905e3d4f | 1486 | |
a47591e4 | 1487 | static int write_iops_log(struct thread_data *td, int try, bool unit_log) |
1c725e70 | 1488 | { |
a47591e4 JA |
1489 | int ret; |
1490 | ||
1491 | if (per_unit_log(td->iops_log) != unit_log) | |
1492 | return 0; | |
1493 | ||
1494 | ret = __write_log(td, td->iops_log, try); | |
1495 | if (!ret) | |
1496 | td->iops_log = NULL; | |
1497 | ||
1498 | return ret; | |
905e3d4f JA |
1499 | } |
1500 | ||
a47591e4 | 1501 | static int write_slat_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1502 | { |
a47591e4 JA |
1503 | int ret; |
1504 | ||
1505 | if (!unit_log) | |
1506 | return 0; | |
1507 | ||
1508 | ret = __write_log(td, td->slat_log, try); | |
1509 | if (!ret) | |
1510 | td->slat_log = NULL; | |
1511 | ||
1512 | return ret; | |
905e3d4f JA |
1513 | } |
1514 | ||
a47591e4 | 1515 | static int write_clat_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1516 | { |
a47591e4 JA |
1517 | int ret; |
1518 | ||
1519 | if (!unit_log) | |
1520 | return 0; | |
1521 | ||
1522 | ret = __write_log(td, td->clat_log, try); | |
1523 | if (!ret) | |
1524 | td->clat_log = NULL; | |
1525 | ||
1526 | return ret; | |
905e3d4f JA |
1527 | } |
1528 | ||
1e613c9c KC |
1529 | static int write_clat_hist_log(struct thread_data *td, int try, bool unit_log) |
1530 | { | |
1531 | int ret; | |
1532 | ||
1533 | if (!unit_log) | |
1534 | return 0; | |
1535 | ||
1536 | ret = __write_log(td, td->clat_hist_log, try); | |
1537 | if (!ret) | |
1538 | td->clat_hist_log = NULL; | |
1539 | ||
1540 | return ret; | |
1541 | } | |
1542 | ||
a47591e4 | 1543 | static int write_lat_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1544 | { |
a47591e4 JA |
1545 | int ret; |
1546 | ||
1547 | if (!unit_log) | |
1548 | return 0; | |
1549 | ||
1550 | ret = __write_log(td, td->lat_log, try); | |
1551 | if (!ret) | |
1552 | td->lat_log = NULL; | |
1553 | ||
1554 | return ret; | |
905e3d4f JA |
1555 | } |
1556 | ||
a47591e4 | 1557 | static int write_bandw_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1558 | { |
a47591e4 JA |
1559 | int ret; |
1560 | ||
1561 | if (per_unit_log(td->bw_log) != unit_log) | |
1562 | return 0; | |
1563 | ||
1564 | ret = __write_log(td, td->bw_log, try); | |
1565 | if (!ret) | |
1566 | td->bw_log = NULL; | |
1567 | ||
1568 | return ret; | |
905e3d4f JA |
1569 | } |
1570 | ||
1571 | enum { | |
1572 | BW_LOG_MASK = 1, | |
1573 | LAT_LOG_MASK = 2, | |
1574 | SLAT_LOG_MASK = 4, | |
1575 | CLAT_LOG_MASK = 8, | |
1576 | IOPS_LOG_MASK = 16, | |
1e613c9c | 1577 | CLAT_HIST_LOG_MASK = 32, |
905e3d4f | 1578 | |
1e613c9c | 1579 | ALL_LOG_NR = 6, |
905e3d4f JA |
1580 | }; |
1581 | ||
1582 | struct log_type { | |
1583 | unsigned int mask; | |
a47591e4 | 1584 | int (*fn)(struct thread_data *, int, bool); |
905e3d4f JA |
1585 | }; |
1586 | ||
1587 | static struct log_type log_types[] = { | |
1588 | { | |
1589 | .mask = BW_LOG_MASK, | |
1590 | .fn = write_bandw_log, | |
1591 | }, | |
1592 | { | |
1593 | .mask = LAT_LOG_MASK, | |
1594 | .fn = write_lat_log, | |
1595 | }, | |
1596 | { | |
1597 | .mask = SLAT_LOG_MASK, | |
1598 | .fn = write_slat_log, | |
1599 | }, | |
1600 | { | |
1601 | .mask = CLAT_LOG_MASK, | |
1602 | .fn = write_clat_log, | |
1603 | }, | |
1604 | { | |
1605 | .mask = IOPS_LOG_MASK, | |
1606 | .fn = write_iops_log, | |
1607 | }, | |
1e613c9c KC |
1608 | { |
1609 | .mask = CLAT_HIST_LOG_MASK, | |
1610 | .fn = write_clat_hist_log, | |
1611 | } | |
905e3d4f JA |
1612 | }; |
1613 | ||
a47591e4 | 1614 | void td_writeout_logs(struct thread_data *td, bool unit_logs) |
905e3d4f | 1615 | { |
ea5409f9 | 1616 | unsigned int log_mask = 0; |
905e3d4f JA |
1617 | unsigned int log_left = ALL_LOG_NR; |
1618 | int old_state, i; | |
1619 | ||
1620 | old_state = td_bump_runstate(td, TD_FINISHING); | |
1621 | ||
a47591e4 | 1622 | finalize_logs(td, unit_logs); |
905e3d4f JA |
1623 | |
1624 | while (log_left) { | |
1625 | int prev_log_left = log_left; | |
1626 | ||
1627 | for (i = 0; i < ALL_LOG_NR && log_left; i++) { | |
1628 | struct log_type *lt = &log_types[i]; | |
1629 | int ret; | |
1630 | ||
ea5409f9 | 1631 | if (!(log_mask & lt->mask)) { |
a47591e4 | 1632 | ret = lt->fn(td, log_left != 1, unit_logs); |
905e3d4f JA |
1633 | if (!ret) { |
1634 | log_left--; | |
ea5409f9 | 1635 | log_mask |= lt->mask; |
905e3d4f JA |
1636 | } |
1637 | } | |
1638 | } | |
1639 | ||
1640 | if (prev_log_left == log_left) | |
1641 | usleep(5000); | |
1642 | } | |
1643 | ||
1644 | td_restore_runstate(td, old_state); | |
1645 | } | |
a47591e4 JA |
1646 | |
1647 | void fio_writeout_logs(bool unit_logs) | |
1648 | { | |
1649 | struct thread_data *td; | |
1650 | int i; | |
1651 | ||
1652 | for_each_td(td, i) | |
1653 | td_writeout_logs(td, unit_logs); | |
1654 | } |