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" |
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 |
31 | static int iolog_flush(struct io_log *log); |
32 | ||
ac9b9101 | 33 | static const char iolog_ver2[] = "fio version 2 iolog"; |
315bbf01 | 34 | static const char iolog_ver3[] = "fio version 3 iolog"; |
ac9b9101 JA |
35 | |
36 | void 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 | 42 | void 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 | ||
57 | void 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 | ||
81 | static 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; |
8b6a404c | 85 | struct timespec ts; |
b85c01f7 | 86 | int ret = 0; |
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 | 100 | |
b85c01f7 KG |
101 | while (delay && !td->terminate) { |
102 | ret = io_u_queued_complete(td, 0); | |
103 | if (ret < 0) | |
104 | td_verror(td, -ret, "io_u_queued_complete"); | |
c3c398e0 SK |
105 | if (td->flags & TD_F_REGROW_LOGS) |
106 | regrow_logs(td); | |
b85c01f7 KG |
107 | if (utime_since_now(&ts) > delay) |
108 | break; | |
30b18672 | 109 | } |
df8fb609 | 110 | |
8b6a404c | 111 | usec = utime_since_now(&ts); |
f4b49c55 JA |
112 | if (usec > orig_delay) |
113 | td->time_offset = usec - orig_delay; | |
df8fb609 JA |
114 | else |
115 | td->time_offset = 0; | |
ac9b9101 JA |
116 | } |
117 | ||
118 | static int ipo_special(struct thread_data *td, struct io_piece *ipo) | |
119 | { | |
120 | struct fio_file *f; | |
121 | int ret; | |
122 | ||
123 | /* | |
124 | * Not a special ipo | |
125 | */ | |
126 | if (ipo->ddir != DDIR_INVAL) | |
127 | return 0; | |
128 | ||
129 | f = td->files[ipo->fileno]; | |
130 | ||
315bbf01 MG |
131 | if (ipo->delay) |
132 | iolog_delay(td, ipo->delay); | |
133 | if (fio_fill_issue_time(td)) | |
134 | fio_gettime(&td->last_issue, NULL); | |
ac9b9101 JA |
135 | switch (ipo->file_action) { |
136 | case FIO_LOG_OPEN_FILE: | |
96d663ad SW |
137 | if (td->o.replay_redirect && fio_file_open(f)) { |
138 | dprint(FD_FILE, "iolog: ignoring re-open of file %s\n", | |
139 | f->file_name); | |
140 | break; | |
141 | } | |
ac9b9101 | 142 | ret = td_io_open_file(td, f); |
f0826678 HP |
143 | if (!ret) { |
144 | if (td->o.dp_type != FIO_DP_NONE) { | |
145 | int dp_init_ret = dp_init(td); | |
146 | ||
147 | if (dp_init_ret != 0) { | |
4150d2e9 | 148 | td_verror(td, abs(dp_init_ret), "dp_init"); |
f0826678 HP |
149 | return -1; |
150 | } | |
151 | } | |
ac9b9101 | 152 | break; |
f0826678 | 153 | } |
ac9b9101 JA |
154 | td_verror(td, ret, "iolog open file"); |
155 | return -1; | |
156 | case FIO_LOG_CLOSE_FILE: | |
157 | td_io_close_file(td, f); | |
158 | break; | |
159 | case FIO_LOG_UNLINK_FILE: | |
38ef9c90 | 160 | td_io_unlink_file(td, f); |
ac9b9101 | 161 | break; |
315bbf01 MG |
162 | case FIO_LOG_ADD_FILE: |
163 | /* | |
164 | * Nothing to do | |
165 | */ | |
166 | break; | |
ac9b9101 JA |
167 | default: |
168 | log_err("fio: bad file action %d\n", ipo->file_action); | |
169 | break; | |
170 | } | |
171 | ||
172 | return 1; | |
173 | } | |
174 | ||
315bbf01 MG |
175 | static bool read_iolog(struct thread_data *td); |
176 | ||
177 | unsigned long long delay_since_ttime(const struct thread_data *td, | |
178 | unsigned long long time) | |
179 | { | |
180 | double tmp; | |
181 | double scale; | |
182 | const unsigned long long *last_ttime = &td->io_log_last_ttime; | |
183 | ||
184 | if (!*last_ttime || td->o.no_stall || time < *last_ttime) | |
185 | return 0; | |
186 | else if (td->o.replay_time_scale == 100) | |
187 | return time - *last_ttime; | |
188 | ||
189 | ||
190 | scale = (double) 100.0 / (double) td->o.replay_time_scale; | |
191 | tmp = time - *last_ttime; | |
192 | return tmp * scale; | |
193 | } | |
98e7161c | 194 | |
ac9b9101 JA |
195 | int read_iolog_get(struct thread_data *td, struct io_u *io_u) |
196 | { | |
197 | struct io_piece *ipo; | |
198 | unsigned long elapsed; | |
3c3ed070 | 199 | |
ac9b9101 JA |
200 | while (!flist_empty(&td->io_log_list)) { |
201 | int ret; | |
b61fdc8a | 202 | |
10f74940 | 203 | if (td->o.read_iolog_chunked) { |
98e7161c | 204 | if (td->io_log_checkmark == td->io_log_current) { |
10f74940 LS |
205 | if (td->io_log_blktrace) { |
206 | if (!read_blktrace(td)) | |
207 | return 1; | |
208 | } else { | |
315bbf01 | 209 | if (!read_iolog(td)) |
10f74940 LS |
210 | return 1; |
211 | } | |
98e7161c AK |
212 | } |
213 | td->io_log_current--; | |
214 | } | |
9342d5f8 | 215 | ipo = flist_first_entry(&td->io_log_list, struct io_piece, list); |
ac9b9101 JA |
216 | flist_del(&ipo->list); |
217 | remove_trim_entry(td, ipo); | |
218 | ||
219 | ret = ipo_special(td, ipo); | |
220 | if (ret < 0) { | |
221 | free(ipo); | |
222 | break; | |
223 | } else if (ret > 0) { | |
224 | free(ipo); | |
225 | continue; | |
226 | } | |
227 | ||
228 | io_u->ddir = ipo->ddir; | |
229 | if (ipo->ddir != DDIR_WAIT) { | |
230 | io_u->offset = ipo->offset; | |
4fff54cc | 231 | io_u->verify_offset = ipo->offset; |
ac9b9101 JA |
232 | io_u->buflen = ipo->len; |
233 | io_u->file = td->files[ipo->fileno]; | |
234 | get_file(io_u->file); | |
5fff9543 | 235 | dprint(FD_IO, "iolog: get %llu/%llu/%s\n", io_u->offset, |
ac9b9101 JA |
236 | io_u->buflen, io_u->file->file_name); |
237 | if (ipo->delay) | |
238 | iolog_delay(td, ipo->delay); | |
a70e681c HP |
239 | |
240 | if (td->o.dp_type != FIO_DP_NONE) | |
241 | dp_fill_dspec_data(td, io_u); | |
ac9b9101 JA |
242 | } else { |
243 | elapsed = mtime_since_genesis(); | |
244 | if (ipo->delay > elapsed) | |
245 | usec_sleep(td, (ipo->delay - elapsed) * 1000); | |
ac9b9101 JA |
246 | } |
247 | ||
248 | free(ipo); | |
3c3ed070 | 249 | |
ac9b9101 JA |
250 | if (io_u->ddir != DDIR_WAIT) |
251 | return 0; | |
252 | } | |
253 | ||
254 | td->done = 1; | |
255 | return 1; | |
256 | } | |
257 | ||
258 | void prune_io_piece_log(struct thread_data *td) | |
259 | { | |
260 | struct io_piece *ipo; | |
e96c0125 | 261 | struct fio_rb_node *n; |
ac9b9101 JA |
262 | |
263 | while ((n = rb_first(&td->io_hist_tree)) != NULL) { | |
264 | ipo = rb_entry(n, struct io_piece, rb_node); | |
265 | rb_erase(n, &td->io_hist_tree); | |
266 | remove_trim_entry(td, ipo); | |
267 | td->io_hist_len--; | |
268 | free(ipo); | |
269 | } | |
270 | ||
271 | while (!flist_empty(&td->io_hist_list)) { | |
4e1020f8 | 272 | ipo = flist_first_entry(&td->io_hist_list, struct io_piece, list); |
ac9b9101 JA |
273 | flist_del(&ipo->list); |
274 | remove_trim_entry(td, ipo); | |
275 | td->io_hist_len--; | |
276 | free(ipo); | |
277 | } | |
278 | } | |
279 | ||
280 | /* | |
281 | * log a successful write, so we can unwind the log for verify | |
282 | */ | |
283 | void log_io_piece(struct thread_data *td, struct io_u *io_u) | |
284 | { | |
e96c0125 | 285 | struct fio_rb_node **p, *parent; |
ac9b9101 JA |
286 | struct io_piece *ipo, *__ipo; |
287 | ||
8812d7f2 | 288 | ipo = calloc(1, sizeof(struct io_piece)); |
ac9b9101 JA |
289 | init_ipo(ipo); |
290 | ipo->file = io_u->file; | |
291 | ipo->offset = io_u->offset; | |
292 | ipo->len = io_u->buflen; | |
da0a7bd2 | 293 | ipo->numberio = io_u->numberio; |
f9401285 JA |
294 | ipo->flags = IP_F_IN_FLIGHT; |
295 | ||
296 | io_u->ipo = ipo; | |
ac9b9101 JA |
297 | |
298 | if (io_u_should_trim(td, io_u)) { | |
299 | flist_add_tail(&ipo->trim_list, &td->trim_list); | |
300 | td->trim_entries++; | |
301 | } | |
302 | ||
303 | /* | |
e43f210a | 304 | * Only sort writes if we don't have a random map in which case we need |
4a06cb41 SW |
305 | * to check for duplicate blocks and drop the old one, which we rely on |
306 | * the rb insert/lookup for handling. | |
ac9b9101 | 307 | */ |
f31feaa2 | 308 | if (file_randommap(td, ipo->file)) { |
ac9b9101 JA |
309 | INIT_FLIST_HEAD(&ipo->list); |
310 | flist_add_tail(&ipo->list, &td->io_hist_list); | |
311 | ipo->flags |= IP_F_ONLIST; | |
312 | td->io_hist_len++; | |
313 | return; | |
314 | } | |
315 | ||
316 | RB_CLEAR_NODE(&ipo->rb_node); | |
317 | ||
318 | /* | |
319 | * Sort the entry into the verification list | |
320 | */ | |
321 | restart: | |
322 | p = &td->io_hist_tree.rb_node; | |
323 | parent = NULL; | |
324 | while (*p) { | |
3b1ff869 | 325 | int overlap = 0; |
ac9b9101 JA |
326 | parent = *p; |
327 | ||
328 | __ipo = rb_entry(parent, struct io_piece, rb_node); | |
329 | if (ipo->file < __ipo->file) | |
330 | p = &(*p)->rb_left; | |
331 | else if (ipo->file > __ipo->file) | |
332 | p = &(*p)->rb_right; | |
3b1ff869 | 333 | else if (ipo->offset < __ipo->offset) { |
ac9b9101 | 334 | p = &(*p)->rb_left; |
3b1ff869 JE |
335 | overlap = ipo->offset + ipo->len > __ipo->offset; |
336 | } | |
337 | else if (ipo->offset > __ipo->offset) { | |
ac9b9101 | 338 | p = &(*p)->rb_right; |
3b1ff869 JE |
339 | overlap = __ipo->offset + __ipo->len > ipo->offset; |
340 | } | |
341 | else | |
342 | overlap = 1; | |
343 | ||
344 | if (overlap) { | |
7abe6a27 | 345 | dprint(FD_IO, "iolog: overlap %llu/%lu, %llu/%lu\n", |
885ac623 JA |
346 | __ipo->offset, __ipo->len, |
347 | ipo->offset, ipo->len); | |
ac9b9101 JA |
348 | td->io_hist_len--; |
349 | rb_erase(parent, &td->io_hist_tree); | |
350 | remove_trim_entry(td, __ipo); | |
e8b74655 SW |
351 | if (!(__ipo->flags & IP_F_IN_FLIGHT)) |
352 | free(__ipo); | |
ac9b9101 JA |
353 | goto restart; |
354 | } | |
355 | } | |
356 | ||
357 | rb_link_node(&ipo->rb_node, parent, p); | |
358 | rb_insert_color(&ipo->rb_node, &td->io_hist_tree); | |
359 | ipo->flags |= IP_F_ONRB; | |
360 | td->io_hist_len++; | |
361 | } | |
362 | ||
890b6656 JA |
363 | void unlog_io_piece(struct thread_data *td, struct io_u *io_u) |
364 | { | |
365 | struct io_piece *ipo = io_u->ipo; | |
366 | ||
66347cfa DE |
367 | if (td->ts.nr_block_infos) { |
368 | uint32_t *info = io_u_block_info(td, io_u); | |
369 | if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) { | |
370 | if (io_u->ddir == DDIR_TRIM) | |
371 | *info = BLOCK_INFO_SET_STATE(*info, | |
372 | BLOCK_STATE_TRIM_FAILURE); | |
373 | else if (io_u->ddir == DDIR_WRITE) | |
374 | *info = BLOCK_INFO_SET_STATE(*info, | |
375 | BLOCK_STATE_WRITE_FAILURE); | |
376 | } | |
377 | } | |
378 | ||
890b6656 JA |
379 | if (!ipo) |
380 | return; | |
381 | ||
382 | if (ipo->flags & IP_F_ONRB) | |
383 | rb_erase(&ipo->rb_node, &td->io_hist_tree); | |
384 | else if (ipo->flags & IP_F_ONLIST) | |
385 | flist_del(&ipo->list); | |
386 | ||
387 | free(ipo); | |
388 | io_u->ipo = NULL; | |
389 | td->io_hist_len--; | |
390 | } | |
391 | ||
94a78d52 | 392 | void trim_io_piece(const struct io_u *io_u) |
890b6656 JA |
393 | { |
394 | struct io_piece *ipo = io_u->ipo; | |
395 | ||
396 | if (!ipo) | |
397 | return; | |
398 | ||
399 | ipo->len = io_u->xfer_buflen - io_u->resid; | |
400 | } | |
401 | ||
ac9b9101 JA |
402 | void write_iolog_close(struct thread_data *td) |
403 | { | |
d087d28f DLM |
404 | if (!td->iolog_f) |
405 | return; | |
406 | ||
ac9b9101 JA |
407 | fflush(td->iolog_f); |
408 | fclose(td->iolog_f); | |
409 | free(td->iolog_buf); | |
410 | td->iolog_f = NULL; | |
411 | td->iolog_buf = NULL; | |
412 | } | |
413 | ||
a21ed2b5 | 414 | int64_t iolog_items_to_fetch(struct thread_data *td) |
1d73ff2a JA |
415 | { |
416 | struct timespec now; | |
417 | uint64_t elapsed; | |
418 | uint64_t for_1s; | |
419 | int64_t items_to_fetch; | |
420 | ||
421 | if (!td->io_log_highmark) | |
422 | return 10; | |
423 | ||
424 | ||
425 | fio_gettime(&now, NULL); | |
426 | elapsed = ntime_since(&td->io_log_highmark_time, &now); | |
427 | if (elapsed) { | |
428 | for_1s = (td->io_log_highmark - td->io_log_current) * 1000000000 / elapsed; | |
429 | items_to_fetch = for_1s - td->io_log_current; | |
430 | if (items_to_fetch < 0) | |
431 | items_to_fetch = 0; | |
432 | } else | |
433 | items_to_fetch = 0; | |
434 | ||
435 | td->io_log_highmark = td->io_log_current + items_to_fetch; | |
436 | td->io_log_checkmark = (td->io_log_highmark + 1) / 2; | |
437 | fio_gettime(&td->io_log_highmark_time, NULL); | |
438 | ||
439 | return items_to_fetch; | |
440 | } | |
441 | ||
315bbf01 MG |
442 | #define io_act(_td, _r) (((_td)->io_log_version == 3 && (r) == 5) || \ |
443 | ((_td)->io_log_version == 2 && (r) == 4)) | |
444 | #define file_act(_td, _r) (((_td)->io_log_version == 3 && (r) == 3) || \ | |
445 | ((_td)->io_log_version == 2 && (r) == 2)) | |
446 | ||
ac9b9101 | 447 | /* |
315bbf01 | 448 | * Read version 2 and 3 iolog data. It is enhanced to include per-file logging, |
ac9b9101 JA |
449 | * syncs, etc. |
450 | */ | |
315bbf01 | 451 | static bool read_iolog(struct thread_data *td) |
ac9b9101 JA |
452 | { |
453 | unsigned long long offset; | |
454 | unsigned int bytes; | |
315bbf01 | 455 | unsigned long long delay = 0; |
dd6500e9 | 456 | int reads, writes, trims, waits, fileno = 0, file_action = 0; /* stupid gcc */ |
dac34079 | 457 | char *rfname, *fname, *act; |
ac9b9101 JA |
458 | char *str, *p; |
459 | enum fio_ddir rw; | |
71e6e5a2 | 460 | bool realloc = false; |
98e7161c | 461 | int64_t items_to_fetch = 0; |
78c0d7a0 | 462 | int syncs; |
ac9b9101 | 463 | |
98e7161c | 464 | if (td->o.read_iolog_chunked) { |
1d73ff2a JA |
465 | items_to_fetch = iolog_items_to_fetch(td); |
466 | if (!items_to_fetch) | |
467 | return true; | |
98e7161c | 468 | } |
1d73ff2a | 469 | |
ac9b9101 JA |
470 | /* |
471 | * Read in the read iolog and store it, reuse the infrastructure | |
472 | * for doing verifications. | |
473 | */ | |
474 | str = malloc(4096); | |
dac34079 | 475 | rfname = fname = malloc(256+16); |
ac9b9101 JA |
476 | act = malloc(256+16); |
477 | ||
dd6500e9 | 478 | syncs = reads = writes = trims = waits = 0; |
98e7161c | 479 | while ((p = fgets(str, 4096, td->io_log_rfile)) != NULL) { |
ac9b9101 JA |
480 | struct io_piece *ipo; |
481 | int r; | |
315bbf01 | 482 | unsigned long long ttime; |
ac9b9101 | 483 | |
315bbf01 MG |
484 | if (td->io_log_version == 3) { |
485 | r = sscanf(p, "%llu %256s %256s %llu %u", &ttime, rfname, act, | |
486 | &offset, &bytes); | |
487 | delay = delay_since_ttime(td, ttime); | |
488 | td->io_log_last_ttime = ttime; | |
489 | /* | |
490 | * "wait" is not allowed with version 3 | |
491 | */ | |
492 | if (!strcmp(act, "wait")) { | |
493 | log_err("iolog: ignoring wait command with" | |
494 | " version 3 for file %s\n", fname); | |
495 | continue; | |
496 | } | |
497 | } else /* version 2 */ | |
498 | r = sscanf(p, "%256s %256s %llu %u", rfname, act, &offset, &bytes); | |
dac34079 JA |
499 | |
500 | if (td->o.replay_redirect) | |
501 | fname = td->o.replay_redirect; | |
502 | ||
315bbf01 | 503 | if (io_act(td, r)) { |
ac9b9101 JA |
504 | /* |
505 | * Check action first | |
506 | */ | |
507 | if (!strcmp(act, "wait")) | |
508 | rw = DDIR_WAIT; | |
ff6131af H |
509 | else if (!strcmp(act, "read")) { |
510 | if (td->o.replay_skip & (1u << DDIR_READ)) | |
511 | continue; | |
ac9b9101 | 512 | rw = DDIR_READ; |
ff6131af H |
513 | } else if (!strcmp(act, "write")) { |
514 | if (td->o.replay_skip & (1u << DDIR_WRITE)) | |
515 | continue; | |
ac9b9101 | 516 | rw = DDIR_WRITE; |
ff6131af H |
517 | } else if (!strcmp(act, "sync")) { |
518 | if (td->o.replay_skip & (1u << DDIR_SYNC)) | |
519 | continue; | |
ac9b9101 | 520 | rw = DDIR_SYNC; |
ff6131af | 521 | } else if (!strcmp(act, "datasync")) |
ac9b9101 | 522 | rw = DDIR_DATASYNC; |
ff6131af H |
523 | else if (!strcmp(act, "trim")) { |
524 | if (td->o.replay_skip & (1u << DDIR_TRIM)) | |
525 | continue; | |
ac9b9101 | 526 | rw = DDIR_TRIM; |
ff6131af | 527 | } else { |
ac9b9101 JA |
528 | log_err("fio: bad iolog file action: %s\n", |
529 | act); | |
530 | continue; | |
531 | } | |
033ace1e | 532 | fileno = get_fileno(td, fname); |
315bbf01 | 533 | } else if (file_act(td, r)) { |
ac9b9101 JA |
534 | rw = DDIR_INVAL; |
535 | if (!strcmp(act, "add")) { | |
96d663ad SW |
536 | if (td->o.replay_redirect && |
537 | get_fileno(td, fname) != -1) { | |
538 | dprint(FD_FILE, "iolog: ignoring" | |
539 | " re-add of file %s\n", fname); | |
540 | } else { | |
c70c7f58 | 541 | fileno = add_file(td, fname, td->subjob_number, 1); |
96d663ad SW |
542 | file_action = FIO_LOG_ADD_FILE; |
543 | } | |
ac9b9101 JA |
544 | } else if (!strcmp(act, "open")) { |
545 | fileno = get_fileno(td, fname); | |
546 | file_action = FIO_LOG_OPEN_FILE; | |
547 | } else if (!strcmp(act, "close")) { | |
548 | fileno = get_fileno(td, fname); | |
549 | file_action = FIO_LOG_CLOSE_FILE; | |
550 | } else { | |
551 | log_err("fio: bad iolog file action: %s\n", | |
552 | act); | |
553 | continue; | |
554 | } | |
555 | } else { | |
315bbf01 | 556 | log_err("bad iolog%d: %s\n", td->io_log_version, p); |
ac9b9101 JA |
557 | continue; |
558 | } | |
559 | ||
560 | if (rw == DDIR_READ) | |
561 | reads++; | |
562 | else if (rw == DDIR_WRITE) { | |
563 | /* | |
564 | * Don't add a write for ro mode | |
565 | */ | |
566 | if (read_only) | |
567 | continue; | |
568 | writes++; | |
dd6500e9 VF |
569 | } else if (rw == DDIR_TRIM) { |
570 | /* | |
571 | * Don't add a trim for ro mode | |
572 | */ | |
573 | if (read_only) | |
574 | continue; | |
575 | trims++; | |
ac9b9101 | 576 | } else if (rw == DDIR_WAIT) { |
74dff443 JA |
577 | if (td->o.no_stall) |
578 | continue; | |
ac9b9101 JA |
579 | waits++; |
580 | } else if (rw == DDIR_INVAL) { | |
78c0d7a0 JA |
581 | } else if (ddir_sync(rw)) { |
582 | syncs++; | |
583 | } else { | |
ac9b9101 JA |
584 | log_err("bad ddir: %d\n", rw); |
585 | continue; | |
586 | } | |
587 | ||
588 | /* | |
589 | * Make note of file | |
590 | */ | |
8812d7f2 | 591 | ipo = calloc(1, sizeof(*ipo)); |
ac9b9101 JA |
592 | init_ipo(ipo); |
593 | ipo->ddir = rw; | |
315bbf01 MG |
594 | if (td->io_log_version == 3) |
595 | ipo->delay = delay; | |
ac9b9101 JA |
596 | if (rw == DDIR_WAIT) { |
597 | ipo->delay = offset; | |
598 | } else { | |
a79f17bf SW |
599 | if (td->o.replay_scale) |
600 | ipo->offset = offset / td->o.replay_scale; | |
601 | else | |
602 | ipo->offset = offset; | |
603 | ipo_bytes_align(td->o.replay_align, ipo); | |
604 | ||
ac9b9101 | 605 | ipo->len = bytes; |
71e6e5a2 AK |
606 | if (rw != DDIR_INVAL && bytes > td->o.max_bs[rw]) { |
607 | realloc = true; | |
ac9b9101 | 608 | td->o.max_bs[rw] = bytes; |
71e6e5a2 | 609 | } |
ac9b9101 JA |
610 | ipo->fileno = fileno; |
611 | ipo->file_action = file_action; | |
70afff59 | 612 | td->o.size += bytes; |
ac9b9101 | 613 | } |
3c3ed070 | 614 | |
ac9b9101 | 615 | queue_io_piece(td, ipo); |
98e7161c AK |
616 | |
617 | if (td->o.read_iolog_chunked) { | |
618 | td->io_log_current++; | |
619 | items_to_fetch--; | |
620 | if (items_to_fetch == 0) | |
621 | break; | |
622 | } | |
ac9b9101 JA |
623 | } |
624 | ||
625 | free(str); | |
626 | free(act); | |
dac34079 | 627 | free(rfname); |
ac9b9101 | 628 | |
98e7161c AK |
629 | if (td->o.read_iolog_chunked) { |
630 | td->io_log_highmark = td->io_log_current; | |
631 | td->io_log_checkmark = (td->io_log_highmark + 1) / 2; | |
632 | fio_gettime(&td->io_log_highmark_time, NULL); | |
633 | } | |
634 | ||
ac9b9101 JA |
635 | if (writes && read_only) { |
636 | log_err("fio: <%s> skips replay of %d writes due to" | |
637 | " read-only\n", td->o.name, writes); | |
638 | writes = 0; | |
639 | } | |
78c0d7a0 JA |
640 | if (syncs) |
641 | td->flags |= TD_F_SYNCS; | |
ac9b9101 | 642 | |
98e7161c AK |
643 | if (td->o.read_iolog_chunked) { |
644 | if (td->io_log_current == 0) { | |
645 | return false; | |
646 | } | |
647 | td->o.td_ddir = TD_DDIR_RW; | |
71e6e5a2 AK |
648 | if (realloc && td->orig_buffer) |
649 | { | |
650 | io_u_quiesce(td); | |
651 | free_io_mem(td); | |
6d8fe6e8 SK |
652 | if (init_io_u_buffers(td)) |
653 | return false; | |
71e6e5a2 | 654 | } |
98e7161c AK |
655 | return true; |
656 | } | |
657 | ||
dd6500e9 | 658 | if (!reads && !writes && !waits && !trims) |
b153f94a | 659 | return false; |
dd6500e9 VF |
660 | |
661 | td->o.td_ddir = 0; | |
662 | if (reads) | |
663 | td->o.td_ddir |= TD_DDIR_READ; | |
664 | if (writes) | |
665 | td->o.td_ddir |= TD_DDIR_WRITE; | |
666 | if (trims) | |
667 | td->o.td_ddir |= TD_DDIR_TRIM; | |
ac9b9101 | 668 | |
b153f94a | 669 | return true; |
ac9b9101 JA |
670 | } |
671 | ||
2f8f4821 AK |
672 | static bool is_socket(const char *path) |
673 | { | |
674 | struct stat buf; | |
8ea203d3 JA |
675 | int r; |
676 | ||
677 | r = stat(path, &buf); | |
2f8f4821 AK |
678 | if (r == -1) |
679 | return false; | |
680 | ||
681 | return S_ISSOCK(buf.st_mode); | |
682 | } | |
683 | ||
684 | static int open_socket(const char *path) | |
685 | { | |
2f8f4821 | 686 | struct sockaddr_un addr; |
8ea203d3 JA |
687 | int ret, fd; |
688 | ||
689 | fd = socket(AF_UNIX, SOCK_STREAM, 0); | |
2f8f4821 AK |
690 | if (fd < 0) |
691 | return fd; | |
8ea203d3 | 692 | |
2f8f4821 | 693 | addr.sun_family = AF_UNIX; |
c7d23f4b | 694 | if (snprintf(addr.sun_path, sizeof(addr.sun_path), "%s", path) >= |
8ea203d3 | 695 | sizeof(addr.sun_path)) { |
c7d23f4b BVA |
696 | log_err("%s: path name %s is too long for a Unix socket\n", |
697 | __func__, path); | |
8ea203d3 JA |
698 | } |
699 | ||
700 | ret = connect(fd, (const struct sockaddr *)&addr, strlen(path) + sizeof(addr.sun_family)); | |
701 | if (!ret) | |
2f8f4821 | 702 | return fd; |
8ea203d3 JA |
703 | |
704 | close(fd); | |
2f8f4821 AK |
705 | return -1; |
706 | } | |
707 | ||
ac9b9101 JA |
708 | /* |
709 | * open iolog, check version, and call appropriate parser | |
710 | */ | |
22d7fdf3 | 711 | static bool init_iolog_read(struct thread_data *td, char *fname) |
ac9b9101 | 712 | { |
22d7fdf3 | 713 | char buffer[256], *p; |
2f8f4821 | 714 | FILE *f = NULL; |
8ea203d3 | 715 | |
c70c7f58 AK |
716 | dprint(FD_IO, "iolog: name=%s\n", fname); |
717 | ||
718 | if (is_socket(fname)) { | |
8ea203d3 JA |
719 | int fd; |
720 | ||
721 | fd = open_socket(fname); | |
722 | if (fd >= 0) | |
2f8f4821 | 723 | f = fdopen(fd, "r"); |
d19c04d1 | 724 | } else if (!strcmp(fname, "-")) { |
725 | f = stdin; | |
2f8f4821 | 726 | } else |
c70c7f58 | 727 | f = fopen(fname, "r"); |
8ea203d3 | 728 | |
ac9b9101 JA |
729 | if (!f) { |
730 | perror("fopen read iolog"); | |
b153f94a | 731 | return false; |
ac9b9101 JA |
732 | } |
733 | ||
734 | p = fgets(buffer, sizeof(buffer), f); | |
735 | if (!p) { | |
736 | td_verror(td, errno, "iolog read"); | |
737 | log_err("fio: unable to read iolog\n"); | |
738 | fclose(f); | |
b153f94a | 739 | return false; |
ac9b9101 | 740 | } |
8ea203d3 | 741 | |
ac9b9101 | 742 | /* |
315bbf01 | 743 | * versions 2 and 3 of the iolog store a specific string as the |
ac9b9101 JA |
744 | * first line, check for that |
745 | */ | |
315bbf01 MG |
746 | if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2))) |
747 | td->io_log_version = 2; | |
748 | else if (!strncmp(iolog_ver3, buffer, strlen(iolog_ver3))) | |
749 | td->io_log_version = 3; | |
750 | else { | |
751 | log_err("fio: iolog version 1 is no longer supported\n"); | |
752 | fclose(f); | |
753 | return false; | |
ac9b9101 JA |
754 | } |
755 | ||
315bbf01 MG |
756 | free_release_files(td); |
757 | td->io_log_rfile = f; | |
758 | return read_iolog(td); | |
ac9b9101 JA |
759 | } |
760 | ||
761 | /* | |
762 | * Set up a log for storing io patterns. | |
763 | */ | |
b153f94a | 764 | static bool init_iolog_write(struct thread_data *td) |
ac9b9101 JA |
765 | { |
766 | struct fio_file *ff; | |
767 | FILE *f; | |
768 | unsigned int i; | |
769 | ||
770 | f = fopen(td->o.write_iolog_file, "a"); | |
771 | if (!f) { | |
772 | perror("fopen write iolog"); | |
b153f94a | 773 | return false; |
ac9b9101 JA |
774 | } |
775 | ||
776 | /* | |
777 | * That's it for writing, setup a log buffer and we're done. | |
778 | */ | |
779 | td->iolog_f = f; | |
780 | td->iolog_buf = malloc(8192); | |
781 | setvbuf(f, td->iolog_buf, _IOFBF, 8192); | |
e8cf24e5 | 782 | fio_gettime(&td->io_log_start_time, NULL); |
ac9b9101 JA |
783 | |
784 | /* | |
785 | * write our version line | |
786 | */ | |
e8cf24e5 | 787 | if (fprintf(f, "%s\n", iolog_ver3) < 0) { |
ac9b9101 | 788 | perror("iolog init\n"); |
b153f94a | 789 | return false; |
ac9b9101 JA |
790 | } |
791 | ||
792 | /* | |
793 | * add all known files | |
794 | */ | |
795 | for_each_file(td, ff, i) | |
796 | log_file(td, ff, FIO_LOG_ADD_FILE); | |
797 | ||
b153f94a | 798 | return true; |
ac9b9101 JA |
799 | } |
800 | ||
b153f94a | 801 | bool init_iolog(struct thread_data *td) |
ac9b9101 | 802 | { |
b153f94a | 803 | bool ret; |
ac9b9101 JA |
804 | |
805 | if (td->o.read_iolog_file) { | |
d95b34a6 | 806 | int need_swap; |
22d7fdf3 | 807 | char * fname = get_name_by_idx(td->o.read_iolog_file, td->subjob_number); |
d95b34a6 | 808 | |
ac9b9101 JA |
809 | /* |
810 | * Check if it's a blktrace file and load that if possible. | |
811 | * Otherwise assume it's a normal log file and load that. | |
812 | */ | |
b61fdc8a JA |
813 | if (is_blktrace(fname, &need_swap)) { |
814 | td->io_log_blktrace = 1; | |
10f74940 | 815 | ret = init_blktrace_read(td, fname, need_swap); |
b61fdc8a JA |
816 | } else { |
817 | td->io_log_blktrace = 0; | |
22d7fdf3 | 818 | ret = init_iolog_read(td, fname); |
b61fdc8a | 819 | } |
3a3e5c6e | 820 | free(fname); |
ac9b9101 JA |
821 | } else if (td->o.write_iolog_file) |
822 | ret = init_iolog_write(td); | |
b153f94a | 823 | else |
43864deb | 824 | ret = true; |
ac9b9101 | 825 | |
b153f94a | 826 | if (!ret) |
f01b34ae JA |
827 | td_verror(td, EINVAL, "failed initializing iolog"); |
828 | ||
c492cb1a | 829 | init_disk_util(td); |
830 | ||
ac9b9101 JA |
831 | return ret; |
832 | } | |
833 | ||
aee2ab67 JA |
834 | void setup_log(struct io_log **log, struct log_params *p, |
835 | const char *filename) | |
ac9b9101 | 836 | { |
e75cc8f3 | 837 | struct io_log *l; |
65a4d15c KC |
838 | int i; |
839 | struct io_u_plat_entry *entry; | |
840 | struct flist_head *list; | |
ac9b9101 | 841 | |
7e419452 | 842 | l = scalloc(1, sizeof(*l)); |
b6eb53b7 | 843 | assert(l); |
7e419452 | 844 | INIT_FLIST_HEAD(&l->io_logs); |
aee2ab67 JA |
845 | l->log_type = p->log_type; |
846 | l->log_offset = p->log_offset; | |
03ec570f | 847 | l->log_prio = p->log_prio; |
14d3134a | 848 | l->log_issue_time = p->log_issue_time; |
aee2ab67 | 849 | l->log_gz = p->log_gz; |
b26317c9 | 850 | l->log_gz_store = p->log_gz_store; |
aee2ab67 | 851 | l->avg_msec = p->avg_msec; |
1e613c9c KC |
852 | l->hist_msec = p->hist_msec; |
853 | l->hist_coarseness = p->hist_coarseness; | |
cb7e0ace | 854 | l->filename = strdup(filename); |
aee2ab67 JA |
855 | l->td = p->td; |
856 | ||
65a4d15c KC |
857 | /* Initialize histogram lists for each r/w direction, |
858 | * with initial io_u_plat of all zeros: | |
859 | */ | |
860 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { | |
d730bc58 | 861 | list = &l->hist_window[i].list; |
65a4d15c KC |
862 | INIT_FLIST_HEAD(list); |
863 | entry = calloc(1, sizeof(struct io_u_plat_entry)); | |
864 | flist_add(&entry->list, list); | |
865 | } | |
866 | ||
1fed2080 | 867 | if (l->td && l->td->o.io_submit_mode != IO_MODE_OFFLOAD) { |
b5aba537 | 868 | unsigned int def_samples = DEF_LOG_ENTRIES; |
a43f4461 | 869 | struct io_logs *__p; |
1fed2080 | 870 | |
a43f4461 | 871 | __p = calloc(1, sizeof(*l->pending)); |
b5aba537 JA |
872 | if (l->td->o.iodepth > DEF_LOG_ENTRIES) |
873 | def_samples = roundup_pow2(l->td->o.iodepth); | |
874 | __p->max_samples = def_samples; | |
a43f4461 JA |
875 | __p->log = calloc(__p->max_samples, log_entry_sz(l)); |
876 | l->pending = __p; | |
1fed2080 JA |
877 | } |
878 | ||
b26317c9 | 879 | if (l->log_offset) |
49e98daa | 880 | l->log_ddir_mask = LOG_OFFSET_SAMPLE_BIT; |
03ec570f DLM |
881 | if (l->log_prio) |
882 | l->log_ddir_mask |= LOG_PRIO_SAMPLE_BIT; | |
acc481b6 AK |
883 | /* |
884 | * The bandwidth-log option generates agg-read_bw.log, | |
885 | * agg-write_bw.log and agg-trim_bw.log for which l->td is NULL. | |
886 | * Check if l->td is valid before dereferencing it. | |
887 | */ | |
888 | if (l->td && l->td->o.log_max == IO_LOG_SAMPLE_BOTH) | |
0e14633c | 889 | l->log_ddir_mask |= LOG_AVG_MAX_SAMPLE_BIT; |
b26317c9 | 890 | |
14d3134a SK |
891 | if (l->log_issue_time) |
892 | l->log_ddir_mask |= LOG_ISSUE_TIME_SAMPLE_BIT; | |
893 | ||
aee2ab67 JA |
894 | INIT_FLIST_HEAD(&l->chunk_list); |
895 | ||
896 | if (l->log_gz && !p->td) | |
897 | l->log_gz = 0; | |
19417eda | 898 | else if (l->log_gz || l->log_gz_store) { |
34febb23 | 899 | mutex_init_pshared(&l->chunk_lock); |
858bce70 | 900 | mutex_init_pshared(&l->deferred_free_lock); |
aee2ab67 JA |
901 | p->td->flags |= TD_F_COMPRESS_LOG; |
902 | } | |
903 | ||
ac9b9101 JA |
904 | *log = l; |
905 | } | |
906 | ||
2e802282 JA |
907 | #ifdef CONFIG_SETVBUF |
908 | static void *set_file_buffer(FILE *f) | |
909 | { | |
910 | size_t size = 1048576; | |
911 | void *buf; | |
912 | ||
913 | buf = malloc(size); | |
914 | setvbuf(f, buf, _IOFBF, size); | |
915 | return buf; | |
916 | } | |
917 | ||
918 | static void clear_file_buffer(void *buf) | |
919 | { | |
920 | free(buf); | |
921 | } | |
922 | #else | |
923 | static void *set_file_buffer(FILE *f) | |
924 | { | |
925 | return NULL; | |
926 | } | |
927 | ||
928 | static void clear_file_buffer(void *buf) | |
929 | { | |
930 | } | |
931 | #endif | |
932 | ||
518dac09 | 933 | void free_log(struct io_log *log) |
cb7e0ace | 934 | { |
7e419452 JA |
935 | while (!flist_empty(&log->io_logs)) { |
936 | struct io_logs *cur_log; | |
937 | ||
938 | cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); | |
939 | flist_del_init(&cur_log->list); | |
940 | free(cur_log->log); | |
2b7625e2 | 941 | sfree(cur_log); |
7e419452 JA |
942 | } |
943 | ||
1fed2080 JA |
944 | if (log->pending) { |
945 | free(log->pending->log); | |
946 | free(log->pending); | |
947 | log->pending = NULL; | |
948 | } | |
949 | ||
950 | free(log->pending); | |
cb7e0ace | 951 | free(log->filename); |
a47591e4 | 952 | sfree(log); |
cb7e0ace JA |
953 | } |
954 | ||
6cc0e5aa AL |
955 | uint64_t hist_sum(int j, int stride, uint64_t *io_u_plat, |
956 | uint64_t *io_u_plat_last) | |
548df930 | 957 | { |
6cc0e5aa | 958 | uint64_t sum; |
f2c972cc | 959 | int k; |
548df930 | 960 | |
868f6f03 KC |
961 | if (io_u_plat_last) { |
962 | for (k = sum = 0; k < stride; k++) | |
963 | sum += io_u_plat[j + k] - io_u_plat_last[j + k]; | |
964 | } else { | |
965 | for (k = sum = 0; k < stride; k++) | |
966 | sum += io_u_plat[j + k]; | |
967 | } | |
548df930 | 968 | |
1e613c9c KC |
969 | return sum; |
970 | } | |
971 | ||
a89ba4b1 JA |
972 | static void flush_hist_samples(FILE *f, int hist_coarseness, void *samples, |
973 | uint64_t sample_size) | |
1e613c9c KC |
974 | { |
975 | struct io_sample *s; | |
14d3134a | 976 | bool log_offset, log_issue_time; |
1e613c9c | 977 | uint64_t i, j, nr_samples; |
65a4d15c | 978 | struct io_u_plat_entry *entry, *entry_before; |
6cc0e5aa AL |
979 | uint64_t *io_u_plat; |
980 | uint64_t *io_u_plat_before; | |
1e613c9c KC |
981 | |
982 | int stride = 1 << hist_coarseness; | |
983 | ||
984 | if (!sample_size) | |
985 | return; | |
986 | ||
14d3134a | 987 | s = __get_sample(samples, 0, 0, 0); |
1e613c9c | 988 | log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0; |
14d3134a | 989 | log_issue_time = (s->__ddir & LOG_ISSUE_TIME_SAMPLE_BIT) != 0; |
1e613c9c | 990 | |
14d3134a | 991 | nr_samples = sample_size / __log_entry_sz(log_offset, log_issue_time); |
1e613c9c KC |
992 | |
993 | for (i = 0; i < nr_samples; i++) { | |
14d3134a | 994 | s = __get_sample(samples, log_offset, log_issue_time, i); |
d730bc58 | 995 | |
af2fde19 | 996 | entry = s->data.plat_entry; |
65a4d15c | 997 | io_u_plat = entry->io_u_plat; |
d730bc58 | 998 | |
65a4d15c KC |
999 | entry_before = flist_first_entry(&entry->list, struct io_u_plat_entry, list); |
1000 | io_u_plat_before = entry_before->io_u_plat; | |
d730bc58 | 1001 | |
5fff9543 JF |
1002 | fprintf(f, "%lu, %u, %llu, ", (unsigned long) s->time, |
1003 | io_sample_ddir(s), (unsigned long long) s->bs); | |
1e613c9c | 1004 | for (j = 0; j < FIO_IO_U_PLAT_NR - stride; j += stride) { |
6cc0e5aa AL |
1005 | fprintf(f, "%llu, ", (unsigned long long) |
1006 | hist_sum(j, stride, io_u_plat, io_u_plat_before)); | |
1e613c9c | 1007 | } |
6cc0e5aa | 1008 | fprintf(f, "%llu\n", (unsigned long long) |
65a4d15c | 1009 | hist_sum(FIO_IO_U_PLAT_NR - stride, stride, io_u_plat, |
d730bc58 JA |
1010 | io_u_plat_before)); |
1011 | ||
65a4d15c KC |
1012 | flist_del(&entry_before->list); |
1013 | free(entry_before); | |
1e613c9c KC |
1014 | } |
1015 | } | |
1016 | ||
3ec6b6da SK |
1017 | static int print_sample_fields(char **p, size_t *left, const char *fmt, ...) { |
1018 | va_list ap; | |
1019 | int ret; | |
1020 | ||
1021 | va_start(ap, fmt); | |
1022 | ret = vsnprintf(*p, *left, fmt, ap); | |
1023 | if (ret < 0 || ret >= *left) { | |
1024 | log_err("sample file write failed: %d\n", ret); | |
65098b11 | 1025 | va_end(ap); |
3ec6b6da SK |
1026 | return -1; |
1027 | } | |
1028 | va_end(ap); | |
1029 | ||
1030 | *p += ret; | |
1031 | *left -= ret; | |
1032 | ||
1033 | return 0; | |
1034 | } | |
1035 | ||
1036 | /* | |
1037 | * flush_samples - Generate output for log samples | |
1038 | * Each sample output is built using a temporary buffer. This buffer size | |
1039 | * assumptions are: | |
1040 | * - Each sample has less than 10 fields | |
1041 | * - Each sample field fits in 25 characters (20 digits for 64 bit number | |
1042 | * and a few bytes delimiter) | |
1043 | */ | |
bead01d7 | 1044 | void flush_samples(FILE *f, void *samples, uint64_t sample_size) |
ac9b9101 | 1045 | { |
b26317c9 | 1046 | struct io_sample *s; |
14d3134a | 1047 | bool log_offset, log_prio, log_avg_max, log_issue_time; |
b26317c9 | 1048 | uint64_t i, nr_samples; |
3ec6b6da SK |
1049 | char buf[256]; |
1050 | char *p; | |
1051 | size_t left; | |
1052 | int ret; | |
b26317c9 JA |
1053 | |
1054 | if (!sample_size) | |
1055 | return; | |
1056 | ||
14d3134a | 1057 | s = __get_sample(samples, 0, 0, 0); |
49e98daa | 1058 | log_offset = (s->__ddir & LOG_OFFSET_SAMPLE_BIT) != 0; |
03ec570f | 1059 | log_prio = (s->__ddir & LOG_PRIO_SAMPLE_BIT) != 0; |
0e14633c | 1060 | log_avg_max = (s->__ddir & LOG_AVG_MAX_SAMPLE_BIT) != 0; |
14d3134a | 1061 | log_issue_time = (s->__ddir & LOG_ISSUE_TIME_SAMPLE_BIT) != 0; |
03ec570f | 1062 | |
14d3134a | 1063 | nr_samples = sample_size / __log_entry_sz(log_offset, log_issue_time); |
ac9b9101 | 1064 | |
aee2ab67 | 1065 | for (i = 0; i < nr_samples; i++) { |
14d3134a | 1066 | s = __get_sample(samples, log_offset, log_issue_time, i); |
3ec6b6da SK |
1067 | p = buf; |
1068 | left = sizeof(buf); | |
1069 | ||
1070 | ret = print_sample_fields(&p, &left, "%" PRIu64 ", %" PRId64, | |
1071 | s->time, s->data.val.val0); | |
1072 | if (ret) | |
1073 | return; | |
1074 | ||
1075 | if (log_avg_max) { | |
1076 | ret = print_sample_fields(&p, &left, ", %" PRId64, | |
1077 | s->data.val.val1); | |
1078 | if (ret) | |
1079 | return; | |
1080 | } | |
1081 | ||
1082 | ret = print_sample_fields(&p, &left, ", %u, %llu", | |
1083 | io_sample_ddir(s), | |
1084 | (unsigned long long) s->bs); | |
1085 | if (ret) | |
1086 | return; | |
1087 | ||
1088 | if (log_offset) { | |
1089 | ret = print_sample_fields(&p, &left, ", %llu", | |
9f0a81a6 | 1090 | (unsigned long long) s->aux[IOS_AUX_OFFSET_INDEX]); |
3ec6b6da SK |
1091 | if (ret) |
1092 | return; | |
1093 | } | |
ac9b9101 | 1094 | |
03ec570f | 1095 | if (log_prio) |
3ec6b6da SK |
1096 | ret = print_sample_fields(&p, &left, ", 0x%04x", |
1097 | s->priority); | |
03ec570f | 1098 | else |
3ec6b6da SK |
1099 | ret = print_sample_fields(&p, &left, ", %u", |
1100 | ioprio_value_is_class_rt(s->priority)); | |
1101 | if (ret) | |
1102 | return; | |
1103 | ||
14d3134a SK |
1104 | if (log_issue_time) { |
1105 | ret = print_sample_fields(&p, &left, ", %llu", | |
1106 | (unsigned long long) s->aux[IOS_AUX_ISSUE_TIME_INDEX]); | |
1107 | if (ret) | |
1108 | return; | |
1109 | } | |
1110 | ||
3ec6b6da | 1111 | fprintf(f, "%s\n", buf); |
ac9b9101 | 1112 | } |
aee2ab67 JA |
1113 | } |
1114 | ||
1115 | #ifdef CONFIG_ZLIB | |
97eabb2a JA |
1116 | |
1117 | struct iolog_flush_data { | |
155f2f02 | 1118 | struct workqueue_work work; |
97eabb2a JA |
1119 | struct io_log *log; |
1120 | void *samples; | |
7e419452 JA |
1121 | uint32_t nr_samples; |
1122 | bool free; | |
97eabb2a JA |
1123 | }; |
1124 | ||
97eabb2a JA |
1125 | #define GZ_CHUNK 131072 |
1126 | ||
1127 | static struct iolog_compress *get_new_chunk(unsigned int seq) | |
1128 | { | |
1129 | struct iolog_compress *c; | |
1130 | ||
1131 | c = malloc(sizeof(*c)); | |
1132 | INIT_FLIST_HEAD(&c->list); | |
1133 | c->buf = malloc(GZ_CHUNK); | |
1134 | c->len = 0; | |
1135 | c->seq = seq; | |
97eabb2a JA |
1136 | return c; |
1137 | } | |
1138 | ||
1139 | static void free_chunk(struct iolog_compress *ic) | |
1140 | { | |
c07826cd JA |
1141 | free(ic->buf); |
1142 | free(ic); | |
97eabb2a JA |
1143 | } |
1144 | ||
b26317c9 | 1145 | static int z_stream_init(z_stream *stream, int gz_hdr) |
aee2ab67 | 1146 | { |
b26317c9 JA |
1147 | int wbits = 15; |
1148 | ||
acb2a69b | 1149 | memset(stream, 0, sizeof(*stream)); |
aee2ab67 JA |
1150 | stream->zalloc = Z_NULL; |
1151 | stream->zfree = Z_NULL; | |
1152 | stream->opaque = Z_NULL; | |
1153 | stream->next_in = Z_NULL; | |
1154 | ||
1a8e7458 JA |
1155 | /* |
1156 | * zlib magic - add 32 for auto-detection of gz header or not, | |
1157 | * if we decide to store files in a gzip friendly format. | |
1158 | */ | |
b26317c9 JA |
1159 | if (gz_hdr) |
1160 | wbits += 32; | |
1161 | ||
1162 | if (inflateInit2(stream, wbits) != Z_OK) | |
aee2ab67 JA |
1163 | return 1; |
1164 | ||
1165 | return 0; | |
1166 | } | |
1167 | ||
1a8e7458 | 1168 | struct inflate_chunk_iter { |
aee2ab67 | 1169 | unsigned int seq; |
dad95da1 | 1170 | int err; |
aee2ab67 JA |
1171 | void *buf; |
1172 | size_t buf_size; | |
1173 | size_t buf_used; | |
1174 | size_t chunk_sz; | |
1175 | }; | |
1176 | ||
b26317c9 | 1177 | static void finish_chunk(z_stream *stream, FILE *f, |
1a8e7458 | 1178 | struct inflate_chunk_iter *iter) |
aee2ab67 | 1179 | { |
aee2ab67 JA |
1180 | int ret; |
1181 | ||
1182 | ret = inflateEnd(stream); | |
1183 | if (ret != Z_OK) | |
50f940f1 JA |
1184 | log_err("fio: failed to end log inflation seq %d (%d)\n", |
1185 | iter->seq, ret); | |
aee2ab67 | 1186 | |
b26317c9 | 1187 | flush_samples(f, iter->buf, iter->buf_used); |
aee2ab67 JA |
1188 | free(iter->buf); |
1189 | iter->buf = NULL; | |
1190 | iter->buf_size = iter->buf_used = 0; | |
1191 | } | |
1192 | ||
1a8e7458 JA |
1193 | /* |
1194 | * Iterative chunk inflation. Handles cases where we cross into a new | |
1195 | * sequence, doing flush finish of previous chunk if needed. | |
1196 | */ | |
1197 | static size_t inflate_chunk(struct iolog_compress *ic, int gz_hdr, FILE *f, | |
1198 | z_stream *stream, struct inflate_chunk_iter *iter) | |
aee2ab67 | 1199 | { |
0c56718d JA |
1200 | size_t ret; |
1201 | ||
0f989d2e | 1202 | dprint(FD_COMPRESS, "inflate chunk size=%lu, seq=%u\n", |
0c56718d JA |
1203 | (unsigned long) ic->len, ic->seq); |
1204 | ||
aee2ab67 JA |
1205 | if (ic->seq != iter->seq) { |
1206 | if (iter->seq) | |
b26317c9 | 1207 | finish_chunk(stream, f, iter); |
aee2ab67 | 1208 | |
b26317c9 | 1209 | z_stream_init(stream, gz_hdr); |
aee2ab67 JA |
1210 | iter->seq = ic->seq; |
1211 | } | |
1212 | ||
1213 | stream->avail_in = ic->len; | |
1214 | stream->next_in = ic->buf; | |
1215 | ||
1216 | if (!iter->buf_size) { | |
1217 | iter->buf_size = iter->chunk_sz; | |
1218 | iter->buf = malloc(iter->buf_size); | |
1219 | } | |
1220 | ||
1221 | while (stream->avail_in) { | |
b26317c9 | 1222 | size_t this_out = iter->buf_size - iter->buf_used; |
aee2ab67 JA |
1223 | int err; |
1224 | ||
b26317c9 | 1225 | stream->avail_out = this_out; |
aee2ab67 JA |
1226 | stream->next_out = iter->buf + iter->buf_used; |
1227 | ||
1228 | err = inflate(stream, Z_NO_FLUSH); | |
1229 | if (err < 0) { | |
1230 | log_err("fio: failed inflating log: %d\n", err); | |
dad95da1 | 1231 | iter->err = err; |
aee2ab67 JA |
1232 | break; |
1233 | } | |
1234 | ||
b26317c9 JA |
1235 | iter->buf_used += this_out - stream->avail_out; |
1236 | ||
1237 | if (!stream->avail_out) { | |
1238 | iter->buf_size += iter->chunk_sz; | |
1239 | iter->buf = realloc(iter->buf, iter->buf_size); | |
1240 | continue; | |
1241 | } | |
1242 | ||
1243 | if (err == Z_STREAM_END) | |
1244 | break; | |
aee2ab67 JA |
1245 | } |
1246 | ||
0c56718d JA |
1247 | ret = (void *) stream->next_in - ic->buf; |
1248 | ||
d0d0cf0a | 1249 | dprint(FD_COMPRESS, "inflated to size=%lu\n", (unsigned long) iter->buf_size); |
0c56718d JA |
1250 | |
1251 | return ret; | |
aee2ab67 JA |
1252 | } |
1253 | ||
1a8e7458 JA |
1254 | /* |
1255 | * Inflate stored compressed chunks, or write them directly to the log | |
1256 | * file if so instructed. | |
1257 | */ | |
dad95da1 | 1258 | static int inflate_gz_chunks(struct io_log *log, FILE *f) |
aee2ab67 | 1259 | { |
1a8e7458 | 1260 | struct inflate_chunk_iter iter = { .chunk_sz = log->log_gz, }; |
aee2ab67 JA |
1261 | z_stream stream; |
1262 | ||
1263 | while (!flist_empty(&log->chunk_list)) { | |
1264 | struct iolog_compress *ic; | |
1265 | ||
9342d5f8 | 1266 | ic = flist_first_entry(&log->chunk_list, struct iolog_compress, list); |
aee2ab67 | 1267 | flist_del(&ic->list); |
b26317c9 | 1268 | |
1a8e7458 JA |
1269 | if (log->log_gz_store) { |
1270 | size_t ret; | |
1271 | ||
0c56718d JA |
1272 | dprint(FD_COMPRESS, "log write chunk size=%lu, " |
1273 | "seq=%u\n", (unsigned long) ic->len, ic->seq); | |
1274 | ||
1a8e7458 | 1275 | ret = fwrite(ic->buf, ic->len, 1, f); |
dad95da1 JA |
1276 | if (ret != 1 || ferror(f)) { |
1277 | iter.err = errno; | |
1a8e7458 | 1278 | log_err("fio: error writing compressed log\n"); |
dad95da1 | 1279 | } |
1a8e7458 JA |
1280 | } else |
1281 | inflate_chunk(ic, log->log_gz_store, f, &stream, &iter); | |
c07826cd JA |
1282 | |
1283 | free_chunk(ic); | |
aee2ab67 JA |
1284 | } |
1285 | ||
1286 | if (iter.seq) { | |
b26317c9 | 1287 | finish_chunk(&stream, f, &iter); |
aee2ab67 JA |
1288 | free(iter.buf); |
1289 | } | |
dad95da1 JA |
1290 | |
1291 | return iter.err; | |
aee2ab67 JA |
1292 | } |
1293 | ||
1a8e7458 JA |
1294 | /* |
1295 | * Open compressed log file and decompress the stored chunks and | |
1296 | * write them to stdout. The chunks are stored sequentially in the | |
1297 | * file, so we iterate over them and do them one-by-one. | |
1298 | */ | |
b26317c9 JA |
1299 | int iolog_file_inflate(const char *file) |
1300 | { | |
1a8e7458 | 1301 | struct inflate_chunk_iter iter = { .chunk_sz = 64 * 1024 * 1024, }; |
b26317c9 JA |
1302 | struct iolog_compress ic; |
1303 | z_stream stream; | |
1304 | struct stat sb; | |
2b8b8f0d | 1305 | size_t ret; |
f302710c JA |
1306 | size_t total; |
1307 | void *buf; | |
b26317c9 JA |
1308 | FILE *f; |
1309 | ||
a7c386ed | 1310 | f = fopen(file, "rb"); |
b26317c9 JA |
1311 | if (!f) { |
1312 | perror("fopen"); | |
1313 | return 1; | |
1314 | } | |
1315 | ||
fd771971 JA |
1316 | if (stat(file, &sb) < 0) { |
1317 | fclose(f); | |
1318 | perror("stat"); | |
1319 | return 1; | |
1320 | } | |
1321 | ||
f302710c | 1322 | ic.buf = buf = malloc(sb.st_size); |
b26317c9 | 1323 | ic.len = sb.st_size; |
b26317c9 JA |
1324 | ic.seq = 1; |
1325 | ||
1326 | ret = fread(ic.buf, ic.len, 1, f); | |
2b8b8f0d | 1327 | if (ret == 0 && ferror(f)) { |
b26317c9 JA |
1328 | perror("fread"); |
1329 | fclose(f); | |
bb1116fe | 1330 | free(buf); |
b26317c9 | 1331 | return 1; |
2b8b8f0d | 1332 | } else if (ferror(f) || (!feof(f) && ret != 1)) { |
b26317c9 JA |
1333 | log_err("fio: short read on reading log\n"); |
1334 | fclose(f); | |
bb1116fe | 1335 | free(buf); |
b26317c9 JA |
1336 | return 1; |
1337 | } | |
1338 | ||
1339 | fclose(f); | |
1340 | ||
1a8e7458 JA |
1341 | /* |
1342 | * Each chunk will return Z_STREAM_END. We don't know how many | |
1343 | * chunks are in the file, so we just keep looping and incrementing | |
1344 | * the sequence number until we have consumed the whole compressed | |
1345 | * file. | |
1346 | */ | |
f302710c JA |
1347 | total = ic.len; |
1348 | do { | |
8a1db9a1 | 1349 | size_t iret; |
f302710c | 1350 | |
8a1db9a1 JA |
1351 | iret = inflate_chunk(&ic, 1, stdout, &stream, &iter); |
1352 | total -= iret; | |
f302710c JA |
1353 | if (!total) |
1354 | break; | |
dad95da1 JA |
1355 | if (iter.err) |
1356 | break; | |
f302710c JA |
1357 | |
1358 | ic.seq++; | |
8a1db9a1 JA |
1359 | ic.len -= iret; |
1360 | ic.buf += iret; | |
f302710c | 1361 | } while (1); |
b26317c9 JA |
1362 | |
1363 | if (iter.seq) { | |
1364 | finish_chunk(&stream, stdout, &iter); | |
1365 | free(iter.buf); | |
1366 | } | |
1367 | ||
f302710c | 1368 | free(buf); |
dad95da1 | 1369 | return iter.err; |
b26317c9 JA |
1370 | } |
1371 | ||
aee2ab67 JA |
1372 | #else |
1373 | ||
dad95da1 | 1374 | static int inflate_gz_chunks(struct io_log *log, FILE *f) |
aee2ab67 | 1375 | { |
b7364e21 | 1376 | return 0; |
aee2ab67 JA |
1377 | } |
1378 | ||
a68c66b2 JA |
1379 | int iolog_file_inflate(const char *file) |
1380 | { | |
1381 | log_err("fio: log inflation not possible without zlib\n"); | |
1382 | return 1; | |
1383 | } | |
1384 | ||
aee2ab67 JA |
1385 | #endif |
1386 | ||
60a25727 | 1387 | void flush_log(struct io_log *log, bool do_append) |
aee2ab67 JA |
1388 | { |
1389 | void *buf; | |
1390 | FILE *f; | |
1391 | ||
a7c386ed | 1392 | /* |
1393 | * If log_gz_store is true, we are writing a binary file. | |
1394 | * Set the mode appropriately (on all platforms) to avoid issues | |
1395 | * on windows (line-ending conversions, etc.) | |
1396 | */ | |
3a5db920 | 1397 | if (!do_append) |
a7c386ed | 1398 | if (log->log_gz_store) |
1399 | f = fopen(log->filename, "wb"); | |
1400 | else | |
1401 | f = fopen(log->filename, "w"); | |
3a5db920 | 1402 | else |
a7c386ed | 1403 | if (log->log_gz_store) |
1404 | f = fopen(log->filename, "ab"); | |
1405 | else | |
1406 | f = fopen(log->filename, "a"); | |
aee2ab67 JA |
1407 | if (!f) { |
1408 | perror("fopen log"); | |
1409 | return; | |
1410 | } | |
1411 | ||
1412 | buf = set_file_buffer(f); | |
1413 | ||
1a8e7458 | 1414 | inflate_gz_chunks(log, f); |
aee2ab67 | 1415 | |
7e419452 JA |
1416 | while (!flist_empty(&log->io_logs)) { |
1417 | struct io_logs *cur_log; | |
1418 | ||
1419 | cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); | |
1420 | flist_del_init(&cur_log->list); | |
1e613c9c | 1421 | |
a9179eeb | 1422 | if (log->td && log == log->td->clat_hist_log) |
1e613c9c | 1423 | flush_hist_samples(f, log->hist_coarseness, cur_log->log, |
868f6f03 | 1424 | log_sample_sz(log, cur_log)); |
1e613c9c | 1425 | else |
868f6f03 | 1426 | flush_samples(f, cur_log->log, log_sample_sz(log, cur_log)); |
1e613c9c | 1427 | |
90d2d53f | 1428 | sfree(cur_log); |
7e419452 | 1429 | } |
ac9b9101 JA |
1430 | |
1431 | fclose(f); | |
2e802282 | 1432 | clear_file_buffer(buf); |
ac9b9101 JA |
1433 | } |
1434 | ||
cb7e0ace | 1435 | static int finish_log(struct thread_data *td, struct io_log *log, int trylock) |
ac9b9101 | 1436 | { |
155f2f02 | 1437 | if (td->flags & TD_F_COMPRESS_LOG) |
7e419452 | 1438 | iolog_flush(log); |
aee2ab67 | 1439 | |
243bfe19 | 1440 | if (trylock) { |
cb7e0ace | 1441 | if (fio_trylock_file(log->filename)) |
243bfe19 JA |
1442 | return 1; |
1443 | } else | |
cb7e0ace | 1444 | fio_lock_file(log->filename); |
243bfe19 | 1445 | |
0cba0f91 | 1446 | if (td->client_type == FIO_CLIENT_TYPE_GUI || is_backend) |
cb7e0ace | 1447 | fio_send_iolog(td, log, log->filename); |
aee2ab67 | 1448 | else |
3a5db920 | 1449 | flush_log(log, !td->o.per_job_logs); |
243bfe19 | 1450 | |
cb7e0ace | 1451 | fio_unlock_file(log->filename); |
518dac09 | 1452 | free_log(log); |
243bfe19 | 1453 | return 0; |
ac9b9101 JA |
1454 | } |
1455 | ||
0cba0f91 JA |
1456 | size_t log_chunk_sizes(struct io_log *log) |
1457 | { | |
1458 | struct flist_head *entry; | |
1459 | size_t ret; | |
1460 | ||
1461 | if (flist_empty(&log->chunk_list)) | |
1462 | return 0; | |
1463 | ||
1464 | ret = 0; | |
1465 | pthread_mutex_lock(&log->chunk_lock); | |
1466 | flist_for_each(entry, &log->chunk_list) { | |
1467 | struct iolog_compress *c; | |
1468 | ||
1469 | c = flist_entry(entry, struct iolog_compress, list); | |
1470 | ret += c->len; | |
1471 | } | |
1472 | pthread_mutex_unlock(&log->chunk_lock); | |
1473 | return ret; | |
1474 | } | |
1475 | ||
aee2ab67 JA |
1476 | #ifdef CONFIG_ZLIB |
1477 | ||
858bce70 JA |
1478 | static void iolog_put_deferred(struct io_log *log, void *ptr) |
1479 | { | |
1480 | if (!ptr) | |
1481 | return; | |
1482 | ||
1483 | pthread_mutex_lock(&log->deferred_free_lock); | |
1484 | if (log->deferred < IOLOG_MAX_DEFER) { | |
1485 | log->deferred_items[log->deferred] = ptr; | |
1486 | log->deferred++; | |
e2446166 | 1487 | } else if (!fio_did_warn(FIO_WARN_IOLOG_DROP)) |
858bce70 | 1488 | log_err("fio: had to drop log entry free\n"); |
858bce70 JA |
1489 | pthread_mutex_unlock(&log->deferred_free_lock); |
1490 | } | |
1491 | ||
1492 | static void iolog_free_deferred(struct io_log *log) | |
1493 | { | |
1494 | int i; | |
1495 | ||
1496 | if (!log->deferred) | |
1497 | return; | |
1498 | ||
1499 | pthread_mutex_lock(&log->deferred_free_lock); | |
1500 | ||
1501 | for (i = 0; i < log->deferred; i++) { | |
1502 | free(log->deferred_items[i]); | |
1503 | log->deferred_items[i] = NULL; | |
1504 | } | |
1505 | ||
1506 | log->deferred = 0; | |
1507 | pthread_mutex_unlock(&log->deferred_free_lock); | |
1508 | } | |
1509 | ||
78e93aa6 | 1510 | static int gz_work(struct iolog_flush_data *data) |
c143980a | 1511 | { |
040238ec | 1512 | struct iolog_compress *c = NULL; |
aee2ab67 JA |
1513 | struct flist_head list; |
1514 | unsigned int seq; | |
1515 | z_stream stream; | |
1516 | size_t total = 0; | |
d73ac887 | 1517 | int ret; |
aee2ab67 JA |
1518 | |
1519 | INIT_FLIST_HEAD(&list); | |
1520 | ||
7e419452 | 1521 | memset(&stream, 0, sizeof(stream)); |
aee2ab67 JA |
1522 | stream.zalloc = Z_NULL; |
1523 | stream.zfree = Z_NULL; | |
1524 | stream.opaque = Z_NULL; | |
1525 | ||
b26317c9 | 1526 | ret = deflateInit(&stream, Z_DEFAULT_COMPRESSION); |
b26317c9 | 1527 | if (ret != Z_OK) { |
aee2ab67 | 1528 | log_err("fio: failed to init gz stream\n"); |
152ea035 | 1529 | goto err; |
aee2ab67 JA |
1530 | } |
1531 | ||
1532 | seq = ++data->log->chunk_seq; | |
b26317c9 | 1533 | |
aee2ab67 JA |
1534 | stream.next_in = (void *) data->samples; |
1535 | stream.avail_in = data->nr_samples * log_entry_sz(data->log); | |
1536 | ||
d0d0cf0a JA |
1537 | dprint(FD_COMPRESS, "deflate input size=%lu, seq=%u, log=%s\n", |
1538 | (unsigned long) stream.avail_in, seq, | |
1539 | data->log->filename); | |
aee2ab67 | 1540 | do { |
040238ec | 1541 | if (c) |
0caccfa7 JA |
1542 | dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, |
1543 | (unsigned long) c->len); | |
aee2ab67 JA |
1544 | c = get_new_chunk(seq); |
1545 | stream.avail_out = GZ_CHUNK; | |
1546 | stream.next_out = c->buf; | |
1547 | ret = deflate(&stream, Z_NO_FLUSH); | |
1548 | if (ret < 0) { | |
1549 | log_err("fio: deflate log (%d)\n", ret); | |
dad95da1 JA |
1550 | free_chunk(c); |
1551 | goto err; | |
aee2ab67 JA |
1552 | } |
1553 | ||
1554 | c->len = GZ_CHUNK - stream.avail_out; | |
1555 | flist_add_tail(&c->list, &list); | |
1556 | total += c->len; | |
1557 | } while (stream.avail_in); | |
1558 | ||
1559 | stream.next_out = c->buf + c->len; | |
1560 | stream.avail_out = GZ_CHUNK - c->len; | |
1561 | ||
1562 | ret = deflate(&stream, Z_FINISH); | |
6fa3ad51 JA |
1563 | if (ret < 0) { |
1564 | /* | |
1565 | * Z_BUF_ERROR is special, it just means we need more | |
1566 | * output space. We'll handle that below. Treat any other | |
1567 | * error as fatal. | |
1568 | */ | |
1569 | if (ret != Z_BUF_ERROR) { | |
1570 | log_err("fio: deflate log (%d)\n", ret); | |
1571 | flist_del(&c->list); | |
1572 | free_chunk(c); | |
1573 | goto err; | |
1574 | } | |
1575 | } | |
1576 | ||
1577 | total -= c->len; | |
1578 | c->len = GZ_CHUNK - stream.avail_out; | |
1579 | total += c->len; | |
0caccfa7 | 1580 | dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, (unsigned long) c->len); |
040238ec | 1581 | |
6fa3ad51 | 1582 | if (ret != Z_STREAM_END) { |
aee2ab67 JA |
1583 | do { |
1584 | c = get_new_chunk(seq); | |
1585 | stream.avail_out = GZ_CHUNK; | |
1586 | stream.next_out = c->buf; | |
1587 | ret = deflate(&stream, Z_FINISH); | |
1588 | c->len = GZ_CHUNK - stream.avail_out; | |
0c56718d | 1589 | total += c->len; |
aee2ab67 | 1590 | flist_add_tail(&c->list, &list); |
0caccfa7 JA |
1591 | dprint(FD_COMPRESS, "seq=%d, chunk=%lu\n", seq, |
1592 | (unsigned long) c->len); | |
aee2ab67 JA |
1593 | } while (ret != Z_STREAM_END); |
1594 | } | |
1595 | ||
0c56718d JA |
1596 | dprint(FD_COMPRESS, "deflated to size=%lu\n", (unsigned long) total); |
1597 | ||
aee2ab67 JA |
1598 | ret = deflateEnd(&stream); |
1599 | if (ret != Z_OK) | |
1600 | log_err("fio: deflateEnd %d\n", ret); | |
1601 | ||
858bce70 | 1602 | iolog_put_deferred(data->log, data->samples); |
aee2ab67 JA |
1603 | |
1604 | if (!flist_empty(&list)) { | |
1605 | pthread_mutex_lock(&data->log->chunk_lock); | |
1606 | flist_splice_tail(&list, &data->log->chunk_list); | |
1607 | pthread_mutex_unlock(&data->log->chunk_lock); | |
1608 | } | |
1609 | ||
dad95da1 JA |
1610 | ret = 0; |
1611 | done: | |
7e419452 | 1612 | if (data->free) |
dab41b1c | 1613 | sfree(data); |
dad95da1 JA |
1614 | return ret; |
1615 | err: | |
1616 | while (!flist_empty(&list)) { | |
1617 | c = flist_first_entry(list.next, struct iolog_compress, list); | |
1618 | flist_del(&c->list); | |
1619 | free_chunk(c); | |
1620 | } | |
1621 | ret = 1; | |
1622 | goto done; | |
aee2ab67 JA |
1623 | } |
1624 | ||
78e93aa6 JA |
1625 | /* |
1626 | * Invoked from our compress helper thread, when logging would have exceeded | |
1627 | * the specified memory limitation. Compresses the previously stored | |
1628 | * entries. | |
1629 | */ | |
1630 | static int gz_work_async(struct submit_worker *sw, struct workqueue_work *work) | |
1631 | { | |
1632 | return gz_work(container_of(work, struct iolog_flush_data, work)); | |
1633 | } | |
1634 | ||
c08f9fe2 JA |
1635 | static int gz_init_worker(struct submit_worker *sw) |
1636 | { | |
1637 | struct thread_data *td = sw->wq->td; | |
1638 | ||
1639 | if (!fio_option_is_set(&td->o, log_gz_cpumask)) | |
1640 | return 0; | |
1641 | ||
1642 | if (fio_setaffinity(gettid(), td->o.log_gz_cpumask) == -1) { | |
1643 | log_err("gz: failed to set CPU affinity\n"); | |
1644 | return 1; | |
1645 | } | |
1646 | ||
1647 | return 0; | |
1648 | } | |
1649 | ||
155f2f02 | 1650 | static struct workqueue_ops log_compress_wq_ops = { |
78e93aa6 | 1651 | .fn = gz_work_async, |
c08f9fe2 | 1652 | .init_worker_fn = gz_init_worker, |
24660963 | 1653 | .nice = 1, |
155f2f02 JA |
1654 | }; |
1655 | ||
24660963 | 1656 | int iolog_compress_init(struct thread_data *td, struct sk_out *sk_out) |
155f2f02 JA |
1657 | { |
1658 | if (!(td->flags & TD_F_COMPRESS_LOG)) | |
1659 | return 0; | |
1660 | ||
24660963 | 1661 | workqueue_init(td, &td->log_compress_wq, &log_compress_wq_ops, 1, sk_out); |
155f2f02 JA |
1662 | return 0; |
1663 | } | |
1664 | ||
1665 | void iolog_compress_exit(struct thread_data *td) | |
1666 | { | |
1667 | if (!(td->flags & TD_F_COMPRESS_LOG)) | |
1668 | return; | |
1669 | ||
1670 | workqueue_exit(&td->log_compress_wq); | |
1671 | } | |
1672 | ||
1a8e7458 | 1673 | /* |
949ae6dc JA |
1674 | * Queue work item to compress the existing log entries. We reset the |
1675 | * current log to a small size, and reference the existing log in the | |
1676 | * data that we queue for compression. Once compression has been done, | |
03900b0b | 1677 | * this old log is freed. Will not return until the log compression |
1678 | * has completed, and will flush all previous logs too | |
1a8e7458 | 1679 | */ |
7e419452 | 1680 | static int iolog_flush(struct io_log *log) |
aee2ab67 | 1681 | { |
aee2ab67 | 1682 | struct iolog_flush_data *data; |
aee2ab67 | 1683 | |
03900b0b | 1684 | workqueue_flush(&log->td->log_compress_wq); |
aee2ab67 JA |
1685 | data = malloc(sizeof(*data)); |
1686 | if (!data) | |
1687 | return 1; | |
1688 | ||
1689 | data->log = log; | |
7e419452 | 1690 | data->free = false; |
aee2ab67 | 1691 | |
7e419452 JA |
1692 | while (!flist_empty(&log->io_logs)) { |
1693 | struct io_logs *cur_log; | |
949ae6dc | 1694 | |
7e419452 JA |
1695 | cur_log = flist_first_entry(&log->io_logs, struct io_logs, list); |
1696 | flist_del_init(&cur_log->list); | |
1697 | ||
1698 | data->samples = cur_log->log; | |
1699 | data->nr_samples = cur_log->nr_samples; | |
1700 | ||
2b7625e2 | 1701 | sfree(cur_log); |
aee2ab67 | 1702 | |
78e93aa6 | 1703 | gz_work(data); |
7e419452 | 1704 | } |
aee2ab67 | 1705 | |
7e419452 | 1706 | free(data); |
aee2ab67 JA |
1707 | return 0; |
1708 | } | |
1709 | ||
7e419452 JA |
1710 | int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log) |
1711 | { | |
1712 | struct iolog_flush_data *data; | |
1713 | ||
dab41b1c | 1714 | data = smalloc(sizeof(*data)); |
7e419452 JA |
1715 | if (!data) |
1716 | return 1; | |
1717 | ||
1718 | data->log = log; | |
1719 | ||
1720 | data->samples = cur_log->log; | |
1721 | data->nr_samples = cur_log->nr_samples; | |
1722 | data->free = true; | |
1723 | ||
1724 | cur_log->nr_samples = cur_log->max_samples = 0; | |
1725 | cur_log->log = NULL; | |
1726 | ||
1727 | workqueue_enqueue(&log->td->log_compress_wq, &data->work); | |
858bce70 JA |
1728 | |
1729 | iolog_free_deferred(log); | |
1730 | ||
7e419452 JA |
1731 | return 0; |
1732 | } | |
aee2ab67 JA |
1733 | #else |
1734 | ||
7e419452 JA |
1735 | static int iolog_flush(struct io_log *log) |
1736 | { | |
1737 | return 1; | |
1738 | } | |
1739 | ||
1740 | int iolog_cur_flush(struct io_log *log, struct io_logs *cur_log) | |
aee2ab67 JA |
1741 | { |
1742 | return 1; | |
1743 | } | |
1744 | ||
24660963 | 1745 | int iolog_compress_init(struct thread_data *td, struct sk_out *sk_out) |
52618a2a JA |
1746 | { |
1747 | return 0; | |
1748 | } | |
1749 | ||
1750 | void iolog_compress_exit(struct thread_data *td) | |
1751 | { | |
1752 | } | |
1753 | ||
aee2ab67 JA |
1754 | #endif |
1755 | ||
7e419452 JA |
1756 | struct io_logs *iolog_cur_log(struct io_log *log) |
1757 | { | |
1758 | if (flist_empty(&log->io_logs)) | |
1759 | return NULL; | |
1760 | ||
1761 | return flist_last_entry(&log->io_logs, struct io_logs, list); | |
1762 | } | |
1763 | ||
1764 | uint64_t iolog_nr_samples(struct io_log *iolog) | |
1765 | { | |
1766 | struct flist_head *entry; | |
1767 | uint64_t ret = 0; | |
1768 | ||
1769 | flist_for_each(entry, &iolog->io_logs) { | |
1770 | struct io_logs *cur_log; | |
1771 | ||
1772 | cur_log = flist_entry(entry, struct io_logs, list); | |
1773 | ret += cur_log->nr_samples; | |
1774 | } | |
1775 | ||
1776 | return ret; | |
1777 | } | |
1778 | ||
1c725e70 | 1779 | static int __write_log(struct thread_data *td, struct io_log *log, int try) |
905e3d4f | 1780 | { |
1c725e70 JA |
1781 | if (log) |
1782 | return finish_log(td, log, try); | |
905e3d4f | 1783 | |
1c725e70 JA |
1784 | return 0; |
1785 | } | |
905e3d4f | 1786 | |
a47591e4 | 1787 | static int write_iops_log(struct thread_data *td, int try, bool unit_log) |
1c725e70 | 1788 | { |
a47591e4 JA |
1789 | int ret; |
1790 | ||
1791 | if (per_unit_log(td->iops_log) != unit_log) | |
1792 | return 0; | |
1793 | ||
1794 | ret = __write_log(td, td->iops_log, try); | |
1795 | if (!ret) | |
1796 | td->iops_log = NULL; | |
1797 | ||
1798 | return ret; | |
905e3d4f JA |
1799 | } |
1800 | ||
a47591e4 | 1801 | static int write_slat_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1802 | { |
a47591e4 JA |
1803 | int ret; |
1804 | ||
1805 | if (!unit_log) | |
1806 | return 0; | |
1807 | ||
1808 | ret = __write_log(td, td->slat_log, try); | |
1809 | if (!ret) | |
1810 | td->slat_log = NULL; | |
1811 | ||
1812 | return ret; | |
905e3d4f JA |
1813 | } |
1814 | ||
a47591e4 | 1815 | static int write_clat_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1816 | { |
a47591e4 JA |
1817 | int ret; |
1818 | ||
1819 | if (!unit_log) | |
1820 | return 0; | |
1821 | ||
1822 | ret = __write_log(td, td->clat_log, try); | |
1823 | if (!ret) | |
1824 | td->clat_log = NULL; | |
1825 | ||
1826 | return ret; | |
905e3d4f JA |
1827 | } |
1828 | ||
1e613c9c KC |
1829 | static int write_clat_hist_log(struct thread_data *td, int try, bool unit_log) |
1830 | { | |
1831 | int ret; | |
1832 | ||
1833 | if (!unit_log) | |
1834 | return 0; | |
1835 | ||
1836 | ret = __write_log(td, td->clat_hist_log, try); | |
1837 | if (!ret) | |
1838 | td->clat_hist_log = NULL; | |
1839 | ||
1840 | return ret; | |
1841 | } | |
1842 | ||
a47591e4 | 1843 | static int write_lat_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1844 | { |
a47591e4 JA |
1845 | int ret; |
1846 | ||
1847 | if (!unit_log) | |
1848 | return 0; | |
1849 | ||
1850 | ret = __write_log(td, td->lat_log, try); | |
1851 | if (!ret) | |
1852 | td->lat_log = NULL; | |
1853 | ||
1854 | return ret; | |
905e3d4f JA |
1855 | } |
1856 | ||
a47591e4 | 1857 | static int write_bandw_log(struct thread_data *td, int try, bool unit_log) |
905e3d4f | 1858 | { |
a47591e4 JA |
1859 | int ret; |
1860 | ||
1861 | if (per_unit_log(td->bw_log) != unit_log) | |
1862 | return 0; | |
1863 | ||
1864 | ret = __write_log(td, td->bw_log, try); | |
1865 | if (!ret) | |
1866 | td->bw_log = NULL; | |
1867 | ||
1868 | return ret; | |
905e3d4f JA |
1869 | } |
1870 | ||
1871 | enum { | |
1872 | BW_LOG_MASK = 1, | |
1873 | LAT_LOG_MASK = 2, | |
1874 | SLAT_LOG_MASK = 4, | |
1875 | CLAT_LOG_MASK = 8, | |
1876 | IOPS_LOG_MASK = 16, | |
1e613c9c | 1877 | CLAT_HIST_LOG_MASK = 32, |
905e3d4f | 1878 | |
1e613c9c | 1879 | ALL_LOG_NR = 6, |
905e3d4f JA |
1880 | }; |
1881 | ||
1882 | struct log_type { | |
1883 | unsigned int mask; | |
a47591e4 | 1884 | int (*fn)(struct thread_data *, int, bool); |
905e3d4f JA |
1885 | }; |
1886 | ||
1887 | static struct log_type log_types[] = { | |
1888 | { | |
1889 | .mask = BW_LOG_MASK, | |
1890 | .fn = write_bandw_log, | |
1891 | }, | |
1892 | { | |
1893 | .mask = LAT_LOG_MASK, | |
1894 | .fn = write_lat_log, | |
1895 | }, | |
1896 | { | |
1897 | .mask = SLAT_LOG_MASK, | |
1898 | .fn = write_slat_log, | |
1899 | }, | |
1900 | { | |
1901 | .mask = CLAT_LOG_MASK, | |
1902 | .fn = write_clat_log, | |
1903 | }, | |
1904 | { | |
1905 | .mask = IOPS_LOG_MASK, | |
1906 | .fn = write_iops_log, | |
1907 | }, | |
1e613c9c KC |
1908 | { |
1909 | .mask = CLAT_HIST_LOG_MASK, | |
1910 | .fn = write_clat_hist_log, | |
1911 | } | |
905e3d4f JA |
1912 | }; |
1913 | ||
a47591e4 | 1914 | void td_writeout_logs(struct thread_data *td, bool unit_logs) |
905e3d4f | 1915 | { |
ea5409f9 | 1916 | unsigned int log_mask = 0; |
905e3d4f JA |
1917 | unsigned int log_left = ALL_LOG_NR; |
1918 | int old_state, i; | |
1919 | ||
1920 | old_state = td_bump_runstate(td, TD_FINISHING); | |
1921 | ||
a47591e4 | 1922 | finalize_logs(td, unit_logs); |
905e3d4f JA |
1923 | |
1924 | while (log_left) { | |
1925 | int prev_log_left = log_left; | |
1926 | ||
1927 | for (i = 0; i < ALL_LOG_NR && log_left; i++) { | |
1928 | struct log_type *lt = &log_types[i]; | |
1929 | int ret; | |
1930 | ||
ea5409f9 | 1931 | if (!(log_mask & lt->mask)) { |
a47591e4 | 1932 | ret = lt->fn(td, log_left != 1, unit_logs); |
905e3d4f JA |
1933 | if (!ret) { |
1934 | log_left--; | |
ea5409f9 | 1935 | log_mask |= lt->mask; |
905e3d4f JA |
1936 | } |
1937 | } | |
1938 | } | |
1939 | ||
1940 | if (prev_log_left == log_left) | |
1941 | usleep(5000); | |
1942 | } | |
1943 | ||
1944 | td_restore_runstate(td, old_state); | |
1945 | } | |
a47591e4 JA |
1946 | |
1947 | void fio_writeout_logs(bool unit_logs) | |
1948 | { | |
da8f124f | 1949 | for_each_td(td) { |
a47591e4 | 1950 | td_writeout_logs(td, unit_logs); |
da8f124f | 1951 | } end_for_each(); |
a47591e4 | 1952 | } |