Calloc() cleanup
[fio.git] / blktrace.c
CommitLineData
fb7b71a3
JA
1/*
2 * blktrace support code for fio
3 */
4#include <stdio.h>
5#include <stdlib.h>
5e6c2067
JA
6#include <sys/stat.h>
7#include <dirent.h>
8c1fdf04 8
01743ee1 9#include "flist.h"
fb7b71a3
JA
10#include "fio.h"
11#include "blktrace_api.h"
12
2da7df1d 13#define TRACE_FIFO_SIZE 8192
e2887563
JA
14
15/*
16 * fifo refill frontend, to avoid reading data in trace sized bites
17 */
18static int refill_fifo(struct thread_data *td, struct fifo *fifo, int fd)
19{
20 char buf[TRACE_FIFO_SIZE];
f12b323f 21 unsigned int total;
e2887563
JA
22 int ret;
23
f12b323f
JA
24 total = sizeof(buf);
25 if (total > fifo_room(fifo))
26 total = fifo_room(fifo);
e2887563 27
f12b323f
JA
28 ret = read(fd, buf, total);
29 if (ret < 0) {
30 td_verror(td, errno, "read blktrace file");
31 return -1;
e2887563
JA
32 }
33
f12b323f
JA
34 if (ret > 0)
35 ret = fifo_put(fifo, buf, ret);
36
bd6f78b2 37 dprint(FD_BLKTRACE, "refill: filled %d bytes\n", ret);
f12b323f 38 return ret;
e2887563
JA
39}
40
41/*
42 * Retrieve 'len' bytes from the fifo, refilling if necessary.
43 */
44static int trace_fifo_get(struct thread_data *td, struct fifo *fifo, int fd,
45 void *buf, unsigned int len)
46{
f12b323f
JA
47 if (fifo_len(fifo) < len) {
48 int ret = refill_fifo(td, fifo, fd);
e2887563 49
f12b323f
JA
50 if (ret < 0)
51 return ret;
52 }
e2887563
JA
53
54 return fifo_get(fifo, buf, len);
55}
56
8c1fdf04
JA
57/*
58 * Just discard the pdu by seeking past it.
59 */
f12b323f
JA
60static int discard_pdu(struct thread_data *td, struct fifo *fifo, int fd,
61 struct blk_io_trace *t)
fb7b71a3
JA
62{
63 if (t->pdu_len == 0)
64 return 0;
65
bd6f78b2 66 dprint(FD_BLKTRACE, "discard pdu len %u\n", t->pdu_len);
f12b323f 67 return trace_fifo_get(td, fifo, fd, NULL, t->pdu_len);
fb7b71a3
JA
68}
69
8c1fdf04
JA
70/*
71 * Check if this is a blktrace binary data file. We read a single trace
72 * into memory and check for the magic signature.
73 */
d95b34a6 74int is_blktrace(const char *filename, int *need_swap)
fb7b71a3
JA
75{
76 struct blk_io_trace t;
77 int fd, ret;
78
79 fd = open(filename, O_RDONLY);
4dced407 80 if (fd < 0)
fb7b71a3 81 return 0;
fb7b71a3
JA
82
83 ret = read(fd, &t, sizeof(t));
84 close(fd);
85
86 if (ret < 0) {
87 perror("read blktrace");
88 return 0;
89 } else if (ret != sizeof(t)) {
90 log_err("fio: short read on blktrace file\n");
91 return 0;
92 }
93
d95b34a6
JA
94 if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) {
95 *need_swap = 0;
fb7b71a3 96 return 1;
d95b34a6
JA
97 }
98
99 /*
100 * Maybe it needs to be endian swapped...
101 */
102 t.magic = fio_swap32(t.magic);
103 if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) {
104 *need_swap = 1;
105 return 1;
106 }
fb7b71a3
JA
107
108 return 0;
109}
110
d1c46c04
DN
111static int lookup_device(struct thread_data *td, char *path, unsigned int maj,
112 unsigned int min)
5e6c2067
JA
113{
114 struct dirent *dir;
115 struct stat st;
116 int found = 0;
117 DIR *D;
118
119 D = opendir(path);
120 if (!D)
121 return 0;
122
123 while ((dir = readdir(D)) != NULL) {
124 char full_path[256];
125
126 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
127 continue;
128
b9fd788f 129 sprintf(full_path, "%s%s%s", path, FIO_OS_PATH_SEPARATOR, dir->d_name);
5e6c2067
JA
130 if (lstat(full_path, &st) == -1) {
131 perror("lstat");
132 break;
133 }
134
135 if (S_ISDIR(st.st_mode)) {
d1c46c04 136 found = lookup_device(td, full_path, maj, min);
5e6c2067
JA
137 if (found) {
138 strcpy(path, full_path);
139 break;
140 }
141 }
142
143 if (!S_ISBLK(st.st_mode))
144 continue;
145
d1c46c04
DN
146 /*
147 * If replay_redirect is set then always return this device
148 * upon lookup which overrides the device lookup based on
149 * major minor in the actual blktrace
150 */
151 if (td->o.replay_redirect) {
152 dprint(FD_BLKTRACE, "device lookup: %d/%d\n overridden"
153 " with: %s", maj, min,
154 td->o.replay_redirect);
155 strcpy(path, td->o.replay_redirect);
156 found = 1;
157 break;
158 }
159
5e6c2067 160 if (maj == major(st.st_rdev) && min == minor(st.st_rdev)) {
5ec10eaa 161 dprint(FD_BLKTRACE, "device lookup: %d/%d\n", maj, min);
5e6c2067
JA
162 strcpy(path, full_path);
163 found = 1;
164 break;
165 }
166 }
167
168 closedir(D);
169 return found;
170}
171
c69aa91f
JA
172#define FMINORBITS 20
173#define FMINORMASK ((1U << FMINORBITS) - 1)
174#define FMAJOR(dev) ((unsigned int) ((dev) >> FMINORBITS))
175#define FMINOR(dev) ((unsigned int) ((dev) & FMINORMASK))
eeb9c2aa 176
89ac1d48 177static void trace_add_open_close_event(struct thread_data *td, int fileno, enum file_log_act action)
691c8fb0
JA
178{
179 struct io_piece *ipo;
180
181 ipo = calloc(1, sizeof(*ipo));
0d29de83 182 init_ipo(ipo);
691c8fb0
JA
183
184 ipo->ddir = DDIR_INVAL;
185 ipo->fileno = fileno;
89ac1d48 186 ipo->file_action = action;
01743ee1 187 flist_add_tail(&ipo->list, &td->io_log_list);
691c8fb0
JA
188}
189
89ac1d48 190static int trace_add_file(struct thread_data *td, __u32 device)
5e6c2067 191{
89ac1d48 192 static unsigned int last_maj, last_min, last_fileno;
c69aa91f
JA
193 unsigned int maj = FMAJOR(device);
194 unsigned int min = FMINOR(device);
5e6c2067
JA
195 struct fio_file *f;
196 char dev[256];
197 unsigned int i;
198
199 if (last_maj == maj && last_min == min)
89ac1d48 200 return last_fileno;
5e6c2067
JA
201
202 last_maj = maj;
203 last_min = min;
204
205 /*
206 * check for this file in our list
207 */
208 for_each_file(td, f, i)
89ac1d48
SL
209 if (f->major == maj && f->minor == min) {
210 last_fileno = f->fileno;
211 return last_fileno;
212 }
5e6c2067
JA
213
214 strcpy(dev, "/dev");
d1c46c04 215 if (lookup_device(td, dev, maj, min)) {
691c8fb0
JA
216 int fileno;
217
bd6f78b2 218 dprint(FD_BLKTRACE, "add devices %s\n", dev);
49ffb4a2 219 fileno = add_file_exclusive(td, dev);
89ac1d48
SL
220 trace_add_open_close_event(td, fileno, FIO_LOG_OPEN_FILE);
221 last_fileno = fileno;
bd6f78b2 222 }
f01b34ae 223
89ac1d48 224 return last_fileno;
5e6c2067
JA
225}
226
8c1fdf04
JA
227/*
228 * Store blk_io_trace data in an ipo for later retrieval.
229 */
fdefd987 230static void store_ipo(struct thread_data *td, unsigned long long offset,
89ac1d48
SL
231 unsigned int bytes, int rw, unsigned long long ttime,
232 int fileno)
fdefd987
JA
233{
234 struct io_piece *ipo = malloc(sizeof(*ipo));
235
0d29de83
JA
236 init_ipo(ipo);
237
a2eea81b
JA
238 /*
239 * the 512 is wrong here, it should be the hardware sector size...
240 */
241 ipo->offset = offset * 512;
fdefd987 242 ipo->len = bytes;
8c1fdf04 243 ipo->delay = ttime / 1000;
fdefd987
JA
244 if (rw)
245 ipo->ddir = DDIR_WRITE;
246 else
247 ipo->ddir = DDIR_READ;
89ac1d48 248 ipo->fileno = fileno;
fdefd987 249
bd6f78b2
JA
250 dprint(FD_BLKTRACE, "store ddir=%d, off=%llu, len=%lu, delay=%lu\n",
251 ipo->ddir, ipo->offset,
252 ipo->len, ipo->delay);
691c8fb0 253 queue_io_piece(td, ipo);
fdefd987
JA
254}
255
0b9d69ec 256static void handle_trace_notify(struct blk_io_trace *t)
cd991b9e 257{
691c8fb0
JA
258 switch (t->action) {
259 case BLK_TN_PROCESS:
d95b34a6
JA
260 log_info("blktrace: got process notify: %x, %d\n",
261 t->action, t->pid);
691c8fb0
JA
262 break;
263 case BLK_TN_TIMESTAMP:
d95b34a6
JA
264 log_info("blktrace: got timestamp notify: %x, %d\n",
265 t->action, t->pid);
691c8fb0 266 break;
ff58fced
JA
267 case BLK_TN_MESSAGE:
268 break;
691c8fb0
JA
269 default:
270 dprint(FD_BLKTRACE, "unknown trace act %x\n", t->action);
271 break;
272 }
273}
5b3023b8 274
ff58fced
JA
275static void handle_trace_discard(struct thread_data *td, struct blk_io_trace *t,
276 unsigned long long ttime, unsigned long *ios)
277{
278 struct io_piece *ipo = malloc(sizeof(*ipo));
89ac1d48 279 int fileno;
ff58fced 280
0d29de83 281 init_ipo(ipo);
89ac1d48 282 fileno = trace_add_file(td, t->device);
ff58fced
JA
283
284 ios[DDIR_WRITE]++;
285 td->o.size += t->bytes;
286
287 memset(ipo, 0, sizeof(*ipo));
288 INIT_FLIST_HEAD(&ipo->list);
289
290 /*
291 * the 512 is wrong here, it should be the hardware sector size...
292 */
293 ipo->offset = t->sector * 512;
294 ipo->len = t->bytes;
295 ipo->delay = ttime / 1000;
296 ipo->ddir = DDIR_TRIM;
89ac1d48 297 ipo->fileno = fileno;
ff58fced
JA
298
299 dprint(FD_BLKTRACE, "store discard, off=%llu, len=%lu, delay=%lu\n",
300 ipo->offset, ipo->len,
301 ipo->delay);
302 queue_io_piece(td, ipo);
303}
304
691c8fb0
JA
305static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t,
306 unsigned long long ttime, unsigned long *ios,
307 unsigned int *bs)
308{
309 int rw;
89ac1d48 310 int fileno;
5b3023b8 311
89ac1d48 312 fileno = trace_add_file(td, t->device);
5b3023b8
JA
313
314 rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
315
316 if (t->bytes > bs[rw])
317 bs[rw] = t->bytes;
318
319 ios[rw]++;
320 td->o.size += t->bytes;
89ac1d48 321 store_ipo(td, t->sector, t->bytes, rw, ttime, fileno);
cd991b9e
JA
322}
323
691c8fb0
JA
324/*
325 * We only care for queue traces, most of the others are side effects
326 * due to internal workings of the block layer.
327 */
328static void handle_trace(struct thread_data *td, struct blk_io_trace *t,
329 unsigned long long ttime, unsigned long *ios,
330 unsigned int *bs)
331{
332 if ((t->action & 0xffff) != __BLK_TA_QUEUE)
333 return;
334 if (t->action & BLK_TC_ACT(BLK_TC_PC))
335 return;
336
337 if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
aec2de20 338 handle_trace_notify(t);
ff58fced
JA
339 else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
340 handle_trace_discard(td, t, ttime, ios);
691c8fb0
JA
341 else
342 handle_trace_fs(td, t, ttime, ios, bs);
343}
344
d95b34a6
JA
345static void byteswap_trace(struct blk_io_trace *t)
346{
347 t->magic = fio_swap32(t->magic);
348 t->sequence = fio_swap32(t->sequence);
349 t->time = fio_swap64(t->time);
350 t->sector = fio_swap64(t->sector);
351 t->bytes = fio_swap32(t->bytes);
352 t->action = fio_swap32(t->action);
353 t->pid = fio_swap32(t->pid);
354 t->device = fio_swap32(t->device);
355 t->cpu = fio_swap32(t->cpu);
356 t->error = fio_swap16(t->error);
357 t->pdu_len = fio_swap16(t->pdu_len);
358}
359
8c1fdf04
JA
360/*
361 * Load a blktrace file by reading all the blk_io_trace entries, and storing
362 * them as io_pieces like the fio text version would do.
363 */
d95b34a6 364int load_blktrace(struct thread_data *td, const char *filename, int need_swap)
fb7b71a3 365{
a61eddec 366 unsigned long long ttime, delay;
fb7b71a3 367 struct blk_io_trace t;
4241ea8f 368 unsigned long ios[2], skipped_writes;
a61eddec 369 unsigned int cpu;
d84f8d49 370 unsigned int rw_bs[2];
e2887563 371 struct fifo *fifo;
89ac1d48
SL
372 int fd, i;
373 struct fio_file *f;
fb7b71a3
JA
374
375 fd = open(filename, O_RDONLY);
376 if (fd < 0) {
377 td_verror(td, errno, "open blktrace file");
378 return 1;
379 }
380
e2887563
JA
381 fifo = fifo_alloc(TRACE_FIFO_SIZE);
382
6df8adaa
JA
383 td->o.size = 0;
384
a61eddec 385 cpu = 0;
d84f8d49
JA
386 ttime = 0;
387 ios[0] = ios[1] = 0;
388 rw_bs[0] = rw_bs[1] = 0;
4241ea8f 389 skipped_writes = 0;
fb7b71a3 390 do {
e2887563 391 int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t));
fb7b71a3 392
e2887563 393 if (ret < 0)
8c1fdf04 394 goto err;
e2887563
JA
395 else if (!ret)
396 break;
397 else if (ret < (int) sizeof(t)) {
398 log_err("fio: short fifo get\n");
fb7b71a3 399 break;
fb7b71a3
JA
400 }
401
d95b34a6
JA
402 if (need_swap)
403 byteswap_trace(&t);
404
fb7b71a3 405 if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
5ec10eaa
JA
406 log_err("fio: bad magic in blktrace data: %x\n",
407 t.magic);
8c1fdf04 408 goto err;
fb7b71a3
JA
409 }
410 if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
5ec10eaa
JA
411 log_err("fio: bad blktrace version %d\n",
412 t.magic & 0xff);
8c1fdf04 413 goto err;
fb7b71a3 414 }
f12b323f
JA
415 ret = discard_pdu(td, fifo, fd, &t);
416 if (ret < 0) {
fb7b71a3 417 td_verror(td, ret, "blktrace lseek");
8c1fdf04 418 goto err;
f12b323f
JA
419 } else if (t.pdu_len != ret) {
420 log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
421 goto err;
fb7b71a3 422 }
691c8fb0
JA
423 if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) {
424 if (!ttime) {
425 ttime = t.time;
426 cpu = t.cpu;
427 }
428
429 delay = 0;
430 if (cpu == t.cpu)
431 delay = t.time - ttime;
432 if ((t.action & BLK_TC_ACT(BLK_TC_WRITE)) && read_only)
433 skipped_writes++;
64bbb865
DN
434 else {
435 /*
436 * set delay to zero if no_stall enabled for
437 * fast replay
438 */
439 if (td->o.no_stall)
440 delay = 0;
441
691c8fb0 442 handle_trace(td, &t, delay, ios, rw_bs);
64bbb865 443 }
691c8fb0 444
8c1fdf04 445 ttime = t.time;
a61eddec 446 cpu = t.cpu;
a6edd638
JA
447 } else {
448 delay = 0;
4241ea8f 449 handle_trace(td, &t, delay, ios, rw_bs);
a6edd638 450 }
fb7b71a3
JA
451 } while (1);
452
85a47ca2 453 for (i = 0; i < td->files_index; i++) {
f01b34ae 454 f = td->files[i];
89ac1d48 455 trace_add_open_close_event(td, f->fileno, FIO_LOG_CLOSE_FILE);
85a47ca2 456 }
89ac1d48 457
38470f85 458 fifo_free(fifo);
fb7b71a3 459 close(fd);
8c1fdf04 460
f01b34ae
JA
461 if (!td->files_index) {
462 log_err("fio: did not find replay device(s)\n");
463 return 1;
464 }
465
4241ea8f 466 if (skipped_writes)
5ec10eaa
JA
467 log_err("fio: %s skips replay of %lu writes due to read-only\n",
468 td->o.name, skipped_writes);
4241ea8f 469
8c1fdf04
JA
470 if (!ios[DDIR_READ] && !ios[DDIR_WRITE]) {
471 log_err("fio: found no ios in blktrace data\n");
472 return 1;
d84f8d49 473 } else if (ios[DDIR_READ] && !ios[DDIR_READ]) {
8c1fdf04 474 td->o.td_ddir = TD_DDIR_READ;
d84f8d49
JA
475 td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
476 } else if (!ios[DDIR_READ] && ios[DDIR_WRITE]) {
8c1fdf04 477 td->o.td_ddir = TD_DDIR_WRITE;
d84f8d49
JA
478 td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
479 } else {
8c1fdf04 480 td->o.td_ddir = TD_DDIR_RW;
d84f8d49
JA
481 td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ];
482 td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE];
483 }
8c1fdf04
JA
484
485 /*
486 * We need to do direct/raw ios to the device, to avoid getting
487 * read-ahead in our way.
488 */
489 td->o.odirect = 1;
490
fb7b71a3 491 return 0;
8c1fdf04
JA
492err:
493 close(fd);
38470f85 494 fifo_free(fifo);
8c1fdf04 495 return 1;
fb7b71a3 496}