Commit | Line | Data |
---|---|---|
fb7b71a3 JA |
1 | /* |
2 | * blktrace support code for fio | |
3 | */ | |
4 | #include <stdio.h> | |
5 | #include <stdlib.h> | |
26e616d0 JA |
6 | #include <sys/ioctl.h> |
7 | #include <linux/fs.h> | |
8c1fdf04 | 8 | |
01743ee1 | 9 | #include "flist.h" |
fb7b71a3 | 10 | #include "fio.h" |
a3e59412 | 11 | #include "blktrace.h" |
fb7b71a3 | 12 | #include "blktrace_api.h" |
984f30c9 | 13 | #include "oslib/linux-dev-lookup.h" |
fb7b71a3 | 14 | |
2da7df1d | 15 | #define TRACE_FIFO_SIZE 8192 |
e2887563 JA |
16 | |
17 | /* | |
18 | * fifo refill frontend, to avoid reading data in trace sized bites | |
19 | */ | |
20 | static int refill_fifo(struct thread_data *td, struct fifo *fifo, int fd) | |
21 | { | |
22 | char buf[TRACE_FIFO_SIZE]; | |
f12b323f | 23 | unsigned int total; |
e2887563 JA |
24 | int ret; |
25 | ||
f12b323f JA |
26 | total = sizeof(buf); |
27 | if (total > fifo_room(fifo)) | |
28 | total = fifo_room(fifo); | |
e2887563 | 29 | |
f12b323f JA |
30 | ret = read(fd, buf, total); |
31 | if (ret < 0) { | |
32 | td_verror(td, errno, "read blktrace file"); | |
33 | return -1; | |
e2887563 JA |
34 | } |
35 | ||
f12b323f JA |
36 | if (ret > 0) |
37 | ret = fifo_put(fifo, buf, ret); | |
38 | ||
bd6f78b2 | 39 | dprint(FD_BLKTRACE, "refill: filled %d bytes\n", ret); |
f12b323f | 40 | return ret; |
e2887563 JA |
41 | } |
42 | ||
43 | /* | |
44 | * Retrieve 'len' bytes from the fifo, refilling if necessary. | |
45 | */ | |
46 | static int trace_fifo_get(struct thread_data *td, struct fifo *fifo, int fd, | |
47 | void *buf, unsigned int len) | |
48 | { | |
f12b323f JA |
49 | if (fifo_len(fifo) < len) { |
50 | int ret = refill_fifo(td, fifo, fd); | |
e2887563 | 51 | |
f12b323f JA |
52 | if (ret < 0) |
53 | return ret; | |
54 | } | |
e2887563 JA |
55 | |
56 | return fifo_get(fifo, buf, len); | |
57 | } | |
58 | ||
8c1fdf04 JA |
59 | /* |
60 | * Just discard the pdu by seeking past it. | |
61 | */ | |
f12b323f JA |
62 | static int discard_pdu(struct thread_data *td, struct fifo *fifo, int fd, |
63 | struct blk_io_trace *t) | |
fb7b71a3 JA |
64 | { |
65 | if (t->pdu_len == 0) | |
66 | return 0; | |
67 | ||
bd6f78b2 | 68 | dprint(FD_BLKTRACE, "discard pdu len %u\n", t->pdu_len); |
f12b323f | 69 | return trace_fifo_get(td, fifo, fd, NULL, t->pdu_len); |
fb7b71a3 JA |
70 | } |
71 | ||
8c1fdf04 JA |
72 | /* |
73 | * Check if this is a blktrace binary data file. We read a single trace | |
74 | * into memory and check for the magic signature. | |
75 | */ | |
b153f94a | 76 | bool is_blktrace(const char *filename, int *need_swap) |
fb7b71a3 JA |
77 | { |
78 | struct blk_io_trace t; | |
79 | int fd, ret; | |
80 | ||
81 | fd = open(filename, O_RDONLY); | |
4dced407 | 82 | if (fd < 0) |
b153f94a | 83 | return false; |
fb7b71a3 JA |
84 | |
85 | ret = read(fd, &t, sizeof(t)); | |
86 | close(fd); | |
87 | ||
88 | if (ret < 0) { | |
89 | perror("read blktrace"); | |
b153f94a | 90 | return false; |
fb7b71a3 JA |
91 | } else if (ret != sizeof(t)) { |
92 | log_err("fio: short read on blktrace file\n"); | |
b153f94a | 93 | return false; |
fb7b71a3 JA |
94 | } |
95 | ||
d95b34a6 JA |
96 | if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) { |
97 | *need_swap = 0; | |
b153f94a | 98 | return true; |
d95b34a6 JA |
99 | } |
100 | ||
101 | /* | |
102 | * Maybe it needs to be endian swapped... | |
103 | */ | |
104 | t.magic = fio_swap32(t.magic); | |
105 | if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC) { | |
106 | *need_swap = 1; | |
b153f94a | 107 | return true; |
d95b34a6 | 108 | } |
fb7b71a3 | 109 | |
b153f94a | 110 | return false; |
fb7b71a3 JA |
111 | } |
112 | ||
c69aa91f JA |
113 | #define FMINORBITS 20 |
114 | #define FMINORMASK ((1U << FMINORBITS) - 1) | |
115 | #define FMAJOR(dev) ((unsigned int) ((dev) >> FMINORBITS)) | |
116 | #define FMINOR(dev) ((unsigned int) ((dev) & FMINORMASK)) | |
eeb9c2aa | 117 | |
89ac1d48 | 118 | static void trace_add_open_close_event(struct thread_data *td, int fileno, enum file_log_act action) |
691c8fb0 JA |
119 | { |
120 | struct io_piece *ipo; | |
121 | ||
122 | ipo = calloc(1, sizeof(*ipo)); | |
0d29de83 | 123 | init_ipo(ipo); |
691c8fb0 JA |
124 | |
125 | ipo->ddir = DDIR_INVAL; | |
126 | ipo->fileno = fileno; | |
89ac1d48 | 127 | ipo->file_action = action; |
01743ee1 | 128 | flist_add_tail(&ipo->list, &td->io_log_list); |
691c8fb0 JA |
129 | } |
130 | ||
26e616d0 | 131 | static int get_dev_blocksize(const char *dev, unsigned int *bs) |
5e6c2067 | 132 | { |
26e616d0 JA |
133 | int fd; |
134 | ||
135 | fd = open(dev, O_RDONLY); | |
136 | if (fd < 0) | |
137 | return 1; | |
138 | ||
139 | if (ioctl(fd, BLKSSZGET, bs) < 0) { | |
140 | close(fd); | |
141 | return 1; | |
142 | } | |
143 | ||
144 | close(fd); | |
145 | return 0; | |
146 | } | |
147 | ||
148 | static int trace_add_file(struct thread_data *td, __u32 device, | |
149 | unsigned int *bs) | |
150 | { | |
151 | static unsigned int last_maj, last_min, last_fileno, last_bs; | |
c69aa91f JA |
152 | unsigned int maj = FMAJOR(device); |
153 | unsigned int min = FMINOR(device); | |
5e6c2067 | 154 | struct fio_file *f; |
5e6c2067 | 155 | unsigned int i; |
26e616d0 | 156 | char dev[256]; |
5e6c2067 | 157 | |
26e616d0 JA |
158 | if (last_maj == maj && last_min == min) { |
159 | *bs = last_bs; | |
89ac1d48 | 160 | return last_fileno; |
26e616d0 | 161 | } |
5e6c2067 JA |
162 | |
163 | last_maj = maj; | |
164 | last_min = min; | |
165 | ||
166 | /* | |
167 | * check for this file in our list | |
168 | */ | |
26e616d0 | 169 | for_each_file(td, f, i) { |
89ac1d48 SL |
170 | if (f->major == maj && f->minor == min) { |
171 | last_fileno = f->fileno; | |
26e616d0 JA |
172 | last_bs = f->bs; |
173 | goto out; | |
89ac1d48 | 174 | } |
26e616d0 | 175 | } |
5e6c2067 JA |
176 | |
177 | strcpy(dev, "/dev"); | |
40bafa33 | 178 | if (blktrace_lookup_device(td->o.replay_redirect, dev, maj, min)) { |
26e616d0 | 179 | unsigned int this_bs; |
691c8fb0 JA |
180 | int fileno; |
181 | ||
40bafa33 JA |
182 | if (td->o.replay_redirect) |
183 | dprint(FD_BLKTRACE, "device lookup: %d/%d\n overridden" | |
184 | " with: %s\n", maj, min, | |
185 | td->o.replay_redirect); | |
186 | else | |
187 | dprint(FD_BLKTRACE, "device lookup: %d/%d\n", maj, min); | |
188 | ||
bd6f78b2 | 189 | dprint(FD_BLKTRACE, "add devices %s\n", dev); |
49ffb4a2 | 190 | fileno = add_file_exclusive(td, dev); |
26e616d0 JA |
191 | |
192 | if (get_dev_blocksize(dev, &this_bs)) | |
193 | this_bs = 512; | |
194 | ||
b53f2c54 | 195 | td->o.open_files++; |
5903e7b7 JA |
196 | td->files[fileno]->major = maj; |
197 | td->files[fileno]->minor = min; | |
26e616d0 | 198 | td->files[fileno]->bs = this_bs; |
89ac1d48 | 199 | trace_add_open_close_event(td, fileno, FIO_LOG_OPEN_FILE); |
26e616d0 | 200 | |
89ac1d48 | 201 | last_fileno = fileno; |
26e616d0 | 202 | last_bs = this_bs; |
bd6f78b2 | 203 | } |
f01b34ae | 204 | |
26e616d0 JA |
205 | out: |
206 | *bs = last_bs; | |
89ac1d48 | 207 | return last_fileno; |
5e6c2067 JA |
208 | } |
209 | ||
0c63576e JA |
210 | static void t_bytes_align(struct thread_options *o, struct blk_io_trace *t) |
211 | { | |
212 | if (!o->replay_align) | |
213 | return; | |
214 | ||
215 | t->bytes = (t->bytes + o->replay_align - 1) & ~(o->replay_align - 1); | |
216 | } | |
217 | ||
8c1fdf04 JA |
218 | /* |
219 | * Store blk_io_trace data in an ipo for later retrieval. | |
220 | */ | |
fdefd987 | 221 | static void store_ipo(struct thread_data *td, unsigned long long offset, |
89ac1d48 | 222 | unsigned int bytes, int rw, unsigned long long ttime, |
26e616d0 | 223 | int fileno, unsigned int bs) |
fdefd987 | 224 | { |
8812d7f2 | 225 | struct io_piece *ipo; |
fdefd987 | 226 | |
8812d7f2 | 227 | ipo = calloc(1, sizeof(*ipo)); |
0d29de83 JA |
228 | init_ipo(ipo); |
229 | ||
26e616d0 | 230 | ipo->offset = offset * bs; |
0c63576e JA |
231 | if (td->o.replay_scale) |
232 | ipo->offset = ipo->offset / td->o.replay_scale; | |
a79f17bf | 233 | ipo_bytes_align(td->o.replay_align, ipo); |
fdefd987 | 234 | ipo->len = bytes; |
8c1fdf04 | 235 | ipo->delay = ttime / 1000; |
fdefd987 JA |
236 | if (rw) |
237 | ipo->ddir = DDIR_WRITE; | |
238 | else | |
239 | ipo->ddir = DDIR_READ; | |
89ac1d48 | 240 | ipo->fileno = fileno; |
fdefd987 | 241 | |
bd6f78b2 JA |
242 | dprint(FD_BLKTRACE, "store ddir=%d, off=%llu, len=%lu, delay=%lu\n", |
243 | ipo->ddir, ipo->offset, | |
244 | ipo->len, ipo->delay); | |
691c8fb0 | 245 | queue_io_piece(td, ipo); |
fdefd987 JA |
246 | } |
247 | ||
0b9d69ec | 248 | static void handle_trace_notify(struct blk_io_trace *t) |
cd991b9e | 249 | { |
691c8fb0 JA |
250 | switch (t->action) { |
251 | case BLK_TN_PROCESS: | |
24653680 | 252 | dprint(FD_BLKTRACE, "got process notify: %x, %d\n", |
d95b34a6 | 253 | t->action, t->pid); |
691c8fb0 JA |
254 | break; |
255 | case BLK_TN_TIMESTAMP: | |
24653680 | 256 | dprint(FD_BLKTRACE, "got timestamp notify: %x, %d\n", |
d95b34a6 | 257 | t->action, t->pid); |
691c8fb0 | 258 | break; |
ff58fced JA |
259 | case BLK_TN_MESSAGE: |
260 | break; | |
691c8fb0 JA |
261 | default: |
262 | dprint(FD_BLKTRACE, "unknown trace act %x\n", t->action); | |
263 | break; | |
264 | } | |
265 | } | |
5b3023b8 | 266 | |
24653680 JA |
267 | static void handle_trace_discard(struct thread_data *td, |
268 | struct blk_io_trace *t, | |
269 | unsigned long long ttime, | |
26e616d0 | 270 | unsigned long *ios, unsigned int *rw_bs) |
ff58fced | 271 | { |
8812d7f2 | 272 | struct io_piece *ipo; |
26e616d0 | 273 | unsigned int bs; |
89ac1d48 | 274 | int fileno; |
ff58fced | 275 | |
d7235efb JA |
276 | if (td->o.replay_skip & (1u << DDIR_TRIM)) |
277 | return; | |
278 | ||
8812d7f2 | 279 | ipo = calloc(1, sizeof(*ipo)); |
0d29de83 | 280 | init_ipo(ipo); |
26e616d0 | 281 | fileno = trace_add_file(td, t->device, &bs); |
ff58fced | 282 | |
24653680 | 283 | ios[DDIR_TRIM]++; |
26e616d0 JA |
284 | if (t->bytes > rw_bs[DDIR_TRIM]) |
285 | rw_bs[DDIR_TRIM] = t->bytes; | |
24653680 | 286 | |
ff58fced JA |
287 | td->o.size += t->bytes; |
288 | ||
ff58fced JA |
289 | INIT_FLIST_HEAD(&ipo->list); |
290 | ||
26e616d0 | 291 | ipo->offset = t->sector * bs; |
0c63576e JA |
292 | if (td->o.replay_scale) |
293 | ipo->offset = ipo->offset / td->o.replay_scale; | |
a79f17bf | 294 | ipo_bytes_align(td->o.replay_align, ipo); |
ff58fced JA |
295 | ipo->len = t->bytes; |
296 | ipo->delay = ttime / 1000; | |
297 | ipo->ddir = DDIR_TRIM; | |
89ac1d48 | 298 | ipo->fileno = fileno; |
ff58fced JA |
299 | |
300 | dprint(FD_BLKTRACE, "store discard, off=%llu, len=%lu, delay=%lu\n", | |
301 | ipo->offset, ipo->len, | |
302 | ipo->delay); | |
303 | queue_io_piece(td, ipo); | |
304 | } | |
305 | ||
19a8064e JA |
306 | static void dump_trace(struct blk_io_trace *t) |
307 | { | |
308 | log_err("blktrace: ignoring zero byte trace: action=%x\n", t->action); | |
309 | } | |
310 | ||
691c8fb0 JA |
311 | static void handle_trace_fs(struct thread_data *td, struct blk_io_trace *t, |
312 | unsigned long long ttime, unsigned long *ios, | |
26e616d0 | 313 | unsigned int *rw_bs) |
691c8fb0 | 314 | { |
26e616d0 | 315 | unsigned int bs; |
691c8fb0 | 316 | int rw; |
89ac1d48 | 317 | int fileno; |
5b3023b8 | 318 | |
26e616d0 | 319 | fileno = trace_add_file(td, t->device, &bs); |
5b3023b8 JA |
320 | |
321 | rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0; | |
322 | ||
d7235efb JA |
323 | if (rw) { |
324 | if (td->o.replay_skip & (1u << DDIR_WRITE)) | |
325 | return; | |
326 | } else { | |
327 | if (td->o.replay_skip & (1u << DDIR_READ)) | |
328 | return; | |
329 | } | |
330 | ||
19a8064e JA |
331 | if (!t->bytes) { |
332 | if (!fio_did_warn(FIO_WARN_BTRACE_ZERO)) | |
333 | dump_trace(t); | |
334 | return; | |
335 | } | |
c1f22c21 JA |
336 | |
337 | if (t->bytes > rw_bs[rw]) | |
26e616d0 | 338 | rw_bs[rw] = t->bytes; |
5b3023b8 JA |
339 | |
340 | ios[rw]++; | |
341 | td->o.size += t->bytes; | |
26e616d0 | 342 | store_ipo(td, t->sector, t->bytes, rw, ttime, fileno, bs); |
cd991b9e JA |
343 | } |
344 | ||
65c42cc8 | 345 | static void handle_trace_flush(struct thread_data *td, struct blk_io_trace *t, |
811f5421 | 346 | unsigned long long ttime, unsigned long *ios) |
65c42cc8 JA |
347 | { |
348 | struct io_piece *ipo; | |
349 | unsigned int bs; | |
350 | int fileno; | |
351 | ||
d7235efb JA |
352 | if (td->o.replay_skip & (1u << DDIR_SYNC)) |
353 | return; | |
354 | ||
65c42cc8 JA |
355 | ipo = calloc(1, sizeof(*ipo)); |
356 | init_ipo(ipo); | |
357 | fileno = trace_add_file(td, t->device, &bs); | |
358 | ||
359 | ipo->delay = ttime / 1000; | |
360 | ipo->ddir = DDIR_SYNC; | |
361 | ipo->fileno = fileno; | |
362 | ||
811f5421 | 363 | ios[DDIR_SYNC]++; |
65c42cc8 JA |
364 | dprint(FD_BLKTRACE, "store flush delay=%lu\n", ipo->delay); |
365 | queue_io_piece(td, ipo); | |
366 | } | |
367 | ||
691c8fb0 JA |
368 | /* |
369 | * We only care for queue traces, most of the others are side effects | |
370 | * due to internal workings of the block layer. | |
371 | */ | |
372 | static void handle_trace(struct thread_data *td, struct blk_io_trace *t, | |
24653680 | 373 | unsigned long *ios, unsigned int *bs) |
691c8fb0 | 374 | { |
24653680 | 375 | static unsigned long long last_ttime; |
0c63576e | 376 | unsigned long long delay = 0; |
24653680 | 377 | |
691c8fb0 JA |
378 | if ((t->action & 0xffff) != __BLK_TA_QUEUE) |
379 | return; | |
24653680 JA |
380 | |
381 | if (!(t->action & BLK_TC_ACT(BLK_TC_NOTIFY))) { | |
6dd7fa77 | 382 | if (!last_ttime || td->o.no_stall) |
24653680 | 383 | delay = 0; |
6dd7fa77 | 384 | else if (td->o.replay_time_scale == 100) |
24653680 | 385 | delay = t->time - last_ttime; |
6dd7fa77 JA |
386 | else { |
387 | double tmp = t->time - last_ttime; | |
388 | double scale; | |
389 | ||
390 | scale = (double) 100.0 / (double) td->o.replay_time_scale; | |
391 | tmp *= scale; | |
392 | delay = tmp; | |
24653680 | 393 | } |
6dd7fa77 | 394 | last_ttime = t->time; |
24653680 | 395 | } |
691c8fb0 | 396 | |
0c63576e JA |
397 | t_bytes_align(&td->o, t); |
398 | ||
691c8fb0 | 399 | if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY)) |
aec2de20 | 400 | handle_trace_notify(t); |
ff58fced | 401 | else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD)) |
24653680 | 402 | handle_trace_discard(td, t, delay, ios, bs); |
65c42cc8 | 403 | else if (t->action & BLK_TC_ACT(BLK_TC_FLUSH)) |
811f5421 | 404 | handle_trace_flush(td, t, delay, ios); |
691c8fb0 | 405 | else |
24653680 | 406 | handle_trace_fs(td, t, delay, ios, bs); |
691c8fb0 JA |
407 | } |
408 | ||
d95b34a6 JA |
409 | static void byteswap_trace(struct blk_io_trace *t) |
410 | { | |
411 | t->magic = fio_swap32(t->magic); | |
412 | t->sequence = fio_swap32(t->sequence); | |
413 | t->time = fio_swap64(t->time); | |
414 | t->sector = fio_swap64(t->sector); | |
415 | t->bytes = fio_swap32(t->bytes); | |
416 | t->action = fio_swap32(t->action); | |
417 | t->pid = fio_swap32(t->pid); | |
418 | t->device = fio_swap32(t->device); | |
419 | t->cpu = fio_swap32(t->cpu); | |
420 | t->error = fio_swap16(t->error); | |
421 | t->pdu_len = fio_swap16(t->pdu_len); | |
422 | } | |
423 | ||
b153f94a | 424 | static bool t_is_write(struct blk_io_trace *t) |
24653680 JA |
425 | { |
426 | return (t->action & BLK_TC_ACT(BLK_TC_WRITE | BLK_TC_DISCARD)) != 0; | |
427 | } | |
428 | ||
a6eaf6c9 JA |
429 | static enum fio_ddir t_get_ddir(struct blk_io_trace *t) |
430 | { | |
431 | if (t->action & BLK_TC_ACT(BLK_TC_READ)) | |
432 | return DDIR_READ; | |
433 | else if (t->action & BLK_TC_ACT(BLK_TC_WRITE)) | |
434 | return DDIR_WRITE; | |
435 | else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD)) | |
436 | return DDIR_TRIM; | |
437 | ||
438 | return DDIR_INVAL; | |
439 | } | |
440 | ||
441 | static void depth_inc(struct blk_io_trace *t, int *depth) | |
442 | { | |
443 | enum fio_ddir ddir; | |
444 | ||
445 | ddir = t_get_ddir(t); | |
446 | if (ddir != DDIR_INVAL) | |
447 | depth[ddir]++; | |
448 | } | |
449 | ||
450 | static void depth_dec(struct blk_io_trace *t, int *depth) | |
451 | { | |
452 | enum fio_ddir ddir; | |
453 | ||
454 | ddir = t_get_ddir(t); | |
455 | if (ddir != DDIR_INVAL) | |
456 | depth[ddir]--; | |
457 | } | |
458 | ||
459 | static void depth_end(struct blk_io_trace *t, int *this_depth, int *depth) | |
460 | { | |
461 | enum fio_ddir ddir = DDIR_INVAL; | |
462 | ||
463 | ddir = t_get_ddir(t); | |
464 | if (ddir != DDIR_INVAL) { | |
465 | depth[ddir] = max(depth[ddir], this_depth[ddir]); | |
466 | this_depth[ddir] = 0; | |
467 | } | |
468 | } | |
469 | ||
8c1fdf04 JA |
470 | /* |
471 | * Load a blktrace file by reading all the blk_io_trace entries, and storing | |
472 | * them as io_pieces like the fio text version would do. | |
473 | */ | |
b153f94a | 474 | bool load_blktrace(struct thread_data *td, const char *filename, int need_swap) |
fb7b71a3 JA |
475 | { |
476 | struct blk_io_trace t; | |
811f5421 JA |
477 | unsigned long ios[DDIR_RWDIR_SYNC_CNT] = { }; |
478 | unsigned int rw_bs[DDIR_RWDIR_CNT] = { }; | |
479 | unsigned long skipped_writes; | |
e2887563 | 480 | struct fifo *fifo; |
811f5421 | 481 | int fd, i, old_state, max_depth; |
89ac1d48 | 482 | struct fio_file *f; |
811f5421 JA |
483 | int this_depth[DDIR_RWDIR_CNT] = { }; |
484 | int depth[DDIR_RWDIR_CNT] = { }; | |
fb7b71a3 JA |
485 | |
486 | fd = open(filename, O_RDONLY); | |
487 | if (fd < 0) { | |
488 | td_verror(td, errno, "open blktrace file"); | |
b153f94a | 489 | return false; |
fb7b71a3 JA |
490 | } |
491 | ||
e2887563 JA |
492 | fifo = fifo_alloc(TRACE_FIFO_SIZE); |
493 | ||
8edd973d | 494 | old_state = td_bump_runstate(td, TD_SETTING_UP); |
5903e7b7 | 495 | |
6df8adaa | 496 | td->o.size = 0; |
4241ea8f | 497 | skipped_writes = 0; |
fb7b71a3 | 498 | do { |
e2887563 | 499 | int ret = trace_fifo_get(td, fifo, fd, &t, sizeof(t)); |
fb7b71a3 | 500 | |
e2887563 | 501 | if (ret < 0) |
8c1fdf04 | 502 | goto err; |
e2887563 JA |
503 | else if (!ret) |
504 | break; | |
505 | else if (ret < (int) sizeof(t)) { | |
506 | log_err("fio: short fifo get\n"); | |
fb7b71a3 | 507 | break; |
fb7b71a3 JA |
508 | } |
509 | ||
d95b34a6 JA |
510 | if (need_swap) |
511 | byteswap_trace(&t); | |
512 | ||
fb7b71a3 | 513 | if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) { |
5ec10eaa JA |
514 | log_err("fio: bad magic in blktrace data: %x\n", |
515 | t.magic); | |
8c1fdf04 | 516 | goto err; |
fb7b71a3 JA |
517 | } |
518 | if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) { | |
5ec10eaa JA |
519 | log_err("fio: bad blktrace version %d\n", |
520 | t.magic & 0xff); | |
8c1fdf04 | 521 | goto err; |
fb7b71a3 | 522 | } |
f12b323f JA |
523 | ret = discard_pdu(td, fifo, fd, &t); |
524 | if (ret < 0) { | |
fb7b71a3 | 525 | td_verror(td, ret, "blktrace lseek"); |
8c1fdf04 | 526 | goto err; |
f12b323f JA |
527 | } else if (t.pdu_len != ret) { |
528 | log_err("fio: discarded %d of %d\n", ret, t.pdu_len); | |
529 | goto err; | |
fb7b71a3 | 530 | } |
691c8fb0 | 531 | if ((t.action & BLK_TC_ACT(BLK_TC_NOTIFY)) == 0) { |
eb5fdcf1 | 532 | if ((t.action & 0xffff) == __BLK_TA_QUEUE) |
a6eaf6c9 JA |
533 | depth_inc(&t, this_depth); |
534 | else if (((t.action & 0xffff) == __BLK_TA_BACKMERGE) || | |
535 | ((t.action & 0xffff) == __BLK_TA_FRONTMERGE)) | |
536 | depth_dec(&t, this_depth); | |
537 | else if ((t.action & 0xffff) == __BLK_TA_COMPLETE) | |
538 | depth_end(&t, this_depth, depth); | |
691c8fb0 | 539 | |
24653680 | 540 | if (t_is_write(&t) && read_only) { |
691c8fb0 | 541 | skipped_writes++; |
24653680 | 542 | continue; |
64bbb865 | 543 | } |
a6edd638 | 544 | } |
24653680 JA |
545 | |
546 | handle_trace(td, &t, ios, rw_bs); | |
fb7b71a3 JA |
547 | } while (1); |
548 | ||
3643a1f2 | 549 | for_each_file(td, f, i) |
89ac1d48 | 550 | trace_add_open_close_event(td, f->fileno, FIO_LOG_CLOSE_FILE); |
89ac1d48 | 551 | |
38470f85 | 552 | fifo_free(fifo); |
fb7b71a3 | 553 | close(fd); |
8c1fdf04 | 554 | |
8edd973d | 555 | td_restore_runstate(td, old_state); |
5903e7b7 | 556 | |
f01b34ae JA |
557 | if (!td->files_index) { |
558 | log_err("fio: did not find replay device(s)\n"); | |
b153f94a | 559 | return false; |
f01b34ae JA |
560 | } |
561 | ||
eb5fdcf1 JA |
562 | /* |
563 | * For stacked devices, we don't always get a COMPLETE event so | |
564 | * the depth grows to insane values. Limit it to something sane(r). | |
565 | */ | |
a6eaf6c9 JA |
566 | max_depth = 0; |
567 | for (i = 0; i < DDIR_RWDIR_CNT; i++) { | |
568 | if (depth[i] > 1024) | |
569 | depth[i] = 1024; | |
570 | else if (!depth[i] && ios[i]) | |
571 | depth[i] = 1; | |
572 | max_depth = max(depth[i], max_depth); | |
573 | } | |
eb5fdcf1 | 574 | |
4241ea8f | 575 | if (skipped_writes) |
5ec10eaa JA |
576 | log_err("fio: %s skips replay of %lu writes due to read-only\n", |
577 | td->o.name, skipped_writes); | |
4241ea8f | 578 | |
811f5421 JA |
579 | if (!ios[DDIR_READ] && !ios[DDIR_WRITE] && !ios[DDIR_TRIM] && |
580 | !ios[DDIR_SYNC]) { | |
8c1fdf04 | 581 | log_err("fio: found no ios in blktrace data\n"); |
b153f94a | 582 | return false; |
24653680 | 583 | } else if (ios[DDIR_READ] && !ios[DDIR_WRITE]) { |
8c1fdf04 | 584 | td->o.td_ddir = TD_DDIR_READ; |
d84f8d49 JA |
585 | td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ]; |
586 | } else if (!ios[DDIR_READ] && ios[DDIR_WRITE]) { | |
8c1fdf04 | 587 | td->o.td_ddir = TD_DDIR_WRITE; |
d84f8d49 JA |
588 | td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE]; |
589 | } else { | |
8c1fdf04 | 590 | td->o.td_ddir = TD_DDIR_RW; |
d84f8d49 JA |
591 | td->o.max_bs[DDIR_READ] = rw_bs[DDIR_READ]; |
592 | td->o.max_bs[DDIR_WRITE] = rw_bs[DDIR_WRITE]; | |
24653680 | 593 | td->o.max_bs[DDIR_TRIM] = rw_bs[DDIR_TRIM]; |
d84f8d49 | 594 | } |
8c1fdf04 JA |
595 | |
596 | /* | |
597 | * We need to do direct/raw ios to the device, to avoid getting | |
76a00ec6 JA |
598 | * read-ahead in our way. But only do so if the minimum block size |
599 | * is a multiple of 4k, otherwise we don't know if it's safe to do so. | |
8c1fdf04 | 600 | */ |
76a00ec6 | 601 | if (!fio_option_is_set(&td->o, odirect) && !(td_min_bs(td) & 4095)) |
a6eaf6c9 | 602 | td->o.odirect = 1; |
8c1fdf04 | 603 | |
eb5fdcf1 | 604 | /* |
8a16f59b | 605 | * If depth wasn't manually set, use probed depth |
eb5fdcf1 | 606 | */ |
8a16f59b | 607 | if (!fio_option_is_set(&td->o, iodepth)) |
a6eaf6c9 | 608 | td->o.iodepth = td->o.iodepth_low = max_depth; |
eb5fdcf1 | 609 | |
b153f94a | 610 | return true; |
8c1fdf04 JA |
611 | err: |
612 | close(fd); | |
38470f85 | 613 | fifo_free(fifo); |
b153f94a | 614 | return false; |
fb7b71a3 | 615 | } |