2 * Code related to writing an iolog of what a thread is doing, and to
3 * later read that back and replay
12 static const char iolog_ver2[] = "fio version 2 iolog";
14 void queue_io_piece(struct thread_data *td, struct io_piece *ipo)
16 flist_add_tail(&ipo->list, &td->io_log_list);
17 td->total_io_size += ipo->len;
20 void log_io_u(struct thread_data *td, struct io_u *io_u)
22 const char *act[] = { "read", "write", "sync" };
24 assert(io_u->ddir < 3);
26 if (!td->o.write_iolog_file)
29 fprintf(td->iolog_f, "%s %s %llu %lu\n", io_u->file->file_name,
30 act[io_u->ddir], io_u->offset,
34 void log_file(struct thread_data *td, struct fio_file *f,
35 enum file_log_act what)
37 const char *act[] = { "add", "open", "close" };
41 if (!td->o.write_iolog_file)
46 * this happens on the pre-open/close done before the job starts
51 fprintf(td->iolog_f, "%s %s\n", f->file_name, act[what]);
54 static void iolog_delay(struct thread_data *td, unsigned long delay)
56 unsigned long usec = utime_since_now(&td->last_issue);
64 * less than 100 usec delay, just regard it as noise
69 usec_sleep(td, delay);
72 static int ipo_special(struct thread_data *td, struct io_piece *ipo)
80 if (ipo->ddir != DDIR_INVAL)
83 f = td->files[ipo->fileno];
85 switch (ipo->file_action) {
86 case FIO_LOG_OPEN_FILE:
87 ret = td_io_open_file(td, f);
92 td_verror(td, ret, "iolog open file");
94 case FIO_LOG_CLOSE_FILE:
95 td_io_close_file(td, f);
97 case FIO_LOG_UNLINK_FILE:
101 log_err("fio: bad file action %d\n", ipo->file_action);
108 int read_iolog_get(struct thread_data *td, struct io_u *io_u)
110 struct io_piece *ipo;
112 while (!flist_empty(&td->io_log_list)) {
115 ipo = flist_entry(td->io_log_list.next, struct io_piece, list);
116 flist_del(&ipo->list);
118 ret = ipo_special(td, ipo);
122 } else if (ret > 0) {
127 io_u->offset = ipo->offset;
128 io_u->buflen = ipo->len;
129 io_u->ddir = ipo->ddir;
130 io_u->file = td->files[ipo->fileno];
131 get_file(io_u->file);
133 dprint(FD_IO, "iolog: get %llu/%lu/%s\n", io_u->offset,
134 io_u->buflen, io_u->file->file_name);
137 iolog_delay(td, ipo->delay);
147 void prune_io_piece_log(struct thread_data *td)
149 struct io_piece *ipo;
152 while ((n = rb_first(&td->io_hist_tree)) != NULL) {
153 ipo = rb_entry(n, struct io_piece, rb_node);
154 rb_erase(n, &td->io_hist_tree);
158 while (!flist_empty(&td->io_hist_list)) {
159 ipo = flist_entry(td->io_hist_list.next, struct io_piece, list);
160 flist_del(&ipo->list);
166 * log a successful write, so we can unwind the log for verify
168 void log_io_piece(struct thread_data *td, struct io_u *io_u)
170 struct rb_node **p, *parent;
171 struct io_piece *ipo, *__ipo;
173 ipo = malloc(sizeof(struct io_piece));
174 ipo->file = io_u->file;
175 ipo->offset = io_u->offset;
176 ipo->len = io_u->buflen;
179 * We don't need to sort the entries, if:
181 * Sequential writes, or
182 * Random writes that lay out the file as it goes along
184 * For both these cases, just reading back data in the order we
185 * wrote it out is the fastest.
187 if (!td_random(td) || !td->o.overwrite) {
188 INIT_FLIST_HEAD(&ipo->list);
189 flist_add_tail(&ipo->list, &td->io_hist_list);
193 RB_CLEAR_NODE(&ipo->rb_node);
194 p = &td->io_hist_tree.rb_node;
198 * Sort the entry into the verification list
203 __ipo = rb_entry(parent, struct io_piece, rb_node);
204 if (ipo->offset <= __ipo->offset)
210 rb_link_node(&ipo->rb_node, parent, p);
211 rb_insert_color(&ipo->rb_node, &td->io_hist_tree);
214 void write_iolog_close(struct thread_data *td)
220 td->iolog_buf = NULL;
224 * Read version 2 iolog data. It is enhanced to include per-file logging,
227 static int read_iolog2(struct thread_data *td, FILE *f)
229 unsigned long long offset;
231 int reads, writes, fileno = 0, file_action = 0; /* stupid gcc */
236 free_release_files(td);
239 * Read in the read iolog and store it, reuse the infrastructure
240 * for doing verifications.
243 fname = malloc(256+16);
244 act = malloc(256+16);
247 while ((p = fgets(str, 4096, f)) != NULL) {
248 struct io_piece *ipo;
251 r = sscanf(p, "%256s %256s %llu %u", fname, act, &offset,
257 if (!strcmp(act, "read"))
259 else if (!strcmp(act, "write"))
261 else if (!strcmp(act, "sync"))
264 log_err("fio: bad iolog file action: %s\n",
270 if (!strcmp(act, "add")) {
272 fileno = add_file(td, fname);
273 file_action = FIO_LOG_ADD_FILE;
275 } else if (!strcmp(act, "open")) {
276 fileno = get_fileno(td, fname);
277 file_action = FIO_LOG_OPEN_FILE;
278 } else if (!strcmp(act, "close")) {
279 fileno = get_fileno(td, fname);
280 file_action = FIO_LOG_CLOSE_FILE;
282 log_err("fio: bad iolog file action: %s\n",
287 log_err("bad iolog2: %s", p);
293 else if (rw == DDIR_WRITE) {
295 * Don't add a write for ro mode
300 } else if (rw != DDIR_SYNC && rw != DDIR_INVAL) {
301 log_err("bad ddir: %d\n", rw);
308 ipo = malloc(sizeof(*ipo));
309 memset(ipo, 0, sizeof(*ipo));
310 INIT_FLIST_HEAD(&ipo->list);
311 ipo->offset = offset;
314 if (bytes > td->o.max_bs[rw])
315 td->o.max_bs[rw] = bytes;
316 if (rw == DDIR_INVAL) {
317 ipo->fileno = fileno;
318 ipo->file_action = file_action;
320 queue_io_piece(td, ipo);
327 if (writes && read_only) {
328 log_err("fio: <%s> skips replay of %d writes due to"
329 " read-only\n", td->o.name, writes);
333 if (!reads && !writes)
335 else if (reads && !writes)
336 td->o.td_ddir = TD_DDIR_READ;
337 else if (!reads && writes)
338 td->o.td_ddir = TD_DDIR_WRITE;
340 td->o.td_ddir = TD_DDIR_RW;
346 * open iolog, check version, and call appropriate parser
348 static int init_iolog_read(struct thread_data *td)
350 char buffer[256], *p;
354 f = fopen(td->o.read_iolog_file, "r");
356 perror("fopen read iolog");
360 p = fgets(buffer, sizeof(buffer), f);
362 td_verror(td, errno, "iolog read");
363 log_err("fio: unable to read iolog\n");
368 * version 2 of the iolog stores a specific string as the
369 * first line, check for that
371 if (!strncmp(iolog_ver2, buffer, strlen(iolog_ver2)))
372 ret = read_iolog2(td, f);
374 log_err("fio: iolog version 1 is no longer supported\n");
383 * Setup a log for storing io patterns.
385 static int init_iolog_write(struct thread_data *td)
391 f = fopen(td->o.write_iolog_file, "w+");
393 perror("fopen write iolog");
398 * That's it for writing, setup a log buffer and we're done.
401 td->iolog_buf = malloc(8192);
402 setvbuf(f, td->iolog_buf, _IOFBF, 8192);
405 * write our version line
407 if (fprintf(f, "%s\n", iolog_ver2) < 0) {
408 perror("iolog init\n");
413 * add all known files
415 for_each_file(td, ff, i)
416 log_file(td, ff, FIO_LOG_ADD_FILE);
421 int init_iolog(struct thread_data *td)
425 if (td->io_ops->flags & FIO_DISKLESSIO)
428 if (td->o.read_iolog_file) {
430 * Check if it's a blktrace file and load that if possible.
431 * Otherwise assume it's a normal log file and load that.
433 if (is_blktrace(td->o.read_iolog_file))
434 ret = load_blktrace(td, td->o.read_iolog_file);
436 ret = init_iolog_read(td);
437 } else if (td->o.write_iolog_file)
438 ret = init_iolog_write(td);
443 void setup_log(struct io_log **log)
445 struct io_log *l = malloc(sizeof(*l));
448 l->max_samples = 1024;
449 l->log = malloc(l->max_samples * sizeof(struct io_sample));
453 void __finish_log(struct io_log *log, const char *name)
458 f = fopen(name, "w");
464 for (i = 0; i < log->nr_samples; i++) {
465 fprintf(f, "%lu, %lu, %u\n", log->log[i].time, log->log[i].val,
474 void finish_log(struct thread_data *td, struct io_log *log, const char *name)
476 char file_name[256], *p;
478 snprintf(file_name, 200, "%s_%s.log", td->o.name, name);
479 p = basename(file_name);
480 __finish_log(log, p);