Commit | Line | Data |
---|---|---|
9f8f2064 JA |
1 | #include <stdio.h> |
2 | #include <string.h> | |
3 | #include <sys/time.h> | |
4 | #include <sys/types.h> | |
5 | #include <sys/stat.h> | |
6 | #include <dirent.h> | |
7 | #include <libgen.h> | |
8 | #include <math.h> | |
9 | ||
10 | #include "fio.h" | |
c97bd0fa | 11 | #include "smalloc.h" |
9f8f2064 JA |
12 | |
13 | static int last_majdev, last_mindev; | |
c97bd0fa | 14 | static struct disk_util *last_du; |
9f8f2064 | 15 | |
01743ee1 | 16 | static struct flist_head disk_list = FLIST_HEAD_INIT(disk_list); |
9f8f2064 | 17 | |
c97bd0fa JA |
18 | static void disk_util_free(struct disk_util *du) |
19 | { | |
20 | if (du == last_du) | |
21 | last_du = NULL; | |
22 | ||
23 | fio_mutex_remove(du->lock); | |
24 | sfree(du->name); | |
25 | sfree(du); | |
26 | } | |
27 | ||
9f8f2064 JA |
28 | static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus) |
29 | { | |
30 | unsigned in_flight; | |
31 | char line[256]; | |
32 | FILE *f; | |
33 | char *p; | |
5ec10eaa | 34 | int ret; |
9f8f2064 | 35 | |
cd991b9e JA |
36 | dprint(FD_DISKUTIL, "open stat file: %s\n", du->path); |
37 | ||
9f8f2064 JA |
38 | f = fopen(du->path, "r"); |
39 | if (!f) | |
40 | return 1; | |
41 | ||
42 | p = fgets(line, sizeof(line), f); | |
43 | if (!p) { | |
44 | fclose(f); | |
45 | return 1; | |
46 | } | |
47 | ||
cd991b9e JA |
48 | dprint(FD_DISKUTIL, "%s: %s", du->path, p); |
49 | ||
50 | ret = sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], | |
5ec10eaa JA |
51 | &dus->merges[0], &dus->sectors[0], |
52 | &dus->ticks[0], &dus->ios[1], | |
53 | &dus->merges[1], &dus->sectors[1], | |
54 | &dus->ticks[1], &in_flight, | |
55 | &dus->io_ticks, &dus->time_in_queue); | |
9f8f2064 | 56 | fclose(f); |
cd991b9e JA |
57 | dprint(FD_DISKUTIL, "%s: stat read ok? %d\n", du->path, ret == 1); |
58 | return ret != 11; | |
9f8f2064 JA |
59 | } |
60 | ||
61 | static void update_io_tick_disk(struct disk_util *du) | |
62 | { | |
63 | struct disk_util_stat __dus, *dus, *ldus; | |
64 | struct timeval t; | |
65 | ||
66 | if (get_io_ticks(du, &__dus)) | |
67 | return; | |
c97bd0fa JA |
68 | if (!du->users) |
69 | return; | |
9f8f2064 JA |
70 | |
71 | dus = &du->dus; | |
72 | ldus = &du->last_dus; | |
73 | ||
74 | dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]); | |
75 | dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]); | |
76 | dus->ios[0] += (__dus.ios[0] - ldus->ios[0]); | |
77 | dus->ios[1] += (__dus.ios[1] - ldus->ios[1]); | |
78 | dus->merges[0] += (__dus.merges[0] - ldus->merges[0]); | |
79 | dus->merges[1] += (__dus.merges[1] - ldus->merges[1]); | |
80 | dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]); | |
81 | dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]); | |
82 | dus->io_ticks += (__dus.io_ticks - ldus->io_ticks); | |
83 | dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue); | |
84 | ||
85 | fio_gettime(&t, NULL); | |
86 | du->msec += mtime_since(&du->time, &t); | |
87 | memcpy(&du->time, &t, sizeof(t)); | |
88 | memcpy(ldus, &__dus, sizeof(__dus)); | |
89 | } | |
90 | ||
91 | void update_io_ticks(void) | |
92 | { | |
01743ee1 | 93 | struct flist_head *entry; |
9f8f2064 JA |
94 | struct disk_util *du; |
95 | ||
cd991b9e JA |
96 | dprint(FD_DISKUTIL, "update io ticks\n"); |
97 | ||
01743ee1 JA |
98 | flist_for_each(entry, &disk_list) { |
99 | du = flist_entry(entry, struct disk_util, list); | |
9f8f2064 JA |
100 | update_io_tick_disk(du); |
101 | } | |
102 | } | |
103 | ||
e11c410c | 104 | static struct disk_util *disk_util_exists(int major, int minor) |
9f8f2064 | 105 | { |
01743ee1 | 106 | struct flist_head *entry; |
9f8f2064 JA |
107 | struct disk_util *du; |
108 | ||
01743ee1 JA |
109 | flist_for_each(entry, &disk_list) { |
110 | du = flist_entry(entry, struct disk_util, list); | |
9f8f2064 JA |
111 | |
112 | if (major == du->major && minor == du->minor) | |
e11c410c | 113 | return du; |
9f8f2064 JA |
114 | } |
115 | ||
e11c410c | 116 | return NULL; |
9f8f2064 JA |
117 | } |
118 | ||
67423013 ST |
119 | static int get_device_numbers(char *file_name, int *maj, int *min) |
120 | { | |
121 | struct stat st; | |
122 | int majdev, mindev; | |
123 | char tempname[PATH_MAX], *p; | |
124 | ||
125 | if (!lstat(file_name, &st)) { | |
126 | if (S_ISBLK(st.st_mode)) { | |
127 | majdev = major(st.st_rdev); | |
128 | mindev = minor(st.st_rdev); | |
129 | } else if (S_ISCHR(st.st_mode)) { | |
130 | majdev = major(st.st_rdev); | |
131 | mindev = minor(st.st_rdev); | |
132 | if (fio_lookup_raw(st.st_rdev, &majdev, &mindev)) | |
133 | return -1; | |
134 | } else if (S_ISFIFO(st.st_mode)) | |
135 | return -1; | |
136 | else { | |
137 | majdev = major(st.st_dev); | |
138 | mindev = minor(st.st_dev); | |
139 | } | |
140 | } else { | |
141 | /* | |
142 | * must be a file, open "." in that path | |
143 | */ | |
144 | strncpy(tempname, file_name, PATH_MAX - 1); | |
145 | p = dirname(tempname); | |
146 | if (stat(p, &st)) { | |
147 | perror("disk util stat"); | |
148 | return -1; | |
149 | } | |
150 | ||
151 | majdev = major(st.st_dev); | |
152 | mindev = minor(st.st_dev); | |
153 | } | |
154 | ||
155 | *min = mindev; | |
156 | *maj = majdev; | |
157 | ||
158 | return 0; | |
159 | } | |
160 | ||
161 | static int read_block_dev_entry(char *path, int *maj, int *min) | |
162 | { | |
163 | char line[256], *p; | |
164 | FILE *f; | |
165 | ||
166 | f = fopen(path, "r"); | |
167 | if (!f) { | |
168 | perror("open path"); | |
169 | return 1; | |
170 | } | |
171 | ||
172 | p = fgets(line, sizeof(line), f); | |
173 | fclose(f); | |
174 | ||
175 | if (!p) | |
176 | return 1; | |
177 | ||
178 | if (sscanf(p, "%u:%u", maj, min) != 2) | |
179 | return 1; | |
180 | ||
181 | return 0; | |
182 | } | |
183 | ||
c97bd0fa JA |
184 | static struct disk_util *__init_per_file_disk_util(struct thread_data *td, |
185 | int majdev, int mindev, char * path); | |
67423013 ST |
186 | |
187 | static void find_add_disk_slaves(struct thread_data *td, char *path, | |
188 | struct disk_util *masterdu) | |
189 | { | |
190 | DIR *dirhandle = NULL; | |
191 | struct dirent *dirent = NULL; | |
192 | char slavesdir[PATH_MAX], temppath[PATH_MAX], slavepath[PATH_MAX]; | |
193 | struct disk_util *slavedu = NULL; | |
194 | int majdev, mindev; | |
195 | ssize_t linklen; | |
196 | ||
197 | sprintf(slavesdir, "%s/%s",path, "slaves"); | |
198 | dirhandle = opendir(slavesdir); | |
199 | if (!dirhandle) | |
200 | return; | |
201 | ||
202 | while ((dirent = readdir(dirhandle)) != NULL) { | |
203 | if (!strcmp(dirent->d_name, ".") || !strcmp(dirent->d_name, "..")) | |
204 | continue; | |
205 | ||
206 | sprintf(temppath, "%s/%s", slavesdir, dirent->d_name); | |
207 | /* Can we always assume that the slaves device entries | |
208 | * are links to the real directories for the slave | |
209 | * devices? | |
210 | */ | |
211 | if ((linklen = readlink(temppath, slavepath, PATH_MAX-1)) < 0) { | |
212 | perror("readlink() for slave device."); | |
213 | return; | |
214 | } | |
215 | slavepath[linklen]='\0'; | |
216 | ||
217 | sprintf(temppath, "%s/%s/dev", slavesdir, slavepath); | |
218 | if (read_block_dev_entry(temppath, &majdev, &mindev)) { | |
219 | perror("Error getting slave device numbers."); | |
220 | return; | |
221 | } | |
222 | ||
223 | sprintf(temppath, "%s/%s", slavesdir, slavepath); | |
224 | __init_per_file_disk_util(td, majdev, mindev, temppath); | |
225 | slavedu = disk_util_exists(majdev, mindev); | |
226 | ||
227 | /* Should probably use an assert here. slavedu should | |
228 | * always be present at this point. */ | |
229 | if (slavedu) | |
230 | flist_add_tail(&slavedu->slavelist, &masterdu->slaves); | |
231 | } | |
232 | ||
233 | closedir(dirhandle); | |
234 | return; | |
235 | } | |
236 | ||
c97bd0fa JA |
237 | static struct disk_util *disk_util_add(struct thread_data * td, int majdev, |
238 | int mindev, char *path) | |
9f8f2064 JA |
239 | { |
240 | struct disk_util *du, *__du; | |
01743ee1 | 241 | struct flist_head *entry; |
9f8f2064 | 242 | |
cd991b9e JA |
243 | dprint(FD_DISKUTIL, "add maj/min %d/%d: %s\n", majdev, mindev, path); |
244 | ||
c97bd0fa | 245 | du = smalloc(sizeof(*du)); |
9f8f2064 | 246 | memset(du, 0, sizeof(*du)); |
01743ee1 | 247 | INIT_FLIST_HEAD(&du->list); |
9f8f2064 | 248 | sprintf(du->path, "%s/stat", path); |
c97bd0fa | 249 | du->name = smalloc_strdup(basename(path)); |
e11c410c | 250 | du->sysfs_root = path; |
9f8f2064 JA |
251 | du->major = majdev; |
252 | du->minor = mindev; | |
67423013 ST |
253 | INIT_FLIST_HEAD(&du->slavelist); |
254 | INIT_FLIST_HEAD(&du->slaves); | |
c97bd0fa JA |
255 | du->lock = fio_mutex_init(1); |
256 | du->users = 0; | |
9f8f2064 | 257 | |
01743ee1 JA |
258 | flist_for_each(entry, &disk_list) { |
259 | __du = flist_entry(entry, struct disk_util, list); | |
9f8f2064 | 260 | |
cd991b9e JA |
261 | dprint(FD_DISKUTIL, "found %s in list\n", __du->name); |
262 | ||
9f8f2064 | 263 | if (!strcmp(du->name, __du->name)) { |
c97bd0fa JA |
264 | disk_util_free(du); |
265 | return __du; | |
9f8f2064 JA |
266 | } |
267 | } | |
268 | ||
cd991b9e JA |
269 | dprint(FD_DISKUTIL, "add %s to list\n", du->name); |
270 | ||
9f8f2064 JA |
271 | fio_gettime(&du->time, NULL); |
272 | get_io_ticks(du, &du->last_dus); | |
273 | ||
01743ee1 | 274 | flist_add_tail(&du->list, &disk_list); |
67423013 | 275 | find_add_disk_slaves(td, path, du); |
c97bd0fa | 276 | return du; |
9f8f2064 JA |
277 | } |
278 | ||
67423013 | 279 | |
9f8f2064 JA |
280 | static int check_dev_match(int majdev, int mindev, char *path) |
281 | { | |
282 | int major, minor; | |
9f8f2064 | 283 | |
67423013 | 284 | if (read_block_dev_entry(path, &major, &minor)) |
9f8f2064 | 285 | return 1; |
9f8f2064 | 286 | |
c97bd0fa | 287 | if (majdev == major && mindev == minor) |
9f8f2064 | 288 | return 0; |
9f8f2064 | 289 | |
9f8f2064 JA |
290 | return 1; |
291 | } | |
292 | ||
ac0be2ac | 293 | static int find_block_dir(int majdev, int mindev, char *path, int link_ok) |
9f8f2064 JA |
294 | { |
295 | struct dirent *dir; | |
296 | struct stat st; | |
297 | int found = 0; | |
298 | DIR *D; | |
299 | ||
300 | D = opendir(path); | |
301 | if (!D) | |
302 | return 0; | |
303 | ||
304 | while ((dir = readdir(D)) != NULL) { | |
305 | char full_path[256]; | |
306 | ||
307 | if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) | |
308 | continue; | |
309 | ||
310 | sprintf(full_path, "%s/%s", path, dir->d_name); | |
311 | ||
312 | if (!strcmp(dir->d_name, "dev")) { | |
313 | if (!check_dev_match(majdev, mindev, full_path)) { | |
314 | found = 1; | |
315 | break; | |
316 | } | |
317 | } | |
318 | ||
ac0be2ac JA |
319 | if (link_ok) { |
320 | if (stat(full_path, &st) == -1) { | |
321 | perror("stat"); | |
322 | break; | |
323 | } | |
324 | } else { | |
325 | if (lstat(full_path, &st) == -1) { | |
326 | perror("stat"); | |
327 | break; | |
328 | } | |
9f8f2064 JA |
329 | } |
330 | ||
331 | if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode)) | |
332 | continue; | |
333 | ||
ac0be2ac | 334 | found = find_block_dir(majdev, mindev, full_path, 0); |
9f8f2064 JA |
335 | if (found) { |
336 | strcpy(path, full_path); | |
337 | break; | |
338 | } | |
339 | } | |
340 | ||
341 | closedir(D); | |
342 | return found; | |
343 | } | |
344 | ||
c97bd0fa JA |
345 | static struct disk_util *__init_per_file_disk_util(struct thread_data *td, |
346 | int majdev, int mindev, | |
347 | char *path) | |
9f8f2064 JA |
348 | { |
349 | struct stat st; | |
67423013 | 350 | char tmp[PATH_MAX]; |
9f8f2064 JA |
351 | char *p; |
352 | ||
67423013 ST |
353 | /* |
354 | * If there's a ../queue/ directory there, we are inside a partition. | |
355 | * Check if that is the case and jump back. For loop/md/dm etc we | |
356 | * are already in the right spot. | |
357 | */ | |
358 | sprintf(tmp, "%s/../queue", path); | |
359 | if (!stat(tmp, &st)) { | |
360 | p = dirname(path); | |
361 | sprintf(tmp, "%s/queue", p); | |
362 | if (stat(tmp, &st)) { | |
363 | log_err("unknown sysfs layout\n"); | |
c97bd0fa | 364 | return NULL; |
9f8f2064 | 365 | } |
67423013 ST |
366 | strncpy(tmp, p, PATH_MAX - 1); |
367 | sprintf(path, "%s", tmp); | |
9f8f2064 JA |
368 | } |
369 | ||
67423013 ST |
370 | if (td->o.ioscheduler && !td->sysfs_root) |
371 | td->sysfs_root = strdup(path); | |
372 | ||
c97bd0fa | 373 | return disk_util_add(td, majdev, mindev, path); |
67423013 ST |
374 | } |
375 | ||
c97bd0fa JA |
376 | static struct disk_util *init_per_file_disk_util(struct thread_data *td, |
377 | char *filename) | |
67423013 ST |
378 | { |
379 | ||
380 | char foo[PATH_MAX]; | |
381 | struct disk_util *du; | |
382 | int mindev, majdev; | |
383 | ||
c97bd0fa JA |
384 | if (get_device_numbers(filename, &majdev, &mindev)) |
385 | return NULL; | |
67423013 ST |
386 | |
387 | dprint(FD_DISKUTIL, "%s belongs to maj/min %d/%d\n", filename, majdev, | |
388 | mindev); | |
cd991b9e | 389 | |
e11c410c JA |
390 | du = disk_util_exists(majdev, mindev); |
391 | if (du) { | |
392 | if (td->o.ioscheduler && !td->sysfs_root) | |
393 | td->sysfs_root = strdup(du->sysfs_root); | |
394 | ||
c97bd0fa | 395 | return du; |
e11c410c | 396 | } |
9f8f2064 JA |
397 | |
398 | /* | |
399 | * for an fs without a device, we will repeatedly stat through | |
400 | * sysfs which can take oodles of time for thousands of files. so | |
401 | * cache the last lookup and compare with that before going through | |
402 | * everything again. | |
403 | */ | |
404 | if (mindev == last_mindev && majdev == last_majdev) | |
c97bd0fa | 405 | return last_du; |
9f8f2064 JA |
406 | |
407 | last_mindev = mindev; | |
408 | last_majdev = majdev; | |
5ec10eaa | 409 | |
9f8f2064 | 410 | sprintf(foo, "/sys/block"); |
ac0be2ac | 411 | if (!find_block_dir(majdev, mindev, foo, 1)) |
c97bd0fa | 412 | return NULL; |
9f8f2064 | 413 | |
c97bd0fa | 414 | return __init_per_file_disk_util(td, majdev, mindev, foo); |
67423013 | 415 | } |
9f8f2064 | 416 | |
c97bd0fa JA |
417 | static struct disk_util *__init_disk_util(struct thread_data *td, |
418 | struct fio_file *f) | |
67423013 ST |
419 | { |
420 | return init_per_file_disk_util(td, f->file_name); | |
9f8f2064 JA |
421 | } |
422 | ||
423 | void init_disk_util(struct thread_data *td) | |
424 | { | |
425 | struct fio_file *f; | |
426 | unsigned int i; | |
427 | ||
428 | if (!td->o.do_disk_util || | |
429 | (td->io_ops->flags & (FIO_DISKLESSIO | FIO_NODISKUTIL))) | |
430 | return; | |
431 | ||
432 | for_each_file(td, f, i) | |
c97bd0fa | 433 | f->du = __init_disk_util(td, f); |
9f8f2064 JA |
434 | } |
435 | ||
043b74b9 ST |
436 | static void aggregate_slaves_stats(struct disk_util *masterdu) |
437 | { | |
438 | struct disk_util_stat *dus; | |
439 | struct flist_head *entry; | |
440 | struct disk_util *slavedu; | |
441 | double util, max_util = 0; | |
442 | int slavecount = 0; | |
443 | ||
5e28df3c JA |
444 | unsigned merges[2] = { 0, }; |
445 | unsigned ticks[2] = { 0, }; | |
446 | unsigned time_in_queue = { 0, }; | |
447 | unsigned long long sectors[2] = { 0, }; | |
448 | unsigned ios[2] = { 0, }; | |
043b74b9 ST |
449 | |
450 | flist_for_each(entry, &masterdu->slaves) { | |
451 | slavedu = flist_entry(entry, struct disk_util, slavelist); | |
452 | dus = &slavedu->dus; | |
453 | ios[0] += dus->ios[0]; | |
454 | ios[1] += dus->ios[1]; | |
455 | merges[0] += dus->merges[0]; | |
456 | merges[1] += dus->merges[1]; | |
457 | sectors[0] += dus->sectors[0]; | |
458 | sectors[1] += dus->sectors[1]; | |
459 | ticks[0] += dus->ticks[0]; | |
460 | ticks[1] += dus->ticks[1]; | |
461 | time_in_queue += dus->time_in_queue; | |
462 | ++slavecount; | |
463 | ||
464 | util = (double) (100 * dus->io_ticks / (double) slavedu->msec); | |
465 | /* System utilization is the utilization of the | |
466 | * component with the highest utilization. | |
467 | */ | |
468 | if (util > max_util) | |
469 | max_util = util; | |
470 | ||
471 | } | |
472 | ||
473 | if (max_util > 100.0) | |
474 | max_util = 100.0; | |
475 | ||
476 | log_info(", aggrios=%u/%u, aggrmerge=%u/%u, aggrticks=%u/%u," | |
477 | " aggrin_queue=%u, aggrutil=%3.2f%%", | |
478 | ios[0]/slavecount, ios[1]/slavecount, | |
479 | merges[0]/slavecount, merges[1]/slavecount, | |
480 | ticks[0]/slavecount, ticks[1]/slavecount, | |
481 | time_in_queue/slavecount, max_util); | |
482 | ||
483 | } | |
484 | ||
9f8f2064 JA |
485 | void show_disk_util(void) |
486 | { | |
487 | struct disk_util_stat *dus; | |
01743ee1 | 488 | struct flist_head *entry, *next; |
9f8f2064 JA |
489 | struct disk_util *du; |
490 | double util; | |
491 | ||
01743ee1 | 492 | if (flist_empty(&disk_list)) |
0a839f30 JA |
493 | return; |
494 | ||
9f8f2064 JA |
495 | log_info("\nDisk stats (read/write):\n"); |
496 | ||
01743ee1 JA |
497 | flist_for_each(entry, &disk_list) { |
498 | du = flist_entry(entry, struct disk_util, list); | |
9f8f2064 JA |
499 | dus = &du->dus; |
500 | ||
501 | util = (double) 100 * du->dus.io_ticks / (double) du->msec; | |
502 | if (util > 100.0) | |
503 | util = 100.0; | |
504 | ||
67423013 ST |
505 | /* If this node is the slave of a master device, as |
506 | * happens in case of software RAIDs, inward-indent | |
507 | * this stats line to reflect a master-slave | |
508 | * relationship. Because the master device gets added | |
509 | * before the slave devices, we can safely assume that | |
510 | * the master's stats line has been displayed in a | |
511 | * previous iteration of this loop. | |
512 | */ | |
513 | if(!flist_empty(&du->slavelist)) | |
514 | log_info(" "); | |
515 | ||
5ec10eaa | 516 | log_info(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, " |
043b74b9 | 517 | "in_queue=%u, util=%3.2f%%", du->name, |
5ec10eaa JA |
518 | dus->ios[0], dus->ios[1], |
519 | dus->merges[0], dus->merges[1], | |
520 | dus->ticks[0], dus->ticks[1], | |
521 | dus->time_in_queue, util); | |
043b74b9 ST |
522 | |
523 | /* If the device has slaves, aggregate the stats for | |
524 | * those slave devices also. | |
525 | */ | |
526 | if(!flist_empty(&du->slaves)) | |
527 | aggregate_slaves_stats(du); | |
528 | ||
529 | log_info("\n"); | |
9f8f2064 JA |
530 | } |
531 | ||
532 | /* | |
533 | * now free the list | |
534 | */ | |
01743ee1 JA |
535 | flist_for_each_safe(entry, next, &disk_list) { |
536 | flist_del(entry); | |
537 | du = flist_entry(entry, struct disk_util, list); | |
c97bd0fa | 538 | disk_util_free(du); |
9f8f2064 JA |
539 | } |
540 | } |