1 // SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause)
2 // Copyright (c) 2020 Wenbo Zhang
3 //
4 // Based on biolatency(8) from BCC by Brendan Gregg.
5 // 15-Jun-2020 Wenbo Zhang Created this.
6 #include <argp.h>
7 #include <signal.h>
8 #include <stdio.h>
9 #include <unistd.h>
10 #include <fcntl.h>
11 #include <time.h>
12 #include <bpf/libbpf.h>
13 #include <sys/resource.h>
14 #include <bpf/bpf.h>
15 #include "blk_types.h"
16 #include "biolatency.h"
17 #include "biolatency.skel.h"
18 #include "trace_helpers.h"
19
20 #define ARRAY_SIZE(x) (sizeof(x) / sizeof(*(x)))
21
22 static struct env {
23 char *disk;
24 time_t interval;
25 int times;
26 bool timestamp;
27 bool queued;
28 bool per_disk;
29 bool per_flag;
30 bool milliseconds;
31 bool verbose;
32 char *cgroupspath;
33 bool cg;
34 } env = {
35 .interval = 99999999,
36 .times = 99999999,
37 };
38
39 static volatile bool exiting;
40
41 const char *argp_program_version = "biolatency 0.1";
42 const char *argp_program_bug_address =
43 "https://github.com/iovisor/bcc/tree/master/libbpf-tools";
44 const char argp_program_doc[] =
45 "Summarize block device I/O latency as a histogram.\n"
46 "\n"
47 "USAGE: biolatency [--help] [-T] [-m] [-Q] [-D] [-F] [-d DISK] [-c CG] [interval] [count]\n"
48 "\n"
49 "EXAMPLES:\n"
50 " biolatency # summarize block I/O latency as a histogram\n"
51 " biolatency 1 10 # print 1 second summaries, 10 times\n"
52 " biolatency -mT 1 # 1s summaries, milliseconds, and timestamps\n"
53 " biolatency -Q # include OS queued time in I/O time\n"
54 " biolatency -D # show each disk device separately\n"
55 " biolatency -F # show I/O flags separately\n"
56 " biolatency -d sdc # Trace sdc only\n"
57 " biolatency -c CG # Trace process under cgroupsPath CG\n";
58
59 static const struct argp_option opts[] = {
60 { "timestamp", 'T', NULL, 0, "Include timestamp on output" },
61 { "milliseconds", 'm', NULL, 0, "Millisecond histogram" },
62 { "queued", 'Q', NULL, 0, "Include OS queued time in I/O time" },
63 { "disk", 'D', NULL, 0, "Print a histogram per disk device" },
64 { "flag", 'F', NULL, 0, "Print a histogram per set of I/O flags" },
65 { "disk", 'd', "DISK", 0, "Trace this disk only" },
66 { "verbose", 'v', NULL, 0, "Verbose debug output" },
67 { "cgroup", 'c', "/sys/fs/cgroup/unified", 0, "Trace process in cgroup path"},
68 { NULL, 'h', NULL, OPTION_HIDDEN, "Show the full help" },
69 {},
70 };
71
parse_arg(int key,char * arg,struct argp_state * state)72 static error_t parse_arg(int key, char *arg, struct argp_state *state)
73 {
74 static int pos_args;
75
76 switch (key) {
77 case 'h':
78 argp_state_help(state, stderr, ARGP_HELP_STD_HELP);
79 break;
80 case 'v':
81 env.verbose = true;
82 break;
83 case 'm':
84 env.milliseconds = true;
85 break;
86 case 'Q':
87 env.queued = true;
88 break;
89 case 'D':
90 env.per_disk = true;
91 break;
92 case 'F':
93 env.per_flag = true;
94 break;
95 case 'T':
96 env.timestamp = true;
97 break;
98 case 'c':
99 env.cgroupspath = arg;
100 env.cg = true;
101 break;
102 case 'd':
103 env.disk = arg;
104 if (strlen(arg) + 1 > DISK_NAME_LEN) {
105 fprintf(stderr, "invaild disk name: too long\n");
106 argp_usage(state);
107 }
108 break;
109 case ARGP_KEY_ARG:
110 errno = 0;
111 if (pos_args == 0) {
112 env.interval = strtol(arg, NULL, 10);
113 if (errno) {
114 fprintf(stderr, "invalid internal\n");
115 argp_usage(state);
116 }
117 } else if (pos_args == 1) {
118 env.times = strtol(arg, NULL, 10);
119 if (errno) {
120 fprintf(stderr, "invalid times\n");
121 argp_usage(state);
122 }
123 } else {
124 fprintf(stderr,
125 "unrecognized positional argument: %s\n", arg);
126 argp_usage(state);
127 }
128 pos_args++;
129 break;
130 default:
131 return ARGP_ERR_UNKNOWN;
132 }
133 return 0;
134 }
135
libbpf_print_fn(enum libbpf_print_level level,const char * format,va_list args)136 static int libbpf_print_fn(enum libbpf_print_level level, const char *format, va_list args)
137 {
138 if (level == LIBBPF_DEBUG && !env.verbose)
139 return 0;
140 return vfprintf(stderr, format, args);
141 }
142
sig_handler(int sig)143 static void sig_handler(int sig)
144 {
145 exiting = true;
146 }
147
print_cmd_flags(int cmd_flags)148 static void print_cmd_flags(int cmd_flags)
149 {
150 static struct { int bit; const char *str; } flags[] = {
151 { REQ_NOWAIT, "NoWait-" },
152 { REQ_BACKGROUND, "Background-" },
153 { REQ_RAHEAD, "ReadAhead-" },
154 { REQ_PREFLUSH, "PreFlush-" },
155 { REQ_FUA, "FUA-" },
156 { REQ_INTEGRITY, "Integrity-" },
157 { REQ_IDLE, "Idle-" },
158 { REQ_NOMERGE, "NoMerge-" },
159 { REQ_PRIO, "Priority-" },
160 { REQ_META, "Metadata-" },
161 { REQ_SYNC, "Sync-" },
162 };
163 static const char *ops[] = {
164 [REQ_OP_READ] = "Read",
165 [REQ_OP_WRITE] = "Write",
166 [REQ_OP_FLUSH] = "Flush",
167 [REQ_OP_DISCARD] = "Discard",
168 [REQ_OP_SECURE_ERASE] = "SecureErase",
169 [REQ_OP_ZONE_RESET] = "ZoneReset",
170 [REQ_OP_WRITE_SAME] = "WriteSame",
171 [REQ_OP_ZONE_RESET_ALL] = "ZoneResetAll",
172 [REQ_OP_WRITE_ZEROES] = "WriteZeroes",
173 [REQ_OP_ZONE_OPEN] = "ZoneOpen",
174 [REQ_OP_ZONE_CLOSE] = "ZoneClose",
175 [REQ_OP_ZONE_FINISH] = "ZoneFinish",
176 [REQ_OP_SCSI_IN] = "SCSIIn",
177 [REQ_OP_SCSI_OUT] = "SCSIOut",
178 [REQ_OP_DRV_IN] = "DrvIn",
179 [REQ_OP_DRV_OUT] = "DrvOut",
180 };
181 int i;
182
183 printf("flags = ");
184
185 for (i = 0; i < ARRAY_SIZE(flags); i++) {
186 if (cmd_flags & flags[i].bit)
187 printf("%s", flags[i].str);
188 }
189
190 if ((cmd_flags & REQ_OP_MASK) < ARRAY_SIZE(ops))
191 printf("%s", ops[cmd_flags & REQ_OP_MASK]);
192 else
193 printf("Unknown");
194 }
195
196 static
print_log2_hists(struct bpf_map * hists,struct partitions * partitions)197 int print_log2_hists(struct bpf_map *hists, struct partitions *partitions)
198 {
199 struct hist_key lookup_key = { .cmd_flags = -1 }, next_key;
200 const char *units = env.milliseconds ? "msecs" : "usecs";
201 const struct partition *partition;
202 int err, fd = bpf_map__fd(hists);
203 struct hist hist;
204
205 while (!bpf_map_get_next_key(fd, &lookup_key, &next_key)) {
206 err = bpf_map_lookup_elem(fd, &next_key, &hist);
207 if (err < 0) {
208 fprintf(stderr, "failed to lookup hist: %d\n", err);
209 return -1;
210 }
211 if (env.per_disk) {
212 partition = partitions__get_by_dev(partitions,
213 next_key.dev);
214 printf("\ndisk = %s\t", partition ? partition->name :
215 "Unknown");
216 }
217 if (env.per_flag)
218 print_cmd_flags(next_key.cmd_flags);
219 printf("\n");
220 print_log2_hist(hist.slots, MAX_SLOTS, units);
221 lookup_key = next_key;
222 }
223
224 lookup_key.cmd_flags = -1;
225 while (!bpf_map_get_next_key(fd, &lookup_key, &next_key)) {
226 err = bpf_map_delete_elem(fd, &next_key);
227 if (err < 0) {
228 fprintf(stderr, "failed to cleanup hist : %d\n", err);
229 return -1;
230 }
231 lookup_key = next_key;
232 }
233
234 return 0;
235 }
236
main(int argc,char ** argv)237 int main(int argc, char **argv)
238 {
239 struct partitions *partitions = NULL;
240 const struct partition *partition;
241 static const struct argp argp = {
242 .options = opts,
243 .parser = parse_arg,
244 .doc = argp_program_doc,
245 };
246 struct biolatency_bpf *obj;
247 struct tm *tm;
248 char ts[32];
249 time_t t;
250 int err;
251 int idx, cg_map_fd;
252 int cgfd = -1;
253
254 err = argp_parse(&argp, argc, argv, 0, NULL, NULL);
255 if (err)
256 return err;
257
258 libbpf_set_strict_mode(LIBBPF_STRICT_ALL);
259 libbpf_set_print(libbpf_print_fn);
260
261 obj = biolatency_bpf__open();
262 if (!obj) {
263 fprintf(stderr, "failed to open BPF object\n");
264 return 1;
265 }
266
267 partitions = partitions__load();
268 if (!partitions) {
269 fprintf(stderr, "failed to load partitions info\n");
270 goto cleanup;
271 }
272
273 /* initialize global data (filtering options) */
274 if (env.disk) {
275 partition = partitions__get_by_name(partitions, env.disk);
276 if (!partition) {
277 fprintf(stderr, "invaild partition name: not exist\n");
278 goto cleanup;
279 }
280 obj->rodata->filter_dev = true;
281 obj->rodata->targ_dev = partition->dev;
282 }
283 obj->rodata->targ_per_disk = env.per_disk;
284 obj->rodata->targ_per_flag = env.per_flag;
285 obj->rodata->targ_ms = env.milliseconds;
286 obj->rodata->targ_queued = env.queued;
287 obj->rodata->filter_cg = env.cg;
288
289 err = biolatency_bpf__load(obj);
290 if (err) {
291 fprintf(stderr, "failed to load BPF object: %d\n", err);
292 goto cleanup;
293 }
294
295 /* update cgroup path fd to map */
296 if (env.cg) {
297 idx = 0;
298 cg_map_fd = bpf_map__fd(obj->maps.cgroup_map);
299 cgfd = open(env.cgroupspath, O_RDONLY);
300 if (cgfd < 0) {
301 fprintf(stderr, "Failed opening Cgroup path: %s", env.cgroupspath);
302 goto cleanup;
303 }
304 if (bpf_map_update_elem(cg_map_fd, &idx, &cgfd, BPF_ANY)) {
305 fprintf(stderr, "Failed adding target cgroup to map");
306 goto cleanup;
307 }
308 }
309
310 if (env.queued) {
311 obj->links.block_rq_insert = bpf_program__attach(obj->progs.block_rq_insert);
312 if (!obj->links.block_rq_insert) {
313 err = -errno;
314 fprintf(stderr, "failed to attach: %s\n", strerror(-err));
315 goto cleanup;
316 }
317 }
318 obj->links.block_rq_issue = bpf_program__attach(obj->progs.block_rq_issue);
319 if (!obj->links.block_rq_issue) {
320 err = -errno;
321 fprintf(stderr, "failed to attach: %s\n", strerror(-err));
322 goto cleanup;
323 }
324 obj->links.block_rq_complete = bpf_program__attach(obj->progs.block_rq_complete);
325 if (!obj->links.block_rq_complete) {
326 err = -errno;
327 fprintf(stderr, "failed to attach: %s\n", strerror(-err));
328 goto cleanup;
329 }
330
331 signal(SIGINT, sig_handler);
332
333 printf("Tracing block device I/O... Hit Ctrl-C to end.\n");
334
335 /* main: poll */
336 while (1) {
337 sleep(env.interval);
338 printf("\n");
339
340 if (env.timestamp) {
341 time(&t);
342 tm = localtime(&t);
343 strftime(ts, sizeof(ts), "%H:%M:%S", tm);
344 printf("%-8s\n", ts);
345 }
346
347 err = print_log2_hists(obj->maps.hists, partitions);
348 if (err)
349 break;
350
351 if (exiting || --env.times == 0)
352 break;
353 }
354
355 cleanup:
356 biolatency_bpf__destroy(obj);
357 partitions__free(partitions);
358 if (cgfd > 0)
359 close(cgfd);
360
361 return err != 0;
362 }
363