#include <sched.h>
#include <ctype.h>
#include <getopt.h>
-#include <sys/mman.h>
#include "blktrace.h"
+static char blktrace_version[] = "0.90";
+
#define BUF_SIZE (128 *1024)
#define BUF_NR (4)
char *long_form;
};
-struct mask_map mask_maps[] = {
+static struct mask_map mask_maps[] = {
DECLARE_MASK_MAP(READ),
DECLARE_MASK_MAP(WRITE),
DECLARE_MASK_MAP(BARRIER),
DECLARE_MASK_MAP(PC),
};
-#define S_OPTS "d:a:A:r:o:kw:"
+#define S_OPTS "d:a:A:r:o:kw:vb:n:"
static struct option l_opts[] = {
{
.name = "dev",
.flag = NULL,
.val = 'w'
},
+ {
+ .name = "version",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'v'
+ },
+ {
+ .name = "buffer size (in KiB)",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'b'
+ },
+ {
+ .name = "nr of sub buffers",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'n'
+ },
};
struct thread_information {
static char *output_name;
static int act_mask = ~0U;
static int kill_running_trace;
-static int use_mmap;
+static unsigned int buf_size = BUF_SIZE;
+static unsigned int buf_nr = BUF_NR;
#define is_done() (*(volatile int *)(&done))
static volatile int done;
static int find_mask_map(char *string)
{
- int i;
+ unsigned int i;
for (i = 0; i < sizeof(mask_maps)/sizeof(mask_maps[0]); i++)
if (COMPARE_MASK_MAP(&mask_maps[i], string))
struct blk_user_trace_setup buts;
memset(&buts, 0, sizeof(buts));
- buts.buf_size = BUF_SIZE;
- buts.buf_nr = BUF_NR;
+ buts.buf_size = buf_size;
+ buts.buf_nr = buf_nr;
buts.act_mask = act_mask;
if (ioctl(dip->fd, BLKSTARTTRACE, &buts) < 0) {
stop_trace(dip);
}
-static int get_data_read(struct thread_information *tip, void *buf, int len)
+static int read_data(struct thread_information *tip, void *buf, int len)
{
char *p = buf;
int ret, bytes_left = len;
while (!is_done() && bytes_left > 0) {
ret = read(tip->fd, p, bytes_left);
- if (ret == len)
+ if (ret == bytes_left)
return 0;
if (ret < 0) {
perror(tip->fn);
fprintf(stderr,"Thread %d failed read of %s\n",
tip->cpu, tip->fn);
- exit_trace(1);
+ break;
} else if (ret > 0) {
- fprintf(stderr,"Thread %d misread %s %d,%d\n",
- tip->cpu, tip->fn, ret, len);
- exit_trace(1);
- } else {
p += ret;
bytes_left -= ret;
- }
-
- usleep(10000);
+ } else
+ usleep(1000);
}
return -1;
}
-static int get_data_mmap(struct thread_information *tip, void *buf, int len,
- int check_magic)
+static int write_data(int fd, void *buf, unsigned int buf_len)
{
- if (len > (BUF_SIZE * (tip->buf_subbuf + 1)) - tip->buf_offset) {
- tip->buf_subbuf++;
- if (tip->buf_subbuf == BUF_NR)
- tip->buf_subbuf = 0;
-
- tip->buf_offset = tip->buf_subbuf * BUF_SIZE;
- }
-
- while (1) {
- struct blk_io_trace *t = buf;
-
- memcpy(buf, tip->buf + tip->buf_offset, len);
+ int ret, bytes_left;
+ char *p = buf;
- if (!check_magic)
+ bytes_left = buf_len;
+ while (bytes_left > 0) {
+ ret = write(fd, p, bytes_left);
+ if (ret == bytes_left)
break;
- if (CHECK_MAGIC(t) && t->sequence >= tip->sequence) {
- tip->sequence = t->sequence;
- break;
+ if (ret < 0) {
+ perror("write");
+ return 1;
+ } else if (ret > 0) {
+ p += ret;
+ bytes_left -= ret;
+ } else {
+ fprintf(stderr, "Zero write?\n");
+ return 1;
}
-
- if (is_done())
- return -1;
-
- usleep(10000);
}
- tip->buf_offset += len;
return 0;
}
-static int get_data(struct thread_information *tip, void *buf, int len,
- int check_magic)
-{
- if (tip->buf)
- return get_data_mmap(tip, buf, len, check_magic);
- else
- return get_data_read(tip, buf, len);
-}
-
-static void *extract_data(struct thread_information *tip, char *ofn, int nb)
+static void *extract_data(struct thread_information *tip, int nb)
{
unsigned char *buf;
buf = malloc(nb);
- if (!get_data(tip, buf, nb, 0))
+ if (!read_data(tip, buf, nb))
return buf;
free(buf);
- exit_trace(1);
return NULL;
}
+/*
+ * trace may start inside 'bit' or may need to be gotten further on
+ */
+static int get_event_slow(struct thread_information *tip,
+ struct blk_io_trace *bit)
+{
+ const int inc = sizeof(__u32);
+ struct blk_io_trace foo;
+ int offset;
+ void *p;
+
+ /*
+ * check is trace is inside
+ */
+ offset = 0;
+ p = bit;
+ while (offset < sizeof(*bit)) {
+ p += inc;
+ offset += inc;
+
+ memcpy(&foo, p, inc);
+
+ if (CHECK_MAGIC(&foo))
+ break;
+ }
+
+ /*
+ * part trace found inside, read the rest
+ */
+ if (offset < sizeof(*bit)) {
+ int good_bytes = sizeof(*bit) - offset;
+
+ memmove(bit, p, good_bytes);
+ p = (void *) bit + good_bytes;
+
+ return read_data(tip, p, offset);
+ }
+
+ /*
+ * nothing found, keep looking for start of trace
+ */
+ do {
+ if (read_data(tip, bit, sizeof(bit->magic)))
+ return -1;
+ } while (!CHECK_MAGIC(bit));
+
+ /*
+ * now get the rest of it
+ */
+ p = &bit->sequence;
+ if (!read_data(tip, p, sizeof(*bit) - inc))
+ return -1;
+
+ return 0;
+}
+
+/*
+ * Sometimes relayfs screws us a little, if an event crosses a sub buffer
+ * boundary. So keep looking forward in the trace data until an event
+ * is found
+ */
+static int get_event(struct thread_information *tip, struct blk_io_trace *bit)
+{
+ /*
+ * optimize for the common fast case, a full trace read that
+ * succeeds
+ */
+ if (read_data(tip, bit, sizeof(*bit)))
+ return -1;
+
+ if (CHECK_MAGIC(bit))
+ return 0;
+
+ /*
+ * ok that didn't work, the event may start somewhere inside the
+ * trace itself
+ */
+ return get_event_slow(tip, bit);
+}
+
static inline void tip_fd_unlock(struct thread_information *tip)
{
if (tip->fd_lock)
static void *extract(void *arg)
{
struct thread_information *tip = arg;
- int ret, pdu_len;
- char dp[64], *pdu_data;
+ int pdu_len;
+ char *pdu_data;
struct blk_io_trace t;
pid_t pid = getpid();
cpu_set_t cpu_mask;
exit_trace(1);
}
- if (use_mmap) {
- tip->buf = mmap(NULL, BUF_SIZE * BUF_NR, PROT_READ,
- MAP_PRIVATE | MAP_POPULATE, tip->fd, 0);
- if (tip->buf == MAP_FAILED) {
- perror("mmap");
- exit_trace(1);
- }
- }
-
pdu_data = NULL;
while (!is_done()) {
- if (get_data(tip, &t, sizeof(t), 1))
+ if (get_event(tip, &t))
break;
if (verify_trace(&t))
- exit_trace(1);
+ break;
pdu_len = t.pdu_len;
trace_to_be(&t);
- if (pdu_len)
- pdu_data = extract_data(tip, dp, pdu_len);
+ if (pdu_len) {
+ pdu_data = extract_data(tip, pdu_len);
+ if (!pdu_data)
+ break;
+ }
/*
* now we have both trace and payload, get a lock on the
*/
tip_fd_lock(tip);
- ret = write(tip->ofd, &t, sizeof(t));
- if (ret < 0) {
- fprintf(stderr,"Thread %d failed write\n", tip->cpu);
+ if (write_data(tip->ofd, &t, sizeof(t))) {
tip_fd_unlock(tip);
- exit_trace(1);
+ break;
}
- if (pdu_data) {
- ret = write(tip->ofd, pdu_data, pdu_len);
- if (ret != pdu_len) {
- perror("write pdu data");
- tip_fd_unlock(tip);
- exit_trace(1);
- }
+ if (pdu_data && write_data(tip->ofd, pdu_data, pdu_len)) {
+ tip_fd_unlock(tip);
+ break;
+ }
+ tip_fd_unlock(tip);
+
+ if (pdu_data) {
free(pdu_data);
pdu_data = NULL;
}
- tip_fd_unlock(tip);
tip->events_processed++;
}
+ exit_trace(1);
return NULL;
}
tip->ofd = dup(STDOUT_FILENO);
tip->fd_lock = &stdout_mutex;
} else {
- if (output_name)
- sprintf(op, "%s.%d", output_name, tip->cpu);
- else
- sprintf(op, "%s_out.%d",
+ if (output_name) {
+ sprintf(op, "%s.blktrace.%d", output_name,
+ tip->cpu);
+ } else {
+ sprintf(op, "%s.blktrace.%d",
dip->buts_name, tip->cpu);
+ }
tip->ofd = open(op, O_CREAT|O_TRUNC|O_WRONLY, 0644);
}
static void close_thread(struct thread_information *tip)
{
- if (tip->buf)
- munmap(tip->buf, BUF_SIZE * BUF_NR);
-
if (tip->fd != -1)
close(tip->fd);
if (tip->ofd != -1)
printf(" Total: %20lld events\n", events_processed);
}
}
-
+
+static char usage_str[] = \
+ "-d <dev> [ -r relay path ] [ -o <output> ] [-k ] [ -w time ]\n" \
+ "[ -a action ] [ -A action mask ] [ -v ]\n\n" \
+ "\t-d Use specified device. May also be given last after options\n" \
+ "\t-r Path to mounted relayfs, defaults to /relay\n" \
+ "\t-o File(s) to send output to\n" \
+ "\t-k Kill a running trace\n" \
+ "\t-w Stop after defined time, in seconds\n" \
+ "\t-a Only trace specified actions. See documentation\n" \
+ "\t-A Give trace mask as a single value. See documentation\n" \
+ "\t-b Sub buffer size in KiB\n" \
+ "\t-n Number of sub buffers\n" \
+ "\t-v Print program version info\n\n";
+
static void show_usage(char *program)
{
- fprintf(stderr,"Usage: %s [-d <dev>] "
- "[-a <trace> [-a <trace>]] <dev>\n",
- program);
+ fprintf(stderr, "Usage: %s %s %s",program, blktrace_version, usage_str);
}
-static void handle_sigint(int sig)
+static void handle_sigint(__attribute__((__unused__)) int sig)
{
done = 1;
}
return 1;
}
break;
-
+ case 'v':
+ printf("%s version %s\n", argv[0], blktrace_version);
+ return 0;
+ case 'b':
+ buf_size = atoi(optarg);
+ if (buf_size <= 0) {
+ fprintf(stderr,
+ "Invalid buffer size (%d)\n", buf_size);
+ return 1;
+ }
+ buf_size <<= 10;
+ break;
+ case 'n':
+ buf_nr = atoi(optarg);
+ if (buf_nr <= 0) {
+ fprintf(stderr,
+ "Invalid buffer nr (%d)\n", buf_nr);
+ return 1;
+ }
+ break;
default:
show_usage(argv[0]);
return 1;