#include <sys/ioctl.h>
#include <sys/param.h>
#include <sys/statfs.h>
+#include <sys/poll.h>
+#include <sys/mman.h>
+#include <sys/socket.h>
#include <stdio.h>
#include <stdlib.h>
#include <sched.h>
#include <ctype.h>
#include <getopt.h>
-#include <sys/mman.h>
+#include <errno.h>
+#include <netinet/in.h>
+#include <arpa/inet.h>
+#include <netdb.h>
+#include <sys/sendfile.h>
#include "blktrace.h"
+#include "barrier.h"
-static char blktrace_version[] = "0.90";
+static char blktrace_version[] = "0.99";
-#define BUF_SIZE (128 *1024)
+/*
+ * You may want to increase this even more, if you are logging at a high
+ * rate and see skipped/missed events
+ */
+#define BUF_SIZE (512 * 1024)
#define BUF_NR (4)
-#define RELAYFS_TYPE 0xF0B4A981
-
-#define DECLARE_MASK_MAP(mask) { BLK_TC_##mask, #mask, "BLK_TC_"#mask }
-#define COMPARE_MASK_MAP(mmp, str) \
- (!strcasecmp((mmp)->short_form, (str)) || \
- !strcasecmp((mmp)->long_form, (str)))
-
-#define VALID_SET(x) ((1 <= (x)) && ((x) < (1 << BLK_TC_SHIFT)))
+#define OFILE_BUF (128 * 1024)
-struct mask_map {
- int mask;
- char *short_form;
- char *long_form;
-};
-
-struct mask_map mask_maps[] = {
- DECLARE_MASK_MAP(READ),
- DECLARE_MASK_MAP(WRITE),
- DECLARE_MASK_MAP(BARRIER),
- DECLARE_MASK_MAP(SYNC),
- DECLARE_MASK_MAP(QUEUE),
- DECLARE_MASK_MAP(REQUEUE),
- DECLARE_MASK_MAP(ISSUE),
- DECLARE_MASK_MAP(COMPLETE),
- DECLARE_MASK_MAP(FS),
- DECLARE_MASK_MAP(PC),
-};
+#define RELAYFS_TYPE 0xF0B4A981
-#define S_OPTS "d:a:A:r:o:kw:v"
+#define S_OPTS "d:a:A:r:o:kw:Vb:n:D:lh:p:"
static struct option l_opts[] = {
{
.name = "dev",
.name = "version",
.has_arg = no_argument,
.flag = NULL,
- .val = 'v'
+ .val = 'V'
+ },
+ {
+ .name = "buffer-size",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'b'
},
+ {
+ .name = "num-sub-buffers",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'n'
+ },
+ {
+ .name = "output-dir",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'D'
+ },
+ {
+ .name = "listen",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'l'
+ },
+ {
+ .name = "host",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'h'
+ },
+ {
+ .name = "port",
+ .has_arg = required_argument,
+ .flag = NULL,
+ .val = 'p'
+ },
+ {
+ .name = NULL,
+ }
+};
+
+struct tip_subbuf {
+ void *buf;
+ unsigned int len;
+ unsigned int max_len;
+};
+
+#define FIFO_SIZE (1024) /* should be plenty big! */
+#define CL_SIZE (128) /* cache line, any bigger? */
+
+struct tip_subbuf_fifo {
+ int tail __attribute__((aligned(CL_SIZE)));
+ int head __attribute__((aligned(CL_SIZE)));
+ struct tip_subbuf *q[FIFO_SIZE];
};
struct thread_information {
pthread_t thread;
int fd;
+ void *fd_buf;
char fn[MAXPATHLEN + 64];
- void *buf;
- unsigned long buf_offset;
- unsigned int buf_subbuf;
- unsigned int sequence;
- pthread_mutex_t *fd_lock;
- int ofd;
+ FILE *ofile;
+ char *ofile_buffer;
+ int ofile_stdout;
+ int ofile_mmap;
unsigned long events_processed;
+ unsigned long long data_read;
struct device_information *device;
+
+ int exited;
+
+ /*
+ * piped fifo buffers
+ */
+ struct tip_subbuf_fifo fifo;
+ struct tip_subbuf *leftover_ts;
+
+ /*
+ * mmap controlled output files
+ */
+ unsigned long long fs_size;
+ unsigned long long fs_max_size;
+ unsigned long fs_off;
+ void *fs_buf;
+ unsigned long fs_buf_len;
};
struct device_information {
int fd;
char *path;
char buts_name[32];
- int trace_started;
+ volatile int trace_started;
+ unsigned long drop_count;
struct thread_information *threads;
};
/* command line option globals */
static char *relay_path;
static char *output_name;
+static char *output_dir;
static int act_mask = ~0U;
static int kill_running_trace;
-static int use_mmap;
+static unsigned long buf_size = BUF_SIZE;
+static unsigned long buf_nr = BUF_NR;
+static unsigned int page_size;
#define is_done() (*(volatile int *)(&done))
static volatile int done;
-static pthread_mutex_t stdout_mutex = PTHREAD_MUTEX_INITIALIZER;
+#define is_trace_stopped() (*(volatile int *)(&trace_stopped))
+static volatile int trace_stopped;
+
+#define is_stat_shown() (*(volatile int *)(&stat_shown))
+static volatile int stat_shown;
+
+int data_is_native = -1;
static void exit_trace(int status);
-static int find_mask_map(char *string)
+#define dip_tracing(dip) (*(volatile int *)(&(dip)->trace_started))
+#define dip_set_tracing(dip, v) ((dip)->trace_started = (v))
+
+#define __for_each_dip(__d, __i, __e) \
+ for (__i = 0, __d = device_information; __i < __e; __i++, __d++)
+
+#define for_each_dip(__d, __i) __for_each_dip(__d, __i, ndevs)
+#define for_each_tip(__d, __t, __j) \
+ for (__j = 0, __t = (__d)->threads; __j < ncpus; __j++, __t++)
+
+/*
+ * networking stuff follows. we include a magic number so we know whether
+ * to endianness convert or not
+ */
+struct blktrace_net_hdr {
+ u32 magic; /* same as trace magic */
+ char buts_name[32]; /* trace name */
+ u32 cpu; /* for which cpu */
+ u32 max_cpus;
+ u32 len; /* length of following trace data */
+};
+
+#define TRACE_NET_PORT (8462)
+
+enum {
+ Net_none = 0,
+ Net_server,
+ Net_client,
+};
+
+/*
+ * network cmd line params
+ */
+static char hostname[MAXHOSTNAMELEN];
+static int net_port = TRACE_NET_PORT;
+static int net_mode = 0;
+
+static int net_in_fd = -1;
+static int net_out_fd = -1;
+
+static void handle_sigint(__attribute__((__unused__)) int sig)
{
- int i;
+ done = 1;
+}
- for (i = 0; i < sizeof(mask_maps)/sizeof(mask_maps[0]); i++)
- if (COMPARE_MASK_MAP(&mask_maps[i], string))
- return mask_maps[i].mask;
+static int get_dropped_count(const char *buts_name)
+{
+ int fd;
+ char tmp[MAXPATHLEN + 64];
- return -1;
+ snprintf(tmp, sizeof(tmp), "%s/block/%s/dropped",
+ relay_path, buts_name);
+
+ fd = open(tmp, O_RDONLY);
+ if (fd < 0) {
+ /*
+ * this may be ok, if the kernel doesn't support dropped counts
+ */
+ if (errno == ENOENT)
+ return 0;
+
+ fprintf(stderr, "Couldn't open dropped file %s\n", tmp);
+ return -1;
+ }
+
+ if (read(fd, tmp, sizeof(tmp)) < 0) {
+ perror(tmp);
+ close(fd);
+ return -1;
+ }
+
+ close(fd);
+
+ return atoi(tmp);
}
static int start_trace(struct device_information *dip)
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) {
- perror("BLKSTARTTRACE");
+ if (ioctl(dip->fd, BLKTRACESETUP, &buts) < 0) {
+ perror("BLKTRACESETUP");
+ return 1;
+ }
+
+ if (ioctl(dip->fd, BLKTRACESTART) < 0) {
+ perror("BLKTRACESTART");
return 1;
}
memcpy(dip->buts_name, buts.name, sizeof(dip->buts_name));
- dip->trace_started = 1;
+ dip_set_tracing(dip, 1);
return 0;
}
static void stop_trace(struct device_information *dip)
{
- if (dip->trace_started || kill_running_trace) {
- if (ioctl(dip->fd, BLKSTOPTRACE) < 0)
- perror("BLKSTOPTRACE");
+ if (dip_tracing(dip) || kill_running_trace) {
+ dip_set_tracing(dip, 0);
+
+ if (ioctl(dip->fd, BLKTRACESTOP) < 0)
+ perror("BLKTRACESTOP");
+ if (ioctl(dip->fd, BLKTRACETEARDOWN) < 0)
+ perror("BLKTRACETEARDOWN");
+
close(dip->fd);
- dip->trace_started = 0;
+ dip->fd = -1;
}
}
struct device_information *dip;
int i;
- for (dip = device_information, i = 0; i < ndevs; i++, dip++)
+ for_each_dip(dip, i) {
+ dip->drop_count = get_dropped_count(dip->buts_name);
stop_trace(dip);
+ }
}
-static int get_data_read(struct thread_information *tip, void *buf, int len)
+static void wait_for_data(struct thread_information *tip)
{
- char *p = buf;
- int ret, bytes_left = len;
+ struct pollfd pfd = { .fd = tip->fd, .events = POLLIN };
- while (!is_done() && bytes_left > 0) {
- ret = read(tip->fd, p, bytes_left);
- if (ret == len)
- return 0;
+ do {
+ poll(&pfd, 1, 100);
+ if (pfd.revents & POLLIN)
+ break;
+ if (tip->ofile_stdout)
+ break;
+ } while (!is_done());
+}
- if (ret < 0) {
- perror(tip->fn);
- fprintf(stderr,"Thread %d failed read of %s\n",
- tip->cpu, tip->fn);
- exit_trace(1);
- } else if (ret > 0) {
- fprintf(stderr,"Thread %d misread %s %d,%d\n",
- tip->cpu, tip->fn, ret, len);
- exit_trace(1);
+static int read_data_file(struct thread_information *tip, void *buf, int len)
+{
+ int ret = 0;
+
+ do {
+ wait_for_data(tip);
+
+ ret = read(tip->fd, buf, len);
+ if (!ret)
+ continue;
+ else if (ret > 0)
+ return ret;
+ else {
+ if (errno != EAGAIN) {
+ perror(tip->fn);
+ fprintf(stderr,"Thread %d failed read of %s\n",
+ tip->cpu, tip->fn);
+ break;
+ }
+ continue;
+ }
+ } while (!is_done());
+
+ return ret;
+
+}
+
+static int read_data_net(struct thread_information *tip, void *buf, int len)
+{
+ unsigned int bytes_left = len;
+ int ret = 0;
+
+ do {
+ ret = recv(net_in_fd, buf, bytes_left, MSG_WAITALL);
+
+ if (!ret)
+ continue;
+ else if (ret < 0) {
+ if (errno != EAGAIN) {
+ perror(tip->fn);
+ fprintf(stderr, "server: failed read\n");
+ return 0;
+ }
+ continue;
} else {
- p += ret;
+ buf += ret;
bytes_left -= ret;
}
+ } while (!is_done() && bytes_left);
- usleep(10000);
+ return len;
+}
+
+static int read_data(struct thread_information *tip, void *buf, int len)
+{
+ if (net_mode == Net_server)
+ return read_data_net(tip, buf, len);
+
+ return read_data_file(tip, buf, len);
+}
+
+static inline struct tip_subbuf *
+subbuf_fifo_dequeue(struct thread_information *tip)
+{
+ const int head = tip->fifo.head;
+ const int next = (head + 1) & (FIFO_SIZE - 1);
+
+ if (head != tip->fifo.tail) {
+ struct tip_subbuf *ts = tip->fifo.q[head];
+
+ store_barrier();
+ tip->fifo.head = next;
+ return ts;
}
- return -1;
+ return NULL;
}
-static int get_data_mmap(struct thread_information *tip, void *buf, int len,
- int check_magic)
+static inline int subbuf_fifo_queue(struct thread_information *tip,
+ struct tip_subbuf *ts)
{
- if (len > (BUF_SIZE * (tip->buf_subbuf + 1)) - tip->buf_offset) {
- tip->buf_subbuf++;
- if (tip->buf_subbuf == BUF_NR)
- tip->buf_subbuf = 0;
+ const int tail = tip->fifo.tail;
+ const int next = (tail + 1) & (FIFO_SIZE - 1);
- tip->buf_offset = tip->buf_subbuf * BUF_SIZE;
+ if (next != tip->fifo.head) {
+ tip->fifo.q[tail] = ts;
+ store_barrier();
+ tip->fifo.tail = next;
+ return 0;
}
- while (1) {
- struct blk_io_trace *t = buf;
+ fprintf(stderr, "fifo too small!\n");
+ return 1;
+}
- memcpy(buf, tip->buf + tip->buf_offset, len);
+/*
+ * For file output, truncate and mmap the file appropriately
+ */
+static int mmap_subbuf(struct thread_information *tip, unsigned int maxlen)
+{
+ int ofd = fileno(tip->ofile);
+ int ret;
+
+ /*
+ * extend file, if we have to. use chunks of 16 subbuffers.
+ */
+ if (tip->fs_off + buf_size > tip->fs_buf_len) {
+ if (tip->fs_buf) {
+ munlock(tip->fs_buf, tip->fs_buf_len);
+ munmap(tip->fs_buf, tip->fs_buf_len);
+ tip->fs_buf = NULL;
+ }
- if (!check_magic)
- break;
+ tip->fs_off = tip->fs_size & (page_size - 1);
+ tip->fs_buf_len = (16 * buf_size) - tip->fs_off;
+ tip->fs_max_size += tip->fs_buf_len;
- if (CHECK_MAGIC(t) && t->sequence >= tip->sequence) {
- tip->sequence = t->sequence;
- break;
+ if (ftruncate(ofd, tip->fs_max_size) < 0) {
+ perror("ftruncate");
+ return -1;
}
-
- if (is_done())
+
+ tip->fs_buf = mmap(NULL, tip->fs_buf_len, PROT_WRITE,
+ MAP_SHARED, ofd, tip->fs_size - tip->fs_off);
+ if (tip->fs_buf == MAP_FAILED) {
+ perror("mmap");
return -1;
+ }
+ mlock(tip->fs_buf, tip->fs_buf_len);
+ }
- usleep(10000);
+ ret = read_data(tip, tip->fs_buf + tip->fs_off, maxlen);
+ if (ret >= 0) {
+ tip->data_read += ret;
+ tip->fs_size += ret;
+ tip->fs_off += ret;
+ return 0;
}
- tip->buf_offset += len;
- return 0;
+ return -1;
}
-static int get_data(struct thread_information *tip, void *buf, int len,
- int check_magic)
+/*
+ * Use the copy approach for pipes and network
+ */
+static int get_subbuf(struct thread_information *tip)
{
- if (tip->buf)
- return get_data_mmap(tip, buf, len, check_magic);
- else
- return get_data_read(tip, buf, len);
-}
+ struct tip_subbuf *ts = malloc(sizeof(*ts));
+ int ret;
-static void *extract_data(struct thread_information *tip, char *ofn, int nb)
-{
- unsigned char *buf;
+ ts->buf = malloc(buf_size);
+ ts->max_len = buf_size;
- buf = malloc(nb);
- if (!get_data(tip, buf, nb, 0))
- return buf;
+ ret = read_data(tip, ts->buf, ts->max_len);
+ if (ret > 0) {
+ ts->len = ret;
+ return subbuf_fifo_queue(tip, ts);
+ }
- free(buf);
- exit_trace(1);
- return NULL;
+ return ret;
}
-static inline void tip_fd_unlock(struct thread_information *tip)
+static void close_thread(struct thread_information *tip)
{
- if (tip->fd_lock)
- pthread_mutex_unlock(tip->fd_lock);
+ if (tip->fd != -1)
+ close(tip->fd);
+ if (tip->ofile)
+ fclose(tip->ofile);
+ if (tip->ofile_buffer)
+ free(tip->ofile_buffer);
+ if (tip->fd_buf)
+ free(tip->fd_buf);
+
+ tip->fd = -1;
+ tip->ofile = NULL;
+ tip->ofile_buffer = NULL;
+ tip->fd_buf = NULL;
}
-static inline void tip_fd_lock(struct thread_information *tip)
+static void tip_ftrunc_final(struct thread_information *tip)
{
- if (tip->fd_lock)
- pthread_mutex_lock(tip->fd_lock);
+ /*
+ * truncate to right size and cleanup mmap
+ */
+ if (tip->ofile_mmap) {
+ int ofd = fileno(tip->ofile);
+
+ if (tip->fs_buf)
+ munmap(tip->fs_buf, tip->fs_buf_len);
+
+ ftruncate(ofd, tip->fs_size);
+ }
}
-static void *extract(void *arg)
+static void *thread_main(void *arg)
{
struct thread_information *tip = arg;
- int ret, pdu_len;
- char dp[64], *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);
+ while (!is_done()) {
+ if (tip->ofile_mmap && net_mode != Net_client) {
+ if (mmap_subbuf(tip, buf_size))
+ break;
+ } else {
+ if (get_subbuf(tip))
+ break;
}
}
- pdu_data = NULL;
- while (!is_done()) {
- if (get_data(tip, &t, sizeof(t), 1))
- break;
+ tip_ftrunc_final(tip);
+ tip->exited = 1;
+ return NULL;
+}
+
+static int write_data_net(int fd, void *buf, unsigned int buf_len)
+{
+ unsigned int bytes_left = buf_len;
+ int ret;
+
+ while (bytes_left) {
+ ret = send(fd, buf, bytes_left, 0);
+ if (ret < 0) {
+ perror("send");
+ return 1;
+ }
- if (verify_trace(&t))
- exit_trace(1);
+ buf += ret;
+ bytes_left -= ret;
+ }
- pdu_len = t.pdu_len;
+ return 0;
+}
- trace_to_be(&t);
+static int flush_subbuf_net(struct thread_information *tip,
+ struct tip_subbuf *ts)
+{
+ struct blktrace_net_hdr hdr;
- if (pdu_len)
- pdu_data = extract_data(tip, dp, pdu_len);
+ hdr.magic = BLK_IO_TRACE_MAGIC;
+ strcpy(hdr.buts_name, tip->device->buts_name);
+ hdr.cpu = tip->cpu;
+ hdr.max_cpus = ncpus;
+ hdr.len = ts->len;
- /*
- * now we have both trace and payload, get a lock on the
- * output descriptor and send it off
- */
- tip_fd_lock(tip);
+ if (write_data_net(net_out_fd, &hdr, sizeof(hdr)))
+ return 1;
+
+ if (write_data_net(net_out_fd, ts->buf, ts->len))
+ return 1;
+
+ free(ts->buf);
+ free(ts);
+ return 0;
+}
+
+static int write_data(struct thread_information *tip, void *buf,
+ unsigned int buf_len)
+{
+ int ret;
+
+ if (!buf_len)
+ return 0;
+
+ while (1) {
+ ret = fwrite(buf, buf_len, 1, tip->ofile);
+ if (ret == 1)
+ break;
- ret = write(tip->ofd, &t, sizeof(t));
if (ret < 0) {
- fprintf(stderr,"Thread %d failed write\n", tip->cpu);
- tip_fd_unlock(tip);
- exit_trace(1);
+ perror("write");
+ return 1;
}
+ }
- 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 (tip->ofile_stdout)
+ fflush(tip->ofile);
+
+ return 0;
+}
- free(pdu_data);
- pdu_data = NULL;
+static int flush_subbuf_file(struct thread_information *tip,
+ struct tip_subbuf *ts)
+{
+ unsigned int offset = 0;
+ struct blk_io_trace *t;
+ int pdu_len, events = 0;
+
+ /*
+ * surplus from last run
+ */
+ if (tip->leftover_ts) {
+ struct tip_subbuf *prev_ts = tip->leftover_ts;
+
+ if (prev_ts->len + ts->len > prev_ts->max_len) {
+ prev_ts->max_len += ts->len;
+ prev_ts->buf = realloc(prev_ts->buf, prev_ts->max_len);
}
- tip_fd_unlock(tip);
+ memcpy(prev_ts->buf + prev_ts->len, ts->buf, ts->len);
+ prev_ts->len += ts->len;
+
+ free(ts->buf);
+ free(ts);
+
+ ts = prev_ts;
+ tip->leftover_ts = NULL;
+ }
+
+ while (offset + sizeof(*t) <= ts->len) {
+ t = ts->buf + offset;
+
+ if (verify_trace(t)) {
+ write_data(tip, ts->buf, offset);
+ return -1;
+ }
+
+ pdu_len = t->pdu_len;
+
+ if (offset + sizeof(*t) + pdu_len > ts->len)
+ break;
+
+ offset += sizeof(*t) + pdu_len;
tip->events_processed++;
+ tip->data_read += sizeof(*t) + pdu_len;
+ events++;
}
- return NULL;
+ if (write_data(tip, ts->buf, offset))
+ return -1;
+
+ /*
+ * leftover bytes, save them for next time
+ */
+ if (offset != ts->len) {
+ tip->leftover_ts = ts;
+ ts->len -= offset;
+ memmove(ts->buf, ts->buf + offset, ts->len);
+ } else {
+ free(ts->buf);
+ free(ts);
+ }
+
+ return events;
+}
+
+static int write_tip_events(struct thread_information *tip)
+{
+ struct tip_subbuf *ts = subbuf_fifo_dequeue(tip);
+
+ if (ts) {
+ if (net_mode == Net_client)
+ return flush_subbuf_net(tip, ts);
+
+ return flush_subbuf_file(tip, ts);
+ }
+
+ return 0;
+}
+
+/*
+ * scans the tips we know and writes out the subbuffers we accumulate
+ */
+static void get_and_write_events(void)
+{
+ struct device_information *dip;
+ struct thread_information *tip;
+ int i, j, events, ret, tips_running;
+
+ while (!is_done()) {
+ events = 0;
+
+ for_each_dip(dip, i) {
+ for_each_tip(dip, tip, j) {
+ ret = write_tip_events(tip);
+ if (ret > 0)
+ events += ret;
+ }
+ }
+
+ if (!events)
+ usleep(10);
+ }
+
+ /*
+ * reap stored events
+ */
+ do {
+ events = 0;
+ tips_running = 0;
+ for_each_dip(dip, i) {
+ for_each_tip(dip, tip, j) {
+ ret = write_tip_events(tip);
+ if (ret > 0)
+ events += ret;
+ tips_running += !tip->exited;
+ }
+ }
+ usleep(10);
+ } while (events || tips_running);
+}
+
+static void wait_for_threads(void)
+{
+ /*
+ * for piped or network output, poll and fetch data for writeout.
+ * for files, we just wait around for trace threads to exit
+ */
+ if ((output_name && !strcmp(output_name, "-")) ||
+ net_mode == Net_client)
+ get_and_write_events();
+ else {
+ struct device_information *dip;
+ struct thread_information *tip;
+ int i, j, tips_running;
+
+ do {
+ tips_running = 0;
+ usleep(1000);
+
+ for_each_dip(dip, i)
+ for_each_tip(dip, tip, j)
+ tips_running += !tip->exited;
+ } while (tips_running);
+ }
+}
+
+static void fill_ofname(char *dst, char *buts_name, int cpu)
+{
+ int len = 0;
+
+ if (output_dir)
+ len = sprintf(dst, "%s/", output_dir);
+
+ if (output_name)
+ sprintf(dst + len, "%s.blktrace.%d", output_name, cpu);
+ else
+ sprintf(dst + len, "%s.blktrace.%d", buts_name, cpu);
}
static int start_threads(struct device_information *dip)
{
struct thread_information *tip;
- char op[64];
int j, pipeline = output_name && !strcmp(output_name, "-");
+ int mode, vbuf_size;
+ char op[64];
- for (tip = dip->threads, j = 0; j < ncpus; j++, tip++) {
+ for_each_tip(dip, tip, j) {
tip->cpu = j;
tip->device = dip;
- tip->fd_lock = NULL;
tip->events_processed = 0;
+ memset(&tip->fifo, 0, sizeof(tip->fifo));
+ tip->leftover_ts = NULL;
if (pipeline) {
- tip->ofd = dup(STDOUT_FILENO);
- tip->fd_lock = &stdout_mutex;
+ tip->ofile = fdopen(STDOUT_FILENO, "w");
+ tip->ofile_stdout = 1;
+ tip->ofile_mmap = 0;
+ mode = _IOLBF;
+ vbuf_size = 512;
} else {
- 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);
+ fill_ofname(op, dip->buts_name, tip->cpu);
+ tip->ofile = fopen(op, "w+");
+ tip->ofile_stdout = 0;
+ tip->ofile_mmap = 1;
+ mode = _IOFBF;
+ vbuf_size = OFILE_BUF;
}
- if (tip->ofd < 0) {
+ if (tip->ofile == NULL) {
perror(op);
return 1;
}
- if (pthread_create(&tip->thread, NULL, extract, tip)) {
+ tip->ofile_buffer = malloc(vbuf_size);
+ if (setvbuf(tip->ofile, tip->ofile_buffer, mode, vbuf_size)) {
+ perror("setvbuf");
+ close_thread(tip);
+ return 1;
+ }
+
+ if (pthread_create(&tip->thread, NULL, thread_main, tip)) {
perror("pthread_create");
- close(tip->ofd);
+ close_thread(tip);
return 1;
}
}
return 0;
}
-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)
- close(tip->ofd);
-
- tip->fd = tip->ofd = -1;
-}
-
static void stop_threads(struct device_information *dip)
{
struct thread_information *tip;
- long ret;
- int j;
+ unsigned long ret;
+ int i;
- for (tip = dip->threads, j = 0; j < ncpus; j++, tip++) {
- if (pthread_join(tip->thread, (void *) &ret))
- perror("thread_join");
+ for_each_tip(dip, tip, i) {
+ (void) pthread_join(tip->thread, (void *) &ret);
close_thread(tip);
}
}
struct device_information *dip;
int i;
- for (dip = device_information, i = 0; i < ndevs; i++, dip++)
+ for_each_dip(dip, i)
stop_threads(dip);
}
static void stop_all_tracing(void)
{
struct device_information *dip;
- struct thread_information *tip;
- int i, j;
+ int i;
- for (dip = device_information, i = 0; i < ndevs; i++, dip++) {
- for (tip = dip->threads, j = 0; j < ncpus; j++, tip++)
- close_thread(tip);
+ for_each_dip(dip, i)
stop_trace(dip);
- }
}
static void exit_trace(int status)
{
- stop_all_tracing();
+ if (!is_trace_stopped()) {
+ trace_stopped = 1;
+ stop_all_threads();
+ stop_all_tracing();
+ }
+
exit(status);
}
struct device_information *dip;
int i;
- for (dip = device_information, i = 0; i < ndevs; i++, dip++) {
- dip->fd = open(dip->path, O_RDONLY);
+ for_each_dip(dip, i) {
+ dip->fd = open(dip->path, O_RDONLY | O_NONBLOCK);
if (dip->fd < 0) {
perror(dip->path);
return 1;
}
}
+
return 0;
}
return 1;
}
- for (dip = device_information, i = 0; i < ndevs; i++, dip++) {
+ for_each_dip(dip, i) {
if (start_trace(dip)) {
close(dip->fd);
fprintf(stderr, "Failed to start trace on %s\n",
break;
}
}
+
if (i != ndevs) {
- for (dip = device_information, j = 0; j < i; j++, dip++)
+ __for_each_dip(dip, j, i)
stop_trace(dip);
+
return 1;
}
- for (dip = device_information, i = 0; i < ndevs; i++, dip++) {
+ for_each_dip(dip, i) {
dip->threads = thread_information + (i * ncpus);
if (start_threads(dip)) {
fprintf(stderr, "Failed to start worker threads\n");
break;
}
}
+
if (i != ndevs) {
- for (dip = device_information, j = 0; j < i; j++, dip++)
+ __for_each_dip(dip, j, i)
stop_threads(dip);
- for (dip = device_information, i = 0; i < ndevs; i++, dip++)
+ for_each_dip(dip, i)
stop_trace(dip);
+
return 1;
}
static void show_stats(void)
{
- int i, j;
struct device_information *dip;
struct thread_information *tip;
- unsigned long long events_processed;
+ unsigned long long events_processed, data_read;
+ unsigned long total_drops;
+ int i, j, no_stdout = 0;
- if (output_name && !strcmp(output_name, "-"))
+ if (is_stat_shown())
return;
- for (dip = device_information, i = 0; i < ndevs; i++, dip++) {
- printf("Device: %s\n", dip->path);
+ if (output_name && !strcmp(output_name, "-"))
+ no_stdout = 1;
+
+ stat_shown = 1;
+
+ total_drops = 0;
+ for_each_dip(dip, i) {
+ if (!no_stdout)
+ printf("Device: %s\n", dip->path);
events_processed = 0;
- for (tip = dip->threads, j = 0; j < ncpus; j++, tip++) {
- printf(" CPU%3d: %20ld events\n",
- tip->cpu, tip->events_processed);
+ data_read = 0;
+ for_each_tip(dip, tip, j) {
+ if (!no_stdout)
+ printf(" CPU%3d: %20lu events, %8llu KiB data\n",
+ tip->cpu, tip->events_processed,
+ tip->data_read >> 10);
events_processed += tip->events_processed;
+ data_read += tip->data_read;
+ }
+ total_drops += dip->drop_count;
+ if (!no_stdout)
+ printf(" Total: %20llu events (dropped %lu), %8llu KiB data\n",
+ events_processed, dip->drop_count,
+ data_read >> 10);
+ }
+
+ if (total_drops)
+ fprintf(stderr, "You have dropped events, consider using a larger buffer size (-b)\n");
+}
+
+static struct device_information *net_get_dip(char *buts_name)
+{
+ struct device_information *dip;
+ int i;
+
+ for (i = 0; i < ndevs; i++) {
+ dip = &device_information[i];
+
+ if (!strcmp(dip->buts_name, buts_name))
+ return dip;
+ }
+
+ device_information = realloc(device_information, (ndevs + 1) * sizeof(*dip));
+ dip = &device_information[ndevs];
+ strcpy(dip->buts_name, buts_name);
+ ndevs++;
+ dip->threads = malloc(ncpus * sizeof(struct thread_information));
+ memset(dip->threads, 0, ncpus * sizeof(struct thread_information));
+
+ /*
+ * open all files
+ */
+ for (i = 0; i < ncpus; i++) {
+ struct thread_information *tip = &dip->threads[i];
+ char op[64];
+
+ tip->cpu = i;
+ tip->ofile_stdout = 0;
+ tip->ofile_mmap = 1;
+ tip->device = dip;
+
+ fill_ofname(op, dip->buts_name, tip->cpu);
+
+ tip->ofile = fopen(op, "w+");
+ if (!tip->ofile) {
+ perror("fopen");
+ return NULL;
+ }
+ }
+
+ return dip;
+}
+
+static struct thread_information *net_get_tip(struct blktrace_net_hdr *bnh)
+{
+ struct device_information *dip;
+
+ ncpus = bnh->max_cpus;
+ dip = net_get_dip(bnh->buts_name);
+ return &dip->threads[bnh->cpu];
+}
+
+static int net_get_header(struct blktrace_net_hdr *bnh)
+{
+ int fl = fcntl(net_in_fd, F_GETFL);
+ int bytes_left, ret;
+ void *p = bnh;
+
+ fcntl(net_in_fd, F_SETFL, fl | O_NONBLOCK);
+ bytes_left = sizeof(*bnh);
+ while (bytes_left && !is_done()) {
+ ret = recv(net_in_fd, p, bytes_left, MSG_WAITALL);
+ if (ret < 0) {
+ if (errno != EAGAIN) {
+ perror("recv header");
+ return 1;
+ }
+ usleep(100);
+ continue;
+ } else if (!ret) {
+ usleep(100);
+ continue;
+ } else {
+ p += ret;
+ bytes_left -= ret;
+ }
+ }
+ fcntl(net_in_fd, F_SETFL, fl & ~O_NONBLOCK);
+ return 0;
+}
+
+static int net_server_loop(void)
+{
+ struct thread_information *tip;
+ struct blktrace_net_hdr bnh;
+
+ if (net_get_header(&bnh))
+ return 1;
+
+ if (data_is_native == -1 && check_data_endianness(bnh.magic)) {
+ fprintf(stderr, "server: received data is bad\n");
+ return 1;
+ }
+
+ if (!data_is_native) {
+ bnh.cpu = be32_to_cpu(bnh.cpu);
+ bnh.len = be32_to_cpu(bnh.len);
+ }
+
+ tip = net_get_tip(&bnh);
+ if (!tip)
+ return 1;
+
+ if (mmap_subbuf(tip, bnh.len))
+ return 1;
+
+ return 0;
+}
+
+/*
+ * Start here when we are in server mode - just fetch data from the network
+ * and dump to files
+ */
+static int net_server(void)
+{
+ struct sockaddr_in addr;
+ socklen_t socklen;
+ int fd, opt, i, j;
+
+ fd = socket(AF_INET, SOCK_STREAM, 0);
+ if (fd < 0) {
+ perror("server: socket");
+ return 1;
+ }
+
+ opt = 1;
+ if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof(opt)) < 0) {
+ perror("setsockopt");
+ return 1;
+ }
+
+ memset(&addr, 0, sizeof(addr));
+ addr.sin_family = AF_INET;
+ addr.sin_addr.s_addr = htonl(INADDR_ANY);
+ addr.sin_port = htons(net_port);
+
+ if (bind(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
+ perror("bind");
+ return 1;
+ }
+
+ if (listen(fd, 1) < 0) {
+ perror("listen");
+ return 1;
+ }
+
+ printf("blktrace: waiting for incoming connection...\n");
+
+ socklen = sizeof(addr);
+ net_in_fd = accept(fd, (struct sockaddr *) &addr, &socklen);
+ if (net_in_fd < 0) {
+ perror("accept");
+ return 1;
+ }
+
+ signal(SIGINT, handle_sigint);
+ signal(SIGHUP, handle_sigint);
+ signal(SIGTERM, handle_sigint);
+ signal(SIGALRM, handle_sigint);
+
+ printf("blktrace: connected!\n");
+
+ while (!is_done()) {
+ if (net_server_loop())
+ break;
+ }
+
+ for (i = 0; i < ndevs; i++) {
+ struct device_information *dip = &device_information[i];
+
+ for (j = 0; j < ncpus; j++)
+ tip_ftrunc_final(&dip->threads[j]);
+ }
+
+ return 0;
+}
+
+/*
+ * Setup outgoing network connection where we will transmit data
+ */
+static int net_setup_client(void)
+{
+ struct sockaddr_in addr;
+ int fd;
+
+ fd = socket(AF_INET, SOCK_STREAM, 0);
+ if (fd < 0) {
+ perror("client: socket");
+ return 1;
+ }
+
+ memset(&addr, 0, sizeof(addr));
+ addr.sin_family = AF_INET;
+ addr.sin_port = htons(net_port);
+
+ if (inet_aton(hostname, &addr.sin_addr) != 1) {
+ struct hostent *hent = gethostbyname(hostname);
+ if (!hent) {
+ perror("gethostbyname");
+ return 1;
}
- printf(" Total: %20lld events\n", events_processed);
+
+ memcpy(&addr.sin_addr, hent->h_addr, 4);
+ strcpy(hostname, hent->h_name);
}
+
+ printf("blktrace: connecting to %s\n", hostname);
+
+ if (connect(fd, (struct sockaddr *) &addr, sizeof(addr)) < 0) {
+ perror("client: connect");
+ return 1;
+ }
+
+ printf("blktrace: connected!\n");
+ net_out_fd = fd;
+ return 0;
}
static char usage_str[] = \
"\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-D Directory to prepend to output file names\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 %s %s",program, blktrace_version, usage_str);
}
-static void handle_sigint(int sig)
-{
- done = 1;
-}
-
int main(int argc, char *argv[])
{
static char default_relay_path[] = "/relay";
break;
case 'A':
- if ((sscanf(optarg, "%x", &i) != 1) || !VALID_SET(i)) {
+ if ((sscanf(optarg, "%x", &i) != 1) ||
+ !valid_act_opt(i)) {
fprintf(stderr,
"Invalid set action mask %s/0x%x\n",
optarg, i);
return 1;
}
break;
- case 'v':
+ case 'V':
printf("%s version %s\n", argv[0], blktrace_version);
return 0;
+ case 'b':
+ buf_size = strtoul(optarg, NULL, 10);
+ if (buf_size <= 0 || buf_size > 16*1024) {
+ fprintf(stderr,
+ "Invalid buffer size (%lu)\n",buf_size);
+ return 1;
+ }
+ buf_size <<= 10;
+ break;
+ case 'n':
+ buf_nr = strtoul(optarg, NULL, 10);
+ if (buf_nr <= 0) {
+ fprintf(stderr,
+ "Invalid buffer nr (%lu)\n", buf_nr);
+ return 1;
+ }
+ break;
+ case 'D':
+ output_dir = optarg;
+ break;
+ case 'h':
+ net_mode = Net_client;
+ strcpy(hostname, optarg);
+ break;
+ case 'l':
+ net_mode = Net_server;
+ break;
+ case 'p':
+ net_port = atoi(optarg);
+ break;
default:
show_usage(argv[0]);
return 1;
}
}
+ setlocale(LC_NUMERIC, "en_US");
+
+ page_size = getpagesize();
+
+ if (net_mode == Net_server)
+ return net_server();
+
while (optind < argc) {
if (resize_devices(argv[optind++]) != 0)
return 1;
fprintf(stderr,"%s does not appear to be a valid path\n",
relay_path);
return 1;
- } else if (st.f_type != RELAYFS_TYPE) {
+ } else if (st.f_type != (long) RELAYFS_TYPE) {
fprintf(stderr,"%s does not appear to be a relay filesystem\n",
relay_path);
return 1;
return 0;
}
- setlocale(LC_NUMERIC, "en_US");
-
ncpus = sysconf(_SC_NPROCESSORS_ONLN);
if (ncpus < 0) {
fprintf(stderr, "sysconf(_SC_NPROCESSORS_ONLN) failed\n");
return 1;
}
- if (start_devices() != 0)
- return 1;
-
signal(SIGINT, handle_sigint);
signal(SIGHUP, handle_sigint);
signal(SIGTERM, handle_sigint);
signal(SIGALRM, handle_sigint);
+ if (net_mode == Net_client && net_setup_client())
+ return 1;
+
+ if (start_devices() != 0)
+ return 1;
+
atexit(stop_all_tracing);
if (stop_watch)
alarm(stop_watch);
- while (!is_done())
- sleep(1);
+ wait_for_threads();
+
+ if (!is_trace_stopped()) {
+ trace_stopped = 1;
+ stop_all_threads();
+ stop_all_traces();
+ }
- stop_all_threads();
- stop_all_traces();
show_stats();
return 0;
projects / blktrace.git / blobdiff