This changes the SHM log format, it doesn’t use 0-bytes to separate
entries anymore. Instead of using lots of printf() calls in i3-dump-log,
we now do precisely one big write().
So, to be clear: i3-dump-log and i3 both need to be upgraded.
Mismatching versions will lead to garbage output (no crashes of i3, just
garbage output).
The -f flag uses an inter-process pthread_cond_t in the shared memory
header to broadcast the arrival of new messages to all i3-dump-log
processes. This internally uses futexes and thus doesn’t even mean a
kernel call in most cases. inter-process pthread_cond_ts require NPTL
(the Native Posix Thread Library, introduce in Linux 2.6).
* vim:ts=4:sw=4:expandtab
*
* i3 - an improved dynamic tiling window manager
* vim:ts=4:sw=4:expandtab
*
* i3 - an improved dynamic tiling window manager
- * © 2009-2011 Michael Stapelberg and contributors (see also: LICENSE)
+ * © 2009-2012 Michael Stapelberg and contributors (see also: LICENSE)
*
* i3-dump-log/main.c: Dumps the i3 SHM log to stdout.
*
*
* i3-dump-log/main.c: Dumps the i3 SHM log to stdout.
*
#include "shmlog.h"
#include <i3/ipc.h>
#include "shmlog.h"
#include <i3/ipc.h>
+static uint32_t offset_next_write,
+ wrap_count;
+
+static i3_shmlog_header *header;
+static char *logbuffer,
+ *walk;
+
+static int check_for_wrap(void) {
+ if (wrap_count == header->wrap_count)
+ return 0;
+
+ /* The log wrapped. Print the remaining content and reset walk to the top
+ * of the log. */
+ wrap_count = header->wrap_count;
+ write(STDOUT_FILENO, walk, ((logbuffer + header->offset_last_wrap) - walk));
+ walk = logbuffer + sizeof(i3_shmlog_header);
+ return 1;
+}
+
+static void print_till_end(void) {
+ check_for_wrap();
+ int n = write(STDOUT_FILENO, walk, ((logbuffer + header->offset_next_write) - walk));
+ if (n > 0) {
+ walk += n;
+ }
+}
+
int main(int argc, char *argv[]) {
int o, option_index = 0;
int main(int argc, char *argv[]) {
int o, option_index = 0;
+ bool verbose = false,
+ follow = false;
static struct option long_options[] = {
{"version", no_argument, 0, 'v'},
{"verbose", no_argument, 0, 'V'},
static struct option long_options[] = {
{"version", no_argument, 0, 'v'},
{"verbose", no_argument, 0, 'V'},
+ {"follow", no_argument, 0, 'f'},
{"help", no_argument, 0, 'h'},
{0, 0, 0, 0}
};
{"help", no_argument, 0, 'h'},
{0, 0, 0, 0}
};
- char *options_string = "s:vVh";
+ char *options_string = "s:vfVh";
while ((o = getopt_long(argc, argv, options_string, long_options, &option_index)) != -1) {
if (o == 'v') {
while ((o = getopt_long(argc, argv, options_string, long_options, &option_index)) != -1) {
if (o == 'v') {
return 0;
} else if (o == 'V') {
verbose = true;
return 0;
} else if (o == 'V') {
verbose = true;
+ } else if (o == 'f') {
+ follow = true;
} else if (o == 'h') {
printf("i3-dump-log " I3_VERSION "\n");
} else if (o == 'h') {
printf("i3-dump-log " I3_VERSION "\n");
- printf("i3-dump-log [-s <socket>]\n");
+ printf("i3-dump-log [-f] [-s <socket>]\n");
- int logbuffer_shm = shm_open(shmname, O_RDONLY, 0);
+ /* NB: While we must never read, we need O_RDWR for the pthread condvar. */
+ int logbuffer_shm = shm_open(shmname, O_RDWR, 0);
if (logbuffer_shm == -1)
err(EXIT_FAILURE, "Could not shm_open SHM segment for the i3 log (%s)", shmname);
if (fstat(logbuffer_shm, &statbuf) != 0)
err(EXIT_FAILURE, "stat(%s)", shmname);
if (logbuffer_shm == -1)
err(EXIT_FAILURE, "Could not shm_open SHM segment for the i3 log (%s)", shmname);
if (fstat(logbuffer_shm, &statbuf) != 0)
err(EXIT_FAILURE, "stat(%s)", shmname);
- char *logbuffer = mmap(NULL, statbuf.st_size, PROT_READ, MAP_SHARED, logbuffer_shm, 0);
+ /* NB: While we must never read, we need O_RDWR for the pthread condvar. */
+ logbuffer = mmap(NULL, statbuf.st_size, PROT_READ | PROT_WRITE, MAP_SHARED, logbuffer_shm, 0);
if (logbuffer == MAP_FAILED)
err(EXIT_FAILURE, "Could not mmap SHM segment for the i3 log");
if (logbuffer == MAP_FAILED)
err(EXIT_FAILURE, "Could not mmap SHM segment for the i3 log");
- i3_shmlog_header *header = (i3_shmlog_header*)logbuffer;
+ header = (i3_shmlog_header*)logbuffer;
if (verbose)
printf("next_write = %d, last_wrap = %d, logbuffer_size = %d, shmname = %s\n",
header->offset_next_write, header->offset_last_wrap, header->size, shmname);
if (verbose)
printf("next_write = %d, last_wrap = %d, logbuffer_size = %d, shmname = %s\n",
header->offset_next_write, header->offset_last_wrap, header->size, shmname);
- int chars;
- char *walk = logbuffer + header->offset_next_write;
- /* Skip the first line, it very likely is mangled. Not a problem, though,
- * the log is chatty enough to have plenty lines left. */
- while (*walk != '\0')
- walk++;
+ walk = logbuffer + header->offset_next_write;
- /* Print the oldest log lines. We use printf("%s") to stop on \0. */
- while (walk < (logbuffer + header->offset_last_wrap)) {
- chars = printf("%s", walk);
- /* Shortcut: If there are two consecutive \0 bytes, this part of the
- * buffer was never touched. To not call printf() for every byte of the
- * buffer, we directly exit the loop. */
- if (*walk == '\0' && *(walk+1) == '\0')
- break;
- walk += (chars > 0 ? chars : 1);
+ /* We first need to print old content in case there was at least one
+ * wrapping already. */
+
+ if (*walk != '\0') {
+ /* In case there was a write to the buffer already, skip the first
+ * old line, it very likely is mangled. Not a problem, though, the log
+ * is chatty enough to have plenty lines left. */
+ while (*walk != '\n')
+ walk++;
+ walk++;
+ /* In case there was no wrapping, this is a no-op, otherwise it prints the
+ * old lines. */
+ wrap_count = 0;
+ check_for_wrap();
+
/* Then start from the beginning and print the newer lines */
walk = logbuffer + sizeof(i3_shmlog_header);
/* Then start from the beginning and print the newer lines */
walk = logbuffer + sizeof(i3_shmlog_header);
- while (walk < (logbuffer + header->offset_next_write)) {
- chars = printf("%s", walk);
- walk += (chars > 0 ? chars : 1);
+ print_till_end();
+
+ if (follow) {
+ /* Since pthread_cond_wait() expects a mutex, we need to provide one.
+ * To not lock i3 (that’s bad, mhkay?) we just define one outside of
+ * the shared memory. */
+ pthread_mutex_t dummy_mutex = PTHREAD_MUTEX_INITIALIZER;
+ pthread_mutex_lock(&dummy_mutex);
+ while (1) {
+ pthread_cond_wait(&(header->condvar), &dummy_mutex);
+ /* If this was not a spurious wakeup, print the new lines. */
+ if (header->offset_next_write != offset_next_write) {
+ offset_next_write = header->offset_next_write;
+ print_till_end();
+ }
+ }
#define _I3_SHMLOG_H
#include <stdint.h>
#define _I3_SHMLOG_H
#include <stdint.h>
+/*
+ * Header of the shmlog file. Used by i3/src/log.c and i3/i3-dump-log/main.c.
+ *
+ */
typedef struct i3_shmlog_header {
typedef struct i3_shmlog_header {
+ /* Byte offset where the next line will be written to. */
uint32_t offset_next_write;
uint32_t offset_next_write;
+
+ /* Byte offset where the last wrap occured. */
uint32_t offset_last_wrap;
uint32_t offset_last_wrap;
+
+ /* The size of the logfile in bytes. Since the size is limited to 25 MiB
+ * an uint32_t is sufficient. */
+
+ /* wrap counter. We need it to reliably signal to clients that we just
+ * wrapped (clients cannot use offset_last_wrap because that might
+ * coincidentally be exactly the same as previously). Overflows can happen
+ * and don’t matter — clients use an equality check (==). */
+ uint32_t wrap_count;
+
+ /* pthread condvar which will be broadcasted whenever there is a new
+ * message in the log. i3-dump-log uses this to implement -f (follow, like
+ * tail -f) in an efficient way. */
+ pthread_cond_t condvar;
} i3_shmlog_header;
#endif
} i3_shmlog_header;
#endif
i3_HEADERS_CMDPARSER := $(wildcard include/GENERATED_*.h)
i3_HEADERS := $(filter-out $(i3_HEADERS_CMDPARSER),$(wildcard include/*.h))
i3_CFLAGS = $(XCB_CFLAGS) $(XCB_KBD_CFLAGS) $(XCB_WM_CFLAGS) $(X11_CFLAGS) $(XCURSOR_CFLAGS) $(YAJL_CFLAGS) $(LIBEV_CFLAGS) $(PCRE_CFLAGS) $(LIBSN_CFLAGS)
i3_HEADERS_CMDPARSER := $(wildcard include/GENERATED_*.h)
i3_HEADERS := $(filter-out $(i3_HEADERS_CMDPARSER),$(wildcard include/*.h))
i3_CFLAGS = $(XCB_CFLAGS) $(XCB_KBD_CFLAGS) $(XCB_WM_CFLAGS) $(X11_CFLAGS) $(XCURSOR_CFLAGS) $(YAJL_CFLAGS) $(LIBEV_CFLAGS) $(PCRE_CFLAGS) $(LIBSN_CFLAGS)
-i3_LIBS = $(XCB_LIBS) $(XCB_KBD_LIBS) $(XCB_WM_LIBS) $(X11_LIBS) $(XCURSOR_LIBS) $(YAJL_LIBS) $(LIBEV_LIBS) $(PCRE_LIBS) $(LIBSN_LIBS) -lm
+i3_LIBS = $(XCB_LIBS) $(XCB_KBD_LIBS) $(XCB_WM_LIBS) $(X11_LIBS) $(XCURSOR_LIBS) $(YAJL_LIBS) $(LIBEV_LIBS) $(PCRE_LIBS) $(LIBSN_LIBS) -lm -lpthread
# When using clang, we use pre-compiled headers to speed up the build. With
# gcc, this actually makes the build slower.
# When using clang, we use pre-compiled headers to speed up the build. With
# gcc, this actually makes the build slower.
#include <sys/mman.h>
#include <sys/stat.h>
#include <errno.h>
#include <sys/mman.h>
#include <sys/stat.h>
#include <errno.h>
#if defined(__APPLE__)
#include <sys/types.h>
#include <sys/sysctl.h>
#if defined(__APPLE__)
#include <sys/types.h>
#include <sys/sysctl.h>
static char *logbuffer;
/* A pointer (within logbuffer) where data will be written to next. */
static char *logwalk;
static char *logbuffer;
/* A pointer (within logbuffer) where data will be written to next. */
static char *logwalk;
+/* A pointer to the shmlog header */
+static i3_shmlog_header *header;
/* A pointer to the byte where we last wrapped. Necessary to not print the
* left-overs at the end of the ringbuffer. */
static char *loglastwrap;
/* A pointer to the byte where we last wrapped. Necessary to not print the
* left-overs at the end of the ringbuffer. */
static char *loglastwrap;
*
*/
static void store_log_markers(void) {
*
*/
static void store_log_markers(void) {
- i3_shmlog_header *header = (i3_shmlog_header*)logbuffer;
-
header->offset_next_write = (logwalk - logbuffer);
header->offset_last_wrap = (loglastwrap - logbuffer);
header->size = logbuffer_size;
header->offset_next_write = (logwalk - logbuffer);
header->offset_last_wrap = (loglastwrap - logbuffer);
header->size = logbuffer_size;
logbuffer = NULL;
return;
}
logbuffer = NULL;
return;
}
+
+ /* Initialize with 0-bytes, just to be sure… */
+ memset(logbuffer, '\0', logbuffer_size);
+
+ header = (i3_shmlog_header*)logbuffer;
+
+ pthread_condattr_t cond_attr;
+ pthread_condattr_init(&cond_attr);
+ if (pthread_condattr_setpshared(&cond_attr, PTHREAD_PROCESS_SHARED) != 0)
+ ELOG("pthread_condattr_setpshared() failed, i3-dump-log -f will not work!\n");
+ pthread_cond_init(&(header->condvar), &cond_attr);
+
logwalk = logbuffer + sizeof(i3_shmlog_header);
loglastwrap = logbuffer + logbuffer_size;
store_log_markers();
logwalk = logbuffer + sizeof(i3_shmlog_header);
loglastwrap = logbuffer + logbuffer_size;
store_log_markers();
fprintf(stderr, "BUG: single log message > 4k\n");
}
fprintf(stderr, "BUG: single log message > 4k\n");
}
- /* If there is no space for the current message (plus trailing
- * nullbyte) in the ringbuffer, we need to wrap and write to the
- * beginning again. */
- if ((len+1) >= (logbuffer_size - (logwalk - logbuffer))) {
+ /* If there is no space for the current message in the ringbuffer, we
+ * need to wrap and write to the beginning again. */
+ if (len >= (logbuffer_size - (logwalk - logbuffer))) {
loglastwrap = logwalk;
logwalk = logbuffer + sizeof(i3_shmlog_header);
loglastwrap = logwalk;
logwalk = logbuffer + sizeof(i3_shmlog_header);
+ store_log_markers();
+ header->wrap_count++;
- /* Copy the buffer, terminate it, move the write pointer to the byte after
- * our current message. */
+ /* Copy the buffer, move the write pointer to the byte after our
+ * current message. */
strncpy(logwalk, message, len);
strncpy(logwalk, message, len);
- logwalk[len] = '\0';
- logwalk += len + 1;
+ /* Wake up all (i3-dump-log) processes waiting for condvar. */
+ pthread_cond_broadcast(&(header->condvar));
+
if (print)
fwrite(message, len, 1, stdout);
}
if (print)
fwrite(message, len, 1, stdout);
}
projects / i3 / i3 / commitdiff