When blk_cgroup option is enabled in trace_options file, the trace
got by blktrace will include the io cgroup info for trace action or
trace message. The io cgroup info is a file handle exported by kernfs
(union kernfs_node_id), so we use open_by_handle_at() and procfs to
get the path of the corresponding io cgroup and output the cgroup path
before the pid. Maybe we also need to add a cache for the translation
from kernfs (ino, gen) tuple to cgroup path.
The following lines are snippet from output with cgroup info:
253,0 3 664 0.528123310 /t1 0 C R 790832 + 8 [0]
253,0 3 665 0.528208698 1521 A R 452720 + 8 <- (253,1) 450672
253,0 3 666 0.528211593 /t1 1521 Q R 452720 + 8 [fio]
253,0 3 667 0.528215253 /t1 1521 G R 452720 + 8 [fio]
253,0 3 668 0.528219125 1521 P N [fio]
253,0 3 669 0.528221590 1521 UT N [fio] 1
253,0 3 670 0.528223067 /t1 1521 I R 452720 + 8 [fio]
253,0 3 671 0.528226553 /t1 1521 D R 452720 + 8 [fio]
253,0 5 608 0.533095375 / 0 C R 3868224 + 8 [0]
253,0 5 609 0.533164175 1523 A R 4641296 + 8 <- (253,1) 4639248
253,0 5 610 0.533166445 / 1523 Q R 4641296 + 8 [fio]
253,0 5 611 0.533169825 / 1523 G R 4641296 + 8 [fio]
Signed-off-by: Hou Tao <houtao1@xxxxxxxxxx>
---
blkparse.c | 103 +++++++++++++++++++++++++++++++++++++++++++++++++++++++--
blkparse_fmt.c | 10 ++++--
blktrace.h | 2 ++
doc/blkparse.1 | 9 ++++-
4 files changed, 118 insertions(+), 6 deletions(-)
diff --git a/blkparse.c b/blkparse.c
index fa2f520..35958d2 100644
--- a/blkparse.c
+++ b/blkparse.c
@@ -148,6 +148,12 @@ static struct option l_opts[] = {
.val = 'F'
},
{
+ .name = "cgroup",
+ .has_arg = no_argument,
+ .flag = NULL,
+ .val = 'g',
+ },
+ {
.name = "hash-by-name",
.has_arg = no_argument,
.flag = NULL,
@@ -292,6 +298,15 @@ static char *pipename;
static int text_output = 1;
+int show_cgroup = 0;
+static const char *unknown_cg = "<...>";
+static const char *root_cg = "/";
+static int cg_mount_fd;
+static struct file_handle *cg_handle;
+static int cg_path_skip;
+static char cg_path[NAME_MAX + 1];
+
+
#define is_done() (*(volatile int *)(&done))
static volatile int done;
@@ -572,6 +587,45 @@ static struct process_pid_map *add_ppm_hash(pid_t pid, const char *name)
return ppm;
}
+void get_cg_path(struct blk_io_trace *bit, const char **path)
+{
+#define FILEID_INO32_GEN 1
+ int fd = -1;
+ union kernfs_node_id *id;
+ char fd_path[NAME_MAX + 1];
+ ssize_t cnt;
+
+ cg_handle->handle_bytes = sizeof(*id);
+ cg_handle->handle_type = FILEID_INO32_GEN;
+ id = (void *)&cg_handle[1];
+
+ memcpy(id, (void *)&bit[1], sizeof(*id));
+
+ fd = open_by_handle_at(cg_mount_fd, cg_handle, 0);
+ if (fd < 0)
+ goto err_out;
+
+ snprintf(fd_path, sizeof(fd_path), "/proc/%d/fd/%d", getpid(), fd);
+ cnt = readlink(fd_path, cg_path, sizeof(cg_path) - 1);
+ if (cnt < 0)
+ goto err_out;
+
+ close(fd);
+
+ cg_path[cnt] = '\0';
+ if (cg_path_skip < cnt)
+ *path = cg_path + cg_path_skip;
+ else
+ *path = root_cg;
+
+ return;
+
+err_out:
+ if (fd >= 0)
+ close(fd);
+ *path = unknown_cg;
+}
+
static void handle_notify(struct blk_io_trace *bit)
{
void *payload;
@@ -607,16 +661,22 @@ static void handle_notify(struct blk_io_trace *bit)
case BLK_TN_MESSAGE:
if (pdu_len > 0) {
char msg[pdu_len+1];
+ const char *cg_path;
memcpy(msg, (char *)payload, pdu_len);
msg[pdu_len] = '\0';
+ if (show_cgroup && (bit->action & __BLK_TA_CGROUP))
+ get_cg_path(bit, &cg_path);
+ else
+ cg_path = "";
+
fprintf(ofp,
- "%3d,%-3d %2d %8s %5d.%09lu %5u %2s %3s %s\n",
+ "%3d,%-3d %2d %8s %5d.%09lu %s%5u %2s %3s %s\n",
MAJOR(bit->device), MINOR(bit->device),
bit->cpu, "0", (int) SECONDS(bit->time),
(unsigned long) NANO_SECONDS(bit->time),
- 0, "m", "N", msg);
+ cg_path, 0, "m", "N", msg);
}
break;
@@ -2743,7 +2803,36 @@ static int is_pipe(const char *str)
return 0;
}
-#define S_OPTS "a:A:b:D:d:f:F:hi:o:Oqstw:vVM"
+static int init_cg_res(void)
+{
+ const char *path;
+
+ cg_handle = malloc(sizeof(*cg_handle) + sizeof(union kernfs_node_id));
+ if (!cg_handle) {
+ perror("malloc");
+ return -1;
+ }
+
+ path = "/sys/fs/cgroup/blkio";
+ cg_mount_fd = open(path, O_RDONLY);
+ if (cg_mount_fd < 0) {
+ if (errno == ENOENT) {
+ path = "/sys/fs/cgroup/unified";
+ cg_mount_fd = open(path, O_RDONLY);
+ }
+
+ if (cg_mount_fd < 0) {
+ perror("open io cgroup");
+ return -1;
+ }
+ }
+
+ cg_path_skip = strlen(path);
+
+ return 0;
+}
+
+#define S_OPTS "a:A:b:D:d:f:F:ghi:o:Oqstw:vVM"
static char usage_str[] = "\n\n" \
"-i <file> | --input=<file>\n" \
"[ -a <action field> | --act-mask=<action field> ]\n" \
@@ -2753,6 +2842,7 @@ static char usage_str[] = "\n\n" \
"[ -D <dir> | --input-directory=<dir> ]\n" \
"[ -f <format> | --format=<format> ]\n" \
"[ -F <spec> | --format-spec=<spec> ]\n" \
+ "[ -g | --cgroup\n" \
"[ -h | --hash-by-name ]\n" \
"[ -o <file> | --output=<file> ]\n" \
"[ -O | --no-text-output ]\n" \
@@ -2771,6 +2861,7 @@ static char usage_str[] = "\n\n" \
"\t-f Output format. Customize the output format. The format field\n" \
"\t identifies can be found in the documentation\n" \
"\t-F Format specification. Can be found in the documentation\n" \
+ "\t-g Show the io cgroup of trace action or trace message\n" \
"\t-h Hash processes by name, not pid\n" \
"\t-i Input file containing trace data, or '-' for stdin\n" \
"\t-o Output file. If not given, output is stdout\n" \
@@ -2840,6 +2931,9 @@ int main(int argc, char *argv[])
if (rb_batch <= 0)
rb_batch = RB_BATCH_DEFAULT;
break;
+ case 'g':
+ show_cgroup = 1;
+ break;
case 's':
per_process_stats = 1;
break;
@@ -2898,6 +2992,9 @@ int main(int argc, char *argv[])
if (act_mask_tmp != 0)
act_mask = act_mask_tmp;
+ if (show_cgroup && init_cg_res())
+ return 1;
+
memset(&rb_sort_root, 0, sizeof(rb_sort_root));
signal(SIGINT, handle_sigint);
diff --git a/blkparse_fmt.c b/blkparse_fmt.c
index 8dc20ca..58c2721 100644
--- a/blkparse_fmt.c
+++ b/blkparse_fmt.c
@@ -312,6 +312,7 @@ static void process_default(char *act, struct per_cpu_info *pci,
struct blk_io_trace_remap r = { .device_from = 0, };
char rwbs[8];
char *name;
+ const char *cg_path;
fill_rwbs(rwbs, t);
@@ -324,13 +325,18 @@ static void process_default(char *act, struct per_cpu_info *pci,
t->device = r.device_to;
}
+ if (show_cgroup && (t->action & __BLK_TA_CGROUP))
+ get_cg_path(t, &cg_path);
+ else
+ cg_path = "";
+
/*
* The header is always the same
*/
- fprintf(ofp, "%3d,%-3d %2d %8d %5d.%09lu %5u %2s %3s ",
+ fprintf(ofp, "%3d,%-3d %2d %8d %5d.%09lu %s%5u %2s %3s ",
MAJOR(t->device), MINOR(t->device), pci->cpu, t->sequence,
(int) SECONDS(t->time), (unsigned long) NANO_SECONDS(t->time),
- t->pid, act, rwbs);
+ cg_path, t->pid, act, rwbs);
name = find_process_name(t->pid);
diff --git a/blktrace.h b/blktrace.h
index c5ed618..6874edb 100644
--- a/blktrace.h
+++ b/blktrace.h
@@ -68,6 +68,7 @@ struct per_cpu_info {
extern FILE *ofp;
extern int data_is_native;
extern struct timespec abs_start_time;
+extern int show_cgroup;
#define CHECK_MAGIC(t) (((t)->magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
#define SUPPORTED_VERSION (0x07)
@@ -152,5 +153,6 @@ extern char *find_process_name(pid_t);
extern void pdu_start_len(struct blk_io_trace *bit,
void **pdu, __u16 *pdu_len);
+extern void get_cg_path(struct blk_io_trace *bit, const char **path);
#endif
diff --git a/doc/blkparse.1 b/doc/blkparse.1
index be9b34b..de2f9cd 100644
--- a/doc/blkparse.1
+++ b/doc/blkparse.1
@@ -114,11 +114,18 @@ event type. The single\-character \fItyp\fR field is one of the
action specifiers described in ACTION IDENTIFIERS.
.RE
+\-g
+.br
+\-\-cgroup
+.RS
+Show the io cgroup of trace action or trace message
+.RE
+
\-M
.br
\-\-no-msgs
.RS
-When \-d is specified, this will stop messages from being output to the
+When \-M is specified, this will stop messages from being output to the
file. (Can seriously reduce the size of the resultant file when using
the CFQ I/O scheduler.)
.RE
--
2.9.5
--
To unsubscribe from this list: send the line "unsubscribe linux-btrace" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
