Merge branch 'pattern-match'

Unify pattern match behavior so it users can use a single pattern
matching type consistently.  Now all of function filter (both user and
kernel) as well as events specification use REGEX match.  It can be
changed using --match option.

Fixes #328.

Signed-off-by: Namhyung Kim <namhyung@gmail.com>

cmd-info.c

@@ -788,6 +788,37 @@ static int read_record_date(void *arg)
 	return 0;
 }
 
+static int fill_pattern_type(void *arg)
+{
+	struct fill_handler_arg *fha = arg;
+
+	dprintf(fha->fd, "pattern_type:%s\n",
+		get_filter_pattern(fha->opts->patt_type));
+
+	return 0;
+}
+
+static int read_pattern_type(void *arg)
+{
+	struct ftrace_file_handle *handle = arg;
+	struct uftrace_info *info = &handle->info;
+	char buf[4096];
+	size_t len;
+
+	if (fgets(buf, sizeof(buf), handle->fp) == NULL)
+		return -1;
+
+	if (strncmp(buf, "pattern_type:", 13))
+		return -1;
+
+	len = strlen(&buf[13]);
+	if (buf[13 + len - 1] == '\n')
+		buf[13 + len - 1] = '\0';
+
+	info->patt_type = parse_filter_pattern(&buf[13]);
+	return 0;
+}
+
 struct uftrace_info_handler {
 	enum uftrace_info_bits bit;
 	int (*handler)(void *arg);
@@ -818,6 +849,7 @@ void fill_uftrace_info(uint64_t *info_mask, int fd, struct opts *opts, int statu
 		{ LOADINFO,	fill_loadinfo },
 		{ ARG_SPEC,	fill_arg_spec },
 		{ RECORD_DATE,	fill_record_date },
+		{ PATTERN_TYPE, fill_pattern_type },
 	};
 
 	for (i = 0; i < ARRAY_SIZE(fill_handlers); i++) {
@@ -856,6 +888,7 @@ int read_uftrace_info(uint64_t info_mask, struct ftrace_file_handle *handle)
 		{ LOADINFO,	read_loadinfo },
 		{ ARG_SPEC,	read_arg_spec },
 		{ RECORD_DATE,	read_record_date },
+		{ PATTERN_TYPE, read_pattern_type },
 	};
 
 	memset(&handle->info, 0, sizeof(handle->info));
@@ -1003,6 +1036,9 @@ int command_info(int argc, char *argv[], struct opts *opts)
 			pr_out(fmt, "auto-args", "true");
 	}
 
+	if (handle.hdr.info_mask & (1UL << PATTERN_TYPE))
+		pr_out(fmt, "pattern", get_filter_pattern(handle.info.patt_type));
+
 	if (handle.hdr.info_mask & (1UL << EXIT_STATUS)) {
 		int status = handle.info.exit_status;
 

cmd-record.c

@@ -17,7 +17,6 @@
 #include <sys/eventfd.h>
 #include <sys/resource.h>
 #include <sys/epoll.h>
-#include <fnmatch.h>
 
 #include "uftrace.h"
 #include "libmcount/mcount.h"
@@ -243,6 +242,9 @@ static void setup_child_environ(struct opts *opts, int pfd)
 	if (opts->script_file)
 		setenv("UFTRACE_SCRIPT", opts->script_file, 1);
 
+	if (opts->patt_type != PATT_REGEX)
+		setenv("UFTRACE_PATTERN", get_filter_pattern(opts->patt_type), 1);
+
 	if (opts->lib_path)
 		snprintf(buf, sizeof(buf), "%s/libmcount/", opts->lib_path);
 	else
@@ -1470,7 +1472,8 @@ static void check_binary(struct opts *opts)
 	close(fd);
 }
 
-static bool check_linux_perf_event(char *events)
+static bool check_linux_schedule_event(char *events,
+				       enum uftrace_pattern_type ptype)
 {
 	struct strv strv = STRV_INIT;
 	char *evt;
@@ -1483,10 +1486,17 @@ static bool check_linux_perf_event(char *events)
 	strv_split(&strv, events, ";");
 
 	strv_for_each(&strv, evt, i) {
-		if (fnmatch(evt, "linux:schedule", 0) == 0) {
+		struct uftrace_pattern patt;
+
+		init_filter_pattern(ptype, &patt, evt);
+
+		if (match_filter_pattern(&patt, "linux:schedule"))
 			found = true;
+
+		free_filter_pattern(&patt);
+
+		if (found)
 			break;
-		}
 	}
 
 	strv_free(&strv);
@@ -1856,7 +1866,8 @@ int command_record(int argc, char *argv[], struct opts *opts)
 
 	check_binary(opts);
 
-	has_perf_event = check_linux_perf_event(opts->event);
+	has_perf_event = check_linux_schedule_event(opts->event,
+						    opts->patt_type);
 
 	fflush(stdout);
 

cmd-replay.c

@@ -939,8 +939,12 @@ static bool skip_sys_exit(struct opts *opts, struct ftrace_task_handle *task)
 
 	ip = task->func_stack[0].addr;
 	sym = find_symtabs(&task->h->sessions.first->symtabs, ip);
+	if (sym == NULL)
+		return false;
 
-	if (sym && !strncmp(sym->name, "sys_exit", 8))
+	if (!strncmp(sym->name, "sys_exit", 8))
+		return true;
+	if (!strcmp(sym->name, "do_syscall_64"))
 		return true;
 
 	return false;

cmd-script.c

@@ -169,7 +169,7 @@ int command_script(int argc, char *argv[], struct opts *opts)
 	fstack_setup_filters(opts, &handle);
 
 	/* initialize script */
-	if (script_init(opts->script_file) < 0)
+	if (script_init(opts->script_file, opts->patt_type) < 0)
 		return -1;
 
 	while (read_rstack(&handle, &task) == 0 && !uftrace_done) {

doc/uftrace-dump.md

@@ -65,6 +65,9 @@ OPTIONS
 \--demangle=*TYPE*
 :   Use demangled C++ symbol names for filters, triggers, arguments and/or return values.  Possible values are "full", "simple" and "no".  Default is "simple" which ignores function arguments and template parameters.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 EXAMPLE
 =======

doc/uftrace-graph.md

@@ -64,6 +64,9 @@ OPTIONS
 \--demangle=*TYPE*
 :   Use demangled C++ symbol names for filters, triggers, arguments and/or return values.  Possible values are "full", "simple" and "no".  Default is "simple" which ignores function arguments and template parameters.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 EXAMPLES
 ========

doc/uftrace-live.md

@@ -139,6 +139,9 @@ OPTIONS
 \--libname
 :   Show library name along with function name.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 FILTERS
 =======

doc/uftrace-record.md

@@ -108,6 +108,9 @@ OPTIONS
 -S *SCRIPT_PATH*, \--script=*SCRIPT_PATH*
 :   Add a script to do addtional work at the entry and exit of function.  The type of script is detected by the postfix such as '.py' for python.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 FILTERS
 =======
@@ -203,7 +206,7 @@ You can also set triggers on filtered functions.  See *TRIGGERS* section below f
 
 When kernel function tracing is enabled, you can also set the filters on kernel functions by marking the symbol with the `@kernel` modifier.  The following example will show all user functions and the (kernel) page fault handler.
 
-    $ sudo uftrace -k -F '*page_fault@kernel' ./abc
+    $ sudo uftrace -k -F '.*page_fault@kernel' ./abc
     # DURATION    TID     FUNCTION
                [14721] | main() {
       7.713 us [14721] |   __do_page_fault();

doc/uftrace-replay.md

@@ -65,7 +65,7 @@ OPTIONS
 :   Do not show comments of returned functions.
 
 -k, \--kernel
-:   Trace kernel functions (and events) as well as user functions (and events).  This options has no meaning and so is deprecated now.  It will always show kernel functions and events if exists.  If you want to hide kernel functions, please use `-N .@kernel` to filter out all kernel functions.
+:   Trace kernel functions (and events) as well as user functions (and events).  This options has no meaning and so is deprecated now.  It will always show kernel functions and events if exists.  If you want to hide kernel functions, please use `-N .@kernel` to filter out all kernel functions (for the regex match).
 
 \--kernel-full
 :   Show all kernel functions and events occurred outside of user functions.  This option is the inverse of `--kernel-skip-out`.
@@ -82,6 +82,9 @@ OPTIONS
 \--libname
 :   Show libname name along with function name.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 FILTERS
 =======

doc/uftrace-report.md

@@ -76,6 +76,9 @@ OPTIONS
 \--demangle=*TYPE*
 :   Use demangled C++ symbol names for filters, triggers, arguments and/or return values.  Possible values are "full", "simple" and "no".  Default is "simple" which ignores function arguments and template parameters.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 EXAMPLE
 =======

doc/uftrace-script.md

@@ -46,6 +46,9 @@ OPTIONS
 \--record COMMAND [*command-options*]
 :   Record a new trace before running a given script.
 
+--match=*TYPE*
+:   Use pattern match using TYPE.  Possible types are `regex` and `glob`.  Default is `regex`.
+
 
 EXAMPLES
 ========
@@ -117,7 +120,7 @@ The below is another example that shows the different output compared to previou
 
 The python script above can be modified to do more output customization.
 
-The python script can have an optional "UFTRACE_FUNCS" list which can have name (or regex pattern) of functions to run the script.  If it exists, only matched functions will run the script.  For example, if you add following lines to the script, it will run only for functions with a single letter name.
+The python script can have an optional "UFTRACE_FUNCS" list which can have name (or pattern depending on the --match option) of functions to run the script.  If it exists, only matched functions will run the script.  For example, if you add following lines to the script, it will run only for functions with a single letter name.
 
     $ echo 'UFTRACE_FUNCS = [ "^.$" ]' >> replay.py
     $ uftrace script -S replay.py

libmcount/dynamic.c

@@ -1,6 +1,5 @@
 #include <string.h>
 #include <link.h>
-#include <regex.h>
 
 /* This should be defined before #include "utils.h" */
 #define PR_FMT     "dynamic"
@@ -92,7 +91,8 @@ static int prepare_dynamic_update(void)
 	return ret;
 }
 
-static int do_dynamic_update(struct symtabs *symtabs, char *patch_funcs)
+static int do_dynamic_update(struct symtabs *symtabs, char *patch_funcs,
+			     enum uftrace_pattern_type ptype)
 {
 	char *name, *nopatched_name = NULL;
 	struct symtab *symtab = &symtabs->symtab;
@@ -105,26 +105,17 @@ static int do_dynamic_update(struct symtabs *symtabs, char *patch_funcs)
 	strv_split(&funcs, patch_funcs, ";");
 
 	strv_for_each(&funcs, name, j) {
-		bool is_regex;
 		bool found = false;
-		regex_t re;
 		unsigned i;
 		struct sym *sym;
+		struct uftrace_pattern patt;
 
-		is_regex = strpbrk(name, REGEX_CHARS);
-		if (is_regex) {
-			if (regcomp(&re, name, REG_NOSUB | REG_EXTENDED)) {
-				pr_dbg("regex pattern failed: %s\n", name);
-				strv_free(&funcs);
-				return -1;
-			}
-		}
+		init_filter_pattern(ptype, &patt, name);
 
 		for (i = 0; i < symtab->nr_sym; i++) {
 			sym = &symtab->sym[i];
 
-			if ((is_regex && regexec(&re, sym->name, 0, NULL, 0)) ||
-			    (!is_regex && strcmp(name, sym->name)))
+			if (!match_filter_pattern(&patt, sym->name))
 				continue;
 
 			found = true;
@@ -147,14 +138,14 @@ static int do_dynamic_update(struct symtabs *symtabs, char *patch_funcs)
 		if (!found)
 			stats.nomatch++;
 
-		if (is_regex)
-			regfree(&re);
+		free_filter_pattern(&patt);
 	}
 
-	if (stats.failed || stats.skipped || stats.nomatch)
+	if (stats.failed || stats.skipped || stats.nomatch) {
 		pr_out("%s cannot be patched dynamically\n",
 		       (stats.failed + stats.skipped + stats.nomatch) > 1 ?
 		       "some functions" : nopatched_name);
+	}
 
 	strv_free(&funcs);
 	return 0;
@@ -184,7 +175,8 @@ static float calc_percent(int n, int total)
 	return 100.0 * n / total;
 }
 
-int mcount_dynamic_update(struct symtabs *symtabs, char *patch_funcs)
+int mcount_dynamic_update(struct symtabs *symtabs, char *patch_funcs,
+			  enum uftrace_pattern_type ptype)
 {
 	int ret = 0;
 	int success;
@@ -194,7 +186,7 @@ int mcount_dynamic_update(struct symtabs *symtabs, char *patch_funcs)
 		return -1;
 	}
 
-	ret = do_dynamic_update(symtabs, patch_funcs);
+	ret = do_dynamic_update(symtabs, patch_funcs, ptype);
 
 	success = stats.total - stats.failed - stats.skipped;
 	pr_dbg("dynamic update stats:\n");

libmcount/event.c

@@ -5,7 +5,6 @@
 #include <link.h>
 #include <libelf.h>
 #include <gelf.h>
-#include <fnmatch.h>
 
 /* This should be defined before #include "utils.h" */
 #define PR_FMT     "event"
@@ -15,15 +14,16 @@
 #include "libmcount/internal.h"
 #include "utils/utils.h"
 #include "utils/list.h"
+#include "utils/filter.h"
 
 #define SDT_SECT  ".note.stapsdt"
 #define SDT_NAME  "stapsdt"
 #define SDT_TYPE  3
 
 struct event_spec {
 	struct list_head list;
-	char *provider;
-	char *event;
+	struct uftrace_pattern provider;
+	struct uftrace_pattern event;
 };
 
 struct stapsdt {
@@ -126,9 +126,9 @@ static int search_sdt_event(struct dl_phdr_info *info, size_t sz, void *data)
 		}
 
 		list_for_each_entry(spec, spec_list, list) {
-			if (fnmatch(spec->provider, vendor, 0) != 0)
+			if (!match_filter_pattern(&spec->provider, vendor))
 				continue;
-			if (fnmatch(spec->event, event, 0) != 0)
+			if (!match_filter_pattern(&spec->event, event))
 				continue;
 			break;
 		}
@@ -161,7 +161,8 @@ static int search_sdt_event(struct dl_phdr_info *info, size_t sz, void *data)
 	goto out;
 }
 
-int mcount_setup_events(char *dirname, char *event_str)
+int mcount_setup_events(char *dirname, char *event_str,
+			enum uftrace_pattern_type ptype)
 {
 	int ret = 0;
 	FILE *fp;
@@ -187,8 +188,9 @@ int mcount_setup_events(char *dirname, char *event_str)
 				continue;
 
 			es = xmalloc(sizeof(*es));
-			es->provider = spec;
-			es->event = sep;
+
+			init_filter_pattern(ptype, &es->provider, spec);
+			init_filter_pattern(ptype, &es->event, sep);
 			list_add_tail(&es->list, &specs);
 		}
 		else {
@@ -200,6 +202,9 @@ int mcount_setup_events(char *dirname, char *event_str)
 
 	list_for_each_entry_safe(es, tmp, &specs, list) {
 		list_del(&es->list);
+
+		free_filter_pattern(&es->provider);
+		free_filter_pattern(&es->event);
 		free(es);
 	}
 	strv_free(&strv);