[PATCH v2] trace-cmd: Remove all die()s from trace-cmd library

[PATCH v2] trace-cmd: Remove all die()s from trace-cmd library

[Tzvetomir Stoyanov (VMware)]

The die() call is used for a fatal error. It terminates the current
program with exit(). The program should not be terminated from a library
routine, the die() call should not be used in the trace-cmd library
context.

But as all these logs indicate a fatal error in the library, it is
useful to have them. The die() call is replaced with _lib_fatal()
library internal function, which prints the error message without
terminating the application. If the library is compiled with
  LIB_DIE
defined, the _lib_fatal() call behaves like die() - calls exit() to
terminate the application.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
---
v2 changes:
 - Replaced die() with _lib_fatal().

 lib/trace-cmd/include/trace-cmd-local.h |  2 ++
 lib/trace-cmd/trace-input.c             | 29 +++++++++++++-------
 lib/trace-cmd/trace-util.c              | 36 ++++++++++++-------------
 3 files changed, 38 insertions(+), 29 deletions(-)

diff --git a/lib/trace-cmd/include/trace-cmd-local.h b/lib/trace-cmd/include/trace-cmd-local.h
index d265e4c8..da9d8d58 100644
--- a/lib/trace-cmd/include/trace-cmd-local.h
+++ b/lib/trace-cmd/include/trace-cmd-local.h
@@ -12,6 +12,8 @@
 /* Can be overridden */
 void warning(const char *fmt, ...);
 
+void _lib_fatal(const char *fmt, ...);
+
 #define __packed __attribute__((packed))
 
 /* trace.dat file format version */
diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c
index 2093a3dc..3687de09 100644
--- a/lib/trace-cmd/trace-input.c
+++ b/lib/trace-cmd/trace-input.c
@@ -16,6 +16,7 @@
 #include <linux/time64.h>
 
 #include "trace-write-local.h"
+#include "trace-cmd-local.h"
 #include "trace-local.h"
 #include "kbuffer.h"
 #include "list.h"
@@ -1087,8 +1088,10 @@ static void __free_page(struct tracecmd_input *handle, struct page *page)
 	struct page **pages;
 	int index;
 
-	if (!page->ref_count)
-		die("Page ref count is zero!\n");
+	if (!page->ref_count) {
+		_lib_fatal("Page ref count is zero!\n");
+		return;
+	}
 
 	page->ref_count--;
 	if (page->ref_count)
@@ -1146,16 +1149,20 @@ void tracecmd_free_record(struct tep_record *record)
 	if (!record)
 		return;
 
-	if (!record->ref_count)
-		die("record ref count is zero!");
+	if (!record->ref_count) {
+		_lib_fatal("record ref count is zero!");
+		return;
+	}
 
 	record->ref_count--;
 
 	if (record->ref_count)
 		return;
 
-	if (record->locked)
-		die("freeing record when it is locked!");
+	if (record->locked) {
+		_lib_fatal("freeing record when it is locked!");
+		return;
+	}
 
 	record->data = NULL;
 
@@ -1318,7 +1325,7 @@ static int get_page(struct tracecmd_input *handle, int cpu,
 
 	if (offset & (handle->page_size - 1)) {
 		errno = -EINVAL;
-		die("bad page offset %llx", offset);
+		_lib_fatal("bad page offset %llx", offset);
 		return -1;
 	}
 
@@ -1326,7 +1333,7 @@ static int get_page(struct tracecmd_input *handle, int cpu,
 	    offset > handle->cpu_data[cpu].file_offset +
 	    handle->cpu_data[cpu].file_size) {
 		errno = -EINVAL;
-		die("bad page offset %llx", offset);
+		_lib_fatal("bad page offset %llx", offset);
 		return -1;
 	}
 
@@ -1891,8 +1898,10 @@ tracecmd_peek_data(struct tracecmd_input *handle, int cpu)
 	if (handle->cpu_data[cpu].next) {
 
 		record = handle->cpu_data[cpu].next;
-		if (!record->data)
-			die("Something freed the record");
+		if (!record->data) {
+			_lib_fatal("Something freed the record");
+			return NULL;
+		}
 
 		if (handle->cpu_data[cpu].timestamp == record->ts)
 			return record;
diff --git a/lib/trace-cmd/trace-util.c b/lib/trace-cmd/trace-util.c
index 538adbc2..e8ed22fa 100644
--- a/lib/trace-cmd/trace-util.c
+++ b/lib/trace-cmd/trace-util.c
@@ -353,12 +353,12 @@ trace_load_plugins(struct tep_handle *tep, int flags)
 	return list;
 }
 
-void __noreturn __vdie(const char *fmt, va_list ap)
+int __vlib_fatal(const char *fmt, va_list ap)
 {
 	int ret = errno;
 
 	if (errno)
-		perror("trace-cmd");
+		perror("libtracecmd");
 	else
 		ret = -1;
 
@@ -366,36 +366,32 @@ void __noreturn __vdie(const char *fmt, va_list ap)
 	vfprintf(stderr, fmt, ap);
 
 	fprintf(stderr, "\n");
-	exit(ret);
+	return ret;
 }
 
-void __noreturn __die(const char *fmt, ...)
+#ifdef LIB_DIE
+
+void __noreturn _lib_fatal(const char *fmt, ...)
 {
 	va_list ap;
+	int ret;
 
 	va_start(ap, fmt);
-	__vdie(fmt, ap);
+	ret = __vlib_fatal(fmt, ap);
 	va_end(ap);
+	exit(ret);
 }
 
-void __weak __noreturn die(const char *fmt, ...)
+#else
+void _lib_fatal(const char *fmt, ...)
 {
 	va_list ap;
 
 	va_start(ap, fmt);
-	__vdie(fmt, ap);
+	__vlib_fatal(fmt, ap);
 	va_end(ap);
 }
-
-void __weak *malloc_or_die(unsigned int size)
-{
-	void *data;
-
-	data = malloc(size);
-	if (!data)
-		die("malloc");
-	return data;
-}
+#endif
 
 #define LOG_BUF_SIZE 1024
 static void __plog(const char *prefix, const char *fmt, va_list ap, FILE *fp)
@@ -546,8 +542,10 @@ int tracecmd_count_cpus(void)
 	pbuf = buf;
 
 	fp = fopen("/proc/cpuinfo", "r");
-	if (!fp)
-		die("Can not read cpuinfo");
+	if (!fp) {
+		_lib_fatal("Can not read cpuinfo");
+		return 0;
+	}
 
 	while ((r = getline(&pbuf, pn, fp)) >= 0) {
 		char *p;
-- 
2.30.2

Re: [PATCH v2] trace-cmd: Remove all die()s from trace-cmd library

[Steven Rostedt]

On Wed, 24 Mar 2021 15:51:43 +0200
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:

> @@ -366,36 +366,32 @@ void __noreturn __vdie(const char *fmt, va_list ap)
>  	vfprintf(stderr, fmt, ap);
>  
>  	fprintf(stderr, "\n");
> -	exit(ret);
> +	return ret;
>  }
>  
> -void __noreturn __die(const char *fmt, ...)
> +#ifdef LIB_DIE

I wonder if it would be more useful not to have it as a compile time
define, but instead a global variable:

static bool lib_die;

void tracecmd_die_on_error(void)
{
	lib_die = true;
}

Make the "die()" functions into static variables here.

void __hidden _lib_fatal(const char *fmt, ...)
{
	va_list ap;

	va_start(ap, fmt);
	if (lib_die)
		__vdie(fmt, ap);
	else
		__vwarning(fmt, ap);
	va_end(ap);
}

That way if we run trace-cmd with the --debug option, it could then call
the tracecmd_die_on_error() function, and it will crash on any "fatal"
places.

-- Steve


> +
> +void __noreturn _lib_fatal(const char *fmt, ...)
>  {
>  	va_list ap;
> +	int ret;
>  
>  	va_start(ap, fmt);
> -	__vdie(fmt, ap);
> +	ret = __vlib_fatal(fmt, ap);
>  	va_end(ap);
> +	exit(ret);
>  }
>  
> -void __weak __noreturn die(const char *fmt, ...)
> +#else
> +void _lib_fatal(const char *fmt, ...)
>  {
>  	va_list ap;
>  
>  	va_start(ap, fmt);
> -	__vdie(fmt, ap);
> +	__vlib_fatal(fmt, ap);
>  	va_end(ap);
>  }
> -
> -void __weak *malloc_or_die(unsigned int size)
> -{
> -	void *data;
> -
> -	data = malloc(size);
> -	if (!data)
> -		die("malloc");
> -	return data;
> -}
> +#endif
>  

Re: [PATCH v2] trace-cmd: Remove all die()s from trace-cmd library

[Tzvetomir Stoyanov]

On Wed, Mar 24, 2021 at 11:53 PM Steven Rostedt <rostedt@goodmis.org> wrote:
>
> On Wed, 24 Mar 2021 15:51:43 +0200
> "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:
>
> > @@ -366,36 +366,32 @@ void __noreturn __vdie(const char *fmt, va_list ap)
> >       vfprintf(stderr, fmt, ap);
> >
> >       fprintf(stderr, "\n");
> > -     exit(ret);
> > +     return ret;
> >  }
> >
> > -void __noreturn __die(const char *fmt, ...)
> > +#ifdef LIB_DIE
>
> I wonder if it would be more useful not to have it as a compile time
> define, but instead a global variable:
>
> static bool lib_die;
>
> void tracecmd_die_on_error(void)
> {
>         lib_die = true;
> }
>

There is tracecmd_set_debug() API already, It can be used to trigger
that die() logic. It makes sense for the library to do a real die()
when running in debug mode, if a fatal error is encountered.

> Make the "die()" functions into static variables here.
>
> void __hidden _lib_fatal(const char *fmt, ...)
> {
>         va_list ap;
>
>         va_start(ap, fmt);
>         if (lib_die)
>                 __vdie(fmt, ap);
>         else
>                 __vwarning(fmt, ap);
>         va_end(ap);
> }
>
> That way if we run trace-cmd with the --debug option, it could then call
> the tracecmd_die_on_error() function, and it will crash on any "fatal"
> places.
>
> -- Steve
>
>
> > +
> > +void __noreturn _lib_fatal(const char *fmt, ...)
> >  {
> >       va_list ap;
> > +     int ret;
> >
> >       va_start(ap, fmt);
> > -     __vdie(fmt, ap);
> > +     ret = __vlib_fatal(fmt, ap);
> >       va_end(ap);
> > +     exit(ret);
> >  }
> >
> > -void __weak __noreturn die(const char *fmt, ...)
> > +#else
> > +void _lib_fatal(const char *fmt, ...)
> >  {
> >       va_list ap;
> >
> >       va_start(ap, fmt);
> > -     __vdie(fmt, ap);
> > +     __vlib_fatal(fmt, ap);
> >       va_end(ap);
> >  }
> > -
> > -void __weak *malloc_or_die(unsigned int size)
> > -{
> > -     void *data;
> > -
> > -     data = malloc(size);
> > -     if (!data)
> > -             die("malloc");
> > -     return data;
> > -}
> > +#endif
> >



-- 
Tzvetomir (Ceco) Stoyanov
VMware Open Source Technology Center