diff mbox series

[v3] trace-cmd: Save the tracee memory map into the trace.dat file.

Message ID 20190619083915.2049-1-tz.stoyanov@gmail.com (mailing list archive)
State Superseded
Headers show
Series [v3] trace-cmd: Save the tracee memory map into the trace.dat file. | expand

Commit Message

Tzvetomir Stoyanov (VMware) June 19, 2019, 8:39 a.m. UTC
From: "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com>

A new trace-cmd record option is added: "--mmap". When it is set with
combination of -F or -P options, the memory map of the traced applications
is stored in the trace.dat file. A new API tracecmd_search_task_mmap()
can be used to look up into stored memory maps. The map is retrieved from
/proc/<pid>/maps file.

Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
---
[
 v3 changes:
   - Changed tracecmd_search_task_mmap() API to return not only the library
     name, but also the start and end memory addresses.
   - Renamed *tracee* to *task*
   - Improved resources cleanup, in case of an error.
   - Removed (this) changelog from the commit message.

 v2 changes:
   - Replaced usage of tracecmd_add_option_v() with tracecmd_add_option() API.
   - Added checks to prevent buffer overflow when sscanf (... "%s", buf) is used.
   - Return error in case memory allocation fails.
   - Return error if option string is not in the expected format.
   - Sort memory maps and use binary search to find matching library in the map.
]

 include/trace-cmd/trace-cmd.h  |   9 ++
 lib/trace-cmd/trace-input.c    | 149 ++++++++++++++++++++++++++++++-
 tracecmd/include/trace-local.h |  10 +++
 tracecmd/trace-record.c        | 158 +++++++++++++++++++++++++++++++++
 4 files changed, 323 insertions(+), 3 deletions(-)

Comments

Yordan Karadzhov June 19, 2019, 2:19 p.m. UTC | #1
On 19.06.19 г. 11:39 ч., tz.stoyanov@gmail.com wrote:
> From: "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com>
> 
> A new trace-cmd record option is added: "--mmap". When it is set with
> combination of -F or -P options, the memory map of the traced applications
> is stored in the trace.dat file. A new API tracecmd_search_task_mmap()
> can be used to look up into stored memory maps. The map is retrieved from
> /proc/<pid>/maps file.
> 
> Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
> ---
> [
>   v3 changes:
>     - Changed tracecmd_search_task_mmap() API to return not only the library
>       name, but also the start and end memory addresses.
>     - Renamed *tracee* to *task*
>     - Improved resources cleanup, in case of an error.
>     - Removed (this) changelog from the commit message.
> 
>   v2 changes:
>     - Replaced usage of tracecmd_add_option_v() with tracecmd_add_option() API.
>     - Added checks to prevent buffer overflow when sscanf (... "%s", buf) is used.
>     - Return error in case memory allocation fails.
>     - Return error if option string is not in the expected format.
>     - Sort memory maps and use binary search to find matching library in the map.
> ]
> 
>   include/trace-cmd/trace-cmd.h  |   9 ++
>   lib/trace-cmd/trace-input.c    | 149 ++++++++++++++++++++++++++++++-
>   tracecmd/include/trace-local.h |  10 +++
>   tracecmd/trace-record.c        | 158 +++++++++++++++++++++++++++++++++
>   4 files changed, 323 insertions(+), 3 deletions(-)
> 
> diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
> index 6f62ab9..17edb9d 100644
> --- a/include/trace-cmd/trace-cmd.h
> +++ b/include/trace-cmd/trace-cmd.h
> @@ -82,6 +82,7 @@ enum {
>   	TRACECMD_OPTION_OFFSET,
>   	TRACECMD_OPTION_CPUCOUNT,
>   	TRACECMD_OPTION_VERSION,
> +	TRACECMD_OPTION_PIDMMAPS,
>   };

...

>   
> +static int get_pid_mmaps(int pid)
> +{
> +	struct buffer_instance *instance = &top_instance;
> +	struct pid_mem_maps *maps = instance->mem_maps;
> +	struct pid_mem_maps *m;
> +	unsigned long long begin, end, inode, tmp;
> +	struct lib_mem_map *map;
> +	char mapname[PATH_MAX+1];
> +	char fname[PATH_MAX+1];
> +	char buf[PATH_MAX+100];
> +	char perm[5];
> +	char dev[6];
> +	FILE *f;
> +	int ret;
> +	int res;
> +	int i;
> +
> +	sprintf(fname, "/proc/%d/exe", pid);
> +	ret = readlink(fname, mapname, PATH_MAX);
> +	if (ret >= PATH_MAX || ret < 0)
> +		return -ENOENT;
> +
> +	sprintf(fname, "/proc/%d/maps", pid);
> +	f = fopen(fname, "r");
> +	if (!f)
> +		return -ENOENT;
> +
> +	while (maps) {
> +		if (pid == maps->pid)
> +			break;
> +		maps = maps->next;
> +	}
> +
> +	ret = -ENOMEM;
> +	if (!maps) {
> +		maps = calloc(1, sizeof(*maps));
> +		if (!maps)
> +			goto out_fail;
> +		maps->pid = pid;
> +		maps->next = instance->mem_maps;
> +		instance->mem_maps = maps;
> +	} else {
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			free(maps->lib_maps[i].lib_name);
> +		free(maps->lib_maps);
> +		maps->lib_maps = NULL;
> +		maps->nr_lib_maps = 0;
> +		free(maps->proc_name);
> +	}
> +
> +	maps->proc_name = strdup(mapname);
> +	if (!maps->proc_name)
> +		goto out;
> +
> +	while (fgets(buf, sizeof(buf), f)) {
> +		mapname[0] = '\0';
> +		res = sscanf(buf, "%llx-%llx %4s %llx %5s %lld %s",

We tried testing the patch and we found that the memory map file is not 
properly parsed. The problem seems to get solved if we replace the line 
above with
		res = sscanf(buf, "%llx-%llx %4s %llx %s %lld %s",

Thanks!
Yordan

> +			     &begin, &end, perm, &tmp, dev, &inode, mapname);
> +		if (res == 7 && mapname[0] != '\0') {
> +			map = realloc(maps->lib_maps,
> +				      (maps->nr_lib_maps+1)*sizeof(*map));
> +			if (!map)
> +				goto out_fail;
> +			map[maps->nr_lib_maps].end = end;
> +			map[maps->nr_lib_maps].start = begin;
> +			map[maps->nr_lib_maps].lib_name = strdup(mapname);
> +			if (!map[maps->nr_lib_maps].lib_name)
> +				goto out_fail;
> +			maps->lib_maps = map;
> +			maps->nr_lib_maps++;
> +		}
> +	}
> +out:
> +	fclose(f);
> +	return 0;
> +
> +out_fail:
> +	fclose(f);
> +	if (maps) {
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			free(maps->lib_maps[i].lib_name);
> +		if (instance->mem_maps != maps) {
> +			m = instance->mem_maps;
> +			while (m) {
> +				if (m->next == maps) {
> +					m->next = maps->next;
> +					break;
> +				}
> +				m = m->next;
> +			}
> +		} else
> +			instance->mem_maps = maps->next;
> +		free(maps->lib_maps);
> +		maps->lib_maps = NULL;
> +		maps->nr_lib_maps = 0;
> +		free(maps->proc_name);
> +		maps->proc_name = NULL;
> +		free(maps);
> +	}
> +	return ret;
> +}
> +
> +static void get_filter_pid_mmaps(void)
> +{
> +	struct filter_pids *p;
> +
> +	for (p = filter_pids; p; p = p->next) {
> +		if (p->exclude)
> +			continue;
> +		get_pid_mmaps(p->pid);
> +	}
> +}
> +
>   static void update_task_filter(void)
>   {
>   	struct buffer_instance *instance;
> @@ -1070,6 +1184,9 @@ static void update_task_filter(void)
>   	if (no_filter)
>   		return;
>   
> +	if (get_mmap && filter_pids)
> +		get_filter_pid_mmaps();
> +
>   	if (filter_task)
>   		add_filter_pid(pid, 0);
>   
> @@ -1264,6 +1381,8 @@ static void ptrace_wait(enum trace_type type, int main_pid)
>   				break;
>   
>   			case PTRACE_EVENT_EXIT:
> +				if (get_mmap)
> +					get_pid_mmaps(main_pid);
>   				ptrace(PTRACE_GETEVENTMSG, pid, NULL, &cstatus);
>   				ptrace(PTRACE_DETACH, pid, NULL, NULL);
>   				break;
> @@ -3094,6 +3213,33 @@ static void append_buffer(struct tracecmd_output *handle,
>   	}
>   }
>   
> +
> +static void
> +add_pid_mem_maps(struct tracecmd_output *handle, struct buffer_instance *instance)
> +{
> +	struct pid_mem_maps *maps = instance->mem_maps;
> +	struct trace_seq s;
> +	int i;
> +
> +	trace_seq_init(&s);
> +	while (maps) {
> +		if (!maps->nr_lib_maps)
> +			continue;
> +		trace_seq_reset(&s);
> +		trace_seq_printf(&s, "%x %x %s\n",
> +				 maps->pid, maps->nr_lib_maps, maps->proc_name);
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			trace_seq_printf(&s, "%llx %llx %s\n",
> +					maps->lib_maps[i].start,
> +					maps->lib_maps[i].end,
> +					maps->lib_maps[i].lib_name);
> +		tracecmd_add_option(handle, TRACECMD_OPTION_PIDMMAPS,
> +				    s.len+1, s.buffer);
> +		maps = maps->next;
> +	}
> +	trace_seq_destroy(&s);
> +}
> +
>   static void
>   add_buffer_stat(struct tracecmd_output *handle, struct buffer_instance *instance)
>   {
> @@ -3287,6 +3433,10 @@ static void record_data(struct common_record_context *ctx)
>   		if (!no_top_instance() && !top_instance.msg_handle)
>   			print_stat(&top_instance);
>   
> +		for_all_instances(instance) {
> +			add_pid_mem_maps(handle, instance);
> +		}
> +
>   		tracecmd_append_cpu_data(handle, local_cpu_count, temp_files);
>   
>   		for (i = 0; i < max_cpu_count; i++)
> @@ -4397,6 +4547,7 @@ void update_first_instance(struct buffer_instance *instance, int topt)
>   }
>   
>   enum {
> +	OPT_mmap		= 244,
>   	OPT_quiet		= 245,
>   	OPT_debug		= 246,
>   	OPT_no_filter		= 247,
> @@ -4627,6 +4778,7 @@ static void parse_record_options(int argc,
>   			{"debug", no_argument, NULL, OPT_debug},
>   			{"quiet", no_argument, NULL, OPT_quiet},
>   			{"help", no_argument, NULL, '?'},
> +			{"mmap", no_argument, NULL, OPT_mmap},
>   			{"module", required_argument, NULL, OPT_module},
>   			{NULL, 0, NULL, 0}
>   		};
> @@ -4858,6 +5010,9 @@ static void parse_record_options(int argc,
>   		case 'i':
>   			ignore_event_not_found = 1;
>   			break;
> +		case OPT_mmap:
> +			get_mmap = 1;
> +			break;
>   		case OPT_date:
>   			ctx->date = 1;
>   			if (ctx->data_flags & DATA_FL_OFFSET)
> @@ -4924,6 +5079,9 @@ static void parse_record_options(int argc,
>   		add_func(&ctx->instance->filter_funcs,
>   			 ctx->instance->filter_mod, "*");
>   
> +	if (filter_task && get_mmap)
> +		do_ptrace = 1;
> +
>   	if (do_ptrace && !filter_task && (filter_pid < 0))
>   		die(" -c can only be used with -F (or -P with event-fork support)");
>   	if (ctx->do_child && !filter_task &&! filter_pid)
>
Yordan Karadzhov June 24, 2019, 10:14 a.m. UTC | #2
On 19.06.19 г. 11:39 ч., tz.stoyanov@gmail.com wrote:
> From: "Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com>
> 
> A new trace-cmd record option is added: "--mmap". When it is set with
> combination of -F or -P options, the memory map of the traced applications
> is stored in the trace.dat file. A new API tracecmd_search_task_mmap()
> can be used to look up into stored memory maps. The map is retrieved from
> /proc/<pid>/maps file.
> 
> Signed-off-by: Tzvetomir Stoyanov (VMware) <tz.stoyanov@gmail.com>
> ---
> [
>   v3 changes:
>     - Changed tracecmd_search_task_mmap() API to return not only the library
>       name, but also the start and end memory addresses.
>     - Renamed *tracee* to *task*
>     - Improved resources cleanup, in case of an error.
>     - Removed (this) changelog from the commit message.
> 
>   v2 changes:
>     - Replaced usage of tracecmd_add_option_v() with tracecmd_add_option() API.
>     - Added checks to prevent buffer overflow when sscanf (... "%s", buf) is used.
>     - Return error in case memory allocation fails.
>     - Return error if option string is not in the expected format.
>     - Sort memory maps and use binary search to find matching library in the map.
> ]
> 
>   include/trace-cmd/trace-cmd.h  |   9 ++
>   lib/trace-cmd/trace-input.c    | 149 ++++++++++++++++++++++++++++++-
>   tracecmd/include/trace-local.h |  10 +++
>   tracecmd/trace-record.c        | 158 +++++++++++++++++++++++++++++++++
>   4 files changed, 323 insertions(+), 3 deletions(-)
> 
> diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
> index 6f62ab9..17edb9d 100644
> --- a/include/trace-cmd/trace-cmd.h
> +++ b/include/trace-cmd/trace-cmd.h
> @@ -82,6 +82,7 @@ enum {
>   	TRACECMD_OPTION_OFFSET,
>   	TRACECMD_OPTION_CPUCOUNT,
>   	TRACECMD_OPTION_VERSION,
> +	TRACECMD_OPTION_PIDMMAPS,
>   };
>   
>   enum {
> @@ -97,6 +98,12 @@ struct tracecmd_ftrace {
>   	int long_size;
>   };
>   
> +struct lib_mem_map {
> +	unsigned long long	start;
> +	unsigned long long	end;
> +	char			*lib_name;
> +};
> +
>   typedef void (*tracecmd_show_data_func)(struct tracecmd_input *handle,
>   					struct tep_record *record);
>   typedef void (*tracecmd_handle_init_func)(struct tracecmd_input *handle,
> @@ -208,6 +215,8 @@ unsigned long long tracecmd_page_ts(struct tracecmd_input *handle,
>   unsigned int tracecmd_record_ts_delta(struct tracecmd_input *handle,
>   				      struct tep_record *record);
>   
> +struct lib_mem_map *tracecmd_search_task_mmap(struct tracecmd_input *handle,
> +					      int pid, unsigned long long addr);
>   #ifndef SWIG
>   /* hack for function graph work around */
>   extern __thread struct tracecmd_input *tracecmd_curr_thread_handle;
> diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c
> index 61566ba..7acecf3 100644
> --- a/lib/trace-cmd/trace-input.c
> +++ b/lib/trace-cmd/trace-input.c
> @@ -101,6 +101,7 @@ struct tracecmd_input {
>   	struct tracecmd_ftrace	finfo;
>   
>   	struct hook_list	*hooks;
> +	struct pid_mem_maps	*pid_mmaps;
>   	/* file information */
>   	size_t			header_files_start;
>   	size_t			ftrace_files_start;
> @@ -2134,6 +2135,147 @@ void tracecmd_set_ts2secs(struct tracecmd_input *handle,
>   	handle->use_trace_clock = false;
>   }
>   
> +static int trace_pid_mmap_cmp(const void *a, const void *b)
> +{
> +	struct lib_mem_map *map_a = (struct lib_mem_map *)a;
> +	struct lib_mem_map *map_b = (struct lib_mem_map *)b;
> +
> +	if (map_a->start > map_b->start)
> +		return 1;
> +	if (map_a->start < map_b->start)
> +		return -1;
> +	return 0;
> +}
> +
> +#define STR_MMAP_LINE_MAX	(PATH_MAX+34)
> +static int trace_pid_mmap_load(struct tracecmd_input *handle, char *buf)
> +{
> +	struct pid_mem_maps *maps = NULL;
> +	char mapname[STR_MMAP_LINE_MAX];
> +	char *line;
> +	int res;
> +	int ret;
> +	int i;
> +
> +	maps = calloc(1, sizeof(*maps));
> +	if (!maps)
> +		return -ENOMEM;
> +
> +	ret  = -EINVAL;
> +	line = strchr(buf, '\n');
> +	if (!line)
> +		goto out_fail;
> +
> +	*line = '\0';
> +	if (strlen(buf) > STR_MMAP_LINE_MAX)
> +		goto out_fail;
> +
> +	res = sscanf(buf, "%x %x %s", &maps->pid, &maps->nr_lib_maps, mapname);
> +	if (res != 3)
> +		goto out_fail;
> +
> +	ret  = -ENOMEM;
> +	maps->proc_name = strdup(mapname);
> +	if (!maps->proc_name)
> +		goto out_fail;
> +
> +	maps->lib_maps = calloc(maps->nr_lib_maps, sizeof(struct lib_mem_map));
> +	if (!maps->lib_maps)
> +		goto out_fail;
> +
> +	buf = line+1;

according to the kernel coding style this line must look lo

	buf = line + 1;
	

> +	line = strchr(buf, '\n');
> +	for (i = 0; i < maps->nr_lib_maps; i++) {
> +		if (!line)
> +			break;
> +		*line = '\0';
> +		if (strlen(buf) > STR_MMAP_LINE_MAX)
> +			break;
> +		res = sscanf(buf, "%llx %llx %s", &maps->lib_maps[i].start,
> +			     &maps->lib_maps[i].end, mapname);
> +		if (res != 3)
> +			break;
> +		maps->lib_maps[i].lib_name = strdup(mapname);
> +		if (!maps->lib_maps[i].lib_name)
> +			goto out_fail;
> +		buf = line+1;

Same here.

> +		line = strchr(buf, '\n');
> +	}
> +
> +	ret  = -EINVAL;
> +	if (i != maps->nr_lib_maps)
> +		goto out_fail;
> +
> +	qsort(maps->lib_maps, maps->nr_lib_maps,
> +	      sizeof(*maps->lib_maps), trace_pid_mmap_cmp);
> +
> +	maps->next = handle->pid_mmaps;
> +	handle->pid_mmaps = maps;

We need to have

void free_pid_mmaps(struct pid_mem_maps *maps)

and than have this function called in tracecmd_close(). Otherwise the 
allocated memory will leak.


> +
> +	return 0;
> +
> +out_fail:
> +	if (maps->lib_maps) {
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			free(maps->lib_maps[i].lib_name);
> +		free(maps->lib_maps);
> +	}
> +	free(maps->proc_name);
> +	free(maps);
> +
> +	return ret;
> +}
> +
> +static int trace_pid_mmap_search(const void *a, const void *b)
> +{
> +	struct lib_mem_map *key = (struct lib_mem_map *)a;
> +	struct lib_mem_map *map = (struct lib_mem_map *)b;
> +
> +	if (key->start >= map->end)
> +		return 1;
> +	if (key->start < map->start)
> +		return -1;
> +	return 0;
> +}
> +
> +/**
> + * tracecmd_search_task_mmap - Search task memory address map
> + * @handle: input handle to the trace.dat file
> + * @pid: pid of the task
> + * @addr: address from the task memory space.
> + *
> + * Map of the task memory can be saved in the trace.dat file, using the option
> + * "--mmap". If there is such information, this API can be used to look up into
> + * this memory map to find what library is loaded at the given @addr.
> + *
> + * A pointer to struct lib_mem_map is returned, containing the name of
> + * the library at given task @addr and the library start and end addresses.
> + */
> +struct lib_mem_map *tracecmd_search_task_mmap(struct tracecmd_input *handle,
> +					      int pid, unsigned long long addr)
> +{
> +	struct pid_mem_maps *maps;
> +	struct lib_mem_map *lib;
> +	struct lib_mem_map key;
> +
> +	if (!handle || !handle->pid_mmaps)
> +		return NULL;
> +
> +	maps = handle->pid_mmaps;
> +	while (maps) {
> +		if (maps->pid == pid)
> +			break;
> +		maps = maps->next;
> +	}
> +	if (!maps || !maps->nr_lib_maps || !maps->lib_maps)
> +		return NULL;
> +	key.start = addr;
> +	lib = bsearch(&key, maps->lib_maps, maps->nr_lib_maps,
> +		      sizeof(*maps->lib_maps), trace_pid_mmap_search);
> +
> +	return lib;
> +}
> +
>   static int handle_options(struct tracecmd_input *handle)
>   {
>   	unsigned long long offset;
> @@ -2221,9 +2363,6 @@ static int handle_options(struct tracecmd_input *handle)
>   		case TRACECMD_OPTION_UNAME:
>   			handle->uname = strdup(buf);
>   			break;
> -		case TRACECMD_OPTION_VERSION:
> -			handle->version = strdup(buf);
> -			break;
>   		case TRACECMD_OPTION_HOOK:
>   			hook = tracecmd_create_event_hook(buf);
>   			hook->next = handle->hooks;
> @@ -2233,6 +2372,10 @@ static int handle_options(struct tracecmd_input *handle)
>   			cpus = *(int *)buf;
>   			handle->cpus = tep_read_number(handle->pevent, &cpus, 4);
>   			break;
> +		case TRACECMD_OPTION_PIDMMAPS:
> +			if (buf[size] == '\0')
> +				trace_pid_mmap_load(handle, buf);
> +			break;
>   		default:
>   			warning("unknown option %d", option);
>   			break;
> diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h
> index 1cad3cc..ae1632c 100644
> --- a/tracecmd/include/trace-local.h
> +++ b/tracecmd/include/trace-local.h
> @@ -157,6 +157,14 @@ struct func_list {
>   	const char *mod;
>   };
>   
> +struct pid_mem_maps {
> +	struct pid_mem_maps	*next;
> +	struct lib_mem_map	*lib_maps;
> +	unsigned int		nr_lib_maps;
> +	char			*proc_name;
> +	int			pid;
> +};
> +
>   struct buffer_instance {
>   	struct buffer_instance	*next;
>   	const char		*name;
> @@ -183,6 +191,8 @@ struct buffer_instance {
>   	struct tracecmd_msg_handle *msg_handle;
>   	struct tracecmd_output *network_handle;
>   
> +	struct pid_mem_maps	*mem_maps;
> +
>   	char			*max_graph_depth;
>   
>   	int			flags;
> diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
> index 5dc6f17..edabf53 100644
> --- a/tracecmd/trace-record.c
> +++ b/tracecmd/trace-record.c
> @@ -84,6 +84,7 @@ static int max_kb;
>   static bool use_tcp;
>   
>   static int do_ptrace;
> +static int get_mmap;
>   
>   static int filter_task;
>   static int filter_pid = -1;
> @@ -1062,6 +1063,119 @@ static char *make_pid_filter(char *curr_filter, const char *field)
>   	return filter;
>   }
>   
> +static int get_pid_mmaps(int pid)
> +{
> +	struct buffer_instance *instance = &top_instance;
> +	struct pid_mem_maps *maps = instance->mem_maps;
> +	struct pid_mem_maps *m;
> +	unsigned long long begin, end, inode, tmp;
> +	struct lib_mem_map *map;
> +	char mapname[PATH_MAX+1];
> +	char fname[PATH_MAX+1];
> +	char buf[PATH_MAX+100];
> +	char perm[5];
> +	char dev[6];
> +	FILE *f;
> +	int ret;
> +	int res;
> +	int i;
> +
> +	sprintf(fname, "/proc/%d/exe", pid);
> +	ret = readlink(fname, mapname, PATH_MAX);
> +	if (ret >= PATH_MAX || ret < 0)
> +		return -ENOENT;
> +

readlink() will not append a null byte to the end of mapname.
You have to do this by hand.

	mapname[ret] = '\0';


> +	sprintf(fname, "/proc/%d/maps", pid);
> +	f = fopen(fname, "r");
> +	if (!f)
> +		return -ENOENT;
> +
> +	while (maps) {
> +		if (pid == maps->pid)
> +			break;
> +		maps = maps->next;
> +	}
> +
> +	ret = -ENOMEM;
> +	if (!maps) {
> +		maps = calloc(1, sizeof(*maps));
> +		if (!maps)
> +			goto out_fail;
> +		maps->pid = pid;
> +		maps->next = instance->mem_maps;
> +		instance->mem_maps = maps;
> +	} else {
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			free(maps->lib_maps[i].lib_name);
> +		free(maps->lib_maps);
> +		maps->lib_maps = NULL;
> +		maps->nr_lib_maps = 0;
> +		free(maps->proc_name);
> +	}
> +
> +	maps->proc_name = strdup(mapname);
> +	if (!maps->proc_name)
> +		goto out;
> +
> +	while (fgets(buf, sizeof(buf), f)) {
> +		mapname[0] = '\0';
> +		res = sscanf(buf, "%llx-%llx %4s %llx %5s %lld %s",
> +			     &begin, &end, perm, &tmp, dev, &inode, mapname);
> +		if (res == 7 && mapname[0] != '\0') {
> +			map = realloc(maps->lib_maps,
> +				      (maps->nr_lib_maps+1)*sizeof(*map));

minor style correction:
(maps->nr_lib_maps + 1) * sizeof(*map));


> +			if (!map)
> +				goto out_fail;
> +			map[maps->nr_lib_maps].end = end;
> +			map[maps->nr_lib_maps].start = begin;
> +			map[maps->nr_lib_maps].lib_name = strdup(mapname);
> +			if (!map[maps->nr_lib_maps].lib_name)
> +				goto out_fail;
> +			maps->lib_maps = map;
> +			maps->nr_lib_maps++;
> +		}
> +	}
> +out:
> +	fclose(f);
> +	return 0;
> +
Not shure if silently freing only tha map that failed to alocate is the 
proper thinng to do. Maybe we  should have an error message printed and 
then free all maps?
> +out_fail:
> +	fclose(f);
> +	if (maps) {
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			free(maps->lib_maps[i].lib_name);
> +		if (instance->mem_maps != maps) {
> +			m = instance->mem_maps;
> +			while (m) {
> +				if (m->next == maps) {
> +					m->next = maps->next;
> +					break;
> +				}
> +				m = m->next;
> +			}
> +		} else
> +			instance->mem_maps = maps->next;
> +		free(maps->lib_maps);
> +		maps->lib_maps = NULL;
> +		maps->nr_lib_maps = 0;
> +		free(maps->proc_name);
> +		maps->proc_name = NULL;
> +		free(maps);

Not sure if silently freeing only the map that failed to allocate memory 
is the proper thing to do. Maybe we should have an error message printed 
and then free all maps?

> +	}
> +	return ret;
> +}
> +
> +static void get_filter_pid_mmaps(void)
> +{
> +	struct filter_pids *p;
> +
> +	for (p = filter_pids; p; p = p->next) {
> +		if (p->exclude)
> +			continue;
> +		get_pid_mmaps(p->pid);
> +	}
> +}
> +
>   static void update_task_filter(void)
>   {
>   	struct buffer_instance *instance;
> @@ -1070,6 +1184,9 @@ static void update_task_filter(void)
>   	if (no_filter)
>   		return;
>   
> +	if (get_mmap && filter_pids)
> +		get_filter_pid_mmaps();
> +
>   	if (filter_task)
>   		add_filter_pid(pid, 0);
>   
> @@ -1264,6 +1381,8 @@ static void ptrace_wait(enum trace_type type, int main_pid)
>   				break;
>   
>   			case PTRACE_EVENT_EXIT:
> +				if (get_mmap)
> +					get_pid_mmaps(main_pid);
>   				ptrace(PTRACE_GETEVENTMSG, pid, NULL, &cstatus);
>   				ptrace(PTRACE_DETACH, pid, NULL, NULL);
>   				break;
> @@ -3094,6 +3213,33 @@ static void append_buffer(struct tracecmd_output *handle,
>   	}
>   }
>   
> +
> +static void
> +add_pid_mem_maps(struct tracecmd_output *handle, struct buffer_instance *instance)
> +{
> +	struct pid_mem_maps *maps = instance->mem_maps;
> +	struct trace_seq s;
> +	int i;
> +
> +	trace_seq_init(&s);
> +	while (maps) {
> +		if (!maps->nr_lib_maps)
> +			continue;
> +		trace_seq_reset(&s);
> +		trace_seq_printf(&s, "%x %x %s\n",
> +				 maps->pid, maps->nr_lib_maps, maps->proc_name);
> +		for (i = 0; i < maps->nr_lib_maps; i++)
> +			trace_seq_printf(&s, "%llx %llx %s\n",
> +					maps->lib_maps[i].start,
> +					maps->lib_maps[i].end,
> +					maps->lib_maps[i].lib_name);
> +		tracecmd_add_option(handle, TRACECMD_OPTION_PIDMMAPS,
> +				    s.len+1, s.buffer);

s.len + 1, s.buffer);

Thanks!
Yordan

> +		maps = maps->next;
> +	}
> +	trace_seq_destroy(&s);
> +}
> +
>   static void
>   add_buffer_stat(struct tracecmd_output *handle, struct buffer_instance *instance)
>   {
> @@ -3287,6 +3433,10 @@ static void record_data(struct common_record_context *ctx)
>   		if (!no_top_instance() && !top_instance.msg_handle)
>   			print_stat(&top_instance);
>   
> +		for_all_instances(instance) {
> +			add_pid_mem_maps(handle, instance);
> +		}
> +
>   		tracecmd_append_cpu_data(handle, local_cpu_count, temp_files);
>   
>   		for (i = 0; i < max_cpu_count; i++)
> @@ -4397,6 +4547,7 @@ void update_first_instance(struct buffer_instance *instance, int topt)
>   }
>   
>   enum {
> +	OPT_mmap		= 244,
>   	OPT_quiet		= 245,
>   	OPT_debug		= 246,
>   	OPT_no_filter		= 247,
> @@ -4627,6 +4778,7 @@ static void parse_record_options(int argc,
>   			{"debug", no_argument, NULL, OPT_debug},
>   			{"quiet", no_argument, NULL, OPT_quiet},
>   			{"help", no_argument, NULL, '?'},
> +			{"mmap", no_argument, NULL, OPT_mmap},
>   			{"module", required_argument, NULL, OPT_module},
>   			{NULL, 0, NULL, 0}
>   		};
> @@ -4858,6 +5010,9 @@ static void parse_record_options(int argc,
>   		case 'i':
>   			ignore_event_not_found = 1;
>   			break;
> +		case OPT_mmap:
> +			get_mmap = 1;
> +			break;
>   		case OPT_date:
>   			ctx->date = 1;
>   			if (ctx->data_flags & DATA_FL_OFFSET)
> @@ -4924,6 +5079,9 @@ static void parse_record_options(int argc,
>   		add_func(&ctx->instance->filter_funcs,
>   			 ctx->instance->filter_mod, "*");
>   
> +	if (filter_task && get_mmap)
> +		do_ptrace = 1;
> +
>   	if (do_ptrace && !filter_task && (filter_pid < 0))
>   		die(" -c can only be used with -F (or -P with event-fork support)");
>   	if (ctx->do_child && !filter_task &&! filter_pid)
>
diff mbox series

Patch

diff --git a/include/trace-cmd/trace-cmd.h b/include/trace-cmd/trace-cmd.h
index 6f62ab9..17edb9d 100644
--- a/include/trace-cmd/trace-cmd.h
+++ b/include/trace-cmd/trace-cmd.h
@@ -82,6 +82,7 @@  enum {
 	TRACECMD_OPTION_OFFSET,
 	TRACECMD_OPTION_CPUCOUNT,
 	TRACECMD_OPTION_VERSION,
+	TRACECMD_OPTION_PIDMMAPS,
 };
 
 enum {
@@ -97,6 +98,12 @@  struct tracecmd_ftrace {
 	int long_size;
 };
 
+struct lib_mem_map {
+	unsigned long long	start;
+	unsigned long long	end;
+	char			*lib_name;
+};
+
 typedef void (*tracecmd_show_data_func)(struct tracecmd_input *handle,
 					struct tep_record *record);
 typedef void (*tracecmd_handle_init_func)(struct tracecmd_input *handle,
@@ -208,6 +215,8 @@  unsigned long long tracecmd_page_ts(struct tracecmd_input *handle,
 unsigned int tracecmd_record_ts_delta(struct tracecmd_input *handle,
 				      struct tep_record *record);
 
+struct lib_mem_map *tracecmd_search_task_mmap(struct tracecmd_input *handle,
+					      int pid, unsigned long long addr);
 #ifndef SWIG
 /* hack for function graph work around */
 extern __thread struct tracecmd_input *tracecmd_curr_thread_handle;
diff --git a/lib/trace-cmd/trace-input.c b/lib/trace-cmd/trace-input.c
index 61566ba..7acecf3 100644
--- a/lib/trace-cmd/trace-input.c
+++ b/lib/trace-cmd/trace-input.c
@@ -101,6 +101,7 @@  struct tracecmd_input {
 	struct tracecmd_ftrace	finfo;
 
 	struct hook_list	*hooks;
+	struct pid_mem_maps	*pid_mmaps;
 	/* file information */
 	size_t			header_files_start;
 	size_t			ftrace_files_start;
@@ -2134,6 +2135,147 @@  void tracecmd_set_ts2secs(struct tracecmd_input *handle,
 	handle->use_trace_clock = false;
 }
 
+static int trace_pid_mmap_cmp(const void *a, const void *b)
+{
+	struct lib_mem_map *map_a = (struct lib_mem_map *)a;
+	struct lib_mem_map *map_b = (struct lib_mem_map *)b;
+
+	if (map_a->start > map_b->start)
+		return 1;
+	if (map_a->start < map_b->start)
+		return -1;
+	return 0;
+}
+
+#define STR_MMAP_LINE_MAX	(PATH_MAX+34)
+static int trace_pid_mmap_load(struct tracecmd_input *handle, char *buf)
+{
+	struct pid_mem_maps *maps = NULL;
+	char mapname[STR_MMAP_LINE_MAX];
+	char *line;
+	int res;
+	int ret;
+	int i;
+
+	maps = calloc(1, sizeof(*maps));
+	if (!maps)
+		return -ENOMEM;
+
+	ret  = -EINVAL;
+	line = strchr(buf, '\n');
+	if (!line)
+		goto out_fail;
+
+	*line = '\0';
+	if (strlen(buf) > STR_MMAP_LINE_MAX)
+		goto out_fail;
+
+	res = sscanf(buf, "%x %x %s", &maps->pid, &maps->nr_lib_maps, mapname);
+	if (res != 3)
+		goto out_fail;
+
+	ret  = -ENOMEM;
+	maps->proc_name = strdup(mapname);
+	if (!maps->proc_name)
+		goto out_fail;
+
+	maps->lib_maps = calloc(maps->nr_lib_maps, sizeof(struct lib_mem_map));
+	if (!maps->lib_maps)
+		goto out_fail;
+
+	buf = line+1;
+	line = strchr(buf, '\n');
+	for (i = 0; i < maps->nr_lib_maps; i++) {
+		if (!line)
+			break;
+		*line = '\0';
+		if (strlen(buf) > STR_MMAP_LINE_MAX)
+			break;
+		res = sscanf(buf, "%llx %llx %s", &maps->lib_maps[i].start,
+			     &maps->lib_maps[i].end, mapname);
+		if (res != 3)
+			break;
+		maps->lib_maps[i].lib_name = strdup(mapname);
+		if (!maps->lib_maps[i].lib_name)
+			goto out_fail;
+		buf = line+1;
+		line = strchr(buf, '\n');
+	}
+
+	ret  = -EINVAL;
+	if (i != maps->nr_lib_maps)
+		goto out_fail;
+
+	qsort(maps->lib_maps, maps->nr_lib_maps,
+	      sizeof(*maps->lib_maps), trace_pid_mmap_cmp);
+
+	maps->next = handle->pid_mmaps;
+	handle->pid_mmaps = maps;
+
+	return 0;
+
+out_fail:
+	if (maps->lib_maps) {
+		for (i = 0; i < maps->nr_lib_maps; i++)
+			free(maps->lib_maps[i].lib_name);
+		free(maps->lib_maps);
+	}
+	free(maps->proc_name);
+	free(maps);
+
+	return ret;
+}
+
+static int trace_pid_mmap_search(const void *a, const void *b)
+{
+	struct lib_mem_map *key = (struct lib_mem_map *)a;
+	struct lib_mem_map *map = (struct lib_mem_map *)b;
+
+	if (key->start >= map->end)
+		return 1;
+	if (key->start < map->start)
+		return -1;
+	return 0;
+}
+
+/**
+ * tracecmd_search_task_mmap - Search task memory address map
+ * @handle: input handle to the trace.dat file
+ * @pid: pid of the task
+ * @addr: address from the task memory space.
+ *
+ * Map of the task memory can be saved in the trace.dat file, using the option
+ * "--mmap". If there is such information, this API can be used to look up into
+ * this memory map to find what library is loaded at the given @addr.
+ *
+ * A pointer to struct lib_mem_map is returned, containing the name of
+ * the library at given task @addr and the library start and end addresses.
+ */
+struct lib_mem_map *tracecmd_search_task_mmap(struct tracecmd_input *handle,
+					      int pid, unsigned long long addr)
+{
+	struct pid_mem_maps *maps;
+	struct lib_mem_map *lib;
+	struct lib_mem_map key;
+
+	if (!handle || !handle->pid_mmaps)
+		return NULL;
+
+	maps = handle->pid_mmaps;
+	while (maps) {
+		if (maps->pid == pid)
+			break;
+		maps = maps->next;
+	}
+	if (!maps || !maps->nr_lib_maps || !maps->lib_maps)
+		return NULL;
+	key.start = addr;
+	lib = bsearch(&key, maps->lib_maps, maps->nr_lib_maps,
+		      sizeof(*maps->lib_maps), trace_pid_mmap_search);
+
+	return lib;
+}
+
 static int handle_options(struct tracecmd_input *handle)
 {
 	unsigned long long offset;
@@ -2221,9 +2363,6 @@  static int handle_options(struct tracecmd_input *handle)
 		case TRACECMD_OPTION_UNAME:
 			handle->uname = strdup(buf);
 			break;
-		case TRACECMD_OPTION_VERSION:
-			handle->version = strdup(buf);
-			break;
 		case TRACECMD_OPTION_HOOK:
 			hook = tracecmd_create_event_hook(buf);
 			hook->next = handle->hooks;
@@ -2233,6 +2372,10 @@  static int handle_options(struct tracecmd_input *handle)
 			cpus = *(int *)buf;
 			handle->cpus = tep_read_number(handle->pevent, &cpus, 4);
 			break;
+		case TRACECMD_OPTION_PIDMMAPS:
+			if (buf[size] == '\0')
+				trace_pid_mmap_load(handle, buf);
+			break;
 		default:
 			warning("unknown option %d", option);
 			break;
diff --git a/tracecmd/include/trace-local.h b/tracecmd/include/trace-local.h
index 1cad3cc..ae1632c 100644
--- a/tracecmd/include/trace-local.h
+++ b/tracecmd/include/trace-local.h
@@ -157,6 +157,14 @@  struct func_list {
 	const char *mod;
 };
 
+struct pid_mem_maps {
+	struct pid_mem_maps	*next;
+	struct lib_mem_map	*lib_maps;
+	unsigned int		nr_lib_maps;
+	char			*proc_name;
+	int			pid;
+};
+
 struct buffer_instance {
 	struct buffer_instance	*next;
 	const char		*name;
@@ -183,6 +191,8 @@  struct buffer_instance {
 	struct tracecmd_msg_handle *msg_handle;
 	struct tracecmd_output *network_handle;
 
+	struct pid_mem_maps	*mem_maps;
+
 	char			*max_graph_depth;
 
 	int			flags;
diff --git a/tracecmd/trace-record.c b/tracecmd/trace-record.c
index 5dc6f17..edabf53 100644
--- a/tracecmd/trace-record.c
+++ b/tracecmd/trace-record.c
@@ -84,6 +84,7 @@  static int max_kb;
 static bool use_tcp;
 
 static int do_ptrace;
+static int get_mmap;
 
 static int filter_task;
 static int filter_pid = -1;
@@ -1062,6 +1063,119 @@  static char *make_pid_filter(char *curr_filter, const char *field)
 	return filter;
 }
 
+static int get_pid_mmaps(int pid)
+{
+	struct buffer_instance *instance = &top_instance;
+	struct pid_mem_maps *maps = instance->mem_maps;
+	struct pid_mem_maps *m;
+	unsigned long long begin, end, inode, tmp;
+	struct lib_mem_map *map;
+	char mapname[PATH_MAX+1];
+	char fname[PATH_MAX+1];
+	char buf[PATH_MAX+100];
+	char perm[5];
+	char dev[6];
+	FILE *f;
+	int ret;
+	int res;
+	int i;
+
+	sprintf(fname, "/proc/%d/exe", pid);
+	ret = readlink(fname, mapname, PATH_MAX);
+	if (ret >= PATH_MAX || ret < 0)
+		return -ENOENT;
+
+	sprintf(fname, "/proc/%d/maps", pid);
+	f = fopen(fname, "r");
+	if (!f)
+		return -ENOENT;
+
+	while (maps) {
+		if (pid == maps->pid)
+			break;
+		maps = maps->next;
+	}
+
+	ret = -ENOMEM;
+	if (!maps) {
+		maps = calloc(1, sizeof(*maps));
+		if (!maps)
+			goto out_fail;
+		maps->pid = pid;
+		maps->next = instance->mem_maps;
+		instance->mem_maps = maps;
+	} else {
+		for (i = 0; i < maps->nr_lib_maps; i++)
+			free(maps->lib_maps[i].lib_name);
+		free(maps->lib_maps);
+		maps->lib_maps = NULL;
+		maps->nr_lib_maps = 0;
+		free(maps->proc_name);
+	}
+
+	maps->proc_name = strdup(mapname);
+	if (!maps->proc_name)
+		goto out;
+
+	while (fgets(buf, sizeof(buf), f)) {
+		mapname[0] = '\0';
+		res = sscanf(buf, "%llx-%llx %4s %llx %5s %lld %s",
+			     &begin, &end, perm, &tmp, dev, &inode, mapname);
+		if (res == 7 && mapname[0] != '\0') {
+			map = realloc(maps->lib_maps,
+				      (maps->nr_lib_maps+1)*sizeof(*map));
+			if (!map)
+				goto out_fail;
+			map[maps->nr_lib_maps].end = end;
+			map[maps->nr_lib_maps].start = begin;
+			map[maps->nr_lib_maps].lib_name = strdup(mapname);
+			if (!map[maps->nr_lib_maps].lib_name)
+				goto out_fail;
+			maps->lib_maps = map;
+			maps->nr_lib_maps++;
+		}
+	}
+out:
+	fclose(f);
+	return 0;
+
+out_fail:
+	fclose(f);
+	if (maps) {
+		for (i = 0; i < maps->nr_lib_maps; i++)
+			free(maps->lib_maps[i].lib_name);
+		if (instance->mem_maps != maps) {
+			m = instance->mem_maps;
+			while (m) {
+				if (m->next == maps) {
+					m->next = maps->next;
+					break;
+				}
+				m = m->next;
+			}
+		} else
+			instance->mem_maps = maps->next;
+		free(maps->lib_maps);
+		maps->lib_maps = NULL;
+		maps->nr_lib_maps = 0;
+		free(maps->proc_name);
+		maps->proc_name = NULL;
+		free(maps);
+	}
+	return ret;
+}
+
+static void get_filter_pid_mmaps(void)
+{
+	struct filter_pids *p;
+
+	for (p = filter_pids; p; p = p->next) {
+		if (p->exclude)
+			continue;
+		get_pid_mmaps(p->pid);
+	}
+}
+
 static void update_task_filter(void)
 {
 	struct buffer_instance *instance;
@@ -1070,6 +1184,9 @@  static void update_task_filter(void)
 	if (no_filter)
 		return;
 
+	if (get_mmap && filter_pids)
+		get_filter_pid_mmaps();
+
 	if (filter_task)
 		add_filter_pid(pid, 0);
 
@@ -1264,6 +1381,8 @@  static void ptrace_wait(enum trace_type type, int main_pid)
 				break;
 
 			case PTRACE_EVENT_EXIT:
+				if (get_mmap)
+					get_pid_mmaps(main_pid);
 				ptrace(PTRACE_GETEVENTMSG, pid, NULL, &cstatus);
 				ptrace(PTRACE_DETACH, pid, NULL, NULL);
 				break;
@@ -3094,6 +3213,33 @@  static void append_buffer(struct tracecmd_output *handle,
 	}
 }
 
+
+static void
+add_pid_mem_maps(struct tracecmd_output *handle, struct buffer_instance *instance)
+{
+	struct pid_mem_maps *maps = instance->mem_maps;
+	struct trace_seq s;
+	int i;
+
+	trace_seq_init(&s);
+	while (maps) {
+		if (!maps->nr_lib_maps)
+			continue;
+		trace_seq_reset(&s);
+		trace_seq_printf(&s, "%x %x %s\n",
+				 maps->pid, maps->nr_lib_maps, maps->proc_name);
+		for (i = 0; i < maps->nr_lib_maps; i++)
+			trace_seq_printf(&s, "%llx %llx %s\n",
+					maps->lib_maps[i].start,
+					maps->lib_maps[i].end,
+					maps->lib_maps[i].lib_name);
+		tracecmd_add_option(handle, TRACECMD_OPTION_PIDMMAPS,
+				    s.len+1, s.buffer);
+		maps = maps->next;
+	}
+	trace_seq_destroy(&s);
+}
+
 static void
 add_buffer_stat(struct tracecmd_output *handle, struct buffer_instance *instance)
 {
@@ -3287,6 +3433,10 @@  static void record_data(struct common_record_context *ctx)
 		if (!no_top_instance() && !top_instance.msg_handle)
 			print_stat(&top_instance);
 
+		for_all_instances(instance) {
+			add_pid_mem_maps(handle, instance);
+		}
+
 		tracecmd_append_cpu_data(handle, local_cpu_count, temp_files);
 
 		for (i = 0; i < max_cpu_count; i++)
@@ -4397,6 +4547,7 @@  void update_first_instance(struct buffer_instance *instance, int topt)
 }
 
 enum {
+	OPT_mmap		= 244,
 	OPT_quiet		= 245,
 	OPT_debug		= 246,
 	OPT_no_filter		= 247,
@@ -4627,6 +4778,7 @@  static void parse_record_options(int argc,
 			{"debug", no_argument, NULL, OPT_debug},
 			{"quiet", no_argument, NULL, OPT_quiet},
 			{"help", no_argument, NULL, '?'},
+			{"mmap", no_argument, NULL, OPT_mmap},
 			{"module", required_argument, NULL, OPT_module},
 			{NULL, 0, NULL, 0}
 		};
@@ -4858,6 +5010,9 @@  static void parse_record_options(int argc,
 		case 'i':
 			ignore_event_not_found = 1;
 			break;
+		case OPT_mmap:
+			get_mmap = 1;
+			break;
 		case OPT_date:
 			ctx->date = 1;
 			if (ctx->data_flags & DATA_FL_OFFSET)
@@ -4924,6 +5079,9 @@  static void parse_record_options(int argc,
 		add_func(&ctx->instance->filter_funcs,
 			 ctx->instance->filter_mod, "*");
 
+	if (filter_task && get_mmap)
+		do_ptrace = 1;
+
 	if (do_ptrace && !filter_task && (filter_pid < 0))
 		die(" -c can only be used with -F (or -P with event-fork support)");
 	if (ctx->do_child && !filter_task &&! filter_pid)