/*
Copyright (C) 2013- The University of Notre Dame
This software is distributed under the GNU General Public License.
See the file COPYING for details.
*/

#include <assert.h>
#include <stdio.h>
#include <string.h>
#include <strings.h>
#include <errno.h>
#include <limits.h>
#include <math.h>
#include <sys/time.h>

#include "debug.h"
#include "path_disk_size_info.h"
#include "macros.h"
#include "stringtools.h"
#include "xxmalloc.h"
#include "load_average.h"


#include "rmonitor_poll_internal.h"
#include "host_memory_info.h"
#include "path_disk_size_info.h"
#include "load_average.h"

#ifndef PATH_MAX
#define PATH_MAX 1024
#endif

/***
 * Helper functions
***/

#define ANON_MAPS_NAME "[anon]"

uint64_t usecs_since_epoch()
{
	uint64_t usecs;
	struct timeval time;

	gettimeofday(&time, NULL);

	usecs  = time.tv_sec;
	usecs *= ONE_SECOND;
	usecs += time.tv_usec;

	return usecs;
}

/***
 * Functions to track the whole process tree.  They call the
 * functions defined just above, accumulating the resources of
 * all the processes.
***/

void rmonitor_poll_all_processes_once(struct itable *processes, struct rmonitor_process_info *acc)
{
	uint64_t pid;
	struct rmonitor_process_info *p;

	bzero(acc, sizeof( struct rmonitor_process_info ));

	itable_firstkey(processes);
	while(itable_nextkey(processes, &pid, (void **) &p))
	{
		int status = rmonitor_poll_process_once(p);

		/* do not consider process if some error is found. */
		if(status != 0)
			continue;

		acc_mem_usage(&acc->mem, &p->mem);
		acc_cpu_time_usage(&acc->cpu, &p->cpu);
		acc_ctxsw_usage(&acc->ctx, &p->ctx);

		acc_sys_io_usage(&acc->io, &p->io);
		acc_map_io_usage(&acc->io, &p->io);
	}

	rmonitor_get_loadavg(&acc->load);
}

void rmonitor_poll_all_wds_once(struct hash_table *wdirs, struct rmonitor_wdir_info *acc, int max_time_for_measurement)
{
	struct rmonitor_wdir_info *d;
	char *path;

	bzero(acc, sizeof( struct rmonitor_wdir_info ));

	if(hash_table_size(wdirs) > 0) {
		if(max_time_for_measurement > 0) {
			/* split time available across all directories to measure. */
			max_time_for_measurement = MAX(1, max_time_for_measurement/hash_table_size(wdirs));
		}

		hash_table_firstkey(wdirs);
		while(hash_table_nextkey(wdirs, &path, (void **) &d))
		{
			int status = rmonitor_poll_wd_once(d, max_time_for_measurement);
			/* do not consider directory if some error is found. */
			if(status != 0)
				continue;

			acc_wd_usage(acc, d);
		}
	}
}

void rmonitor_poll_all_fss_once(struct itable *filesysms, struct rmonitor_filesys_info *acc)
{
	struct rmonitor_filesys_info *f;
	uint64_t dev_id;

	bzero(acc, sizeof( struct rmonitor_filesys_info ));

	itable_firstkey(filesysms);
	while(itable_nextkey(filesysms, &dev_id, (void **) &f))
	{
		int status = rmonitor_poll_fs_once(f);
		/* do not consider fs if some error is found. */
		if(status != 0)
			continue;

		acc_dsk_usage(&acc->disk, &f->disk);
	}
}


/***
 * Functions to monitor a single process, workind directory, or
 * filesystem.
***/

int rmonitor_poll_process_once(struct rmonitor_process_info *p)
{
	int status = 0;

	debug(D_RMON, "monitoring process: %d\n", p->pid);

	status |= rmonitor_get_cpu_time_usage(p->pid, &p->cpu);
	status |= rmonitor_get_ctxsw_usage(p->pid, &p->ctx);
	status |= rmonitor_get_mem_usage(p->pid, &p->mem);
	status |= rmonitor_get_sys_io_usage(p->pid, &p->io);

	return status;
}

int rmonitor_poll_wd_once(struct rmonitor_wdir_info *d, int max_time_for_measurement)
{
	debug(D_RMON, "monitoring dir %s\n", d->path);

	int status = rmonitor_get_wd_usage(d, max_time_for_measurement);

	return status;
}

int rmonitor_poll_fs_once(struct rmonitor_filesys_info *f)
{
	int status = rmonitor_get_dsk_usage(f->path, &f->disk);

	if(!status) {
		f->disk.f_bfree  = f->disk_initial.f_bfree  - f->disk.f_bfree;
		f->disk.f_bavail = f->disk_initial.f_bavail - f->disk.f_bavail;
		f->disk.f_ffree  = f->disk_initial.f_ffree  - f->disk.f_ffree;
	}

	return status;
}

/***
 * Utility functions (open log files, proc files, measure time)
 ***/

FILE *open_proc_file(pid_t pid, char *filename)
{
		FILE *fproc;
		char fproc_path[PATH_MAX];

#if defined(CCTOOLS_OPSYS_DARWIN)
		return NULL;
#endif

		if(pid > -1)
		{
			sprintf(fproc_path, "/proc/%d/%s", pid, filename);
		}
		else
		{
			sprintf(fproc_path, "/proc/%s", filename);
		}

		if((fproc = fopen(fproc_path, "r")) == NULL)
		{
				debug(D_RMON, "could not process file %s : %s\n", fproc_path, strerror(errno));
				return NULL;
		}

		return fproc;
}

/* Parse a /proc file looking for line attribute: value */
int rmonitor_get_int_attribute(FILE *fstatus, char *attribute, uint64_t *value, int rewind_flag)
{
	char proc_attr_line[PATH_MAX];
	int not_found = 1;
	int n = strlen(attribute);

	if(!fstatus)
		return not_found;

	proc_attr_line[PATH_MAX - 2] = '\0';
	proc_attr_line[PATH_MAX - 1] = '\0';

	if(rewind_flag)
		rewind(fstatus);

	while( fgets(proc_attr_line, PATH_MAX - 2, fstatus) )
	{
		if(strncmp(attribute, proc_attr_line, n) == 0)
		{
			//We make sure that fgets got a whole line
			if(proc_attr_line[PATH_MAX - 2] == '\n')
				proc_attr_line[PATH_MAX - 2] = '\0';
			if(strlen(proc_attr_line) == PATH_MAX - 2)
				return -1;

			sscanf(proc_attr_line, "%*s %" SCNu64, value);
			not_found = 0;
			break;
		}
	}

	return not_found;
}

uint64_t clicks_to_usecs(uint64_t clicks)
{
	return ((clicks * ONE_SECOND) / sysconf(_SC_CLK_TCK));
}

/***
 * Low level resource monitor functions.
 ***/

/***
 Get children of a process by polling, rather than capturing fork return value
 is number of children found. The array of pid values is returned in *children.
 ***/
int rmonitor_get_children(pid_t pid, uint64_t **children)
{
	/* /proc/[pid]/task/[pid]/children */

	char *fchildren_path = string_format("/proc/%d/task/%d/children", pid, pid);
	FILE *fstat = fopen(fchildren_path, "r");
	free(fchildren_path);

	if(!fstat) {
		return 0;
	}

	int count = 0;
	int max   = 0;
	uint64_t child;

	uint64_t *child_list = NULL;

	while(fscanf(fstat, "%" PRIu64, &child) == 1) {
		count++;
		if(count >= max) {
			max = 2*count;
			child_list = realloc(child_list, max*sizeof(uint64_t));
		}

		child_list[count - 1] = child;
	}

	*children = child_list;

	fclose(fstat);

	return count;
}


int rmonitor_get_start_time(pid_t pid, uint64_t *start_time)
{
	/* /proc/[pid]/stat */

	uint64_t start_clicks;
	double uptime;

	FILE *fstat = open_proc_file(pid, "stat");
	if(!fstat)
		return 1;

	int n;
	n = fscanf(fstat,
			"%*s" /* pid */ "%*s" /* cmd line */ "%*s" /* state */ "%*s" /* pid of parent */
			"%*s" /* group ID */ "%*s" /* session id */ "%*s" /* tty pid */ "%*s" /* tty group ID */
			"%*s" /* linux/sched.h flags */ "%*s %*s %*s %*s" /* faults */
			"%*s" /* user mode time (in clock ticks)   (field 14)  */
			"%*s" /* kernel mode time (in clock ticks) (field 15) */
			"%*s" /* time (clock ticks) waiting for children in user mode */
			"%*s" /* time (clock ticks) waiting for children in kernel mode */
			"%*s" /* priority */ "%*s" /* nice */
			"%*s" /* num threads */ "%*s" /* always 0 */
			"%" SCNu64 /* clock ticks since start     (field 22) */
			/* .... */,
			&start_clicks);
	fclose(fstat);

	if(n != 1)
		return 1;

	FILE *fuptime = open_proc_file(-1, "uptime");
	if(!fuptime)
		return 1;

	n = fscanf(fuptime, "%lf %*s", &uptime);
	fclose(fuptime);

	if(n != 1)
		return 1;

	uint64_t origin = usecs_since_epoch() - (uptime * ONE_SECOND);
	*start_time     = origin + clicks_to_usecs(start_clicks);

	return 0;
}

int rmonitor_get_cpu_time_usage(pid_t pid, struct rmonitor_cpu_time_info *cpu)
{
	/* /proc/[pid]/stat */

	uint64_t kernel, user;


	FILE *fstat = open_proc_file(pid, "stat");
	if(!fstat)
		return 1;

	int n;
	n = fscanf(fstat,
			"%*s" /* pid */ "%*s" /* cmd line */ "%*s" /* state */ "%*s" /* pid of parent */
			"%*s" /* group ID */ "%*s" /* session id */ "%*s" /* tty pid */ "%*s" /* tty group ID */
			"%*s" /* linux/sched.h flags */ "%*s %*s %*s %*s" /* faults */
			"%" SCNu64 /* user mode time (in clock ticks) */
			"%" SCNu64 /* kernel mode time (in clock ticks) */
			/* .... */,
			&kernel, &user);
	fclose(fstat);

	if(n != 2)
		return 1;

	uint64_t accum = clicks_to_usecs(kernel) + clicks_to_usecs(user);

	cpu->delta       = accum  - cpu->accumulated;
	cpu->accumulated = accum;

	return 0;
}

void acc_cpu_time_usage(struct rmonitor_cpu_time_info *acc, struct rmonitor_cpu_time_info *other)
{
	acc->delta += other->delta;
}

int rmonitor_get_ctxsw_usage(pid_t pid, struct rmonitor_ctxsw_info *switches)
{
	/* /proc/[pid]/stat */

	int notfound;
	uint64_t vol_switches = 0;
	uint64_t nonvol_switches = 0;

	FILE *fstat = open_proc_file(pid, "status");
	if(!fstat) {
		return 0;
	}

	notfound = 0;

	notfound |= rmonitor_get_int_attribute(fstat, "voluntary_ctxt_switches:", &vol_switches, 1);
	notfound |= rmonitor_get_int_attribute(fstat, "nonvoluntary_ctxt_switches:", &nonvol_switches, 0);

	uint64_t accum = vol_switches + nonvol_switches;

	switches->delta       = accum  - switches->accumulated;
	switches->accumulated = accum;

	fclose(fstat);
	return notfound;
}

void acc_ctxsw_usage(struct rmonitor_ctxsw_info *acc, struct rmonitor_ctxsw_info *other)
{
	acc->delta += other->delta;
}

int rmonitor_get_loadavg(struct rmonitor_load_info *load)
{
	double last_minute;
	int status = getloadavg(&last_minute, 1);

	if(status != 1) {
		last_minute = -1;
	}

	load->last_minute = last_minute;
	load->cpus        = load_average_get_cpus();

	return 0;
}

int rmonitor_get_mem_usage(pid_t pid, struct rmonitor_mem_info *mem)
{
	// /proc/[pid]/status:

	FILE *fmem = open_proc_file(pid, "status");
	if(!fmem)
		return 1;

	int status = 0;
	/* in kB */
	status |= rmonitor_get_int_attribute(fmem, "VmPeak:", &mem->virtual,  1);
	status |= rmonitor_get_int_attribute(fmem, "VmHWM:",  &mem->resident, 1);
	status |= rmonitor_get_int_attribute(fmem, "VmLib:",  &mem->shared,   1);
	status |= rmonitor_get_int_attribute(fmem, "VmExe:",  &mem->text,     1);
	status |= rmonitor_get_int_attribute(fmem, "VmData:", &mem->data,     1);

	/* from smaps when reading maps. */
	mem->swap = 0;

	fclose(fmem);

	/* in MB */
	mem->virtual  = DIV_INT_ROUND_UP(mem->virtual,  1024);
	mem->resident = DIV_INT_ROUND_UP(mem->resident, 1024);
	mem->text     = DIV_INT_ROUND_UP(mem->text,     1024);
	mem->data     = DIV_INT_ROUND_UP(mem->data,     1024);
	mem->shared   = DIV_INT_ROUND_UP(mem->shared,   1024);

	return status;
}

void acc_mem_usage(struct rmonitor_mem_info *acc, struct rmonitor_mem_info *other)
{
		acc->virtual  += other->virtual;
		acc->resident += other->resident;
		acc->data     += other->data;
		acc->swap     += other->swap;
		acc->shared   += other->shared;
}

struct rmonitor_mem_info *rmonitor_get_map_info(FILE *fmem, int rewind_flag) {
	static int anon_map_count = 0;

	if(!fmem)
		return NULL;

	if(rewind_flag)
		rewind(fmem);

	struct rmonitor_mem_info *info = malloc(sizeof(struct rmonitor_mem_info));

	uint64_t offset;
	char map_info_line[PATH_MAX];
	char map_name_found[PATH_MAX];
	while( fgets(map_info_line, PATH_MAX, fmem) )
	{
		// start-end                 perm   offset device inode                       path
		// 560019f25000-56001a127000 r-xp 00000000 08:01 266469                     /usr/bin/vim.basic

		int n;
		n = sscanf(map_info_line, "%llx-%llx %*s %llx %*s %*s %s", (long long unsigned int *) &(info->map_start), (long long unsigned int *)  &(info->map_end), (long long unsigned int *) &offset, map_name_found);

		/* continue if we do not get at least start, end, and offset */
		if(n < 3)
			continue;

		/* file maps are always an absolute pathname. consider maps without a filename as different. */
		if(n < 4 || map_name_found[0] != '/') {
			info->map_name = string_format("ANON_MAPS_NAME.%d", anon_map_count);
			anon_map_count++;
		} else {
			info->map_name = xxstrdup(map_name_found);
		}

		// move boundaries to origin
		info->map_end   = info->map_end - info->map_start + offset;
		info->map_start = offset;

		return info;
	}

	free(info);

	return NULL;
}

static double rmonitor_mem_info_priority(void *item) {
	assert(item);
	struct rmonitor_mem_info *i = item;
	return -1*(i->map_start);
}

int rmonitor_get_mmaps_usage(pid_t pid, struct hash_table *maps)
{
	// /proc/[pid]/smaps:

	FILE *fmem = open_proc_file(pid, "smaps");
	if(!fmem)
		return 1;

	struct rmonitor_mem_info *info;
	while((info = rmonitor_get_map_info(fmem, 0))) {

		uint64_t rss, pss, swap, ref;
		uint64_t private_dirty, private_clean;

		int status = 0;

		/* order is important, this is how the fields appear in smaps */
		/* in kB! */
		status |= rmonitor_get_int_attribute(fmem, "Rss:",           &rss, 0);
		status |= rmonitor_get_int_attribute(fmem, "Pss:",           &pss, 0);
		status |= rmonitor_get_int_attribute(fmem, "Private_Clean:", &private_clean, 0);
		status |= rmonitor_get_int_attribute(fmem, "Private_Dirty:", &private_dirty, 0);
		status |= rmonitor_get_int_attribute(fmem, "Referenced:",    &ref, 0);
		status |= rmonitor_get_int_attribute(fmem, "Swap:",          &swap, 0);

		/* error reading a field, we simply skip the record. */
		if(status) {
			free(info);
			continue;
		}

		info->resident   = rss;
		info->referenced = ref;
		info->swap       = swap;

		/* private and shared may or may not be currently resident, (e.g.,
		 swap). That is: rss = private + shared - swap = referenced - swap.  In
		 the following, we try to compute private and shared that are actually
		 resident. Since we do not have enough information, we assume the worst
		 case that all private pages are resident. If swap is zero, then
		 resident private and resident shared will have the correct values. */

		info->private  = MIN(private_dirty + private_clean, rss);
		info->shared   = MAX(rss - info->private, 0);

		/* add the info to a sorted list per map, by start. Overlaping maps will be merged later. */
		struct list *infos = hash_table_lookup(maps, info->map_name);
		if(!infos) {
			infos = list_create();
			hash_table_insert(maps, info->map_name, infos);
		}

		list_push_priority(infos, rmonitor_mem_info_priority, info);
	}

	fclose(fmem);

	return 0;
}

int rmonitor_poll_maps_once(struct itable *processes, struct rmonitor_mem_info *mem) {
	/* set result to 0. */
	bzero(mem, sizeof(struct rmonitor_mem_info));

	struct hash_table *maps_per_file = hash_table_create(0, 0);

	uint64_t pid;
	struct rmonitor_process_info *pinfo;
	itable_firstkey(processes);
	while(itable_nextkey(processes, &pid, (void *) &pinfo)) {
		rmonitor_get_mmaps_usage(pid, maps_per_file);
	}

	/* Accumulate the maps we just found per file. First, we merge together all
	 * the maps segment that overlap. With this, we do not overcount private
	 * segments, but do consider that segments are shared as little as
	 * possible.
	 *
	 * After this merging, we determine upper bounds, such as:
	 *
	 * virtual >= referenced >= private >= referenced - private >= shared.
	 *
	 * If swap size is zero, then virtual and private have the exact values,
	 * and we have worst case counts for referenced and shared.
	 *
	 * When we accumulate resident for the result, we do it from private +
	 * shared, rather than the resident reported originally as this would
	 * overcount shared.
	 *
	 * There could be a way to use Pss (proportional resident) to improve these
	 * bounds.
	 */

	char *map_name;
	struct list *infos;

	hash_table_firstkey(maps_per_file);
	while(hash_table_nextkey(maps_per_file, &map_name, (void *) &infos )) {

		struct rmonitor_mem_info *info, *next;
		while((info = list_pop_head(infos))) {
			while((next = list_peek_head(infos))) {
				/* do we need to merge with the next segment? */
				if(info->map_end > next->map_start) {
					info->private  += next->private;
					info->shared   += next->shared;
					info->resident += next->resident;
					info->referenced += next->referenced;
					info->swap     += next->swap;

					info->map_end = MAX(info->map_end, next->map_end);

					list_pop_head(infos);
					if(next->map_name)
						free(next->map_name);
					free(next);
				} else {
					break;
				}
			}

			/* a series of upper bounds: */
			/* by adding referenced, we assumed a worst case of non-sharing
			 * memory, but referenced cannot be larger than the virtual size: */
			info->virtual  = DIV_INT_ROUND_UP(info->map_end - info->map_start, 1024); /* bytes to kB. */
			info->referenced = MIN(info->referenced, info->virtual);

			/* similarly, resident cannot be larger than referenced. */
			info->resident = MIN(info->resident, info->referenced);

			/* and, resident private cannot be larger than resident. */
			info->private  = MIN(info->private, info->resident);

			/* lastly, resident shared memory cannot be larger than the whole
			 * resident size minus the resident private memory. */
			info->shared = MIN(info->shared, info->resident - info->private);

			/* once the individual values have been found, we added together to the result. */
			mem->virtual     += info->virtual;
			mem->referenced  += info->referenced;
			mem->shared      += info->shared;
			mem->private     += info->private;

			/* note that we add private + shared, rather than resident,
			 * otherwise we will overcount shared. */
			mem->resident += info->private + info->shared;

			if(info->map_name)
				free(info->map_name);
			free(info);
		}

		list_delete(infos);
	}

	hash_table_delete(maps_per_file);

	/* all the values computed are in kB, we convert to MB. */
	mem->virtual      = DIV_INT_ROUND_UP(mem->virtual,  1024);
	mem->shared       = DIV_INT_ROUND_UP(mem->shared,   1024);
	mem->private      = DIV_INT_ROUND_UP(mem->private,  1024);
	mem->resident     = DIV_INT_ROUND_UP(mem->resident, 1024);

	return 0;
}


int rmonitor_get_sys_io_usage(pid_t pid, struct rmonitor_io_info *io)
{
	/* /proc/[pid]/io: if process dies before we read the file,
	   then info is lost, as if the process did not read or write
	   any characters.
	*/

	FILE *fio = open_proc_file(pid, "io");
	uint64_t cread, cwritten;
	int rstatus, wstatus;

	io->delta_chars_read = 0;
	io->delta_chars_written = 0;

	if(!fio)
		return 1;

	/* We really want "bytes_read", but there are issues with
	 * distributed filesystems. Instead, we also count page
	 * faulting in another function below. */
	rstatus  = rmonitor_get_int_attribute(fio, "rchar", &cread, 1);
	wstatus  = rmonitor_get_int_attribute(fio, "write_bytes", &cwritten, 1);

	fclose(fio);

	if(rstatus || wstatus)
		return 1;

	io->delta_chars_read    = cread    - io->chars_read;
	io->delta_chars_written = cwritten - io->chars_written;

	io->chars_read = cread;
	io->chars_written = cwritten;

	return 0;
}

void acc_sys_io_usage(struct rmonitor_io_info *acc, struct rmonitor_io_info *other)
{
	acc->delta_chars_read    += other->delta_chars_read;
	acc->delta_chars_written += other->delta_chars_written;
}

/* We compute the resident memory changes from mmap files. */
int rmonitor_get_map_io_usage(pid_t pid, struct rmonitor_io_info *io)
{
	/* /proc/[pid]/smaps */

	uint64_t kbytes_resident_accum;
	uint64_t kbytes_resident;

	kbytes_resident_accum    = 0;
	io->delta_bytes_faulted = 0;

	FILE *fsmaps = open_proc_file(pid, "smaps");
	if(!fsmaps)
	{
		return 1;
	}

	char dummy_line[1024];

	/* Look for next mmap file */
	while(fgets(dummy_line, 1024, fsmaps))
		if(strchr(dummy_line, '/'))
			if(rmonitor_get_int_attribute(fsmaps, "Rss:", &kbytes_resident, 0) == 0)
				kbytes_resident_accum += kbytes_resident;

	if((kbytes_resident_accum * 1024) > io->bytes_faulted)
		io->delta_bytes_faulted = (kbytes_resident_accum * 1024) - io->bytes_faulted;

	/* in bytes */
	io->bytes_faulted = (kbytes_resident_accum * 1024);

	fclose(fsmaps);

	return 0;
}

void acc_map_io_usage(struct rmonitor_io_info *acc, struct rmonitor_io_info *other)
{
	acc->delta_bytes_faulted += other->delta_bytes_faulted;
}


/***
 * Low level resource monitor functions.
 ***/

int rmonitor_get_dsk_usage(const char *path, struct statfs *disk)
{
	char cwd[PATH_MAX];

	debug(D_RMON, "statfs on path: %s\n", path);

	if(statfs(path, disk) > 0)
	{
		debug(D_RMON, "could not statfs on %s : %s\n", cwd, strerror(errno));
		return 1;
	}

	return 0;
}

void acc_dsk_usage(struct statfs *acc, struct statfs *other)
{
	acc->f_bfree  += other->f_bfree;
	acc->f_bavail += other->f_bavail;
	acc->f_ffree  += other->f_ffree;
}

int rmonitor_get_wd_usage(struct rmonitor_wdir_info *d, int max_time_for_measurement)
{
	/* We need a pointer to a pointer, which it is not possible from a struct. Use a dummy variable. */
	struct path_disk_size_info *state = d->state;
	int status = path_disk_size_info_get_r(d->path, max_time_for_measurement, &state);

	d->state = state;

	d->files = d->state->last_file_count_complete;
	d->byte_count = d->state->last_byte_size_complete;

	return status;
}

void acc_wd_usage(struct rmonitor_wdir_info *acc, struct rmonitor_wdir_info *other)
{
	acc->files       += other->files;
	acc->byte_count  += other->byte_count;
}

char *rmonitor_get_command_line(pid_t pid)
{
	/* /proc/[pid]/cmdline */

	FILE *fline = open_proc_file(pid, "cmdline");
	if(!fline)
		return NULL;

	char cmdline[PATH_MAX];
	ssize_t cmdline_len = read(fileno(fline), cmdline, PATH_MAX);

	if(cmdline_len < 1)
		return NULL;

	int i;
	for(i=0; i < cmdline_len - 1; i++) { /* -1 because cmdline ends with two \0. */
		if(cmdline[i] == '\0')
			cmdline[i] = ' ';
	}
        fclose(fline);

	return xxstrdup(cmdline);
}

void rmonitor_info_to_rmsummary(struct rmsummary *tr, struct rmonitor_process_info *p, struct rmonitor_wdir_info *d, struct rmonitor_filesys_info *f, uint64_t start_time)
{
	tr->start = ((double) start_time) / ONE_SECOND;
	tr->end = ((double) usecs_since_epoch()) / ONE_SECOND;

	tr->wall_time = tr->end - tr->start;
	tr->cpu_time = ((double) p->cpu.accumulated) / ONE_SECOND;

	tr->cores = 0;
	tr->cores_avg = 0;

	/* set both cores and cores_avg to avg, as info does not come from time windows. */
	if(tr->wall_time > 0) {
		// set cores = cpu_time/wall_time;
		tr->cores = DIV_INT_ROUND_UP(tr->cpu_time, tr->wall_time);
		tr->cores_avg = tr->cores;
	}

	tr->context_switches = p->ctx.accumulated;
	tr->max_concurrent_processes = -1;
	tr->total_processes = -1;

	tr->virtual_memory = p->mem.virtual;
	tr->memory =         p->mem.resident;
	tr->swap_memory =    p->mem.virtual;

	// assigning values read/written in MB
	tr->bytes_read = ((double) (p->io.chars_read + p->io.bytes_faulted)) / ONE_MEGABYTE;
	tr->bytes_written = ((double) p->io.chars_written) / ONE_MEGABYTE;

	tr->machine_load = p->load.last_minute;
	tr->machine_cpus = p->load.cpus;

	if(d) {
		tr->total_files = d->files;
		tr->disk = ((double) d->byte_count) / ONE_MEGABYTE;
	} else {
		tr->total_files = -1;
		tr->disk = -1;
	}

	if(f) {
		tr->fs_nodes = f->disk.f_ffree;
	} else {
		tr->fs_nodes = -1;
	}

	tr->machine_load = p->load.last_minute;
	tr->machine_cpus = p->load.cpus;
}

struct rmsummary *rmonitor_measure_process(pid_t pid) {
	int err;

	struct rmsummary *tr = rmsummary_create(-1);

	struct rmonitor_process_info p;
	p.pid = pid;

	err = rmonitor_poll_process_once(&p);
	if(err != 0)
		return NULL;

	char cwd_link[PATH_MAX];
	char cwd_org[PATH_MAX];

	struct rmonitor_wdir_info *d = NULL;
	snprintf(cwd_link, PATH_MAX, "/proc/%d/cwd", pid);
	ssize_t n = readlink(cwd_link, cwd_org, PATH_MAX - 1);

	if(n != -1)  {
		cwd_org[n] = '\0';
		d = malloc(sizeof(struct rmonitor_wdir_info));
		d->path  = cwd_org;
		d->state = NULL;

		rmonitor_poll_wd_once(d, -1);
	}

	uint64_t start;
	err = rmonitor_get_start_time(pid, &start);
	if(err != 0)
		return NULL;

	rmonitor_info_to_rmsummary(tr, &p, d, NULL, start);
	tr->command = rmonitor_get_command_line(pid);

	if(d) {
		path_disk_size_info_delete_state(d->state);
		free(d);
	}

	return tr;
}

int rmonitor_measure_process_update_to_peak(struct rmsummary *tr, pid_t pid) {

	struct rmsummary *now = rmonitor_measure_process(pid);

	if(!now)
		return 0;

	rmsummary_merge_max(tr, now);

	rmsummary_delete(now);

	return 1;
}

struct rmsummary *rmonitor_measure_host(char *path) {
	uint64_t free_mem;
	uint64_t total_mem;
	int64_t total_disk;
	int64_t file_count;
	struct rmsummary *tr = rmsummary_create(-1);

	if(path) {
		path_disk_size_info_get(path, &total_disk, &file_count);
		tr->disk = ((double) total_disk) / ONE_MEGABYTE;
		tr->total_files = file_count;
	}

	host_memory_info_get(&free_mem, &total_mem);
	tr->memory = ((double) total_mem) / ONE_MEGABYTE;
	tr->cores = load_average_get_cpus();
	rmsummary_read_env_vars(tr);

	return tr;
}

struct rmsummary *rmonitor_collate_minimonitor(uint64_t start_time, int current_ps, int total_processes, struct rmonitor_process_info *p, struct rmonitor_mem_info *m, struct rmonitor_wdir_info *d) {
	struct rmsummary *tr = rmsummary_create(-1);

	tr->start = ((double) start_time) / ONE_SECOND;
	tr->end = ((double) usecs_since_epoch()) / ONE_SECOND;

	tr->wall_time = tr->end - tr->start;
	tr->cpu_time = ((double) p->cpu.accumulated) / ONE_SECOND;

	tr->cores = 0;

	if(tr->wall_time > 0) {
		// set cores = cpu_time/wall_time
		tr->cores = DIV_INT_ROUND_UP(tr->cpu_time, tr->wall_time);
	}

	tr->context_switches = p->ctx.accumulated;
	tr->max_concurrent_processes = current_ps;
	tr->total_processes = total_processes;

	/* we use max here, as /proc/pid/smaps that fills *m is not always
	 * available. This causes /proc/pid/status to become a conservative
	 * fallback. */
	if(m->resident > 0) {
		tr->virtual_memory = m->virtual;
		tr->memory =         m->resident;
		tr->swap_memory =    m->virtual;
	}
	else {
		tr->virtual_memory = p->mem.virtual;
		tr->memory =         p->mem.resident;
		tr->swap_memory =    p->mem.virtual;
	}

	// assigning values read/written in MB
	tr->bytes_read = ((double) (p->io.chars_read + p->io.bytes_faulted)) / ONE_MEGABYTE;
	tr->bytes_written = ((double) p->io.chars_written) / ONE_MEGABYTE;

	// set in resource_monitor.c from messages of the helper
	// tr->bytes_received = total_bytes_rx;
	// tr->bytes_sent     = total_bytes_tx;
	// tr->bandwidth = average_bandwidth(1);

	tr->total_files = d->files;
	tr->disk = ((double) d->byte_count) / ONE_MEGABYTE;

	tr->machine_load = p->load.last_minute;
	tr->machine_cpus = p->load.cpus;

	return tr;
}

struct rmsummary *rmonitor_minimonitor(minimonitor_op op, uint64_t pid) {
	static struct itable                 *processes = NULL;
	static struct rmonitor_process_info *p_acc      = NULL;
	static struct rmonitor_mem_info     *m_acc      = NULL;
	static struct rmonitor_wdir_info    *d_acc      = NULL;

	static uint64_t first_pid  = 0;
	static uint64_t start_time = 0;
	static uint64_t total_processes = 0;

	struct rmsummary *result = NULL;

	if(!processes) {
		processes = itable_create(0);
		p_acc     = calloc(1, sizeof(*p_acc)); //Automatic zeroed.
		m_acc     = calloc(1, sizeof(*m_acc));
		d_acc     = calloc(1, sizeof(*d_acc));
	}

	char cwd_link[PATH_MAX];
	char cwd_org[PATH_MAX];

	struct rmonitor_process_info *p;
	switch(op) {
		case MINIMONITOR_RESET:
			if(processes) {
				itable_firstkey(processes);
				while(itable_nextkey(processes, &pid, (void **) &p)) {
					itable_remove(processes, pid);
					free(p);
				}
				first_pid = 0;
				total_processes = 0;
				memset(p_acc, 0, sizeof(*p_acc));
				memset(m_acc, 0, sizeof(*m_acc));
				path_disk_size_info_delete_state(d_acc->state);
			}
			break;
		case MINIMONITOR_ADD_PID:
			p = itable_lookup(processes, pid);
			if(!p) {
				p = calloc(1, sizeof(struct rmonitor_process_info));
				p->pid = pid;
				itable_insert(processes, p->pid, (void *) p);
				total_processes++;
				if(first_pid == 0) {
					first_pid = pid;
					if(start_time < 1) {
						rmonitor_get_start_time(pid, &start_time);
					}

					snprintf(cwd_link, PATH_MAX, "/proc/%" PRIu64 "/cwd", pid);
					size_t n = readlink(cwd_link, cwd_org, PATH_MAX - 1);
					if(n > 0)  {
						cwd_org[n] = '\0';
						d_acc->path  = cwd_org;
						d_acc->state = NULL;
					}
				}
			}
			break;
		case MINIMONITOR_REMOVE_PID:
			p = itable_lookup(processes, pid);
			if(p) {
				itable_remove(processes, pid);
				free(p);
				if(pid == first_pid) {
					first_pid = 0;
				}
			}
			break;
		case MINIMONITOR_MEASURE:
			if(itable_size(processes) > 0) {
				rmonitor_poll_all_processes_once(processes, p_acc);
				rmonitor_poll_maps_once(processes, m_acc);
				rmonitor_poll_wd_once(d_acc, 1);

				result = rmonitor_collate_minimonitor(start_time, itable_size(processes), total_processes, p_acc, m_acc, d_acc);
			}
			break;
	}

	return result;
}

/* vim: set noexpandtab tabstop=4: */