sigar/src/os/darwin/darwin_sigar.c
2009-07-01 14:29:20 -07:00

3440 lines
88 KiB
C

/*
* Copyright (C) [2004, 2005, 2006], Hyperic, Inc.
* This file is part of SIGAR.
*
* SIGAR is free software; you can redistribute it and/or modify
* it under the terms version 2 of the GNU General Public License as
* published by the Free Software Foundation. This program is distributed
* in the hope that it will be useful, but WITHOUT ANY WARRANTY; without
* even the implied warranty of MERCHANTABILITY or FITNESS FOR A
* PARTICULAR PURPOSE. See the GNU General Public License for more
* details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
* USA.
*/
#include "sigar.h"
#include "sigar_private.h"
#include "sigar_util.h"
#include "sigar_os.h"
#include <sys/param.h>
#include <sys/mount.h>
#include <nfs/rpcv2.h>
#include <nfs/nfsproto.h>
#ifdef DARWIN
#include <dlfcn.h>
#include <mach/mach_init.h>
#include <mach/message.h>
#include <mach/kern_return.h>
#include <mach/mach_host.h>
#include <mach/mach_traps.h>
#include <mach/mach_port.h>
#include <mach/task.h>
#include <mach/thread_act.h>
#include <mach/thread_info.h>
#include <mach/vm_map.h>
#include <mach/shared_memory_server.h>
#include <mach-o/dyld.h>
#define __OPENTRANSPORTPROVIDERS__
#include <Gestalt.h>
#include <CFString.h>
#include <CoreFoundation/CoreFoundation.h>
#include <IOKit/IOBSD.h>
#include <IOKit/IOKitLib.h>
#include <IOKit/IOTypes.h>
#include <IOKit/storage/IOBlockStorageDriver.h>
#else
#include <sys/dkstat.h>
#include <sys/types.h>
#include <sys/param.h>
#include <sys/user.h>
#include <sys/vmmeter.h>
#include <fcntl.h>
#include <stdio.h>
#endif
#if defined(__FreeBSD__) && (__FreeBSD_version >= 500013)
#define SIGAR_FREEBSD5_NFSSTAT
#include <nfsclient/nfs.h>
#include <nfsserver/nfs.h>
#else
#include <nfs/nfs.h>
#endif
#include <sys/ioctl.h>
#include <sys/mount.h>
#include <sys/resource.h>
#include <sys/stat.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <net/if.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <net/route.h>
#include <netinet/in.h>
#include <dirent.h>
#include <errno.h>
#include <sys/socketvar.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/in_pcb.h>
#include <netinet/tcp.h>
#include <netinet/tcp_timer.h>
#ifdef __NetBSD__
#include <netinet/ip_var.h>
#include <sys/lwp.h>
#include <sys/mount.h>
#define SRUN LSRUN
#define SSLEEP LSSLEEP
#define SDEAD LSDEAD
#define SONPROC LSONPROC
#define SSUSPENDED LSSUSPENDED
#include <sys/sched.h>
#endif
#include <netinet/tcp_var.h>
#include <netinet/tcp_fsm.h>
#define NMIB(mib) (sizeof(mib)/sizeof(mib[0]))
#ifdef __FreeBSD__
# if (__FreeBSD_version >= 500013)
# define SIGAR_FREEBSD5
# else
# define SIGAR_FREEBSD4
# endif
#endif
#if defined(SIGAR_FREEBSD5)
#define KI_FD ki_fd
#define KI_PID ki_pid
#define KI_PPID ki_ppid
#define KI_PRI ki_pri.pri_user
#define KI_NICE ki_nice
#define KI_COMM ki_comm
#define KI_STAT ki_stat
#define KI_UID ki_ruid
#define KI_GID ki_rgid
#define KI_EUID ki_svuid
#define KI_EGID ki_svgid
#define KI_SIZE ki_size
#define KI_RSS ki_rssize
#define KI_TSZ ki_tsize
#define KI_DSZ ki_dsize
#define KI_SSZ ki_ssize
#define KI_FLAG ki_flag
#define KI_START ki_start
#elif defined(DARWIN) || defined(SIGAR_FREEBSD4) || defined(__OpenBSD__) || defined(__NetBSD__)
#define KI_FD kp_proc.p_fd
#define KI_PID kp_proc.p_pid
#define KI_PPID kp_eproc.e_ppid
#define KI_PRI kp_proc.p_priority
#define KI_NICE kp_proc.p_nice
#define KI_COMM kp_proc.p_comm
#define KI_STAT kp_proc.p_stat
#define KI_UID kp_eproc.e_pcred.p_ruid
#define KI_GID kp_eproc.e_pcred.p_rgid
#define KI_EUID kp_eproc.e_pcred.p_svuid
#define KI_EGID kp_eproc.e_pcred.p_svgid
#define KI_SIZE XXX
#define KI_RSS kp_eproc.e_vm.vm_rssize
#define KI_TSZ kp_eproc.e_vm.vm_tsize
#define KI_DSZ kp_eproc.e_vm.vm_dsize
#define KI_SSZ kp_eproc.e_vm.vm_ssize
#define KI_FLAG kp_eproc.e_flag
#define KI_START kp_proc.p_starttime
#endif
#ifndef DARWIN
#define PROCFS_STATUS(status) \
((((status) != SIGAR_OK) && !sigar->proc_mounted) ? \
SIGAR_ENOTIMPL : status)
static int get_koffsets(sigar_t *sigar)
{
int i;
struct nlist klist[] = {
{ "_cp_time" },
{ "_cnt" },
#if defined(__OpenBSD__) || defined(__NetBSD__)
{ "_tcpstat" },
{ "_tcbtable" },
#endif
{ NULL }
};
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
kvm_nlist(sigar->kmem, klist);
for (i=0; i<KOFFSET_MAX; i++) {
sigar->koffsets[i] = klist[i].n_value;
}
return SIGAR_OK;
}
static int kread(sigar_t *sigar, void *data, int size, long offset)
{
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
if (kvm_read(sigar->kmem, offset, data, size) != size) {
return errno;
}
return SIGAR_OK;
}
#endif
int sigar_os_open(sigar_t **sigar)
{
int mib[2];
int ncpu;
size_t len;
struct timeval boottime;
#ifndef DARWIN
struct stat sb;
#endif
len = sizeof(ncpu);
mib[0] = CTL_HW;
mib[1] = HW_NCPU;
if (sysctl(mib, NMIB(mib), &ncpu, &len, NULL, 0) < 0) {
return errno;
}
len = sizeof(boottime);
mib[0] = CTL_KERN;
mib[1] = KERN_BOOTTIME;
if (sysctl(mib, NMIB(mib), &boottime, &len, NULL, 0) < 0) {
return errno;
}
*sigar = malloc(sizeof(**sigar));
#ifdef DARWIN
(*sigar)->mach_port = mach_host_self();
# ifdef DARWIN_HAS_LIBPROC_H
if (((*sigar)->libproc = dlopen("/usr/lib/libproc.dylib", 0))) {
(*sigar)->proc_pidinfo = dlsym((*sigar)->libproc, "proc_pidinfo");
(*sigar)->proc_pidfdinfo = dlsym((*sigar)->libproc, "proc_pidfdinfo");
}
# endif
#else
(*sigar)->kmem = kvm_open(NULL, NULL, NULL, O_RDONLY, NULL);
if (stat("/proc/curproc", &sb) < 0) {
(*sigar)->proc_mounted = 0;
}
else {
(*sigar)->proc_mounted = 1;
}
#endif
#ifndef DARWIN
get_koffsets(*sigar);
#endif
(*sigar)->ncpu = ncpu;
(*sigar)->lcpu = -1;
(*sigar)->argmax = 0;
(*sigar)->boot_time = boottime.tv_sec; /* XXX seems off a bit */
(*sigar)->pagesize = getpagesize();
#ifdef __FreeBSD__
(*sigar)->ticks = 100; /* sysconf(_SC_CLK_TCK) == 128 !? */
#else
(*sigar)->ticks = sysconf(_SC_CLK_TCK);
#endif
(*sigar)->last_pid = -1;
(*sigar)->pinfo = NULL;
return SIGAR_OK;
}
int sigar_os_close(sigar_t *sigar)
{
if (sigar->pinfo) {
free(sigar->pinfo);
}
#ifndef DARWIN
if (sigar->kmem) {
kvm_close(sigar->kmem);
}
#endif
free(sigar);
return SIGAR_OK;
}
char *sigar_os_error_string(sigar_t *sigar, int err)
{
switch (err) {
case SIGAR_EPERM_KMEM:
return "Failed to open /dev/kmem for reading";
case SIGAR_EPROC_NOENT:
return "/proc filesystem is not mounted";
default:
return NULL;
}
}
/* ARG_MAX in FreeBSD 6.0 == 262144, which blows up the stack */
#define SIGAR_ARG_MAX 65536
static size_t sigar_argmax_get(sigar_t *sigar)
{
#ifdef KERN_ARGMAX
int mib[] = { CTL_KERN, KERN_ARGMAX };
size_t size = sizeof(sigar->argmax);
if (sigar->argmax != 0) {
return sigar->argmax;
}
if (sysctl(mib, NMIB(mib), &sigar->argmax, &size, NULL, 0) == 0) {
return sigar->argmax;
}
#endif
return SIGAR_ARG_MAX;
}
#if defined(DARWIN)
static int sigar_vmstat(sigar_t *sigar, vm_statistics_data_t *vmstat)
{
kern_return_t status;
mach_msg_type_number_t count = sizeof(*vmstat) / sizeof(integer_t);
status = host_statistics(sigar->mach_port, HOST_VM_INFO,
(host_info_t)vmstat, &count);
if (status == KERN_SUCCESS) {
return SIGAR_OK;
}
else {
return errno;
}
}
#elif defined(__FreeBSD__)
static int sigar_vmstat(sigar_t *sigar, struct vmmeter *vmstat)
{
int status;
size_t size = sizeof(unsigned int);
status = kread(sigar, vmstat, sizeof(*vmstat),
sigar->koffsets[KOFFSET_VMMETER]);
if (status == SIGAR_OK) {
return SIGAR_OK;
}
SIGAR_ZERO(vmstat);
/* derived from src/usr.bin/vmstat/vmstat.c */
/* only collect the ones we actually use */
#define GET_VM_STATS(cat, name, used) \
if (used) sysctlbyname("vm.stats." #cat "." #name, &vmstat->name, &size, NULL, 0)
/* sys */
GET_VM_STATS(sys, v_swtch, 0);
GET_VM_STATS(sys, v_trap, 0);
GET_VM_STATS(sys, v_syscall, 0);
GET_VM_STATS(sys, v_intr, 0);
GET_VM_STATS(sys, v_soft, 0);
/* vm */
GET_VM_STATS(vm, v_vm_faults, 0);
GET_VM_STATS(vm, v_cow_faults, 0);
GET_VM_STATS(vm, v_cow_optim, 0);
GET_VM_STATS(vm, v_zfod, 0);
GET_VM_STATS(vm, v_ozfod, 0);
GET_VM_STATS(vm, v_swapin, 1);
GET_VM_STATS(vm, v_swapout, 1);
GET_VM_STATS(vm, v_swappgsin, 0);
GET_VM_STATS(vm, v_swappgsout, 0);
GET_VM_STATS(vm, v_vnodein, 1);
GET_VM_STATS(vm, v_vnodeout, 1);
GET_VM_STATS(vm, v_vnodepgsin, 0);
GET_VM_STATS(vm, v_vnodepgsout, 0);
GET_VM_STATS(vm, v_intrans, 0);
GET_VM_STATS(vm, v_reactivated, 0);
GET_VM_STATS(vm, v_pdwakeups, 0);
GET_VM_STATS(vm, v_pdpages, 0);
GET_VM_STATS(vm, v_dfree, 0);
GET_VM_STATS(vm, v_pfree, 0);
GET_VM_STATS(vm, v_tfree, 0);
GET_VM_STATS(vm, v_page_size, 0);
GET_VM_STATS(vm, v_page_count, 0);
GET_VM_STATS(vm, v_free_reserved, 0);
GET_VM_STATS(vm, v_free_target, 0);
GET_VM_STATS(vm, v_free_min, 0);
GET_VM_STATS(vm, v_free_count, 1);
GET_VM_STATS(vm, v_wire_count, 0);
GET_VM_STATS(vm, v_active_count, 0);
GET_VM_STATS(vm, v_inactive_target, 0);
GET_VM_STATS(vm, v_inactive_count, 0);
GET_VM_STATS(vm, v_cache_count, 0);
GET_VM_STATS(vm, v_cache_min, 0);
GET_VM_STATS(vm, v_cache_max, 0);
GET_VM_STATS(vm, v_pageout_free_min, 0);
GET_VM_STATS(vm, v_interrupt_free_min, 0);
GET_VM_STATS(vm, v_forks, 0);
GET_VM_STATS(vm, v_vforks, 0);
GET_VM_STATS(vm, v_rforks, 0);
GET_VM_STATS(vm, v_kthreads, 0);
GET_VM_STATS(vm, v_forkpages, 0);
GET_VM_STATS(vm, v_vforkpages, 0);
GET_VM_STATS(vm, v_rforkpages, 0);
GET_VM_STATS(vm, v_kthreadpages, 0);
#undef GET_VM_STATS
return SIGAR_OK;
}
#elif defined(__OpenBSD__) || defined(__NetBSD__)
static int sigar_vmstat(sigar_t *sigar, struct uvmexp *vmstat)
{
size_t size = sizeof(*vmstat);
int mib[] = { CTL_VM, VM_UVMEXP };
if (sysctl(mib, NMIB(mib), vmstat, &size, NULL, 0) < 0) {
return errno;
}
else {
return SIGAR_OK;
}
}
#endif
int sigar_mem_get(sigar_t *sigar, sigar_mem_t *mem)
{
#ifdef DARWIN
vm_statistics_data_t vmstat;
uint64_t mem_total;
#else
unsigned long mem_total;
#endif
#if defined(__FreeBSD__)
struct vmmeter vmstat;
#elif defined(__OpenBSD__) || defined(__NetBSD__)
struct uvmexp vmstat;
#endif
int mib[2];
size_t len;
int status;
mib[0] = CTL_HW;
mib[1] = HW_PAGESIZE;
len = sizeof(sigar->pagesize);
if (sysctl(mib, NMIB(mib), &sigar->pagesize, &len, NULL, 0) < 0) {
return errno;
}
#ifdef DARWIN
mib[1] = HW_MEMSIZE;
#else
mib[1] = HW_PHYSMEM;
#endif
len = sizeof(mem_total);
if (sysctl(mib, NMIB(mib), &mem_total, &len, NULL, 0) < 0) {
return errno;
}
mem->total = mem_total;
#if defined(DARWIN)
if ((status = sigar_vmstat(sigar, &vmstat)) != SIGAR_OK) {
return status;
}
mem->free = vmstat.free_count;
mem->free *= sigar->pagesize;
#elif defined(__FreeBSD__)
if ((status = sigar_vmstat(sigar, &vmstat)) == SIGAR_OK) {
mem->free = vmstat.v_free_count;
mem->free *= sigar->pagesize;
}
#elif defined(__OpenBSD__) || defined(__NetBSD__)
if ((status = sigar_vmstat(sigar, &vmstat)) != SIGAR_OK) {
return status;
}
mem->free = vmstat.free;
#endif
mem->used = mem->total - mem->free;
mem->actual_free = mem->free;
mem->actual_used = mem->used;
sigar_mem_calc_ram(sigar, mem);
return SIGAR_OK;
}
#define SWI_MAXMIB 3
#ifdef SIGAR_FREEBSD5
/* code in this function is based on FreeBSD 5.3 kvm_getswapinfo.c */
static int getswapinfo_sysctl(struct kvm_swap *swap_ary,
int swap_max)
{
int ti, ttl;
size_t mibi, len, size;
int soid[SWI_MAXMIB];
struct xswdev xsd;
struct kvm_swap tot;
int unswdev, dmmax;
/* XXX this can be optimized by using os_open */
size = sizeof(dmmax);
if (sysctlbyname("vm.dmmax", &dmmax, &size, NULL, 0) == -1) {
return errno;
}
mibi = SWI_MAXMIB - 1;
if (sysctlnametomib("vm.swap_info", soid, &mibi) == -1) {
return errno;
}
bzero(&tot, sizeof(tot));
for (unswdev = 0;; unswdev++) {
soid[mibi] = unswdev;
len = sizeof(xsd);
if (sysctl(soid, mibi + 1, &xsd, &len, NULL, 0) == -1) {
if (errno == ENOENT) {
break;
}
return errno;
}
#if 0
if (len != sizeof(xsd)) {
_kvm_err(kd, kd->program, "struct xswdev has unexpected "
"size; kernel and libkvm out of sync?");
return -1;
}
if (xsd.xsw_version != XSWDEV_VERSION) {
_kvm_err(kd, kd->program, "struct xswdev version "
"mismatch; kernel and libkvm out of sync?");
return -1;
}
#endif
ttl = xsd.xsw_nblks - dmmax;
if (unswdev < swap_max - 1) {
bzero(&swap_ary[unswdev], sizeof(swap_ary[unswdev]));
swap_ary[unswdev].ksw_total = ttl;
swap_ary[unswdev].ksw_used = xsd.xsw_used;
swap_ary[unswdev].ksw_flags = xsd.xsw_flags;
}
tot.ksw_total += ttl;
tot.ksw_used += xsd.xsw_used;
}
ti = unswdev;
if (ti >= swap_max) {
ti = swap_max - 1;
}
if (ti >= 0) {
swap_ary[ti] = tot;
}
return SIGAR_OK;
}
#else
#define getswapinfo_sysctl(swap_ary, swap_max) SIGAR_ENOTIMPL
#endif
#define SIGAR_FS_BLOCKS_TO_BYTES(val, bsize) ((val * bsize) >> 1)
#ifdef DARWIN
#define VM_DIR "/private/var/vm"
#define SWAPFILE "swapfile"
static int sigar_swap_fs_get(sigar_t *sigar, sigar_swap_t *swap) /* <= 10.3 */
{
DIR *dirp;
struct dirent *ent;
char swapfile[SSTRLEN(VM_DIR) + SSTRLEN("/") + SSTRLEN(SWAPFILE) + 12];
struct stat swapstat;
struct statfs vmfs;
sigar_uint64_t val, bsize;
swap->used = swap->total = swap->free = 0;
if (!(dirp = opendir(VM_DIR))) {
return errno;
}
/* looking for "swapfile0", "swapfile1", etc. */
while ((ent = readdir(dirp))) {
char *ptr = swapfile;
if ((ent->d_namlen < SSTRLEN(SWAPFILE)+1) || /* n/a, see comment above */
(ent->d_namlen > SSTRLEN(SWAPFILE)+11)) /* ensure no overflow */
{
continue;
}
if (!strnEQ(ent->d_name, SWAPFILE, SSTRLEN(SWAPFILE))) {
continue;
}
/* sprintf(swapfile, "%s/%s", VM_DIR, ent->d_name) */
memcpy(ptr, VM_DIR, SSTRLEN(VM_DIR));
ptr += SSTRLEN(VM_DIR);
*ptr++ = '/';
memcpy(ptr, ent->d_name, ent->d_namlen+1);
if (stat(swapfile, &swapstat) < 0) {
continue;
}
swap->used += swapstat.st_size;
}
closedir(dirp);
if (statfs(VM_DIR, &vmfs) < 0) {
return errno;
}
bsize = vmfs.f_bsize / 512;
val = vmfs.f_bfree;
swap->total = SIGAR_FS_BLOCKS_TO_BYTES(val, bsize) + swap->used;
swap->free = swap->total - swap->used;
return SIGAR_OK;
}
static int sigar_swap_sysctl_get(sigar_t *sigar, sigar_swap_t *swap)
{
#ifdef VM_SWAPUSAGE /* => 10.4 */
struct xsw_usage sw_usage;
size_t size = sizeof(sw_usage);
int mib[] = { CTL_VM, VM_SWAPUSAGE };
if (sysctl(mib, NMIB(mib), &sw_usage, &size, NULL, 0) != 0) {
return errno;
}
swap->total = sw_usage.xsu_total;
swap->used = sw_usage.xsu_used;
swap->free = sw_usage.xsu_avail;
return SIGAR_OK;
#else
return SIGAR_ENOTIMPL; /* <= 10.3 */
#endif
}
#endif /* DARWIN */
int sigar_swap_get(sigar_t *sigar, sigar_swap_t *swap)
{
int status;
#if defined(DARWIN)
vm_statistics_data_t vmstat;
if (sigar_swap_sysctl_get(sigar, swap) != SIGAR_OK) {
status = sigar_swap_fs_get(sigar, swap); /* <= 10.3 */
if (status != SIGAR_OK) {
return status;
}
}
if ((status = sigar_vmstat(sigar, &vmstat)) != SIGAR_OK) {
return status;
}
swap->page_in = vmstat.pageins;
swap->page_out = vmstat.pageouts;
#elif defined(__FreeBSD__)
struct kvm_swap kswap[1];
struct vmmeter vmstat;
if (getswapinfo_sysctl(kswap, 1) != SIGAR_OK) {
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
if (kvm_getswapinfo(sigar->kmem, kswap, 1, 0) < 0) {
return errno;
}
}
if (kswap[0].ksw_total == 0) {
swap->total = 0;
swap->used = 0;
swap->free = 0;
return SIGAR_OK;
}
swap->total = kswap[0].ksw_total * sigar->pagesize;
swap->used = kswap[0].ksw_used * sigar->pagesize;
swap->free = swap->total - swap->used;
if ((status = sigar_vmstat(sigar, &vmstat)) == SIGAR_OK) {
swap->page_in = vmstat.v_swapin + vmstat.v_vnodein;
swap->page_out = vmstat.v_swapout + vmstat.v_vnodeout;
}
else {
swap->page_in = swap->page_out = -1;
}
#elif defined(__OpenBSD__) || defined(__NetBSD__)
struct uvmexp vmstat;
if ((status = sigar_vmstat(sigar, &vmstat)) != SIGAR_OK) {
return status;
}
swap->total = vmstat.swpages * sigar->pagesize;
swap->used = vmstat.swpginuse * sigar->pagesize;
swap->free = swap->total - swap->used;
swap->page_in = vmstat.pageins;
swap->page_out = vmstat.pdpageouts;
#endif
return SIGAR_OK;
}
#ifndef KERN_CPTIME
#define KERN_CPTIME KERN_CP_TIME
#endif
#if defined(__NetBSD__)
typedef uint64_t cp_time_t;
#else
typedef unsigned long cp_time_t;
#endif
int sigar_cpu_get(sigar_t *sigar, sigar_cpu_t *cpu)
{
#if defined(DARWIN)
kern_return_t status;
mach_msg_type_number_t count = HOST_CPU_LOAD_INFO_COUNT;
host_cpu_load_info_data_t cpuload;
status = host_statistics(sigar->mach_port, HOST_CPU_LOAD_INFO,
(host_info_t)&cpuload, &count);
if (status != KERN_SUCCESS) {
return errno;
}
cpu->user = SIGAR_TICK2MSEC(cpuload.cpu_ticks[CPU_STATE_USER]);
cpu->sys = SIGAR_TICK2MSEC(cpuload.cpu_ticks[CPU_STATE_SYSTEM]);
cpu->idle = SIGAR_TICK2MSEC(cpuload.cpu_ticks[CPU_STATE_IDLE]);
cpu->nice = SIGAR_TICK2MSEC(cpuload.cpu_ticks[CPU_STATE_NICE]);
cpu->wait = 0; /*N/A*/
cpu->irq = 0; /*N/A*/
cpu->soft_irq = 0; /*N/A*/
cpu->stolen = 0; /*N/A*/
cpu->total = cpu->user + cpu->nice + cpu->sys + cpu->idle;
#elif defined(__FreeBSD__) || (__OpenBSD__) || defined(__NetBSD__)
int status;
cp_time_t cp_time[CPUSTATES];
size_t size = sizeof(cp_time);
# if defined(__OpenBSD__) || defined(__NetBSD__)
int mib[] = { CTL_KERN, KERN_CPTIME };
if (sysctl(mib, NMIB(mib), &cp_time, &size, NULL, 0) == -1) {
status = errno;
}
# else
/* try sysctl first, does not require /dev/kmem perms */
if (sysctlbyname("kern.cp_time", &cp_time, &size, NULL, 0) == -1) {
status = kread(sigar, &cp_time, sizeof(cp_time),
sigar->koffsets[KOFFSET_CPUINFO]);
}
# endif
else {
status = SIGAR_OK;
}
if (status != SIGAR_OK) {
return status;
}
cpu->user = SIGAR_TICK2MSEC(cp_time[CP_USER]);
cpu->nice = SIGAR_TICK2MSEC(cp_time[CP_NICE]);
cpu->sys = SIGAR_TICK2MSEC(cp_time[CP_SYS]);
cpu->idle = SIGAR_TICK2MSEC(cp_time[CP_IDLE]);
cpu->wait = 0; /*N/A*/
cpu->irq = SIGAR_TICK2MSEC(cp_time[CP_INTR]);
cpu->soft_irq = 0; /*N/A*/
cpu->stolen = 0; /*N/A*/
cpu->total = cpu->user + cpu->nice + cpu->sys + cpu->idle + cpu->irq;
#endif
return SIGAR_OK;
}
#if defined(__FreeBSD__) && (__FreeBSD_version >= 700000)
#define HAVE_KERN_CP_TIMES /* kern.cp_times came later than 7.0, not sure exactly when */
static int sigar_cp_times_get(sigar_t *sigar, sigar_cpu_list_t *cpulist)
{
int maxcpu, status;
size_t len = sizeof(maxcpu), size;
long *times;
if (sysctlbyname("kern.smp.maxcpus", &maxcpu, &len, NULL, 0) == -1) {
return errno;
}
size = sizeof(long) * maxcpu * CPUSTATES;
times = malloc(size);
if (sysctlbyname("kern.cp_times", times, &size, NULL, 0) == -1) {
status = errno;
}
else {
int i, maxid = (size / CPUSTATES / sizeof(long));
long *cp_time = times;
status = SIGAR_OK;
for (i=0; i<maxid; i++) {
sigar_cpu_t *cpu;
SIGAR_CPU_LIST_GROW(cpulist);
cpu = &cpulist->data[cpulist->number++];
cpu->user = SIGAR_TICK2MSEC(cp_time[CP_USER]);
cpu->nice = SIGAR_TICK2MSEC(cp_time[CP_NICE]);
cpu->sys = SIGAR_TICK2MSEC(cp_time[CP_SYS]);
cpu->idle = SIGAR_TICK2MSEC(cp_time[CP_IDLE]);
cpu->wait = 0; /*N/A*/
cpu->irq = SIGAR_TICK2MSEC(cp_time[CP_INTR]);
cpu->soft_irq = 0; /*N/A*/
cpu->stolen = 0; /*N/A*/
cpu->total = cpu->user + cpu->nice + cpu->sys + cpu->idle + cpu->irq;
cp_time += CPUSTATES;
}
}
free(times);
return status;
}
#endif
int sigar_cpu_list_get(sigar_t *sigar, sigar_cpu_list_t *cpulist)
{
#ifdef DARWIN
kern_return_t status;
mach_msg_type_number_t count;
processor_cpu_load_info_data_t *cpuload;
natural_t i, ncpu;
status = host_processor_info(sigar->mach_port,
PROCESSOR_CPU_LOAD_INFO,
&ncpu,
(processor_info_array_t*)&cpuload,
&count);
if (status != KERN_SUCCESS) {
return errno;
}
sigar_cpu_list_create(cpulist);
for (i=0; i<ncpu; i++) {
sigar_cpu_t *cpu;
SIGAR_CPU_LIST_GROW(cpulist);
cpu = &cpulist->data[cpulist->number++];
cpu->user = SIGAR_TICK2MSEC(cpuload[i].cpu_ticks[CPU_STATE_USER]);
cpu->sys = SIGAR_TICK2MSEC(cpuload[i].cpu_ticks[CPU_STATE_SYSTEM]);
cpu->idle = SIGAR_TICK2MSEC(cpuload[i].cpu_ticks[CPU_STATE_IDLE]);
cpu->nice = SIGAR_TICK2MSEC(cpuload[i].cpu_ticks[CPU_STATE_NICE]);
cpu->wait = 0; /*N/A*/
cpu->irq = 0; /*N/A*/
cpu->soft_irq = 0; /*N/A*/
cpu->stolen = 0; /*N/A*/
cpu->total = cpu->user + cpu->nice + cpu->sys + cpu->idle;
}
vm_deallocate(mach_task_self(), (vm_address_t)cpuload, count);
return SIGAR_OK;
#else
int status, i;
sigar_cpu_t *cpu;
sigar_cpu_list_create(cpulist);
#ifdef HAVE_KERN_CP_TIMES
if ((status = sigar_cp_times_get(sigar, cpulist)) == SIGAR_OK) {
return SIGAR_OK;
}
#endif
/* XXX no multi cpu in freebsd < 7.0, howbout others?
* for now just report all metrics on the 1st cpu
* 0's for the rest
*/
cpu = &cpulist->data[cpulist->number++];
status = sigar_cpu_get(sigar, cpu);
if (status != SIGAR_OK) {
return status;
}
for (i=1; i<sigar->ncpu; i++) {
SIGAR_CPU_LIST_GROW(cpulist);
cpu = &cpulist->data[cpulist->number++];
SIGAR_ZERO(cpu);
}
return SIGAR_OK;
#endif
}
int sigar_uptime_get(sigar_t *sigar,
sigar_uptime_t *uptime)
{
uptime->uptime = time(NULL) - sigar->boot_time;
return SIGAR_OK;
}
int sigar_loadavg_get(sigar_t *sigar,
sigar_loadavg_t *loadavg)
{
getloadavg(loadavg->loadavg, 3);
return SIGAR_OK;
}
#if defined(DARWIN) && defined(DARWIN_HAS_LIBPROC_H)
static int proc_fdinfo_get(sigar_t *sigar, sigar_pid_t pid, int *num)
{
int rsize;
const int init_size = PROC_PIDLISTFD_SIZE * 32;
if (!sigar->libproc) {
return SIGAR_ENOTIMPL;
}
if (sigar->ifconf_len == 0) {
sigar->ifconf_len = init_size;
sigar->ifconf_buf = malloc(sigar->ifconf_len);
}
while (1) {
rsize = sigar->proc_pidinfo(pid, PROC_PIDLISTFDS, 0,
sigar->ifconf_buf, sigar->ifconf_len);
if (rsize <= 0) {
return errno;
}
if ((rsize + PROC_PIDLISTFD_SIZE) < sigar->ifconf_len) {
break;
}
sigar->ifconf_len += init_size;
sigar->ifconf_buf = realloc(sigar->ifconf_buf, sigar->ifconf_len);
}
*num = rsize / PROC_PIDLISTFD_SIZE;
return SIGAR_OK;
}
#endif
#ifndef KERN_PROC_PROC
/* freebsd 4.x */
#define KERN_PROC_PROC KERN_PROC_ALL
#endif
int sigar_os_proc_list_get(sigar_t *sigar,
sigar_proc_list_t *proclist)
{
#if defined(DARWIN) || defined(SIGAR_FREEBSD5) || defined(__OpenBSD__) || defined(__NetBSD__)
int mib[4] = { CTL_KERN, KERN_PROC, KERN_PROC_PROC, 0 };
int i, num;
size_t len;
struct kinfo_proc *proc;
if (sysctl(mib, NMIB(mib), NULL, &len, NULL, 0) < 0) {
return errno;
}
proc = malloc(len);
if (sysctl(mib, NMIB(mib), proc, &len, NULL, 0) < 0) {
free(proc);
return errno;
}
num = len/sizeof(*proc);
for (i=0; i<num; i++) {
if (proc[i].KI_FLAG & P_SYSTEM) {
continue;
}
if (proc[i].KI_PID == 0) {
continue;
}
SIGAR_PROC_LIST_GROW(proclist);
proclist->data[proclist->number++] = proc[i].KI_PID;
}
free(proc);
return SIGAR_OK;
#else
int i, num;
struct kinfo_proc *proc;
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
proc = kvm_getprocs(sigar->kmem, KERN_PROC_PROC, 0, &num);
for (i=0; i<num; i++) {
if (proc[i].KI_FLAG & P_SYSTEM) {
continue;
}
SIGAR_PROC_LIST_GROW(proclist);
proclist->data[proclist->number++] = proc[i].KI_PID;
}
#endif
return SIGAR_OK;
}
static int sigar_get_pinfo(sigar_t *sigar, sigar_pid_t pid)
{
#if defined(__OpenBSD__) || defined(__NetBSD__)
int mib[] = { CTL_KERN, KERN_PROC2, KERN_PROC_PID, 0, sizeof(*sigar->pinfo), 1 };
#else
int mib[] = { CTL_KERN, KERN_PROC, KERN_PROC_PID, 0 };
#endif
size_t len = sizeof(*sigar->pinfo);
time_t timenow = time(NULL);
mib[3] = pid;
if (sigar->pinfo == NULL) {
sigar->pinfo = malloc(len);
}
if (sigar->last_pid == pid) {
if ((timenow - sigar->last_getprocs) < SIGAR_LAST_PROC_EXPIRE) {
return SIGAR_OK;
}
}
sigar->last_pid = pid;
sigar->last_getprocs = timenow;
if (sysctl(mib, NMIB(mib), sigar->pinfo, &len, NULL, 0) < 0) {
return errno;
}
return SIGAR_OK;
}
#ifdef DARWIN
#define GLOBAL_SHARED_SIZE (SHARED_TEXT_REGION_SIZE + SHARED_DATA_REGION_SIZE)
#endif
int sigar_proc_mem_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_mem_t *procmem)
{
#if defined(DARWIN)
mach_port_t task, self = mach_task_self();
kern_return_t status;
task_basic_info_data_t info;
task_events_info_data_t events;
mach_msg_type_number_t count;
# ifdef DARWIN_HAS_LIBPROC_H
struct proc_taskinfo pti;
struct proc_regioninfo pri;
if (sigar->libproc) {
int sz =
sigar->proc_pidinfo(pid, PROC_PIDTASKINFO, 0, &pti, sizeof(pti));
if (sz == sizeof(pti)) {
procmem->size = pti.pti_virtual_size;
procmem->resident = pti.pti_resident_size;
procmem->page_faults = pti.pti_faults;
procmem->minor_faults = SIGAR_FIELD_NOTIMPL;
procmem->major_faults = SIGAR_FIELD_NOTIMPL;
procmem->share = SIGAR_FIELD_NOTIMPL;
sz = sigar->proc_pidinfo(pid, PROC_PIDREGIONINFO, 0, &pri, sizeof(pri));
if (sz == sizeof(pri)) {
if ((pri.pri_share_mode == SM_EMPTY) &&
(procmem->size > GLOBAL_SHARED_SIZE))
{
procmem->size -= GLOBAL_SHARED_SIZE; /* SIGAR-123 */
}
}
return SIGAR_OK;
}
}
# endif
status = task_for_pid(self, pid, &task);
if (status != KERN_SUCCESS) {
return errno;
}
count = TASK_BASIC_INFO_COUNT;
status = task_info(task, TASK_BASIC_INFO, (task_info_t)&info, &count);
if (status != KERN_SUCCESS) {
return errno;
}
count = TASK_EVENTS_INFO_COUNT;
status = task_info(task, TASK_EVENTS_INFO, (task_info_t)&events, &count);
if (status == KERN_SUCCESS) {
procmem->page_faults = events.faults;
}
else {
procmem->page_faults = SIGAR_FIELD_NOTIMPL;
}
procmem->minor_faults = SIGAR_FIELD_NOTIMPL;
procmem->major_faults = SIGAR_FIELD_NOTIMPL;
if (task != self) {
mach_port_deallocate(self, task);
}
procmem->size = info.virtual_size;
procmem->resident = info.resident_size;
procmem->share = SIGAR_FIELD_NOTIMPL;
return SIGAR_OK;
#elif defined(__FreeBSD__)
int status = sigar_get_pinfo(sigar, pid);
bsd_pinfo_t *pinfo = sigar->pinfo;
if (status != SIGAR_OK) {
return status;
}
procmem->size =
(pinfo->KI_TSZ + pinfo->KI_DSZ + pinfo->KI_SSZ) * sigar->pagesize;
procmem->resident = pinfo->KI_RSS * sigar->pagesize;
procmem->share = SIGAR_FIELD_NOTIMPL;
procmem->page_faults = SIGAR_FIELD_NOTIMPL;
procmem->minor_faults = SIGAR_FIELD_NOTIMPL;
procmem->major_faults = SIGAR_FIELD_NOTIMPL;
#elif defined(__OpenBSD__) || defined(__NetBSD__)
int status = sigar_get_pinfo(sigar, pid);
bsd_pinfo_t *pinfo = sigar->pinfo;
if (status != SIGAR_OK) {
return status;
}
procmem->size =
(pinfo->p_vm_tsize + pinfo->p_vm_dsize + pinfo->p_vm_ssize) * sigar->pagesize;
procmem->resident = pinfo->p_vm_rssize * sigar->pagesize;
procmem->share = SIGAR_FIELD_NOTIMPL;
procmem->minor_faults = pinfo->p_uru_minflt;
procmem->major_faults = pinfo->p_uru_majflt;
procmem->page_faults = procmem->minor_faults + procmem->major_faults;
#endif
return SIGAR_OK;
}
int sigar_proc_cred_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_cred_t *proccred)
{
int status = sigar_get_pinfo(sigar, pid);
bsd_pinfo_t *pinfo = sigar->pinfo;
if (status != SIGAR_OK) {
return status;
}
#if defined(__OpenBSD__) || defined(__NetBSD__)
proccred->uid = pinfo->p_ruid;
proccred->gid = pinfo->p_rgid;
proccred->euid = pinfo->p_uid;
proccred->egid = pinfo->p_gid;
#else
proccred->uid = pinfo->KI_UID;
proccred->gid = pinfo->KI_GID;
proccred->euid = pinfo->KI_EUID;
proccred->egid = pinfo->KI_EGID;
#endif
return SIGAR_OK;
}
#define tv2msec(tv) \
(((sigar_uint64_t)tv.tv_sec * SIGAR_MSEC) + (((sigar_uint64_t)tv.tv_usec) / 1000))
#ifdef DARWIN
#define tval2msec(tval) \
((tval.seconds * SIGAR_MSEC) + (tval.microseconds / 1000))
#define tval2nsec(tval) \
(SIGAR_SEC2NANO((tval).seconds) + SIGAR_MICROSEC2NANO((tval).microseconds))
static int get_proc_times(sigar_t *sigar, sigar_pid_t pid, sigar_proc_time_t *time)
{
unsigned int count;
time_value_t utime = {0, 0}, stime = {0, 0};
task_basic_info_data_t ti;
task_thread_times_info_data_t tti;
task_port_t task, self;
kern_return_t status;
# ifdef DARWIN_HAS_LIBPROC_H
if (sigar->libproc) {
struct proc_taskinfo pti;
int sz =
sigar->proc_pidinfo(pid, PROC_PIDTASKINFO, 0, &pti, sizeof(pti));
if (sz == sizeof(pti)) {
time->user = SIGAR_NSEC2MSEC(pti.pti_total_user);
time->sys = SIGAR_NSEC2MSEC(pti.pti_total_system);
time->total = time->user + time->sys;
return SIGAR_OK;
}
}
# endif
self = mach_task_self();
status = task_for_pid(self, pid, &task);
if (status != KERN_SUCCESS) {
return errno;
}
count = TASK_BASIC_INFO_COUNT;
status = task_info(task, TASK_BASIC_INFO,
(task_info_t)&ti, &count);
if (status != KERN_SUCCESS) {
if (task != self) {
mach_port_deallocate(self, task);
}
return errno;
}
count = TASK_THREAD_TIMES_INFO_COUNT;
status = task_info(task, TASK_THREAD_TIMES_INFO,
(task_info_t)&tti, &count);
if (status != KERN_SUCCESS) {
if (task != self) {
mach_port_deallocate(self, task);
}
return errno;
}
time_value_add(&utime, &ti.user_time);
time_value_add(&stime, &ti.system_time);
time_value_add(&utime, &tti.user_time);
time_value_add(&stime, &tti.system_time);
time->user = tval2msec(utime);
time->sys = tval2msec(stime);
time->total = time->user + time->sys;
return SIGAR_OK;
}
#endif
int sigar_proc_time_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_time_t *proctime)
{
#ifdef SIGAR_FREEBSD4
struct user user;
#endif
int status = sigar_get_pinfo(sigar, pid);
bsd_pinfo_t *pinfo = sigar->pinfo;
if (status != SIGAR_OK) {
return status;
}
#if defined(DARWIN)
if ((status = get_proc_times(sigar, pid, proctime)) != SIGAR_OK) {
return status;
}
proctime->start_time = tv2msec(pinfo->KI_START);
#elif defined(SIGAR_FREEBSD5)
proctime->user = tv2msec(pinfo->ki_rusage.ru_utime);
proctime->sys = tv2msec(pinfo->ki_rusage.ru_stime);
proctime->total = proctime->user + proctime->sys;
proctime->start_time = tv2msec(pinfo->KI_START);
#elif defined(SIGAR_FREEBSD4)
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
status = kread(sigar, &user, sizeof(user),
(u_long)pinfo->kp_proc.p_addr);
if (status != SIGAR_OK) {
return status;
}
proctime->user = tv2msec(user.u_stats.p_ru.ru_utime);
proctime->sys = tv2msec(user.u_stats.p_ru.ru_stime);
proctime->total = proctime->user + proctime->sys;
proctime->start_time = tv2msec(user.u_stats.p_start);
#elif defined(__OpenBSD__) || defined(__NetBSD__)
/* XXX *_usec */
proctime->user = pinfo->p_uutime_sec * SIGAR_MSEC;
proctime->sys = pinfo->p_ustime_sec * SIGAR_MSEC;
proctime->total = proctime->user + proctime->sys;
proctime->start_time = pinfo->p_ustart_sec * SIGAR_MSEC;
#endif
return SIGAR_OK;
}
#ifdef DARWIN
/* thread state mapping derived from ps.tproj */
static const char const thread_states[] = {
/*0*/ '-',
/*1*/ SIGAR_PROC_STATE_RUN,
/*2*/ SIGAR_PROC_STATE_ZOMBIE,
/*3*/ SIGAR_PROC_STATE_SLEEP,
/*4*/ SIGAR_PROC_STATE_IDLE,
/*5*/ SIGAR_PROC_STATE_STOP,
/*6*/ SIGAR_PROC_STATE_STOP,
/*7*/ '?'
};
static int thread_state_get(thread_basic_info_data_t *info)
{
switch (info->run_state) {
case TH_STATE_RUNNING:
return 1;
case TH_STATE_UNINTERRUPTIBLE:
return 2;
case TH_STATE_WAITING:
return (info->sleep_time > 20) ? 4 : 3;
case TH_STATE_STOPPED:
return 5;
case TH_STATE_HALTED:
return 6;
default:
return 7;
}
}
static int sigar_proc_threads_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_state_t *procstate)
{
mach_port_t task, self = mach_task_self();
kern_return_t status;
thread_array_t threads;
mach_msg_type_number_t count, i;
int state = TH_STATE_HALTED + 1;
status = task_for_pid(self, pid, &task);
if (status != KERN_SUCCESS) {
return errno;
}
status = task_threads(task, &threads, &count);
if (status != KERN_SUCCESS) {
return errno;
}
procstate->threads = count;
for (i=0; i<count; i++) {
mach_msg_type_number_t info_count = THREAD_BASIC_INFO_COUNT;
thread_basic_info_data_t info;
status = thread_info(threads[i], THREAD_BASIC_INFO,
(thread_info_t)&info, &info_count);
if (status == KERN_SUCCESS) {
int tstate = thread_state_get(&info);
if (tstate < state) {
state = tstate;
}
}
}
vm_deallocate(self, (vm_address_t)threads, sizeof(thread_t) * count);
procstate->state = thread_states[state];
return SIGAR_OK;
}
#endif
int sigar_proc_state_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_state_t *procstate)
{
int status = sigar_get_pinfo(sigar, pid);
bsd_pinfo_t *pinfo = sigar->pinfo;
#if defined(__OpenBSD__) || defined(__NetBSD__)
int state = pinfo->p_stat;
#else
int state = pinfo->KI_STAT;
#endif
if (status != SIGAR_OK) {
return status;
}
#if defined(__OpenBSD__) || defined(__NetBSD__)
SIGAR_SSTRCPY(procstate->name, pinfo->p_comm);
procstate->ppid = pinfo->p_ppid;
procstate->priority = pinfo->p_priority;
procstate->nice = pinfo->p_nice;
procstate->tty = pinfo->p_tdev;
procstate->threads = SIGAR_FIELD_NOTIMPL;
procstate->processor = pinfo->p_cpuid;
#else
SIGAR_SSTRCPY(procstate->name, pinfo->KI_COMM);
procstate->ppid = pinfo->KI_PPID;
procstate->priority = pinfo->KI_PRI;
procstate->nice = pinfo->KI_NICE;
procstate->tty = SIGAR_FIELD_NOTIMPL; /*XXX*/
procstate->threads = SIGAR_FIELD_NOTIMPL;
procstate->processor = SIGAR_FIELD_NOTIMPL;
#endif
#ifdef DARWIN
status = sigar_proc_threads_get(sigar, pid, procstate);
if (status == SIGAR_OK) {
return status;
}
#endif
switch (state) {
case SIDL:
procstate->state = 'D';
break;
case SRUN:
#ifdef SONPROC
case SONPROC:
#endif
procstate->state = 'R';
break;
case SSLEEP:
procstate->state = 'S';
break;
case SSTOP:
procstate->state = 'T';
break;
case SZOMB:
procstate->state = 'Z';
break;
default:
procstate->state = '?';
break;
}
return SIGAR_OK;
}
#if defined(DARWIN)
typedef struct {
char *buf, *ptr, *end;
int count;
} sigar_kern_proc_args_t;
static void sigar_kern_proc_args_destroy(sigar_kern_proc_args_t *kargs)
{
if (kargs->buf) {
free(kargs->buf);
kargs->buf = NULL;
}
}
/* re-usable hack for use by proc_args and proc_env */
static int sigar_kern_proc_args_get(sigar_t *sigar,
sigar_pid_t pid,
char *exe,
sigar_kern_proc_args_t *kargs)
{
/*
* derived from:
* http://darwinsource.opendarwin.org/10.4.1/adv_cmds-79.1/ps.tproj/print.c
*/
int mib[3], len;
size_t size = sigar_argmax_get(sigar);
kargs->buf = malloc(size);
mib[0] = CTL_KERN;
mib[1] = KERN_PROCARGS2;
mib[2] = pid;
if (sysctl(mib, NMIB(mib), kargs->buf, &size, NULL, 0) < 0) {
sigar_kern_proc_args_destroy(kargs);
return errno;
}
kargs->end = &kargs->buf[size];
memcpy(&kargs->count, kargs->buf, sizeof(kargs->count));
kargs->ptr = kargs->buf + sizeof(kargs->count);
len = strlen(kargs->ptr);
if (exe) {
memcpy(exe, kargs->ptr, len+1);
}
kargs->ptr += len+1;
if (kargs->ptr == kargs->end) {
sigar_kern_proc_args_destroy(kargs);
return exe ? SIGAR_OK : ENOENT;
}
for (; kargs->ptr < kargs->end; kargs->ptr++) {
if (*kargs->ptr != '\0') {
break; /* start of argv[0] */
}
}
if (kargs->ptr == kargs->end) {
sigar_kern_proc_args_destroy(kargs);
return exe ? SIGAR_OK : ENOENT;
}
return SIGAR_OK;
}
static int kern_proc_args_skip_argv(sigar_kern_proc_args_t *kargs)
{
char *ptr = kargs->ptr;
char *end = kargs->end;
int count = kargs->count;
/* skip over argv */
while ((ptr < end) && (count-- > 0)) {
int alen = strlen(ptr)+1;
ptr += alen;
}
kargs->ptr = ptr;
kargs->end = end;
kargs->count = 0;
if (ptr >= end) {
return ENOENT;
}
return SIGAR_OK;
}
#endif
int sigar_os_proc_args_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_args_t *procargs)
{
#if defined(DARWIN)
int status, count;
sigar_kern_proc_args_t kargs;
char *ptr, *end;
status = sigar_kern_proc_args_get(sigar, pid, NULL, &kargs);
if (status != SIGAR_OK) {
return status;
}
count = kargs.count;
ptr = kargs.ptr;
end = kargs.end;
while ((ptr < end) && (count-- > 0)) {
int slen = strlen(ptr);
int alen = slen+1;
char *arg;
/*
* trim trailing whitespace.
* seen w/ postgresql, probably related
* to messing with argv[0]
*/
while (*(ptr + (slen-1)) == ' ') {
if (--slen <= 0) {
break;
}
}
arg = malloc(slen+1);
SIGAR_PROC_ARGS_GROW(procargs);
memcpy(arg, ptr, slen);
*(arg+slen) = '\0';
procargs->data[procargs->number++] = arg;
ptr += alen;
}
sigar_kern_proc_args_destroy(&kargs);
return SIGAR_OK;
#elif defined(__FreeBSD__) || defined(__NetBSD__)
char buffer[SIGAR_ARG_MAX+1], *ptr=buffer;
size_t len = sizeof(buffer);
# ifdef __NetBSD__
int mib[] = { CTL_KERN, KERN_PROC_ARGS, 0, KERN_PROC_ARGV };
mib[2] = pid;
# else
int mib[] = { CTL_KERN, KERN_PROC, KERN_PROC_ARGS, 0 };
mib[3] = pid;
# endif
if (sysctl(mib, NMIB(mib), buffer, &len, NULL, 0) < 0) {
return errno;
}
if (len == 0) {
procargs->number = 0;
return SIGAR_OK;
}
buffer[len] = '\0';
while (len > 0) {
int alen = strlen(ptr)+1;
char *arg = malloc(alen);
SIGAR_PROC_ARGS_GROW(procargs);
memcpy(arg, ptr, alen);
procargs->data[procargs->number++] = arg;
len -= alen;
if (len > 0) {
ptr += alen;
}
}
return SIGAR_OK;
#elif defined(__OpenBSD__)
char buffer[SIGAR_ARG_MAX+1], **ptr=(char **)buffer;
size_t len = sizeof(buffer);
int mib[] = { CTL_KERN, KERN_PROC_ARGS, 0, KERN_PROC_ARGV };
mib[2] = pid;
if (sysctl(mib, NMIB(mib), buffer, &len, NULL, 0) < 0) {
return errno;
}
if (len == 0) {
procargs->number = 0;
return SIGAR_OK;
}
for (; *ptr; ptr++) {
int alen = strlen(*ptr)+1;
char *arg = malloc(alen);
SIGAR_PROC_ARGS_GROW(procargs);
memcpy(arg, *ptr, alen);
procargs->data[procargs->number++] = arg;
}
return SIGAR_OK;
#else
return SIGAR_ENOTIMPL;
#endif
}
int sigar_proc_env_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_env_t *procenv)
{
#ifdef DARWIN
int status, count;
sigar_kern_proc_args_t kargs;
char *ptr, *end;
status = sigar_kern_proc_args_get(sigar, pid, NULL, &kargs);
if (status != SIGAR_OK) {
return status;
}
status = kern_proc_args_skip_argv(&kargs);
if (status != SIGAR_OK) {
sigar_kern_proc_args_destroy(&kargs);
return status;
}
count = kargs.count;
ptr = kargs.ptr;
end = kargs.end;
/* into environ */
while (ptr < end) {
char *val = strchr(ptr, '=');
int klen, vlen, status;
char key[256]; /* XXX is there a max key size? */
if (val == NULL) {
/* not key=val format */
break;
}
klen = val - ptr;
SIGAR_SSTRCPY(key, ptr);
key[klen] = '\0';
++val;
vlen = strlen(val);
status = procenv->env_getter(procenv->data,
key, klen, val, vlen);
if (status != SIGAR_OK) {
/* not an error; just stop iterating */
break;
}
ptr += (klen + 1 + vlen + 1);
if (*ptr == '\0') {
break;
}
}
sigar_kern_proc_args_destroy(&kargs);
return SIGAR_OK;
#else
char **env;
struct kinfo_proc *pinfo;
int num;
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
pinfo = kvm_getprocs(sigar->kmem, KERN_PROC_PID, pid, &num);
if (!pinfo || (num < 1)) {
return errno;
}
if (!(env = kvm_getenvv(sigar->kmem, pinfo, 9086))) {
return errno;
}
while (*env) {
char *ptr = *env++;
char *val = strchr(ptr, '=');
int klen, vlen, status;
char key[128]; /* XXX is there a max key size? */
if (val == NULL) {
/* not key=val format */
procenv->env_getter(procenv->data, ptr, strlen(ptr), NULL, 0);
break;
}
klen = val - ptr;
SIGAR_SSTRCPY(key, ptr);
key[klen] = '\0';
++val;
vlen = strlen(val);
status = procenv->env_getter(procenv->data,
key, klen, val, vlen);
if (status != SIGAR_OK) {
/* not an error; just stop iterating */
break;
}
ptr += (klen + 1 + vlen + 1);
}
return SIGAR_OK;
#endif
}
int sigar_proc_fd_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_fd_t *procfd)
{
#ifdef __FreeBSD__
int status;
bsd_pinfo_t *pinfo;
struct filedesc filed;
#if 0
struct file **ofiles;
int nfiles, i;
size_t size;
#endif
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
if ((status = sigar_get_pinfo(sigar, pid)) != SIGAR_OK) {
return status;
}
pinfo = sigar->pinfo;
status = kread(sigar, &filed, sizeof(filed), (u_long)pinfo->KI_FD);
if (status != SIGAR_OK) {
return status;
}
#if 0
nfiles = filed.fd_lastfile+1;
size = sizeof(*ofiles) * nfiles;
ofiles = malloc(size);
status = kread(sigar, ofiles, size, (u_long)filed.fd_ofiles);
if (status != SIGAR_OK) {
free(ofiles);
return status;
}
procfd->total = 0;
for (i=0; i<filed.fd_lastfile; i++) {
if (!ofiles[i]) {
continue;
}
procfd->total++;
}
free(ofiles);
#else
/* seems the same as the above */
procfd->total = filed.fd_lastfile;
#endif
return SIGAR_OK;
#else
return SIGAR_ENOTIMPL;
#endif
}
int sigar_proc_exe_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_exe_t *procexe)
{
#ifdef DARWIN
int status;
sigar_kern_proc_args_t kargs;
status = sigar_kern_proc_args_get(sigar, pid, procexe->name, &kargs);
if (status != SIGAR_OK) {
return status;
}
procexe->cwd[0] = '\0';
procexe->root[0] = '\0';
/* attempt to determine cwd from $PWD */
status = kern_proc_args_skip_argv(&kargs);
if (status == SIGAR_OK) {
char *ptr = kargs.ptr;
char *end = kargs.end;
/* into environ */
while (ptr < end) {
int len = strlen(ptr);
if ((len > 4) &&
(ptr[0] == 'P') &&
(ptr[1] == 'W') &&
(ptr[2] == 'D') &&
(ptr[3] == '='))
{
memcpy(procexe->cwd, ptr+4, len-3);
break;
}
ptr += len+1;
}
}
return SIGAR_OK;
#else
int len;
char name[1024];
procexe->cwd[0] = '\0';
procexe->root[0] = '\0';
(void)SIGAR_PROC_FILENAME(name, pid, "/file");
if ((len = readlink(name, procexe->name,
sizeof(procexe->name)-1)) < 0)
{
return PROCFS_STATUS(errno);
}
procexe->name[len] = '\0';
return SIGAR_OK;
#endif
}
#ifdef DARWIN
static int sigar_dlinfo_modules(sigar_t *sigar, sigar_proc_modules_t *procmods)
{
uint32_t i, count = _dyld_image_count();
for (i=0; i<count; i++) {
int status;
const char *name =
_dyld_get_image_name(i);
if (name == NULL) {
continue;
}
status =
procmods->module_getter(procmods->data,
(char *)name, strlen(name));
if (status != SIGAR_OK) {
/* not an error; just stop iterating */
break;
}
}
return SIGAR_OK;
}
#endif /* DARWIN */
int sigar_proc_modules_get(sigar_t *sigar, sigar_pid_t pid,
sigar_proc_modules_t *procmods)
{
#if defined(SIGAR_HAS_DLINFO_MODULES) || defined(DARWIN)
if (pid == sigar_pid_get(sigar)) {
return sigar_dlinfo_modules(sigar, procmods);
}
#endif
return SIGAR_ENOTIMPL;
}
#define SIGAR_MICROSEC2NANO(s) \
((sigar_uint64_t)(s) * (sigar_uint64_t)1000)
#define TIME_NSEC(t) \
(SIGAR_SEC2NANO((t).tv_sec) + SIGAR_MICROSEC2NANO((t).tv_usec))
int sigar_thread_cpu_get(sigar_t *sigar,
sigar_uint64_t id,
sigar_thread_cpu_t *cpu)
{
#if defined(DARWIN)
mach_port_t self = mach_thread_self();
thread_basic_info_data_t info;
mach_msg_type_number_t count = THREAD_BASIC_INFO_COUNT;
kern_return_t status;
status = thread_info(self, THREAD_BASIC_INFO,
(thread_info_t)&info, &count);
if (status != KERN_SUCCESS) {
return errno;
}
mach_port_deallocate(mach_task_self(), self);
cpu->user = tval2nsec(info.user_time);
cpu->sys = tval2nsec(info.system_time);
cpu->total = cpu->user + cpu->sys;
#elif defined(__NetBSD__)
return SIGAR_ENOTIMPL; /* http://tinyurl.com/chbvln */
#else
/* XXX this is not per-thread, it is for the whole-process.
* just want to use for the shell time command at the moment.
*/
struct rusage usage;
getrusage(RUSAGE_SELF, &usage);
cpu->user = TIME_NSEC(usage.ru_utime);
cpu->sys = TIME_NSEC(usage.ru_stime);
cpu->total = TIME_NSEC(usage.ru_utime) + TIME_NSEC(usage.ru_stime);
#endif
return SIGAR_OK;
}
int sigar_os_fs_type_get(sigar_file_system_t *fsp)
{
char *type = fsp->sys_type_name;
/* see sys/disklabel.h */
switch (*type) {
case 'f':
if (strEQ(type, "ffs")) {
fsp->type = SIGAR_FSTYPE_LOCAL_DISK;
}
break;
case 'h':
if (strEQ(type, "hfs")) {
fsp->type = SIGAR_FSTYPE_LOCAL_DISK;
}
break;
case 'u':
if (strEQ(type, "ufs")) {
fsp->type = SIGAR_FSTYPE_LOCAL_DISK;
}
break;
}
return fsp->type;
}
static void get_fs_options(char *opts, int osize, long flags)
{
*opts = '\0';
if (flags & MNT_RDONLY) strncat(opts, "ro", osize);
else strncat(opts, "rw", osize);
if (flags & MNT_SYNCHRONOUS) strncat(opts, ",sync", osize);
if (flags & MNT_NOEXEC) strncat(opts, ",noexec", osize);
if (flags & MNT_NOSUID) strncat(opts, ",nosuid", osize);
#ifdef MNT_NODEV
if (flags & MNT_NODEV) strncat(opts, ",nodev", osize);
#endif
#ifdef MNT_UNION
if (flags & MNT_UNION) strncat(opts, ",union", osize);
#endif
if (flags & MNT_ASYNC) strncat(opts, ",async", osize);
#ifdef MNT_NOATIME
if (flags & MNT_NOATIME) strncat(opts, ",noatime", osize);
#endif
#ifdef MNT_NOCLUSTERR
if (flags & MNT_NOCLUSTERR) strncat(opts, ",noclusterr", osize);
#endif
#ifdef MNT_NOCLUSTERW
if (flags & MNT_NOCLUSTERW) strncat(opts, ",noclusterw", osize);
#endif
#ifdef MNT_NOSYMFOLLOW
if (flags & MNT_NOSYMFOLLOW) strncat(opts, ",nosymfollow", osize);
#endif
#ifdef MNT_SUIDDIR
if (flags & MNT_SUIDDIR) strncat(opts, ",suiddir", osize);
#endif
#ifdef MNT_SOFTDEP
if (flags & MNT_SOFTDEP) strncat(opts, ",soft-updates", osize);
#endif
if (flags & MNT_LOCAL) strncat(opts, ",local", osize);
if (flags & MNT_QUOTA) strncat(opts, ",quota", osize);
if (flags & MNT_ROOTFS) strncat(opts, ",rootfs", osize);
#ifdef MNT_USER
if (flags & MNT_USER) strncat(opts, ",user", osize);
#endif
#ifdef MNT_IGNORE
if (flags & MNT_IGNORE) strncat(opts, ",ignore", osize);
#endif
if (flags & MNT_EXPORTED) strncat(opts, ",nfs", osize);
}
#ifdef __NetBSD__
#define sigar_statfs statvfs
#define sigar_getfsstat getvfsstat
#define sigar_f_flags f_flag
#else
#define sigar_statfs statfs
#define sigar_getfsstat getfsstat
#define sigar_f_flags f_flags
#endif
int sigar_file_system_list_get(sigar_t *sigar,
sigar_file_system_list_t *fslist)
{
struct sigar_statfs *fs;
int num, i;
int is_debug = SIGAR_LOG_IS_DEBUG(sigar);
long len;
if ((num = sigar_getfsstat(NULL, 0, MNT_NOWAIT)) < 0) {
return errno;
}
len = sizeof(*fs) * num;
fs = malloc(len);
if ((num = sigar_getfsstat(fs, len, MNT_NOWAIT)) < 0) {
free(fs);
return errno;
}
sigar_file_system_list_create(fslist);
for (i=0; i<num; i++) {
sigar_file_system_t *fsp;
#ifdef MNT_AUTOMOUNTED
if (fs[i].sigar_f_flags & MNT_AUTOMOUNTED) {
if (is_debug) {
sigar_log_printf(sigar, SIGAR_LOG_DEBUG,
"[file_system_list] skipping automounted %s: %s",
fs[i].f_fstypename, fs[i].f_mntonname);
}
continue;
}
#endif
#ifdef MNT_RDONLY
if (fs[i].sigar_f_flags & MNT_RDONLY) {
/* e.g. ftp mount or .dmg image */
if (is_debug) {
sigar_log_printf(sigar, SIGAR_LOG_DEBUG,
"[file_system_list] skipping readonly %s: %s",
fs[i].f_fstypename, fs[i].f_mntonname);
}
continue;
}
#endif
SIGAR_FILE_SYSTEM_LIST_GROW(fslist);
fsp = &fslist->data[fslist->number++];
SIGAR_SSTRCPY(fsp->dir_name, fs[i].f_mntonname);
SIGAR_SSTRCPY(fsp->dev_name, fs[i].f_mntfromname);
SIGAR_SSTRCPY(fsp->sys_type_name, fs[i].f_fstypename);
get_fs_options(fsp->options, sizeof(fsp->options)-1, fs[i].sigar_f_flags);
sigar_fs_type_init(fsp);
}
free(fs);
return SIGAR_OK;
}
#ifdef DARWIN
#define IoStatGetValue(key, val) \
if ((number = (CFNumberRef)CFDictionaryGetValue(stats, CFSTR(kIOBlockStorageDriverStatistics##key)))) \
CFNumberGetValue(number, kCFNumberSInt64Type, &val)
#endif
int sigar_disk_usage_get(sigar_t *sigar, const char *name,
sigar_disk_usage_t *disk)
{
#if defined(DARWIN)
kern_return_t status;
io_registry_entry_t parent;
io_service_t service;
CFDictionaryRef props;
CFNumberRef number;
sigar_iodev_t *iodev = sigar_iodev_get(sigar, name);
char dname[256], *ptr;
SIGAR_DISK_STATS_INIT(disk);
if (!iodev) {
return ESRCH;
}
/* "/dev/disk0s1" -> "disk0" */ /* XXX better way? */
ptr = &iodev->name[SSTRLEN(SIGAR_DEV_PREFIX)];
SIGAR_SSTRCPY(dname, ptr);
ptr = dname;
if (strnEQ(ptr, "disk", 4)) {
ptr += 4;
if ((ptr = strchr(ptr, 's')) && isdigit(*(ptr+1))) {
*ptr = '\0';
}
}
if (SIGAR_LOG_IS_DEBUG(sigar)) {
sigar_log_printf(sigar, SIGAR_LOG_DEBUG,
"[disk_usage] map %s -> %s",
iodev->name, dname);
}
/* e.g. name == "disk0" */
service = IOServiceGetMatchingService(kIOMasterPortDefault,
IOBSDNameMatching(kIOMasterPortDefault, 0, dname));
if (!service) {
return errno;
}
status = IORegistryEntryGetParentEntry(service, kIOServicePlane, &parent);
if (status != KERN_SUCCESS) {
IOObjectRelease(service);
return status;
}
status = IORegistryEntryCreateCFProperties(parent,
(CFMutableDictionaryRef *)&props,
kCFAllocatorDefault,
kNilOptions);
if (props) {
CFDictionaryRef stats =
(CFDictionaryRef)CFDictionaryGetValue(props,
CFSTR(kIOBlockStorageDriverStatisticsKey));
if (stats) {
IoStatGetValue(ReadsKey, disk->reads);
IoStatGetValue(BytesReadKey, disk->read_bytes);
IoStatGetValue(TotalReadTimeKey, disk->rtime);
IoStatGetValue(WritesKey, disk->writes);
IoStatGetValue(BytesWrittenKey, disk->write_bytes);
IoStatGetValue(TotalWriteTimeKey, disk->wtime);
disk->time = disk->rtime + disk->wtime;
}
CFRelease(props);
}
IOObjectRelease(service);
IOObjectRelease(parent);
return SIGAR_OK;
#else
return SIGAR_ENOTIMPL;
#endif
}
int sigar_file_system_usage_get(sigar_t *sigar,
const char *dirname,
sigar_file_system_usage_t *fsusage)
{
int status = sigar_statvfs(sigar, dirname, fsusage);
if (status != SIGAR_OK) {
return status;
}
fsusage->use_percent = sigar_file_system_usage_calc_used(sigar, fsusage);
#if defined(__FreeBSD__)
fsusage->disk_reads = buf.f_syncreads + buf.f_asyncreads;
fsusage->disk_writes = buf.f_syncwrites + buf.f_asyncwrites;
fsusage->disk_read_bytes = SIGAR_FIELD_NOTIMPL;
fsusage->disk_write_bytes = SIGAR_FIELD_NOTIMPL;
fsusage->disk_queue = SIGAR_FIELD_NOTIMPL;
#elif defined(DARWIN)
sigar_disk_usage_get(sigar, dirname, &fsusage->disk);
#else
SIGAR_DISK_STATS_INIT(&fsusage->disk);
#endif
return SIGAR_OK;
}
#ifdef DARWIN
#define CTL_HW_FREQ "hw.cpufrequency"
#else
/* XXX FreeBSD 5.x+ only? */
#define CTL_HW_FREQ "machdep.tsc_freq"
#endif
int sigar_cpu_info_list_get(sigar_t *sigar,
sigar_cpu_info_list_t *cpu_infos)
{
int i;
unsigned int mhz;
int cache_size=SIGAR_FIELD_NOTIMPL;
size_t size;
char model[128], vendor[128], *ptr;
size = sizeof(mhz);
(void)sigar_cpu_core_count(sigar);
#if defined(DARWIN)
{
int mib[] = { CTL_HW, HW_CPU_FREQ };
size = sizeof(mhz);
if (sysctl(mib, NMIB(mib), &mhz, &size, NULL, 0) < 0) {
mhz = SIGAR_FIELD_NOTIMPL;
}
}
#elif defined(__FreeBSD__)
if (sysctlbyname(CTL_HW_FREQ, &mhz, &size, NULL, 0) < 0) {
mhz = SIGAR_FIELD_NOTIMPL;
}
#else
/*XXX OpenBSD*/
mhz = SIGAR_FIELD_NOTIMPL;
#endif
if (mhz != SIGAR_FIELD_NOTIMPL) {
mhz /= 1000000;
}
size = sizeof(model);
#ifdef __OpenBSD__
if (1) {
#else
if (sysctlbyname("hw.model", &model, &size, NULL, 0) < 0) {
#endif
int mib[] = { CTL_HW, HW_MODEL };
size = sizeof(model);
if (sysctl(mib, NMIB(mib), &model[0], &size, NULL, 0) < 0) {
#ifdef DARWIN
strcpy(model, "powerpc");
#else
strcpy(model, "Unknown");
#endif
}
}
if (mhz == SIGAR_FIELD_NOTIMPL) {
/* freebsd4 */
mhz = sigar_cpu_mhz_from_model(model);
}
#ifdef DARWIN
size = sizeof(vendor);
if (sysctlbyname("machdep.cpu.vendor", &vendor, &size, NULL, 0) < 0) {
SIGAR_SSTRCPY(vendor, "Apple");
}
else {
/* GenuineIntel -> Intel */
if (strstr(vendor, "Intel")) {
SIGAR_SSTRCPY(vendor, "Intel");
}
}
#endif
if ((ptr = strchr(model, ' '))) {
if (strstr(model, "Intel")) {
SIGAR_SSTRCPY(vendor, "Intel");
}
else if (strstr(model, "AMD")) {
SIGAR_SSTRCPY(vendor, "AMD");
}
else {
SIGAR_SSTRCPY(vendor, "Unknown");
}
SIGAR_SSTRCPY(model, ptr+1);
}
#ifdef DARWIN
{
int mib[] = { CTL_HW, HW_L2CACHESIZE }; /* in bytes */
size = sizeof(cache_size);
if (sysctl(mib, NMIB(mib), &cache_size, &size, NULL, 0) < 0) {
cache_size = SIGAR_FIELD_NOTIMPL;
}
else {
cache_size /= 1024; /* convert to KB */
}
}
#endif
sigar_cpu_info_list_create(cpu_infos);
for (i=0; i<sigar->ncpu; i++) {
sigar_cpu_info_t *info;
SIGAR_CPU_INFO_LIST_GROW(cpu_infos);
info = &cpu_infos->data[cpu_infos->number++];
SIGAR_SSTRCPY(info->vendor, vendor);
SIGAR_SSTRCPY(info->model, model);
sigar_cpu_model_adjust(sigar, info);
info->mhz = mhz;
info->cache_size = cache_size;
info->total_cores = sigar->ncpu;
info->cores_per_socket = sigar->lcpu;
info->total_sockets = sigar_cpu_socket_count(sigar);
}
return SIGAR_OK;
}
#define rt_s_addr(sa) ((struct sockaddr_in *)(sa))->sin_addr.s_addr
#ifndef SA_SIZE
#define SA_SIZE(sa) \
( (!(sa) || ((struct sockaddr *)(sa))->sa_len == 0) ? \
sizeof(long) : \
1 + ( (((struct sockaddr *)(sa))->sa_len - 1) | (sizeof(long) - 1) ) )
#endif
int sigar_net_route_list_get(sigar_t *sigar,
sigar_net_route_list_t *routelist)
{
size_t needed;
int bit;
char *buf, *next, *lim;
struct rt_msghdr *rtm;
int mib[6] = { CTL_NET, PF_ROUTE, 0, 0, NET_RT_DUMP, 0 };
if (sysctl(mib, NMIB(mib), NULL, &needed, NULL, 0) < 0) {
return errno;
}
buf = malloc(needed);
if (sysctl(mib, NMIB(mib), buf, &needed, NULL, 0) < 0) {
free(buf);
return errno;
}
sigar_net_route_list_create(routelist);
lim = buf + needed;
for (next = buf; next < lim; next += rtm->rtm_msglen) {
struct sockaddr *sa;
sigar_net_route_t *route;
rtm = (struct rt_msghdr *)next;
if (rtm->rtm_type != RTM_GET) {
continue;
}
sa = (struct sockaddr *)(rtm + 1);
if (sa->sa_family != AF_INET) {
continue;
}
SIGAR_NET_ROUTE_LIST_GROW(routelist);
route = &routelist->data[routelist->number++];
SIGAR_ZERO(route);
route->flags = rtm->rtm_flags;
for (bit=RTA_DST;
bit && ((char *)sa < lim);
bit <<= 1)
{
if ((rtm->rtm_addrs & bit) == 0) {
continue;
}
switch (bit) {
case RTA_DST:
sigar_net_address_set(route->destination,
rt_s_addr(sa));
break;
case RTA_GATEWAY:
if (sa->sa_family == AF_INET) {
sigar_net_address_set(route->gateway,
rt_s_addr(sa));
}
break;
case RTA_NETMASK:
sigar_net_address_set(route->mask,
rt_s_addr(sa));
break;
case RTA_IFA:
break;
}
sa = (struct sockaddr *)((char *)sa + SA_SIZE(sa));
}
}
free(buf);
return SIGAR_OK;
}
typedef enum {
IFMSG_ITER_LIST,
IFMSG_ITER_GET
} ifmsg_iter_e;
typedef struct {
const char *name;
ifmsg_iter_e type;
union {
sigar_net_interface_list_t *iflist;
struct if_msghdr *ifm;
} data;
} ifmsg_iter_t;
static int sigar_ifmsg_init(sigar_t *sigar)
{
int mib[] = { CTL_NET, PF_ROUTE, 0, AF_INET, NET_RT_IFLIST, 0 };
size_t len;
if (sysctl(mib, NMIB(mib), NULL, &len, NULL, 0) < 0) {
return errno;
}
if (sigar->ifconf_len < len) {
sigar->ifconf_buf = realloc(sigar->ifconf_buf, len);
sigar->ifconf_len = len;
}
if (sysctl(mib, NMIB(mib), sigar->ifconf_buf, &len, NULL, 0) < 0) {
return errno;
}
return SIGAR_OK;
}
static int has_ifaddr(char *name)
{
int sock, status;
struct ifreq ifr;
if ((sock = socket(AF_INET, SOCK_DGRAM, 0)) < 0) {
return errno;
}
SIGAR_SSTRCPY(ifr.ifr_name, name);
if (ioctl(sock, SIOCGIFADDR, &ifr) == 0) {
status = SIGAR_OK;
}
else {
status = errno;
}
close(sock);
return status;
}
static int sigar_ifmsg_iter(sigar_t *sigar, ifmsg_iter_t *iter)
{
char *end = sigar->ifconf_buf + sigar->ifconf_len;
char *ptr = sigar->ifconf_buf;
if (iter->type == IFMSG_ITER_LIST) {
sigar_net_interface_list_create(iter->data.iflist);
}
while (ptr < end) {
char *name;
struct sockaddr_dl *sdl;
struct if_msghdr *ifm = (struct if_msghdr *)ptr;
if (ifm->ifm_type != RTM_IFINFO) {
break;
}
ptr += ifm->ifm_msglen;
while (ptr < end) {
struct if_msghdr *next = (struct if_msghdr *)ptr;
if (next->ifm_type != RTM_NEWADDR) {
break;
}
ptr += next->ifm_msglen;
}
sdl = (struct sockaddr_dl *)(ifm + 1);
if (sdl->sdl_family != AF_LINK) {
continue;
}
switch (iter->type) {
case IFMSG_ITER_LIST:
if (sdl->sdl_type == IFT_OTHER) {
if (has_ifaddr(sdl->sdl_data) != SIGAR_OK) {
break;
}
}
else if (!((sdl->sdl_type == IFT_ETHER) ||
(sdl->sdl_type == IFT_LOOP)))
{
break; /* XXX deal w/ other weirdo interfaces */
}
SIGAR_NET_IFLIST_GROW(iter->data.iflist);
/* sdl_data doesn't include a trailing \0, it is only sdl_nlen long */
name = malloc(sdl->sdl_nlen+1);
memcpy(name, sdl->sdl_data, sdl->sdl_nlen);
name[sdl->sdl_nlen] = '\0'; /* add the missing \0 */
iter->data.iflist->data[iter->data.iflist->number++] = name;
break;
case IFMSG_ITER_GET:
if (strEQ(iter->name, sdl->sdl_data)) {
iter->data.ifm = ifm;
return SIGAR_OK;
}
}
}
switch (iter->type) {
case IFMSG_ITER_LIST:
return SIGAR_OK;
case IFMSG_ITER_GET:
default:
return ENXIO;
}
}
int sigar_net_interface_list_get(sigar_t *sigar,
sigar_net_interface_list_t *iflist)
{
int status;
ifmsg_iter_t iter;
if ((status = sigar_ifmsg_init(sigar)) != SIGAR_OK) {
return status;
}
iter.type = IFMSG_ITER_LIST;
iter.data.iflist = iflist;
return sigar_ifmsg_iter(sigar, &iter);
}
int sigar_net_interface_config_get(sigar_t *sigar, const char *name,
sigar_net_interface_config_t *ifconfig)
{
int sock;
int status;
ifmsg_iter_t iter;
struct if_msghdr *ifm;
struct sockaddr_dl *sdl;
struct ifreq ifr;
if (!name) {
return sigar_net_interface_config_primary_get(sigar, ifconfig);
}
if (sigar->ifconf_len == 0) {
if ((status = sigar_ifmsg_init(sigar)) != SIGAR_OK) {
return status;
}
}
SIGAR_ZERO(ifconfig);
iter.type = IFMSG_ITER_GET;
iter.name = name;
if ((status = sigar_ifmsg_iter(sigar, &iter)) != SIGAR_OK) {
return status;
}
if ((sock = socket(AF_INET, SOCK_DGRAM, 0)) < 0) {
return errno;
}
ifm = iter.data.ifm;
SIGAR_SSTRCPY(ifconfig->name, name);
sdl = (struct sockaddr_dl *)(ifm + 1);
sigar_net_address_mac_set(ifconfig->hwaddr,
LLADDR(sdl),
sdl->sdl_alen);
ifconfig->flags = ifm->ifm_flags;
ifconfig->mtu = ifm->ifm_data.ifi_mtu;
ifconfig->metric = ifm->ifm_data.ifi_metric;
SIGAR_SSTRCPY(ifr.ifr_name, name);
#define ifr_s_addr(ifr) \
((struct sockaddr_in *)&ifr.ifr_addr)->sin_addr.s_addr
if (!ioctl(sock, SIOCGIFADDR, &ifr)) {
sigar_net_address_set(ifconfig->address,
ifr_s_addr(ifr));
}
if (!ioctl(sock, SIOCGIFNETMASK, &ifr)) {
sigar_net_address_set(ifconfig->netmask,
ifr_s_addr(ifr));
}
if (ifconfig->flags & IFF_LOOPBACK) {
sigar_net_address_set(ifconfig->destination,
ifconfig->address.addr.in);
sigar_net_address_set(ifconfig->broadcast, 0);
SIGAR_SSTRCPY(ifconfig->type,
SIGAR_NIC_LOOPBACK);
}
else {
if (!ioctl(sock, SIOCGIFDSTADDR, &ifr)) {
sigar_net_address_set(ifconfig->destination,
ifr_s_addr(ifr));
}
if (!ioctl(sock, SIOCGIFBRDADDR, &ifr)) {
sigar_net_address_set(ifconfig->broadcast,
ifr_s_addr(ifr));
}
SIGAR_SSTRCPY(ifconfig->type,
SIGAR_NIC_ETHERNET);
}
close(sock);
/* XXX can we get a better description like win32? */
SIGAR_SSTRCPY(ifconfig->description,
ifconfig->name);
return SIGAR_OK;
}
int sigar_net_interface_stat_get(sigar_t *sigar, const char *name,
sigar_net_interface_stat_t *ifstat)
{
int status;
ifmsg_iter_t iter;
struct if_msghdr *ifm;
if ((status = sigar_ifmsg_init(sigar)) != SIGAR_OK) {
return status;
}
iter.type = IFMSG_ITER_GET;
iter.name = name;
if ((status = sigar_ifmsg_iter(sigar, &iter)) != SIGAR_OK) {
return status;
}
ifm = iter.data.ifm;
ifstat->rx_bytes = ifm->ifm_data.ifi_ibytes;
ifstat->rx_packets = ifm->ifm_data.ifi_ipackets;
ifstat->rx_errors = ifm->ifm_data.ifi_ierrors;
ifstat->rx_dropped = ifm->ifm_data.ifi_iqdrops;
ifstat->rx_overruns = SIGAR_FIELD_NOTIMPL;
ifstat->rx_frame = SIGAR_FIELD_NOTIMPL;
ifstat->tx_bytes = ifm->ifm_data.ifi_obytes;
ifstat->tx_packets = ifm->ifm_data.ifi_opackets;
ifstat->tx_errors = ifm->ifm_data.ifi_oerrors;
ifstat->tx_collisions = ifm->ifm_data.ifi_collisions;
ifstat->tx_dropped = SIGAR_FIELD_NOTIMPL;
ifstat->tx_overruns = SIGAR_FIELD_NOTIMPL;
ifstat->tx_carrier = SIGAR_FIELD_NOTIMPL;
ifstat->speed = ifm->ifm_data.ifi_baudrate;
return SIGAR_OK;
}
static int net_connection_state_get(int state)
{
switch (state) {
case TCPS_CLOSED:
return SIGAR_TCP_CLOSE;
case TCPS_LISTEN:
return SIGAR_TCP_LISTEN;
case TCPS_SYN_SENT:
return SIGAR_TCP_SYN_SENT;
case TCPS_SYN_RECEIVED:
return SIGAR_TCP_SYN_RECV;
case TCPS_ESTABLISHED:
return SIGAR_TCP_ESTABLISHED;
case TCPS_CLOSE_WAIT:
return SIGAR_TCP_CLOSE_WAIT;
case TCPS_FIN_WAIT_1:
return SIGAR_TCP_FIN_WAIT1;
case TCPS_CLOSING:
return SIGAR_TCP_CLOSING;
case TCPS_LAST_ACK:
return SIGAR_TCP_LAST_ACK;
case TCPS_FIN_WAIT_2:
return SIGAR_TCP_FIN_WAIT2;
case TCPS_TIME_WAIT:
return SIGAR_TCP_TIME_WAIT;
default:
return SIGAR_TCP_UNKNOWN;
}
}
#if defined(__OpenBSD__) || defined(__NetBSD__)
static int net_connection_get(sigar_net_connection_walker_t *walker, int proto)
{
int status;
int istcp = 0, type;
int flags = walker->flags;
struct inpcbtable table;
struct inpcb *head, *next, *prev;
sigar_t *sigar = walker->sigar;
u_long offset;
switch (proto) {
case IPPROTO_TCP:
offset = sigar->koffsets[KOFFSET_TCBTABLE];
istcp = 1;
type = SIGAR_NETCONN_TCP;
break;
case IPPROTO_UDP:
default:
return SIGAR_ENOTIMPL;
}
status = kread(sigar, &table, sizeof(table), offset);
if (status != SIGAR_OK) {
return status;
}
prev = head =
(struct inpcb *)&CIRCLEQ_FIRST(&((struct inpcbtable *)offset)->inpt_queue);
next = (struct inpcb *)CIRCLEQ_FIRST(&table.inpt_queue);
while (next != head) {
struct inpcb inpcb;
struct tcpcb tcpcb;
struct socket socket;
status = kread(sigar, &inpcb, sizeof(inpcb), (long)next);
prev = next;
next = (struct inpcb *)CIRCLEQ_NEXT(&inpcb, inp_queue);
kread(sigar, &socket, sizeof(socket), (u_long)inpcb.inp_socket);
if ((((flags & SIGAR_NETCONN_SERVER) && socket.so_qlimit) ||
((flags & SIGAR_NETCONN_CLIENT) && !socket.so_qlimit)))
{
sigar_net_connection_t conn;
SIGAR_ZERO(&conn);
if (istcp) {
kread(sigar, &tcpcb, sizeof(tcpcb), (u_long)inpcb.inp_ppcb);
}
#ifdef __NetBSD__
if (inpcb.inp_af == AF_INET6) {
/*XXX*/
continue;
}
#else
if (inpcb.inp_flags & INP_IPV6) {
sigar_net_address6_set(conn.local_address,
&inpcb.inp_laddr6.s6_addr);
sigar_net_address6_set(conn.remote_address,
&inpcb.inp_faddr6.s6_addr);
}
#endif
else {
sigar_net_address_set(conn.local_address,
inpcb.inp_laddr.s_addr);
sigar_net_address_set(conn.remote_address,
inpcb.inp_faddr.s_addr);
}
conn.local_port = ntohs(inpcb.inp_lport);
conn.remote_port = ntohs(inpcb.inp_fport);
conn.receive_queue = socket.so_rcv.sb_cc;
conn.send_queue = socket.so_snd.sb_cc;
conn.uid = socket.so_pgid;
conn.type = type;
if (!istcp) {
conn.state = SIGAR_TCP_UNKNOWN;
if (walker->add_connection(walker, &conn) != SIGAR_OK) {
break;
}
continue;
}
conn.state = net_connection_state_get(tcpcb.t_state);
if (walker->add_connection(walker, &conn) != SIGAR_OK) {
break;
}
}
}
return SIGAR_OK;
}
#else
static int net_connection_get(sigar_net_connection_walker_t *walker, int proto)
{
int flags = walker->flags;
int type, istcp = 0;
char *buf;
const char *mibvar;
struct tcpcb *tp = NULL;
struct inpcb *inp;
struct xinpgen *xig, *oxig;
struct xsocket *so;
size_t len;
switch (proto) {
case IPPROTO_TCP:
mibvar = "net.inet.tcp.pcblist";
istcp = 1;
type = SIGAR_NETCONN_TCP;
break;
case IPPROTO_UDP:
mibvar = "net.inet.udp.pcblist";
type = SIGAR_NETCONN_UDP;
break;
default:
mibvar = "net.inet.raw.pcblist";
type = SIGAR_NETCONN_RAW;
break;
}
len = 0;
if (sysctlbyname(mibvar, 0, &len, 0, 0) < 0) {
return errno;
}
if ((buf = malloc(len)) == 0) {
return errno;
}
if (sysctlbyname(mibvar, buf, &len, 0, 0) < 0) {
free(buf);
return errno;
}
oxig = xig = (struct xinpgen *)buf;
for (xig = (struct xinpgen *)((char *)xig + xig->xig_len);
xig->xig_len > sizeof(struct xinpgen);
xig = (struct xinpgen *)((char *)xig + xig->xig_len))
{
if (istcp) {
struct xtcpcb *cb = (struct xtcpcb *)xig;
tp = &cb->xt_tp;
inp = &cb->xt_inp;
so = &cb->xt_socket;
}
else {
struct xinpcb *cb = (struct xinpcb *)xig;
inp = &cb->xi_inp;
so = &cb->xi_socket;
}
if (so->xso_protocol != proto) {
continue;
}
if (inp->inp_gencnt > oxig->xig_gen) {
continue;
}
if ((((flags & SIGAR_NETCONN_SERVER) && so->so_qlimit) ||
((flags & SIGAR_NETCONN_CLIENT) && !so->so_qlimit)))
{
sigar_net_connection_t conn;
SIGAR_ZERO(&conn);
if (inp->inp_vflag & INP_IPV6) {
sigar_net_address6_set(conn.local_address,
&inp->in6p_laddr.s6_addr);
sigar_net_address6_set(conn.remote_address,
&inp->in6p_faddr.s6_addr);
}
else {
sigar_net_address_set(conn.local_address,
inp->inp_laddr.s_addr);
sigar_net_address_set(conn.remote_address,
inp->inp_faddr.s_addr);
}
conn.local_port = ntohs(inp->inp_lport);
conn.remote_port = ntohs(inp->inp_fport);
conn.receive_queue = so->so_rcv.sb_cc;
conn.send_queue = so->so_snd.sb_cc;
conn.uid = so->so_pgid;
conn.type = type;
if (!istcp) {
conn.state = SIGAR_TCP_UNKNOWN;
if (walker->add_connection(walker, &conn) != SIGAR_OK) {
break;
}
continue;
}
conn.state = net_connection_state_get(tp->t_state);
if (walker->add_connection(walker, &conn) != SIGAR_OK) {
break;
}
}
}
free(buf);
return SIGAR_OK;
}
#endif
int sigar_net_connection_walk(sigar_net_connection_walker_t *walker)
{
int flags = walker->flags;
int status;
if (flags & SIGAR_NETCONN_TCP) {
status = net_connection_get(walker, IPPROTO_TCP);
if (status != SIGAR_OK) {
return status;
}
}
if (flags & SIGAR_NETCONN_UDP) {
status = net_connection_get(walker, IPPROTO_UDP);
if (status != SIGAR_OK) {
return status;
}
}
return SIGAR_OK;
}
SIGAR_DECLARE(int)
sigar_tcp_get(sigar_t *sigar,
sigar_tcp_t *tcp)
{
struct tcpstat mib;
#if !defined(TCPCTL_STATS) && (defined(__OpenBSD__) || defined(__NetBSD__))
int status =
kread(sigar, &mib, sizeof(mib),
sigar->koffsets[KOFFSET_TCPSTAT]);
if (status != SIGAR_OK) {
return status;
}
#else
int var[4] = { CTL_NET, PF_INET, IPPROTO_TCP, TCPCTL_STATS };
size_t len = sizeof(mib);
if (sysctl(var, NMIB(var), &mib, &len, NULL, 0) < 0) {
return errno;
}
#endif
tcp->active_opens = mib.tcps_connattempt;
tcp->passive_opens = mib.tcps_accepts;
tcp->attempt_fails = mib.tcps_conndrops;
tcp->estab_resets = mib.tcps_drops;
if (sigar_tcp_curr_estab(sigar, tcp) != SIGAR_OK) {
tcp->curr_estab = -1;
}
tcp->in_segs = mib.tcps_rcvtotal;
tcp->out_segs = mib.tcps_sndtotal - mib.tcps_sndrexmitpack;
tcp->retrans_segs = mib.tcps_sndrexmitpack;
tcp->in_errs =
mib.tcps_rcvbadsum +
mib.tcps_rcvbadoff +
mib.tcps_rcvmemdrop +
mib.tcps_rcvshort;
tcp->out_rsts = -1; /* XXX mib.tcps_sndctrl - mib.tcps_closed; ? */
return SIGAR_OK;
}
#ifndef SIGAR_FREEBSD5_NFSSTAT
static int get_nfsstats(struct nfsstats *stats)
{
size_t len = sizeof(*stats);
int mib[] = { CTL_VFS, 2, NFS_NFSSTATS };
if (sysctl(mib, NMIB(mib), stats, &len, NULL, 0) < 0) {
return errno;
}
else {
return SIGAR_OK;
}
}
#endif
#if defined(__OpenBSD__)
typedef uint64_t rpc_cnt_t;
#else
typedef int rpc_cnt_t;
#endif
static void map_nfs_stats(sigar_nfs_v3_t *nfs, rpc_cnt_t *rpc)
{
nfs->null = rpc[NFSPROC_NULL];
nfs->getattr = rpc[NFSPROC_GETATTR];
nfs->setattr = rpc[NFSPROC_SETATTR];
nfs->lookup = rpc[NFSPROC_LOOKUP];
nfs->access = rpc[NFSPROC_ACCESS];
nfs->readlink = rpc[NFSPROC_READLINK];
nfs->read = rpc[NFSPROC_READ];
nfs->write = rpc[NFSPROC_WRITE];
nfs->create = rpc[NFSPROC_CREATE];
nfs->mkdir = rpc[NFSPROC_MKDIR];
nfs->symlink = rpc[NFSPROC_SYMLINK];
nfs->mknod = rpc[NFSPROC_MKNOD];
nfs->remove = rpc[NFSPROC_REMOVE];
nfs->rmdir = rpc[NFSPROC_RMDIR];
nfs->rename = rpc[NFSPROC_RENAME];
nfs->link = rpc[NFSPROC_LINK];
nfs->readdir = rpc[NFSPROC_READDIR];
nfs->readdirplus = rpc[NFSPROC_READDIRPLUS];
nfs->fsstat = rpc[NFSPROC_FSSTAT];
nfs->fsinfo = rpc[NFSPROC_FSINFO];
nfs->pathconf = rpc[NFSPROC_PATHCONF];
nfs->commit = rpc[NFSPROC_COMMIT];
}
int sigar_nfs_client_v2_get(sigar_t *sigar,
sigar_nfs_client_v2_t *nfs)
{
return SIGAR_ENOTIMPL;
}
int sigar_nfs_server_v2_get(sigar_t *sigar,
sigar_nfs_server_v2_t *nfs)
{
return SIGAR_ENOTIMPL;
}
int sigar_nfs_client_v3_get(sigar_t *sigar,
sigar_nfs_client_v3_t *nfs)
{
#ifdef SIGAR_FREEBSD5_NFSSTAT
struct nfsstats stats;
size_t size = sizeof(stats);
if (sysctlbyname("vfs.nfs.nfsstats", &stats, &size, NULL, 0) == -1) {
return errno;
}
map_nfs_stats((sigar_nfs_v3_t *)nfs, &stats.rpccnt[0]);
#else
int status;
struct nfsstats stats;
if ((status = get_nfsstats(&stats)) != SIGAR_OK) {
return status;
}
map_nfs_stats((sigar_nfs_v3_t *)nfs, &stats.rpccnt[0]);
#endif
return SIGAR_OK;
}
int sigar_nfs_server_v3_get(sigar_t *sigar,
sigar_nfs_server_v3_t *nfs)
{
#ifdef SIGAR_FREEBSD5_NFSSTAT
struct nfsrvstats stats;
size_t size = sizeof(stats);
if (sysctlbyname("vfs.nfsrv.nfsrvstats", &stats, &size, NULL, 0) == -1) {
return errno;
}
map_nfs_stats((sigar_nfs_v3_t *)nfs, &stats.srvrpccnt[0]);
#else
int status;
struct nfsstats stats;
if ((status = get_nfsstats(&stats)) != SIGAR_OK) {
return status;
}
map_nfs_stats((sigar_nfs_v3_t *)nfs, &stats.srvrpccnt[0]);
#endif
return SIGAR_OK;
}
#if defined(__FreeBSD__)
#define _KERNEL
#include <sys/file.h>
#undef _KERNEL
/* derived from
* /usr/ports/security/pidentd/work/pidentd-3.0.16/src/k_freebsd2.c
*/
int sigar_proc_port_get(sigar_t *sigar, int protocol,
unsigned long port, sigar_pid_t *pid)
{
struct nlist nl[2];
struct inpcbhead tcb;
struct socket *sockp = NULL;
struct kinfo_proc *pinfo;
struct inpcb *head, pcbp;
int i, nentries, status;
if (protocol != SIGAR_NETCONN_TCP) {
return SIGAR_ENOTIMPL;
}
if (!sigar->kmem) {
return SIGAR_EPERM_KMEM;
}
nl[0].n_name = "_tcb"; /* XXX cache */
nl[1].n_name = "";
if (kvm_nlist(sigar->kmem, nl) < 0) {
return errno;
}
status = kread(sigar, &tcb, sizeof(tcb), nl[0].n_value);
if (status != SIGAR_OK) {
return status;
}
for (head = tcb.lh_first; head != NULL;
head = pcbp.inp_list.le_next)
{
status = kread(sigar, &pcbp, sizeof(pcbp), (long)head);
if (status != SIGAR_OK) {
return status;
}
if (!(pcbp.inp_vflag & INP_IPV4)) {
continue;
}
if (pcbp.inp_fport != 0) {
continue;
}
if (ntohs(pcbp.inp_lport) == port) {
sockp = pcbp.inp_socket;
break;
}
}
if (!sockp) {
return ENOENT;
}
pinfo = kvm_getprocs(sigar->kmem, KERN_PROC_PROC, 0, &nentries);
if (!pinfo) {
return errno;
}
for (i=0; i<nentries; i++) {
if (pinfo[i].KI_FLAG & P_SYSTEM) {
continue;
}
if (pinfo[i].KI_FD) {
struct filedesc pfd;
struct file **ofiles, ofile;
int j, osize;
status = kread(sigar, &pfd, sizeof(pfd), (long)pinfo[i].KI_FD);
if (status != SIGAR_OK) {
return status;
}
osize = pfd.fd_nfiles * sizeof(struct file *);
ofiles = malloc(osize); /* XXX reuse */
if (!ofiles) {
return errno;
}
status = kread(sigar, ofiles, osize, (long)pfd.fd_ofiles);
if (status != SIGAR_OK) {
free(ofiles);
return status;
}
for (j=0; j<pfd.fd_nfiles; j++) {
if (!ofiles[j]) {
continue;
}
status = kread(sigar, &ofile, sizeof(ofile), (long)ofiles[j]);
if (status != SIGAR_OK) {
free(ofiles);
return status;
}
if (ofile.f_count == 0) {
continue;
}
if (ofile.f_type == DTYPE_SOCKET &&
(struct socket *)ofile.f_data == sockp)
{
*pid = pinfo[i].KI_PID;
free(ofiles);
return SIGAR_OK;
}
}
free(ofiles);
}
}
return ENOENT;
}
#elif defined(DARWIN) && defined(DARWIN_HAS_LIBPROC_H)
int sigar_proc_port_get(sigar_t *sigar, int protocol,
unsigned long port, sigar_pid_t *pid)
{
sigar_proc_list_t pids;
int i, status, found=0;
if (!sigar->libproc) {
return SIGAR_ENOTIMPL;
}
status = sigar_proc_list_get(sigar, &pids);
if (status != SIGAR_OK) {
return status;
}
for (i=0; i<pids.number; i++) {
int n, num=0;
struct proc_fdinfo *fdinfo;
status = proc_fdinfo_get(sigar, pids.data[i], &num);
if (status != SIGAR_OK) {
continue;
}
fdinfo = (struct proc_fdinfo *)sigar->ifconf_buf;
for (n=0; n<num; n++) {
struct proc_fdinfo *fdp = &fdinfo[n];
struct socket_fdinfo si;
int rsize, family;
unsigned long lport;
if (fdp->proc_fdtype != PROX_FDTYPE_SOCKET) {
continue;
}
rsize = sigar->proc_pidfdinfo(pids.data[i], fdp->proc_fd,
PROC_PIDFDSOCKETINFO, &si, sizeof(si));
if (rsize != sizeof(si)) {
continue;
}
if (si.psi.soi_kind != SOCKINFO_TCP) {
continue;
}
if (si.psi.soi_proto.pri_tcp.tcpsi_state != TSI_S_LISTEN) {
continue;
}
family = si.psi.soi_family;
if (!((family == AF_INET) || (family == AF_INET6))) {
continue;
}
lport = ntohs(si.psi.soi_proto.pri_tcp.tcpsi_ini.insi_lport);
if (lport == port) {
*pid = pids.data[i];
found = 1;
break;
}
}
}
sigar_proc_list_destroy(sigar, &pids);
return found ? SIGAR_OK : ENOENT;
}
#else
int sigar_proc_port_get(sigar_t *sigar, int protocol,
unsigned long port, sigar_pid_t *pid)
{
return SIGAR_ENOTIMPL;
}
#endif
int sigar_os_sys_info_get(sigar_t *sigar,
sigar_sys_info_t *sysinfo)
{
#ifdef DARWIN
char *codename = NULL;
SInt32 version, version_major, version_minor, version_fix;
SIGAR_SSTRCPY(sysinfo->name, "MacOSX");
SIGAR_SSTRCPY(sysinfo->vendor_name, "Mac OS X");
SIGAR_SSTRCPY(sysinfo->vendor, "Apple");
if (Gestalt(gestaltSystemVersion, &version) == noErr) {
if (version >= 0x00001040) {
Gestalt('sys1' /*gestaltSystemVersionMajor*/, &version_major);
Gestalt('sys2' /*gestaltSystemVersionMinor*/, &version_minor);
Gestalt('sys3' /*gestaltSystemVersionBugFix*/, &version_fix);
}
else {
version_fix = version & 0xf;
version >>= 4;
version_minor = version & 0xf;
version >>= 4;
version_major = version - (version >> 4) * 6;
}
}
else {
return SIGAR_ENOTIMPL;
}
snprintf(sysinfo->vendor_version,
sizeof(sysinfo->vendor_version),
"%d.%d",
(int)version_major, (int)version_minor);
snprintf(sysinfo->version,
sizeof(sysinfo->version),
"%s.%d",
sysinfo->vendor_version, (int)version_fix);
if (version_major == 10) {
switch (version_minor) {
case 2:
codename = "Jaguar";
break;
case 3:
codename = "Panther";
break;
case 4:
codename = "Tiger";
break;
case 5:
codename = "Leopard";
break;
default:
codename = "Unknown";
break;
}
}
else {
return SIGAR_ENOTIMPL;
}
SIGAR_SSTRCPY(sysinfo->vendor_code_name, codename);
snprintf(sysinfo->description,
sizeof(sysinfo->description),
"%s %s",
sysinfo->vendor_name, sysinfo->vendor_code_name);
#else
char *ptr;
#if defined(__FreeBSD__)
SIGAR_SSTRCPY(sysinfo->name, "FreeBSD");
#elif defined(__OpenBSD__)
SIGAR_SSTRCPY(sysinfo->name, "OpenBSD");
#elif defined(__NetBSD__)
SIGAR_SSTRCPY(sysinfo->name, "NetBSD");
#else
SIGAR_SSTRCPY(sysinfo->name, "Unknown");
#endif
SIGAR_SSTRCPY(sysinfo->vendor_name, sysinfo->name);
SIGAR_SSTRCPY(sysinfo->vendor, sysinfo->name);
SIGAR_SSTRCPY(sysinfo->vendor_version,
sysinfo->version);
if ((ptr = strstr(sysinfo->vendor_version, "-"))) {
/* STABLE, RELEASE, CURRENT */
*ptr++ = '\0';
SIGAR_SSTRCPY(sysinfo->vendor_code_name, ptr);
}
snprintf(sysinfo->description,
sizeof(sysinfo->description),
"%s %s",
sysinfo->name, sysinfo->version);
#endif
return SIGAR_OK;
}