lookup_result_type<Type> lookup(const pmInDom indom,
const instance_id_type instance_id,
const lookup_flags flags = require_active)
{
lookup_result_type<Type> result;
void * opaque;
result.name = NULL;
result.status = pmdaCacheLookup(indom, instance_id, &result.name, &opaque);
if (result.status < 0) {
throw pcp::exception(result.status);
}
if ((flags & require_active) && (result.status != PMDA_CACHE_ACTIVE)) {
std::ostringstream message;
message << "Cache entry " << indom << ':' << instance_id << " (";
if (result.name == NULL) {
message << "NULL";
} else {
message << '"' << result.name << '"';
}
message << ") inactive";
throw pcp::exception(result.status, message.str());
}
result.instance_id = instance_id;
result.opaque = static_cast<Type>(opaque);
return result;
}
int
cgroup_mounts_subsys(const char *system, char *buffer, int length)
{
pmInDom mounts = INDOM(CGROUP_MOUNTS_INDOM);
pmInDom subsys = INDOM(CGROUP_SUBSYS_INDOM);
filesys_t *fs;
char *name;
int sts;
/* Iterate over cgroup.mounts.subsys indom, comparing the value
* with the given subsys - if a match is found, return the inst
* name, else NULL.
*/
pmdaCacheOp(mounts, PMDA_CACHE_WALK_REWIND);
while ((sts = pmdaCacheOp(mounts, PMDA_CACHE_WALK_NEXT)) != -1) {
if (!pmdaCacheLookup(mounts, sts, &name, (void **)&fs))
continue;
if (strcmp(system, cgroup_find_subsys(subsys, fs)) != 0)
continue;
snprintf(buffer, length, "%s%s/", proc_statspath, name);
buffer[length-1] = '\0';
return strlen(buffer);
}
return 0;
}
int
proc_slabinfo_fetch(pmInDom indom, int item, unsigned int inst, pmAtomValue *ap)
{
slab_cache_t *slab_cache = NULL;
int sts;
sts = pmdaCacheLookup(indom, inst, NULL, (void **)&slab_cache);
if (sts < 0)
return sts;
if (sts == PMDA_CACHE_INACTIVE)
return PM_ERR_INST;
switch (item) {
case 0: /* mem.slabinfo.objects.active */
ap->ull = slab_cache->num_active_objs;
break;
case 1: /* mem.slabinfo.objects.total */
ap->ull = slab_cache->total_objs;
break;
case 2: /* mem.slabinfo.objects.size */
if (slab_cache->seen < 11) /* version 1.1 or later only */
return 0;
ap->ul = slab_cache->object_size;
break;
case 3: /* mem.slabinfo.slabs.active */
if (slab_cache->seen < 11) /* version 1.1 or later only */
return 0;
ap->ul = slab_cache->num_active_slabs;
break;
case 4: /* mem.slabinfo.slabs.total */
if (slab_cache->seen == 11) /* version 1.1 only */
return 0;
ap->ul = slab_cache->total_slabs;
break;
case 5: /* mem.slabinfo.slabs.pages_per_slab */
if (slab_cache->seen < 11) /* version 1.1 or later only */
return 0;
ap->ul = slab_cache->pages_per_slab;
break;
case 6: /* mem.slabinfo.slabs.objects_per_slab */
if (slab_cache->seen != 20) /* version 2.0 only */
return 0;
ap->ul = slab_cache->objects_per_slab;
break;
case 7: /* mem.slabinfo.slabs.total_size */
if (slab_cache->seen < 11) /* version 1.1 or later only */
return 0;
ap->ull = slab_cache->total_size;
break;
default:
return PM_ERR_PMID;
}
return 1;
}
int
netlink_fetch(pmdaMetric *pm, int inst, pmAtomValue *av)
{
char *lname;
metricdesc_t *md = pm->m_user;
kstat_t *k;
char *stat = (char *)md->md_offset;
if (pmdaCacheLookup(indomtab[NETLINK_INDOM].it_indom, inst, &lname,
(void **)&k) != PMDA_CACHE_ACTIVE)
return PM_ERR_INST;
if (k) {
kstat_named_t *kn = kstat_data_lookup(k, stat);
if (kn == NULL) {
fprintf(stderr, "No kstat called %s for %s\n", stat, lname);
return 0;
}
switch (pm->m_desc.type) {
case PM_TYPE_32:
if (kn->data_type == KSTAT_DATA_INT32) {
av->l = kn->value.i32;
return 1;
}
break;
case PM_TYPE_U32:
if (kn->data_type == KSTAT_DATA_UINT32) {
av->ul = kn->value.ui32;
return 1;
}
break;
case PM_TYPE_64:
if (kn->data_type == KSTAT_DATA_INT64) {
av->ll = kn->value.i64;
return 1;
}
break;
case PM_TYPE_U64:
if (kn->data_type == KSTAT_DATA_UINT64) {
av->ull = kn->value.ui64;
return 1;
}
break;
}
}
return 0;
}
int
refresh_net_ioctl(pmInDom indom, linux_container_t *cp, int *need_refresh)
{
net_interface_t *netip;
char *p;
int sts = 0;
for (pmdaCacheOp(indom, PMDA_CACHE_WALK_REWIND);;) {
if ((sts = pmdaCacheOp(indom, PMDA_CACHE_WALK_NEXT)) < 0)
break;
if (!pmdaCacheLookup(indom, sts, &p, (void **)&netip) || !p)
continue;
refresh_net_dev_ioctl(p, netip, cp, need_refresh);
}
return sts;
}
/*
* For block devices we have one instance domain for dev_t
* based lookup, and another for (real) name lookup.
* The reason we need this is that the blkio cgroup stats
* are exported using the major:minor numbers, and not the
* device names - we must perform that mapping ourselves.
* In some places (value refresh) we need to lookup the blk
* name from device major/minor, in other places (instances
* refresh) we need the usual external instid:name lookup.
*/
static void
refresh_cgroup_devices(void)
{
pmInDom diskindom = INDOM(DISK_INDOM);
pmInDom devtindom = INDOM(DEVT_INDOM);
char buf[MAXPATHLEN];
FILE *fp;
pmdaCacheOp(devtindom, PMDA_CACHE_INACTIVE);
pmdaCacheOp(diskindom, PMDA_CACHE_INACTIVE);
if ((fp = proc_statsfile("/proc/diskstats", buf, sizeof(buf))) == NULL)
return;
while (fgets(buf, sizeof(buf), fp) != NULL) {
unsigned int major, minor, unused;
device_t *dev = NULL;
char namebuf[1024];
int inst;
if (sscanf(buf, "%u %u %s %u", &major, &minor, namebuf, &unused) != 4)
continue;
if (_pm_isloop(namebuf) || _pm_isramdisk(namebuf))
continue;
if (pmdaCacheLookupName(diskindom, namebuf, &inst, (void **)&dev) < 0 ||
dev == NULL) {
if (!(dev = (device_t *)malloc(sizeof(device_t)))) {
__pmNoMem("device", sizeof(device_t), PM_RECOV_ERR);
continue;
}
dev->major = major;
dev->minor = minor;
}
/* keeping track of all fields (major/minor/inst/name) */
pmdaCacheStore(diskindom, PMDA_CACHE_ADD, namebuf, dev);
(void)pmdaCacheLookupName(diskindom, namebuf, &dev->inst, NULL);
(void)pmdaCacheLookup(diskindom, dev->inst, &dev->name, NULL);
snprintf(buf, sizeof(buf), "%u:%u", major, minor);
pmdaCacheStore(devtindom, PMDA_CACHE_ADD, buf, (void *)dev);
if (pmDebug & DBG_TRACE_APPL0)
fprintf(stderr, "refresh_devices: \"%s\" \"%d:%d\" inst=%d\n",
dev->name, dev->major, dev->minor, dev->inst);
}
fclose(fp);
}
/*
* This separate indom provides the addresses for all interfaces including
* aliases (e.g. eth0, eth0:0, eth0:1, etc) - this is what ifconfig does.
*/
void
clear_net_addr_indom(pmInDom indom)
{
net_addr_t *p;
int inst;
for (pmdaCacheOp(indom, PMDA_CACHE_WALK_REWIND);;) {
if ((inst = pmdaCacheOp(indom, PMDA_CACHE_WALK_NEXT)) < 0)
break;
if (!pmdaCacheLookup(indom, inst, NULL, (void **)&p) || !p)
continue;
p->has_inet = 0;
p->has_ipv6 = 0;
p->has_hw = 0;
}
pmdaCacheOp(indom, PMDA_CACHE_INACTIVE);
}
/*
* Primary driver interface - finds any/all mount points for a given
* cgroup subsystem and iteratively expands all of the cgroups below
* them. The setup callback inactivates each indoms contents, while
* the refresh callback is called once per cgroup (with path/name) -
* its role is to refresh the values for that one named cgroup.
*/
void
refresh_cgroups(const char *subsys, const char *container,
int length, cgroup_setup_t setup, cgroup_refresh_t refresh)
{
int sts;
filesys_t *fs;
pmInDom mounts = INDOM(CGROUP_MOUNTS_INDOM);
pmdaCacheOp(mounts, PMDA_CACHE_WALK_REWIND);
while ((sts = pmdaCacheOp(mounts, PMDA_CACHE_WALK_NEXT)) != -1) {
if (!pmdaCacheLookup(mounts, sts, NULL, (void **)&fs))
continue;
if (scan_filesys_options(fs->options, subsys) == NULL)
continue;
setup();
cgroup_scan(fs->path, "", refresh, container, length);
}
}
int
proc_partitions_fetch(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
{
__pmID_int *idp = (__pmID_int *)&(mdesc->m_desc.pmid);
int i;
partitions_entry_t *p = NULL;
if (inst != PM_IN_NULL) {
if (pmdaCacheLookup(mdesc->m_desc.indom, inst, NULL, (void **)&p) < 0)
return PM_ERR_INST;
}
switch (idp->cluster) {
case CLUSTER_STAT:
/*
* disk.{dev,all} remain in CLUSTER_STAT for backward compatibility
*/
switch(idp->item) {
case 4: /* disk.dev.read */
if (p == NULL)
return PM_ERR_INST;
_pm_assign_ulong(atom, p->rd_ios);
break;
case 5: /* disk.dev.write */
if (p == NULL)
return PM_ERR_INST;
_pm_assign_ulong(atom, p->wr_ios);
break;
case 6: /* disk.dev.blkread */
if (p == NULL)
return PM_ERR_INST;
atom->ull = p->rd_sectors;
break;
case 7: /* disk.dev.blkwrite */
if (p == NULL)
return PM_ERR_INST;
atom->ull = p->wr_sectors;
break;
case 28: /* disk.dev.total */
if (p == NULL)
return PM_ERR_INST;
atom->ull = p->rd_ios + p->wr_ios;
break;
case 36: /* disk.dev.blktotal */
if (p == NULL)
return PM_ERR_INST;
atom->ull = p->rd_sectors + p->wr_sectors;
break;
case 38: /* disk.dev.read_bytes */
if (p == NULL)
return PM_ERR_INST;
atom->ull = p->rd_sectors / 2;
break;
case 39: /* disk.dev.write_bytes */
if (p == NULL)
return PM_ERR_INST;
atom->ull = p->wr_sectors / 2;
break;
case 40: /* disk.dev.total_bytes */
if (p == NULL)
return PM_ERR_INST;
atom->ull = (p->rd_sectors + p->wr_sectors) / 2;
break;
case 46: /* disk.dev.avactive ... already msec from /proc/diskstats */
if (p == NULL)
return PM_ERR_INST;
atom->ul = p->io_ticks;
break;
case 47: /* disk.dev.aveq ... already msec from /proc/diskstats */
if (p == NULL)
return PM_ERR_INST;
atom->ul = p->aveq;
break;
case 49: /* disk.dev.read_merge */
if (p == NULL)
return PM_ERR_INST;
_pm_assign_ulong(atom, p->rd_merges);
break;
case 50: /* disk.dev.write_merge */
if (p == NULL)
return PM_ERR_INST;
_pm_assign_ulong(atom, p->wr_merges);
break;
case 59: /* disk.dev.scheduler */
if (p == NULL)
return PM_ERR_INST;
atom->cp = _pm_ioscheduler(p->namebuf);
break;
case 72: /* disk.dev.read_rawactive already ms from /proc/diskstats */
if (p == NULL)
return PM_ERR_INST;
atom->ul = p->rd_ticks;
break;
case 73: /* disk.dev.write_rawactive already ms from /proc/diskstats */
if (p == NULL)
return PM_ERR_INST;
atom->ul = p->wr_ticks;
break;
default:
/* disk.all.* is a singular instance domain */
atom->ull = 0;
for (pmdaCacheOp(INDOM(DISK_INDOM), PMDA_CACHE_WALK_REWIND);;) {
if ((i = pmdaCacheOp(INDOM(DISK_INDOM), PMDA_CACHE_WALK_NEXT)) < 0)
break;
if (!pmdaCacheLookup(INDOM(DISK_INDOM), i, NULL, (void **)&p) || !p)
continue;
switch (idp->item) {
case 24: /* disk.all.read */
atom->ull += p->rd_ios;
break;
case 25: /* disk.all.write */
atom->ull += p->wr_ios;
break;
case 26: /* disk.all.blkread */
atom->ull += p->rd_sectors;
break;
case 27: /* disk.all.blkwrite */
atom->ull += p->wr_sectors;
break;
case 29: /* disk.all.total */
atom->ull += p->rd_ios + p->wr_ios;
break;
case 37: /* disk.all.blktotal */
atom->ull += p->rd_sectors + p->wr_sectors;
break;
case 41: /* disk.all.read_bytes */
atom->ull += p->rd_sectors / 2;
break;
case 42: /* disk.all.write_bytes */
atom->ull += p->wr_sectors / 2;
break;
case 43: /* disk.all.total_bytes */
atom->ull += (p->rd_sectors + p->wr_sectors) / 2;
break;
case 44: /* disk.all.avactive ... already msec from /proc/diskstats */
atom->ull += p->io_ticks;
break;
case 45: /* disk.all.aveq ... already msec from /proc/diskstats */
atom->ull += p->aveq;
break;
case 51: /* disk.all.read_merge */
atom->ull += p->rd_merges;
break;
case 52: /* disk.all.write_merge */
atom->ull += p->wr_merges;
break;
case 74: /* disk.all.read_rawactive ... already msec from /proc/diskstats */
atom->ull += p->rd_ticks;
break;
case 75: /* disk.all.write_rawactive ... already msec from /proc/diskstats */
atom->ull += p->wr_ticks;
break;
default:
return PM_ERR_PMID;
}
} /* loop */
}
break;
case CLUSTER_PARTITIONS:
if (p == NULL)
return PM_ERR_INST;
switch(idp->item) {
/* disk.partitions */
case 0: /* disk.partitions.read */
atom->ul = p->rd_ios;
break;
case 1: /* disk.partitions.write */
atom->ul = p->wr_ios;
break;
case 2: /* disk.partitions.total */
atom->ul = p->wr_ios +
p->rd_ios;
break;
case 3: /* disk.partitions.blkread */
atom->ul = p->rd_sectors;
break;
case 4: /* disk.partitions.blkwrite */
atom->ul = p->wr_sectors;
break;
case 5: /* disk.partitions.blktotal */
atom->ul = p->rd_sectors +
p->wr_sectors;
break;
case 6: /* disk.partitions.read_bytes */
atom->ul = p->rd_sectors / 2;
break;
case 7: /* disk.partitions.write_bytes */
atom->ul = p->wr_sectors / 2;
break;
case 8: /* disk.partitions.total_bytes */
atom->ul = (p->rd_sectors +
p->wr_sectors) / 2;
break;
default:
return PM_ERR_PMID;
}
break;
case CLUSTER_DM:
if (p == NULL)
return PM_ERR_INST;
switch(idp->item) {
case 0: /* disk.dm.read */
atom->ull = p->rd_ios;
break;
case 1: /* disk.dm.write */
atom->ull = p->wr_ios;
break;
case 2: /* disk.dm.total */
atom->ull = p->rd_ios + p->wr_ios;
break;
case 3: /* disk.dm.blkread */
atom->ull = p->rd_sectors;
break;
case 4: /* disk.dm.blkwrite */
atom->ull = p->wr_sectors;
break;
case 5: /* disk.dm.blktotal */
atom->ull = p->rd_sectors + p->wr_sectors;
break;
case 6: /* disk.dm.read_bytes */
atom->ull = p->rd_sectors / 2;
break;
case 7: /* disk.dm.write_bytes */
atom->ull = p->wr_sectors / 2;
break;
case 8: /* disk.dm.total_bytes */
atom->ull = (p->rd_sectors + p->wr_sectors) / 2;
break;
case 9: /* disk.dm.read_merge */
atom->ull = p->rd_merges;
break;
case 10: /* disk.dm.write_merge */
atom->ull = p->wr_merges;
break;
case 11: /* disk.dm.avactive */
atom->ull = p->io_ticks;
break;
case 12: /* disk.dm.aveq */
atom->ull = p->aveq;
break;
case 13: /* hinv.map.dmname */
atom->cp = p->dmname;
break;
case 14: /* disk.dm.read_rawactive */
atom->ul = p->rd_ticks;
break;
case 15: /* disk.dm.write_rawactive */
atom->ul = p->wr_ticks;
break;
default:
return PM_ERR_PMID;
}
break;
default: /* switch cluster */
return PM_ERR_PMID;
}
return 1;
}
int
pmdaInstance(pmInDom indom, int inst, char *name, __pmInResult **result, pmdaExt *pmda)
{
int i;
int namelen;
int err = 0;
__pmInResult *res;
pmdaIndom *idp = NULL; /* initialize to pander to gcc */
int have_cache = 0;
int myinst;
char *np;
if (pmdaCacheOp(indom, PMDA_CACHE_CHECK)) {
have_cache = 1;
}
else {
/*
* check this is an instance domain we know about -- code below
* assumes this test is complete
*/
for (i = 0; i < pmda->e_nindoms; i++) {
if (pmda->e_indoms[i].it_indom == indom)
break;
}
if (i >= pmda->e_nindoms)
return PM_ERR_INDOM;
idp = &pmda->e_indoms[i];
}
if ((res = (__pmInResult *)malloc(sizeof(*res))) == NULL)
return -oserror();
res->indom = indom;
if (name == NULL && inst == PM_IN_NULL)
res->numinst = __pmdaCntInst(indom, pmda);
else
res->numinst = 1;
if (inst == PM_IN_NULL) {
if ((res->instlist = (int *)malloc(res->numinst * sizeof(res->instlist[0]))) == NULL) {
free(res);
return -oserror();
}
}
else
res->instlist = NULL;
if (name == NULL) {
if ((res->namelist = (char **)malloc(res->numinst * sizeof(res->namelist[0]))) == NULL) {
__pmFreeInResult(res);
return -oserror();
}
for (i = 0; i < res->numinst; i++)
res->namelist[0] = NULL;
}
else
res->namelist = NULL;
if (name == NULL && inst == PM_IN_NULL) {
/* return inst and name for everything */
if (have_cache) {
pmdaCacheOp(indom, PMDA_CACHE_WALK_REWIND);
i = 0;
while (i < res->numinst && (myinst = pmdaCacheOp(indom, PMDA_CACHE_WALK_NEXT)) != -1) {
if (pmdaCacheLookup(indom, myinst, &np, NULL) != PMDA_CACHE_ACTIVE)
continue;
res->instlist[i] = myinst;
if ((res->namelist[i++] = strdup(np)) == NULL) {
__pmFreeInResult(res);
return -oserror();
}
}
}
else {
for (i = 0; i < res->numinst; i++) {
res->instlist[i] = idp->it_set[i].i_inst;
if ((res->namelist[i] = strdup(idp->it_set[i].i_name)) == NULL) {
__pmFreeInResult(res);
return -oserror();
}
}
}
}
else if (name == NULL) {
/* given an inst, return the name */
if (have_cache) {
if (pmdaCacheLookup(indom, inst, &np, NULL) == PMDA_CACHE_ACTIVE) {
if ((res->namelist[0] = strdup(np)) == NULL) {
__pmFreeInResult(res);
return -oserror();
}
}
else
err = 1;
}
else {
for (i = 0; i < idp->it_numinst; i++) {
if (inst == idp->it_set[i].i_inst) {
if ((res->namelist[0] = strdup(idp->it_set[i].i_name)) == NULL) {
__pmFreeInResult(res);
return -oserror();
}
break;
}
}
if (i == idp->it_numinst)
err = 1;
}
}
else if (inst == PM_IN_NULL && (namelen = (int)strlen(name)) > 0) {
if (have_cache) {
if (pmdaCacheLookupName(indom, name, &myinst, NULL) == PMDA_CACHE_ACTIVE)
res->instlist[0] = myinst;
else
err = 1;
}
else {
/* given a name, return an inst. If the name contains spaces,
* only exact matches are good enough for us, otherwise, we're
* prepared to accept a match upto the first space in the
* instance name on the assumption that pmdas will play by the
* rules and guarantee the first "word" in the instance name
* is unique. That allows for the things like "1 5 15" to match
* instances for kernel.all.load["1 minute","5 minute","15 minutes"]
*/
char * nspace = strchr (name, ' ');
for (i = 0; i < idp->it_numinst; i++) {
char *instname = idp->it_set[i].i_name;
if (strcmp(name, instname) == 0) {
/* accept an exact match */
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_LIBPMDA) {
fprintf(stderr,
"pmdaInstance: exact match name=%s id=%d\n",
name, idp->it_set[i].i_inst);
}
#endif
res->instlist[0] = idp->it_set[i].i_inst;
break;
}
else if (nspace == NULL) {
/* all of name must match instname up to the the first
* space in instname. */
char *p = strchr(instname, ' ');
if (p != NULL) {
int len = (int)(p - instname);
if (namelen == len && strncmp(name, instname, len) == 0) {
#ifdef PCP_DEBUG
if (pmDebug & DBG_TRACE_LIBPMDA) {
fprintf(stderr, "pmdaInstance: matched argument name=\"%s\" with indom id=%d name=\"%s\" len=%d\n",
name, idp->it_set[i].i_inst, instname, len);
}
#endif
res->instlist[0] = idp->it_set[i].i_inst;
break;
}
}
}
}
if (i == idp->it_numinst)
err = 1;
}
}
else
err = 1;
if (err == 1) {
/* bogus arguments or instance id/name */
__pmFreeInResult(res);
return PM_ERR_INST;
}
*result = res;
return 0;
}
int
refresh_proc_slabinfo(pmInDom slab_indom, proc_slabinfo_t *slabinfo)
{
slab_cache_t sbuf, *s;
char buf[BUFSIZ];
char name[128];
char *w, *p;
FILE *fp;
int i, sts = 0, indom_change = 0;
static int major_version = -1;
static int minor_version = 0;
for (pmdaCacheOp(slab_indom, PMDA_CACHE_WALK_REWIND);;) {
if ((i = pmdaCacheOp(slab_indom, PMDA_CACHE_WALK_NEXT)) < 0)
break;
if (!pmdaCacheLookup(slab_indom, i, NULL, (void **)&s) || !s)
continue;
s->seen = 0;
}
pmdaCacheOp(slab_indom, PMDA_CACHE_INACTIVE);
if ((fp = linux_statsfile("/proc/slabinfo", buf, sizeof(buf))) == NULL)
return -oserror();
/* skip header */
if (fgets(buf, sizeof(buf), fp) == NULL) {
/* oops, no header! */
fclose(fp);
return -oserror();
}
if (major_version < 0) {
major_version = minor_version = 0;
if (strstr(buf, "slabinfo - version:")) {
for (p = buf; *p; p++) {
if (isdigit((int)*p)) {
sscanf(p, "%d.%d", &major_version, &minor_version);
break;
}
}
}
}
while (fgets(buf, sizeof(buf), fp) != NULL) {
/* try to convert whitespace in cache names to underscores, */
/* by looking for alphabetic chars which follow whitespace. */
if (buf[0] == '#')
continue;
for (w = NULL, p = buf; *p != '\0'; p++) {
if (isspace((int)*p))
w = p;
else if (isdigit((int)*p))
break;
else if (isalpha((int)*p) && w) {
for (; w && w != p; w++)
*w = '_';
w = NULL;
}
}
memset(&sbuf, 0, sizeof(slab_cache_t));
if (major_version == 1 && minor_version == 0) {
/*
* <name> <active_objs> <num_objs>
* (generally 2.2 kernels)
*/
i = sscanf(buf, "%s %lu %lu", name,
(unsigned long *)&sbuf.num_active_objs,
(unsigned long *)&sbuf.total_objs);
if (i != 3) {
sts = PM_ERR_APPVERSION;
break;
}
}
else if (major_version == 1 && minor_version == 1) {
/*
* <name> <active_objs> <num_objs> <objsize> <active_slabs> <num_slabs> <pagesperslab>
* (generally 2.4 kernels)
*/
i = sscanf(buf, "%s %lu %lu %u %u %u %u", name,
(unsigned long *)&sbuf.num_active_objs,
(unsigned long *)&sbuf.total_objs,
&sbuf.object_size,
&sbuf.num_active_slabs,
&sbuf.total_slabs,
&sbuf.pages_per_slab);
if (i != 7) {
sts = PM_ERR_APPVERSION;
break;
}
sbuf.total_size = sbuf.pages_per_slab * sbuf.num_active_slabs;
sbuf.total_size <<= _pm_pageshift;
}
else if (major_version == 2 && minor_version >= 0 && minor_version <= 1) {
/*
* <name> <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab> .. and more
* (generally for kernels up to at least 2.6.11)
*/
i = sscanf(buf, "%s %lu %lu %u %u %u", name,
(unsigned long *)&sbuf.num_active_objs,
(unsigned long *)&sbuf.total_objs,
&sbuf.object_size,
&sbuf.objects_per_slab,
&sbuf.pages_per_slab);
if (i != 6) {
sts = PM_ERR_APPVERSION;
break;
}
sbuf.total_size = sbuf.pages_per_slab * sbuf.num_active_objs;
sbuf.total_size <<= _pm_pageshift;
sbuf.total_size /= sbuf.objects_per_slab;
}
else {
/* no support */
sts = PM_ERR_APPVERSION;
break;
}
sts = pmdaCacheLookupName(slab_indom, name, &i, (void **)&s);
if (sts < 0 || !s) {
/* new cache has appeared */
if ((s = calloc(1, sizeof(*s))) == NULL)
continue;
#if PCP_DEBUG
if (pmDebug & DBG_TRACE_LIBPMDA)
fprintf(stderr, "refresh_slabinfo: added \"%s\"\n", name);
#endif
indom_change++;
}
s->num_active_objs = sbuf.num_active_objs;
s->total_objs = sbuf.total_objs;
s->object_size = sbuf.object_size;
s->num_active_slabs = sbuf.num_active_slabs;
s->total_slabs = sbuf.total_slabs;
s->pages_per_slab = sbuf.pages_per_slab;
s->objects_per_slab = sbuf.objects_per_slab;
s->total_size = sbuf.total_size;
s->seen = major_version * 10 + minor_version;
pmdaCacheStore(slab_indom, PMDA_CACHE_ADD, name, s);
}
fclose(fp);
if (indom_change)
pmdaCacheOp(slab_indom, PMDA_CACHE_SAVE);
return sts;
}
int
do_netif_metrics(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
{
struct if_msghdr *ifm;
int sts;
if (!valid)
return 0;
if (inst != PM_IN_NULL) {
/*
* per-network interface metrics
*/
sts = pmdaCacheLookup(indomtab[NETIF_INDOM].it_indom, inst, NULL, (void **)&ifm);
if (sts == PMDA_CACHE_ACTIVE) {
sts = 1;
/* cluster and domain already checked, just need item ... */
switch (pmid_item(mdesc->m_desc.pmid)) {
case 0: /* network.interface.mtu */
atom->ull = ifm->ifm_data.ifi_mtu;
break;
case 1: /* network.interface.up */
atom->ul = (ifm->ifm_flags & IFF_UP) == IFF_UP;
break;
case 2: /* network.interface.baudrate */
atom->ull = ifm->ifm_data.ifi_baudrate;
break;
case 3: /* network.interface.in.bytes */
atom->ull = ifm->ifm_data.ifi_ibytes;
break;
case 4: /* network.interface.in.packets */
atom->ull = ifm->ifm_data.ifi_ipackets;
break;
case 5: /* network.interface.in.mcasts */
atom->ull = ifm->ifm_data.ifi_imcasts;
break;
case 6: /* network.interface.in.errors */
atom->ull = ifm->ifm_data.ifi_ierrors;
break;
case 7: /* network.interface.in.drops */
atom->ull = ifm->ifm_data.ifi_iqdrops;
break;
case 8: /* network.interface.out.bytes */
atom->ull = ifm->ifm_data.ifi_obytes;
break;
case 9: /* network.interface.out.packets */
atom->ull = ifm->ifm_data.ifi_opackets;
break;
case 10: /* network.interface.out.mcasts */
atom->ull = ifm->ifm_data.ifi_omcasts;
break;
case 11: /* network.interface.out.errors */
atom->ull = ifm->ifm_data.ifi_oerrors;
break;
case 12: /* network.interface.out.collisions */
atom->ull = ifm->ifm_data.ifi_collisions;
break;
case 13: /* network.interface.total.bytes */
atom->ull = ifm->ifm_data.ifi_ibytes + ifm->ifm_data.ifi_obytes;
break;
case 14: /* network.interface.total.packets */
atom->ull = ifm->ifm_data.ifi_ipackets + ifm->ifm_data.ifi_opackets;
break;
case 15: /* network.interface.total.mcasts */
atom->ull = ifm->ifm_data.ifi_imcasts + ifm->ifm_data.ifi_omcasts;
break;
case 16: /* network.interface.total.errors */
atom->ull = ifm->ifm_data.ifi_ierrors + ifm->ifm_data.ifi_oerrors;
break;
default:
sts = PM_ERR_PMID;
break;
}
}
else
sts = 0;
}
else {
/*
* most network interface metrics don't have an instance domain
*
* cluster and domain already checked, just need item ...
*/
switch (pmid_item(mdesc->m_desc.pmid)) {
case 17: /* hinv.interface */
atom->ul = valid;
sts = 1;
break;
default:
sts = PM_ERR_INST;
break;
}
}
return sts;
}
/*
* We use /proc/stat as a single source of truth regarding online/offline
* state for CPUs (its per-CPU stats are for online CPUs only).
* This drives the contents of the CPU indom for all per-CPU metrics, so
* it is important to ensure this refresh routine is called first before
* refreshing any other per-CPU metrics (e.g. interrupts, softnet).
*/
int
refresh_proc_stat(proc_stat_t *proc_stat)
{
pernode_t *np;
percpu_t *cp;
pmInDom cpus, nodes;
char buf[MAXPATHLEN], *name;
int n = 0, i, size;
static int fd = -1; /* kept open until exit(), unless testing */
static char *statbuf;
static int maxstatbuf;
static char **bufindex;
static int nbufindex;
static int maxbufindex;
cpu_node_setup();
cpus = INDOM(CPU_INDOM);
pmdaCacheOp(cpus, PMDA_CACHE_INACTIVE);
nodes = INDOM(NODE_INDOM);
/* reset per-node aggregate CPU utilisation stats */
for (pmdaCacheOp(nodes, PMDA_CACHE_WALK_REWIND);;) {
if ((i = pmdaCacheOp(nodes, PMDA_CACHE_WALK_NEXT)) < 0)
break;
if (!pmdaCacheLookup(nodes, i, NULL, (void **)&np) || !np)
continue;
memset(&np->stat, 0, sizeof(np->stat));
}
/* in test mode we replace procfs files (keeping fd open thwarts that) */
if (fd >= 0 && (linux_test_mode & LINUX_TEST_STATSPATH)) {
close(fd);
fd = -1;
}
if (fd >= 0) {
if (lseek(fd, 0, SEEK_SET) < 0)
return -oserror();
} else {
snprintf(buf, sizeof(buf), "%s/proc/stat", linux_statspath);
if ((fd = open(buf, O_RDONLY)) < 0)
return -oserror();
}
for (;;) {
while (n >= maxstatbuf) {
size = maxstatbuf + 512;
if ((statbuf = (char *)realloc(statbuf, size)) == NULL)
return -ENOMEM;
maxstatbuf = size;
}
size = (statbuf + maxstatbuf) - (statbuf + n);
if ((i = read(fd, statbuf + n, size)) > 0)
n += i;
else
break;
}
statbuf[n] = '\0';
if (bufindex == NULL) {
size = 16 * sizeof(char *);
if ((bufindex = (char **)malloc(size)) == NULL)
return -ENOMEM;
maxbufindex = 16;
}
nbufindex = 0;
bufindex[nbufindex] = statbuf;
for (i = 0; i < n; i++) {
if (statbuf[i] == '\n' || statbuf[i] == '\0') {
statbuf[i] = '\0';
if (nbufindex + 1 >= maxbufindex) {
size = (maxbufindex + 4) * sizeof(char *);
if ((bufindex = (char **)realloc(bufindex, size)) == NULL)
return -ENOMEM;
maxbufindex += 4;
}
bufindex[++nbufindex] = statbuf + i + 1;
}
}
#define ALLCPU_FMT "cpu %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu"
n = sscanf((const char *)bufindex[0], ALLCPU_FMT,
&proc_stat->all.user, &proc_stat->all.nice,
&proc_stat->all.sys, &proc_stat->all.idle,
&proc_stat->all.wait, &proc_stat->all.irq,
&proc_stat->all.sirq, &proc_stat->all.steal,
&proc_stat->all.guest, &proc_stat->all.guest_nice);
#define PERCPU_FMT "cpu%u %llu %llu %llu %llu %llu %llu %llu %llu %llu %llu"
/*
* per-CPU stats
* e.g. cpu0 95379 4 20053 6502503
* 2.6 kernels have 3 additional fields for wait, irq and soft_irq.
* More recent (2008) 2.6 kernels have an extra field for guest and
* also (since 2009) guest_nice.
* In the single-CPU system case, don't bother scanning, use "all";
* this handles non-SMP kernels with no line starting with "cpu0".
*/
if ((size = pmdaCacheOp(cpus, PMDA_CACHE_SIZE)) == 1) {
pmdaCacheLookup(cpus, 0, &name, (void **)&cp);
memcpy(&cp->stat, &proc_stat->all, sizeof(cp->stat));
pmdaCacheStore(cpus, PMDA_CACHE_ADD, name, (void *)cp);
pmdaCacheLookup(nodes, 0, NULL, (void **)&np);
memcpy(&np->stat, &proc_stat->all, sizeof(np->stat));
}
else {
for (n = 0; n < nbufindex; n++) {
if (strncmp("cpu", bufindex[n], 3) != 0 ||
!isdigit((int)bufindex[n][3]))
continue;
cp = NULL;
np = NULL;
i = atoi(&bufindex[n][3]); /* extract CPU identifier */
if (pmdaCacheLookup(cpus, i, &name, (void **)&cp) < 0 || !cp)
continue;
memset(&cp->stat, 0, sizeof(cp->stat));
sscanf(bufindex[n], PERCPU_FMT, &i,
&cp->stat.user, &cp->stat.nice, &cp->stat.sys,
&cp->stat.idle, &cp->stat.wait, &cp->stat.irq,
&cp->stat.sirq, &cp->stat.steal, &cp->stat.guest,
&cp->stat.guest_nice);
pmdaCacheStore(cpus, PMDA_CACHE_ADD, name, (void *)cp);
/* update per-node aggregate CPU utilisation stats as well */
if (pmdaCacheLookup(nodes, cp->nodeid, NULL, (void **)&np) < 0)
continue;
np->stat.user += cp->stat.user;
np->stat.nice += cp->stat.nice;
np->stat.sys += cp->stat.sys;
np->stat.idle += cp->stat.idle;
np->stat.wait += cp->stat.wait;
np->stat.irq += cp->stat.irq;
np->stat.sirq += cp->stat.sirq;
np->stat.steal += cp->stat.steal;
np->stat.guest += cp->stat.guest;
np->stat.guest_nice += cp->stat.guest_nice;
}
}
i = size;
#define PAGE_FMT "page %u %u" /* NB: moved to /proc/vmstat in 2.6 kernels */
if ((i = find_line_format(PAGE_FMT, 5, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], PAGE_FMT,
&proc_stat->page[0], &proc_stat->page[1]);
#define SWAP_FMT "swap %u %u" /* NB: moved to /proc/vmstat in 2.6 kernels */
if ((i = find_line_format(SWAP_FMT, 5, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], SWAP_FMT,
&proc_stat->swap[0], &proc_stat->swap[1]);
#define INTR_FMT "intr %llu" /* (export 1st 'total interrupts' value only) */
if ((i = find_line_format(INTR_FMT, 5, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], INTR_FMT, &proc_stat->intr);
#define CTXT_FMT "ctxt %llu"
if ((i = find_line_format(CTXT_FMT, 5, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], CTXT_FMT, &proc_stat->ctxt);
#define BTIME_FMT "btime %lu"
if ((i = find_line_format(BTIME_FMT, 6, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], BTIME_FMT, &proc_stat->btime);
#define PROCESSES_FMT "processes %lu"
if ((i = find_line_format(PROCESSES_FMT, 10, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], PROCESSES_FMT, &proc_stat->processes);
#define RUNNING_FMT "procs_running %lu"
if ((i = find_line_format(RUNNING_FMT, 14, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], RUNNING_FMT, &proc_stat->procs_running);
#define BLOCKED_FMT "procs_blocked %lu"
if ((i = find_line_format(BLOCKED_FMT, 14, bufindex, nbufindex, i)) >= 0)
sscanf((const char *)bufindex[i], BLOCKED_FMT, &proc_stat->procs_blocked);
/* success */
return 0;
}