/* * vmstat_solaris2.c * UCD SNMP module for systemStats section of UCD-SNMP-MIB for SunOS/Solaris * Jochen Kmietsch * with fixes and additions from the UCD-SNMP community * Uses some ideas from xosview and top * Some comments paraphrased from the SUN man pages * Version 0.1 initial release (Dec 1999) * Version 0.2 added support for multiprocessor machines (Jan 2000) * Version 0.3 some reliability enhancements and compile time fixes (Feb 2000) * Version 0.4 portability issue and raw cpu value support (Jun 2000) * Version 0.5 64-bit Solaris support and new data gathering routine (Aug 2000) * Version 0.6 Memory savings, overroll precautions and lint checks (Aug 2000) * Version 0.7 More raw counters and some cosmetic changes (Jan 2001) * */ /* * To make lint skip the debug code and stop complaining */ #ifdef __lint #define SNMP_NO_DEBUGGING 1 #endif /* * Includes start here */ /* * Standard includes */ #include #include #include #include #include /* * kstat and sysinfo structs */ #include #include #include #include #include #include "mibdefs.h" #include "util_funcs.h" /* * Header file for this module */ #include "vmstat.h" #include "vmstat_solaris2.h" /* * Includes end here */ /* * Global structures start here */ /* * A structure to save data gathered from the kernel kstat interface to. * We used to have the sys/sysinfo.h cpu_stat_t here but we did not need * all of it, some in a different size and some additional ones so we build * our own */ struct cpu_stat_snapshot { hrtime_t css_time; unsigned int css_cpus; unsigned long long css_swapin; unsigned long long css_swapout; unsigned long long css_blocks_read; unsigned long long css_blocks_write; unsigned long long css_interrupts; unsigned long long css_context_sw; unsigned long long css_cpu[CPU_STATES]; }; /* * Global structures end here */ /* * Global variables start here */ /* * From kstat.h: * Provides access to the kernel statistics library by * initializing a kstat control structure and returning a pointer * to this structure. This pointer must be used as the kc argument in * following function calls from libkstat (here kc is called kstat_fd). * Pointer to structure to be opened with kstat_open in main procedure. * We share this one with memory_solaris2 and kernel_sunos5, where it's * defined. */ extern kstat_ctl_t *kstat_fd; /* * Variables for the calculated values, filled in update_stats * Need to be global since we need them in more than one function */ static ulong swapin; static ulong swapout; static ulong blocks_read; static ulong blocks_write; static ulong interrupts; static ulong context_sw; /* * Since MIB wants CPU_SYSTEM, which is CPU_KERNEL + CPU_WAIT */ static long cpu_perc[CPU_STATES + 1]; /* * How many snapshots we have already taken, needed for the first * POLL_INTERVAL * POLL_VALUES seconds of agent running */ static unsigned int number_of_snapshots; /* * The place to store the snapshots of system data in */ static struct cpu_stat_snapshot snapshot[POLL_VALUES + 1]; /* * And one for the raw counters, which we fill when the raw values are * requested, as opposed to the absolute values, which are taken every * POLL_INTERVAL seconds and calculated over POLL_INTERVAL * POLL_VALUES time */ static struct cpu_stat_snapshot raw_values; /* * Global variables end here */ /* * Functions start here */ /* * Function prototype */ static void update_stats(unsigned int registrationNumber, void *clientarg); static int take_snapshot(struct cpu_stat_snapshot *css); /* * init_vmstat_solaris2 starts here * Init function for this module, from prototype * Defines variables handled by this module, defines root OID for * this module and registers it with the agent */ FindVarMethod var_extensible_vmstat; void init_vmstat_solaris2(void) { /* * Which variables do we service ? */ struct variable2 extensible_vmstat_variables[] = { {MIBINDEX, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {MIBINDEX}}, {ERRORNAME, ASN_OCTET_STR, RONLY, var_extensible_vmstat, 1, {ERRORNAME}}, {SWAPIN, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {SWAPIN}}, {SWAPOUT, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {SWAPOUT}}, {IOSENT, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {IOSENT}}, {IORECEIVE, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {IORECEIVE}}, {SYSINTERRUPTS, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {SYSINTERRUPTS}}, {SYSCONTEXT, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {SYSCONTEXT}}, {CPUUSER, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {CPUUSER}}, {CPUSYSTEM, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {CPUSYSTEM}}, {CPUIDLE, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {CPUIDLE}}, {CPURAWUSER, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {CPURAWUSER}}, {CPURAWSYSTEM, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {CPURAWSYSTEM}}, {CPURAWIDLE, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {CPURAWIDLE}}, {CPURAWWAIT, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {CPURAWWAIT}}, {CPURAWKERNEL, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {CPURAWKERNEL}}, {IORAWSENT, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {IORAWSENT}}, {IORAWRECEIVE, ASN_COUNTER, RONLY, var_extensible_vmstat, 1, {IORAWRECEIVE}}, /* * Future use: * {ERRORFLAG, ASN_INTEGER, RONLY, var_extensible_vmstat, 1, {ERRORFLAG }}, * {ERRORMSG, ASN_OCTET_STR, RONLY, var_extensible_vmstat, 1, {ERRORMSG }} */ }; /* * Define the OID pointer to the top of the mib tree that we're * registering underneath */ oid vmstat_variables_oid[] = { UCDAVIS_MIB, 11 }; /* * register ourselves with the agent to handle our mib tree * LINTED Trust me, I know what I'm doing */ REGISTER_MIB("ucd-snmp/vmstat", extensible_vmstat_variables, variable2, vmstat_variables_oid); /* * First check whether shared kstat contol is NULL, if so, try to open our * own. */ if (kstat_fd == NULL) { kstat_fd = kstat_open(); } /* * Then check whether either shared kstat was found or we succeeded in * opening our own. */ if (kstat_fd == NULL) { snmp_log(LOG_ERR, "vmstat_solaris2 (init): kstat_open() failed and no shared kstat control found.\n"); } /* * Start with some useful data */ update_stats(0, NULL); /* * update_stats is run every POLL_INTERVAL seconds using this routine * (see 'man snmp_alarm') * This is only executed once to get some useful data in the beginning */ if (snmp_alarm_register(5, NULL, update_stats, NULL) == 0) { snmp_log(LOG_WARNING, "vmstat_solaris2 (init): snmp_alarm_register failed.\n"); } /* * This is the one that runs update_stats every POLL_INTERVAL seconds */ if (snmp_alarm_register(POLL_INTERVAL, SA_REPEAT, update_stats, NULL) == 0) { snmp_log(LOG_ERR, "vmstat_solaris2 (init): snmp_alarm_register failed, cannot service requests.\n"); } } /* init_vmstat_solaris2 ends here */ /* * Data collection function take_snapshot starts here * Get data from kernel and save into the snapshot strutcs * Argument is the snapshot struct to save to. Global anyway, but looks nicer */ static int take_snapshot(struct cpu_stat_snapshot *css) { /* * Variables start here */ /* * From sys/kstat.h (included from kstat.h): * Pointer to current kstat */ kstat_t *ksp; /* * Counters */ unsigned int cpu_num = 0; /* * High resolution time counter */ hrtime_t current_time; /* * see sys/sysinfo.h, holds CPU data */ cpu_stat_t cs; /* * The usual stuff to count on, err, by */ int i; /* * Variables end here */ /* * Function starts here */ /* * Get time */ current_time = gethrtime(); /* * If we have just gotten the data, return the values from last run (skip if-clause) * This happens on a snmpwalk request. No need to read the kstat again * if we just did it less than 2 seconds ago * Jumps into if-clause either when snapshot is empty or when too old */ if ((css->css_time == 0) || (current_time > css->css_time + 2000000000)) { /* * Make sure we clean up before we put new data into snapshot */ memset(css, 0, sizeof *css); /* * Update timer */ css->css_time = current_time; /* * Look thru all the cpu slots on the machine whether they holds a CPU * and if so, get the data from that CPU * We walk through the whole kstat chain and sum up all the found cpu_stat kstats, * there's one for every CPU in a machine */ for (ksp = kstat_fd->kc_chain; ksp != NULL; ksp = ksp->ks_next) { /* * If we encounter an invalid kstat, skip it and continue with next one */ if (ksp->ks_flags & KSTAT_FLAG_INVALID) { continue; } if (strcmp(ksp->ks_module, "cpu_stat") == 0) { /* * Yeah, we found a CPU. */ cpu_num++; /* * Read data from kstat into cs structure * kstat_fd is the control structure, ksp the kstat we are reading * and cs the buffer we are writing to. */ if ((ksp->ks_type != KSTAT_TYPE_RAW) || (ksp->ks_data_size != sizeof cs) || (kstat_read(kstat_fd, ksp, &cs) == -1)) { snmp_log(LOG_ERR, "vmstat_solaris2 (take_snapshot): could not read cs structure.\n"); return (-1); } /* * Get the data from the cs structure and sum it up in our own structure */ css->css_swapin += (unsigned long long) cs.cpu_vminfo.swapin; css->css_swapout += (unsigned long long) cs.cpu_vminfo.swapout; css->css_blocks_read += (unsigned long long) cs.cpu_sysinfo.bread; css->css_blocks_write += (unsigned long long) cs.cpu_sysinfo.bwrite; css->css_interrupts += (unsigned long long) cs.cpu_sysinfo.intr; css->css_context_sw += (unsigned long long) cs.cpu_sysinfo.pswitch; /* * We need a for-loop for the CPU values */ cs.cpu_sysinfo.cpu[CPU_WAIT] = cs.cpu_sysinfo.wait[W_IO] + cs.cpu_sysinfo.wait[W_PIO]; for (i = 0; i < CPU_STATES; i++) { css->css_cpu[i] += (unsigned long long) cs.cpu_sysinfo.cpu[i]; } /* end for */ } /* end if */ } /* end for */ /* * Increment number of CPUs we gathered data from, for future use */ css->css_cpus = cpu_num; } /* * All engines running at warp speed, no problems (if there are any engines, that is) */ return (cpu_num > 0 ? 0 : -1); } /* take_snapshot ends here */ /* * This gets called every POLL_INTERVAL seconds to update the snapshots. It takes a new snapshot and * drops the oldest one. This way we move the time window so we always take the values over * POLL_INTERVAL * POLL_VALUES seconds and update the data used every POLL_INTERVAL seconds * The alarm timer is in the init function of this module (snmp_alarm_register) * ARGSUSED0 */ static void update_stats(unsigned int registrationNumber, void *clientarg) { /* * The time between the samples we compare */ hrtime_t time_diff; /* * Easier to use these than the snapshots, short hand pointers */ struct cpu_stat_snapshot *css_old, *css_new; /* * The usual stuff to count on, err, by */ int i; /* * Kstat chain id, to check whether kstat chain changed */ kid_t kid; /* * The sum of the CPU ticks that have passed on the different CPU states, so we can calculate * the percentages of each state */ unsigned long long cpu_sum = 0; DEBUGMSGTL(("ucd-snmp/vmstat_solaris2.c:update_stats", "updating stats\n")); /* * Just in case someone added (or removed) some CPUs during operation (or other kstat chain changes) */ kid = kstat_chain_update(kstat_fd); if (kid != 0) { if (kid == -1) { snmp_log(LOG_WARNING, "vmstat_solaris2 (update_stats): Could not update kstat chain.\n"); } else { /* * On some machines this floods the logfile, thus commented out * snmp_log(LOG_INFO, "vmstat_solaris2 (update_stats): Kstat chain changed."); */ } } /* * Take the current snapshot */ if (take_snapshot(&snapshot[0]) == -1) { snmp_log(LOG_WARNING, "vmstat_solaris2 (update_stats): Something went wrong with take_snapshot.\n"); return; } /* * Do we have some data we can use ? An issue right after the start of the agent */ if (number_of_snapshots > 0) { /* * Huh, the number of CPUs changed during run time. That is indeed s.th. worth noting, we * output a humorous (more or less) syslog message and need to retake the snapshots */ if (snapshot[0].css_cpus != snapshot[1].css_cpus) { if (snapshot[0].css_cpus > snapshot[1].css_cpus) { snmp_log(LOG_NOTICE, "vmstat_solaris2 (update_stats): Cool ! Number of CPUs increased, must be hot-pluggable.\n"); } else { snmp_log(LOG_NOTICE, "vmstat_solaris2 (update_stats): Lost at least one CPU, RIP.\n"); } /* * Make all snapshots but the current one invalid */ number_of_snapshots = 1; /* * Move the current one in the "first" [1] slot */ memmove(&snapshot[1], &snapshot[0], sizeof snapshot[0]); /* * Erase the current one */ memset(&snapshot[0], 0, sizeof snapshot[0]); /* * Try to get a new snapshot in five seconds so we can return s.th. useful */ if (snmp_alarm_register(5, NULL, update_stats, NULL) == 0) { snmp_log(LOG_WARNING, "vmstat_solaris2 (update_stats): snmp_alarm_register failed.\n"); } return; } /* * Short hand pointers */ css_new = &snapshot[0]; css_old = &snapshot[number_of_snapshots]; /* * How much time has passed between the snapshots we get the values from ? * Time is in nanoseconds so a few zeros here to juggle with * But the hrtime is not subject to change (s.b. setting the clock), unlike the normal time */ time_diff = (snapshot[0].css_time - snapshot[number_of_snapshots].css_time) / 1000000; DEBUGMSGTL(("ucd-snmp/vmstat_solaris2.c:update_stats", "time_diff: %lld\n", time_diff)); /* * swapin and swapout are in pages, MIB wants kB/s,so we just need to get kB and seconds * For the others we need to get value per second * getpagesize() returns pagesize in bytes * decided to use sysconf(_SC_PAGESIZE) instead to get around an #ifndef (I don't like those) * that was needed b/c some old Solaris versions don't have getpagesize() */ /* * LINTED cast needed, really */ swapin = (uint_t) ((css_new->css_swapin - css_old->css_swapin) * (hrtime_t) 1000 * sysconf(_SC_PAGESIZE) / 1024 / time_diff); /* * LINTED cast needed, really */ swapout = (uint_t) ((css_new->css_swapout - css_old->css_swapout) * (hrtime_t) 1000 * sysconf(_SC_PAGESIZE) / 1024 / time_diff); /* * LINTED cast needed, really */ blocks_read = (uint_t) ((css_new->css_blocks_read - css_old->css_blocks_read) * (hrtime_t) 1000 / time_diff); /* * LINTED cast needed, really */ blocks_write = (uint_t) ((css_new->css_blocks_write - css_old->css_blocks_write) * (hrtime_t) 1000 / time_diff); /* * LINTED cast needed, really */ interrupts = (uint_t) ((css_new->css_interrupts - css_old->css_interrupts) * (hrtime_t) 1000 / time_diff); /* * LINTED cast needed, really */ context_sw = (uint_t) ((css_new->css_context_sw - css_old->css_context_sw) * (hrtime_t) 1000 / time_diff); /* * Loop thru all the CPU_STATES and get the differences */ for (i = 0; i < CPU_STATES; i++) { cpu_sum += (css_new->css_cpu[i] - css_old->css_cpu[i]); } /* * Now calculate the absolute percentage values * Looks somewhat complicated sometimes but tries to get around using floats to increase speed */ for (i = 0; i < CPU_STATES; i++) { /* * Since we don't return fractions we use + 0.5 to get between 99 and 101 percent adding the values * together, otherwise we would get less than 100 most of the time */ /* * LINTED has to be 'long' */ cpu_perc[i] = (long) (((css_new->css_cpu[i] - css_old->css_cpu[i]) * 100 + (cpu_sum / 2)) / cpu_sum); } /* * As said before, MIB wants CPU_SYSTEM which is CPU_KERNEL + CPU_WAIT */ /* * LINTED has to be 'long' */ cpu_perc[CPU_SYSTEM] = (long) ((((css_new->css_cpu[CPU_KERNEL] - css_old->css_cpu[CPU_KERNEL]) + (css_new->css_cpu[CPU_WAIT] - css_old->css_cpu[CPU_WAIT])) * 100 + (cpu_sum / 2)) / cpu_sum); } /* * Make the current one the first one and move the whole thing one place down */ memmove(&snapshot[1], &snapshot[0], (size_t) (((char *) &snapshot[POLL_VALUES]) - ((char *) &snapshot[0]))); /* * Erase the current one */ memset(&snapshot[0], 0, sizeof snapshot[0]); /* * Only important on start up, we keep track of how many snapshots we have taken so far */ if (number_of_snapshots < POLL_VALUES) { number_of_snapshots++; } } /* update_stats ends here */ /* * *var_extensible_vmstat starts here * The guts of the module, this routine gets called to service a request */ unsigned char * var_extensible_vmstat(struct variable *vp, oid * name, size_t * length, int exact, size_t * var_len, WriteMethod ** write_method) { /* * Needed for returning the values */ static long long_ret; static char errmsg[300]; /* * set to 0 as default */ long_ret = 0; /* * generic check whether the options passed make sense and whether the * right variable is requested */ if (header_generic(vp, name, length, exact, var_len, write_method) != MATCH_SUCCEEDED) { return (NULL); } /* * The function that actually returns s.th. */ switch (vp->magic) { case MIBINDEX: long_ret = 1; return ((u_char *) (&long_ret)); case ERRORNAME: /* dummy name */ sprintf(errmsg, "systemStats"); *var_len = strlen(errmsg); return ((u_char *) (errmsg)); case SWAPIN: return ((u_char *) (&swapin)); case SWAPOUT: return ((u_char *) (&swapout)); case IOSENT: return ((u_char *) (&blocks_write)); case IORECEIVE: return ((u_char *) (&blocks_read)); case SYSINTERRUPTS: return ((u_char *) (&interrupts)); case SYSCONTEXT: return ((u_char *) (&context_sw)); case CPUUSER: return ((u_char *) (&cpu_perc[CPU_USER])); case CPUSYSTEM: return ((u_char *) (&cpu_perc[CPU_SYSTEM])); case CPUIDLE: return ((u_char *) (&cpu_perc[CPU_IDLE])); case CPURAWUSER: take_snapshot(&raw_values); /* * LINTED has to be 'long' */ long_ret = (long) (raw_values.css_cpu[CPU_USER] / raw_values.css_cpus); return ((u_char *) (&long_ret)); /* * We are missing CPURAWNICE, Solaris does not account for this in the kernel so this OID can not * be returned. Also, these values will roll over sooner or later and then return inaccurate data * but the MIB wants Integer32 so we cannot put a counter here * (Has been changed to Counter32 in the latest MIB version!) */ case CPURAWSYSTEM: take_snapshot(&raw_values); /* * LINTED has to be 'long' */ long_ret = (long) ((raw_values.css_cpu[CPU_KERNEL] + raw_values.css_cpu[CPU_WAIT]) / raw_values.css_cpus); return ((u_char *) (&long_ret)); case CPURAWIDLE: take_snapshot(&raw_values); /* * LINTED has to be 'long' */ long_ret = (long) (raw_values.css_cpu[CPU_IDLE] / raw_values.css_cpus); return ((u_char *) (&long_ret)); case CPURAWWAIT: take_snapshot(&raw_values); /* * LINTED has to be 'long' */ long_ret = (long) (raw_values.css_cpu[CPU_WAIT] / raw_values.css_cpus); return ((u_char *) (&long_ret)); case CPURAWKERNEL: take_snapshot(&raw_values); /* * LINTED has to be 'long' */ long_ret = (long) (raw_values.css_cpu[CPU_KERNEL] / raw_values.css_cpus); return ((u_char *) (&long_ret)); case IORAWSENT: long_ret = (long) (raw_values.css_blocks_write); return ((u_char *) (&long_ret)); case IORAWRECEIVE: long_ret = (long) (raw_values.css_blocks_read); return ((u_char *) (&long_ret)); /* * reserved for future use */ /* * case ERRORFLAG: * return((u_char *) (&long_ret)); * case ERRORMSG: * return((u_char *) (&long_ret)); */ default: snmp_log(LOG_ERR, "vmstat_solaris2: Error in request, no match for %d.\n", vp->magic); } return (NULL); } /* *var_extensible_vmstat ends here */ /* * Functions end here */ /* * Program ends here */