Solaris: condense separate process vs lwp handling down to a single workflow

solaris/SolarisProcessList.c

@@ -15,13 +15,11 @@ in the source distribution for its full text.
 #include <sys/types.h>
 #include <sys/user.h>
 #include <err.h>
-#include <fcntl.h>
 #include <limits.h>
 #include <string.h>
 #include <procfs.h>
 #include <errno.h>
 #include <pwd.h>
-#include <dirent.h>
 #include <math.h>
 #include <time.h>
 
@@ -238,139 +236,96 @@ void ProcessList_delete(ProcessList* this) {
    free(this);
 }
 
-void ProcessList_enumerateLWPs(Process* proc, ProcessList* pl, lwpsinfo_t *_lwpsinfo, struct timeval tv) {
-   Process *lwp;
-   SolarisProcess *slwp;
-   SolarisProcess *sproc = (SolarisProcess*) proc;
-   id_t lwpid_real; 
-   pid_t lwpid;
-   bool preExisting = false;   
+int SolarisProcessList_walkproc(psinfo_t *_psinfo, lwpsinfo_t *_lwpsinfo, void *listptr) {
+   struct timeval tv;
    struct tm date;
+   bool preExistingP = false;
+   bool preExistingL = false;
+   bool preExisting;
+   Process *cproc;
+   SolarisProcess *csproc;
 
-   // With 10 bits to spare, we can only list up to 1023 unique LWPs per process
-   lwpid_real = _lwpsinfo->pr_lwpid;
-   if (lwpid_real > 1023) return;
-   lwpid   = proc->pid + lwpid_real;
-   lwp     = ProcessList_getProcess(pl, lwpid, &preExisting, (Process_New) SolarisProcess_new);
-   slwp    = (SolarisProcess*) lwp;
+   // Setup process list
+   ProcessList *pl = (ProcessList*) listptr;
+   SolarisProcessList *spl = (SolarisProcessList*) listptr;
 
-   // Common items set for both new and refreshed LWPs
-   slwp->zoneid            = sproc->zoneid;
-   lwp->percent_cpu        = ((uint16_t)_lwpsinfo->pr_pctcpu/(double)32768)*(double)100.0;
-   lwp->pgrp               = proc->pgrp;
-   lwp->st_uid             = proc->st_uid;
-   lwp->user               = UsersTable_getRef(pl->usersTable, lwp->st_uid);
-   lwp->session            = proc->session;
-   lwp->comm               = xStrdup(proc->comm);
-   lwp->commLen            = strnlen(proc->comm,PRFNSZ);
-   slwp->zname             = sproc->zname;
-   lwp->tty_nr             = proc->tty_nr;
-   lwp->priority           = _lwpsinfo->pr_pri;
-   lwp->nice               = _lwpsinfo->pr_nice;
-   lwp->processor          = _lwpsinfo->pr_onpro;
-   lwp->state              = _lwpsinfo->pr_sname;
-   lwp->time               = _lwpsinfo->pr_time.tv_sec;
-   slwp->taskid            = sproc->taskid;
-   slwp->projid            = sproc->projid;
-   slwp->poolid            = sproc->poolid;
-   slwp->contid            = sproc->contid;
-   lwp->show               = false;
+   // Setup Process entry
+   Process *proc           = ProcessList_getProcess(pl, _psinfo->pr_pid * 1024, &preExistingP, (Process_New) SolarisProcess_new);
+   SolarisProcess *sproc   = (SolarisProcess*) proc;
 
-   // Tasks done only for NEW LWPs  
-   if (!preExisting) {
-      slwp->is_lwp         = true; 
-      lwp->basenameOffset  = -1;
-      slwp->kernel         = sproc->kernel;
-      // Fake values used for sorting
-      // Only set once because threads don't generally
-      // move... between... processes.
-      lwp->pid             = lwpid;
-      lwp->ppid            = proc->pid;
-      lwp->tgid            = proc->pid;
-      // Corresponding real values used for display
-      slwp->realpid        = sproc->realpid;
-      slwp->realppid       = sproc->realpid;
-      slwp->lwpid          = lwpid_real;
-      // Not tracked per thread
-      lwp->percent_mem     = (double)0.0;
-      lwp->nlwp            = 0;
-      lwp->m_resident      = 0;
-      lwp->m_size          = 0;
-      lwp->starttime_ctime = _lwpsinfo->pr_start.tv_sec;
-      (void) localtime_r((time_t*) &lwp->starttime_ctime, &date);
-      strftime(lwp->starttime_show, 7, ((lwp->starttime_ctime > tv.tv_sec - 86400) ? "%R " : "%b%d "), &date);
-      ProcessList_add(pl, lwp);
+   // Setup LWP entry
+   id_t lwpid_real = _lwpsinfo->pr_lwpid;
+   if (lwpid_real > 1023) return 0;
+   pid_t lwpid   = proc->pid + lwpid_real;
+   Process *lwp            = ProcessList_getProcess(pl, lwpid, &preExistingL, (Process_New) SolarisProcess_new);
+   SolarisProcess *slwp    = (SolarisProcess*) lwp;
+
+   bool onMasterLWP = (_lwpsinfo->pr_lwpid == _psinfo->pr_lwp.pr_lwpid);
+
+   // Determine whether we're updating proc info or LWP info
+   // based on whether or not we're on the representative LWP
+   // for a given proc
+   if (onMasterLWP) {
+      cproc = proc;
+      csproc = sproc;
+      preExisting = preExistingP;
+   } else {
+      cproc = lwp;
+      csproc = slwp;
+      preExisting = preExistingL;
    }
 
-   // Top-level process only gets this for the representative LWP
-   if (lwp->state == 'O') proc->state = 'O';
-   if (slwp->kernel  && !pl->settings->hideKernelThreads)   lwp->show = true;
-   if (!slwp->kernel && !pl->settings->hideUserlandThreads) lwp->show = true;
-   lwp->updated = true;
+   gettimeofday(&tv, NULL);
+
+   // Common code pass 1
+   cproc->show              = false;
+   csproc->zoneid           = _psinfo->pr_zoneid;
+   csproc->zname            = SolarisProcessList_readZoneName(spl->kd,sproc);
+   csproc->taskid           = _psinfo->pr_taskid;
+   csproc->projid           = _psinfo->pr_projid;
+   csproc->poolid           = _psinfo->pr_poolid;
+   csproc->contid           = _psinfo->pr_contract;
+   cproc->priority          = _lwpsinfo->pr_pri;
+   cproc->nice              = _lwpsinfo->pr_nice;
+   cproc->processor         = _lwpsinfo->pr_onpro;
+   cproc->state             = _lwpsinfo->pr_sname;
+   // This could potentially get bungled if the master LWP is not the first
+   // one enumerated.  Unaware of any workaround right now.
+   if ((cproc->state == 'O') && !onMasterLWP) proc->state = 'O';
+   // NOTE: This 'percentage' is a 16-bit BINARY FRACTIONS where 1.0 = 0x8000
+   // Source: https://docs.oracle.com/cd/E19253-01/816-5174/proc-4/index.html
+   // (accessed on 18 November 2017)
+   cproc->percent_mem       = ((uint16_t)_psinfo->pr_pctmem/(double)32768)*(double)100.0;
+   cproc->st_uid            = _psinfo->pr_euid;
+   cproc->user              = UsersTable_getRef(pl->usersTable, proc->st_uid);
+   cproc->pgrp              = _psinfo->pr_pgid;
+   cproc->nlwp              = _psinfo->pr_nlwp;
+   cproc->comm              = xStrdup(_psinfo->pr_fname);
+   cproc->commLen           = strnlen(_psinfo->pr_fname,PRFNSZ);
+   cproc->tty_nr            = _psinfo->pr_ttydev;
+   cproc->m_resident        = _psinfo->pr_rssize/PAGE_SIZE_KB;
+   cproc->m_size            = _psinfo->pr_size/PAGE_SIZE_KB;
 
-   return;
-}
+   if (!preExisting) {
+      csproc->realpid          = _psinfo->pr_pid;
+      csproc->lwpid            = lwpid_real;
+   }
 
-int SolarisProcessList_walkproc(psinfo_t *_psinfo, lwpsinfo_t *_lwpsinfo, void *listptr) {
-   ProcessList *pl = (ProcessList*) listptr;
-   SolarisProcessList *spl = (SolarisProcessList*) listptr;
-   bool preExisting = false;
-   Process *proc = ProcessList_getProcess(pl, _psinfo->pr_pid * 1024, &preExisting, (Process_New) SolarisProcess_new);
-   SolarisProcess *sproc = (SolarisProcess*) proc;
-   struct timeval tv;
-   struct tm date;
+   // End common code pass 1
 
-   // Are we on the representative LWP?
-   if (_lwpsinfo->pr_lwpid == _psinfo->pr_lwp.pr_lwpid) { 
-      // Common items set for both new and refreshed processes
+   if (onMasterLWP) { // Are we on the representative LWP?
       proc->ppid            = (_psinfo->pr_ppid * 1024);
       proc->tgid            = (_psinfo->pr_ppid * 1024);
       sproc->realppid       = _psinfo->pr_ppid;
-      sproc->zoneid         = _psinfo->pr_zoneid;
-      sproc->zname          = SolarisProcessList_readZoneName(spl->kd,sproc);
-      // NOTE: These 'percentages' are 16-bit BINARY FRACTIONS where 1.0 = 0x8000
-      // Source: https://docs.oracle.com/cd/E19253-01/816-5174/proc-4/index.html
-      // (accessed on 18 November 2017)
+      // See note above (in common section) about this BINARY FRACTION
       proc->percent_cpu     = ((uint16_t)_psinfo->pr_pctcpu/(double)32768)*(double)100.0;
-      proc->percent_mem     = ((uint16_t)_psinfo->pr_pctmem/(double)32768)*(double)100.0;
-      proc->st_uid          = _psinfo->pr_euid;
-      proc->user            = UsersTable_getRef(pl->usersTable, proc->st_uid);
-      proc->pgrp            = _psinfo->pr_pgid;
-      proc->nlwp            = _psinfo->pr_nlwp;
-      proc->comm            = xStrdup(_psinfo->pr_fname);
-      proc->commLen         = strnlen(_psinfo->pr_fname,PRFNSZ);
-      proc->tty_nr          = _psinfo->pr_ttydev;
-      proc->m_resident      = _psinfo->pr_rssize/PAGE_SIZE_KB;
-      proc->m_size          = _psinfo->pr_size/PAGE_SIZE_KB;
-      proc->priority        = _psinfo->pr_lwp.pr_pri;
-      proc->nice            = _psinfo->pr_lwp.pr_nice;
-      proc->processor       = _psinfo->pr_lwp.pr_onpro;
-      proc->state           = _psinfo->pr_lwp.pr_sname;
       proc->time            = _psinfo->pr_time.tv_sec;
-      sproc->taskid         = _psinfo->pr_taskid;
-      sproc->projid         = _psinfo->pr_projid;
-      sproc->poolid         = _psinfo->pr_poolid;
-      sproc->contid         = _psinfo->pr_contract; 
-
-      // Tasks done only for NEW processes
-      if(!preExisting) {
+      if(!preExisting) { // Tasks done only for NEW processes
          sproc->is_lwp = false;
-         // Fake PID values used for sorting, since Solaris LWPs lack unique PIDs
-         proc->pid             = (_psinfo->pr_pid * 1024);
-         // Corresponding real values used for display
-         sproc->realpid        = _psinfo->pr_pid;
-         sproc->lwpid          = 1;
          proc->starttime_ctime = _psinfo->pr_start.tv_sec;
-         if ((sproc->realppid <= 0) && !(sproc->realpid <= 1)) {
-            sproc->kernel = true;
-         } else {
-            sproc->kernel = false;
-         }
-         (void) localtime_r((time_t*) &proc->starttime_ctime, &date);
-         strftime(proc->starttime_show, 7, ((proc->starttime_ctime > tv.tv_sec - 86400) ? "%R " : "%b%d "), &date);
-         ProcessList_add(pl, proc);
       }
 
+      // Update proc and thread counts based on settings
       if (sproc->kernel && !pl->settings->hideKernelThreads) {
          pl->kernelThreads += proc->nlwp;
          pl->totalTasks += proc->nlwp+1;
@@ -385,11 +340,39 @@ int SolarisProcessList_walkproc(psinfo_t *_psinfo, lwpsinfo_t *_lwpsinfo, void *
          }
       }
       proc->show = !(pl->settings->hideKernelThreads && sproc->kernel);
-      proc->updated = true;
+   } else { // We are not in the master LWP, so jump to the LWP handling code
+      lwp->percent_cpu        = ((uint16_t)_lwpsinfo->pr_pctcpu/(double)32768)*(double)100.0;
+      lwp->time               = _lwpsinfo->pr_time.tv_sec;
+      if (!preExisting) { // Tasks done only for NEW LWPs
+         slwp->is_lwp         = true; 
+         lwp->basenameOffset  = -1;
+         lwp->ppid            = proc->pid;
+         lwp->tgid            = proc->pid;
+         slwp->realppid       = sproc->realpid;
+         lwp->starttime_ctime = _lwpsinfo->pr_start.tv_sec;
+      }
+
+      // Top-level process only gets this for the representative LWP
+      if (slwp->kernel  && !pl->settings->hideKernelThreads)   lwp->show = true;
+      if (!slwp->kernel && !pl->settings->hideUserlandThreads) lwp->show = true;
+   } // Top-level LWP or subordinate LWP
+
+   // Common code pass 2
+
+   if (!preExisting) {
+      if ((sproc->realppid <= 0) && !(sproc->realpid <= 1)) {
+         csproc->kernel = true;
       } else {
-      // We are not in the rep. LWP, so jump to the LWP handling code
-      ProcessList_enumerateLWPs(proc, pl, _lwpsinfo, tv);
+         csproc->kernel = false;
       }
+      (void) localtime_r((time_t*) &cproc->starttime_ctime, &date);
+      strftime(cproc->starttime_show, 7, ((cproc->starttime_ctime > tv.tv_sec - 86400) ? "%R " : "%b%d "), &date);
+      ProcessList_add(pl, cproc);
+   }
+   cproc->updated = true;
+
+   // End common code pass 2
+
    return 0;
 }
 
@@ -397,6 +380,6 @@ void ProcessList_goThroughEntries(ProcessList* this) {
    SolarisProcessList_scanCPUTime(this);
    SolarisProcessList_scanMemoryInfo(this);
    this->kernelThreads = 1;
-   proc_walk( &SolarisProcessList_walkproc, this, PR_WALK_LWP );
+   proc_walk(&SolarisProcessList_walkproc, this, PR_WALK_LWP);
 }
 

solaris/SolarisProcessList.h

@@ -50,8 +50,6 @@ ProcessList* ProcessList_new(UsersTable* usersTable, Hashtable* pidWhiteList, ui
 
 void ProcessList_delete(ProcessList* this);
 
-void ProcessList_enumerateLWPs(Process* proc, ProcessList* pl, lwpsinfo_t *_lwpsinfo, struct timeval tv);
-
 int SolarisProcessList_walkproc(psinfo_t *_psinfo, lwpsinfo_t *_lwpsinfo, void *listptr);
 
 void ProcessList_goThroughEntries(ProcessList* this);