source: trunk/minix/commands/ps/ps.c@ 12

/* ps - print status                    Author: Peter Valkenburg */

/* Ps.c, Peter Valkenburg (valke@psy.vu.nl), january 1990.
 *
 * This is a V7 ps(1) look-alike for MINIX >= 1.5.0.
 * It does not support the 'k' option (i.e. cannot read memory from core file).
 * If you want to compile this for non-IBM PC architectures, the header files
 * require that you have your CHIP, MACHINE etc. defined.
 * Full syntax:
 *      ps [-][aeflx]
 * Option `a' gives all processes, `l' for detailed info, `x' includes even
 * processes without a terminal.
 * The `f' and `e' options were added by Kees Bot for the convenience of 
 * Solaris users accustomed to these options. The `e' option is equivalent to 
 * `a' and `f' is equivalent to  -l. These do not appear in the usage message.
 *
 * VERY IMPORTANT NOTE:
 *      To compile ps, the kernel/, fs/ and pm/ source directories must be in
 *      ../../ relative to the directory where ps is compiled (normally the
 *      tools source directory).
 *
 *      If you want your ps to be useable by anyone, you can arrange the
 *      following access permissions (note the protected memory files and set
 *      *group* id on ps):
 *      -rwxr-sr-x  1 bin   kmem       11916 Jul  4 15:31 /bin/ps
 *      crw-r-----  1 bin   kmem      1,   1 Jan  1  1970 /dev/mem
 *      crw-r-----  1 bin   kmem      1,   2 Jan  1  1970 /dev/kmem
 */

/* Some technical comments on this implementation:
 *
 * Most fields are similar to V7 ps(1), except for CPU, NICE, PRI which are
 * absent, RECV which replaces WCHAN, and PGRP that is an extra.
 * The info is obtained from the following fields of proc, mproc and fproc:
 * F    - kernel status field, p_rts_flags
 * S    - kernel status field, p_rts_flags; mm status field, mp_flags (R if p_rts_flags
 *        is 0; Z if mp_flags == ZOMBIE; T if mp_flags == STOPPED; else W).
 * UID  - mm eff uid field, mp_effuid
 * PID  - mm pid field, mp_pid
 * PPID - mm parent process index field, mp_parent (used as index in proc).
 * PGRP - mm process group field, mp_procgrp
 * SZ   - kernel text size + physical stack address - physical data address
 *                         + stack size
 *        p_memmap[T].mem_len + p_memmap[S].mem_phys - p_memmap[D].mem_phys
 *                         + p_memmap[S].mem_len
 * RECV - kernel process index field for message receiving, p_getfrom
 *        If sleeping, mm's mp_flags, or fs's fp_task are used for more info.
 * TTY  - fs controlling tty device field, fp_tty.
 * TIME - kernel user + system times fields, user_time + sys_time
 * CMD  - system process index (converted to mnemonic name by using the p_name
 *        field), or user process argument list (obtained by reading the stack
 *        frame; the resulting address is used to get the argument vector from
 *        user space and converted into a concatenated argument list).
 */

#include <minix/config.h>
#include <minix/endpoint.h>
#include <limits.h>
#include <timers.h>
#include <sys/types.h>

#include <minix/const.h>
#include <minix/type.h>
#include <minix/ipc.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>

#include <minix/com.h>
#include <fcntl.h>
#include <a.out.h>
#include <dirent.h>
#include <sys/stat.h>
#include <sys/ioctl.h>
#include <signal.h>
#include <stdio.h>
#include <ttyent.h>

#include "../../kernel/const.h"
#include "../../kernel/type.h"
#include "../../kernel/proc.h"

#include "../../servers/pm/mproc.h"
#include "../../servers/fs/fproc.h"
#include "../../servers/fs/const.h"


/*----- ps's local stuff below this line ------*/


#define mindev(dev)     (((dev)>>MINOR) & 0377) /* yield minor device */
#define majdev(dev)     (((dev)>>MAJOR) & 0377) /* yield major device */

#define TTY_MAJ         4       /* major device of console */

/* Structure for tty name info. */
typedef struct {
  char tty_name[NAME_MAX + 1];  /* file name in /dev */
  dev_t tty_dev;                /* major/minor pair */
} ttyinfo_t;

ttyinfo_t *ttyinfo;             /* ttyinfo holds actual tty info */
size_t n_ttyinfo;               /* Number of tty info slots */

/* Macro to convert memory offsets to rounded kilo-units */
#define off_to_k(off)   ((unsigned) (((off) + 512) / 1024))


/* Number of tasks and processes and addresses of the main process tables. */
int nr_tasks, nr_procs;         
vir_bytes proc_addr, mproc_addr, fproc_addr;    
extern int errno;

/* Process tables of the kernel, MM, and FS. */
struct proc *ps_proc;
struct mproc *ps_mproc;
struct fproc *ps_fproc;

/* Where is INIT? */
int init_proc_nr;
#define low_user init_proc_nr

#define KMEM_PATH       "/dev/kmem"     /* opened for kernel proc table */
#define MEM_PATH        "/dev/mem"      /* opened for pm/fs + user processes */

int kmemfd, memfd;              /* file descriptors of [k]mem */

/* Short and long listing formats:
 *
 *   PID TTY  TIME CMD
 * ppppp tttmmm:ss cccccccccc...
 *
 *   F S UID   PID  PPID  PGRP   SZ       RECV TTY  TIME CMD
 * fff s uuu ppppp ppppp ppppp ssss rrrrrrrrrr tttmmm:ss cccccccc...
 */
#define S_HEADER "  PID TTY  TIME CMD\n"
#define S_FORMAT "%5s %3s %s %s\n"
#define L_HEADER "  F S UID   PID  PPID  PGRP     SZ         RECV TTY  TIME CMD\n"
#define L_FORMAT "%3o %c %3d %5s %5d %5d %6d %12s %3s %s %s\n"


struct pstat {                  /* structure filled by pstat() */
  dev_t ps_dev;                 /* major/minor of controlling tty */
  uid_t ps_ruid;                /* real uid */
  uid_t ps_euid;                /* effective uid */
  pid_t ps_pid;                 /* process id */
  pid_t ps_ppid;                /* parent process id */
  int ps_pgrp;                  /* process group id */
  int ps_flags;                 /* kernel flags */
  int ps_mflags;                /* mm flags */
  int ps_ftask;                 /* (possibly pseudo) fs suspend task */
  char ps_state;                /* process state */
  vir_bytes ps_tsize;           /* text size (in bytes) */
  vir_bytes ps_dsize;           /* data size (in bytes) */
  vir_bytes ps_ssize;           /* stack size (in bytes) */
  phys_bytes ps_vtext;          /* virtual text offset */
  phys_bytes ps_vdata;          /* virtual data offset */
  phys_bytes ps_vstack;         /* virtual stack offset */
  phys_bytes ps_text;           /* physical text offset */
  phys_bytes ps_data;           /* physical data offset */
  phys_bytes ps_stack;          /* physical stack offset */
  int ps_recv;                  /* process number to receive from */
  time_t ps_utime;              /* accumulated user time */
  time_t ps_stime;              /* accumulated system time */
  char *ps_args;                /* concatenated argument string */
  vir_bytes ps_procargs;        /* initial stack frame from MM */
};

/* Ps_state field values in pstat struct above */
#define Z_STATE         'Z'     /* Zombie */
#define W_STATE         'W'     /* Waiting */
#define S_STATE         'S'     /* Sleeping */
#define R_STATE         'R'     /* Runnable */
#define T_STATE         'T'     /* stopped (Trace) */

_PROTOTYPE(char *tname, (Dev_t dev_nr ));
_PROTOTYPE(char *taskname, (int p_nr ));
_PROTOTYPE(char *prrecv, (struct pstat *bufp ));
_PROTOTYPE(void disaster, (int sig ));
_PROTOTYPE(int main, (int argc, char *argv []));
_PROTOTYPE(char *get_args, (struct pstat *bufp ));
_PROTOTYPE(int pstat, (int p_nr, struct pstat *bufp ));
_PROTOTYPE(int addrread, (int fd, phys_clicks base, vir_bytes addr, 
                                                    char *buf, int nbytes ));
_PROTOTYPE(void usage, (char *pname ));
_PROTOTYPE(void err, (char *s ));
_PROTOTYPE(int gettynames, (void));


/*
 * Tname returns mnemonic string for dev_nr. This is "?" for maj/min pairs that
 * are not found.  It uses the ttyinfo array (prepared by gettynames).
 * Tname assumes that the first three letters of the tty's name can be omitted
 * and returns the rest (except for the console, which yields "co").
 */
char *tname(dev_nr)
Dev_t dev_nr;
{
  int i;

  if (majdev(dev_nr) == TTY_MAJ && mindev(dev_nr) == 0) return "co";

  for (i = 0; i < n_ttyinfo && ttyinfo[i].tty_name[0] != '\0'; i++)
        if (ttyinfo[i].tty_dev == dev_nr)
                return ttyinfo[i].tty_name + 3;

  return "?";
}

/* Return canonical task name of task p_nr; overwritten on each call (yucch) */
char *taskname(p_nr)
int p_nr;
{
  return ps_proc[_ENDPOINT_P(p_nr) + nr_tasks].p_name;
}

/* Prrecv prints the RECV field for process with pstat buffer pointer bufp.
 * This is either "ANY", "taskname", or "(blockreason) taskname".
 */
char *prrecv(bufp)
struct pstat *bufp;
{
  char *blkstr, *task;          /* reason for blocking and task */
  static char recvstr[20];

  if (bufp->ps_recv == ANY) return "ANY";

  task = taskname(bufp->ps_recv);
  if (bufp->ps_state != S_STATE) return task;

  blkstr = "?";
  if (bufp->ps_recv == PM_PROC_NR) {
        if (bufp->ps_mflags & PAUSED)
                blkstr = "pause";
        else if (bufp->ps_mflags & WAITING)
                blkstr = "wait";
        else if (bufp->ps_mflags & SIGSUSPENDED)
                blkstr = "ssusp";
  } else if (bufp->ps_recv == FS_PROC_NR) {
        if (-bufp->ps_ftask == XPIPE)
                blkstr = "pipe";
        else if (-bufp->ps_ftask == XPOPEN)
                blkstr = "popen";
        else if (-bufp->ps_ftask == XLOCK)
                blkstr = "flock";
        else if(-bufp->ps_ftask == XSELECT)
                blkstr = "select";
        else if(-bufp->ps_ftask >= 0)
                blkstr = taskname(-bufp->ps_ftask);
        else
                blkstr = "??";
  }
  (void) sprintf(recvstr, "(%s) %s", blkstr, task);
  return recvstr;
}

/* If disaster is called some of the system parameters imported into ps are
 * probably wrong.  This tends to result in memory faults.
 */
void disaster(sig)
int sig;
{
  fprintf(stderr, "Ooops, got signal %d\n", sig);
  fprintf(stderr, "Was ps recompiled since the last kernel change?\n");
  exit(3);
}

/* Main interprets arguments, gets system addresses, opens [k]mem, reads in
 * process tables from kernel/pm/fs and calls pstat() for relevant entries.
 */
int main(argc, argv)
int argc;
char *argv[];
{
  int i;
  struct pstat buf;
  int db_fd;
  int uid = getuid();           /* real uid of caller */
  char *opt;
  int opt_all = FALSE;          /* -a */
  int opt_long = FALSE;         /* -l */
  int opt_notty = FALSE;        /* -x */
  char *ke_path;                /* paths of kernel, */
  char *mm_path;                /* mm, */
  char *fs_path;                /* and fs used in ps -U */
  char pid[2 + sizeof(pid_t) * 3];
  unsigned long ustime;
  char cpu[sizeof(clock_t) * 3 + 1 + 2];
  struct kinfo kinfo;
  int s;

  (void) signal(SIGSEGV, disaster);     /* catch a common crash */

  /* Parse arguments; a '-' need not be present (V7/BSD compatability) */
  for (i = 1; i < argc; i++) {
        opt = argv[i];
        if (opt[0] == '-') opt++;
        while (*opt != 0) switch (*opt++) {
                case 'a':       opt_all = TRUE;                 break;
                case 'e':       opt_all = opt_notty = TRUE;     break;
                case 'f':
                case 'l':       opt_long = TRUE;                break;
                case 'x':       opt_notty = TRUE;               break;
                default:        usage(argv[0]);
        }
  }

  /* Open memory devices and get PS info from the kernel */
  if ((kmemfd = open(KMEM_PATH, O_RDONLY)) == -1) err(KMEM_PATH);
  if ((memfd = open(MEM_PATH, O_RDONLY)) == -1) err(MEM_PATH);
  if (gettynames() == -1) err("Can't get tty names");

  getsysinfo(PM_PROC_NR, SI_PROC_ADDR, &mproc_addr);
  getsysinfo(FS_PROC_NR, SI_PROC_ADDR, &fproc_addr);
  getsysinfo(PM_PROC_NR, SI_KINFO, &kinfo);
  proc_addr = kinfo.proc_addr;
  nr_tasks = kinfo.nr_tasks;    
  nr_procs = kinfo.nr_procs;

  /* Allocate memory for process tables */
  ps_proc = (struct proc *) malloc((nr_tasks + nr_procs) * sizeof(ps_proc[0]));
  ps_mproc = (struct mproc *) malloc(nr_procs * sizeof(ps_mproc[0]));
  ps_fproc = (struct fproc *) malloc(nr_procs * sizeof(ps_fproc[0]));
  if (ps_proc == NULL || ps_mproc == NULL || ps_fproc == NULL)
        err("Out of memory");

  /* Get kernel process table */
  if (addrread(kmemfd, (phys_clicks) 0,
                proc_addr, (char *) ps_proc,
                (nr_tasks + nr_procs) * sizeof(ps_proc[0]))
                        != (nr_tasks + nr_procs) * sizeof(ps_proc[0]))
        err("Can't get kernel proc table from /dev/kmem");

  /* Get mm/fs process tables */
  if (addrread(memfd, ps_proc[nr_tasks + PM_PROC_NR].p_memmap[D].mem_phys,
                mproc_addr, (char *) ps_mproc,
                nr_procs * sizeof(ps_mproc[0]))
                        != nr_procs * sizeof(ps_mproc[0]))
        err("Can't get mm proc table from /dev/mem");
  if (addrread(memfd, ps_proc[nr_tasks + FS_PROC_NR].p_memmap[D].mem_phys,
                fproc_addr, (char *) ps_fproc,
                nr_procs * sizeof(ps_fproc[0]))
                        != nr_procs * sizeof(ps_fproc[0]))
        err("Can't get fs proc table from /dev/mem");

  /* We need to know where INIT hangs out. */
  for (i = FS_PROC_NR; i < nr_procs; i++) {
        if (strcmp(ps_proc[nr_tasks + i].p_name, "init") == 0) break;
  }
  init_proc_nr = i;

  /* Now loop through process table and handle each entry */
  printf("%s", opt_long ? L_HEADER : S_HEADER);
  for (i = -nr_tasks; i < nr_procs; i++) {
        if (pstat(i, &buf) != -1 &&
            (opt_all || buf.ps_euid == uid || buf.ps_ruid == uid) &&
            (opt_notty || majdev(buf.ps_dev) == TTY_MAJ)) {
                if (buf.ps_pid == 0 && i != PM_PROC_NR) {
                        sprintf(pid, "(%d)", i);
                } else {
                        sprintf(pid, "%d", buf.ps_pid);
                }

                ustime = (buf.ps_utime + buf.ps_stime) / HZ;
                if (ustime < 60 * 60) {
                        sprintf(cpu, "%2lu:%02lu", ustime / 60, ustime % 60);
                } else
                if (ustime < 100L * 60 * 60) {
                        ustime /= 60;
                        sprintf(cpu, "%2luh%02lu", ustime / 60, ustime % 60);
                } else {
                        sprintf(cpu, "%4luh", ustime / 3600);
                }

                if (opt_long) printf(L_FORMAT,
                               buf.ps_flags, buf.ps_state,
                               buf.ps_euid, pid, buf.ps_ppid, 
                               buf.ps_pgrp,
                               off_to_k((buf.ps_tsize
                                         + buf.ps_stack - buf.ps_data
                                         + buf.ps_ssize)),
                               (buf.ps_flags & RECEIVING ?
                                prrecv(&buf) :
                                ""),
                               tname((Dev_t) buf.ps_dev),
                               cpu,
                               i <= init_proc_nr || buf.ps_args == NULL
                                       ? taskname(i) : buf.ps_args);
                else
                        printf(S_FORMAT,
                               pid, tname((Dev_t) buf.ps_dev),
                               cpu,
                               i <= init_proc_nr || buf.ps_args == NULL
                                       ? taskname(i) : buf.ps_args);
        }
  }
  return(0);
}

char *get_args(bufp)
struct pstat *bufp;
{
  int nargv;
  int cnt;                      /* # of bytes read from stack frame */
  int neos;                     /* # of '\0's seen in argv string space */
  phys_bytes iframe;
  long l;
  char *cp, *args;
  static union stack {
        vir_bytes stk_i;
        char *stk_cp;
        char stk_c;
  } stk[ARG_MAX / sizeof(char *)];
  union stack *sp;

  /* Phys address of the original stack frame. */
  iframe = bufp->ps_procargs - bufp->ps_vstack + bufp->ps_stack;

  /* Calculate the number of bytes to read from user stack */
  l = (phys_bytes) bufp->ps_ssize - (iframe - bufp->ps_stack);
  if (l > ARG_MAX) l = ARG_MAX;
  cnt = l;

  /* Get cnt bytes from user initial stack to local stack buffer */
  if (lseek(memfd, (off_t) iframe, 0) < 0)
        return NULL; 

  if ( read(memfd, (char *)stk, cnt) != cnt ) 
        return NULL;

  sp = stk;
  nargv = (int) sp[0].stk_i;  /* number of argv arguments */

  /* See if argv[0] is with the bytes we read in */
  l = (long) sp[1].stk_cp - (long) bufp->ps_procargs;

  if ( ( l < 0 ) || ( l > cnt ) )  
        return NULL;

  /* l is the offset of the argv[0] argument */
  /* change for concatenation the '\0' to space, for nargv elements */

  args = &((char *) stk)[(int)l]; 
  neos = 0;
  for (cp = args; cp < &((char *) stk)[cnt]; cp++)
        if (*cp == '\0')
                if (++neos >= nargv)
                        break;
                else
                        *cp = ' ';
  if (cp == args) return NULL;
  *cp = '\0';

  return args;

}

/* Pstat collects info on process number p_nr and returns it in buf.
 * It is assumed that tasks do not have entries in fproc/mproc.
 */
int pstat(p_nr, bufp)
int p_nr;
struct pstat *bufp;
{
  int p_ki = p_nr + nr_tasks;   /* kernel proc index */

  if (p_nr < -nr_tasks || p_nr >= nr_procs) return -1;

  if ((ps_proc[p_ki].p_rts_flags == SLOT_FREE)
                                && !(ps_mproc[p_nr].mp_flags & IN_USE))
        return -1;

  bufp->ps_flags = ps_proc[p_ki].p_rts_flags;

  if (p_nr >= low_user) {
        bufp->ps_dev = ps_fproc[p_nr].fp_tty;
        bufp->ps_ftask = ps_fproc[p_nr].fp_task;
  } else {
        bufp->ps_dev = 0;
        bufp->ps_ftask = 0;
  }

  if (p_nr >= 0) {
        bufp->ps_ruid = ps_mproc[p_nr].mp_realuid;
        bufp->ps_euid = ps_mproc[p_nr].mp_effuid;
        bufp->ps_pid = ps_mproc[p_nr].mp_pid;
        bufp->ps_ppid = ps_mproc[ps_mproc[p_nr].mp_parent].mp_pid;
        bufp->ps_pgrp = ps_mproc[p_nr].mp_procgrp;
        bufp->ps_mflags = ps_mproc[p_nr].mp_flags;
  } else {
        bufp->ps_pid = 0;
        bufp->ps_ppid = 0;
        bufp->ps_ruid = bufp->ps_euid = 0;
        bufp->ps_pgrp = 0;
        bufp->ps_mflags = 0;
  }

  /* State is interpretation of combined kernel/mm flags for non-tasks */
  if (p_nr >= low_user) {               /* non-tasks */
        if (ps_mproc[p_nr].mp_flags & ZOMBIE)
                bufp->ps_state = Z_STATE;       /* zombie */
        else if (ps_mproc[p_nr].mp_flags & STOPPED)
                bufp->ps_state = T_STATE;       /* stopped (traced) */
        else if (ps_proc[p_ki].p_rts_flags == 0)
                bufp->ps_state = R_STATE;       /* in run-queue */
        else if (ps_mproc[p_nr].mp_flags & (WAITING | PAUSED | SIGSUSPENDED) ||
                 ps_fproc[p_nr].fp_suspended == SUSPENDED)
                bufp->ps_state = S_STATE;       /* sleeping */
        else
                bufp->ps_state = W_STATE;       /* a short wait */
  } else {                      /* tasks are simple */
        if (ps_proc[p_ki].p_rts_flags == 0)
                bufp->ps_state = R_STATE;       /* in run-queue */
        else
                bufp->ps_state = W_STATE;       /* other i.e. waiting */
  }

  bufp->ps_tsize = (size_t) ps_proc[p_ki].p_memmap[T].mem_len << CLICK_SHIFT;
  bufp->ps_dsize = (size_t) ps_proc[p_ki].p_memmap[D].mem_len << CLICK_SHIFT;
  bufp->ps_ssize = (size_t) ps_proc[p_ki].p_memmap[S].mem_len << CLICK_SHIFT;
  bufp->ps_vtext = (off_t) ps_proc[p_ki].p_memmap[T].mem_vir << CLICK_SHIFT;
  bufp->ps_vdata = (off_t) ps_proc[p_ki].p_memmap[D].mem_vir << CLICK_SHIFT;
  bufp->ps_vstack = (off_t) ps_proc[p_ki].p_memmap[S].mem_vir << CLICK_SHIFT;
  bufp->ps_text = (off_t) ps_proc[p_ki].p_memmap[T].mem_phys << CLICK_SHIFT;
  bufp->ps_data = (off_t) ps_proc[p_ki].p_memmap[D].mem_phys << CLICK_SHIFT;
  bufp->ps_stack = (off_t) ps_proc[p_ki].p_memmap[S].mem_phys << CLICK_SHIFT;

  bufp->ps_recv = _ENDPOINT_P(ps_proc[p_ki].p_getfrom_e);

  bufp->ps_utime = ps_proc[p_ki].p_user_time;
  bufp->ps_stime = ps_proc[p_ki].p_sys_time;

  bufp->ps_procargs = ps_mproc[p_nr].mp_procargs;

  if (bufp->ps_state == Z_STATE)
        bufp->ps_args = "<defunct>";
  else if (p_nr > init_proc_nr)
        bufp->ps_args = get_args(bufp);

  return 0;
}

/* Addrread reads nbytes from offset addr to click base of fd into buf. */
int addrread(fd, base, addr, buf, nbytes)
int fd;
phys_clicks base;
vir_bytes addr;
char *buf;
int nbytes;
{
  if (lseek(fd, ((off_t) base << CLICK_SHIFT) + addr, 0) < 0)
        return -1;

  return read(fd, buf, nbytes);
}

void usage(pname)
char *pname;
{
  fprintf(stderr, "Usage: %s [-][aeflx]\n", pname);
  exit(1);
}

void err(s)
char *s;
{
  extern int errno;

  if (errno == 0)
        fprintf(stderr, "ps: %s\n", s);
  else
        fprintf(stderr, "ps: %s: %s\n", s, strerror(errno));

  exit(2);
}

/* Fill ttyinfo by fstatting character specials in /dev. */
int gettynames()
{
  static char dev_path[] = "/dev/";
  struct stat statbuf;
  static char path[sizeof(dev_path) + NAME_MAX];
  int index;
  struct ttyent *ttyp;

  index = 0;
  while ((ttyp = getttyent()) != NULL) {
        strcpy(path, dev_path);
        strcat(path, ttyp->ty_name);
        if (stat(path, &statbuf) == -1 || !S_ISCHR(statbuf.st_mode))
                continue;
        if (index >= n_ttyinfo) {
                n_ttyinfo= (index+16) * 2;
                ttyinfo = realloc(ttyinfo, n_ttyinfo * sizeof(ttyinfo[0]));
                if (ttyinfo == NULL) err("Out of memory");
        }
        ttyinfo[index].tty_dev = statbuf.st_rdev;
        strcpy(ttyinfo[index].tty_name, ttyp->ty_name);
        index++;
  }
  endttyent();
  while (index < n_ttyinfo) ttyinfo[index++].tty_dev= 0;

  return 0;
}