vserver 1.9.5.x5
[linux-2.6.git] / kernel / acct.c
index f0e2bd0..32e39ac 100644 (file)
@@ -53,6 +53,7 @@
 #include <linux/vfs.h>
 #include <linux/jiffies.h>
 #include <linux/times.h>
+#include <linux/syscalls.h>
 #include <asm/uaccess.h>
 #include <asm/div64.h>
 #include <linux/blkdev.h> /* sector_div */
@@ -302,6 +303,69 @@ static comp_t encode_comp_t(unsigned long value)
        return exp;
 }
 
+#if ACCT_VERSION==1 || ACCT_VERSION==2
+/*
+ * encode an u64 into a comp2_t (24 bits)
+ *
+ * Format: 5 bit base 2 exponent, 20 bits mantissa.
+ * The leading bit of the mantissa is not stored, but implied for
+ * non-zero exponents.
+ * Largest encodable value is 50 bits.
+ */
+
+#define MANTSIZE2       20                      /* 20 bit mantissa. */
+#define EXPSIZE2        5                       /* 5 bit base 2 exponent. */
+#define MAXFRACT2       ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
+#define MAXEXP2         ((1 <<EXPSIZE2) - 1)    /* Maximum exponent. */
+
+static comp2_t encode_comp2_t(u64 value)
+{
+        int exp, rnd;
+
+        exp = (value > (MAXFRACT2>>1));
+        rnd = 0;
+        while (value > MAXFRACT2) {
+                rnd = value & 1;
+                value >>= 1;
+                exp++;
+        }
+
+        /*
+         * If we need to round up, do it (and handle overflow correctly).
+         */
+        if (rnd && (++value > MAXFRACT2)) {
+                value >>= 1;
+                exp++;
+        }
+
+        if (exp > MAXEXP2) {
+                /* Overflow. Return largest representable number instead. */
+                return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
+        } else {
+                return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
+        }
+}
+#endif
+
+#if ACCT_VERSION==3
+/*
+ * encode an u64 into a 32 bit IEEE float
+ */
+static u32 encode_float(u64 value)
+{
+       unsigned exp = 190;
+       unsigned u;
+
+       if (value==0) return 0;
+       while ((s64)value > 0){
+               value <<= 1;
+               exp--;
+       }
+       u = (u32)(value >> 40) & 0x7fffffu;
+       return u | (exp << 23);
+}
+#endif
+
 /*
  *  Write an accounting entry for an exiting process
  *
@@ -316,11 +380,13 @@ static comp_t encode_comp_t(unsigned long value)
  */
 static void do_acct_process(long exitcode, struct file *file)
 {
-       struct acct ac;
+       acct_t ac;
        mm_segment_t fs;
        unsigned long vsize;
        unsigned long flim;
        u64 elapsed;
+       u64 run_time;
+       struct timespec uptime;
 
        /*
         * First check to see if there is enough free_space to continue
@@ -333,20 +399,55 @@ static void do_acct_process(long exitcode, struct file *file)
         * Fill the accounting struct with the needed info as recorded
         * by the different kernel functions.
         */
-       memset((caddr_t)&ac, 0, sizeof(struct acct));
+       memset((caddr_t)&ac, 0, sizeof(acct_t));
 
+       ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER;
        strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
 
-       elapsed = jiffies_64_to_clock_t(get_jiffies_64() - current->start_time);
+       /* calculate run_time in nsec*/
+       do_posix_clock_monotonic_gettime(&uptime);
+       run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec;
+       run_time -= (u64)current->start_time.tv_sec*NSEC_PER_SEC
+                                       + current->start_time.tv_nsec;
+       /* convert nsec -> AHZ */
+       elapsed = nsec_to_AHZ(run_time);
+#if ACCT_VERSION==3
+       ac.ac_etime = encode_float(elapsed);
+#else
        ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
                               (unsigned long) elapsed : (unsigned long) -1l);
-       do_div(elapsed, USER_HZ);
+#endif
+#if ACCT_VERSION==1 || ACCT_VERSION==2
+       {
+               /* new enlarged etime field */
+               comp2_t etime = encode_comp2_t(elapsed);
+               ac.ac_etime_hi = etime >> 16;
+               ac.ac_etime_lo = (u16) etime;
+       }
+#endif
+       do_div(elapsed, AHZ);
        ac.ac_btime = xtime.tv_sec - elapsed;
-       ac.ac_utime = encode_comp_t(jiffies_to_clock_t(current->utime));
-       ac.ac_stime = encode_comp_t(jiffies_to_clock_t(current->stime));
+       ac.ac_utime = encode_comp_t(jiffies_to_AHZ(
+                                           current->signal->utime +
+                                           current->group_leader->utime));
+       ac.ac_stime = encode_comp_t(jiffies_to_AHZ(
+                                           current->signal->stime +
+                                           current->group_leader->stime));
        /* we really need to bite the bullet and change layout */
        ac.ac_uid = current->uid;
        ac.ac_gid = current->gid;
+#if ACCT_VERSION==2
+       ac.ac_ahz = AHZ;
+#endif
+#if ACCT_VERSION==1 || ACCT_VERSION==2
+       /* backward-compatible 16 bit fields */
+       ac.ac_uid16 = current->uid;
+       ac.ac_gid16 = current->gid;
+#endif
+#if ACCT_VERSION==3
+       ac.ac_pid = current->tgid;
+       ac.ac_ppid = current->parent->tgid;
+#endif
 
        read_lock(&tasklist_lock);      /* pin current->signal */
        ac.ac_tty = current->signal->tty ?
@@ -378,8 +479,10 @@ static void do_acct_process(long exitcode, struct file *file)
        ac.ac_mem = encode_comp_t(vsize);
        ac.ac_io = encode_comp_t(0 /* current->io_usage */);    /* %% */
        ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
-       ac.ac_minflt = encode_comp_t(current->min_flt);
-       ac.ac_majflt = encode_comp_t(current->maj_flt);
+       ac.ac_minflt = encode_comp_t(current->signal->min_flt +
+                                    current->group_leader->min_flt);
+       ac.ac_majflt = encode_comp_t(current->signal->maj_flt +
+                                    current->group_leader->maj_flt);
        ac.ac_swaps = encode_comp_t(0);
        ac.ac_exitcode = exitcode;
 
@@ -392,11 +495,11 @@ static void do_acct_process(long exitcode, struct file *file)
        /*
         * Accounting records are not subject to resource limits.
         */
-       flim = current->rlim[RLIMIT_FSIZE].rlim_cur;
-       current->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
+       flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
+       current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
        file->f_op->write(file, (char *)&ac,
-                              sizeof(struct acct), &file->f_pos);
-       current->rlim[RLIMIT_FSIZE].rlim_cur = flim;
+                              sizeof(acct_t), &file->f_pos);
+       current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
        set_fs(fs);
 }
 
@@ -425,3 +528,36 @@ void acct_process(long exitcode)
        do_acct_process(exitcode, file);
        fput(file);
 }
+
+
+/*
+ * acct_update_integrals
+ *    -  update mm integral fields in task_struct
+ */
+void acct_update_integrals(void)
+{
+       struct task_struct *tsk = current;
+
+       if (likely(tsk->mm)) {
+               long delta = tsk->stime - tsk->acct_stimexpd;
+
+               if (delta == 0)
+                       return;
+               tsk->acct_stimexpd = tsk->stime;
+               tsk->acct_rss_mem1 += delta * tsk->mm->rss;
+               tsk->acct_vm_mem1 += delta * tsk->mm->total_vm;
+       }
+}
+
+/*
+ * acct_clear_integrals
+ *    - clear the mm integral fields in task_struct
+ */
+void acct_clear_integrals(struct task_struct *tsk)
+{
+       if (tsk) {
+               tsk->acct_stimexpd = 0;
+               tsk->acct_rss_mem1 = 0;
+               tsk->acct_vm_mem1 = 0;
+       }
+}