kernel/acct.c

   1 /*
   2  *  linux/kernel/acct.c
   3  *
   4  *  BSD Process Accounting for Linux
   5  *
   6  *  Author: Marco van Wieringen <mvw@planets.elm.net>
   7  *
   8  *  Some code based on ideas and code from:
   9  *  Thomas K. Dyas <tdyas@eden.rutgers.edu>
  10  *
  11  *  This file implements BSD-style process accounting. Whenever any
  12  *  process exits, an accounting record of type "struct acct" is
  13  *  written to the file specified with the acct() system call. It is
  14  *  up to user-level programs to do useful things with the accounting
  15  *  log. The kernel just provides the raw accounting information.
  16  *
  17  * (C) Copyright 1995 - 1997 Marco van Wieringen - ELM Consultancy B.V.
  18  *
  19  *  Plugged two leaks. 1) It didn't return acct_file into the free_filps if
  20  *  the file happened to be read-only. 2) If the accounting was suspended
  21  *  due to the lack of space it happily allowed to reopen it and completely
  22  *  lost the old acct_file. 3/10/98, Al Viro.
  23  *
  24  *  Now we silently close acct_file on attempt to reopen. Cleaned sys_acct().
  25  *  XTerms and EMACS are manifestations of pure evil. 21/10/98, AV.
  26  *
  27  *  Fixed a nasty interaction with with sys_umount(). If the accointing
  28  *  was suspeneded we failed to stop it on umount(). Messy.
  29  *  Another one: remount to readonly didn't stop accounting.
  30  *      Question: what should we do if we have CAP_SYS_ADMIN but not
  31  *  CAP_SYS_PACCT? Current code does the following: umount returns -EBUSY
  32  *  unless we are messing with the root. In that case we are getting a
  33  *  real mess with do_remount_sb(). 9/11/98, AV.
  34  *
  35  *  Fixed a bunch of races (and pair of leaks). Probably not the best way,
  36  *  but this one obviously doesn't introduce deadlocks. Later. BTW, found
  37  *  one race (and leak) in BSD implementation.
  38  *  OK, that's better. ANOTHER race and leak in BSD variant. There always
  39  *  is one more bug... 10/11/98, AV.
  40  *
  41  *      Oh, fsck... Oopsable SMP race in do_process_acct() - we must hold
  42  * ->mmap_sem to walk the vma list of current->mm. Nasty, since it leaks
  43  * a struct file opened for write. Fixed. 2/6/2000, AV.
  44  */
  45
  46 #include <linux/config.h>
  47 #include <linux/errno.h>
  48 #include <linux/kernel.h>
  49
  50 #ifdef CONFIG_BSD_PROCESS_ACCT
  51 #include <linux/mm.h>
  52 #include <linux/slab.h>
  53 #include <linux/acct.h>
  54 #include <linux/smp_lock.h>
  55 #include <linux/file.h>
  56
  57 #include <asm/uaccess.h>
  58
  59 /*
  60  * These constants control the amount of freespace that suspend and
  61  * resume the process accounting system, and the time delay between
  62  * each check.
  63  * Turned into sysctl-controllable parameters. AV, 12/11/98
  64  */
  65
  66 int acct_parm[3] = {4, 2, 30};
  67 #define RESUME          (acct_parm[0])  /* >foo% free space - resume */
  68 #define SUSPEND         (acct_parm[1])  /* <foo% free space - suspend */
  69 #define ACCT_TIMEOUT    (acct_parm[2])  /* foo second timeout between checks */
  70
  71 /*
  72  * External references and all of the globals.
  73  */
  74
  75 static volatile int acct_active;
  76 static volatile int acct_needcheck;
  77 static struct file *acct_file;
  78 static struct timer_list acct_timer;
  79 static void do_acct_process(long, struct file *);
  80
  81 /*
  82  * Called whenever the timer says to check the free space.
  83  */
  84 static void acct_timeout(unsigned long unused)
  85 {
  86         acct_needcheck = 1;
  87 }
  88
  89 /*
  90  * Check the amount of free space and suspend/resume accordingly.
  91  */
  92 static int check_free_space(struct file *file)
  93 {
  94         struct statfs sbuf;
  95         int res;
  96         int act;
  97
  98         lock_kernel();
  99         res = acct_active;
 100         if (!file || !acct_needcheck)
 101                 goto out;
 102         unlock_kernel();
 103
 104         /* May block */
 105         if (vfs_statfs(file->f_dentry->d_inode->i_sb, &sbuf))
 106                 return res;
 107
 108         if (sbuf.f_bavail <= SUSPEND * sbuf.f_blocks / 100)
 109                 act = -1;
 110         else if (sbuf.f_bavail >= RESUME * sbuf.f_blocks / 100)
 111                 act = 1;
 112         else
 113                 act = 0;
 114
 115         /*
 116          * If some joker switched acct_file under us we'ld better be
 117          * silent and _not_ touch anything.
 118          */
 119         lock_kernel();
 120         if (file != acct_file) {
 121                 if (act)
 122                         res = act>0;
 123                 goto out;
 124         }
 125
 126         if (acct_active) {
 127                 if (act < 0) {
 128                         acct_active = 0;
 129                         printk(KERN_INFO "Process accounting paused\n");
 130                 }
 131         } else {
 132                 if (act > 0) {
 133                         acct_active = 1;
 134                         printk(KERN_INFO "Process accounting resumed\n");
 135                 }
 136         }
 137
 138         del_timer(&acct_timer);
 139         acct_needcheck = 0;
 140         acct_timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 141         add_timer(&acct_timer);
 142         res = acct_active;
 143 out:
 144         unlock_kernel();
 145         return res;
 146 }
 147
 148 /*
 149  *  sys_acct() is the only system call needed to implement process
 150  *  accounting. It takes the name of the file where accounting records
 151  *  should be written. If the filename is NULL, accounting will be
 152  *  shutdown.
 153  */
 154 asmlinkage long sys_acct(const char *name)
 155 {
 156         struct file *file = NULL, *old_acct = NULL;
 157         char *tmp;
 158         int error;
 159
 160         if (!capable(CAP_SYS_PACCT))
 161                 return -EPERM;
 162
 163         if (name) {
 164                 tmp = getname(name);
 165                 error = PTR_ERR(tmp);
 166                 if (IS_ERR(tmp))
 167                         goto out;
 168                 /* Difference from BSD - they don't do O_APPEND */
 169                 file = filp_open(tmp, O_WRONLY|O_APPEND, 0);
 170                 putname(tmp);
 171                 if (IS_ERR(file)) {
 172                         error = PTR_ERR(file);
 173                         goto out;
 174                 }
 175                 error = -EACCES;
 176                 if (!S_ISREG(file->f_dentry->d_inode->i_mode))
 177                         goto out_err;
 178
 179                 error = -EIO;
 180                 if (!file->f_op->write)
 181                         goto out_err;
 182         }
 183
 184         error = 0;
 185         lock_kernel();
 186         if (acct_file) {
 187                 old_acct = acct_file;
 188                 del_timer(&acct_timer);
 189                 acct_active = 0;
 190                 acct_needcheck = 0;
 191                 acct_file = NULL;
 192         }
 193         if (name) {
 194                 acct_file = file;
 195                 acct_needcheck = 0;
 196                 acct_active = 1;
 197                 /* It's been deleted if it was used before so this is safe */
 198                 init_timer(&acct_timer);
 199                 acct_timer.function = acct_timeout;
 200                 acct_timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 201                 add_timer(&acct_timer);
 202         }
 203         unlock_kernel();
 204         if (old_acct) {
 205                 do_acct_process(0,old_acct);
 206                 filp_close(old_acct, NULL);
 207         }
 208 out:
 209         return error;
 210 out_err:
 211         filp_close(file, NULL);
 212         goto out;
 213 }
 214
 215 void acct_auto_close(kdev_t dev)
 216 {
 217         lock_kernel();
 218         if (acct_file && acct_file->f_dentry->d_inode->i_dev == dev)
 219                 sys_acct(NULL);
 220         unlock_kernel();
 221 }
 222
 223 /*
 224  *  encode an unsigned long into a comp_t
 225  *
 226  *  This routine has been adopted from the encode_comp_t() function in
 227  *  the kern_acct.c file of the FreeBSD operating system. The encoding
 228  *  is a 13-bit fraction with a 3-bit (base 8) exponent.
 229  */
 230
 231 #define MANTSIZE        13                      /* 13 bit mantissa. */
 232 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 233 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 234
 235 static comp_t encode_comp_t(unsigned long value)
 236 {
 237         int exp, rnd;
 238
 239         exp = rnd = 0;
 240         while (value > MAXFRACT) {
 241                 rnd = value & (1 << (EXPSIZE - 1));     /* Round up? */
 242                 value >>= EXPSIZE;      /* Base 8 exponent == 3 bit shift. */
 243                 exp++;
 244         }
 245
 246         /*
 247          * If we need to round up, do it (and handle overflow correctly).
 248          */
 249         if (rnd && (++value > MAXFRACT)) {
 250                 value >>= EXPSIZE;
 251                 exp++;
 252         }
 253
 254         /*
 255          * Clean it up and polish it off.
 256          */
 257         exp <<= MANTSIZE;               /* Shift the exponent into place */
 258         exp += value;                   /* and add on the mantissa. */
 259         return exp;
 260 }
 261
 262 /*
 263  *  Write an accounting entry for an exiting process
 264  *
 265  *  The acct_process() call is the workhorse of the process
 266  *  accounting system. The struct acct is built here and then written
 267  *  into the accounting file. This function should only be called from
 268  *  do_exit().
 269  */
 270
 271 /*
 272  *  do_acct_process does all actual work. Caller holds the reference to file.
 273  */
 274 static void do_acct_process(long exitcode, struct file *file)
 275 {
 276         struct acct ac;
 277         mm_segment_t fs;
 278         unsigned long vsize;
 279
 280         /*
 281          * First check to see if there is enough free_space to continue
 282          * the process accounting system.
 283          */
 284         if (!check_free_space(file))
 285                 return;
 286
 287         /*
 288          * Fill the accounting struct with the needed info as recorded
 289          * by the different kernel functions.
 290          */
 291         memset((caddr_t)&ac, 0, sizeof(struct acct));
 292
 293         strncpy(ac.ac_comm, current->comm, ACCT_COMM);
 294         ac.ac_comm[ACCT_COMM - 1] = '\0';
 295
 296         ac.ac_btime = CT_TO_SECS(current->start_time) + (xtime.tv_sec - (jiffies / HZ));
 297         ac.ac_etime = encode_comp_t(jiffies - current->start_time);
 298         ac.ac_utime = encode_comp_t(current->times.tms_utime);
 299         ac.ac_stime = encode_comp_t(current->times.tms_stime);
 300         ac.ac_uid = current->uid;
 301         ac.ac_gid = current->gid;
 302         ac.ac_tty = (current->tty) ? kdev_t_to_nr(current->tty->device) : 0;
 303
 304         ac.ac_flag = 0;
 305         if (current->flags & PF_FORKNOEXEC)
 306                 ac.ac_flag |= AFORK;
 307         if (current->flags & PF_SUPERPRIV)
 308                 ac.ac_flag |= ASU;
 309         if (current->flags & PF_DUMPCORE)
 310                 ac.ac_flag |= ACORE;
 311         if (current->flags & PF_SIGNALED)
 312                 ac.ac_flag |= AXSIG;
 313
 314         vsize = 0;
 315         if (current->mm) {
 316                 struct vm_area_struct *vma;
 317                 down(&current->mm->mmap_sem);
 318                 vma = current->mm->mmap;
 319                 while (vma) {
 320                         vsize += vma->vm_end - vma->vm_start;
 321                         vma = vma->vm_next;
 322                 }
 323                 up(&current->mm->mmap_sem);
 324         }
 325         vsize = vsize / 1024;
 326         ac.ac_mem = encode_comp_t(vsize);
 327         ac.ac_io = encode_comp_t(0 /* current->io_usage */);    /* %% */
 328         ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
 329         ac.ac_minflt = encode_comp_t(current->min_flt);
 330         ac.ac_majflt = encode_comp_t(current->maj_flt);
 331         ac.ac_swaps = encode_comp_t(current->nswap);
 332         ac.ac_exitcode = exitcode;
 333
 334         /*
 335          * Kernel segment override to datasegment and write it
 336          * to the accounting file.
 337          */
 338         fs = get_fs();
 339         set_fs(KERNEL_DS);
 340         file->f_op->write(file, (char *)&ac,
 341                                sizeof(struct acct), &file->f_pos);
 342         set_fs(fs);
 343 }
 344
 345 /*
 346  * acct_process - now just a wrapper around do_acct_process
 347  */
 348 int acct_process(long exitcode)
 349 {
 350         struct file *file = NULL;
 351         lock_kernel();
 352         if (acct_file) {
 353                 file = acct_file;
 354                 get_file(file);
 355                 unlock_kernel();
 356                 do_acct_process(exitcode, acct_file);
 357                 fput(file);
 358         } else
 359                 unlock_kernel();
 360         return 0;
 361 }
 362
 363 #else
 364 /*
 365  * Dummy system call when BSD process accounting is not configured
 366  * into the kernel.
 367  */
 368
 369 asmlinkage long sys_acct(const char * filename)
 370 {
 371         return -ENOSYS;
 372 }
 373 #endif