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/mm.h>
  48 #include <linux/slab.h>
  49 #include <linux/acct.h>
  50 #include <linux/file.h>
  51 #include <linux/tty.h>
  52 #include <asm/uaccess.h>
  53
  54 /*
  55  * These constants control the amount of freespace that suspend and
  56  * resume the process accounting system, and the time delay between
  57  * each check.
  58  * Turned into sysctl-controllable parameters. AV, 12/11/98
  59  */
  60
  61 int acct_parm[3] = {4, 2, 30};
  62 #define RESUME          (acct_parm[0])  /* >foo% free space - resume */
  63 #define SUSPEND         (acct_parm[1])  /* <foo% free space - suspend */
  64 #define ACCT_TIMEOUT    (acct_parm[2])  /* foo second timeout between checks */
  65
  66 /*
  67  * External references and all of the globals.
  68  */
  69 static void do_acct_process(long, struct file *);
  70
  71 /*
  72  * This structure is used so that all the data protected by lock
  73  * can be placed in the same cache line as the lock.  This primes
  74  * the cache line to have the data after getting the lock.
  75  */
  76 struct acct_glbs {
  77         spinlock_t              lock;
  78         volatile int            active;
  79         volatile int            needcheck;
  80         struct file             *file;
  81         struct timer_list       timer;
  82 };
  83
  84 static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED};
  85
  86 /*
  87  * Called whenever the timer says to check the free space.
  88  */
  89 static void acct_timeout(unsigned long unused)
  90 {
  91         acct_globals.needcheck = 1;
  92 }
  93
  94 /*
  95  * Check the amount of free space and suspend/resume accordingly.
  96  */
  97 static int check_free_space(struct file *file)
  98 {
  99         struct statfs sbuf;
 100         int res;
 101         int act;
 102
 103         spin_lock(&acct_globals.lock);
 104         res = acct_globals.active;
 105         if (!file || !acct_globals.needcheck)
 106                 goto out;
 107         spin_unlock(&acct_globals.lock);
 108
 109         /* May block */
 110         if (vfs_statfs(file->f_dentry->d_inode->i_sb, &sbuf))
 111                 return res;
 112
 113         if (sbuf.f_bavail <= SUSPEND * sbuf.f_blocks / 100)
 114                 act = -1;
 115         else if (sbuf.f_bavail >= RESUME * sbuf.f_blocks / 100)
 116                 act = 1;
 117         else
 118                 act = 0;
 119
 120         /*
 121          * If some joker switched acct_globals.file under us we'ld better be
 122          * silent and _not_ touch anything.
 123          */
 124         spin_lock(&acct_globals.lock);
 125         if (file != acct_globals.file) {
 126                 if (act)
 127                         res = act>0;
 128                 goto out;
 129         }
 130
 131         if (acct_globals.active) {
 132                 if (act < 0) {
 133                         acct_globals.active = 0;
 134                         printk(KERN_INFO "Process accounting paused\n");
 135                 }
 136         } else {
 137                 if (act > 0) {
 138                         acct_globals.active = 1;
 139                         printk(KERN_INFO "Process accounting resumed\n");
 140                 }
 141         }
 142
 143         del_timer(&acct_globals.timer);
 144         acct_globals.needcheck = 0;
 145         acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 146         add_timer(&acct_globals.timer);
 147         res = acct_globals.active;
 148 out:
 149         spin_unlock(&acct_globals.lock);
 150         return res;
 151 }
 152
 153 /*
 154  * Close the old accouting file (if currently open) and then replace
 155  * it with file (if non-NULL).
 156  *
 157  * NOTE: acct_globals.lock MUST be held on entry and exit.
 158  */
 159 void acct_file_reopen(struct file *file)
 160 {
 161         struct file *old_acct = NULL;
 162
 163         if (acct_globals.file) {
 164                 old_acct = acct_globals.file;
 165                 del_timer(&acct_globals.timer);
 166                 acct_globals.active = 0;
 167                 acct_globals.needcheck = 0;
 168                 acct_globals.file = NULL;
 169         }
 170         if (file) {
 171                 acct_globals.file = file;
 172                 acct_globals.needcheck = 0;
 173                 acct_globals.active = 1;
 174                 /* It's been deleted if it was used before so this is safe */
 175                 init_timer(&acct_globals.timer);
 176                 acct_globals.timer.function = acct_timeout;
 177                 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 178                 add_timer(&acct_globals.timer);
 179         }
 180         if (old_acct) {
 181                 spin_unlock(&acct_globals.lock);
 182                 do_acct_process(0, old_acct);
 183                 filp_close(old_acct, NULL);
 184                 spin_lock(&acct_globals.lock);
 185         }
 186 }
 187
 188 /*
 189  *  sys_acct() is the only system call needed to implement process
 190  *  accounting. It takes the name of the file where accounting records
 191  *  should be written. If the filename is NULL, accounting will be
 192  *  shutdown.
 193  */
 194 asmlinkage long sys_acct(const char *name)
 195 {
 196         struct file *file = NULL;
 197         char *tmp;
 198         int error;
 199
 200         if (!capable(CAP_SYS_PACCT))
 201                 return -EPERM;
 202
 203         if (name) {
 204                 tmp = getname(name);
 205                 if (IS_ERR(tmp)) {
 206                         return (PTR_ERR(tmp));
 207                 }
 208                 /* Difference from BSD - they don't do O_APPEND */
 209                 file = filp_open(tmp, O_WRONLY|O_APPEND, 0);
 210                 putname(tmp);
 211                 if (IS_ERR(file)) {
 212                         return (PTR_ERR(file));
 213                 }
 214                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 215                         filp_close(file, NULL);
 216                         return (-EACCES);
 217                 }
 218
 219                 if (!file->f_op->write) {
 220                         filp_close(file, NULL);
 221                         return (-EIO);
 222                 }
 223         }
 224
 225         error = security_ops->acct(file);
 226         if (error)
 227                 return error;
 228
 229         spin_lock(&acct_globals.lock);
 230         acct_file_reopen(file);
 231         spin_unlock(&acct_globals.lock);
 232
 233         return (0);
 234 }
 235
 236 /*
 237  * If the accouting is turned on for a file in the filesystem pointed
 238  * to by sb, turn accouting off.
 239  */
 240 void acct_auto_close(struct super_block *sb)
 241 {
 242         spin_lock(&acct_globals.lock);
 243         if (acct_globals.file &&
 244             acct_globals.file->f_dentry->d_inode->i_sb == sb) {
 245                 acct_file_reopen((struct file *)NULL);
 246         }
 247         spin_unlock(&acct_globals.lock);
 248 }
 249
 250 /*
 251  *  encode an unsigned long into a comp_t
 252  *
 253  *  This routine has been adopted from the encode_comp_t() function in
 254  *  the kern_acct.c file of the FreeBSD operating system. The encoding
 255  *  is a 13-bit fraction with a 3-bit (base 8) exponent.
 256  */
 257
 258 #define MANTSIZE        13                      /* 13 bit mantissa. */
 259 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 260 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 261
 262 static comp_t encode_comp_t(unsigned long value)
 263 {
 264         int exp, rnd;
 265
 266         exp = rnd = 0;
 267         while (value > MAXFRACT) {
 268                 rnd = value & (1 << (EXPSIZE - 1));     /* Round up? */
 269                 value >>= EXPSIZE;      /* Base 8 exponent == 3 bit shift. */
 270                 exp++;
 271         }
 272
 273         /*
 274          * If we need to round up, do it (and handle overflow correctly).
 275          */
 276         if (rnd && (++value > MAXFRACT)) {
 277                 value >>= EXPSIZE;
 278                 exp++;
 279         }
 280
 281         /*
 282          * Clean it up and polish it off.
 283          */
 284         exp <<= MANTSIZE;               /* Shift the exponent into place */
 285         exp += value;                   /* and add on the mantissa. */
 286         return exp;
 287 }
 288
 289 /*
 290  *  Write an accounting entry for an exiting process
 291  *
 292  *  The acct_process() call is the workhorse of the process
 293  *  accounting system. The struct acct is built here and then written
 294  *  into the accounting file. This function should only be called from
 295  *  do_exit().
 296  */
 297
 298 /*
 299  *  do_acct_process does all actual work. Caller holds the reference to file.
 300  */
 301 static void do_acct_process(long exitcode, struct file *file)
 302 {
 303         struct acct ac;
 304         mm_segment_t fs;
 305         unsigned long vsize;
 306         unsigned long flim;
 307
 308         /*
 309          * First check to see if there is enough free_space to continue
 310          * the process accounting system.
 311          */
 312         if (!check_free_space(file))
 313                 return;
 314
 315         /*
 316          * Fill the accounting struct with the needed info as recorded
 317          * by the different kernel functions.
 318          */
 319         memset((caddr_t)&ac, 0, sizeof(struct acct));
 320
 321         strncpy(ac.ac_comm, current->comm, ACCT_COMM);
 322         ac.ac_comm[ACCT_COMM - 1] = '\0';
 323
 324         ac.ac_btime = CT_TO_SECS(current->start_time) +
 325                 (xtime.tv_sec - (jiffies / HZ));
 326         ac.ac_etime = encode_comp_t(jiffies - current->start_time);
 327         ac.ac_utime = encode_comp_t(current->utime);
 328         ac.ac_stime = encode_comp_t(current->stime);
 329         ac.ac_uid = current->uid;
 330         ac.ac_gid = current->gid;
 331         ac.ac_tty = (current->tty) ? kdev_t_to_nr(current->tty->device) : 0;
 332
 333         ac.ac_flag = 0;
 334         if (current->flags & PF_FORKNOEXEC)
 335                 ac.ac_flag |= AFORK;
 336         if (current->flags & PF_SUPERPRIV)
 337                 ac.ac_flag |= ASU;
 338         if (current->flags & PF_DUMPCORE)
 339                 ac.ac_flag |= ACORE;
 340         if (current->flags & PF_SIGNALED)
 341                 ac.ac_flag |= AXSIG;
 342
 343         vsize = 0;
 344         if (current->mm) {
 345                 struct vm_area_struct *vma;
 346                 down_read(&current->mm->mmap_sem);
 347                 vma = current->mm->mmap;
 348                 while (vma) {
 349                         vsize += vma->vm_end - vma->vm_start;
 350                         vma = vma->vm_next;
 351                 }
 352                 up_read(&current->mm->mmap_sem);
 353         }
 354         vsize = vsize / 1024;
 355         ac.ac_mem = encode_comp_t(vsize);
 356         ac.ac_io = encode_comp_t(0 /* current->io_usage */);    /* %% */
 357         ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
 358         ac.ac_minflt = encode_comp_t(current->min_flt);
 359         ac.ac_majflt = encode_comp_t(current->maj_flt);
 360         ac.ac_swaps = encode_comp_t(current->nswap);
 361         ac.ac_exitcode = exitcode;
 362
 363         /*
 364          * Kernel segment override to datasegment and write it
 365          * to the accounting file.
 366          */
 367         fs = get_fs();
 368         set_fs(KERNEL_DS);
 369         /*
 370          * Accounting records are not subject to resource limits.
 371          */
 372         flim = current->rlim[RLIMIT_FSIZE].rlim_cur;
 373         current->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
 374         file->f_op->write(file, (char *)&ac,
 375                                sizeof(struct acct), &file->f_pos);
 376         current->rlim[RLIMIT_FSIZE].rlim_cur = flim;
 377         set_fs(fs);
 378 }
 379
 380 /*
 381  * acct_process - now just a wrapper around do_acct_process
 382  */
 383 int acct_process(long exitcode)
 384 {
 385         struct file *file = NULL;
 386         spin_lock(&acct_globals.lock);
 387         if (acct_globals.file) {
 388                 file = acct_globals.file;
 389                 get_file(file);
 390                 spin_unlock(&acct_globals.lock);
 391                 do_acct_process(exitcode, file);
 392                 fput(file);
 393         } else
 394                 spin_unlock(&acct_globals.lock);
 395         return 0;
 396 }