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 <linux/security.h>
  53 #include <linux/vfs.h>
  54 #include <linux/jiffies.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/div64.h>
  57 #include <linux/blkdev.h> /* sector_div */
  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 static void do_acct_process(long, struct file *);
  75
  76 /*
  77  * This structure is used so that all the data protected by lock
  78  * can be placed in the same cache line as the lock.  This primes
  79  * the cache line to have the data after getting the lock.
  80  */
  81 struct acct_glbs {
  82         spinlock_t              lock;
  83         volatile int            active;
  84         volatile int            needcheck;
  85         struct file             *file;
  86         struct timer_list       timer;
  87 };
  88
  89 static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED};
  90
  91 /*
  92  * Called whenever the timer says to check the free space.
  93  */
  94 static void acct_timeout(unsigned long unused)
  95 {
  96         acct_globals.needcheck = 1;
  97 }
  98
  99 /*
 100  * Check the amount of free space and suspend/resume accordingly.
 101  */
 102 static int check_free_space(struct file *file)
 103 {
 104         struct kstatfs sbuf;
 105         int res;
 106         int act;
 107         sector_t resume;
 108         sector_t suspend;
 109
 110         spin_lock(&acct_globals.lock);
 111         res = acct_globals.active;
 112         if (!file || !acct_globals.needcheck)
 113                 goto out;
 114         spin_unlock(&acct_globals.lock);
 115
 116         /* May block */
 117         if (vfs_statfs(file->f_dentry->d_inode->i_sb, &sbuf))
 118                 return res;
 119         suspend = sbuf.f_blocks * SUSPEND;
 120         resume = sbuf.f_blocks * RESUME;
 121
 122         sector_div(suspend, 100);
 123         sector_div(resume, 100);
 124
 125         if (sbuf.f_bavail <= suspend)
 126                 act = -1;
 127         else if (sbuf.f_bavail >= resume)
 128                 act = 1;
 129         else
 130                 act = 0;
 131
 132         /*
 133          * If some joker switched acct_globals.file under us we'ld better be
 134          * silent and _not_ touch anything.
 135          */
 136         spin_lock(&acct_globals.lock);
 137         if (file != acct_globals.file) {
 138                 if (act)
 139                         res = act>0;
 140                 goto out;
 141         }
 142
 143         if (acct_globals.active) {
 144                 if (act < 0) {
 145                         acct_globals.active = 0;
 146                         printk(KERN_INFO "Process accounting paused\n");
 147                 }
 148         } else {
 149                 if (act > 0) {
 150                         acct_globals.active = 1;
 151                         printk(KERN_INFO "Process accounting resumed\n");
 152                 }
 153         }
 154
 155         del_timer(&acct_globals.timer);
 156         acct_globals.needcheck = 0;
 157         acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 158         add_timer(&acct_globals.timer);
 159         res = acct_globals.active;
 160 out:
 161         spin_unlock(&acct_globals.lock);
 162         return res;
 163 }
 164
 165 /*
 166  * Close the old accouting file (if currently open) and then replace
 167  * it with file (if non-NULL).
 168  *
 169  * NOTE: acct_globals.lock MUST be held on entry and exit.
 170  */
 171 void acct_file_reopen(struct file *file)
 172 {
 173         struct file *old_acct = NULL;
 174
 175         if (acct_globals.file) {
 176                 old_acct = acct_globals.file;
 177                 del_timer(&acct_globals.timer);
 178                 acct_globals.active = 0;
 179                 acct_globals.needcheck = 0;
 180                 acct_globals.file = NULL;
 181         }
 182         if (file) {
 183                 acct_globals.file = file;
 184                 acct_globals.needcheck = 0;
 185                 acct_globals.active = 1;
 186                 /* It's been deleted if it was used before so this is safe */
 187                 init_timer(&acct_globals.timer);
 188                 acct_globals.timer.function = acct_timeout;
 189                 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 190                 add_timer(&acct_globals.timer);
 191         }
 192         if (old_acct) {
 193                 spin_unlock(&acct_globals.lock);
 194                 do_acct_process(0, old_acct);
 195                 filp_close(old_acct, NULL);
 196                 spin_lock(&acct_globals.lock);
 197         }
 198 }
 199
 200 /*
 201  *  sys_acct() is the only system call needed to implement process
 202  *  accounting. It takes the name of the file where accounting records
 203  *  should be written. If the filename is NULL, accounting will be
 204  *  shutdown.
 205  */
 206 asmlinkage long sys_acct(const char *name)
 207 {
 208         struct file *file = NULL;
 209         char *tmp;
 210         int error;
 211
 212         if (!capable(CAP_SYS_PACCT))
 213                 return -EPERM;
 214
 215         if (name) {
 216                 tmp = getname(name);
 217                 if (IS_ERR(tmp)) {
 218                         return (PTR_ERR(tmp));
 219                 }
 220                 /* Difference from BSD - they don't do O_APPEND */
 221                 file = filp_open(tmp, O_WRONLY|O_APPEND, 0);
 222                 putname(tmp);
 223                 if (IS_ERR(file)) {
 224                         return (PTR_ERR(file));
 225                 }
 226                 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
 227                         filp_close(file, NULL);
 228                         return (-EACCES);
 229                 }
 230
 231                 if (!file->f_op->write) {
 232                         filp_close(file, NULL);
 233                         return (-EIO);
 234                 }
 235         }
 236
 237         error = security_acct(file);
 238         if (error)
 239                 return error;
 240
 241         spin_lock(&acct_globals.lock);
 242         acct_file_reopen(file);
 243         spin_unlock(&acct_globals.lock);
 244
 245         return (0);
 246 }
 247
 248 /*
 249  * If the accouting is turned on for a file in the filesystem pointed
 250  * to by sb, turn accouting off.
 251  */
 252 void acct_auto_close(struct super_block *sb)
 253 {
 254         spin_lock(&acct_globals.lock);
 255         if (acct_globals.file &&
 256             acct_globals.file->f_dentry->d_inode->i_sb == sb) {
 257                 acct_file_reopen((struct file *)NULL);
 258         }
 259         spin_unlock(&acct_globals.lock);
 260 }
 261
 262 /*
 263  *  encode an unsigned long into a comp_t
 264  *
 265  *  This routine has been adopted from the encode_comp_t() function in
 266  *  the kern_acct.c file of the FreeBSD operating system. The encoding
 267  *  is a 13-bit fraction with a 3-bit (base 8) exponent.
 268  */
 269
 270 #define MANTSIZE        13                      /* 13 bit mantissa. */
 271 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 272 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 273
 274 static comp_t encode_comp_t(unsigned long value)
 275 {
 276         int exp, rnd;
 277
 278         exp = rnd = 0;
 279         while (value > MAXFRACT) {
 280                 rnd = value & (1 << (EXPSIZE - 1));     /* Round up? */
 281                 value >>= EXPSIZE;      /* Base 8 exponent == 3 bit shift. */
 282                 exp++;
 283         }
 284
 285         /*
 286          * If we need to round up, do it (and handle overflow correctly).
 287          */
 288         if (rnd && (++value > MAXFRACT)) {
 289                 value >>= EXPSIZE;
 290                 exp++;
 291         }
 292
 293         /*
 294          * Clean it up and polish it off.
 295          */
 296         exp <<= MANTSIZE;               /* Shift the exponent into place */
 297         exp += value;                   /* and add on the mantissa. */
 298         return exp;
 299 }
 300
 301 /*
 302  *  Write an accounting entry for an exiting process
 303  *
 304  *  The acct_process() call is the workhorse of the process
 305  *  accounting system. The struct acct is built here and then written
 306  *  into the accounting file. This function should only be called from
 307  *  do_exit().
 308  */
 309
 310 /*
 311  *  do_acct_process does all actual work. Caller holds the reference to file.
 312  */
 313 static void do_acct_process(long exitcode, struct file *file)
 314 {
 315         struct acct ac;
 316         mm_segment_t fs;
 317         unsigned long vsize;
 318         unsigned long flim;
 319         u64 elapsed;
 320
 321         /*
 322          * First check to see if there is enough free_space to continue
 323          * the process accounting system.
 324          */
 325         if (!check_free_space(file))
 326                 return;
 327
 328         /*
 329          * Fill the accounting struct with the needed info as recorded
 330          * by the different kernel functions.
 331          */
 332         memset((caddr_t)&ac, 0, sizeof(struct acct));
 333
 334         strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
 335
 336         elapsed = get_jiffies_64() - current->start_time;
 337         ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
 338                                (unsigned long) elapsed : (unsigned long) -1l);
 339         do_div(elapsed, HZ);
 340         ac.ac_btime = xtime.tv_sec - elapsed;
 341         ac.ac_utime = encode_comp_t(current->utime);
 342         ac.ac_stime = encode_comp_t(current->stime);
 343         ac.ac_uid = current->uid;
 344         ac.ac_gid = current->gid;
 345         ac.ac_tty = current->tty ? current->tty->device : 0;
 346
 347         ac.ac_flag = 0;
 348         if (current->flags & PF_FORKNOEXEC)
 349                 ac.ac_flag |= AFORK;
 350         if (current->flags & PF_SUPERPRIV)
 351                 ac.ac_flag |= ASU;
 352         if (current->flags & PF_DUMPCORE)
 353                 ac.ac_flag |= ACORE;
 354         if (current->flags & PF_SIGNALED)
 355                 ac.ac_flag |= AXSIG;
 356
 357         vsize = 0;
 358         if (current->mm) {
 359                 struct vm_area_struct *vma;
 360                 down_read(&current->mm->mmap_sem);
 361                 vma = current->mm->mmap;
 362                 while (vma) {
 363                         vsize += vma->vm_end - vma->vm_start;
 364                         vma = vma->vm_next;
 365                 }
 366                 up_read(&current->mm->mmap_sem);
 367         }
 368         vsize = vsize / 1024;
 369         ac.ac_mem = encode_comp_t(vsize);
 370         ac.ac_io = encode_comp_t(0 /* current->io_usage */);    /* %% */
 371         ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
 372         ac.ac_minflt = encode_comp_t(current->min_flt);
 373         ac.ac_majflt = encode_comp_t(current->maj_flt);
 374         ac.ac_swaps = encode_comp_t(current->nswap);
 375         ac.ac_exitcode = exitcode;
 376
 377         /*
 378          * Kernel segment override to datasegment and write it
 379          * to the accounting file.
 380          */
 381         fs = get_fs();
 382         set_fs(KERNEL_DS);
 383         /*
 384          * Accounting records are not subject to resource limits.
 385          */
 386         flim = current->rlim[RLIMIT_FSIZE].rlim_cur;
 387         current->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
 388         file->f_op->write(file, (char *)&ac,
 389                                sizeof(struct acct), &file->f_pos);
 390         current->rlim[RLIMIT_FSIZE].rlim_cur = flim;
 391         set_fs(fs);
 392 }
 393
 394 /*
 395  * acct_process - now just a wrapper around do_acct_process
 396  */
 397 void acct_process(long exitcode)
 398 {
 399         struct file *file = NULL;
 400
 401         /*
 402          * accelerate the common fastpath:
 403          */
 404         if (!acct_globals.file)
 405                 return;
 406
 407         spin_lock(&acct_globals.lock);
 408         file = acct_globals.file;
 409         if (unlikely(!file)) {
 410                 spin_unlock(&acct_globals.lock);
 411                 return;
 412         }
 413         get_file(file);
 414         spin_unlock(&acct_globals.lock);
 415
 416         do_acct_process(exitcode, file);
 417         fput(file);
 418 }