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/mm.h>
  47 #include <linux/slab.h>
  48 #include <linux/acct.h>
  49 #include <linux/capability.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 <linux/times.h>
  56 #include <linux/syscalls.h>
  57 #include <linux/mount.h>
  58 #include <asm/uaccess.h>
  59 #include <asm/div64.h>
  60 #include <linux/blkdev.h> /* sector_div */
  61 #include <linux/pid_namespace.h>
  62
  63 /*
  64  * These constants control the amount of freespace that suspend and
  65  * resume the process accounting system, and the time delay between
  66  * each check.
  67  * Turned into sysctl-controllable parameters. AV, 12/11/98
  68  */
  69
  70 int acct_parm[3] = {4, 2, 30};
  71 #define RESUME          (acct_parm[0])  /* >foo% free space - resume */
  72 #define SUSPEND         (acct_parm[1])  /* <foo% free space - suspend */
  73 #define ACCT_TIMEOUT    (acct_parm[2])  /* foo second timeout between checks */
  74
  75 /*
  76  * External references and all of the globals.
  77  */
  78 static void do_acct_process(struct bsd_acct_struct *acct,
  79                 struct pid_namespace *ns, struct file *);
  80
  81 /*
  82  * This structure is used so that all the data protected by lock
  83  * can be placed in the same cache line as the lock.  This primes
  84  * the cache line to have the data after getting the lock.
  85  */
  86 struct bsd_acct_struct {
  87         volatile int            active;
  88         volatile int            needcheck;
  89         struct file             *file;
  90         struct pid_namespace    *ns;
  91         struct timer_list       timer;
  92 };
  93
  94 static DEFINE_SPINLOCK(acct_lock);
  95
  96 static struct bsd_acct_struct acct_globals __cacheline_aligned;
  97
  98 /*
  99  * Called whenever the timer says to check the free space.
 100  */
 101 static void acct_timeout(unsigned long x)
 102 {
 103         struct bsd_acct_struct *acct = (struct bsd_acct_struct *)x;
 104         acct->needcheck = 1;
 105 }
 106
 107 /*
 108  * Check the amount of free space and suspend/resume accordingly.
 109  */
 110 static int check_free_space(struct bsd_acct_struct *acct, struct file *file)
 111 {
 112         struct kstatfs sbuf;
 113         int res;
 114         int act;
 115         sector_t resume;
 116         sector_t suspend;
 117
 118         spin_lock(&acct_lock);
 119         res = acct->active;
 120         if (!file || !acct->needcheck)
 121                 goto out;
 122         spin_unlock(&acct_lock);
 123
 124         /* May block */
 125         if (vfs_statfs(file->f_path.dentry, &sbuf))
 126                 return res;
 127         suspend = sbuf.f_blocks * SUSPEND;
 128         resume = sbuf.f_blocks * RESUME;
 129
 130         sector_div(suspend, 100);
 131         sector_div(resume, 100);
 132
 133         if (sbuf.f_bavail <= suspend)
 134                 act = -1;
 135         else if (sbuf.f_bavail >= resume)
 136                 act = 1;
 137         else
 138                 act = 0;
 139
 140         /*
 141          * If some joker switched acct->file under us we'ld better be
 142          * silent and _not_ touch anything.
 143          */
 144         spin_lock(&acct_lock);
 145         if (file != acct->file) {
 146                 if (act)
 147                         res = act>0;
 148                 goto out;
 149         }
 150
 151         if (acct->active) {
 152                 if (act < 0) {
 153                         acct->active = 0;
 154                         printk(KERN_INFO "Process accounting paused\n");
 155                 }
 156         } else {
 157                 if (act > 0) {
 158                         acct->active = 1;
 159                         printk(KERN_INFO "Process accounting resumed\n");
 160                 }
 161         }
 162
 163         del_timer(&acct->timer);
 164         acct->needcheck = 0;
 165         acct->timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 166         add_timer(&acct->timer);
 167         res = acct->active;
 168 out:
 169         spin_unlock(&acct_lock);
 170         return res;
 171 }
 172
 173 /*
 174  * Close the old accounting file (if currently open) and then replace
 175  * it with file (if non-NULL).
 176  *
 177  * NOTE: acct_lock MUST be held on entry and exit.
 178  */
 179 static void acct_file_reopen(struct bsd_acct_struct *acct, struct file *file)
 180 {
 181         struct file *old_acct = NULL;
 182         struct pid_namespace *old_ns = NULL;
 183
 184         if (acct->file) {
 185                 old_acct = acct->file;
 186                 old_ns = acct->ns;
 187                 del_timer(&acct->timer);
 188                 acct->active = 0;
 189                 acct->needcheck = 0;
 190                 acct->file = NULL;
 191         }
 192         if (file) {
 193                 acct->file = file;
 194                 acct->ns = get_pid_ns(task_active_pid_ns(current));
 195                 acct->needcheck = 0;
 196                 acct->active = 1;
 197                 /* It's been deleted if it was used before so this is safe */
 198                 setup_timer(&acct->timer, acct_timeout, (unsigned long)acct);
 199                 acct->timer.expires = jiffies + ACCT_TIMEOUT*HZ;
 200                 add_timer(&acct->timer);
 201         }
 202         if (old_acct) {
 203                 mnt_unpin(old_acct->f_path.mnt);
 204                 spin_unlock(&acct_lock);
 205                 do_acct_process(acct, old_ns, old_acct);
 206                 filp_close(old_acct, NULL);
 207                 put_pid_ns(old_ns);
 208                 spin_lock(&acct_lock);
 209         }
 210 }
 211
 212 static int acct_on(char *name)
 213 {
 214         struct file *file;
 215         int error;
 216
 217         /* Difference from BSD - they don't do O_APPEND */
 218         file = filp_open(name, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
 219         if (IS_ERR(file))
 220                 return PTR_ERR(file);
 221
 222         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) {
 223                 filp_close(file, NULL);
 224                 return -EACCES;
 225         }
 226
 227         if (!file->f_op->write) {
 228                 filp_close(file, NULL);
 229                 return -EIO;
 230         }
 231
 232         error = security_acct(file);
 233         if (error) {
 234                 filp_close(file, NULL);
 235                 return error;
 236         }
 237
 238         spin_lock(&acct_lock);
 239         mnt_pin(file->f_path.mnt);
 240         acct_file_reopen(&acct_globals, file);
 241         spin_unlock(&acct_lock);
 242
 243         mntput(file->f_path.mnt); /* it's pinned, now give up active reference */
 244
 245         return 0;
 246 }
 247
 248 /**
 249  * sys_acct - enable/disable process accounting
 250  * @name: file name for accounting records or NULL to shutdown accounting
 251  *
 252  * Returns 0 for success or negative errno values for failure.
 253  *
 254  * sys_acct() is the only system call needed to implement process
 255  * accounting. It takes the name of the file where accounting records
 256  * should be written. If the filename is NULL, accounting will be
 257  * shutdown.
 258  */
 259 asmlinkage long sys_acct(const char __user *name)
 260 {
 261         int error;
 262
 263         if (!capable(CAP_SYS_PACCT))
 264                 return -EPERM;
 265
 266         if (name) {
 267                 char *tmp = getname(name);
 268                 if (IS_ERR(tmp))
 269                         return (PTR_ERR(tmp));
 270                 error = acct_on(tmp);
 271                 putname(tmp);
 272         } else {
 273                 error = security_acct(NULL);
 274                 if (!error) {
 275                         spin_lock(&acct_lock);
 276                         acct_file_reopen(&acct_globals, NULL);
 277                         spin_unlock(&acct_lock);
 278                 }
 279         }
 280         return error;
 281 }
 282
 283 /**
 284  * acct_auto_close - turn off a filesystem's accounting if it is on
 285  * @m: vfsmount being shut down
 286  *
 287  * If the accounting is turned on for a file in the subtree pointed to
 288  * to by m, turn accounting off.  Done when m is about to die.
 289  */
 290 void acct_auto_close_mnt(struct vfsmount *m)
 291 {
 292         spin_lock(&acct_lock);
 293         if (acct_globals.file && acct_globals.file->f_path.mnt == m)
 294                 acct_file_reopen(&acct_globals, NULL);
 295         spin_unlock(&acct_lock);
 296 }
 297
 298 /**
 299  * acct_auto_close - turn off a filesystem's accounting if it is on
 300  * @sb: super block for the filesystem
 301  *
 302  * If the accounting is turned on for a file in the filesystem pointed
 303  * to by sb, turn accounting off.
 304  */
 305 void acct_auto_close(struct super_block *sb)
 306 {
 307         spin_lock(&acct_lock);
 308         if (acct_globals.file &&
 309             acct_globals.file->f_path.mnt->mnt_sb == sb) {
 310                 acct_file_reopen(&acct_globals, NULL);
 311         }
 312         spin_unlock(&acct_lock);
 313 }
 314
 315 /*
 316  *  encode an unsigned long into a comp_t
 317  *
 318  *  This routine has been adopted from the encode_comp_t() function in
 319  *  the kern_acct.c file of the FreeBSD operating system. The encoding
 320  *  is a 13-bit fraction with a 3-bit (base 8) exponent.
 321  */
 322
 323 #define MANTSIZE        13                      /* 13 bit mantissa. */
 324 #define EXPSIZE         3                       /* Base 8 (3 bit) exponent. */
 325 #define MAXFRACT        ((1 << MANTSIZE) - 1)   /* Maximum fractional value. */
 326
 327 static comp_t encode_comp_t(unsigned long value)
 328 {
 329         int exp, rnd;
 330
 331         exp = rnd = 0;
 332         while (value > MAXFRACT) {
 333                 rnd = value & (1 << (EXPSIZE - 1));     /* Round up? */
 334                 value >>= EXPSIZE;      /* Base 8 exponent == 3 bit shift. */
 335                 exp++;
 336         }
 337
 338         /*
 339          * If we need to round up, do it (and handle overflow correctly).
 340          */
 341         if (rnd && (++value > MAXFRACT)) {
 342                 value >>= EXPSIZE;
 343                 exp++;
 344         }
 345
 346         /*
 347          * Clean it up and polish it off.
 348          */
 349         exp <<= MANTSIZE;               /* Shift the exponent into place */
 350         exp += value;                   /* and add on the mantissa. */
 351         return exp;
 352 }
 353
 354 #if ACCT_VERSION==1 || ACCT_VERSION==2
 355 /*
 356  * encode an u64 into a comp2_t (24 bits)
 357  *
 358  * Format: 5 bit base 2 exponent, 20 bits mantissa.
 359  * The leading bit of the mantissa is not stored, but implied for
 360  * non-zero exponents.
 361  * Largest encodable value is 50 bits.
 362  */
 363
 364 #define MANTSIZE2       20                      /* 20 bit mantissa. */
 365 #define EXPSIZE2        5                       /* 5 bit base 2 exponent. */
 366 #define MAXFRACT2       ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
 367 #define MAXEXP2         ((1 <<EXPSIZE2) - 1)    /* Maximum exponent. */
 368
 369 static comp2_t encode_comp2_t(u64 value)
 370 {
 371         int exp, rnd;
 372
 373         exp = (value > (MAXFRACT2>>1));
 374         rnd = 0;
 375         while (value > MAXFRACT2) {
 376                 rnd = value & 1;
 377                 value >>= 1;
 378                 exp++;
 379         }
 380
 381         /*
 382          * If we need to round up, do it (and handle overflow correctly).
 383          */
 384         if (rnd && (++value > MAXFRACT2)) {
 385                 value >>= 1;
 386                 exp++;
 387         }
 388
 389         if (exp > MAXEXP2) {
 390                 /* Overflow. Return largest representable number instead. */
 391                 return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
 392         } else {
 393                 return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
 394         }
 395 }
 396 #endif
 397
 398 #if ACCT_VERSION==3
 399 /*
 400  * encode an u64 into a 32 bit IEEE float
 401  */
 402 static u32 encode_float(u64 value)
 403 {
 404         unsigned exp = 190;
 405         unsigned u;
 406
 407         if (value==0) return 0;
 408         while ((s64)value > 0){
 409                 value <<= 1;
 410                 exp--;
 411         }
 412         u = (u32)(value >> 40) & 0x7fffffu;
 413         return u | (exp << 23);
 414 }
 415 #endif
 416
 417 /*
 418  *  Write an accounting entry for an exiting process
 419  *
 420  *  The acct_process() call is the workhorse of the process
 421  *  accounting system. The struct acct is built here and then written
 422  *  into the accounting file. This function should only be called from
 423  *  do_exit() or when switching to a different output file.
 424  */
 425
 426 /*
 427  *  do_acct_process does all actual work. Caller holds the reference to file.
 428  */
 429 static void do_acct_process(struct bsd_acct_struct *acct,
 430                 struct pid_namespace *ns, struct file *file)
 431 {
 432         struct pacct_struct *pacct = &current->signal->pacct;
 433         acct_t ac;
 434         mm_segment_t fs;
 435         unsigned long flim;
 436         u64 elapsed;
 437         u64 run_time;
 438         struct timespec uptime;
 439         struct tty_struct *tty;
 440
 441         /*
 442          * First check to see if there is enough free_space to continue
 443          * the process accounting system.
 444          */
 445         if (!check_free_space(acct, file))
 446                 return;
 447
 448         /*
 449          * Fill the accounting struct with the needed info as recorded
 450          * by the different kernel functions.
 451          */
 452         memset((caddr_t)&ac, 0, sizeof(acct_t));
 453
 454         ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER;
 455         strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
 456
 457         /* calculate run_time in nsec*/
 458         do_posix_clock_monotonic_gettime(&uptime);
 459         run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec;
 460         run_time -= (u64)current->group_leader->start_time.tv_sec * NSEC_PER_SEC
 461                        + current->group_leader->start_time.tv_nsec;
 462         /* convert nsec -> AHZ */
 463         elapsed = nsec_to_AHZ(run_time);
 464 #if ACCT_VERSION==3
 465         ac.ac_etime = encode_float(elapsed);
 466 #else
 467         ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
 468                                (unsigned long) elapsed : (unsigned long) -1l);
 469 #endif
 470 #if ACCT_VERSION==1 || ACCT_VERSION==2
 471         {
 472                 /* new enlarged etime field */
 473                 comp2_t etime = encode_comp2_t(elapsed);
 474                 ac.ac_etime_hi = etime >> 16;
 475                 ac.ac_etime_lo = (u16) etime;
 476         }
 477 #endif
 478         do_div(elapsed, AHZ);
 479         ac.ac_btime = get_seconds() - elapsed;
 480         /* we really need to bite the bullet and change layout */
 481         ac.ac_uid = current->uid;
 482         ac.ac_gid = current->gid;
 483 #if ACCT_VERSION==2
 484         ac.ac_ahz = AHZ;
 485 #endif
 486 #if ACCT_VERSION==1 || ACCT_VERSION==2
 487         /* backward-compatible 16 bit fields */
 488         ac.ac_uid16 = current->uid;
 489         ac.ac_gid16 = current->gid;
 490 #endif
 491 #if ACCT_VERSION==3
 492         ac.ac_pid = task_tgid_nr_ns(current, ns);
 493         rcu_read_lock();
 494         ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent), ns);
 495         rcu_read_unlock();
 496 #endif
 497
 498         spin_lock_irq(&current->sighand->siglock);
 499         tty = current->signal->tty;
 500         ac.ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
 501         ac.ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime)));
 502         ac.ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime)));
 503         ac.ac_flag = pacct->ac_flag;
 504         ac.ac_mem = encode_comp_t(pacct->ac_mem);
 505         ac.ac_minflt = encode_comp_t(pacct->ac_minflt);
 506         ac.ac_majflt = encode_comp_t(pacct->ac_majflt);
 507         ac.ac_exitcode = pacct->ac_exitcode;
 508         spin_unlock_irq(&current->sighand->siglock);
 509         ac.ac_io = encode_comp_t(0 /* current->io_usage */);    /* %% */
 510         ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
 511         ac.ac_swaps = encode_comp_t(0);
 512
 513         /*
 514          * Kernel segment override to datasegment and write it
 515          * to the accounting file.
 516          */
 517         fs = get_fs();
 518         set_fs(KERNEL_DS);
 519         /*
 520          * Accounting records are not subject to resource limits.
 521          */
 522         flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
 523         current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
 524         file->f_op->write(file, (char *)&ac,
 525                                sizeof(acct_t), &file->f_pos);
 526         current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
 527         set_fs(fs);
 528 }
 529
 530 /**
 531  * acct_init_pacct - initialize a new pacct_struct
 532  * @pacct: per-process accounting info struct to initialize
 533  */
 534 void acct_init_pacct(struct pacct_struct *pacct)
 535 {
 536         memset(pacct, 0, sizeof(struct pacct_struct));
 537         pacct->ac_utime = pacct->ac_stime = cputime_zero;
 538 }
 539
 540 /**
 541  * acct_collect - collect accounting information into pacct_struct
 542  * @exitcode: task exit code
 543  * @group_dead: not 0, if this thread is the last one in the process.
 544  */
 545 void acct_collect(long exitcode, int group_dead)
 546 {
 547         struct pacct_struct *pacct = &current->signal->pacct;
 548         unsigned long vsize = 0;
 549
 550         if (group_dead && current->mm) {
 551                 struct vm_area_struct *vma;
 552                 down_read(&current->mm->mmap_sem);
 553                 vma = current->mm->mmap;
 554                 while (vma) {
 555                         vsize += vma->vm_end - vma->vm_start;
 556                         vma = vma->vm_next;
 557                 }
 558                 up_read(&current->mm->mmap_sem);
 559         }
 560
 561         spin_lock_irq(&current->sighand->siglock);
 562         if (group_dead)
 563                 pacct->ac_mem = vsize / 1024;
 564         if (thread_group_leader(current)) {
 565                 pacct->ac_exitcode = exitcode;
 566                 if (current->flags & PF_FORKNOEXEC)
 567                         pacct->ac_flag |= AFORK;
 568         }
 569         if (current->flags & PF_SUPERPRIV)
 570                 pacct->ac_flag |= ASU;
 571         if (current->flags & PF_DUMPCORE)
 572                 pacct->ac_flag |= ACORE;
 573         if (current->flags & PF_SIGNALED)
 574                 pacct->ac_flag |= AXSIG;
 575         pacct->ac_utime = cputime_add(pacct->ac_utime, current->utime);
 576         pacct->ac_stime = cputime_add(pacct->ac_stime, current->stime);
 577         pacct->ac_minflt += current->min_flt;
 578         pacct->ac_majflt += current->maj_flt;
 579         spin_unlock_irq(&current->sighand->siglock);
 580 }
 581
 582 /**
 583  * acct_process - now just a wrapper around do_acct_process
 584  *
 585  * handles process accounting for an exiting task
 586  */
 587 void acct_process(void)
 588 {
 589         struct file *file = NULL;
 590         struct pid_namespace *ns;
 591
 592         /*
 593          * accelerate the common fastpath:
 594          */
 595         if (!acct_globals.file)
 596                 return;
 597
 598         spin_lock(&acct_lock);
 599         file = acct_globals.file;
 600         if (unlikely(!file)) {
 601                 spin_unlock(&acct_lock);
 602                 return;
 603         }
 604         get_file(file);
 605         ns = get_pid_ns(acct_globals.ns);
 606         spin_unlock(&acct_lock);
 607
 608         do_acct_process(&acct_globals, ns, file);
 609         fput(file);
 610         put_pid_ns(ns);
 611 }