4 * BSD Process Accounting for Linux
6 * Author: Marco van Wieringen <mvw@planets.elm.net>
8 * Some code based on ideas and code from:
9 * Thomas K. Dyas <tdyas@eden.rutgers.edu>
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.
17 * (C) Copyright 1995 - 1997 Marco van Wieringen - ELM Consultancy B.V.
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.
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.
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.
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.
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.
46 #include <linux/config.h>
48 #include <linux/slab.h>
49 #include <linux/acct.h>
50 #include <linux/capability.h>
51 #include <linux/file.h>
52 #include <linux/tty.h>
53 #include <linux/security.h>
54 #include <linux/vfs.h>
55 #include <linux/jiffies.h>
56 #include <linux/times.h>
57 #include <linux/syscalls.h>
58 #include <linux/mount.h>
59 #include <asm/uaccess.h>
60 #include <asm/div64.h>
61 #include <linux/blkdev.h> /* sector_div */
64 * These constants control the amount of freespace that suspend and
65 * resume the process accounting system, and the time delay between
67 * Turned into sysctl-controllable parameters. AV, 12/11/98
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 */
76 * External references and all of the globals.
78 static void do_acct_process(long, struct file *);
81 * This structure is used so that all the data protected by lock
82 * can be placed in the same cache line as the lock. This primes
83 * the cache line to have the data after getting the lock.
88 volatile int needcheck;
90 struct timer_list timer;
93 static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED};
96 * Called whenever the timer says to check the free space.
98 static void acct_timeout(unsigned long unused)
100 acct_globals.needcheck = 1;
104 * Check the amount of free space and suspend/resume accordingly.
106 static int check_free_space(struct file *file)
114 spin_lock(&acct_globals.lock);
115 res = acct_globals.active;
116 if (!file || !acct_globals.needcheck)
118 spin_unlock(&acct_globals.lock);
121 if (vfs_statfs(file->f_dentry->d_inode->i_sb, &sbuf))
123 suspend = sbuf.f_blocks * SUSPEND;
124 resume = sbuf.f_blocks * RESUME;
126 sector_div(suspend, 100);
127 sector_div(resume, 100);
129 if (sbuf.f_bavail <= suspend)
131 else if (sbuf.f_bavail >= resume)
137 * If some joker switched acct_globals.file under us we'ld better be
138 * silent and _not_ touch anything.
140 spin_lock(&acct_globals.lock);
141 if (file != acct_globals.file) {
147 if (acct_globals.active) {
149 acct_globals.active = 0;
150 printk(KERN_INFO "Process accounting paused\n");
154 acct_globals.active = 1;
155 printk(KERN_INFO "Process accounting resumed\n");
159 del_timer(&acct_globals.timer);
160 acct_globals.needcheck = 0;
161 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
162 add_timer(&acct_globals.timer);
163 res = acct_globals.active;
165 spin_unlock(&acct_globals.lock);
170 * Close the old accounting file (if currently open) and then replace
171 * it with file (if non-NULL).
173 * NOTE: acct_globals.lock MUST be held on entry and exit.
175 static void acct_file_reopen(struct file *file)
177 struct file *old_acct = NULL;
179 if (acct_globals.file) {
180 old_acct = acct_globals.file;
181 del_timer(&acct_globals.timer);
182 acct_globals.active = 0;
183 acct_globals.needcheck = 0;
184 acct_globals.file = NULL;
187 acct_globals.file = file;
188 acct_globals.needcheck = 0;
189 acct_globals.active = 1;
190 /* It's been deleted if it was used before so this is safe */
191 init_timer(&acct_globals.timer);
192 acct_globals.timer.function = acct_timeout;
193 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
194 add_timer(&acct_globals.timer);
197 mnt_unpin(old_acct->f_vfsmnt);
198 spin_unlock(&acct_globals.lock);
199 do_acct_process(0, old_acct);
200 filp_close(old_acct, NULL);
201 spin_lock(&acct_globals.lock);
205 static int acct_on(char *name)
210 /* Difference from BSD - they don't do O_APPEND */
211 file = filp_open(name, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
213 return PTR_ERR(file);
215 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
216 filp_close(file, NULL);
220 if (!file->f_op->write) {
221 filp_close(file, NULL);
225 error = security_acct(file);
227 filp_close(file, NULL);
231 spin_lock(&acct_globals.lock);
232 mnt_pin(file->f_vfsmnt);
233 acct_file_reopen(file);
234 spin_unlock(&acct_globals.lock);
236 mntput(file->f_vfsmnt); /* it's pinned, now give up active reference */
242 * sys_acct - enable/disable process accounting
243 * @name: file name for accounting records or NULL to shutdown accounting
245 * Returns 0 for success or negative errno values for failure.
247 * sys_acct() is the only system call needed to implement process
248 * accounting. It takes the name of the file where accounting records
249 * should be written. If the filename is NULL, accounting will be
252 asmlinkage long sys_acct(const char __user *name)
256 if (!capable(CAP_SYS_PACCT))
260 char *tmp = getname(name);
262 return (PTR_ERR(tmp));
263 error = acct_on(tmp);
266 error = security_acct(NULL);
268 spin_lock(&acct_globals.lock);
269 acct_file_reopen(NULL);
270 spin_unlock(&acct_globals.lock);
277 * acct_auto_close - turn off a filesystem's accounting if it is on
278 * @m: vfsmount being shut down
280 * If the accounting is turned on for a file in the subtree pointed to
281 * to by m, turn accounting off. Done when m is about to die.
283 void acct_auto_close_mnt(struct vfsmount *m)
285 spin_lock(&acct_globals.lock);
286 if (acct_globals.file && acct_globals.file->f_vfsmnt == m)
287 acct_file_reopen(NULL);
288 spin_unlock(&acct_globals.lock);
292 * acct_auto_close - turn off a filesystem's accounting if it is on
293 * @sb: super block for the filesystem
295 * If the accounting is turned on for a file in the filesystem pointed
296 * to by sb, turn accounting off.
298 void acct_auto_close(struct super_block *sb)
300 spin_lock(&acct_globals.lock);
301 if (acct_globals.file &&
302 acct_globals.file->f_vfsmnt->mnt_sb == sb) {
303 acct_file_reopen(NULL);
305 spin_unlock(&acct_globals.lock);
309 * encode an unsigned long into a comp_t
311 * This routine has been adopted from the encode_comp_t() function in
312 * the kern_acct.c file of the FreeBSD operating system. The encoding
313 * is a 13-bit fraction with a 3-bit (base 8) exponent.
316 #define MANTSIZE 13 /* 13 bit mantissa. */
317 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
318 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
320 static comp_t encode_comp_t(unsigned long value)
325 while (value > MAXFRACT) {
326 rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */
327 value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
332 * If we need to round up, do it (and handle overflow correctly).
334 if (rnd && (++value > MAXFRACT)) {
340 * Clean it up and polish it off.
342 exp <<= MANTSIZE; /* Shift the exponent into place */
343 exp += value; /* and add on the mantissa. */
347 #if ACCT_VERSION==1 || ACCT_VERSION==2
349 * encode an u64 into a comp2_t (24 bits)
351 * Format: 5 bit base 2 exponent, 20 bits mantissa.
352 * The leading bit of the mantissa is not stored, but implied for
353 * non-zero exponents.
354 * Largest encodable value is 50 bits.
357 #define MANTSIZE2 20 /* 20 bit mantissa. */
358 #define EXPSIZE2 5 /* 5 bit base 2 exponent. */
359 #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
360 #define MAXEXP2 ((1 <<EXPSIZE2) - 1) /* Maximum exponent. */
362 static comp2_t encode_comp2_t(u64 value)
366 exp = (value > (MAXFRACT2>>1));
368 while (value > MAXFRACT2) {
375 * If we need to round up, do it (and handle overflow correctly).
377 if (rnd && (++value > MAXFRACT2)) {
383 /* Overflow. Return largest representable number instead. */
384 return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
386 return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
393 * encode an u64 into a 32 bit IEEE float
395 static u32 encode_float(u64 value)
400 if (value==0) return 0;
401 while ((s64)value > 0){
405 u = (u32)(value >> 40) & 0x7fffffu;
406 return u | (exp << 23);
411 * Write an accounting entry for an exiting process
413 * The acct_process() call is the workhorse of the process
414 * accounting system. The struct acct is built here and then written
415 * into the accounting file. This function should only be called from
420 * do_acct_process does all actual work. Caller holds the reference to file.
422 static void do_acct_process(long exitcode, struct file *file)
430 struct timespec uptime;
431 unsigned long jiffies;
434 * First check to see if there is enough free_space to continue
435 * the process accounting system.
437 if (!check_free_space(file))
441 * Fill the accounting struct with the needed info as recorded
442 * by the different kernel functions.
444 memset((caddr_t)&ac, 0, sizeof(acct_t));
446 ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER;
447 strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
449 /* calculate run_time in nsec*/
450 do_posix_clock_monotonic_gettime(&uptime);
451 run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec;
452 run_time -= (u64)current->group_leader->start_time.tv_sec * NSEC_PER_SEC
453 + current->group_leader->start_time.tv_nsec;
454 /* convert nsec -> AHZ */
455 elapsed = nsec_to_AHZ(run_time);
457 ac.ac_etime = encode_float(elapsed);
459 ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
460 (unsigned long) elapsed : (unsigned long) -1l);
462 #if ACCT_VERSION==1 || ACCT_VERSION==2
464 /* new enlarged etime field */
465 comp2_t etime = encode_comp2_t(elapsed);
466 ac.ac_etime_hi = etime >> 16;
467 ac.ac_etime_lo = (u16) etime;
470 do_div(elapsed, AHZ);
471 ac.ac_btime = xtime.tv_sec - elapsed;
472 jiffies = cputime_to_jiffies(cputime_add(current->utime,
473 current->signal->utime));
474 ac.ac_utime = encode_comp_t(jiffies_to_AHZ(jiffies));
475 jiffies = cputime_to_jiffies(cputime_add(current->stime,
476 current->signal->stime));
477 ac.ac_stime = encode_comp_t(jiffies_to_AHZ(jiffies));
478 /* we really need to bite the bullet and change layout */
479 ac.ac_uid = current->uid;
480 ac.ac_gid = current->gid;
484 #if ACCT_VERSION==1 || ACCT_VERSION==2
485 /* backward-compatible 16 bit fields */
486 ac.ac_uid16 = current->uid;
487 ac.ac_gid16 = current->gid;
490 ac.ac_pid = current->tgid;
491 ac.ac_ppid = current->parent->tgid;
494 read_lock(&tasklist_lock); /* pin current->signal */
495 ac.ac_tty = current->signal->tty ?
496 old_encode_dev(tty_devnum(current->signal->tty)) : 0;
497 read_unlock(&tasklist_lock);
500 if (current->flags & PF_FORKNOEXEC)
502 if (current->flags & PF_SUPERPRIV)
504 if (current->flags & PF_DUMPCORE)
506 if (current->flags & PF_SIGNALED)
511 struct vm_area_struct *vma;
512 down_read(¤t->mm->mmap_sem);
513 vma = current->mm->mmap;
515 vsize += vma->vm_end - vma->vm_start;
518 up_read(¤t->mm->mmap_sem);
520 vsize = vsize / 1024;
521 ac.ac_mem = encode_comp_t(vsize);
522 ac.ac_io = encode_comp_t(0 /* current->io_usage */); /* %% */
523 ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
524 ac.ac_minflt = encode_comp_t(current->signal->min_flt +
526 ac.ac_majflt = encode_comp_t(current->signal->maj_flt +
528 ac.ac_swaps = encode_comp_t(0);
529 ac.ac_exitcode = exitcode;
532 * Kernel segment override to datasegment and write it
533 * to the accounting file.
538 * Accounting records are not subject to resource limits.
540 flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
541 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
542 file->f_op->write(file, (char *)&ac,
543 sizeof(acct_t), &file->f_pos);
544 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
549 * acct_process - now just a wrapper around do_acct_process
550 * @exitcode: task exit code
552 * handles process accounting for an exiting task
554 void acct_process(long exitcode)
556 struct file *file = NULL;
559 * accelerate the common fastpath:
561 if (!acct_globals.file)
564 spin_lock(&acct_globals.lock);
565 file = acct_globals.file;
566 if (unlikely(!file)) {
567 spin_unlock(&acct_globals.lock);
571 spin_unlock(&acct_globals.lock);
573 do_acct_process(exitcode, file);
579 * acct_update_integrals - update mm integral fields in task_struct
580 * @tsk: task_struct for accounting
582 void acct_update_integrals(struct task_struct *tsk)
584 if (likely(tsk->mm)) {
586 cputime_to_jiffies(tsk->stime) - tsk->acct_stimexpd;
590 tsk->acct_stimexpd = tsk->stime;
591 tsk->acct_rss_mem1 += delta * get_mm_rss(tsk->mm);
592 tsk->acct_vm_mem1 += delta * tsk->mm->total_vm;
597 * acct_clear_integrals - clear the mm integral fields in task_struct
598 * @tsk: task_struct whose accounting fields are cleared
600 void acct_clear_integrals(struct task_struct *tsk)
603 tsk->acct_stimexpd = 0;
604 tsk->acct_rss_mem1 = 0;
605 tsk->acct_vm_mem1 = 0;