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/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 <asm/uaccess.h>
57 #include <asm/div64.h>
58 #include <linux/blkdev.h> /* sector_div */
61 * These constants control the amount of freespace that suspend and
62 * resume the process accounting system, and the time delay between
64 * Turned into sysctl-controllable parameters. AV, 12/11/98
67 int acct_parm[3] = {4, 2, 30};
68 #define RESUME (acct_parm[0]) /* >foo% free space - resume */
69 #define SUSPEND (acct_parm[1]) /* <foo% free space - suspend */
70 #define ACCT_TIMEOUT (acct_parm[2]) /* foo second timeout between checks */
73 * External references and all of the globals.
75 static void do_acct_process(long, struct file *);
78 * This structure is used so that all the data protected by lock
79 * can be placed in the same cache line as the lock. This primes
80 * the cache line to have the data after getting the lock.
85 volatile int needcheck;
87 struct timer_list timer;
90 static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED};
93 * Called whenever the timer says to check the free space.
95 static void acct_timeout(unsigned long unused)
97 acct_globals.needcheck = 1;
101 * Check the amount of free space and suspend/resume accordingly.
103 static int check_free_space(struct file *file)
111 spin_lock(&acct_globals.lock);
112 res = acct_globals.active;
113 if (!file || !acct_globals.needcheck)
115 spin_unlock(&acct_globals.lock);
118 if (vfs_statfs(file->f_dentry->d_inode->i_sb, &sbuf))
120 suspend = sbuf.f_blocks * SUSPEND;
121 resume = sbuf.f_blocks * RESUME;
123 sector_div(suspend, 100);
124 sector_div(resume, 100);
126 if (sbuf.f_bavail <= suspend)
128 else if (sbuf.f_bavail >= resume)
134 * If some joker switched acct_globals.file under us we'ld better be
135 * silent and _not_ touch anything.
137 spin_lock(&acct_globals.lock);
138 if (file != acct_globals.file) {
144 if (acct_globals.active) {
146 acct_globals.active = 0;
147 printk(KERN_INFO "Process accounting paused\n");
151 acct_globals.active = 1;
152 printk(KERN_INFO "Process accounting resumed\n");
156 del_timer(&acct_globals.timer);
157 acct_globals.needcheck = 0;
158 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
159 add_timer(&acct_globals.timer);
160 res = acct_globals.active;
162 spin_unlock(&acct_globals.lock);
167 * Close the old accouting file (if currently open) and then replace
168 * it with file (if non-NULL).
170 * NOTE: acct_globals.lock MUST be held on entry and exit.
172 void acct_file_reopen(struct file *file)
174 struct file *old_acct = NULL;
176 if (acct_globals.file) {
177 old_acct = acct_globals.file;
178 del_timer(&acct_globals.timer);
179 acct_globals.active = 0;
180 acct_globals.needcheck = 0;
181 acct_globals.file = NULL;
184 acct_globals.file = file;
185 acct_globals.needcheck = 0;
186 acct_globals.active = 1;
187 /* It's been deleted if it was used before so this is safe */
188 init_timer(&acct_globals.timer);
189 acct_globals.timer.function = acct_timeout;
190 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
191 add_timer(&acct_globals.timer);
194 spin_unlock(&acct_globals.lock);
195 do_acct_process(0, old_acct);
196 filp_close(old_acct, NULL);
197 spin_lock(&acct_globals.lock);
202 * sys_acct() is the only system call needed to implement process
203 * accounting. It takes the name of the file where accounting records
204 * should be written. If the filename is NULL, accounting will be
207 asmlinkage long sys_acct(const char *name)
209 struct file *file = NULL;
213 if (!capable(CAP_SYS_PACCT))
219 return (PTR_ERR(tmp));
221 /* Difference from BSD - they don't do O_APPEND */
222 file = filp_open(tmp, O_WRONLY|O_APPEND, 0);
225 return (PTR_ERR(file));
227 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
228 filp_close(file, NULL);
232 if (!file->f_op->write) {
233 filp_close(file, NULL);
238 error = security_acct(file);
241 filp_close(file, NULL);
245 spin_lock(&acct_globals.lock);
246 acct_file_reopen(file);
247 spin_unlock(&acct_globals.lock);
253 * If the accouting is turned on for a file in the filesystem pointed
254 * to by sb, turn accouting off.
256 void acct_auto_close(struct super_block *sb)
258 spin_lock(&acct_globals.lock);
259 if (acct_globals.file &&
260 acct_globals.file->f_dentry->d_inode->i_sb == sb) {
261 acct_file_reopen((struct file *)NULL);
263 spin_unlock(&acct_globals.lock);
267 * encode an unsigned long into a comp_t
269 * This routine has been adopted from the encode_comp_t() function in
270 * the kern_acct.c file of the FreeBSD operating system. The encoding
271 * is a 13-bit fraction with a 3-bit (base 8) exponent.
274 #define MANTSIZE 13 /* 13 bit mantissa. */
275 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
276 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
278 static comp_t encode_comp_t(unsigned long value)
283 while (value > MAXFRACT) {
284 rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */
285 value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
290 * If we need to round up, do it (and handle overflow correctly).
292 if (rnd && (++value > MAXFRACT)) {
298 * Clean it up and polish it off.
300 exp <<= MANTSIZE; /* Shift the exponent into place */
301 exp += value; /* and add on the mantissa. */
306 * Write an accounting entry for an exiting process
308 * The acct_process() call is the workhorse of the process
309 * accounting system. The struct acct is built here and then written
310 * into the accounting file. This function should only be called from
315 * do_acct_process does all actual work. Caller holds the reference to file.
317 static void do_acct_process(long exitcode, struct file *file)
326 * First check to see if there is enough free_space to continue
327 * the process accounting system.
329 if (!check_free_space(file))
333 * Fill the accounting struct with the needed info as recorded
334 * by the different kernel functions.
336 memset((caddr_t)&ac, 0, sizeof(struct acct));
338 strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
340 elapsed = jiffies_64_to_clock_t(get_jiffies_64() - current->start_time);
341 ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
342 (unsigned long) elapsed : (unsigned long) -1l);
343 do_div(elapsed, USER_HZ);
344 ac.ac_btime = xtime.tv_sec - elapsed;
345 ac.ac_utime = encode_comp_t(jiffies_to_clock_t(current->utime));
346 ac.ac_stime = encode_comp_t(jiffies_to_clock_t(current->stime));
347 /* we really need to bite the bullet and change layout */
348 ac.ac_uid = current->uid;
349 ac.ac_gid = current->gid;
351 read_lock(&tasklist_lock); /* pin current->signal */
352 ac.ac_tty = current->signal->tty ?
353 old_encode_dev(tty_devnum(current->signal->tty)) : 0;
354 read_unlock(&tasklist_lock);
357 if (current->flags & PF_FORKNOEXEC)
359 if (current->flags & PF_SUPERPRIV)
361 if (current->flags & PF_DUMPCORE)
363 if (current->flags & PF_SIGNALED)
368 struct vm_area_struct *vma;
369 down_read(¤t->mm->mmap_sem);
370 vma = current->mm->mmap;
372 vsize += vma->vm_end - vma->vm_start;
375 up_read(¤t->mm->mmap_sem);
377 vsize = vsize / 1024;
378 ac.ac_mem = encode_comp_t(vsize);
379 ac.ac_io = encode_comp_t(0 /* current->io_usage */); /* %% */
380 ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
381 ac.ac_minflt = encode_comp_t(current->min_flt);
382 ac.ac_majflt = encode_comp_t(current->maj_flt);
383 ac.ac_swaps = encode_comp_t(0);
384 ac.ac_exitcode = exitcode;
387 * Kernel segment override to datasegment and write it
388 * to the accounting file.
393 * Accounting records are not subject to resource limits.
395 flim = current->rlim[RLIMIT_FSIZE].rlim_cur;
396 current->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
397 file->f_op->write(file, (char *)&ac,
398 sizeof(struct acct), &file->f_pos);
399 current->rlim[RLIMIT_FSIZE].rlim_cur = flim;
404 * acct_process - now just a wrapper around do_acct_process
406 void acct_process(long exitcode)
408 struct file *file = NULL;
411 * accelerate the common fastpath:
413 if (!acct_globals.file)
416 spin_lock(&acct_globals.lock);
417 file = acct_globals.file;
418 if (unlikely(!file)) {
419 spin_unlock(&acct_globals.lock);
423 spin_unlock(&acct_globals.lock);
425 do_acct_process(exitcode, file);