[PATCH] hugetlb: fix absurd HugePages_Rsvd
[linux-2.6.22.y-op.git] / kernel / acct.c
blob0aad5ca36a8125e205e85978aa32e8712ddc957a
1 /*
2 * linux/kernel/acct.c
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/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 */
63 * These constants control the amount of freespace that suspend and
64 * resume the process accounting system, and the time delay between
65 * each check.
66 * Turned into sysctl-controllable parameters. AV, 12/11/98
69 int acct_parm[3] = {4, 2, 30};
70 #define RESUME (acct_parm[0]) /* >foo% free space - resume */
71 #define SUSPEND (acct_parm[1]) /* <foo% free space - suspend */
72 #define ACCT_TIMEOUT (acct_parm[2]) /* foo second timeout between checks */
75 * External references and all of the globals.
77 static void do_acct_process(struct file *);
80 * This structure is used so that all the data protected by lock
81 * can be placed in the same cache line as the lock. This primes
82 * the cache line to have the data after getting the lock.
84 struct acct_glbs {
85 spinlock_t lock;
86 volatile int active;
87 volatile int needcheck;
88 struct file *file;
89 struct timer_list timer;
92 static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED};
95 * Called whenever the timer says to check the free space.
97 static void acct_timeout(unsigned long unused)
99 acct_globals.needcheck = 1;
103 * Check the amount of free space and suspend/resume accordingly.
105 static int check_free_space(struct file *file)
107 struct kstatfs sbuf;
108 int res;
109 int act;
110 sector_t resume;
111 sector_t suspend;
113 spin_lock(&acct_globals.lock);
114 res = acct_globals.active;
115 if (!file || !acct_globals.needcheck)
116 goto out;
117 spin_unlock(&acct_globals.lock);
119 /* May block */
120 if (vfs_statfs(file->f_dentry, &sbuf))
121 return res;
122 suspend = sbuf.f_blocks * SUSPEND;
123 resume = sbuf.f_blocks * RESUME;
125 sector_div(suspend, 100);
126 sector_div(resume, 100);
128 if (sbuf.f_bavail <= suspend)
129 act = -1;
130 else if (sbuf.f_bavail >= resume)
131 act = 1;
132 else
133 act = 0;
136 * If some joker switched acct_globals.file under us we'ld better be
137 * silent and _not_ touch anything.
139 spin_lock(&acct_globals.lock);
140 if (file != acct_globals.file) {
141 if (act)
142 res = act>0;
143 goto out;
146 if (acct_globals.active) {
147 if (act < 0) {
148 acct_globals.active = 0;
149 printk(KERN_INFO "Process accounting paused\n");
151 } else {
152 if (act > 0) {
153 acct_globals.active = 1;
154 printk(KERN_INFO "Process accounting resumed\n");
158 del_timer(&acct_globals.timer);
159 acct_globals.needcheck = 0;
160 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
161 add_timer(&acct_globals.timer);
162 res = acct_globals.active;
163 out:
164 spin_unlock(&acct_globals.lock);
165 return res;
169 * Close the old accounting file (if currently open) and then replace
170 * it with file (if non-NULL).
172 * NOTE: acct_globals.lock MUST be held on entry and exit.
174 static void acct_file_reopen(struct file *file)
176 struct file *old_acct = NULL;
178 if (acct_globals.file) {
179 old_acct = acct_globals.file;
180 del_timer(&acct_globals.timer);
181 acct_globals.active = 0;
182 acct_globals.needcheck = 0;
183 acct_globals.file = NULL;
185 if (file) {
186 acct_globals.file = file;
187 acct_globals.needcheck = 0;
188 acct_globals.active = 1;
189 /* It's been deleted if it was used before so this is safe */
190 init_timer(&acct_globals.timer);
191 acct_globals.timer.function = acct_timeout;
192 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
193 add_timer(&acct_globals.timer);
195 if (old_acct) {
196 mnt_unpin(old_acct->f_vfsmnt);
197 spin_unlock(&acct_globals.lock);
198 do_acct_process(old_acct);
199 filp_close(old_acct, NULL);
200 spin_lock(&acct_globals.lock);
204 static int acct_on(char *name)
206 struct file *file;
207 int error;
209 /* Difference from BSD - they don't do O_APPEND */
210 file = filp_open(name, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
211 if (IS_ERR(file))
212 return PTR_ERR(file);
214 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
215 filp_close(file, NULL);
216 return -EACCES;
219 if (!file->f_op->write) {
220 filp_close(file, NULL);
221 return -EIO;
224 error = security_acct(file);
225 if (error) {
226 filp_close(file, NULL);
227 return error;
230 spin_lock(&acct_globals.lock);
231 mnt_pin(file->f_vfsmnt);
232 acct_file_reopen(file);
233 spin_unlock(&acct_globals.lock);
235 mntput(file->f_vfsmnt); /* it's pinned, now give up active reference */
237 return 0;
241 * sys_acct - enable/disable process accounting
242 * @name: file name for accounting records or NULL to shutdown accounting
244 * Returns 0 for success or negative errno values for failure.
246 * sys_acct() is the only system call needed to implement process
247 * accounting. It takes the name of the file where accounting records
248 * should be written. If the filename is NULL, accounting will be
249 * shutdown.
251 asmlinkage long sys_acct(const char __user *name)
253 int error;
255 if (!capable(CAP_SYS_PACCT))
256 return -EPERM;
258 if (name) {
259 char *tmp = getname(name);
260 if (IS_ERR(tmp))
261 return (PTR_ERR(tmp));
262 error = acct_on(tmp);
263 putname(tmp);
264 } else {
265 error = security_acct(NULL);
266 if (!error) {
267 spin_lock(&acct_globals.lock);
268 acct_file_reopen(NULL);
269 spin_unlock(&acct_globals.lock);
272 return error;
276 * acct_auto_close - turn off a filesystem's accounting if it is on
277 * @m: vfsmount being shut down
279 * If the accounting is turned on for a file in the subtree pointed to
280 * to by m, turn accounting off. Done when m is about to die.
282 void acct_auto_close_mnt(struct vfsmount *m)
284 spin_lock(&acct_globals.lock);
285 if (acct_globals.file && acct_globals.file->f_vfsmnt == m)
286 acct_file_reopen(NULL);
287 spin_unlock(&acct_globals.lock);
291 * acct_auto_close - turn off a filesystem's accounting if it is on
292 * @sb: super block for the filesystem
294 * If the accounting is turned on for a file in the filesystem pointed
295 * to by sb, turn accounting off.
297 void acct_auto_close(struct super_block *sb)
299 spin_lock(&acct_globals.lock);
300 if (acct_globals.file &&
301 acct_globals.file->f_vfsmnt->mnt_sb == sb) {
302 acct_file_reopen(NULL);
304 spin_unlock(&acct_globals.lock);
308 * encode an unsigned long into a comp_t
310 * This routine has been adopted from the encode_comp_t() function in
311 * the kern_acct.c file of the FreeBSD operating system. The encoding
312 * is a 13-bit fraction with a 3-bit (base 8) exponent.
315 #define MANTSIZE 13 /* 13 bit mantissa. */
316 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
317 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
319 static comp_t encode_comp_t(unsigned long value)
321 int exp, rnd;
323 exp = rnd = 0;
324 while (value > MAXFRACT) {
325 rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */
326 value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
327 exp++;
331 * If we need to round up, do it (and handle overflow correctly).
333 if (rnd && (++value > MAXFRACT)) {
334 value >>= EXPSIZE;
335 exp++;
339 * Clean it up and polish it off.
341 exp <<= MANTSIZE; /* Shift the exponent into place */
342 exp += value; /* and add on the mantissa. */
343 return exp;
346 #if ACCT_VERSION==1 || ACCT_VERSION==2
348 * encode an u64 into a comp2_t (24 bits)
350 * Format: 5 bit base 2 exponent, 20 bits mantissa.
351 * The leading bit of the mantissa is not stored, but implied for
352 * non-zero exponents.
353 * Largest encodable value is 50 bits.
356 #define MANTSIZE2 20 /* 20 bit mantissa. */
357 #define EXPSIZE2 5 /* 5 bit base 2 exponent. */
358 #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
359 #define MAXEXP2 ((1 <<EXPSIZE2) - 1) /* Maximum exponent. */
361 static comp2_t encode_comp2_t(u64 value)
363 int exp, rnd;
365 exp = (value > (MAXFRACT2>>1));
366 rnd = 0;
367 while (value > MAXFRACT2) {
368 rnd = value & 1;
369 value >>= 1;
370 exp++;
374 * If we need to round up, do it (and handle overflow correctly).
376 if (rnd && (++value > MAXFRACT2)) {
377 value >>= 1;
378 exp++;
381 if (exp > MAXEXP2) {
382 /* Overflow. Return largest representable number instead. */
383 return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
384 } else {
385 return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
388 #endif
390 #if ACCT_VERSION==3
392 * encode an u64 into a 32 bit IEEE float
394 static u32 encode_float(u64 value)
396 unsigned exp = 190;
397 unsigned u;
399 if (value==0) return 0;
400 while ((s64)value > 0){
401 value <<= 1;
402 exp--;
404 u = (u32)(value >> 40) & 0x7fffffu;
405 return u | (exp << 23);
407 #endif
410 * Write an accounting entry for an exiting process
412 * The acct_process() call is the workhorse of the process
413 * accounting system. The struct acct is built here and then written
414 * into the accounting file. This function should only be called from
415 * do_exit().
419 * do_acct_process does all actual work. Caller holds the reference to file.
421 static void do_acct_process(struct file *file)
423 struct pacct_struct *pacct = &current->signal->pacct;
424 acct_t ac;
425 mm_segment_t fs;
426 unsigned long flim;
427 u64 elapsed;
428 u64 run_time;
429 struct timespec uptime;
432 * First check to see if there is enough free_space to continue
433 * the process accounting system.
435 if (!check_free_space(file))
436 return;
439 * Fill the accounting struct with the needed info as recorded
440 * by the different kernel functions.
442 memset((caddr_t)&ac, 0, sizeof(acct_t));
444 ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER;
445 strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
447 /* calculate run_time in nsec*/
448 do_posix_clock_monotonic_gettime(&uptime);
449 run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec;
450 run_time -= (u64)current->group_leader->start_time.tv_sec * NSEC_PER_SEC
451 + current->group_leader->start_time.tv_nsec;
452 /* convert nsec -> AHZ */
453 elapsed = nsec_to_AHZ(run_time);
454 #if ACCT_VERSION==3
455 ac.ac_etime = encode_float(elapsed);
456 #else
457 ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
458 (unsigned long) elapsed : (unsigned long) -1l);
459 #endif
460 #if ACCT_VERSION==1 || ACCT_VERSION==2
462 /* new enlarged etime field */
463 comp2_t etime = encode_comp2_t(elapsed);
464 ac.ac_etime_hi = etime >> 16;
465 ac.ac_etime_lo = (u16) etime;
467 #endif
468 do_div(elapsed, AHZ);
469 ac.ac_btime = xtime.tv_sec - elapsed;
470 /* we really need to bite the bullet and change layout */
471 ac.ac_uid = current->uid;
472 ac.ac_gid = current->gid;
473 #if ACCT_VERSION==2
474 ac.ac_ahz = AHZ;
475 #endif
476 #if ACCT_VERSION==1 || ACCT_VERSION==2
477 /* backward-compatible 16 bit fields */
478 ac.ac_uid16 = current->uid;
479 ac.ac_gid16 = current->gid;
480 #endif
481 #if ACCT_VERSION==3
482 ac.ac_pid = current->tgid;
483 ac.ac_ppid = current->parent->tgid;
484 #endif
486 mutex_lock(&tty_mutex);
487 /* FIXME: Whoever is responsible for current->signal locking needs
488 to use the same locking all over the kernel and document it */
489 read_lock(&tasklist_lock);
490 ac.ac_tty = current->signal->tty ?
491 old_encode_dev(tty_devnum(current->signal->tty)) : 0;
492 read_unlock(&tasklist_lock);
493 mutex_unlock(&tty_mutex);
495 spin_lock_irq(&current->sighand->siglock);
496 ac.ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime)));
497 ac.ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime)));
498 ac.ac_flag = pacct->ac_flag;
499 ac.ac_mem = encode_comp_t(pacct->ac_mem);
500 ac.ac_minflt = encode_comp_t(pacct->ac_minflt);
501 ac.ac_majflt = encode_comp_t(pacct->ac_majflt);
502 ac.ac_exitcode = pacct->ac_exitcode;
503 spin_unlock_irq(&current->sighand->siglock);
504 ac.ac_io = encode_comp_t(0 /* current->io_usage */); /* %% */
505 ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
506 ac.ac_swaps = encode_comp_t(0);
509 * Kernel segment override to datasegment and write it
510 * to the accounting file.
512 fs = get_fs();
513 set_fs(KERNEL_DS);
515 * Accounting records are not subject to resource limits.
517 flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
518 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
519 file->f_op->write(file, (char *)&ac,
520 sizeof(acct_t), &file->f_pos);
521 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
522 set_fs(fs);
526 * acct_init_pacct - initialize a new pacct_struct
527 * @pacct: per-process accounting info struct to initialize
529 void acct_init_pacct(struct pacct_struct *pacct)
531 memset(pacct, 0, sizeof(struct pacct_struct));
532 pacct->ac_utime = pacct->ac_stime = cputime_zero;
536 * acct_collect - collect accounting information into pacct_struct
537 * @exitcode: task exit code
538 * @group_dead: not 0, if this thread is the last one in the process.
540 void acct_collect(long exitcode, int group_dead)
542 struct pacct_struct *pacct = &current->signal->pacct;
543 unsigned long vsize = 0;
545 if (group_dead && current->mm) {
546 struct vm_area_struct *vma;
547 down_read(&current->mm->mmap_sem);
548 vma = current->mm->mmap;
549 while (vma) {
550 vsize += vma->vm_end - vma->vm_start;
551 vma = vma->vm_next;
553 up_read(&current->mm->mmap_sem);
556 spin_lock_irq(&current->sighand->siglock);
557 if (group_dead)
558 pacct->ac_mem = vsize / 1024;
559 if (thread_group_leader(current)) {
560 pacct->ac_exitcode = exitcode;
561 if (current->flags & PF_FORKNOEXEC)
562 pacct->ac_flag |= AFORK;
564 if (current->flags & PF_SUPERPRIV)
565 pacct->ac_flag |= ASU;
566 if (current->flags & PF_DUMPCORE)
567 pacct->ac_flag |= ACORE;
568 if (current->flags & PF_SIGNALED)
569 pacct->ac_flag |= AXSIG;
570 pacct->ac_utime = cputime_add(pacct->ac_utime, current->utime);
571 pacct->ac_stime = cputime_add(pacct->ac_stime, current->stime);
572 pacct->ac_minflt += current->min_flt;
573 pacct->ac_majflt += current->maj_flt;
574 spin_unlock_irq(&current->sighand->siglock);
578 * acct_process - now just a wrapper around do_acct_process
579 * @exitcode: task exit code
581 * handles process accounting for an exiting task
583 void acct_process(void)
585 struct file *file = NULL;
588 * accelerate the common fastpath:
590 if (!acct_globals.file)
591 return;
593 spin_lock(&acct_globals.lock);
594 file = acct_globals.file;
595 if (unlikely(!file)) {
596 spin_unlock(&acct_globals.lock);
597 return;
599 get_file(file);
600 spin_unlock(&acct_globals.lock);
602 do_acct_process(file);
603 fput(file);