Linux Audio

Check our new training course

Loading...
v3.15
 
  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
 70int 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 */
 78static 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 */
 86struct bsd_acct_struct {
 
 
 
 
 87	int			active;
 88	unsigned long		needcheck;
 89	struct file		*file;
 90	struct pid_namespace	*ns;
 91	struct list_head	list;
 
 92};
 93
 94static DEFINE_SPINLOCK(acct_lock);
 95static LIST_HEAD(acct_list);
 96
 97/*
 98 * Check the amount of free space and suspend/resume accordingly.
 99 */
100static int check_free_space(struct bsd_acct_struct *acct, struct file *file)
101{
102	struct kstatfs sbuf;
103	int res;
104	int act;
105	u64 resume;
106	u64 suspend;
107
108	spin_lock(&acct_lock);
109	res = acct->active;
110	if (!file || time_is_before_jiffies(acct->needcheck))
111		goto out;
112	spin_unlock(&acct_lock);
113
114	/* May block */
115	if (vfs_statfs(&file->f_path, &sbuf))
116		return res;
117	suspend = sbuf.f_blocks * SUSPEND;
118	resume = sbuf.f_blocks * RESUME;
119
120	do_div(suspend, 100);
121	do_div(resume, 100);
122
123	if (sbuf.f_bavail <= suspend)
124		act = -1;
125	else if (sbuf.f_bavail >= resume)
126		act = 1;
127	else
128		act = 0;
129
130	/*
131	 * If some joker switched acct->file under us we'ld better be
132	 * silent and _not_ touch anything.
133	 */
134	spin_lock(&acct_lock);
135	if (file != acct->file) {
136		if (act)
137			res = act>0;
138		goto out;
139	}
140
141	if (acct->active) {
142		if (act < 0) {
 
 
143			acct->active = 0;
144			printk(KERN_INFO "Process accounting paused\n");
145		}
146	} else {
147		if (act > 0) {
 
 
148			acct->active = 1;
149			printk(KERN_INFO "Process accounting resumed\n");
150		}
151	}
152
153	acct->needcheck = jiffies + ACCT_TIMEOUT*HZ;
154	res = acct->active;
155out:
156	spin_unlock(&acct_lock);
157	return res;
158}
159
160/*
161 * Close the old accounting file (if currently open) and then replace
162 * it with file (if non-NULL).
163 *
164 * NOTE: acct_lock MUST be held on entry and exit.
165 */
166static void acct_file_reopen(struct bsd_acct_struct *acct, struct file *file,
167		struct pid_namespace *ns)
 
 
 
 
168{
169	struct file *old_acct = NULL;
170	struct pid_namespace *old_ns = NULL;
171
172	if (acct->file) {
173		old_acct = acct->file;
174		old_ns = acct->ns;
175		acct->active = 0;
176		acct->file = NULL;
177		acct->ns = NULL;
178		list_del(&acct->list);
179	}
180	if (file) {
181		acct->file = file;
182		acct->ns = ns;
183		acct->needcheck = jiffies + ACCT_TIMEOUT*HZ;
184		acct->active = 1;
185		list_add(&acct->list, &acct_list);
186	}
187	if (old_acct) {
188		mnt_unpin(old_acct->f_path.mnt);
189		spin_unlock(&acct_lock);
190		do_acct_process(acct, old_ns, old_acct);
191		filp_close(old_acct, NULL);
192		spin_lock(&acct_lock);
193	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
194}
195
196static int acct_on(struct filename *pathname)
197{
198	struct file *file;
199	struct vfsmount *mnt;
200	struct pid_namespace *ns;
201	struct bsd_acct_struct *acct = NULL;
 
 
 
 
 
 
202
203	/* Difference from BSD - they don't do O_APPEND */
204	file = file_open_name(pathname, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
205	if (IS_ERR(file))
 
206		return PTR_ERR(file);
 
207
208	if (!S_ISREG(file_inode(file)->i_mode)) {
 
209		filp_close(file, NULL);
210		return -EACCES;
211	}
212
213	if (!file->f_op->write) {
 
214		filp_close(file, NULL);
215		return -EIO;
216	}
217
218	ns = task_active_pid_ns(current);
219	if (ns->bacct == NULL) {
220		acct = kzalloc(sizeof(struct bsd_acct_struct), GFP_KERNEL);
221		if (acct == NULL) {
222			filp_close(file, NULL);
223			return -ENOMEM;
224		}
225	}
226
227	spin_lock(&acct_lock);
228	if (ns->bacct == NULL) {
229		ns->bacct = acct;
230		acct = NULL;
 
231	}
232
233	mnt = file->f_path.mnt;
234	mnt_pin(mnt);
235	acct_file_reopen(ns->bacct, file, ns);
236	spin_unlock(&acct_lock);
237
238	mntput(mnt); /* it's pinned, now give up active reference */
239	kfree(acct);
 
 
 
 
 
 
 
 
240
 
 
 
 
 
 
241	return 0;
242}
243
 
 
244/**
245 * sys_acct - enable/disable process accounting
246 * @name: file name for accounting records or NULL to shutdown accounting
247 *
248 * Returns 0 for success or negative errno values for failure.
249 *
250 * sys_acct() is the only system call needed to implement process
251 * accounting. It takes the name of the file where accounting records
252 * should be written. If the filename is NULL, accounting will be
253 * shutdown.
254 */
255SYSCALL_DEFINE1(acct, const char __user *, name)
256{
257	int error = 0;
258
259	if (!capable(CAP_SYS_PACCT))
260		return -EPERM;
261
262	if (name) {
263		struct filename *tmp = getname(name);
 
264		if (IS_ERR(tmp))
265			return (PTR_ERR(tmp));
 
266		error = acct_on(tmp);
 
267		putname(tmp);
268	} else {
269		struct bsd_acct_struct *acct;
270
271		acct = task_active_pid_ns(current)->bacct;
272		if (acct == NULL)
273			return 0;
274
275		spin_lock(&acct_lock);
276		acct_file_reopen(acct, NULL, 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 */
290void acct_auto_close_mnt(struct vfsmount *m)
291{
292	struct bsd_acct_struct *acct;
293
294	spin_lock(&acct_lock);
295restart:
296	list_for_each_entry(acct, &acct_list, list)
297		if (acct->file && acct->file->f_path.mnt == m) {
298			acct_file_reopen(acct, NULL, NULL);
299			goto restart;
300		}
301	spin_unlock(&acct_lock);
302}
303
304/**
305 * acct_auto_close - turn off a filesystem's accounting if it is on
306 * @sb: super block for the filesystem
307 *
308 * If the accounting is turned on for a file in the filesystem pointed
309 * to by sb, turn accounting off.
310 */
311void acct_auto_close(struct super_block *sb)
312{
313	struct bsd_acct_struct *acct;
314
315	spin_lock(&acct_lock);
316restart:
317	list_for_each_entry(acct, &acct_list, list)
318		if (acct->file && acct->file->f_path.dentry->d_sb == sb) {
319			acct_file_reopen(acct, NULL, NULL);
320			goto restart;
321		}
322	spin_unlock(&acct_lock);
323}
324
325void acct_exit_ns(struct pid_namespace *ns)
326{
327	struct bsd_acct_struct *acct = ns->bacct;
328
329	if (acct == NULL)
330		return;
331
332	spin_lock(&acct_lock);
333	if (acct->file != NULL)
334		acct_file_reopen(acct, NULL, NULL);
335	spin_unlock(&acct_lock);
336
337	kfree(acct);
338}
339
340/*
341 *  encode an unsigned long into a comp_t
342 *
343 *  This routine has been adopted from the encode_comp_t() function in
344 *  the kern_acct.c file of the FreeBSD operating system. The encoding
345 *  is a 13-bit fraction with a 3-bit (base 8) exponent.
346 */
347
348#define	MANTSIZE	13			/* 13 bit mantissa. */
349#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
350#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
351
352static comp_t encode_comp_t(unsigned long value)
353{
354	int exp, rnd;
355
356	exp = rnd = 0;
357	while (value > MAXFRACT) {
358		rnd = value & (1 << (EXPSIZE - 1));	/* Round up? */
359		value >>= EXPSIZE;	/* Base 8 exponent == 3 bit shift. */
360		exp++;
361	}
362
363	/*
364	 * If we need to round up, do it (and handle overflow correctly).
365	 */
366	if (rnd && (++value > MAXFRACT)) {
367		value >>= EXPSIZE;
368		exp++;
369	}
370
371	/*
372	 * Clean it up and polish it off.
373	 */
374	exp <<= MANTSIZE;		/* Shift the exponent into place */
375	exp += value;			/* and add on the mantissa. */
376	return exp;
377}
378
379#if ACCT_VERSION==1 || ACCT_VERSION==2
380/*
381 * encode an u64 into a comp2_t (24 bits)
382 *
383 * Format: 5 bit base 2 exponent, 20 bits mantissa.
384 * The leading bit of the mantissa is not stored, but implied for
385 * non-zero exponents.
386 * Largest encodable value is 50 bits.
387 */
388
389#define MANTSIZE2       20                      /* 20 bit mantissa. */
390#define EXPSIZE2        5                       /* 5 bit base 2 exponent. */
391#define MAXFRACT2       ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
392#define MAXEXP2         ((1 <<EXPSIZE2) - 1)    /* Maximum exponent. */
393
394static comp2_t encode_comp2_t(u64 value)
395{
396	int exp, rnd;
397
398	exp = (value > (MAXFRACT2>>1));
399	rnd = 0;
400	while (value > MAXFRACT2) {
401		rnd = value & 1;
402		value >>= 1;
403		exp++;
404	}
405
406	/*
407	 * If we need to round up, do it (and handle overflow correctly).
408	 */
409	if (rnd && (++value > MAXFRACT2)) {
410		value >>= 1;
411		exp++;
412	}
413
414	if (exp > MAXEXP2) {
415		/* Overflow. Return largest representable number instead. */
416		return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
417	} else {
418		return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
419	}
420}
421#endif
422
423#if ACCT_VERSION==3
424/*
425 * encode an u64 into a 32 bit IEEE float
426 */
427static u32 encode_float(u64 value)
428{
429	unsigned exp = 190;
430	unsigned u;
431
432	if (value==0) return 0;
433	while ((s64)value > 0){
 
434		value <<= 1;
435		exp--;
436	}
437	u = (u32)(value >> 40) & 0x7fffffu;
438	return u | (exp << 23);
439}
440#endif
441
442/*
443 *  Write an accounting entry for an exiting process
444 *
445 *  The acct_process() call is the workhorse of the process
446 *  accounting system. The struct acct is built here and then written
447 *  into the accounting file. This function should only be called from
448 *  do_exit() or when switching to a different output file.
449 */
450
451/*
452 *  do_acct_process does all actual work. Caller holds the reference to file.
453 */
454static void do_acct_process(struct bsd_acct_struct *acct,
455		struct pid_namespace *ns, struct file *file)
456{
457	struct pacct_struct *pacct = &current->signal->pacct;
458	acct_t ac;
459	mm_segment_t fs;
460	unsigned long flim;
461	u64 elapsed;
462	u64 run_time;
463	struct timespec uptime;
464	struct tty_struct *tty;
465	const struct cred *orig_cred;
466
467	/* Perform file operations on behalf of whoever enabled accounting */
468	orig_cred = override_creds(file->f_cred);
469
470	/*
471	 * First check to see if there is enough free_space to continue
472	 * the process accounting system.
473	 */
474	if (!check_free_space(acct, file))
475		goto out;
476
477	/*
478	 * Fill the accounting struct with the needed info as recorded
479	 * by the different kernel functions.
480	 */
481	memset(&ac, 0, sizeof(acct_t));
482
483	ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER;
484	strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
485
486	/* calculate run_time in nsec*/
487	do_posix_clock_monotonic_gettime(&uptime);
488	run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec;
489	run_time -= (u64)current->group_leader->start_time.tv_sec * NSEC_PER_SEC
490		       + current->group_leader->start_time.tv_nsec;
491	/* convert nsec -> AHZ */
492	elapsed = nsec_to_AHZ(run_time);
493#if ACCT_VERSION==3
494	ac.ac_etime = encode_float(elapsed);
495#else
496	ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
497	                       (unsigned long) elapsed : (unsigned long) -1l);
498#endif
499#if ACCT_VERSION==1 || ACCT_VERSION==2
500	{
501		/* new enlarged etime field */
502		comp2_t etime = encode_comp2_t(elapsed);
503		ac.ac_etime_hi = etime >> 16;
504		ac.ac_etime_lo = (u16) etime;
 
505	}
506#endif
507	do_div(elapsed, AHZ);
508	ac.ac_btime = get_seconds() - elapsed;
509	/* we really need to bite the bullet and change layout */
510	ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid);
511	ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid);
512#if ACCT_VERSION==2
513	ac.ac_ahz = AHZ;
514#endif
515#if ACCT_VERSION==1 || ACCT_VERSION==2
516	/* backward-compatible 16 bit fields */
517	ac.ac_uid16 = ac.ac_uid;
518	ac.ac_gid16 = ac.ac_gid;
519#endif
520#if ACCT_VERSION==3
521	ac.ac_pid = task_tgid_nr_ns(current, ns);
522	rcu_read_lock();
523	ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent), ns);
524	rcu_read_unlock();
525#endif
526
527	spin_lock_irq(&current->sighand->siglock);
528	tty = current->signal->tty;	/* Safe as we hold the siglock */
529	ac.ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
530	ac.ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime)));
531	ac.ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime)));
532	ac.ac_flag = pacct->ac_flag;
533	ac.ac_mem = encode_comp_t(pacct->ac_mem);
534	ac.ac_minflt = encode_comp_t(pacct->ac_minflt);
535	ac.ac_majflt = encode_comp_t(pacct->ac_majflt);
536	ac.ac_exitcode = pacct->ac_exitcode;
537	spin_unlock_irq(&current->sighand->siglock);
538	ac.ac_io = encode_comp_t(0 /* current->io_usage */);	/* %% */
539	ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
540	ac.ac_swaps = encode_comp_t(0);
 
 
 
 
 
 
 
541
542	/*
543	 * Get freeze protection. If the fs is frozen, just skip the write
544	 * as we could deadlock the system otherwise.
545	 */
546	if (!file_start_write_trylock(file))
547		goto out;
 
 
 
548	/*
549	 * Kernel segment override to datasegment and write it
550	 * to the accounting file.
551	 */
552	fs = get_fs();
553	set_fs(KERNEL_DS);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
554	/*
555	 * Accounting records are not subject to resource limits.
 
556	 */
557	flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
558	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
559	file->f_op->write(file, (char *)&ac,
560			       sizeof(acct_t), &file->f_pos);
561	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
562	set_fs(fs);
563	file_end_write(file);
564out:
 
565	revert_creds(orig_cred);
566}
567
568/**
569 * acct_collect - collect accounting information into pacct_struct
570 * @exitcode: task exit code
571 * @group_dead: not 0, if this thread is the last one in the process.
572 */
573void acct_collect(long exitcode, int group_dead)
574{
575	struct pacct_struct *pacct = &current->signal->pacct;
576	cputime_t utime, stime;
577	unsigned long vsize = 0;
578
579	if (group_dead && current->mm) {
580		struct vm_area_struct *vma;
 
581		down_read(&current->mm->mmap_sem);
582		vma = current->mm->mmap;
583		while (vma) {
584			vsize += vma->vm_end - vma->vm_start;
585			vma = vma->vm_next;
586		}
587		up_read(&current->mm->mmap_sem);
588	}
589
590	spin_lock_irq(&current->sighand->siglock);
591	if (group_dead)
592		pacct->ac_mem = vsize / 1024;
593	if (thread_group_leader(current)) {
594		pacct->ac_exitcode = exitcode;
595		if (current->flags & PF_FORKNOEXEC)
596			pacct->ac_flag |= AFORK;
597	}
598	if (current->flags & PF_SUPERPRIV)
599		pacct->ac_flag |= ASU;
600	if (current->flags & PF_DUMPCORE)
601		pacct->ac_flag |= ACORE;
602	if (current->flags & PF_SIGNALED)
603		pacct->ac_flag |= AXSIG;
 
604	task_cputime(current, &utime, &stime);
605	pacct->ac_utime += utime;
606	pacct->ac_stime += stime;
607	pacct->ac_minflt += current->min_flt;
608	pacct->ac_majflt += current->maj_flt;
609	spin_unlock_irq(&current->sighand->siglock);
610}
611
612static void acct_process_in_ns(struct pid_namespace *ns)
613{
614	struct file *file = NULL;
615	struct bsd_acct_struct *acct;
616
617	acct = ns->bacct;
618	/*
619	 * accelerate the common fastpath:
620	 */
621	if (!acct || !acct->file)
622		return;
623
624	spin_lock(&acct_lock);
625	file = acct->file;
626	if (unlikely(!file)) {
627		spin_unlock(&acct_lock);
628		return;
629	}
630	get_file(file);
631	spin_unlock(&acct_lock);
632
633	do_acct_process(acct, ns, file);
634	fput(file);
635}
636
637/**
638 * acct_process - now just a wrapper around acct_process_in_ns,
639 * which in turn is a wrapper around do_acct_process.
640 *
641 * handles process accounting for an exiting task
642 */
643void acct_process(void)
644{
645	struct pid_namespace *ns;
646
647	/*
648	 * This loop is safe lockless, since current is still
649	 * alive and holds its namespace, which in turn holds
650	 * its parent.
651	 */
652	for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent)
653		acct_process_in_ns(ns);
 
 
 
 
654}
v5.4
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 *  linux/kernel/acct.c
  4 *
  5 *  BSD Process Accounting for Linux
  6 *
  7 *  Author: Marco van Wieringen <mvw@planets.elm.net>
  8 *
  9 *  Some code based on ideas and code from:
 10 *  Thomas K. Dyas <tdyas@eden.rutgers.edu>
 11 *
 12 *  This file implements BSD-style process accounting. Whenever any
 13 *  process exits, an accounting record of type "struct acct" is
 14 *  written to the file specified with the acct() system call. It is
 15 *  up to user-level programs to do useful things with the accounting
 16 *  log. The kernel just provides the raw accounting information.
 17 *
 18 * (C) Copyright 1995 - 1997 Marco van Wieringen - ELM Consultancy B.V.
 19 *
 20 *  Plugged two leaks. 1) It didn't return acct_file into the free_filps if
 21 *  the file happened to be read-only. 2) If the accounting was suspended
 22 *  due to the lack of space it happily allowed to reopen it and completely
 23 *  lost the old acct_file. 3/10/98, Al Viro.
 24 *
 25 *  Now we silently close acct_file on attempt to reopen. Cleaned sys_acct().
 26 *  XTerms and EMACS are manifestations of pure evil. 21/10/98, AV.
 27 *
 28 *  Fixed a nasty interaction with with sys_umount(). If the accointing
 29 *  was suspeneded we failed to stop it on umount(). Messy.
 30 *  Another one: remount to readonly didn't stop accounting.
 31 *	Question: what should we do if we have CAP_SYS_ADMIN but not
 32 *  CAP_SYS_PACCT? Current code does the following: umount returns -EBUSY
 33 *  unless we are messing with the root. In that case we are getting a
 34 *  real mess with do_remount_sb(). 9/11/98, AV.
 35 *
 36 *  Fixed a bunch of races (and pair of leaks). Probably not the best way,
 37 *  but this one obviously doesn't introduce deadlocks. Later. BTW, found
 38 *  one race (and leak) in BSD implementation.
 39 *  OK, that's better. ANOTHER race and leak in BSD variant. There always
 40 *  is one more bug... 10/11/98, AV.
 41 *
 42 *	Oh, fsck... Oopsable SMP race in do_process_acct() - we must hold
 43 * ->mmap_sem to walk the vma list of current->mm. Nasty, since it leaks
 44 * a struct file opened for write. Fixed. 2/6/2000, AV.
 45 */
 46
 47#include <linux/mm.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 <linux/uaccess.h>
 60#include <linux/sched/cputime.h>
 61
 62#include <asm/div64.h>
 63#include <linux/blkdev.h> /* sector_div */
 64#include <linux/pid_namespace.h>
 65#include <linux/fs_pin.h>
 66
 67/*
 68 * These constants control the amount of freespace that suspend and
 69 * resume the process accounting system, and the time delay between
 70 * each check.
 71 * Turned into sysctl-controllable parameters. AV, 12/11/98
 72 */
 73
 74int acct_parm[3] = {4, 2, 30};
 75#define RESUME		(acct_parm[0])	/* >foo% free space - resume */
 76#define SUSPEND		(acct_parm[1])	/* <foo% free space - suspend */
 77#define ACCT_TIMEOUT	(acct_parm[2])	/* foo second timeout between checks */
 78
 79/*
 80 * External references and all of the globals.
 81 */
 
 
 82
 
 
 
 
 
 83struct bsd_acct_struct {
 84	struct fs_pin		pin;
 85	atomic_long_t		count;
 86	struct rcu_head		rcu;
 87	struct mutex		lock;
 88	int			active;
 89	unsigned long		needcheck;
 90	struct file		*file;
 91	struct pid_namespace	*ns;
 92	struct work_struct	work;
 93	struct completion	done;
 94};
 95
 96static void do_acct_process(struct bsd_acct_struct *acct);
 
 97
 98/*
 99 * Check the amount of free space and suspend/resume accordingly.
100 */
101static int check_free_space(struct bsd_acct_struct *acct)
102{
103	struct kstatfs sbuf;
104
105	if (time_is_after_jiffies(acct->needcheck))
 
 
 
 
 
 
106		goto out;
 
107
108	/* May block */
109	if (vfs_statfs(&acct->file->f_path, &sbuf))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
110		goto out;
 
111
112	if (acct->active) {
113		u64 suspend = sbuf.f_blocks * SUSPEND;
114		do_div(suspend, 100);
115		if (sbuf.f_bavail <= suspend) {
116			acct->active = 0;
117			pr_info("Process accounting paused\n");
118		}
119	} else {
120		u64 resume = sbuf.f_blocks * RESUME;
121		do_div(resume, 100);
122		if (sbuf.f_bavail >= resume) {
123			acct->active = 1;
124			pr_info("Process accounting resumed\n");
125		}
126	}
127
128	acct->needcheck = jiffies + ACCT_TIMEOUT*HZ;
 
129out:
130	return acct->active;
 
131}
132
133static void acct_put(struct bsd_acct_struct *p)
134{
135	if (atomic_long_dec_and_test(&p->count))
136		kfree_rcu(p, rcu);
137}
138
139static inline struct bsd_acct_struct *to_acct(struct fs_pin *p)
140{
141	return p ? container_of(p, struct bsd_acct_struct, pin) : NULL;
142}
143
144static struct bsd_acct_struct *acct_get(struct pid_namespace *ns)
145{
146	struct bsd_acct_struct *res;
147again:
148	smp_rmb();
149	rcu_read_lock();
150	res = to_acct(READ_ONCE(ns->bacct));
151	if (!res) {
152		rcu_read_unlock();
153		return NULL;
154	}
155	if (!atomic_long_inc_not_zero(&res->count)) {
156		rcu_read_unlock();
157		cpu_relax();
158		goto again;
159	}
160	rcu_read_unlock();
161	mutex_lock(&res->lock);
162	if (res != to_acct(READ_ONCE(ns->bacct))) {
163		mutex_unlock(&res->lock);
164		acct_put(res);
165		goto again;
 
 
 
 
166	}
167	return res;
168}
169
170static void acct_pin_kill(struct fs_pin *pin)
171{
172	struct bsd_acct_struct *acct = to_acct(pin);
173	mutex_lock(&acct->lock);
174	do_acct_process(acct);
175	schedule_work(&acct->work);
176	wait_for_completion(&acct->done);
177	cmpxchg(&acct->ns->bacct, pin, NULL);
178	mutex_unlock(&acct->lock);
179	pin_remove(pin);
180	acct_put(acct);
181}
182
183static void close_work(struct work_struct *work)
184{
185	struct bsd_acct_struct *acct = container_of(work, struct bsd_acct_struct, work);
186	struct file *file = acct->file;
187	if (file->f_op->flush)
188		file->f_op->flush(file, NULL);
189	__fput_sync(file);
190	complete(&acct->done);
191}
192
193static int acct_on(struct filename *pathname)
194{
195	struct file *file;
196	struct vfsmount *mnt, *internal;
197	struct pid_namespace *ns = task_active_pid_ns(current);
198	struct bsd_acct_struct *acct;
199	struct fs_pin *old;
200	int err;
201
202	acct = kzalloc(sizeof(struct bsd_acct_struct), GFP_KERNEL);
203	if (!acct)
204		return -ENOMEM;
205
206	/* Difference from BSD - they don't do O_APPEND */
207	file = file_open_name(pathname, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
208	if (IS_ERR(file)) {
209		kfree(acct);
210		return PTR_ERR(file);
211	}
212
213	if (!S_ISREG(file_inode(file)->i_mode)) {
214		kfree(acct);
215		filp_close(file, NULL);
216		return -EACCES;
217	}
218
219	if (!(file->f_mode & FMODE_CAN_WRITE)) {
220		kfree(acct);
221		filp_close(file, NULL);
222		return -EIO;
223	}
224	internal = mnt_clone_internal(&file->f_path);
225	if (IS_ERR(internal)) {
226		kfree(acct);
227		filp_close(file, NULL);
228		return PTR_ERR(internal);
 
 
 
229	}
230	err = __mnt_want_write(internal);
231	if (err) {
232		mntput(internal);
233		kfree(acct);
234		filp_close(file, NULL);
235		return err;
236	}
 
237	mnt = file->f_path.mnt;
238	file->f_path.mnt = internal;
 
 
239
240	atomic_long_set(&acct->count, 1);
241	init_fs_pin(&acct->pin, acct_pin_kill);
242	acct->file = file;
243	acct->needcheck = jiffies;
244	acct->ns = ns;
245	mutex_init(&acct->lock);
246	INIT_WORK(&acct->work, close_work);
247	init_completion(&acct->done);
248	mutex_lock_nested(&acct->lock, 1);	/* nobody has seen it yet */
249	pin_insert(&acct->pin, mnt);
250
251	rcu_read_lock();
252	old = xchg(&ns->bacct, &acct->pin);
253	mutex_unlock(&acct->lock);
254	pin_kill(old);
255	__mnt_drop_write(mnt);
256	mntput(mnt);
257	return 0;
258}
259
260static DEFINE_MUTEX(acct_on_mutex);
261
262/**
263 * sys_acct - enable/disable process accounting
264 * @name: file name for accounting records or NULL to shutdown accounting
265 *
266 * Returns 0 for success or negative errno values for failure.
267 *
268 * sys_acct() is the only system call needed to implement process
269 * accounting. It takes the name of the file where accounting records
270 * should be written. If the filename is NULL, accounting will be
271 * shutdown.
272 */
273SYSCALL_DEFINE1(acct, const char __user *, name)
274{
275	int error = 0;
276
277	if (!capable(CAP_SYS_PACCT))
278		return -EPERM;
279
280	if (name) {
281		struct filename *tmp = getname(name);
282
283		if (IS_ERR(tmp))
284			return PTR_ERR(tmp);
285		mutex_lock(&acct_on_mutex);
286		error = acct_on(tmp);
287		mutex_unlock(&acct_on_mutex);
288		putname(tmp);
289	} else {
290		rcu_read_lock();
291		pin_kill(task_active_pid_ns(current)->bacct);
 
 
 
 
 
 
 
292	}
293
294	return error;
295}
296
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
297void acct_exit_ns(struct pid_namespace *ns)
298{
299	rcu_read_lock();
300	pin_kill(ns->bacct);
 
 
 
 
 
 
 
 
 
301}
302
303/*
304 *  encode an unsigned long into a comp_t
305 *
306 *  This routine has been adopted from the encode_comp_t() function in
307 *  the kern_acct.c file of the FreeBSD operating system. The encoding
308 *  is a 13-bit fraction with a 3-bit (base 8) exponent.
309 */
310
311#define	MANTSIZE	13			/* 13 bit mantissa. */
312#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
313#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
314
315static comp_t encode_comp_t(unsigned long value)
316{
317	int exp, rnd;
318
319	exp = rnd = 0;
320	while (value > MAXFRACT) {
321		rnd = value & (1 << (EXPSIZE - 1));	/* Round up? */
322		value >>= EXPSIZE;	/* Base 8 exponent == 3 bit shift. */
323		exp++;
324	}
325
326	/*
327	 * If we need to round up, do it (and handle overflow correctly).
328	 */
329	if (rnd && (++value > MAXFRACT)) {
330		value >>= EXPSIZE;
331		exp++;
332	}
333
334	/*
335	 * Clean it up and polish it off.
336	 */
337	exp <<= MANTSIZE;		/* Shift the exponent into place */
338	exp += value;			/* and add on the mantissa. */
339	return exp;
340}
341
342#if ACCT_VERSION == 1 || ACCT_VERSION == 2
343/*
344 * encode an u64 into a comp2_t (24 bits)
345 *
346 * Format: 5 bit base 2 exponent, 20 bits mantissa.
347 * The leading bit of the mantissa is not stored, but implied for
348 * non-zero exponents.
349 * Largest encodable value is 50 bits.
350 */
351
352#define MANTSIZE2       20                      /* 20 bit mantissa. */
353#define EXPSIZE2        5                       /* 5 bit base 2 exponent. */
354#define MAXFRACT2       ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
355#define MAXEXP2         ((1 << EXPSIZE2) - 1)    /* Maximum exponent. */
356
357static comp2_t encode_comp2_t(u64 value)
358{
359	int exp, rnd;
360
361	exp = (value > (MAXFRACT2>>1));
362	rnd = 0;
363	while (value > MAXFRACT2) {
364		rnd = value & 1;
365		value >>= 1;
366		exp++;
367	}
368
369	/*
370	 * If we need to round up, do it (and handle overflow correctly).
371	 */
372	if (rnd && (++value > MAXFRACT2)) {
373		value >>= 1;
374		exp++;
375	}
376
377	if (exp > MAXEXP2) {
378		/* Overflow. Return largest representable number instead. */
379		return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
380	} else {
381		return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
382	}
383}
384#endif
385
386#if ACCT_VERSION == 3
387/*
388 * encode an u64 into a 32 bit IEEE float
389 */
390static u32 encode_float(u64 value)
391{
392	unsigned exp = 190;
393	unsigned u;
394
395	if (value == 0)
396		return 0;
397	while ((s64)value > 0) {
398		value <<= 1;
399		exp--;
400	}
401	u = (u32)(value >> 40) & 0x7fffffu;
402	return u | (exp << 23);
403}
404#endif
405
406/*
407 *  Write an accounting entry for an exiting process
408 *
409 *  The acct_process() call is the workhorse of the process
410 *  accounting system. The struct acct is built here and then written
411 *  into the accounting file. This function should only be called from
412 *  do_exit() or when switching to a different output file.
413 */
414
415static void fill_ac(acct_t *ac)
 
 
 
 
416{
417	struct pacct_struct *pacct = &current->signal->pacct;
418	u64 elapsed, run_time;
 
 
 
 
 
419	struct tty_struct *tty;
 
 
 
 
 
 
 
 
 
 
 
420
421	/*
422	 * Fill the accounting struct with the needed info as recorded
423	 * by the different kernel functions.
424	 */
425	memset(ac, 0, sizeof(acct_t));
426
427	ac->ac_version = ACCT_VERSION | ACCT_BYTEORDER;
428	strlcpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm));
429
430	/* calculate run_time in nsec*/
431	run_time = ktime_get_ns();
432	run_time -= current->group_leader->start_time;
 
 
433	/* convert nsec -> AHZ */
434	elapsed = nsec_to_AHZ(run_time);
435#if ACCT_VERSION == 3
436	ac->ac_etime = encode_float(elapsed);
437#else
438	ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
439				(unsigned long) elapsed : (unsigned long) -1l);
440#endif
441#if ACCT_VERSION == 1 || ACCT_VERSION == 2
442	{
443		/* new enlarged etime field */
444		comp2_t etime = encode_comp2_t(elapsed);
445
446		ac->ac_etime_hi = etime >> 16;
447		ac->ac_etime_lo = (u16) etime;
448	}
449#endif
450	do_div(elapsed, AHZ);
451	ac->ac_btime = get_seconds() - elapsed;
 
 
 
452#if ACCT_VERSION==2
453	ac->ac_ahz = AHZ;
 
 
 
 
 
 
 
 
 
 
 
454#endif
455
456	spin_lock_irq(&current->sighand->siglock);
457	tty = current->signal->tty;	/* Safe as we hold the siglock */
458	ac->ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
459	ac->ac_utime = encode_comp_t(nsec_to_AHZ(pacct->ac_utime));
460	ac->ac_stime = encode_comp_t(nsec_to_AHZ(pacct->ac_stime));
461	ac->ac_flag = pacct->ac_flag;
462	ac->ac_mem = encode_comp_t(pacct->ac_mem);
463	ac->ac_minflt = encode_comp_t(pacct->ac_minflt);
464	ac->ac_majflt = encode_comp_t(pacct->ac_majflt);
465	ac->ac_exitcode = pacct->ac_exitcode;
466	spin_unlock_irq(&current->sighand->siglock);
467}
468/*
469 *  do_acct_process does all actual work. Caller holds the reference to file.
470 */
471static void do_acct_process(struct bsd_acct_struct *acct)
472{
473	acct_t ac;
474	unsigned long flim;
475	const struct cred *orig_cred;
476	struct file *file = acct->file;
477
478	/*
479	 * Accounting records are not subject to resource limits.
 
480	 */
481	flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
482	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
483	/* Perform file operations on behalf of whoever enabled accounting */
484	orig_cred = override_creds(file->f_cred);
485
486	/*
487	 * First check to see if there is enough free_space to continue
488	 * the process accounting system.
489	 */
490	if (!check_free_space(acct))
491		goto out;
492
493	fill_ac(&ac);
494	/* we really need to bite the bullet and change layout */
495	ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid);
496	ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid);
497#if ACCT_VERSION == 1 || ACCT_VERSION == 2
498	/* backward-compatible 16 bit fields */
499	ac.ac_uid16 = ac.ac_uid;
500	ac.ac_gid16 = ac.ac_gid;
501#endif
502#if ACCT_VERSION == 3
503	{
504		struct pid_namespace *ns = acct->ns;
505
506		ac.ac_pid = task_tgid_nr_ns(current, ns);
507		rcu_read_lock();
508		ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent),
509					     ns);
510		rcu_read_unlock();
511	}
512#endif
513	/*
514	 * Get freeze protection. If the fs is frozen, just skip the write
515	 * as we could deadlock the system otherwise.
516	 */
517	if (file_start_write_trylock(file)) {
518		/* it's been opened O_APPEND, so position is irrelevant */
519		loff_t pos = 0;
520		__kernel_write(file, &ac, sizeof(acct_t), &pos);
521		file_end_write(file);
522	}
 
523out:
524	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
525	revert_creds(orig_cred);
526}
527
528/**
529 * acct_collect - collect accounting information into pacct_struct
530 * @exitcode: task exit code
531 * @group_dead: not 0, if this thread is the last one in the process.
532 */
533void acct_collect(long exitcode, int group_dead)
534{
535	struct pacct_struct *pacct = &current->signal->pacct;
536	u64 utime, stime;
537	unsigned long vsize = 0;
538
539	if (group_dead && current->mm) {
540		struct vm_area_struct *vma;
541
542		down_read(&current->mm->mmap_sem);
543		vma = current->mm->mmap;
544		while (vma) {
545			vsize += vma->vm_end - vma->vm_start;
546			vma = vma->vm_next;
547		}
548		up_read(&current->mm->mmap_sem);
549	}
550
551	spin_lock_irq(&current->sighand->siglock);
552	if (group_dead)
553		pacct->ac_mem = vsize / 1024;
554	if (thread_group_leader(current)) {
555		pacct->ac_exitcode = exitcode;
556		if (current->flags & PF_FORKNOEXEC)
557			pacct->ac_flag |= AFORK;
558	}
559	if (current->flags & PF_SUPERPRIV)
560		pacct->ac_flag |= ASU;
561	if (current->flags & PF_DUMPCORE)
562		pacct->ac_flag |= ACORE;
563	if (current->flags & PF_SIGNALED)
564		pacct->ac_flag |= AXSIG;
565
566	task_cputime(current, &utime, &stime);
567	pacct->ac_utime += utime;
568	pacct->ac_stime += stime;
569	pacct->ac_minflt += current->min_flt;
570	pacct->ac_majflt += current->maj_flt;
571	spin_unlock_irq(&current->sighand->siglock);
572}
573
574static void slow_acct_process(struct pid_namespace *ns)
575{
576	for ( ; ns; ns = ns->parent) {
577		struct bsd_acct_struct *acct = acct_get(ns);
578		if (acct) {
579			do_acct_process(acct);
580			mutex_unlock(&acct->lock);
581			acct_put(acct);
582		}
 
 
 
 
 
 
 
 
583	}
 
 
 
 
 
584}
585
586/**
587 * acct_process
 
588 *
589 * handles process accounting for an exiting task
590 */
591void acct_process(void)
592{
593	struct pid_namespace *ns;
594
595	/*
596	 * This loop is safe lockless, since current is still
597	 * alive and holds its namespace, which in turn holds
598	 * its parent.
599	 */
600	for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) {
601		if (ns->bacct)
602			break;
603	}
604	if (unlikely(ns))
605		slow_acct_process(ns);
606}