Linux Audio

Check our new training course

Loading...
v4.6
 
  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 <linux/uaccess.h>
 
 
 59#include <asm/div64.h>
 60#include <linux/blkdev.h> /* sector_div */
 61#include <linux/pid_namespace.h>
 62#include <linux/fs_pin.h>
 63
 64/*
 65 * These constants control the amount of freespace that suspend and
 66 * resume the process accounting system, and the time delay between
 67 * each check.
 68 * Turned into sysctl-controllable parameters. AV, 12/11/98
 69 */
 70
 71int acct_parm[3] = {4, 2, 30};
 72#define RESUME		(acct_parm[0])	/* >foo% free space - resume */
 73#define SUSPEND		(acct_parm[1])	/* <foo% free space - suspend */
 74#define ACCT_TIMEOUT	(acct_parm[2])	/* foo second timeout between checks */
 75
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 76/*
 77 * External references and all of the globals.
 78 */
 79
 80struct bsd_acct_struct {
 81	struct fs_pin		pin;
 82	atomic_long_t		count;
 83	struct rcu_head		rcu;
 84	struct mutex		lock;
 85	int			active;
 
 86	unsigned long		needcheck;
 87	struct file		*file;
 88	struct pid_namespace	*ns;
 89	struct work_struct	work;
 90	struct completion	done;
 
 91};
 92
 93static void do_acct_process(struct bsd_acct_struct *acct);
 
 94
 95/*
 96 * Check the amount of free space and suspend/resume accordingly.
 97 */
 98static int check_free_space(struct bsd_acct_struct *acct)
 99{
100	struct kstatfs sbuf;
101
102	if (time_is_before_jiffies(acct->needcheck))
103		goto out;
104
105	/* May block */
106	if (vfs_statfs(&acct->file->f_path, &sbuf))
107		goto out;
108
109	if (acct->active) {
110		u64 suspend = sbuf.f_blocks * SUSPEND;
111		do_div(suspend, 100);
112		if (sbuf.f_bavail <= suspend) {
113			acct->active = 0;
114			pr_info("Process accounting paused\n");
115		}
116	} else {
117		u64 resume = sbuf.f_blocks * RESUME;
118		do_div(resume, 100);
119		if (sbuf.f_bavail >= resume) {
120			acct->active = 1;
121			pr_info("Process accounting resumed\n");
122		}
123	}
124
125	acct->needcheck = jiffies + ACCT_TIMEOUT*HZ;
126out:
127	return acct->active;
128}
129
130static void acct_put(struct bsd_acct_struct *p)
131{
132	if (atomic_long_dec_and_test(&p->count))
133		kfree_rcu(p, rcu);
134}
135
136static inline struct bsd_acct_struct *to_acct(struct fs_pin *p)
137{
138	return p ? container_of(p, struct bsd_acct_struct, pin) : NULL;
139}
140
141static struct bsd_acct_struct *acct_get(struct pid_namespace *ns)
142{
143	struct bsd_acct_struct *res;
144again:
145	smp_rmb();
146	rcu_read_lock();
147	res = to_acct(ACCESS_ONCE(ns->bacct));
148	if (!res) {
149		rcu_read_unlock();
150		return NULL;
151	}
152	if (!atomic_long_inc_not_zero(&res->count)) {
153		rcu_read_unlock();
154		cpu_relax();
155		goto again;
156	}
157	rcu_read_unlock();
158	mutex_lock(&res->lock);
159	if (res != to_acct(ACCESS_ONCE(ns->bacct))) {
160		mutex_unlock(&res->lock);
161		acct_put(res);
162		goto again;
163	}
164	return res;
165}
166
167static void acct_pin_kill(struct fs_pin *pin)
168{
169	struct bsd_acct_struct *acct = to_acct(pin);
170	mutex_lock(&acct->lock);
171	do_acct_process(acct);
 
 
 
 
172	schedule_work(&acct->work);
173	wait_for_completion(&acct->done);
174	cmpxchg(&acct->ns->bacct, pin, NULL);
175	mutex_unlock(&acct->lock);
176	pin_remove(pin);
177	acct_put(acct);
178}
179
180static void close_work(struct work_struct *work)
181{
182	struct bsd_acct_struct *acct = container_of(work, struct bsd_acct_struct, work);
183	struct file *file = acct->file;
 
 
 
184	if (file->f_op->flush)
185		file->f_op->flush(file, NULL);
186	__fput_sync(file);
187	complete(&acct->done);
188}
189
190static int acct_on(struct filename *pathname)
191{
192	struct file *file;
193	struct vfsmount *mnt, *internal;
194	struct pid_namespace *ns = task_active_pid_ns(current);
195	struct bsd_acct_struct *acct;
196	struct fs_pin *old;
197	int err;
198
199	acct = kzalloc(sizeof(struct bsd_acct_struct), GFP_KERNEL);
200	if (!acct)
201		return -ENOMEM;
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		kfree(acct);
207		return PTR_ERR(file);
208	}
209
210	if (!S_ISREG(file_inode(file)->i_mode)) {
211		kfree(acct);
212		filp_close(file, NULL);
213		return -EACCES;
214	}
215
 
 
 
 
 
 
 
 
 
 
 
 
 
 
216	if (!(file->f_mode & FMODE_CAN_WRITE)) {
217		kfree(acct);
218		filp_close(file, NULL);
219		return -EIO;
220	}
221	internal = mnt_clone_internal(&file->f_path);
222	if (IS_ERR(internal)) {
223		kfree(acct);
224		filp_close(file, NULL);
225		return PTR_ERR(internal);
226	}
227	err = mnt_want_write(internal);
228	if (err) {
229		mntput(internal);
230		kfree(acct);
231		filp_close(file, NULL);
232		return err;
233	}
234	mnt = file->f_path.mnt;
235	file->f_path.mnt = internal;
236
237	atomic_long_set(&acct->count, 1);
238	init_fs_pin(&acct->pin, acct_pin_kill);
239	acct->file = file;
240	acct->needcheck = jiffies;
241	acct->ns = ns;
242	mutex_init(&acct->lock);
243	INIT_WORK(&acct->work, close_work);
244	init_completion(&acct->done);
245	mutex_lock_nested(&acct->lock, 1);	/* nobody has seen it yet */
246	pin_insert(&acct->pin, mnt);
247
248	rcu_read_lock();
249	old = xchg(&ns->bacct, &acct->pin);
250	mutex_unlock(&acct->lock);
251	pin_kill(old);
252	mnt_drop_write(mnt);
253	mntput(mnt);
254	return 0;
255}
256
257static DEFINE_MUTEX(acct_on_mutex);
258
259/**
260 * sys_acct - enable/disable process accounting
261 * @name: file name for accounting records or NULL to shutdown accounting
262 *
263 * Returns 0 for success or negative errno values for failure.
264 *
265 * sys_acct() is the only system call needed to implement process
266 * accounting. It takes the name of the file where accounting records
267 * should be written. If the filename is NULL, accounting will be
268 * shutdown.
 
 
269 */
270SYSCALL_DEFINE1(acct, const char __user *, name)
271{
272	int error = 0;
273
274	if (!capable(CAP_SYS_PACCT))
275		return -EPERM;
276
277	if (name) {
278		struct filename *tmp = getname(name);
279
280		if (IS_ERR(tmp))
281			return PTR_ERR(tmp);
282		mutex_lock(&acct_on_mutex);
283		error = acct_on(tmp);
284		mutex_unlock(&acct_on_mutex);
285		putname(tmp);
286	} else {
287		rcu_read_lock();
288		pin_kill(task_active_pid_ns(current)->bacct);
289	}
290
291	return error;
292}
293
294void acct_exit_ns(struct pid_namespace *ns)
295{
296	rcu_read_lock();
297	pin_kill(ns->bacct);
298}
299
300/*
301 *  encode an unsigned long into a comp_t
302 *
303 *  This routine has been adopted from the encode_comp_t() function in
304 *  the kern_acct.c file of the FreeBSD operating system. The encoding
305 *  is a 13-bit fraction with a 3-bit (base 8) exponent.
306 */
307
308#define	MANTSIZE	13			/* 13 bit mantissa. */
309#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
310#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
311
312static comp_t encode_comp_t(unsigned long value)
313{
314	int exp, rnd;
315
316	exp = rnd = 0;
317	while (value > MAXFRACT) {
318		rnd = value & (1 << (EXPSIZE - 1));	/* Round up? */
319		value >>= EXPSIZE;	/* Base 8 exponent == 3 bit shift. */
320		exp++;
321	}
322
323	/*
324	 * If we need to round up, do it (and handle overflow correctly).
325	 */
326	if (rnd && (++value > MAXFRACT)) {
327		value >>= EXPSIZE;
328		exp++;
329	}
330
 
 
331	/*
332	 * Clean it up and polish it off.
333	 */
334	exp <<= MANTSIZE;		/* Shift the exponent into place */
335	exp += value;			/* and add on the mantissa. */
336	return exp;
337}
338
339#if ACCT_VERSION == 1 || ACCT_VERSION == 2
340/*
341 * encode an u64 into a comp2_t (24 bits)
342 *
343 * Format: 5 bit base 2 exponent, 20 bits mantissa.
344 * The leading bit of the mantissa is not stored, but implied for
345 * non-zero exponents.
346 * Largest encodable value is 50 bits.
347 */
348
349#define MANTSIZE2       20                      /* 20 bit mantissa. */
350#define EXPSIZE2        5                       /* 5 bit base 2 exponent. */
351#define MAXFRACT2       ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
352#define MAXEXP2         ((1 << EXPSIZE2) - 1)    /* Maximum exponent. */
353
354static comp2_t encode_comp2_t(u64 value)
355{
356	int exp, rnd;
357
358	exp = (value > (MAXFRACT2>>1));
359	rnd = 0;
360	while (value > MAXFRACT2) {
361		rnd = value & 1;
362		value >>= 1;
363		exp++;
364	}
365
366	/*
367	 * If we need to round up, do it (and handle overflow correctly).
368	 */
369	if (rnd && (++value > MAXFRACT2)) {
370		value >>= 1;
371		exp++;
372	}
373
374	if (exp > MAXEXP2) {
375		/* Overflow. Return largest representable number instead. */
376		return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
377	} else {
378		return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
379	}
380}
381#endif
382
383#if ACCT_VERSION == 3
384/*
385 * encode an u64 into a 32 bit IEEE float
386 */
387static u32 encode_float(u64 value)
388{
389	unsigned exp = 190;
390	unsigned u;
391
392	if (value == 0)
393		return 0;
394	while ((s64)value > 0) {
395		value <<= 1;
396		exp--;
397	}
398	u = (u32)(value >> 40) & 0x7fffffu;
399	return u | (exp << 23);
400}
401#endif
402
403/*
404 *  Write an accounting entry for an exiting process
405 *
406 *  The acct_process() call is the workhorse of the process
407 *  accounting system. The struct acct is built here and then written
408 *  into the accounting file. This function should only be called from
409 *  do_exit() or when switching to a different output file.
410 */
411
412static void fill_ac(acct_t *ac)
413{
414	struct pacct_struct *pacct = &current->signal->pacct;
 
 
415	u64 elapsed, run_time;
 
416	struct tty_struct *tty;
417
 
 
 
 
 
 
 
 
 
 
 
 
418	/*
419	 * Fill the accounting struct with the needed info as recorded
420	 * by the different kernel functions.
421	 */
422	memset(ac, 0, sizeof(acct_t));
423
424	ac->ac_version = ACCT_VERSION | ACCT_BYTEORDER;
425	strlcpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm));
426
427	/* calculate run_time in nsec*/
428	run_time = ktime_get_ns();
429	run_time -= current->group_leader->start_time;
430	/* convert nsec -> AHZ */
431	elapsed = nsec_to_AHZ(run_time);
432#if ACCT_VERSION == 3
433	ac->ac_etime = encode_float(elapsed);
434#else
435	ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
436				(unsigned long) elapsed : (unsigned long) -1l);
437#endif
438#if ACCT_VERSION == 1 || ACCT_VERSION == 2
439	{
440		/* new enlarged etime field */
441		comp2_t etime = encode_comp2_t(elapsed);
442
443		ac->ac_etime_hi = etime >> 16;
444		ac->ac_etime_lo = (u16) etime;
445	}
446#endif
447	do_div(elapsed, AHZ);
448	ac->ac_btime = get_seconds() - elapsed;
449#if ACCT_VERSION==2
 
450	ac->ac_ahz = AHZ;
451#endif
452
453	spin_lock_irq(&current->sighand->siglock);
454	tty = current->signal->tty;	/* Safe as we hold the siglock */
455	ac->ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
456	ac->ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime)));
457	ac->ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime)));
458	ac->ac_flag = pacct->ac_flag;
459	ac->ac_mem = encode_comp_t(pacct->ac_mem);
460	ac->ac_minflt = encode_comp_t(pacct->ac_minflt);
461	ac->ac_majflt = encode_comp_t(pacct->ac_majflt);
462	ac->ac_exitcode = pacct->ac_exitcode;
463	spin_unlock_irq(&current->sighand->siglock);
464}
465/*
466 *  do_acct_process does all actual work. Caller holds the reference to file.
467 */
468static void do_acct_process(struct bsd_acct_struct *acct)
469{
470	acct_t ac;
471	unsigned long flim;
472	const struct cred *orig_cred;
473	struct file *file = acct->file;
474
475	/*
476	 * Accounting records are not subject to resource limits.
477	 */
478	flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
479	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
480	/* Perform file operations on behalf of whoever enabled accounting */
481	orig_cred = override_creds(file->f_cred);
482
483	/*
484	 * First check to see if there is enough free_space to continue
485	 * the process accounting system.
486	 */
487	if (!check_free_space(acct))
488		goto out;
489
490	fill_ac(&ac);
491	/* we really need to bite the bullet and change layout */
492	ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid);
493	ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid);
494#if ACCT_VERSION == 1 || ACCT_VERSION == 2
495	/* backward-compatible 16 bit fields */
496	ac.ac_uid16 = ac.ac_uid;
497	ac.ac_gid16 = ac.ac_gid;
498#endif
499#if ACCT_VERSION == 3
500	{
501		struct pid_namespace *ns = acct->ns;
502
503		ac.ac_pid = task_tgid_nr_ns(current, ns);
504		rcu_read_lock();
505		ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent),
506					     ns);
507		rcu_read_unlock();
508	}
509#endif
 
 
 
 
 
 
 
 
 
 
 
510	/*
511	 * Get freeze protection. If the fs is frozen, just skip the write
512	 * as we could deadlock the system otherwise.
 
 
513	 */
514	if (file_start_write_trylock(file)) {
515		/* it's been opened O_APPEND, so position is irrelevant */
516		loff_t pos = 0;
517		__kernel_write(file, (char *)&ac, sizeof(acct_t), &pos);
518		file_end_write(file);
519	}
520out:
 
 
 
 
 
 
 
 
 
 
 
 
521	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
522	revert_creds(orig_cred);
523}
524
525/**
526 * acct_collect - collect accounting information into pacct_struct
527 * @exitcode: task exit code
528 * @group_dead: not 0, if this thread is the last one in the process.
529 */
530void acct_collect(long exitcode, int group_dead)
531{
532	struct pacct_struct *pacct = &current->signal->pacct;
533	cputime_t utime, stime;
534	unsigned long vsize = 0;
535
536	if (group_dead && current->mm) {
 
 
537		struct vm_area_struct *vma;
538
539		down_read(&current->mm->mmap_sem);
540		vma = current->mm->mmap;
541		while (vma) {
542			vsize += vma->vm_end - vma->vm_start;
543			vma = vma->vm_next;
544		}
545		up_read(&current->mm->mmap_sem);
546	}
547
548	spin_lock_irq(&current->sighand->siglock);
549	if (group_dead)
550		pacct->ac_mem = vsize / 1024;
551	if (thread_group_leader(current)) {
552		pacct->ac_exitcode = exitcode;
553		if (current->flags & PF_FORKNOEXEC)
554			pacct->ac_flag |= AFORK;
555	}
556	if (current->flags & PF_SUPERPRIV)
557		pacct->ac_flag |= ASU;
558	if (current->flags & PF_DUMPCORE)
559		pacct->ac_flag |= ACORE;
560	if (current->flags & PF_SIGNALED)
561		pacct->ac_flag |= AXSIG;
 
562	task_cputime(current, &utime, &stime);
563	pacct->ac_utime += utime;
564	pacct->ac_stime += stime;
565	pacct->ac_minflt += current->min_flt;
566	pacct->ac_majflt += current->maj_flt;
567	spin_unlock_irq(&current->sighand->siglock);
568}
569
570static void slow_acct_process(struct pid_namespace *ns)
571{
572	for ( ; ns; ns = ns->parent) {
573		struct bsd_acct_struct *acct = acct_get(ns);
574		if (acct) {
575			do_acct_process(acct);
576			mutex_unlock(&acct->lock);
577			acct_put(acct);
578		}
579	}
580}
581
582/**
583 * acct_process
584 *
585 * handles process accounting for an exiting task
586 */
587void acct_process(void)
588{
589	struct pid_namespace *ns;
590
591	/*
592	 * This loop is safe lockless, since current is still
593	 * alive and holds its namespace, which in turn holds
594	 * its parent.
595	 */
596	for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) {
597		if (ns->bacct)
598			break;
599	}
600	if (unlikely(ns))
601		slow_acct_process(ns);
602}
v6.13.7
  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 sys_umount(). If the accounting
 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_lock 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/pid_namespace.h>
 64#include <linux/fs_pin.h>
 65
 66/*
 67 * These constants control the amount of freespace that suspend and
 68 * resume the process accounting system, and the time delay between
 69 * each check.
 70 * Turned into sysctl-controllable parameters. AV, 12/11/98
 71 */
 72
 73static int acct_parm[3] = {4, 2, 30};
 74#define RESUME		(acct_parm[0])	/* >foo% free space - resume */
 75#define SUSPEND		(acct_parm[1])	/* <foo% free space - suspend */
 76#define ACCT_TIMEOUT	(acct_parm[2])	/* foo second timeout between checks */
 77
 78#ifdef CONFIG_SYSCTL
 79static struct ctl_table kern_acct_table[] = {
 80	{
 81		.procname       = "acct",
 82		.data           = &acct_parm,
 83		.maxlen         = 3*sizeof(int),
 84		.mode           = 0644,
 85		.proc_handler   = proc_dointvec,
 86	},
 87};
 88
 89static __init int kernel_acct_sysctls_init(void)
 90{
 91	register_sysctl_init("kernel", kern_acct_table);
 92	return 0;
 93}
 94late_initcall(kernel_acct_sysctls_init);
 95#endif /* CONFIG_SYSCTL */
 96
 97/*
 98 * External references and all of the globals.
 99 */
100
101struct bsd_acct_struct {
102	struct fs_pin		pin;
103	atomic_long_t		count;
104	struct rcu_head		rcu;
105	struct mutex		lock;
106	bool			active;
107	bool			check_space;
108	unsigned long		needcheck;
109	struct file		*file;
110	struct pid_namespace	*ns;
111	struct work_struct	work;
112	struct completion	done;
113	acct_t			ac;
114};
115
116static void fill_ac(struct bsd_acct_struct *acct);
117static void acct_write_process(struct bsd_acct_struct *acct);
118
119/*
120 * Check the amount of free space and suspend/resume accordingly.
121 */
122static bool check_free_space(struct bsd_acct_struct *acct)
123{
124	struct kstatfs sbuf;
125
126	if (!acct->check_space)
127		return acct->active;
128
129	/* May block */
130	if (vfs_statfs(&acct->file->f_path, &sbuf))
131		return acct->active;
132
133	if (acct->active) {
134		u64 suspend = sbuf.f_blocks * SUSPEND;
135		do_div(suspend, 100);
136		if (sbuf.f_bavail <= suspend) {
137			acct->active = false;
138			pr_info("Process accounting paused\n");
139		}
140	} else {
141		u64 resume = sbuf.f_blocks * RESUME;
142		do_div(resume, 100);
143		if (sbuf.f_bavail >= resume) {
144			acct->active = true;
145			pr_info("Process accounting resumed\n");
146		}
147	}
148
149	acct->needcheck = jiffies + ACCT_TIMEOUT*HZ;
 
150	return acct->active;
151}
152
153static void acct_put(struct bsd_acct_struct *p)
154{
155	if (atomic_long_dec_and_test(&p->count))
156		kfree_rcu(p, rcu);
157}
158
159static inline struct bsd_acct_struct *to_acct(struct fs_pin *p)
160{
161	return p ? container_of(p, struct bsd_acct_struct, pin) : NULL;
162}
163
164static struct bsd_acct_struct *acct_get(struct pid_namespace *ns)
165{
166	struct bsd_acct_struct *res;
167again:
168	smp_rmb();
169	rcu_read_lock();
170	res = to_acct(READ_ONCE(ns->bacct));
171	if (!res) {
172		rcu_read_unlock();
173		return NULL;
174	}
175	if (!atomic_long_inc_not_zero(&res->count)) {
176		rcu_read_unlock();
177		cpu_relax();
178		goto again;
179	}
180	rcu_read_unlock();
181	mutex_lock(&res->lock);
182	if (res != to_acct(READ_ONCE(ns->bacct))) {
183		mutex_unlock(&res->lock);
184		acct_put(res);
185		goto again;
186	}
187	return res;
188}
189
190static void acct_pin_kill(struct fs_pin *pin)
191{
192	struct bsd_acct_struct *acct = to_acct(pin);
193	mutex_lock(&acct->lock);
194	/*
195	 * Fill the accounting struct with the exiting task's info
196	 * before punting to the workqueue.
197	 */
198	fill_ac(acct);
199	schedule_work(&acct->work);
200	wait_for_completion(&acct->done);
201	cmpxchg(&acct->ns->bacct, pin, NULL);
202	mutex_unlock(&acct->lock);
203	pin_remove(pin);
204	acct_put(acct);
205}
206
207static void close_work(struct work_struct *work)
208{
209	struct bsd_acct_struct *acct = container_of(work, struct bsd_acct_struct, work);
210	struct file *file = acct->file;
211
212	/* We were fired by acct_pin_kill() which holds acct->lock. */
213	acct_write_process(acct);
214	if (file->f_op->flush)
215		file->f_op->flush(file, NULL);
216	__fput_sync(file);
217	complete(&acct->done);
218}
219
220static int acct_on(struct filename *pathname)
221{
222	struct file *file;
223	struct vfsmount *mnt, *internal;
224	struct pid_namespace *ns = task_active_pid_ns(current);
225	struct bsd_acct_struct *acct;
226	struct fs_pin *old;
227	int err;
228
229	acct = kzalloc(sizeof(struct bsd_acct_struct), GFP_KERNEL);
230	if (!acct)
231		return -ENOMEM;
232
233	/* Difference from BSD - they don't do O_APPEND */
234	file = file_open_name(pathname, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
235	if (IS_ERR(file)) {
236		kfree(acct);
237		return PTR_ERR(file);
238	}
239
240	if (!S_ISREG(file_inode(file)->i_mode)) {
241		kfree(acct);
242		filp_close(file, NULL);
243		return -EACCES;
244	}
245
246	/* Exclude kernel kernel internal filesystems. */
247	if (file_inode(file)->i_sb->s_flags & (SB_NOUSER | SB_KERNMOUNT)) {
248		kfree(acct);
249		filp_close(file, NULL);
250		return -EINVAL;
251	}
252
253	/* Exclude procfs and sysfs. */
254	if (file_inode(file)->i_sb->s_iflags & SB_I_USERNS_VISIBLE) {
255		kfree(acct);
256		filp_close(file, NULL);
257		return -EINVAL;
258	}
259
260	if (!(file->f_mode & FMODE_CAN_WRITE)) {
261		kfree(acct);
262		filp_close(file, NULL);
263		return -EIO;
264	}
265	internal = mnt_clone_internal(&file->f_path);
266	if (IS_ERR(internal)) {
267		kfree(acct);
268		filp_close(file, NULL);
269		return PTR_ERR(internal);
270	}
271	err = mnt_get_write_access(internal);
272	if (err) {
273		mntput(internal);
274		kfree(acct);
275		filp_close(file, NULL);
276		return err;
277	}
278	mnt = file->f_path.mnt;
279	file->f_path.mnt = internal;
280
281	atomic_long_set(&acct->count, 1);
282	init_fs_pin(&acct->pin, acct_pin_kill);
283	acct->file = file;
284	acct->needcheck = jiffies;
285	acct->ns = ns;
286	mutex_init(&acct->lock);
287	INIT_WORK(&acct->work, close_work);
288	init_completion(&acct->done);
289	mutex_lock_nested(&acct->lock, 1);	/* nobody has seen it yet */
290	pin_insert(&acct->pin, mnt);
291
292	rcu_read_lock();
293	old = xchg(&ns->bacct, &acct->pin);
294	mutex_unlock(&acct->lock);
295	pin_kill(old);
296	mnt_put_write_access(mnt);
297	mntput(mnt);
298	return 0;
299}
300
301static DEFINE_MUTEX(acct_on_mutex);
302
303/**
304 * sys_acct - enable/disable process accounting
305 * @name: file name for accounting records or NULL to shutdown accounting
306 *
 
 
307 * sys_acct() is the only system call needed to implement process
308 * accounting. It takes the name of the file where accounting records
309 * should be written. If the filename is NULL, accounting will be
310 * shutdown.
311 *
312 * Returns: 0 for success or negative errno values for failure.
313 */
314SYSCALL_DEFINE1(acct, const char __user *, name)
315{
316	int error = 0;
317
318	if (!capable(CAP_SYS_PACCT))
319		return -EPERM;
320
321	if (name) {
322		struct filename *tmp = getname(name);
323
324		if (IS_ERR(tmp))
325			return PTR_ERR(tmp);
326		mutex_lock(&acct_on_mutex);
327		error = acct_on(tmp);
328		mutex_unlock(&acct_on_mutex);
329		putname(tmp);
330	} else {
331		rcu_read_lock();
332		pin_kill(task_active_pid_ns(current)->bacct);
333	}
334
335	return error;
336}
337
338void acct_exit_ns(struct pid_namespace *ns)
339{
340	rcu_read_lock();
341	pin_kill(ns->bacct);
342}
343
344/*
345 *  encode an u64 into a comp_t
346 *
347 *  This routine has been adopted from the encode_comp_t() function in
348 *  the kern_acct.c file of the FreeBSD operating system. The encoding
349 *  is a 13-bit fraction with a 3-bit (base 8) exponent.
350 */
351
352#define	MANTSIZE	13			/* 13 bit mantissa. */
353#define	EXPSIZE		3			/* Base 8 (3 bit) exponent. */
354#define	MAXFRACT	((1 << MANTSIZE) - 1)	/* Maximum fractional value. */
355
356static comp_t encode_comp_t(u64 value)
357{
358	int exp, rnd;
359
360	exp = rnd = 0;
361	while (value > MAXFRACT) {
362		rnd = value & (1 << (EXPSIZE - 1));	/* Round up? */
363		value >>= EXPSIZE;	/* Base 8 exponent == 3 bit shift. */
364		exp++;
365	}
366
367	/*
368	 * If we need to round up, do it (and handle overflow correctly).
369	 */
370	if (rnd && (++value > MAXFRACT)) {
371		value >>= EXPSIZE;
372		exp++;
373	}
374
375	if (exp > (((comp_t) ~0U) >> MANTSIZE))
376		return (comp_t) ~0U;
377	/*
378	 * Clean it up and polish it off.
379	 */
380	exp <<= MANTSIZE;		/* Shift the exponent into place */
381	exp += value;			/* and add on the mantissa. */
382	return exp;
383}
384
385#if ACCT_VERSION == 1 || ACCT_VERSION == 2
386/*
387 * encode an u64 into a comp2_t (24 bits)
388 *
389 * Format: 5 bit base 2 exponent, 20 bits mantissa.
390 * The leading bit of the mantissa is not stored, but implied for
391 * non-zero exponents.
392 * Largest encodable value is 50 bits.
393 */
394
395#define MANTSIZE2       20                      /* 20 bit mantissa. */
396#define EXPSIZE2        5                       /* 5 bit base 2 exponent. */
397#define MAXFRACT2       ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
398#define MAXEXP2         ((1 << EXPSIZE2) - 1)    /* Maximum exponent. */
399
400static comp2_t encode_comp2_t(u64 value)
401{
402	int exp, rnd;
403
404	exp = (value > (MAXFRACT2>>1));
405	rnd = 0;
406	while (value > MAXFRACT2) {
407		rnd = value & 1;
408		value >>= 1;
409		exp++;
410	}
411
412	/*
413	 * If we need to round up, do it (and handle overflow correctly).
414	 */
415	if (rnd && (++value > MAXFRACT2)) {
416		value >>= 1;
417		exp++;
418	}
419
420	if (exp > MAXEXP2) {
421		/* Overflow. Return largest representable number instead. */
422		return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
423	} else {
424		return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
425	}
426}
427#elif ACCT_VERSION == 3
 
 
428/*
429 * encode an u64 into a 32 bit IEEE float
430 */
431static u32 encode_float(u64 value)
432{
433	unsigned exp = 190;
434	unsigned u;
435
436	if (value == 0)
437		return 0;
438	while ((s64)value > 0) {
439		value <<= 1;
440		exp--;
441	}
442	u = (u32)(value >> 40) & 0x7fffffu;
443	return u | (exp << 23);
444}
445#endif
446
447/*
448 *  Write an accounting entry for an exiting process
449 *
450 *  The acct_process() call is the workhorse of the process
451 *  accounting system. The struct acct is built here and then written
452 *  into the accounting file. This function should only be called from
453 *  do_exit() or when switching to a different output file.
454 */
455
456static void fill_ac(struct bsd_acct_struct *acct)
457{
458	struct pacct_struct *pacct = &current->signal->pacct;
459	struct file *file = acct->file;
460	acct_t *ac = &acct->ac;
461	u64 elapsed, run_time;
462	time64_t btime;
463	struct tty_struct *tty;
464
465	lockdep_assert_held(&acct->lock);
466
467	if (time_is_after_jiffies(acct->needcheck)) {
468		acct->check_space = false;
469
470		/* Don't fill in @ac if nothing will be written. */
471		if (!acct->active)
472			return;
473	} else {
474		acct->check_space = true;
475	}
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	strscpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm));
485
486	/* calculate run_time in nsec*/
487	run_time = ktime_get_ns();
488	run_time -= current->group_leader->start_time;
489	/* convert nsec -> AHZ */
490	elapsed = nsec_to_AHZ(run_time);
491#if ACCT_VERSION == 3
492	ac->ac_etime = encode_float(elapsed);
493#else
494	ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
495				(unsigned long) elapsed : (unsigned long) -1l);
496#endif
497#if ACCT_VERSION == 1 || ACCT_VERSION == 2
498	{
499		/* new enlarged etime field */
500		comp2_t etime = encode_comp2_t(elapsed);
501
502		ac->ac_etime_hi = etime >> 16;
503		ac->ac_etime_lo = (u16) etime;
504	}
505#endif
506	do_div(elapsed, AHZ);
507	btime = ktime_get_real_seconds() - elapsed;
508	ac->ac_btime = clamp_t(time64_t, btime, 0, U32_MAX);
509#if ACCT_VERSION == 2
510	ac->ac_ahz = AHZ;
511#endif
512
513	spin_lock_irq(&current->sighand->siglock);
514	tty = current->signal->tty;	/* Safe as we hold the siglock */
515	ac->ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
516	ac->ac_utime = encode_comp_t(nsec_to_AHZ(pacct->ac_utime));
517	ac->ac_stime = encode_comp_t(nsec_to_AHZ(pacct->ac_stime));
518	ac->ac_flag = pacct->ac_flag;
519	ac->ac_mem = encode_comp_t(pacct->ac_mem);
520	ac->ac_minflt = encode_comp_t(pacct->ac_minflt);
521	ac->ac_majflt = encode_comp_t(pacct->ac_majflt);
522	ac->ac_exitcode = pacct->ac_exitcode;
523	spin_unlock_irq(&current->sighand->siglock);
 
 
 
 
 
 
 
 
 
 
524
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
525	/* we really need to bite the bullet and change layout */
526	ac->ac_uid = from_kuid_munged(file->f_cred->user_ns, current_uid());
527	ac->ac_gid = from_kgid_munged(file->f_cred->user_ns, current_gid());
528#if ACCT_VERSION == 1 || ACCT_VERSION == 2
529	/* backward-compatible 16 bit fields */
530	ac->ac_uid16 = ac->ac_uid;
531	ac->ac_gid16 = ac->ac_gid;
532#elif ACCT_VERSION == 3
 
533	{
534		struct pid_namespace *ns = acct->ns;
535
536		ac->ac_pid = task_tgid_nr_ns(current, ns);
537		rcu_read_lock();
538		ac->ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent), ns);
 
539		rcu_read_unlock();
540	}
541#endif
542}
543
544static void acct_write_process(struct bsd_acct_struct *acct)
545{
546	struct file *file = acct->file;
547	const struct cred *cred;
548	acct_t *ac = &acct->ac;
549
550	/* Perform file operations on behalf of whoever enabled accounting */
551	cred = override_creds(file->f_cred);
552
553	/*
554	 * First check to see if there is enough free_space to continue
555	 * the process accounting system. Then get freeze protection. If
556	 * the fs is frozen, just skip the write as we could deadlock
557	 * the system otherwise.
558	 */
559	if (check_free_space(acct) && file_start_write_trylock(file)) {
560		/* it's been opened O_APPEND, so position is irrelevant */
561		loff_t pos = 0;
562		__kernel_write(file, ac, sizeof(acct_t), &pos);
563		file_end_write(file);
564	}
565
566	revert_creds(cred);
567}
568
569static void do_acct_process(struct bsd_acct_struct *acct)
570{
571	unsigned long flim;
572
573	/* Accounting records are not subject to resource limits. */
574	flim = rlimit(RLIMIT_FSIZE);
575	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
576	fill_ac(acct);
577	acct_write_process(acct);
578	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
 
579}
580
581/**
582 * acct_collect - collect accounting information into pacct_struct
583 * @exitcode: task exit code
584 * @group_dead: not 0, if this thread is the last one in the process.
585 */
586void acct_collect(long exitcode, int group_dead)
587{
588	struct pacct_struct *pacct = &current->signal->pacct;
589	u64 utime, stime;
590	unsigned long vsize = 0;
591
592	if (group_dead && current->mm) {
593		struct mm_struct *mm = current->mm;
594		VMA_ITERATOR(vmi, mm, 0);
595		struct vm_area_struct *vma;
596
597		mmap_read_lock(mm);
598		for_each_vma(vmi, vma)
 
599			vsize += vma->vm_end - vma->vm_start;
600		mmap_read_unlock(mm);
 
 
601	}
602
603	spin_lock_irq(&current->sighand->siglock);
604	if (group_dead)
605		pacct->ac_mem = vsize / 1024;
606	if (thread_group_leader(current)) {
607		pacct->ac_exitcode = exitcode;
608		if (current->flags & PF_FORKNOEXEC)
609			pacct->ac_flag |= AFORK;
610	}
611	if (current->flags & PF_SUPERPRIV)
612		pacct->ac_flag |= ASU;
613	if (current->flags & PF_DUMPCORE)
614		pacct->ac_flag |= ACORE;
615	if (current->flags & PF_SIGNALED)
616		pacct->ac_flag |= AXSIG;
617
618	task_cputime(current, &utime, &stime);
619	pacct->ac_utime += utime;
620	pacct->ac_stime += stime;
621	pacct->ac_minflt += current->min_flt;
622	pacct->ac_majflt += current->maj_flt;
623	spin_unlock_irq(&current->sighand->siglock);
624}
625
626static void slow_acct_process(struct pid_namespace *ns)
627{
628	for ( ; ns; ns = ns->parent) {
629		struct bsd_acct_struct *acct = acct_get(ns);
630		if (acct) {
631			do_acct_process(acct);
632			mutex_unlock(&acct->lock);
633			acct_put(acct);
634		}
635	}
636}
637
638/**
639 * acct_process - handles process accounting for an exiting task
 
 
640 */
641void acct_process(void)
642{
643	struct pid_namespace *ns;
644
645	/*
646	 * This loop is safe lockless, since current is still
647	 * alive and holds its namespace, which in turn holds
648	 * its parent.
649	 */
650	for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) {
651		if (ns->bacct)
652			break;
653	}
654	if (unlikely(ns))
655		slow_acct_process(ns);
656}