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/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
74 int 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
83 struct 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
96 static void do_acct_process(struct bsd_acct_struct *acct);
97
98 /*
99 * Check the amount of free space and suspend/resume accordingly.
100 */
check_free_space(struct bsd_acct_struct * acct)101 static 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;
129 out:
130 return acct->active;
131 }
132
acct_put(struct bsd_acct_struct * p)133 static 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
to_acct(struct fs_pin * p)139 static 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
acct_get(struct pid_namespace * ns)144 static struct bsd_acct_struct *acct_get(struct pid_namespace *ns)
145 {
146 struct bsd_acct_struct *res;
147 again:
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
acct_pin_kill(struct fs_pin * pin)170 static 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
close_work(struct work_struct * work)183 static 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
acct_on(struct filename * pathname)193 static 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
260 static 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 * sys_acct() is the only system call needed to implement process
267 * accounting. It takes the name of the file where accounting records
268 * should be written. If the filename is NULL, accounting will be
269 * shutdown.
270 *
271 * Returns: 0 for success or negative errno values for failure.
272 */
SYSCALL_DEFINE1(acct,const char __user *,name)273 SYSCALL_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
acct_exit_ns(struct pid_namespace * ns)297 void 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
encode_comp_t(unsigned long value)315 static 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 if (exp > (((comp_t) ~0U) >> MANTSIZE))
335 return (comp_t) ~0U;
336 /*
337 * Clean it up and polish it off.
338 */
339 exp <<= MANTSIZE; /* Shift the exponent into place */
340 exp += value; /* and add on the mantissa. */
341 return exp;
342 }
343
344 #if ACCT_VERSION == 1 || ACCT_VERSION == 2
345 /*
346 * encode an u64 into a comp2_t (24 bits)
347 *
348 * Format: 5 bit base 2 exponent, 20 bits mantissa.
349 * The leading bit of the mantissa is not stored, but implied for
350 * non-zero exponents.
351 * Largest encodable value is 50 bits.
352 */
353
354 #define MANTSIZE2 20 /* 20 bit mantissa. */
355 #define EXPSIZE2 5 /* 5 bit base 2 exponent. */
356 #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
357 #define MAXEXP2 ((1 << EXPSIZE2) - 1) /* Maximum exponent. */
358
encode_comp2_t(u64 value)359 static comp2_t encode_comp2_t(u64 value)
360 {
361 int exp, rnd;
362
363 exp = (value > (MAXFRACT2>>1));
364 rnd = 0;
365 while (value > MAXFRACT2) {
366 rnd = value & 1;
367 value >>= 1;
368 exp++;
369 }
370
371 /*
372 * If we need to round up, do it (and handle overflow correctly).
373 */
374 if (rnd && (++value > MAXFRACT2)) {
375 value >>= 1;
376 exp++;
377 }
378
379 if (exp > MAXEXP2) {
380 /* Overflow. Return largest representable number instead. */
381 return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
382 } else {
383 return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
384 }
385 }
386 #elif ACCT_VERSION == 3
387 /*
388 * encode an u64 into a 32 bit IEEE float
389 */
encode_float(u64 value)390 static 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
fill_ac(acct_t * ac)415 static void fill_ac(acct_t *ac)
416 {
417 struct pacct_struct *pacct = ¤t->signal->pacct;
418 u64 elapsed, run_time;
419 time64_t btime;
420 struct tty_struct *tty;
421
422 /*
423 * Fill the accounting struct with the needed info as recorded
424 * by the different kernel functions.
425 */
426 memset(ac, 0, sizeof(acct_t));
427
428 ac->ac_version = ACCT_VERSION | ACCT_BYTEORDER;
429 strlcpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm));
430
431 /* calculate run_time in nsec*/
432 run_time = ktime_get_ns();
433 run_time -= current->group_leader->start_time;
434 /* convert nsec -> AHZ */
435 elapsed = nsec_to_AHZ(run_time);
436 #if ACCT_VERSION == 3
437 ac->ac_etime = encode_float(elapsed);
438 #else
439 ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
440 (unsigned long) elapsed : (unsigned long) -1l);
441 #endif
442 #if ACCT_VERSION == 1 || ACCT_VERSION == 2
443 {
444 /* new enlarged etime field */
445 comp2_t etime = encode_comp2_t(elapsed);
446
447 ac->ac_etime_hi = etime >> 16;
448 ac->ac_etime_lo = (u16) etime;
449 }
450 #endif
451 do_div(elapsed, AHZ);
452 btime = ktime_get_real_seconds() - elapsed;
453 ac->ac_btime = clamp_t(time64_t, btime, 0, U32_MAX);
454 #if ACCT_VERSION==2
455 ac->ac_ahz = AHZ;
456 #endif
457
458 spin_lock_irq(¤t->sighand->siglock);
459 tty = current->signal->tty; /* Safe as we hold the siglock */
460 ac->ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
461 ac->ac_utime = encode_comp_t(nsec_to_AHZ(pacct->ac_utime));
462 ac->ac_stime = encode_comp_t(nsec_to_AHZ(pacct->ac_stime));
463 ac->ac_flag = pacct->ac_flag;
464 ac->ac_mem = encode_comp_t(pacct->ac_mem);
465 ac->ac_minflt = encode_comp_t(pacct->ac_minflt);
466 ac->ac_majflt = encode_comp_t(pacct->ac_majflt);
467 ac->ac_exitcode = pacct->ac_exitcode;
468 spin_unlock_irq(¤t->sighand->siglock);
469 }
470 /*
471 * do_acct_process does all actual work. Caller holds the reference to file.
472 */
do_acct_process(struct bsd_acct_struct * acct)473 static void do_acct_process(struct bsd_acct_struct *acct)
474 {
475 acct_t ac;
476 unsigned long flim;
477 const struct cred *orig_cred;
478 struct file *file = acct->file;
479
480 /*
481 * Accounting records are not subject to resource limits.
482 */
483 flim = rlimit(RLIMIT_FSIZE);
484 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
485 /* Perform file operations on behalf of whoever enabled accounting */
486 orig_cred = override_creds(file->f_cred);
487
488 /*
489 * First check to see if there is enough free_space to continue
490 * the process accounting system.
491 */
492 if (!check_free_space(acct))
493 goto out;
494
495 fill_ac(&ac);
496 /* we really need to bite the bullet and change layout */
497 ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid);
498 ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid);
499 #if ACCT_VERSION == 1 || ACCT_VERSION == 2
500 /* backward-compatible 16 bit fields */
501 ac.ac_uid16 = ac.ac_uid;
502 ac.ac_gid16 = ac.ac_gid;
503 #elif ACCT_VERSION == 3
504 {
505 struct pid_namespace *ns = acct->ns;
506
507 ac.ac_pid = task_tgid_nr_ns(current, ns);
508 rcu_read_lock();
509 ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent),
510 ns);
511 rcu_read_unlock();
512 }
513 #endif
514 /*
515 * Get freeze protection. If the fs is frozen, just skip the write
516 * as we could deadlock the system otherwise.
517 */
518 if (file_start_write_trylock(file)) {
519 /* it's been opened O_APPEND, so position is irrelevant */
520 loff_t pos = 0;
521 __kernel_write(file, &ac, sizeof(acct_t), &pos);
522 file_end_write(file);
523 }
524 out:
525 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
526 revert_creds(orig_cred);
527 }
528
529 /**
530 * acct_collect - collect accounting information into pacct_struct
531 * @exitcode: task exit code
532 * @group_dead: not 0, if this thread is the last one in the process.
533 */
acct_collect(long exitcode,int group_dead)534 void acct_collect(long exitcode, int group_dead)
535 {
536 struct pacct_struct *pacct = ¤t->signal->pacct;
537 u64 utime, stime;
538 unsigned long vsize = 0;
539
540 if (group_dead && current->mm) {
541 struct vm_area_struct *vma;
542
543 mmap_read_lock(current->mm);
544 vma = current->mm->mmap;
545 while (vma) {
546 vsize += vma->vm_end - vma->vm_start;
547 vma = vma->vm_next;
548 }
549 mmap_read_unlock(current->mm);
550 }
551
552 spin_lock_irq(¤t->sighand->siglock);
553 if (group_dead)
554 pacct->ac_mem = vsize / 1024;
555 if (thread_group_leader(current)) {
556 pacct->ac_exitcode = exitcode;
557 if (current->flags & PF_FORKNOEXEC)
558 pacct->ac_flag |= AFORK;
559 }
560 if (current->flags & PF_SUPERPRIV)
561 pacct->ac_flag |= ASU;
562 if (current->flags & PF_DUMPCORE)
563 pacct->ac_flag |= ACORE;
564 if (current->flags & PF_SIGNALED)
565 pacct->ac_flag |= AXSIG;
566
567 task_cputime(current, &utime, &stime);
568 pacct->ac_utime += utime;
569 pacct->ac_stime += stime;
570 pacct->ac_minflt += current->min_flt;
571 pacct->ac_majflt += current->maj_flt;
572 spin_unlock_irq(¤t->sighand->siglock);
573 }
574
slow_acct_process(struct pid_namespace * ns)575 static void slow_acct_process(struct pid_namespace *ns)
576 {
577 for ( ; ns; ns = ns->parent) {
578 struct bsd_acct_struct *acct = acct_get(ns);
579 if (acct) {
580 do_acct_process(acct);
581 mutex_unlock(&acct->lock);
582 acct_put(acct);
583 }
584 }
585 }
586
587 /**
588 * acct_process - handles process accounting for an exiting task
589 */
acct_process(void)590 void acct_process(void)
591 {
592 struct pid_namespace *ns;
593
594 /*
595 * This loop is safe lockless, since current is still
596 * alive and holds its namespace, which in turn holds
597 * its parent.
598 */
599 for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) {
600 if (ns->bacct)
601 break;
602 }
603 if (unlikely(ns))
604 slow_acct_process(ns);
605 }
606