1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
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>
64#include <linux/pid_namespace.h>
65#include <linux/fs_pin.h>
66
67
68
69
70
71
72
73
74int acct_parm[3] = {4, 2, 30};
75#define RESUME (acct_parm[0])
76#define SUSPEND (acct_parm[1])
77#define ACCT_TIMEOUT (acct_parm[2])
78
79
80
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
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
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
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);
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
264
265
266
267
268
269
270
271
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
305
306
307
308
309
310
311#define MANTSIZE 13
312#define EXPSIZE 3
313#define MAXFRACT ((1 << MANTSIZE) - 1)
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));
322 value >>= EXPSIZE;
323 exp++;
324 }
325
326
327
328
329 if (rnd && (++value > MAXFRACT)) {
330 value >>= EXPSIZE;
331 exp++;
332 }
333
334
335
336
337 exp <<= MANTSIZE;
338 exp += value;
339 return exp;
340}
341
342#if ACCT_VERSION == 1 || ACCT_VERSION == 2
343
344
345
346
347
348
349
350
351
352#define MANTSIZE2 20
353#define EXPSIZE2 5
354#define MAXFRACT2 ((1ul << MANTSIZE2) - 1)
355#define MAXEXP2 ((1 << EXPSIZE2) - 1)
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
371
372 if (rnd && (++value > MAXFRACT2)) {
373 value >>= 1;
374 exp++;
375 }
376
377 if (exp > MAXEXP2) {
378
379 return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
380 } else {
381 return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
382 }
383}
384#elif ACCT_VERSION == 3
385
386
387
388static u32 encode_float(u64 value)
389{
390 unsigned exp = 190;
391 unsigned u;
392
393 if (value == 0)
394 return 0;
395 while ((s64)value > 0) {
396 value <<= 1;
397 exp--;
398 }
399 u = (u32)(value >> 40) & 0x7fffffu;
400 return u | (exp << 23);
401}
402#endif
403
404
405
406
407
408
409
410
411
412
413static void fill_ac(acct_t *ac)
414{
415 struct pacct_struct *pacct = ¤t->signal->pacct;
416 u64 elapsed, run_time;
417 time64_t btime;
418 struct tty_struct *tty;
419
420
421
422
423
424 memset(ac, 0, sizeof(acct_t));
425
426 ac->ac_version = ACCT_VERSION | ACCT_BYTEORDER;
427 strlcpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm));
428
429
430 run_time = ktime_get_ns();
431 run_time -= current->group_leader->start_time;
432
433 elapsed = nsec_to_AHZ(run_time);
434#if ACCT_VERSION == 3
435 ac->ac_etime = encode_float(elapsed);
436#else
437 ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
438 (unsigned long) elapsed : (unsigned long) -1l);
439#endif
440#if ACCT_VERSION == 1 || ACCT_VERSION == 2
441 {
442
443 comp2_t etime = encode_comp2_t(elapsed);
444
445 ac->ac_etime_hi = etime >> 16;
446 ac->ac_etime_lo = (u16) etime;
447 }
448#endif
449 do_div(elapsed, AHZ);
450 btime = ktime_get_real_seconds() - elapsed;
451 ac->ac_btime = clamp_t(time64_t, btime, 0, U32_MAX);
452#if ACCT_VERSION==2
453 ac->ac_ahz = AHZ;
454#endif
455
456 spin_lock_irq(¤t->sighand->siglock);
457 tty = current->signal->tty;
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(¤t->sighand->siglock);
467}
468
469
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
480
481 flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
482 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
483
484 orig_cred = override_creds(file->f_cred);
485
486
487
488
489
490 if (!check_free_space(acct))
491 goto out;
492
493 fill_ac(&ac);
494
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
499 ac.ac_uid16 = ac.ac_uid;
500 ac.ac_gid16 = ac.ac_gid;
501#elif ACCT_VERSION == 3
502 {
503 struct pid_namespace *ns = acct->ns;
504
505 ac.ac_pid = task_tgid_nr_ns(current, ns);
506 rcu_read_lock();
507 ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent),
508 ns);
509 rcu_read_unlock();
510 }
511#endif
512
513
514
515
516 if (file_start_write_trylock(file)) {
517
518 loff_t pos = 0;
519 __kernel_write(file, &ac, sizeof(acct_t), &pos);
520 file_end_write(file);
521 }
522out:
523 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
524 revert_creds(orig_cred);
525}
526
527
528
529
530
531
532void acct_collect(long exitcode, int group_dead)
533{
534 struct pacct_struct *pacct = ¤t->signal->pacct;
535 u64 utime, stime;
536 unsigned long vsize = 0;
537
538 if (group_dead && current->mm) {
539 struct vm_area_struct *vma;
540
541 mmap_read_lock(current->mm);
542 vma = current->mm->mmap;
543 while (vma) {
544 vsize += vma->vm_end - vma->vm_start;
545 vma = vma->vm_next;
546 }
547 mmap_read_unlock(current->mm);
548 }
549
550 spin_lock_irq(¤t->sighand->siglock);
551 if (group_dead)
552 pacct->ac_mem = vsize / 1024;
553 if (thread_group_leader(current)) {
554 pacct->ac_exitcode = exitcode;
555 if (current->flags & PF_FORKNOEXEC)
556 pacct->ac_flag |= AFORK;
557 }
558 if (current->flags & PF_SUPERPRIV)
559 pacct->ac_flag |= ASU;
560 if (current->flags & PF_DUMPCORE)
561 pacct->ac_flag |= ACORE;
562 if (current->flags & PF_SIGNALED)
563 pacct->ac_flag |= AXSIG;
564
565 task_cputime(current, &utime, &stime);
566 pacct->ac_utime += utime;
567 pacct->ac_stime += stime;
568 pacct->ac_minflt += current->min_flt;
569 pacct->ac_majflt += current->maj_flt;
570 spin_unlock_irq(¤t->sighand->siglock);
571}
572
573static void slow_acct_process(struct pid_namespace *ns)
574{
575 for ( ; ns; ns = ns->parent) {
576 struct bsd_acct_struct *acct = acct_get(ns);
577 if (acct) {
578 do_acct_process(acct);
579 mutex_unlock(&acct->lock);
580 acct_put(acct);
581 }
582 }
583}
584
585
586
587
588void acct_process(void)
589{
590 struct pid_namespace *ns;
591
592
593
594
595
596
597 for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) {
598 if (ns->bacct)
599 break;
600 }
601 if (unlikely(ns))
602 slow_acct_process(ns);
603}
604