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#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>
61#include <linux/pid_namespace.h>
62#include <linux/fs_pin.h>
63
64
65
66
67
68
69
70
71int acct_parm[3] = {4, 2, 30};
72#define RESUME (acct_parm[0])
73#define SUSPEND (acct_parm[1])
74#define ACCT_TIMEOUT (acct_parm[2])
75
76
77
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
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
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
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);
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
261
262
263
264
265
266
267
268
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
302
303
304
305
306
307
308#define MANTSIZE 13
309#define EXPSIZE 3
310#define MAXFRACT ((1 << MANTSIZE) - 1)
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));
319 value >>= EXPSIZE;
320 exp++;
321 }
322
323
324
325
326 if (rnd && (++value > MAXFRACT)) {
327 value >>= EXPSIZE;
328 exp++;
329 }
330
331
332
333
334 exp <<= MANTSIZE;
335 exp += value;
336 return exp;
337}
338
339#if ACCT_VERSION == 1 || ACCT_VERSION == 2
340
341
342
343
344
345
346
347
348
349#define MANTSIZE2 20
350#define EXPSIZE2 5
351#define MAXFRACT2 ((1ul << MANTSIZE2) - 1)
352#define MAXEXP2 ((1 << EXPSIZE2) - 1)
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
368
369 if (rnd && (++value > MAXFRACT2)) {
370 value >>= 1;
371 exp++;
372 }
373
374 if (exp > MAXEXP2) {
375
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
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
405
406
407
408
409
410
411
412static void fill_ac(acct_t *ac)
413{
414 struct pacct_struct *pacct = ¤t->signal->pacct;
415 u64 elapsed, run_time;
416 struct tty_struct *tty;
417
418
419
420
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
428 run_time = ktime_get_ns();
429 run_time -= current->group_leader->start_time;
430
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
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(¤t->sighand->siglock);
454 tty = current->signal->tty;
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(¤t->sighand->siglock);
464}
465
466
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
477
478 flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
479 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
480
481 orig_cred = override_creds(file->f_cred);
482
483
484
485
486
487 if (!check_free_space(acct))
488 goto out;
489
490 fill_ac(&ac);
491
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
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
512
513
514 if (file_start_write_trylock(file)) {
515
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
527
528
529
530void acct_collect(long exitcode, int group_dead)
531{
532 struct pacct_struct *pacct = ¤t->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(¤t->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(¤t->mm->mmap_sem);
546 }
547
548 spin_lock_irq(¤t->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(¤t->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
584
585
586
587void acct_process(void)
588{
589 struct pid_namespace *ns;
590
591
592
593
594
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}
603