]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/acct.c
4 * BSD Process Accounting for Linux
6 * Author: Marco van Wieringen <mvw@planets.elm.net>
8 * Some code based on ideas and code from:
9 * Thomas K. Dyas <tdyas@eden.rutgers.edu>
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.
17 * (C) Copyright 1995 - 1997 Marco van Wieringen - ELM Consultancy B.V.
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.
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.
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.
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.
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.
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>
64 * These constants control the amount of freespace that suspend and
65 * resume the process accounting system, and the time delay between
67 * Turned into sysctl-controllable parameters. AV, 12/11/98
70 int 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 */
76 * External references and all of the globals.
78 static void do_acct_process(struct bsd_acct_struct
*acct
,
79 struct pid_namespace
*ns
, struct file
*);
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.
86 struct bsd_acct_struct
{
88 volatile int needcheck
;
90 struct pid_namespace
*ns
;
91 struct timer_list timer
;
94 static DEFINE_SPINLOCK(acct_lock
);
97 * Called whenever the timer says to check the free space.
99 static void acct_timeout(unsigned long x
)
101 struct bsd_acct_struct
*acct
= (struct bsd_acct_struct
*)x
;
106 * Check the amount of free space and suspend/resume accordingly.
108 static int check_free_space(struct bsd_acct_struct
*acct
, struct file
*file
)
116 spin_lock(&acct_lock
);
118 if (!file
|| !acct
->needcheck
)
120 spin_unlock(&acct_lock
);
123 if (vfs_statfs(file
->f_path
.dentry
, &sbuf
))
125 suspend
= sbuf
.f_blocks
* SUSPEND
;
126 resume
= sbuf
.f_blocks
* RESUME
;
128 sector_div(suspend
, 100);
129 sector_div(resume
, 100);
131 if (sbuf
.f_bavail
<= suspend
)
133 else if (sbuf
.f_bavail
>= resume
)
139 * If some joker switched acct->file under us we'ld better be
140 * silent and _not_ touch anything.
142 spin_lock(&acct_lock
);
143 if (file
!= acct
->file
) {
152 printk(KERN_INFO
"Process accounting paused\n");
157 printk(KERN_INFO
"Process accounting resumed\n");
161 del_timer(&acct
->timer
);
163 acct
->timer
.expires
= jiffies
+ ACCT_TIMEOUT
*HZ
;
164 add_timer(&acct
->timer
);
167 spin_unlock(&acct_lock
);
172 * Close the old accounting file (if currently open) and then replace
173 * it with file (if non-NULL).
175 * NOTE: acct_lock MUST be held on entry and exit.
177 static void acct_file_reopen(struct bsd_acct_struct
*acct
, struct file
*file
,
178 struct pid_namespace
*ns
)
180 struct file
*old_acct
= NULL
;
181 struct pid_namespace
*old_ns
= NULL
;
184 old_acct
= acct
->file
;
186 del_timer(&acct
->timer
);
197 /* It's been deleted if it was used before so this is safe */
198 setup_timer(&acct
->timer
, acct_timeout
, (unsigned long)acct
);
199 acct
->timer
.expires
= jiffies
+ ACCT_TIMEOUT
*HZ
;
200 add_timer(&acct
->timer
);
203 mnt_unpin(old_acct
->f_path
.mnt
);
204 spin_unlock(&acct_lock
);
205 do_acct_process(acct
, old_ns
, old_acct
);
206 filp_close(old_acct
, NULL
);
207 spin_lock(&acct_lock
);
211 static int acct_on(char *name
)
215 struct pid_namespace
*ns
;
216 struct bsd_acct_struct
*acct
= NULL
;
218 /* Difference from BSD - they don't do O_APPEND */
219 file
= filp_open(name
, O_WRONLY
|O_APPEND
|O_LARGEFILE
, 0);
221 return PTR_ERR(file
);
223 if (!S_ISREG(file
->f_path
.dentry
->d_inode
->i_mode
)) {
224 filp_close(file
, NULL
);
228 if (!file
->f_op
->write
) {
229 filp_close(file
, NULL
);
233 ns
= task_active_pid_ns(current
);
234 if (ns
->bacct
== NULL
) {
235 acct
= kzalloc(sizeof(struct bsd_acct_struct
), GFP_KERNEL
);
237 filp_close(file
, NULL
);
242 error
= security_acct(file
);
245 filp_close(file
, NULL
);
249 spin_lock(&acct_lock
);
250 if (ns
->bacct
== NULL
) {
255 mnt_pin(file
->f_path
.mnt
);
256 acct_file_reopen(ns
->bacct
, file
, ns
);
257 spin_unlock(&acct_lock
);
259 mntput(file
->f_path
.mnt
); /* it's pinned, now give up active reference */
266 * sys_acct - enable/disable process accounting
267 * @name: file name for accounting records or NULL to shutdown accounting
269 * Returns 0 for success or negative errno values for failure.
271 * sys_acct() is the only system call needed to implement process
272 * accounting. It takes the name of the file where accounting records
273 * should be written. If the filename is NULL, accounting will be
276 asmlinkage
long sys_acct(const char __user
*name
)
280 if (!capable(CAP_SYS_PACCT
))
284 char *tmp
= getname(name
);
286 return (PTR_ERR(tmp
));
287 error
= acct_on(tmp
);
290 struct bsd_acct_struct
*acct
;
292 acct
= task_active_pid_ns(current
)->bacct
;
296 error
= security_acct(NULL
);
298 spin_lock(&acct_lock
);
299 acct_file_reopen(acct
, NULL
, NULL
);
300 spin_unlock(&acct_lock
);
307 * acct_auto_close - turn off a filesystem's accounting if it is on
308 * @m: vfsmount being shut down
310 * If the accounting is turned on for a file in the subtree pointed to
311 * to by m, turn accounting off. Done when m is about to die.
313 void acct_auto_close_mnt(struct vfsmount
*m
)
315 struct bsd_acct_struct
*acct
;
317 acct
= init_pid_ns
.bacct
;
321 spin_lock(&acct_lock
);
322 if (acct
->file
&& acct
->file
->f_path
.mnt
== m
)
323 acct_file_reopen(acct
, NULL
, NULL
);
324 spin_unlock(&acct_lock
);
328 * acct_auto_close - turn off a filesystem's accounting if it is on
329 * @sb: super block for the filesystem
331 * If the accounting is turned on for a file in the filesystem pointed
332 * to by sb, turn accounting off.
334 void acct_auto_close(struct super_block
*sb
)
336 struct bsd_acct_struct
*acct
;
338 acct
= init_pid_ns
.bacct
;
342 spin_lock(&acct_lock
);
343 if (acct
->file
&& acct
->file
->f_path
.mnt
->mnt_sb
== sb
)
344 acct_file_reopen(acct
, NULL
, NULL
);
345 spin_unlock(&acct_lock
);
348 void acct_exit_ns(struct pid_namespace
*ns
)
350 struct bsd_acct_struct
*acct
;
352 spin_lock(&acct_lock
);
355 if (acct
->file
!= NULL
)
356 acct_file_reopen(acct
, NULL
, NULL
);
360 spin_unlock(&acct_lock
);
364 * encode an unsigned long into a comp_t
366 * This routine has been adopted from the encode_comp_t() function in
367 * the kern_acct.c file of the FreeBSD operating system. The encoding
368 * is a 13-bit fraction with a 3-bit (base 8) exponent.
371 #define MANTSIZE 13 /* 13 bit mantissa. */
372 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
373 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
375 static comp_t
encode_comp_t(unsigned long value
)
380 while (value
> MAXFRACT
) {
381 rnd
= value
& (1 << (EXPSIZE
- 1)); /* Round up? */
382 value
>>= EXPSIZE
; /* Base 8 exponent == 3 bit shift. */
387 * If we need to round up, do it (and handle overflow correctly).
389 if (rnd
&& (++value
> MAXFRACT
)) {
395 * Clean it up and polish it off.
397 exp
<<= MANTSIZE
; /* Shift the exponent into place */
398 exp
+= value
; /* and add on the mantissa. */
402 #if ACCT_VERSION==1 || ACCT_VERSION==2
404 * encode an u64 into a comp2_t (24 bits)
406 * Format: 5 bit base 2 exponent, 20 bits mantissa.
407 * The leading bit of the mantissa is not stored, but implied for
408 * non-zero exponents.
409 * Largest encodable value is 50 bits.
412 #define MANTSIZE2 20 /* 20 bit mantissa. */
413 #define EXPSIZE2 5 /* 5 bit base 2 exponent. */
414 #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
415 #define MAXEXP2 ((1 <<EXPSIZE2) - 1) /* Maximum exponent. */
417 static comp2_t
encode_comp2_t(u64 value
)
421 exp
= (value
> (MAXFRACT2
>>1));
423 while (value
> MAXFRACT2
) {
430 * If we need to round up, do it (and handle overflow correctly).
432 if (rnd
&& (++value
> MAXFRACT2
)) {
438 /* Overflow. Return largest representable number instead. */
439 return (1ul << (MANTSIZE2
+EXPSIZE2
-1)) - 1;
441 return (value
& (MAXFRACT2
>>1)) | (exp
<< (MANTSIZE2
-1));
448 * encode an u64 into a 32 bit IEEE float
450 static u32
encode_float(u64 value
)
455 if (value
==0) return 0;
456 while ((s64
)value
> 0){
460 u
= (u32
)(value
>> 40) & 0x7fffffu
;
461 return u
| (exp
<< 23);
466 * Write an accounting entry for an exiting process
468 * The acct_process() call is the workhorse of the process
469 * accounting system. The struct acct is built here and then written
470 * into the accounting file. This function should only be called from
471 * do_exit() or when switching to a different output file.
475 * do_acct_process does all actual work. Caller holds the reference to file.
477 static void do_acct_process(struct bsd_acct_struct
*acct
,
478 struct pid_namespace
*ns
, struct file
*file
)
480 struct pacct_struct
*pacct
= ¤t
->signal
->pacct
;
486 struct timespec uptime
;
487 struct tty_struct
*tty
;
490 * First check to see if there is enough free_space to continue
491 * the process accounting system.
493 if (!check_free_space(acct
, file
))
497 * Fill the accounting struct with the needed info as recorded
498 * by the different kernel functions.
500 memset((caddr_t
)&ac
, 0, sizeof(acct_t
));
502 ac
.ac_version
= ACCT_VERSION
| ACCT_BYTEORDER
;
503 strlcpy(ac
.ac_comm
, current
->comm
, sizeof(ac
.ac_comm
));
505 /* calculate run_time in nsec*/
506 do_posix_clock_monotonic_gettime(&uptime
);
507 run_time
= (u64
)uptime
.tv_sec
*NSEC_PER_SEC
+ uptime
.tv_nsec
;
508 run_time
-= (u64
)current
->group_leader
->start_time
.tv_sec
* NSEC_PER_SEC
509 + current
->group_leader
->start_time
.tv_nsec
;
510 /* convert nsec -> AHZ */
511 elapsed
= nsec_to_AHZ(run_time
);
513 ac
.ac_etime
= encode_float(elapsed
);
515 ac
.ac_etime
= encode_comp_t(elapsed
< (unsigned long) -1l ?
516 (unsigned long) elapsed
: (unsigned long) -1l);
518 #if ACCT_VERSION==1 || ACCT_VERSION==2
520 /* new enlarged etime field */
521 comp2_t etime
= encode_comp2_t(elapsed
);
522 ac
.ac_etime_hi
= etime
>> 16;
523 ac
.ac_etime_lo
= (u16
) etime
;
526 do_div(elapsed
, AHZ
);
527 ac
.ac_btime
= get_seconds() - elapsed
;
528 /* we really need to bite the bullet and change layout */
529 ac
.ac_uid
= current
->uid
;
530 ac
.ac_gid
= current
->gid
;
534 #if ACCT_VERSION==1 || ACCT_VERSION==2
535 /* backward-compatible 16 bit fields */
536 ac
.ac_uid16
= current
->uid
;
537 ac
.ac_gid16
= current
->gid
;
540 ac
.ac_pid
= task_tgid_nr_ns(current
, ns
);
542 ac
.ac_ppid
= task_tgid_nr_ns(rcu_dereference(current
->real_parent
), ns
);
546 spin_lock_irq(¤t
->sighand
->siglock
);
547 tty
= current
->signal
->tty
;
548 ac
.ac_tty
= tty
? old_encode_dev(tty_devnum(tty
)) : 0;
549 ac
.ac_utime
= encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct
->ac_utime
)));
550 ac
.ac_stime
= encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct
->ac_stime
)));
551 ac
.ac_flag
= pacct
->ac_flag
;
552 ac
.ac_mem
= encode_comp_t(pacct
->ac_mem
);
553 ac
.ac_minflt
= encode_comp_t(pacct
->ac_minflt
);
554 ac
.ac_majflt
= encode_comp_t(pacct
->ac_majflt
);
555 ac
.ac_exitcode
= pacct
->ac_exitcode
;
556 spin_unlock_irq(¤t
->sighand
->siglock
);
557 ac
.ac_io
= encode_comp_t(0 /* current->io_usage */); /* %% */
558 ac
.ac_rw
= encode_comp_t(ac
.ac_io
/ 1024);
559 ac
.ac_swaps
= encode_comp_t(0);
562 * Kernel segment override to datasegment and write it
563 * to the accounting file.
568 * Accounting records are not subject to resource limits.
570 flim
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
571 current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
= RLIM_INFINITY
;
572 file
->f_op
->write(file
, (char *)&ac
,
573 sizeof(acct_t
), &file
->f_pos
);
574 current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
= flim
;
579 * acct_init_pacct - initialize a new pacct_struct
580 * @pacct: per-process accounting info struct to initialize
582 void acct_init_pacct(struct pacct_struct
*pacct
)
584 memset(pacct
, 0, sizeof(struct pacct_struct
));
585 pacct
->ac_utime
= pacct
->ac_stime
= cputime_zero
;
589 * acct_collect - collect accounting information into pacct_struct
590 * @exitcode: task exit code
591 * @group_dead: not 0, if this thread is the last one in the process.
593 void acct_collect(long exitcode
, int group_dead
)
595 struct pacct_struct
*pacct
= ¤t
->signal
->pacct
;
596 unsigned long vsize
= 0;
598 if (group_dead
&& current
->mm
) {
599 struct vm_area_struct
*vma
;
600 down_read(¤t
->mm
->mmap_sem
);
601 vma
= current
->mm
->mmap
;
603 vsize
+= vma
->vm_end
- vma
->vm_start
;
606 up_read(¤t
->mm
->mmap_sem
);
609 spin_lock_irq(¤t
->sighand
->siglock
);
611 pacct
->ac_mem
= vsize
/ 1024;
612 if (thread_group_leader(current
)) {
613 pacct
->ac_exitcode
= exitcode
;
614 if (current
->flags
& PF_FORKNOEXEC
)
615 pacct
->ac_flag
|= AFORK
;
617 if (current
->flags
& PF_SUPERPRIV
)
618 pacct
->ac_flag
|= ASU
;
619 if (current
->flags
& PF_DUMPCORE
)
620 pacct
->ac_flag
|= ACORE
;
621 if (current
->flags
& PF_SIGNALED
)
622 pacct
->ac_flag
|= AXSIG
;
623 pacct
->ac_utime
= cputime_add(pacct
->ac_utime
, current
->utime
);
624 pacct
->ac_stime
= cputime_add(pacct
->ac_stime
, current
->stime
);
625 pacct
->ac_minflt
+= current
->min_flt
;
626 pacct
->ac_majflt
+= current
->maj_flt
;
627 spin_unlock_irq(¤t
->sighand
->siglock
);
631 * acct_process - now just a wrapper around do_acct_process
633 * handles process accounting for an exiting task
635 void acct_process(void)
637 struct file
*file
= NULL
;
638 struct pid_namespace
*ns
= task_active_pid_ns(current
);
639 struct bsd_acct_struct
*acct
;
643 * accelerate the common fastpath:
645 if (!acct
|| !acct
->file
)
648 spin_lock(&acct_lock
);
650 if (unlikely(!file
)) {
651 spin_unlock(&acct_lock
);
655 spin_unlock(&acct_lock
);
657 do_acct_process(acct
, ns
, file
);