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.
46 #include <linux/config.h>
48 #include <linux/slab.h>
49 #include <linux/acct.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 <asm/uaccess.h>
56 #include <asm/div64.h>
57 #include <linux/blkdev.h> /* sector_div */
60 * These constants control the amount of freespace that suspend and
61 * resume the process accounting system, and the time delay between
63 * Turned into sysctl-controllable parameters. AV, 12/11/98
66 int acct_parm
[3] = {4, 2, 30};
67 #define RESUME (acct_parm[0]) /* >foo% free space - resume */
68 #define SUSPEND (acct_parm[1]) /* <foo% free space - suspend */
69 #define ACCT_TIMEOUT (acct_parm[2]) /* foo second timeout between checks */
72 * External references and all of the globals.
74 static void do_acct_process(long, struct file
*);
77 * This structure is used so that all the data protected by lock
78 * can be placed in the same cache line as the lock. This primes
79 * the cache line to have the data after getting the lock.
84 volatile int needcheck
;
86 struct timer_list timer
;
89 static struct acct_glbs acct_globals __cacheline_aligned
= {SPIN_LOCK_UNLOCKED
};
92 * Called whenever the timer says to check the free space.
94 static void acct_timeout(unsigned long unused
)
96 acct_globals
.needcheck
= 1;
100 * Check the amount of free space and suspend/resume accordingly.
102 static int check_free_space(struct file
*file
)
110 spin_lock(&acct_globals
.lock
);
111 res
= acct_globals
.active
;
112 if (!file
|| !acct_globals
.needcheck
)
114 spin_unlock(&acct_globals
.lock
);
117 if (vfs_statfs(file
->f_dentry
->d_inode
->i_sb
, &sbuf
))
119 suspend
= sbuf
.f_blocks
* SUSPEND
;
120 resume
= sbuf
.f_blocks
* RESUME
;
122 sector_div(suspend
, 100);
123 sector_div(resume
, 100);
125 if (sbuf
.f_bavail
<= suspend
)
127 else if (sbuf
.f_bavail
>= resume
)
133 * If some joker switched acct_globals.file under us we'ld better be
134 * silent and _not_ touch anything.
136 spin_lock(&acct_globals
.lock
);
137 if (file
!= acct_globals
.file
) {
143 if (acct_globals
.active
) {
145 acct_globals
.active
= 0;
146 printk(KERN_INFO
"Process accounting paused\n");
150 acct_globals
.active
= 1;
151 printk(KERN_INFO
"Process accounting resumed\n");
155 del_timer(&acct_globals
.timer
);
156 acct_globals
.needcheck
= 0;
157 acct_globals
.timer
.expires
= jiffies
+ ACCT_TIMEOUT
*HZ
;
158 add_timer(&acct_globals
.timer
);
159 res
= acct_globals
.active
;
161 spin_unlock(&acct_globals
.lock
);
166 * Close the old accouting file (if currently open) and then replace
167 * it with file (if non-NULL).
169 * NOTE: acct_globals.lock MUST be held on entry and exit.
171 void acct_file_reopen(struct file
*file
)
173 struct file
*old_acct
= NULL
;
175 if (acct_globals
.file
) {
176 old_acct
= acct_globals
.file
;
177 del_timer(&acct_globals
.timer
);
178 acct_globals
.active
= 0;
179 acct_globals
.needcheck
= 0;
180 acct_globals
.file
= NULL
;
183 acct_globals
.file
= file
;
184 acct_globals
.needcheck
= 0;
185 acct_globals
.active
= 1;
186 /* It's been deleted if it was used before so this is safe */
187 init_timer(&acct_globals
.timer
);
188 acct_globals
.timer
.function
= acct_timeout
;
189 acct_globals
.timer
.expires
= jiffies
+ ACCT_TIMEOUT
*HZ
;
190 add_timer(&acct_globals
.timer
);
193 spin_unlock(&acct_globals
.lock
);
194 do_acct_process(0, old_acct
);
195 filp_close(old_acct
, NULL
);
196 spin_lock(&acct_globals
.lock
);
201 * sys_acct() is the only system call needed to implement process
202 * accounting. It takes the name of the file where accounting records
203 * should be written. If the filename is NULL, accounting will be
206 asmlinkage
long sys_acct(const char *name
)
208 struct file
*file
= NULL
;
212 if (!capable(CAP_SYS_PACCT
))
218 return (PTR_ERR(tmp
));
220 /* Difference from BSD - they don't do O_APPEND */
221 file
= filp_open(tmp
, O_WRONLY
|O_APPEND
, 0);
224 return (PTR_ERR(file
));
226 if (!S_ISREG(file
->f_dentry
->d_inode
->i_mode
)) {
227 filp_close(file
, NULL
);
231 if (!file
->f_op
->write
) {
232 filp_close(file
, NULL
);
237 error
= security_acct(file
);
241 spin_lock(&acct_globals
.lock
);
242 acct_file_reopen(file
);
243 spin_unlock(&acct_globals
.lock
);
249 * If the accouting is turned on for a file in the filesystem pointed
250 * to by sb, turn accouting off.
252 void acct_auto_close(struct super_block
*sb
)
254 spin_lock(&acct_globals
.lock
);
255 if (acct_globals
.file
&&
256 acct_globals
.file
->f_dentry
->d_inode
->i_sb
== sb
) {
257 acct_file_reopen((struct file
*)NULL
);
259 spin_unlock(&acct_globals
.lock
);
263 * encode an unsigned long into a comp_t
265 * This routine has been adopted from the encode_comp_t() function in
266 * the kern_acct.c file of the FreeBSD operating system. The encoding
267 * is a 13-bit fraction with a 3-bit (base 8) exponent.
270 #define MANTSIZE 13 /* 13 bit mantissa. */
271 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
272 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
274 static comp_t
encode_comp_t(unsigned long value
)
279 while (value
> MAXFRACT
) {
280 rnd
= value
& (1 << (EXPSIZE
- 1)); /* Round up? */
281 value
>>= EXPSIZE
; /* Base 8 exponent == 3 bit shift. */
286 * If we need to round up, do it (and handle overflow correctly).
288 if (rnd
&& (++value
> MAXFRACT
)) {
294 * Clean it up and polish it off.
296 exp
<<= MANTSIZE
; /* Shift the exponent into place */
297 exp
+= value
; /* and add on the mantissa. */
302 * Write an accounting entry for an exiting process
304 * The acct_process() call is the workhorse of the process
305 * accounting system. The struct acct is built here and then written
306 * into the accounting file. This function should only be called from
311 * do_acct_process does all actual work. Caller holds the reference to file.
313 static void do_acct_process(long exitcode
, struct file
*file
)
322 * First check to see if there is enough free_space to continue
323 * the process accounting system.
325 if (!check_free_space(file
))
329 * Fill the accounting struct with the needed info as recorded
330 * by the different kernel functions.
332 memset((caddr_t
)&ac
, 0, sizeof(struct acct
));
334 strlcpy(ac
.ac_comm
, current
->comm
, sizeof(ac
.ac_comm
));
336 elapsed
= get_jiffies_64() - current
->start_time
;
337 ac
.ac_etime
= encode_comp_t(elapsed
< (unsigned long) -1l ?
338 (unsigned long) elapsed
: (unsigned long) -1l);
340 ac
.ac_btime
= xtime
.tv_sec
- elapsed
;
341 ac
.ac_utime
= encode_comp_t(current
->utime
);
342 ac
.ac_stime
= encode_comp_t(current
->stime
);
343 ac
.ac_uid
= current
->uid
;
344 ac
.ac_gid
= current
->gid
;
345 ac
.ac_tty
= current
->tty
? current
->tty
->device
: 0;
348 if (current
->flags
& PF_FORKNOEXEC
)
350 if (current
->flags
& PF_SUPERPRIV
)
352 if (current
->flags
& PF_DUMPCORE
)
354 if (current
->flags
& PF_SIGNALED
)
359 struct vm_area_struct
*vma
;
360 down_read(¤t
->mm
->mmap_sem
);
361 vma
= current
->mm
->mmap
;
363 vsize
+= vma
->vm_end
- vma
->vm_start
;
366 up_read(¤t
->mm
->mmap_sem
);
368 vsize
= vsize
/ 1024;
369 ac
.ac_mem
= encode_comp_t(vsize
);
370 ac
.ac_io
= encode_comp_t(0 /* current->io_usage */); /* %% */
371 ac
.ac_rw
= encode_comp_t(ac
.ac_io
/ 1024);
372 ac
.ac_minflt
= encode_comp_t(current
->min_flt
);
373 ac
.ac_majflt
= encode_comp_t(current
->maj_flt
);
374 ac
.ac_swaps
= encode_comp_t(current
->nswap
);
375 ac
.ac_exitcode
= exitcode
;
378 * Kernel segment override to datasegment and write it
379 * to the accounting file.
384 * Accounting records are not subject to resource limits.
386 flim
= current
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
387 current
->rlim
[RLIMIT_FSIZE
].rlim_cur
= RLIM_INFINITY
;
388 file
->f_op
->write(file
, (char *)&ac
,
389 sizeof(struct acct
), &file
->f_pos
);
390 current
->rlim
[RLIMIT_FSIZE
].rlim_cur
= flim
;
395 * acct_process - now just a wrapper around do_acct_process
397 void acct_process(long exitcode
)
399 struct file
*file
= NULL
;
402 * accelerate the common fastpath:
404 if (!acct_globals
.file
)
407 spin_lock(&acct_globals
.lock
);
408 file
= acct_globals
.file
;
409 if (unlikely(!file
)) {
410 spin_unlock(&acct_globals
.lock
);
414 spin_unlock(&acct_globals
.lock
);
416 do_acct_process(exitcode
, file
);