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... TODO: move filp_open into open.c and make
40 * parameters sysctl-controllable. 10/11/98, AV.
43 #include <linux/config.h>
44 #include <linux/errno.h>
45 #include <linux/kernel.h>
47 #ifdef CONFIG_BSD_PROCESS_ACCT
49 #include <linux/acct.h>
50 #include <linux/smp_lock.h>
51 #include <linux/file.h>
53 #include <asm/uaccess.h>
56 * These constants control the amount of freespace that suspend and
57 * resume the process accounting system, and the time delay between
59 * Turned into sysctl-controllable parameters. AV, 12/11/98
62 int acct_parm
[3] = {4, 2, 30};
63 #define RESUME (acct_parm[0]) /* >foo% free space - resume */
64 #define SUSPEND (acct_parm[1]) /* <foo% free space - suspend */
65 #define ACCT_TIMEOUT (acct_parm[2]) /* foo second timeout between checks */
68 * External references and all of the globals.
70 void acct_timeout(unsigned long);
72 static volatile int acct_active
= 0;
73 static volatile int acct_needcheck
= 0;
74 static struct file
*acct_file
= NULL
;
75 static struct timer_list acct_timer
= { NULL
, NULL
, 0, 0, acct_timeout
};
76 static int do_acct_process(long, struct file
*);
79 * Called whenever the timer says to check the free space.
81 void acct_timeout(unsigned long unused
)
87 * Check the amount of free space and suspend/resume accordingly.
89 static int check_free_space(struct file
*file
)
93 struct super_block
*sb
;
94 int res
= acct_active
;
97 if (!file
|| !acct_needcheck
)
100 sb
= file
->f_dentry
->d_inode
->i_sb
;
101 if (!sb
->s_op
|| !sb
->s_op
->statfs
)
107 sb
->s_op
->statfs(sb
, &sbuf
, sizeof(struct statfs
));
110 if (sbuf
.f_bavail
<= SUSPEND
* sbuf
.f_blocks
/ 100)
112 else if (sbuf
.f_bavail
>= RESUME
* sbuf
.f_blocks
/ 100)
118 * If some joker switched acct_file under us we'ld better be
119 * silent and _not_ touch anything.
121 if (file
!= acct_file
)
122 return act
? (act
>0) : res
;
127 printk(KERN_INFO
"Process accounting paused\n");
132 printk(KERN_INFO
"Process accounting resumed\n");
136 del_timer(&acct_timer
);
138 acct_timer
.expires
= jiffies
+ ACCT_TIMEOUT
*HZ
;
139 add_timer(&acct_timer
);
144 * sys_acct() is the only system call needed to implement process
145 * accounting. It takes the name of the file where accounting records
146 * should be written. If the filename is NULL, accounting will be
149 asmlinkage
int sys_acct(const char *name
)
151 struct file
*file
= NULL
, *old_acct
= NULL
;
156 if (!capable(CAP_SYS_PACCT
))
161 error
= PTR_ERR(tmp
);
164 /* Difference from BSD - they don't do O_APPEND */
165 file
= filp_open(tmp
, O_WRONLY
|O_APPEND
, 0);
168 error
= PTR_ERR(file
);
172 if (!S_ISREG(file
->f_dentry
->d_inode
->i_mode
))
176 if (!file
->f_op
->write
)
182 old_acct
= acct_file
;
183 del_timer(&acct_timer
);
192 acct_timer
.expires
= jiffies
+ ACCT_TIMEOUT
*HZ
;
193 add_timer(&acct_timer
);
196 do_acct_process(0,old_acct
);
197 filp_close(old_acct
, NULL
);
203 filp_close(file
, NULL
);
207 void acct_auto_close(kdev_t dev
)
209 if (acct_file
&& acct_file
->f_dentry
->d_inode
->i_dev
== dev
)
210 sys_acct((char *)NULL
);
214 * encode an unsigned long into a comp_t
216 * This routine has been adopted from the encode_comp_t() function in
217 * the kern_acct.c file of the FreeBSD operating system. The encoding
218 * is a 13-bit fraction with a 3-bit (base 8) exponent.
221 #define MANTSIZE 13 /* 13 bit mantissa. */
222 #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
223 #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
225 static comp_t
encode_comp_t(unsigned long value
)
230 while (value
> MAXFRACT
) {
231 rnd
= value
& (1 << (EXPSIZE
- 1)); /* Round up? */
232 value
>>= EXPSIZE
; /* Base 8 exponent == 3 bit shift. */
237 * If we need to round up, do it (and handle overflow correctly).
239 if (rnd
&& (++value
> MAXFRACT
)) {
245 * Clean it up and polish it off.
247 exp
<<= MANTSIZE
; /* Shift the exponent into place */
248 exp
+= value
; /* and add on the mantissa. */
253 * Write an accounting entry for an exiting process
255 * The acct_process() call is the workhorse of the process
256 * accounting system. The struct acct is built here and then written
257 * into the accounting file. This function should only be called from
260 #define KSTK_EIP(stack) (((unsigned long *)(stack))[1019])
261 #define KSTK_ESP(stack) (((unsigned long *)(stack))[1022])
264 * do_acct_process does all actual work.
266 static int do_acct_process(long exitcode
, struct file
*file
)
274 * First check to see if there is enough free_space to continue
275 * the process accounting system.
280 if (!check_free_space(file
)) {
286 * Fill the accounting struct with the needed info as recorded
287 * by the different kernel functions.
289 memset((caddr_t
)&ac
, 0, sizeof(struct acct
));
291 strncpy(ac
.ac_comm
, current
->comm
, ACCT_COMM
);
292 ac
.ac_comm
[ACCT_COMM
- 1] = '\0';
294 ac
.ac_btime
= CT_TO_SECS(current
->start_time
) + (xtime
.tv_sec
- (jiffies
/ HZ
));
295 ac
.ac_etime
= encode_comp_t(jiffies
- current
->start_time
);
296 ac
.ac_utime
= encode_comp_t(current
->times
.tms_utime
);
297 ac
.ac_stime
= encode_comp_t(current
->times
.tms_stime
);
298 ac
.ac_uid
= current
->uid
;
299 ac
.ac_gid
= current
->gid
;
300 ac
.ac_tty
= (current
->tty
) ? kdev_t_to_nr(current
->tty
->device
) : 0;
303 if (current
->flags
& PF_FORKNOEXEC
)
305 if (current
->flags
& PF_SUPERPRIV
)
307 if (current
->flags
& PF_DUMPCORE
)
309 if (current
->flags
& PF_SIGNALED
)
314 struct vm_area_struct
*vma
= current
->mm
->mmap
;
316 vsize
+= vma
->vm_end
- vma
->vm_start
;
320 vsize
= vsize
/ 1024;
321 ac
.ac_mem
= encode_comp_t(vsize
);
322 ac
.ac_io
= encode_comp_t(0 /* current->io_usage */); /* %% */
323 ac
.ac_rw
= encode_comp_t(ac
.ac_io
/ 1024);
324 ac
.ac_minflt
= encode_comp_t(current
->min_flt
);
325 ac
.ac_majflt
= encode_comp_t(current
->maj_flt
);
326 ac
.ac_swaps
= encode_comp_t(current
->nswap
);
327 ac
.ac_exitcode
= exitcode
;
330 * Kernel segment override to datasegment and write it
331 * to the accounting file.
335 inode
= file
->f_dentry
->d_inode
;
337 file
->f_op
->write(file
, (char *)&ac
,
338 sizeof(struct acct
), &file
->f_pos
);
346 * acct_process - now just a wrapper around do_acct_process
348 int acct_process(long exitcode
)
350 return do_acct_process(exitcode
, acct_file
);
355 * Dummy system call when BSD process accounting is not configured
359 asmlinkage
int sys_acct(const char * filename
)