[PATCH] PM: Fix SMP races in the freezer
[linux-block.git] / kernel / acct.c
CommitLineData
1da177e4
LT
1/*
2 * linux/kernel/acct.c
3 *
4 * BSD Process Accounting for Linux
5 *
6 * Author: Marco van Wieringen <mvw@planets.elm.net>
7 *
8 * Some code based on ideas and code from:
9 * Thomas K. Dyas <tdyas@eden.rutgers.edu>
10 *
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.
16 *
17 * (C) Copyright 1995 - 1997 Marco van Wieringen - ELM Consultancy B.V.
18 *
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.
23 *
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.
26 *
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.
34 *
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.
40 *
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.
44 */
45
1da177e4
LT
46#include <linux/mm.h>
47#include <linux/slab.h>
48#include <linux/acct.h>
c59ede7b 49#include <linux/capability.h>
1da177e4
LT
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>
7b7b1ace 57#include <linux/mount.h>
1da177e4
LT
58#include <asm/uaccess.h>
59#include <asm/div64.h>
60#include <linux/blkdev.h> /* sector_div */
61
62/*
63 * These constants control the amount of freespace that suspend and
64 * resume the process accounting system, and the time delay between
65 * each check.
66 * Turned into sysctl-controllable parameters. AV, 12/11/98
67 */
68
69int acct_parm[3] = {4, 2, 30};
70#define RESUME (acct_parm[0]) /* >foo% free space - resume */
71#define SUSPEND (acct_parm[1]) /* <foo% free space - suspend */
72#define ACCT_TIMEOUT (acct_parm[2]) /* foo second timeout between checks */
73
74/*
75 * External references and all of the globals.
76 */
f6ec29a4 77static void do_acct_process(struct file *);
1da177e4
LT
78
79/*
80 * This structure is used so that all the data protected by lock
81 * can be placed in the same cache line as the lock. This primes
82 * the cache line to have the data after getting the lock.
83 */
84struct acct_glbs {
85 spinlock_t lock;
86 volatile int active;
87 volatile int needcheck;
88 struct file *file;
89 struct timer_list timer;
90};
91
6cfd76a2
PZ
92static struct acct_glbs acct_globals __cacheline_aligned =
93 {__SPIN_LOCK_UNLOCKED(acct_globals.lock)};
1da177e4
LT
94
95/*
96 * Called whenever the timer says to check the free space.
97 */
98static void acct_timeout(unsigned long unused)
99{
100 acct_globals.needcheck = 1;
101}
102
103/*
104 * Check the amount of free space and suspend/resume accordingly.
105 */
106static int check_free_space(struct file *file)
107{
108 struct kstatfs sbuf;
109 int res;
110 int act;
111 sector_t resume;
112 sector_t suspend;
113
114 spin_lock(&acct_globals.lock);
115 res = acct_globals.active;
116 if (!file || !acct_globals.needcheck)
117 goto out;
118 spin_unlock(&acct_globals.lock);
119
120 /* May block */
f3a43f3f 121 if (vfs_statfs(file->f_path.dentry, &sbuf))
1da177e4
LT
122 return res;
123 suspend = sbuf.f_blocks * SUSPEND;
124 resume = sbuf.f_blocks * RESUME;
125
126 sector_div(suspend, 100);
127 sector_div(resume, 100);
128
129 if (sbuf.f_bavail <= suspend)
130 act = -1;
131 else if (sbuf.f_bavail >= resume)
132 act = 1;
133 else
134 act = 0;
135
136 /*
137 * If some joker switched acct_globals.file under us we'ld better be
138 * silent and _not_ touch anything.
139 */
140 spin_lock(&acct_globals.lock);
141 if (file != acct_globals.file) {
142 if (act)
143 res = act>0;
144 goto out;
145 }
146
147 if (acct_globals.active) {
148 if (act < 0) {
149 acct_globals.active = 0;
150 printk(KERN_INFO "Process accounting paused\n");
151 }
152 } else {
153 if (act > 0) {
154 acct_globals.active = 1;
155 printk(KERN_INFO "Process accounting resumed\n");
156 }
157 }
158
159 del_timer(&acct_globals.timer);
160 acct_globals.needcheck = 0;
161 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
162 add_timer(&acct_globals.timer);
163 res = acct_globals.active;
164out:
165 spin_unlock(&acct_globals.lock);
166 return res;
167}
168
169/*
417ef531 170 * Close the old accounting file (if currently open) and then replace
1da177e4
LT
171 * it with file (if non-NULL).
172 *
173 * NOTE: acct_globals.lock MUST be held on entry and exit.
174 */
175static void acct_file_reopen(struct file *file)
176{
177 struct file *old_acct = NULL;
178
179 if (acct_globals.file) {
180 old_acct = acct_globals.file;
181 del_timer(&acct_globals.timer);
182 acct_globals.active = 0;
183 acct_globals.needcheck = 0;
184 acct_globals.file = NULL;
185 }
186 if (file) {
187 acct_globals.file = file;
188 acct_globals.needcheck = 0;
189 acct_globals.active = 1;
190 /* It's been deleted if it was used before so this is safe */
191 init_timer(&acct_globals.timer);
192 acct_globals.timer.function = acct_timeout;
193 acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ;
194 add_timer(&acct_globals.timer);
195 }
196 if (old_acct) {
f3a43f3f 197 mnt_unpin(old_acct->f_path.mnt);
1da177e4 198 spin_unlock(&acct_globals.lock);
f6ec29a4 199 do_acct_process(old_acct);
1da177e4
LT
200 filp_close(old_acct, NULL);
201 spin_lock(&acct_globals.lock);
202 }
203}
204
7b7b1ace
AV
205static int acct_on(char *name)
206{
207 struct file *file;
208 int error;
209
210 /* Difference from BSD - they don't do O_APPEND */
211 file = filp_open(name, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
212 if (IS_ERR(file))
213 return PTR_ERR(file);
214
f3a43f3f 215 if (!S_ISREG(file->f_path.dentry->d_inode->i_mode)) {
7b7b1ace
AV
216 filp_close(file, NULL);
217 return -EACCES;
218 }
219
220 if (!file->f_op->write) {
221 filp_close(file, NULL);
222 return -EIO;
223 }
224
225 error = security_acct(file);
226 if (error) {
227 filp_close(file, NULL);
228 return error;
229 }
230
231 spin_lock(&acct_globals.lock);
f3a43f3f 232 mnt_pin(file->f_path.mnt);
7b7b1ace
AV
233 acct_file_reopen(file);
234 spin_unlock(&acct_globals.lock);
235
f3a43f3f 236 mntput(file->f_path.mnt); /* it's pinned, now give up active reference */
7b7b1ace
AV
237
238 return 0;
239}
240
417ef531
RD
241/**
242 * sys_acct - enable/disable process accounting
243 * @name: file name for accounting records or NULL to shutdown accounting
244 *
245 * Returns 0 for success or negative errno values for failure.
246 *
247 * sys_acct() is the only system call needed to implement process
248 * accounting. It takes the name of the file where accounting records
249 * should be written. If the filename is NULL, accounting will be
250 * shutdown.
1da177e4
LT
251 */
252asmlinkage long sys_acct(const char __user *name)
253{
1da177e4
LT
254 int error;
255
256 if (!capable(CAP_SYS_PACCT))
257 return -EPERM;
258
259 if (name) {
7b7b1ace
AV
260 char *tmp = getname(name);
261 if (IS_ERR(tmp))
1da177e4 262 return (PTR_ERR(tmp));
7b7b1ace 263 error = acct_on(tmp);
1da177e4 264 putname(tmp);
7b7b1ace
AV
265 } else {
266 error = security_acct(NULL);
267 if (!error) {
268 spin_lock(&acct_globals.lock);
269 acct_file_reopen(NULL);
270 spin_unlock(&acct_globals.lock);
1da177e4
LT
271 }
272 }
7b7b1ace
AV
273 return error;
274}
1da177e4 275
7b7b1ace
AV
276/**
277 * acct_auto_close - turn off a filesystem's accounting if it is on
278 * @m: vfsmount being shut down
279 *
280 * If the accounting is turned on for a file in the subtree pointed to
281 * to by m, turn accounting off. Done when m is about to die.
282 */
283void acct_auto_close_mnt(struct vfsmount *m)
284{
1da177e4 285 spin_lock(&acct_globals.lock);
f3a43f3f 286 if (acct_globals.file && acct_globals.file->f_path.mnt == m)
7b7b1ace 287 acct_file_reopen(NULL);
1da177e4 288 spin_unlock(&acct_globals.lock);
1da177e4
LT
289}
290
417ef531
RD
291/**
292 * acct_auto_close - turn off a filesystem's accounting if it is on
293 * @sb: super block for the filesystem
294 *
295 * If the accounting is turned on for a file in the filesystem pointed
296 * to by sb, turn accounting off.
1da177e4
LT
297 */
298void acct_auto_close(struct super_block *sb)
299{
300 spin_lock(&acct_globals.lock);
301 if (acct_globals.file &&
f3a43f3f 302 acct_globals.file->f_path.mnt->mnt_sb == sb) {
7b7b1ace 303 acct_file_reopen(NULL);
1da177e4
LT
304 }
305 spin_unlock(&acct_globals.lock);
306}
307
308/*
309 * encode an unsigned long into a comp_t
310 *
311 * This routine has been adopted from the encode_comp_t() function in
312 * the kern_acct.c file of the FreeBSD operating system. The encoding
313 * is a 13-bit fraction with a 3-bit (base 8) exponent.
314 */
315
316#define MANTSIZE 13 /* 13 bit mantissa. */
317#define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
318#define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
319
320static comp_t encode_comp_t(unsigned long value)
321{
322 int exp, rnd;
323
324 exp = rnd = 0;
325 while (value > MAXFRACT) {
326 rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */
327 value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
328 exp++;
329 }
330
331 /*
332 * If we need to round up, do it (and handle overflow correctly).
333 */
334 if (rnd && (++value > MAXFRACT)) {
335 value >>= EXPSIZE;
336 exp++;
337 }
338
339 /*
340 * Clean it up and polish it off.
341 */
342 exp <<= MANTSIZE; /* Shift the exponent into place */
343 exp += value; /* and add on the mantissa. */
344 return exp;
345}
346
347#if ACCT_VERSION==1 || ACCT_VERSION==2
348/*
349 * encode an u64 into a comp2_t (24 bits)
350 *
351 * Format: 5 bit base 2 exponent, 20 bits mantissa.
352 * The leading bit of the mantissa is not stored, but implied for
353 * non-zero exponents.
354 * Largest encodable value is 50 bits.
355 */
356
357#define MANTSIZE2 20 /* 20 bit mantissa. */
358#define EXPSIZE2 5 /* 5 bit base 2 exponent. */
359#define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
360#define MAXEXP2 ((1 <<EXPSIZE2) - 1) /* Maximum exponent. */
361
362static comp2_t encode_comp2_t(u64 value)
363{
364 int exp, rnd;
365
366 exp = (value > (MAXFRACT2>>1));
367 rnd = 0;
368 while (value > MAXFRACT2) {
369 rnd = value & 1;
370 value >>= 1;
371 exp++;
372 }
373
374 /*
375 * If we need to round up, do it (and handle overflow correctly).
376 */
377 if (rnd && (++value > MAXFRACT2)) {
378 value >>= 1;
379 exp++;
380 }
381
382 if (exp > MAXEXP2) {
383 /* Overflow. Return largest representable number instead. */
384 return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
385 } else {
386 return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
387 }
388}
389#endif
390
391#if ACCT_VERSION==3
392/*
393 * encode an u64 into a 32 bit IEEE float
394 */
395static u32 encode_float(u64 value)
396{
397 unsigned exp = 190;
398 unsigned u;
399
400 if (value==0) return 0;
401 while ((s64)value > 0){
402 value <<= 1;
403 exp--;
404 }
405 u = (u32)(value >> 40) & 0x7fffffu;
406 return u | (exp << 23);
407}
408#endif
409
410/*
411 * Write an accounting entry for an exiting process
412 *
413 * The acct_process() call is the workhorse of the process
414 * accounting system. The struct acct is built here and then written
415 * into the accounting file. This function should only be called from
416 * do_exit().
417 */
418
419/*
420 * do_acct_process does all actual work. Caller holds the reference to file.
421 */
f6ec29a4 422static void do_acct_process(struct file *file)
1da177e4 423{
0e464814 424 struct pacct_struct *pacct = &current->signal->pacct;
1da177e4
LT
425 acct_t ac;
426 mm_segment_t fs;
1da177e4
LT
427 unsigned long flim;
428 u64 elapsed;
429 u64 run_time;
430 struct timespec uptime;
24ec839c 431 struct tty_struct *tty;
1da177e4
LT
432
433 /*
434 * First check to see if there is enough free_space to continue
435 * the process accounting system.
436 */
437 if (!check_free_space(file))
438 return;
439
440 /*
441 * Fill the accounting struct with the needed info as recorded
442 * by the different kernel functions.
443 */
444 memset((caddr_t)&ac, 0, sizeof(acct_t));
445
446 ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER;
447 strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm));
448
449 /* calculate run_time in nsec*/
450 do_posix_clock_monotonic_gettime(&uptime);
451 run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec;
bb231fe3
KK
452 run_time -= (u64)current->group_leader->start_time.tv_sec * NSEC_PER_SEC
453 + current->group_leader->start_time.tv_nsec;
1da177e4
LT
454 /* convert nsec -> AHZ */
455 elapsed = nsec_to_AHZ(run_time);
456#if ACCT_VERSION==3
457 ac.ac_etime = encode_float(elapsed);
458#else
459 ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
460 (unsigned long) elapsed : (unsigned long) -1l);
461#endif
462#if ACCT_VERSION==1 || ACCT_VERSION==2
463 {
464 /* new enlarged etime field */
465 comp2_t etime = encode_comp2_t(elapsed);
466 ac.ac_etime_hi = etime >> 16;
467 ac.ac_etime_lo = (u16) etime;
468 }
469#endif
470 do_div(elapsed, AHZ);
471 ac.ac_btime = xtime.tv_sec - elapsed;
1da177e4
LT
472 /* we really need to bite the bullet and change layout */
473 ac.ac_uid = current->uid;
474 ac.ac_gid = current->gid;
475#if ACCT_VERSION==2
476 ac.ac_ahz = AHZ;
477#endif
478#if ACCT_VERSION==1 || ACCT_VERSION==2
479 /* backward-compatible 16 bit fields */
480 ac.ac_uid16 = current->uid;
481 ac.ac_gid16 = current->gid;
482#endif
483#if ACCT_VERSION==3
484 ac.ac_pid = current->tgid;
485 ac.ac_ppid = current->parent->tgid;
486#endif
487
a4afee02 488 spin_lock_irq(&current->sighand->siglock);
7bcfa95e
ON
489 tty = current->signal->tty;
490 ac.ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
77787bfb
KK
491 ac.ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime)));
492 ac.ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime)));
f6ec29a4 493 ac.ac_flag = pacct->ac_flag;
0e464814 494 ac.ac_mem = encode_comp_t(pacct->ac_mem);
77787bfb
KK
495 ac.ac_minflt = encode_comp_t(pacct->ac_minflt);
496 ac.ac_majflt = encode_comp_t(pacct->ac_majflt);
f6ec29a4 497 ac.ac_exitcode = pacct->ac_exitcode;
a4afee02 498 spin_unlock_irq(&current->sighand->siglock);
1da177e4
LT
499 ac.ac_io = encode_comp_t(0 /* current->io_usage */); /* %% */
500 ac.ac_rw = encode_comp_t(ac.ac_io / 1024);
1da177e4 501 ac.ac_swaps = encode_comp_t(0);
1da177e4
LT
502
503 /*
504 * Kernel segment override to datasegment and write it
505 * to the accounting file.
506 */
507 fs = get_fs();
508 set_fs(KERNEL_DS);
509 /*
510 * Accounting records are not subject to resource limits.
511 */
512 flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
513 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
514 file->f_op->write(file, (char *)&ac,
515 sizeof(acct_t), &file->f_pos);
516 current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
517 set_fs(fs);
518}
519
0e464814
KK
520/**
521 * acct_init_pacct - initialize a new pacct_struct
1dbe83c3 522 * @pacct: per-process accounting info struct to initialize
0e464814
KK
523 */
524void acct_init_pacct(struct pacct_struct *pacct)
525{
526 memset(pacct, 0, sizeof(struct pacct_struct));
77787bfb 527 pacct->ac_utime = pacct->ac_stime = cputime_zero;
0e464814
KK
528}
529
530/**
531 * acct_collect - collect accounting information into pacct_struct
f6ec29a4
KK
532 * @exitcode: task exit code
533 * @group_dead: not 0, if this thread is the last one in the process.
0e464814 534 */
f6ec29a4 535void acct_collect(long exitcode, int group_dead)
0e464814
KK
536{
537 struct pacct_struct *pacct = &current->signal->pacct;
538 unsigned long vsize = 0;
539
f6ec29a4 540 if (group_dead && current->mm) {
0e464814
KK
541 struct vm_area_struct *vma;
542 down_read(&current->mm->mmap_sem);
543 vma = current->mm->mmap;
544 while (vma) {
545 vsize += vma->vm_end - vma->vm_start;
546 vma = vma->vm_next;
547 }
548 up_read(&current->mm->mmap_sem);
549 }
550
77787bfb 551 spin_lock_irq(&current->sighand->siglock);
f6ec29a4
KK
552 if (group_dead)
553 pacct->ac_mem = vsize / 1024;
554 if (thread_group_leader(current)) {
555 pacct->ac_exitcode = exitcode;
556 if (current->flags & PF_FORKNOEXEC)
557 pacct->ac_flag |= AFORK;
558 }
559 if (current->flags & PF_SUPERPRIV)
560 pacct->ac_flag |= ASU;
561 if (current->flags & PF_DUMPCORE)
562 pacct->ac_flag |= ACORE;
563 if (current->flags & PF_SIGNALED)
564 pacct->ac_flag |= AXSIG;
77787bfb
KK
565 pacct->ac_utime = cputime_add(pacct->ac_utime, current->utime);
566 pacct->ac_stime = cputime_add(pacct->ac_stime, current->stime);
567 pacct->ac_minflt += current->min_flt;
568 pacct->ac_majflt += current->maj_flt;
569 spin_unlock_irq(&current->sighand->siglock);
0e464814
KK
570}
571
417ef531 572/**
1da177e4 573 * acct_process - now just a wrapper around do_acct_process
417ef531
RD
574 * @exitcode: task exit code
575 *
576 * handles process accounting for an exiting task
1da177e4 577 */
7f32a25f 578void acct_process(void)
1da177e4
LT
579{
580 struct file *file = NULL;
581
582 /*
583 * accelerate the common fastpath:
584 */
585 if (!acct_globals.file)
586 return;
587
588 spin_lock(&acct_globals.lock);
589 file = acct_globals.file;
590 if (unlikely(!file)) {
591 spin_unlock(&acct_globals.lock);
592 return;
593 }
594 get_file(file);
595 spin_unlock(&acct_globals.lock);
596
f6ec29a4 597 do_acct_process(file);
1da177e4
LT
598 fput(file);
599}