]>
Commit | Line | Data |
---|---|---|
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 | ||
46 | #include <linux/config.h> | |
47 | #include <linux/mm.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 <linux/times.h> | |
56 | #include <linux/syscalls.h> | |
57 | #include <asm/uaccess.h> | |
58 | #include <asm/div64.h> | |
59 | #include <linux/blkdev.h> /* sector_div */ | |
60 | ||
61 | /* | |
62 | * These constants control the amount of freespace that suspend and | |
63 | * resume the process accounting system, and the time delay between | |
64 | * each check. | |
65 | * Turned into sysctl-controllable parameters. AV, 12/11/98 | |
66 | */ | |
67 | ||
68 | int acct_parm[3] = {4, 2, 30}; | |
69 | #define RESUME (acct_parm[0]) /* >foo% free space - resume */ | |
70 | #define SUSPEND (acct_parm[1]) /* <foo% free space - suspend */ | |
71 | #define ACCT_TIMEOUT (acct_parm[2]) /* foo second timeout between checks */ | |
72 | ||
73 | /* | |
74 | * External references and all of the globals. | |
75 | */ | |
76 | static void do_acct_process(long, struct file *); | |
77 | ||
78 | /* | |
79 | * This structure is used so that all the data protected by lock | |
80 | * can be placed in the same cache line as the lock. This primes | |
81 | * the cache line to have the data after getting the lock. | |
82 | */ | |
83 | struct acct_glbs { | |
84 | spinlock_t lock; | |
85 | volatile int active; | |
86 | volatile int needcheck; | |
87 | struct file *file; | |
88 | struct timer_list timer; | |
89 | }; | |
90 | ||
91 | static struct acct_glbs acct_globals __cacheline_aligned = {SPIN_LOCK_UNLOCKED}; | |
92 | ||
93 | /* | |
94 | * Called whenever the timer says to check the free space. | |
95 | */ | |
96 | static void acct_timeout(unsigned long unused) | |
97 | { | |
98 | acct_globals.needcheck = 1; | |
99 | } | |
100 | ||
101 | /* | |
102 | * Check the amount of free space and suspend/resume accordingly. | |
103 | */ | |
104 | static int check_free_space(struct file *file) | |
105 | { | |
106 | struct kstatfs sbuf; | |
107 | int res; | |
108 | int act; | |
109 | sector_t resume; | |
110 | sector_t suspend; | |
111 | ||
112 | spin_lock(&acct_globals.lock); | |
113 | res = acct_globals.active; | |
114 | if (!file || !acct_globals.needcheck) | |
115 | goto out; | |
116 | spin_unlock(&acct_globals.lock); | |
117 | ||
118 | /* May block */ | |
119 | if (vfs_statfs(file->f_dentry->d_inode->i_sb, &sbuf)) | |
120 | return res; | |
121 | suspend = sbuf.f_blocks * SUSPEND; | |
122 | resume = sbuf.f_blocks * RESUME; | |
123 | ||
124 | sector_div(suspend, 100); | |
125 | sector_div(resume, 100); | |
126 | ||
127 | if (sbuf.f_bavail <= suspend) | |
128 | act = -1; | |
129 | else if (sbuf.f_bavail >= resume) | |
130 | act = 1; | |
131 | else | |
132 | act = 0; | |
133 | ||
134 | /* | |
135 | * If some joker switched acct_globals.file under us we'ld better be | |
136 | * silent and _not_ touch anything. | |
137 | */ | |
138 | spin_lock(&acct_globals.lock); | |
139 | if (file != acct_globals.file) { | |
140 | if (act) | |
141 | res = act>0; | |
142 | goto out; | |
143 | } | |
144 | ||
145 | if (acct_globals.active) { | |
146 | if (act < 0) { | |
147 | acct_globals.active = 0; | |
148 | printk(KERN_INFO "Process accounting paused\n"); | |
149 | } | |
150 | } else { | |
151 | if (act > 0) { | |
152 | acct_globals.active = 1; | |
153 | printk(KERN_INFO "Process accounting resumed\n"); | |
154 | } | |
155 | } | |
156 | ||
157 | del_timer(&acct_globals.timer); | |
158 | acct_globals.needcheck = 0; | |
159 | acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ; | |
160 | add_timer(&acct_globals.timer); | |
161 | res = acct_globals.active; | |
162 | out: | |
163 | spin_unlock(&acct_globals.lock); | |
164 | return res; | |
165 | } | |
166 | ||
167 | /* | |
168 | * Close the old accouting file (if currently open) and then replace | |
169 | * it with file (if non-NULL). | |
170 | * | |
171 | * NOTE: acct_globals.lock MUST be held on entry and exit. | |
172 | */ | |
173 | static void acct_file_reopen(struct file *file) | |
174 | { | |
175 | struct file *old_acct = NULL; | |
176 | ||
177 | if (acct_globals.file) { | |
178 | old_acct = acct_globals.file; | |
179 | del_timer(&acct_globals.timer); | |
180 | acct_globals.active = 0; | |
181 | acct_globals.needcheck = 0; | |
182 | acct_globals.file = NULL; | |
183 | } | |
184 | if (file) { | |
185 | acct_globals.file = file; | |
186 | acct_globals.needcheck = 0; | |
187 | acct_globals.active = 1; | |
188 | /* It's been deleted if it was used before so this is safe */ | |
189 | init_timer(&acct_globals.timer); | |
190 | acct_globals.timer.function = acct_timeout; | |
191 | acct_globals.timer.expires = jiffies + ACCT_TIMEOUT*HZ; | |
192 | add_timer(&acct_globals.timer); | |
193 | } | |
194 | if (old_acct) { | |
195 | spin_unlock(&acct_globals.lock); | |
196 | do_acct_process(0, old_acct); | |
197 | filp_close(old_acct, NULL); | |
198 | spin_lock(&acct_globals.lock); | |
199 | } | |
200 | } | |
201 | ||
202 | /* | |
203 | * sys_acct() is the only system call needed to implement process | |
204 | * accounting. It takes the name of the file where accounting records | |
205 | * should be written. If the filename is NULL, accounting will be | |
206 | * shutdown. | |
207 | */ | |
208 | asmlinkage long sys_acct(const char __user *name) | |
209 | { | |
210 | struct file *file = NULL; | |
211 | char *tmp; | |
212 | int error; | |
213 | ||
214 | if (!capable(CAP_SYS_PACCT)) | |
215 | return -EPERM; | |
216 | ||
217 | if (name) { | |
218 | tmp = getname(name); | |
219 | if (IS_ERR(tmp)) { | |
220 | return (PTR_ERR(tmp)); | |
221 | } | |
222 | /* Difference from BSD - they don't do O_APPEND */ | |
223 | file = filp_open(tmp, O_WRONLY|O_APPEND, 0); | |
224 | putname(tmp); | |
225 | if (IS_ERR(file)) { | |
226 | return (PTR_ERR(file)); | |
227 | } | |
228 | if (!S_ISREG(file->f_dentry->d_inode->i_mode)) { | |
229 | filp_close(file, NULL); | |
230 | return (-EACCES); | |
231 | } | |
232 | ||
233 | if (!file->f_op->write) { | |
234 | filp_close(file, NULL); | |
235 | return (-EIO); | |
236 | } | |
237 | } | |
238 | ||
239 | error = security_acct(file); | |
240 | if (error) { | |
241 | if (file) | |
242 | filp_close(file, NULL); | |
243 | return error; | |
244 | } | |
245 | ||
246 | spin_lock(&acct_globals.lock); | |
247 | acct_file_reopen(file); | |
248 | spin_unlock(&acct_globals.lock); | |
249 | ||
250 | return (0); | |
251 | } | |
252 | ||
253 | /* | |
254 | * If the accouting is turned on for a file in the filesystem pointed | |
255 | * to by sb, turn accouting off. | |
256 | */ | |
257 | void acct_auto_close(struct super_block *sb) | |
258 | { | |
259 | spin_lock(&acct_globals.lock); | |
260 | if (acct_globals.file && | |
261 | acct_globals.file->f_dentry->d_inode->i_sb == sb) { | |
262 | acct_file_reopen((struct file *)NULL); | |
263 | } | |
264 | spin_unlock(&acct_globals.lock); | |
265 | } | |
266 | ||
267 | /* | |
268 | * encode an unsigned long into a comp_t | |
269 | * | |
270 | * This routine has been adopted from the encode_comp_t() function in | |
271 | * the kern_acct.c file of the FreeBSD operating system. The encoding | |
272 | * is a 13-bit fraction with a 3-bit (base 8) exponent. | |
273 | */ | |
274 | ||
275 | #define MANTSIZE 13 /* 13 bit mantissa. */ | |
276 | #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ | |
277 | #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ | |
278 | ||
279 | static comp_t encode_comp_t(unsigned long value) | |
280 | { | |
281 | int exp, rnd; | |
282 | ||
283 | exp = rnd = 0; | |
284 | while (value > MAXFRACT) { | |
285 | rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */ | |
286 | value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ | |
287 | exp++; | |
288 | } | |
289 | ||
290 | /* | |
291 | * If we need to round up, do it (and handle overflow correctly). | |
292 | */ | |
293 | if (rnd && (++value > MAXFRACT)) { | |
294 | value >>= EXPSIZE; | |
295 | exp++; | |
296 | } | |
297 | ||
298 | /* | |
299 | * Clean it up and polish it off. | |
300 | */ | |
301 | exp <<= MANTSIZE; /* Shift the exponent into place */ | |
302 | exp += value; /* and add on the mantissa. */ | |
303 | return exp; | |
304 | } | |
305 | ||
306 | #if ACCT_VERSION==1 || ACCT_VERSION==2 | |
307 | /* | |
308 | * encode an u64 into a comp2_t (24 bits) | |
309 | * | |
310 | * Format: 5 bit base 2 exponent, 20 bits mantissa. | |
311 | * The leading bit of the mantissa is not stored, but implied for | |
312 | * non-zero exponents. | |
313 | * Largest encodable value is 50 bits. | |
314 | */ | |
315 | ||
316 | #define MANTSIZE2 20 /* 20 bit mantissa. */ | |
317 | #define EXPSIZE2 5 /* 5 bit base 2 exponent. */ | |
318 | #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */ | |
319 | #define MAXEXP2 ((1 <<EXPSIZE2) - 1) /* Maximum exponent. */ | |
320 | ||
321 | static comp2_t encode_comp2_t(u64 value) | |
322 | { | |
323 | int exp, rnd; | |
324 | ||
325 | exp = (value > (MAXFRACT2>>1)); | |
326 | rnd = 0; | |
327 | while (value > MAXFRACT2) { | |
328 | rnd = value & 1; | |
329 | value >>= 1; | |
330 | exp++; | |
331 | } | |
332 | ||
333 | /* | |
334 | * If we need to round up, do it (and handle overflow correctly). | |
335 | */ | |
336 | if (rnd && (++value > MAXFRACT2)) { | |
337 | value >>= 1; | |
338 | exp++; | |
339 | } | |
340 | ||
341 | if (exp > MAXEXP2) { | |
342 | /* Overflow. Return largest representable number instead. */ | |
343 | return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1; | |
344 | } else { | |
345 | return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1)); | |
346 | } | |
347 | } | |
348 | #endif | |
349 | ||
350 | #if ACCT_VERSION==3 | |
351 | /* | |
352 | * encode an u64 into a 32 bit IEEE float | |
353 | */ | |
354 | static u32 encode_float(u64 value) | |
355 | { | |
356 | unsigned exp = 190; | |
357 | unsigned u; | |
358 | ||
359 | if (value==0) return 0; | |
360 | while ((s64)value > 0){ | |
361 | value <<= 1; | |
362 | exp--; | |
363 | } | |
364 | u = (u32)(value >> 40) & 0x7fffffu; | |
365 | return u | (exp << 23); | |
366 | } | |
367 | #endif | |
368 | ||
369 | /* | |
370 | * Write an accounting entry for an exiting process | |
371 | * | |
372 | * The acct_process() call is the workhorse of the process | |
373 | * accounting system. The struct acct is built here and then written | |
374 | * into the accounting file. This function should only be called from | |
375 | * do_exit(). | |
376 | */ | |
377 | ||
378 | /* | |
379 | * do_acct_process does all actual work. Caller holds the reference to file. | |
380 | */ | |
381 | static void do_acct_process(long exitcode, struct file *file) | |
382 | { | |
383 | acct_t ac; | |
384 | mm_segment_t fs; | |
385 | unsigned long vsize; | |
386 | unsigned long flim; | |
387 | u64 elapsed; | |
388 | u64 run_time; | |
389 | struct timespec uptime; | |
390 | ||
391 | /* | |
392 | * First check to see if there is enough free_space to continue | |
393 | * the process accounting system. | |
394 | */ | |
395 | if (!check_free_space(file)) | |
396 | return; | |
397 | ||
398 | /* | |
399 | * Fill the accounting struct with the needed info as recorded | |
400 | * by the different kernel functions. | |
401 | */ | |
402 | memset((caddr_t)&ac, 0, sizeof(acct_t)); | |
403 | ||
404 | ac.ac_version = ACCT_VERSION | ACCT_BYTEORDER; | |
405 | strlcpy(ac.ac_comm, current->comm, sizeof(ac.ac_comm)); | |
406 | ||
407 | /* calculate run_time in nsec*/ | |
408 | do_posix_clock_monotonic_gettime(&uptime); | |
409 | run_time = (u64)uptime.tv_sec*NSEC_PER_SEC + uptime.tv_nsec; | |
410 | run_time -= (u64)current->start_time.tv_sec*NSEC_PER_SEC | |
411 | + current->start_time.tv_nsec; | |
412 | /* convert nsec -> AHZ */ | |
413 | elapsed = nsec_to_AHZ(run_time); | |
414 | #if ACCT_VERSION==3 | |
415 | ac.ac_etime = encode_float(elapsed); | |
416 | #else | |
417 | ac.ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ? | |
418 | (unsigned long) elapsed : (unsigned long) -1l); | |
419 | #endif | |
420 | #if ACCT_VERSION==1 || ACCT_VERSION==2 | |
421 | { | |
422 | /* new enlarged etime field */ | |
423 | comp2_t etime = encode_comp2_t(elapsed); | |
424 | ac.ac_etime_hi = etime >> 16; | |
425 | ac.ac_etime_lo = (u16) etime; | |
426 | } | |
427 | #endif | |
428 | do_div(elapsed, AHZ); | |
429 | ac.ac_btime = xtime.tv_sec - elapsed; | |
430 | ac.ac_utime = encode_comp_t(jiffies_to_AHZ( | |
431 | current->signal->utime + | |
432 | current->group_leader->utime)); | |
433 | ac.ac_stime = encode_comp_t(jiffies_to_AHZ( | |
434 | current->signal->stime + | |
435 | current->group_leader->stime)); | |
436 | /* we really need to bite the bullet and change layout */ | |
437 | ac.ac_uid = current->uid; | |
438 | ac.ac_gid = current->gid; | |
439 | #if ACCT_VERSION==2 | |
440 | ac.ac_ahz = AHZ; | |
441 | #endif | |
442 | #if ACCT_VERSION==1 || ACCT_VERSION==2 | |
443 | /* backward-compatible 16 bit fields */ | |
444 | ac.ac_uid16 = current->uid; | |
445 | ac.ac_gid16 = current->gid; | |
446 | #endif | |
447 | #if ACCT_VERSION==3 | |
448 | ac.ac_pid = current->tgid; | |
449 | ac.ac_ppid = current->parent->tgid; | |
450 | #endif | |
451 | ||
452 | read_lock(&tasklist_lock); /* pin current->signal */ | |
453 | ac.ac_tty = current->signal->tty ? | |
454 | old_encode_dev(tty_devnum(current->signal->tty)) : 0; | |
455 | read_unlock(&tasklist_lock); | |
456 | ||
457 | ac.ac_flag = 0; | |
458 | if (current->flags & PF_FORKNOEXEC) | |
459 | ac.ac_flag |= AFORK; | |
460 | if (current->flags & PF_SUPERPRIV) | |
461 | ac.ac_flag |= ASU; | |
462 | if (current->flags & PF_DUMPCORE) | |
463 | ac.ac_flag |= ACORE; | |
464 | if (current->flags & PF_SIGNALED) | |
465 | ac.ac_flag |= AXSIG; | |
466 | ||
467 | vsize = 0; | |
468 | if (current->mm) { | |
469 | struct vm_area_struct *vma; | |
470 | down_read(¤t->mm->mmap_sem); | |
471 | vma = current->mm->mmap; | |
472 | while (vma) { | |
473 | vsize += vma->vm_end - vma->vm_start; | |
474 | vma = vma->vm_next; | |
475 | } | |
476 | up_read(¤t->mm->mmap_sem); | |
477 | } | |
478 | vsize = vsize / 1024; | |
479 | ac.ac_mem = encode_comp_t(vsize); | |
480 | ac.ac_io = encode_comp_t(0 /* current->io_usage */); /* %% */ | |
481 | ac.ac_rw = encode_comp_t(ac.ac_io / 1024); | |
482 | ac.ac_minflt = encode_comp_t(current->signal->min_flt + | |
483 | current->group_leader->min_flt); | |
484 | ac.ac_majflt = encode_comp_t(current->signal->maj_flt + | |
485 | current->group_leader->maj_flt); | |
486 | ac.ac_swaps = encode_comp_t(0); | |
487 | ac.ac_exitcode = exitcode; | |
488 | ||
489 | /* | |
490 | * Kernel segment override to datasegment and write it | |
491 | * to the accounting file. | |
492 | */ | |
493 | fs = get_fs(); | |
494 | set_fs(KERNEL_DS); | |
495 | /* | |
496 | * Accounting records are not subject to resource limits. | |
497 | */ | |
498 | flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
499 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
500 | file->f_op->write(file, (char *)&ac, | |
501 | sizeof(acct_t), &file->f_pos); | |
502 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim; | |
503 | set_fs(fs); | |
504 | } | |
505 | ||
506 | /* | |
507 | * acct_process - now just a wrapper around do_acct_process | |
508 | */ | |
509 | void acct_process(long exitcode) | |
510 | { | |
511 | struct file *file = NULL; | |
512 | ||
513 | /* | |
514 | * accelerate the common fastpath: | |
515 | */ | |
516 | if (!acct_globals.file) | |
517 | return; | |
518 | ||
519 | spin_lock(&acct_globals.lock); | |
520 | file = acct_globals.file; | |
521 | if (unlikely(!file)) { | |
522 | spin_unlock(&acct_globals.lock); | |
523 | return; | |
524 | } | |
525 | get_file(file); | |
526 | spin_unlock(&acct_globals.lock); | |
527 | ||
528 | do_acct_process(exitcode, file); | |
529 | fput(file); | |
530 | } | |
531 | ||
532 | ||
533 | /* | |
534 | * acct_update_integrals | |
535 | * - update mm integral fields in task_struct | |
536 | */ | |
537 | void acct_update_integrals(struct task_struct *tsk) | |
538 | { | |
539 | if (likely(tsk->mm)) { | |
540 | long delta = tsk->stime - tsk->acct_stimexpd; | |
541 | ||
542 | if (delta == 0) | |
543 | return; | |
544 | tsk->acct_stimexpd = tsk->stime; | |
545 | tsk->acct_rss_mem1 += delta * get_mm_counter(tsk->mm, rss); | |
546 | tsk->acct_vm_mem1 += delta * tsk->mm->total_vm; | |
547 | } | |
548 | } | |
549 | ||
550 | /* | |
551 | * acct_clear_integrals | |
552 | * - clear the mm integral fields in task_struct | |
553 | */ | |
554 | void acct_clear_integrals(struct task_struct *tsk) | |
555 | { | |
556 | if (tsk) { | |
557 | tsk->acct_stimexpd = 0; | |
558 | tsk->acct_rss_mem1 = 0; | |
559 | tsk->acct_vm_mem1 = 0; | |
560 | } | |
561 | } |