]>
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 | ||
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> |
7153e402 | 58 | #include <linux/uaccess.h> |
1da177e4 LT |
59 | #include <asm/div64.h> |
60 | #include <linux/blkdev.h> /* sector_div */ | |
5f7b703f | 61 | #include <linux/pid_namespace.h> |
1da177e4 LT |
62 | |
63 | /* | |
64 | * These constants control the amount of freespace that suspend and | |
65 | * resume the process accounting system, and the time delay between | |
66 | * each check. | |
67 | * Turned into sysctl-controllable parameters. AV, 12/11/98 | |
68 | */ | |
69 | ||
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 */ | |
74 | ||
75 | /* | |
76 | * External references and all of the globals. | |
77 | */ | |
6248b1b3 PE |
78 | static void do_acct_process(struct bsd_acct_struct *acct, |
79 | struct pid_namespace *ns, struct file *); | |
1da177e4 LT |
80 | |
81 | /* | |
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. | |
85 | */ | |
081e4c8a | 86 | struct bsd_acct_struct { |
32dc7308 AV |
87 | int active; |
88 | unsigned long needcheck; | |
1da177e4 | 89 | struct file *file; |
5f7b703f | 90 | struct pid_namespace *ns; |
b5a71748 | 91 | struct list_head list; |
1da177e4 LT |
92 | }; |
93 | ||
a75d9797 | 94 | static DEFINE_SPINLOCK(acct_lock); |
b5a71748 | 95 | static LIST_HEAD(acct_list); |
a75d9797 | 96 | |
1da177e4 LT |
97 | /* |
98 | * Check the amount of free space and suspend/resume accordingly. | |
99 | */ | |
6248b1b3 | 100 | static int check_free_space(struct bsd_acct_struct *acct, struct file *file) |
1da177e4 LT |
101 | { |
102 | struct kstatfs sbuf; | |
103 | int res; | |
104 | int act; | |
32dc7308 AV |
105 | u64 resume; |
106 | u64 suspend; | |
1da177e4 | 107 | |
a75d9797 | 108 | spin_lock(&acct_lock); |
6248b1b3 | 109 | res = acct->active; |
32dc7308 | 110 | if (!file || time_is_before_jiffies(acct->needcheck)) |
1da177e4 | 111 | goto out; |
a75d9797 | 112 | spin_unlock(&acct_lock); |
1da177e4 LT |
113 | |
114 | /* May block */ | |
ebabe9a9 | 115 | if (vfs_statfs(&file->f_path, &sbuf)) |
1da177e4 LT |
116 | return res; |
117 | suspend = sbuf.f_blocks * SUSPEND; | |
118 | resume = sbuf.f_blocks * RESUME; | |
119 | ||
32dc7308 AV |
120 | do_div(suspend, 100); |
121 | do_div(resume, 100); | |
1da177e4 LT |
122 | |
123 | if (sbuf.f_bavail <= suspend) | |
124 | act = -1; | |
125 | else if (sbuf.f_bavail >= resume) | |
126 | act = 1; | |
127 | else | |
128 | act = 0; | |
129 | ||
130 | /* | |
6248b1b3 | 131 | * If some joker switched acct->file under us we'ld better be |
1da177e4 LT |
132 | * silent and _not_ touch anything. |
133 | */ | |
a75d9797 | 134 | spin_lock(&acct_lock); |
6248b1b3 | 135 | if (file != acct->file) { |
1da177e4 | 136 | if (act) |
46c0a8ca | 137 | res = act > 0; |
1da177e4 LT |
138 | goto out; |
139 | } | |
140 | ||
6248b1b3 | 141 | if (acct->active) { |
1da177e4 | 142 | if (act < 0) { |
6248b1b3 | 143 | acct->active = 0; |
1da177e4 LT |
144 | printk(KERN_INFO "Process accounting paused\n"); |
145 | } | |
146 | } else { | |
147 | if (act > 0) { | |
6248b1b3 | 148 | acct->active = 1; |
1da177e4 LT |
149 | printk(KERN_INFO "Process accounting resumed\n"); |
150 | } | |
151 | } | |
152 | ||
32dc7308 | 153 | acct->needcheck = jiffies + ACCT_TIMEOUT*HZ; |
6248b1b3 | 154 | res = acct->active; |
1da177e4 | 155 | out: |
a75d9797 | 156 | spin_unlock(&acct_lock); |
1da177e4 LT |
157 | return res; |
158 | } | |
159 | ||
160 | /* | |
417ef531 | 161 | * Close the old accounting file (if currently open) and then replace |
1da177e4 LT |
162 | * it with file (if non-NULL). |
163 | * | |
a75d9797 | 164 | * NOTE: acct_lock MUST be held on entry and exit. |
1da177e4 | 165 | */ |
0b6b030f PE |
166 | static void acct_file_reopen(struct bsd_acct_struct *acct, struct file *file, |
167 | struct pid_namespace *ns) | |
1da177e4 LT |
168 | { |
169 | struct file *old_acct = NULL; | |
5f7b703f | 170 | struct pid_namespace *old_ns = NULL; |
1da177e4 | 171 | |
6248b1b3 PE |
172 | if (acct->file) { |
173 | old_acct = acct->file; | |
174 | old_ns = acct->ns; | |
6248b1b3 | 175 | acct->active = 0; |
6248b1b3 | 176 | acct->file = NULL; |
0b6b030f | 177 | acct->ns = NULL; |
b5a71748 | 178 | list_del(&acct->list); |
1da177e4 LT |
179 | } |
180 | if (file) { | |
6248b1b3 | 181 | acct->file = file; |
0b6b030f | 182 | acct->ns = ns; |
795a2f22 AV |
183 | acct->needcheck = jiffies; |
184 | acct->active = 0; | |
b5a71748 | 185 | list_add(&acct->list, &acct_list); |
1da177e4 LT |
186 | } |
187 | if (old_acct) { | |
f3a43f3f | 188 | mnt_unpin(old_acct->f_path.mnt); |
a75d9797 | 189 | spin_unlock(&acct_lock); |
6248b1b3 | 190 | do_acct_process(acct, old_ns, old_acct); |
1da177e4 | 191 | filp_close(old_acct, NULL); |
a75d9797 | 192 | spin_lock(&acct_lock); |
1da177e4 LT |
193 | } |
194 | } | |
195 | ||
669abf4e | 196 | static int acct_on(struct filename *pathname) |
7b7b1ace AV |
197 | { |
198 | struct file *file; | |
df279ca8 | 199 | struct vfsmount *mnt; |
0b6b030f PE |
200 | struct pid_namespace *ns; |
201 | struct bsd_acct_struct *acct = NULL; | |
7b7b1ace AV |
202 | |
203 | /* Difference from BSD - they don't do O_APPEND */ | |
669abf4e | 204 | file = file_open_name(pathname, O_WRONLY|O_APPEND|O_LARGEFILE, 0); |
7b7b1ace AV |
205 | if (IS_ERR(file)) |
206 | return PTR_ERR(file); | |
207 | ||
496ad9aa | 208 | if (!S_ISREG(file_inode(file)->i_mode)) { |
7b7b1ace AV |
209 | filp_close(file, NULL); |
210 | return -EACCES; | |
211 | } | |
212 | ||
213 | if (!file->f_op->write) { | |
214 | filp_close(file, NULL); | |
215 | return -EIO; | |
216 | } | |
217 | ||
0b6b030f PE |
218 | ns = task_active_pid_ns(current); |
219 | if (ns->bacct == NULL) { | |
220 | acct = kzalloc(sizeof(struct bsd_acct_struct), GFP_KERNEL); | |
221 | if (acct == NULL) { | |
222 | filp_close(file, NULL); | |
223 | return -ENOMEM; | |
224 | } | |
225 | } | |
226 | ||
a75d9797 | 227 | spin_lock(&acct_lock); |
0b6b030f PE |
228 | if (ns->bacct == NULL) { |
229 | ns->bacct = acct; | |
230 | acct = NULL; | |
231 | } | |
232 | ||
df279ca8 RL |
233 | mnt = file->f_path.mnt; |
234 | mnt_pin(mnt); | |
0b6b030f | 235 | acct_file_reopen(ns->bacct, file, ns); |
a75d9797 | 236 | spin_unlock(&acct_lock); |
7b7b1ace | 237 | |
df279ca8 | 238 | mntput(mnt); /* it's pinned, now give up active reference */ |
0b6b030f | 239 | kfree(acct); |
7b7b1ace AV |
240 | |
241 | return 0; | |
242 | } | |
243 | ||
9df7fa16 AV |
244 | static DEFINE_MUTEX(acct_on_mutex); |
245 | ||
417ef531 RD |
246 | /** |
247 | * sys_acct - enable/disable process accounting | |
248 | * @name: file name for accounting records or NULL to shutdown accounting | |
249 | * | |
250 | * Returns 0 for success or negative errno values for failure. | |
251 | * | |
252 | * sys_acct() is the only system call needed to implement process | |
253 | * accounting. It takes the name of the file where accounting records | |
254 | * should be written. If the filename is NULL, accounting will be | |
255 | * shutdown. | |
1da177e4 | 256 | */ |
b290ebe2 | 257 | SYSCALL_DEFINE1(acct, const char __user *, name) |
1da177e4 | 258 | { |
05b90496 | 259 | int error = 0; |
1da177e4 LT |
260 | |
261 | if (!capable(CAP_SYS_PACCT)) | |
262 | return -EPERM; | |
263 | ||
264 | if (name) { | |
91a27b2a | 265 | struct filename *tmp = getname(name); |
7b7b1ace | 266 | if (IS_ERR(tmp)) |
46c0a8ca | 267 | return PTR_ERR(tmp); |
9df7fa16 | 268 | mutex_lock(&acct_on_mutex); |
669abf4e | 269 | error = acct_on(tmp); |
9df7fa16 | 270 | mutex_unlock(&acct_on_mutex); |
1da177e4 | 271 | putname(tmp); |
7b7b1ace | 272 | } else { |
0b6b030f PE |
273 | struct bsd_acct_struct *acct; |
274 | ||
275 | acct = task_active_pid_ns(current)->bacct; | |
276 | if (acct == NULL) | |
277 | return 0; | |
278 | ||
05b90496 EP |
279 | spin_lock(&acct_lock); |
280 | acct_file_reopen(acct, NULL, NULL); | |
281 | spin_unlock(&acct_lock); | |
1da177e4 | 282 | } |
05b90496 | 283 | |
7b7b1ace AV |
284 | return error; |
285 | } | |
1da177e4 | 286 | |
7b7b1ace AV |
287 | /** |
288 | * acct_auto_close - turn off a filesystem's accounting if it is on | |
289 | * @m: vfsmount being shut down | |
290 | * | |
291 | * If the accounting is turned on for a file in the subtree pointed to | |
292 | * to by m, turn accounting off. Done when m is about to die. | |
293 | */ | |
294 | void acct_auto_close_mnt(struct vfsmount *m) | |
295 | { | |
0b6b030f PE |
296 | struct bsd_acct_struct *acct; |
297 | ||
a75d9797 | 298 | spin_lock(&acct_lock); |
b5a71748 PE |
299 | restart: |
300 | list_for_each_entry(acct, &acct_list, list) | |
301 | if (acct->file && acct->file->f_path.mnt == m) { | |
302 | acct_file_reopen(acct, NULL, NULL); | |
303 | goto restart; | |
304 | } | |
a75d9797 | 305 | spin_unlock(&acct_lock); |
1da177e4 LT |
306 | } |
307 | ||
417ef531 RD |
308 | /** |
309 | * acct_auto_close - turn off a filesystem's accounting if it is on | |
310 | * @sb: super block for the filesystem | |
311 | * | |
312 | * If the accounting is turned on for a file in the filesystem pointed | |
313 | * to by sb, turn accounting off. | |
1da177e4 LT |
314 | */ |
315 | void acct_auto_close(struct super_block *sb) | |
316 | { | |
0b6b030f PE |
317 | struct bsd_acct_struct *acct; |
318 | ||
a75d9797 | 319 | spin_lock(&acct_lock); |
b5a71748 PE |
320 | restart: |
321 | list_for_each_entry(acct, &acct_list, list) | |
d8c9584e | 322 | if (acct->file && acct->file->f_path.dentry->d_sb == sb) { |
b5a71748 PE |
323 | acct_file_reopen(acct, NULL, NULL); |
324 | goto restart; | |
325 | } | |
0b6b030f PE |
326 | spin_unlock(&acct_lock); |
327 | } | |
328 | ||
329 | void acct_exit_ns(struct pid_namespace *ns) | |
330 | { | |
11cad320 | 331 | struct bsd_acct_struct *acct = ns->bacct; |
0b6b030f | 332 | |
11cad320 VG |
333 | if (acct == NULL) |
334 | return; | |
0b6b030f | 335 | |
11cad320 VG |
336 | spin_lock(&acct_lock); |
337 | if (acct->file != NULL) | |
338 | acct_file_reopen(acct, NULL, NULL); | |
a75d9797 | 339 | spin_unlock(&acct_lock); |
11cad320 VG |
340 | |
341 | kfree(acct); | |
1da177e4 LT |
342 | } |
343 | ||
344 | /* | |
345 | * encode an unsigned long into a comp_t | |
346 | * | |
347 | * This routine has been adopted from the encode_comp_t() function in | |
348 | * the kern_acct.c file of the FreeBSD operating system. The encoding | |
349 | * is a 13-bit fraction with a 3-bit (base 8) exponent. | |
350 | */ | |
351 | ||
352 | #define MANTSIZE 13 /* 13 bit mantissa. */ | |
353 | #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */ | |
354 | #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */ | |
355 | ||
356 | static comp_t encode_comp_t(unsigned long value) | |
357 | { | |
358 | int exp, rnd; | |
359 | ||
360 | exp = rnd = 0; | |
361 | while (value > MAXFRACT) { | |
362 | rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */ | |
363 | value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */ | |
364 | exp++; | |
365 | } | |
366 | ||
367 | /* | |
6ae965cd DW |
368 | * If we need to round up, do it (and handle overflow correctly). |
369 | */ | |
1da177e4 LT |
370 | if (rnd && (++value > MAXFRACT)) { |
371 | value >>= EXPSIZE; | |
372 | exp++; | |
373 | } | |
374 | ||
375 | /* | |
6ae965cd DW |
376 | * Clean it up and polish it off. |
377 | */ | |
1da177e4 LT |
378 | exp <<= MANTSIZE; /* Shift the exponent into place */ |
379 | exp += value; /* and add on the mantissa. */ | |
380 | return exp; | |
381 | } | |
382 | ||
383 | #if ACCT_VERSION==1 || ACCT_VERSION==2 | |
384 | /* | |
385 | * encode an u64 into a comp2_t (24 bits) | |
386 | * | |
387 | * Format: 5 bit base 2 exponent, 20 bits mantissa. | |
388 | * The leading bit of the mantissa is not stored, but implied for | |
389 | * non-zero exponents. | |
390 | * Largest encodable value is 50 bits. | |
391 | */ | |
392 | ||
393 | #define MANTSIZE2 20 /* 20 bit mantissa. */ | |
394 | #define EXPSIZE2 5 /* 5 bit base 2 exponent. */ | |
395 | #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */ | |
396 | #define MAXEXP2 ((1 <<EXPSIZE2) - 1) /* Maximum exponent. */ | |
397 | ||
398 | static comp2_t encode_comp2_t(u64 value) | |
399 | { | |
6ae965cd DW |
400 | int exp, rnd; |
401 | ||
402 | exp = (value > (MAXFRACT2>>1)); | |
403 | rnd = 0; | |
404 | while (value > MAXFRACT2) { | |
405 | rnd = value & 1; | |
406 | value >>= 1; | |
407 | exp++; | |
408 | } | |
409 | ||
410 | /* | |
411 | * If we need to round up, do it (and handle overflow correctly). | |
412 | */ | |
413 | if (rnd && (++value > MAXFRACT2)) { | |
414 | value >>= 1; | |
415 | exp++; | |
416 | } | |
417 | ||
418 | if (exp > MAXEXP2) { | |
419 | /* Overflow. Return largest representable number instead. */ | |
420 | return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1; | |
421 | } else { | |
422 | return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1)); | |
423 | } | |
1da177e4 LT |
424 | } |
425 | #endif | |
426 | ||
427 | #if ACCT_VERSION==3 | |
428 | /* | |
429 | * encode an u64 into a 32 bit IEEE float | |
430 | */ | |
431 | static u32 encode_float(u64 value) | |
432 | { | |
433 | unsigned exp = 190; | |
434 | unsigned u; | |
435 | ||
436 | if (value==0) return 0; | |
437 | while ((s64)value > 0){ | |
438 | value <<= 1; | |
439 | exp--; | |
440 | } | |
441 | u = (u32)(value >> 40) & 0x7fffffu; | |
442 | return u | (exp << 23); | |
443 | } | |
444 | #endif | |
445 | ||
446 | /* | |
447 | * Write an accounting entry for an exiting process | |
448 | * | |
449 | * The acct_process() call is the workhorse of the process | |
450 | * accounting system. The struct acct is built here and then written | |
451 | * into the accounting file. This function should only be called from | |
bcbe4a07 | 452 | * do_exit() or when switching to a different output file. |
1da177e4 LT |
453 | */ |
454 | ||
cdd37e23 | 455 | static void fill_ac(acct_t *ac) |
1da177e4 | 456 | { |
0e464814 | 457 | struct pacct_struct *pacct = ¤t->signal->pacct; |
ccbf62d8 | 458 | u64 elapsed, run_time; |
24ec839c | 459 | struct tty_struct *tty; |
1da177e4 LT |
460 | |
461 | /* | |
462 | * Fill the accounting struct with the needed info as recorded | |
463 | * by the different kernel functions. | |
464 | */ | |
cdd37e23 | 465 | memset(ac, 0, sizeof(acct_t)); |
1da177e4 | 466 | |
cdd37e23 AV |
467 | ac->ac_version = ACCT_VERSION | ACCT_BYTEORDER; |
468 | strlcpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm)); | |
1da177e4 LT |
469 | |
470 | /* calculate run_time in nsec*/ | |
ccbf62d8 TG |
471 | run_time = ktime_get_ns(); |
472 | run_time -= current->group_leader->start_time; | |
1da177e4 LT |
473 | /* convert nsec -> AHZ */ |
474 | elapsed = nsec_to_AHZ(run_time); | |
475 | #if ACCT_VERSION==3 | |
cdd37e23 | 476 | ac->ac_etime = encode_float(elapsed); |
1da177e4 | 477 | #else |
cdd37e23 | 478 | ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ? |
1da177e4 LT |
479 | (unsigned long) elapsed : (unsigned long) -1l); |
480 | #endif | |
481 | #if ACCT_VERSION==1 || ACCT_VERSION==2 | |
482 | { | |
483 | /* new enlarged etime field */ | |
484 | comp2_t etime = encode_comp2_t(elapsed); | |
cdd37e23 AV |
485 | ac->ac_etime_hi = etime >> 16; |
486 | ac->ac_etime_lo = (u16) etime; | |
1da177e4 LT |
487 | } |
488 | #endif | |
489 | do_div(elapsed, AHZ); | |
cdd37e23 AV |
490 | ac->ac_btime = get_seconds() - elapsed; |
491 | #if ACCT_VERSION==2 | |
492 | ac->ac_ahz = AHZ; | |
493 | #endif | |
494 | ||
495 | spin_lock_irq(¤t->sighand->siglock); | |
496 | tty = current->signal->tty; /* Safe as we hold the siglock */ | |
497 | ac->ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0; | |
498 | ac->ac_utime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_utime))); | |
499 | ac->ac_stime = encode_comp_t(jiffies_to_AHZ(cputime_to_jiffies(pacct->ac_stime))); | |
500 | ac->ac_flag = pacct->ac_flag; | |
501 | ac->ac_mem = encode_comp_t(pacct->ac_mem); | |
502 | ac->ac_minflt = encode_comp_t(pacct->ac_minflt); | |
503 | ac->ac_majflt = encode_comp_t(pacct->ac_majflt); | |
504 | ac->ac_exitcode = pacct->ac_exitcode; | |
505 | spin_unlock_irq(¤t->sighand->siglock); | |
506 | } | |
507 | /* | |
508 | * do_acct_process does all actual work. Caller holds the reference to file. | |
509 | */ | |
510 | static void do_acct_process(struct bsd_acct_struct *acct, | |
511 | struct pid_namespace *ns, struct file *file) | |
512 | { | |
513 | acct_t ac; | |
514 | unsigned long flim; | |
515 | const struct cred *orig_cred; | |
516 | ||
517 | /* | |
518 | * Accounting records are not subject to resource limits. | |
519 | */ | |
520 | flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; | |
521 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY; | |
522 | /* Perform file operations on behalf of whoever enabled accounting */ | |
523 | orig_cred = override_creds(file->f_cred); | |
524 | ||
525 | /* | |
526 | * First check to see if there is enough free_space to continue | |
527 | * the process accounting system. | |
528 | */ | |
529 | if (!check_free_space(acct, file)) | |
530 | goto out; | |
531 | ||
532 | fill_ac(&ac); | |
1da177e4 | 533 | /* we really need to bite the bullet and change layout */ |
f8f3d4de EB |
534 | ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid); |
535 | ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid); | |
1da177e4 LT |
536 | #if ACCT_VERSION==1 || ACCT_VERSION==2 |
537 | /* backward-compatible 16 bit fields */ | |
76aac0e9 DH |
538 | ac.ac_uid16 = ac.ac_uid; |
539 | ac.ac_gid16 = ac.ac_gid; | |
1da177e4 LT |
540 | #endif |
541 | #if ACCT_VERSION==3 | |
5f7b703f | 542 | ac.ac_pid = task_tgid_nr_ns(current, ns); |
a846a195 | 543 | rcu_read_lock(); |
5f7b703f | 544 | ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent), ns); |
a846a195 | 545 | rcu_read_unlock(); |
1da177e4 | 546 | #endif |
5ae98f15 JK |
547 | /* |
548 | * Get freeze protection. If the fs is frozen, just skip the write | |
549 | * as we could deadlock the system otherwise. | |
550 | */ | |
ed44724b AV |
551 | if (file_start_write_trylock(file)) { |
552 | /* it's been opened O_APPEND, so position is irrelevant */ | |
553 | loff_t pos = 0; | |
554 | __kernel_write(file, (char *)&ac, sizeof(acct_t), &pos); | |
555 | file_end_write(file); | |
556 | } | |
d8e180dc | 557 | out: |
ed44724b | 558 | current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim; |
d8e180dc | 559 | revert_creds(orig_cred); |
1da177e4 LT |
560 | } |
561 | ||
0e464814 KK |
562 | /** |
563 | * acct_collect - collect accounting information into pacct_struct | |
f6ec29a4 KK |
564 | * @exitcode: task exit code |
565 | * @group_dead: not 0, if this thread is the last one in the process. | |
0e464814 | 566 | */ |
f6ec29a4 | 567 | void acct_collect(long exitcode, int group_dead) |
0e464814 KK |
568 | { |
569 | struct pacct_struct *pacct = ¤t->signal->pacct; | |
6fac4829 | 570 | cputime_t utime, stime; |
0e464814 KK |
571 | unsigned long vsize = 0; |
572 | ||
f6ec29a4 | 573 | if (group_dead && current->mm) { |
0e464814 KK |
574 | struct vm_area_struct *vma; |
575 | down_read(¤t->mm->mmap_sem); | |
576 | vma = current->mm->mmap; | |
577 | while (vma) { | |
578 | vsize += vma->vm_end - vma->vm_start; | |
579 | vma = vma->vm_next; | |
580 | } | |
581 | up_read(¤t->mm->mmap_sem); | |
582 | } | |
583 | ||
77787bfb | 584 | spin_lock_irq(¤t->sighand->siglock); |
f6ec29a4 KK |
585 | if (group_dead) |
586 | pacct->ac_mem = vsize / 1024; | |
587 | if (thread_group_leader(current)) { | |
588 | pacct->ac_exitcode = exitcode; | |
589 | if (current->flags & PF_FORKNOEXEC) | |
590 | pacct->ac_flag |= AFORK; | |
591 | } | |
592 | if (current->flags & PF_SUPERPRIV) | |
593 | pacct->ac_flag |= ASU; | |
594 | if (current->flags & PF_DUMPCORE) | |
595 | pacct->ac_flag |= ACORE; | |
596 | if (current->flags & PF_SIGNALED) | |
597 | pacct->ac_flag |= AXSIG; | |
6fac4829 FW |
598 | task_cputime(current, &utime, &stime); |
599 | pacct->ac_utime += utime; | |
600 | pacct->ac_stime += stime; | |
77787bfb KK |
601 | pacct->ac_minflt += current->min_flt; |
602 | pacct->ac_majflt += current->maj_flt; | |
603 | spin_unlock_irq(¤t->sighand->siglock); | |
0e464814 KK |
604 | } |
605 | ||
e25ff11f | 606 | static void slow_acct_process(struct pid_namespace *ns) |
1da177e4 | 607 | { |
e25ff11f AV |
608 | for ( ; ns; ns = ns->parent) { |
609 | struct file *file = NULL; | |
610 | struct bsd_acct_struct *acct; | |
1da177e4 | 611 | |
e25ff11f AV |
612 | acct = ns->bacct; |
613 | /* | |
614 | * accelerate the common fastpath: | |
615 | */ | |
616 | if (!acct || !acct->file) | |
617 | continue; | |
1da177e4 | 618 | |
e25ff11f AV |
619 | spin_lock(&acct_lock); |
620 | file = acct->file; | |
621 | if (unlikely(!file)) { | |
622 | spin_unlock(&acct_lock); | |
623 | continue; | |
624 | } | |
625 | get_file(file); | |
a75d9797 | 626 | spin_unlock(&acct_lock); |
1da177e4 | 627 | |
e25ff11f AV |
628 | do_acct_process(acct, ns, file); |
629 | fput(file); | |
630 | } | |
1da177e4 | 631 | } |
7d1e1350 PE |
632 | |
633 | /** | |
e25ff11f | 634 | * acct_process |
7d1e1350 PE |
635 | * |
636 | * handles process accounting for an exiting task | |
637 | */ | |
638 | void acct_process(void) | |
639 | { | |
640 | struct pid_namespace *ns; | |
641 | ||
0c18d7a5 PE |
642 | /* |
643 | * This loop is safe lockless, since current is still | |
644 | * alive and holds its namespace, which in turn holds | |
645 | * its parent. | |
646 | */ | |
e25ff11f AV |
647 | for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) { |
648 | struct bsd_acct_struct *acct = ns->bacct; | |
649 | if (acct && acct->file) | |
650 | break; | |
651 | } | |
652 | if (unlikely(ns)) | |
653 | slow_acct_process(ns); | |
7d1e1350 | 654 | } |