acct.c 15 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606
  1. // SPDX-License-Identifier: GPL-2.0
  2. /*
  3. * linux/kernel/acct.c
  4. *
  5. * BSD Process Accounting for Linux
  6. *
  7. * Author: Marco van Wieringen <mvw@planets.elm.net>
  8. *
  9. * Some code based on ideas and code from:
  10. * Thomas K. Dyas <tdyas@eden.rutgers.edu>
  11. *
  12. * This file implements BSD-style process accounting. Whenever any
  13. * process exits, an accounting record of type "struct acct" is
  14. * written to the file specified with the acct() system call. It is
  15. * up to user-level programs to do useful things with the accounting
  16. * log. The kernel just provides the raw accounting information.
  17. *
  18. * (C) Copyright 1995 - 1997 Marco van Wieringen - ELM Consultancy B.V.
  19. *
  20. * Plugged two leaks. 1) It didn't return acct_file into the free_filps if
  21. * the file happened to be read-only. 2) If the accounting was suspended
  22. * due to the lack of space it happily allowed to reopen it and completely
  23. * lost the old acct_file. 3/10/98, Al Viro.
  24. *
  25. * Now we silently close acct_file on attempt to reopen. Cleaned sys_acct().
  26. * XTerms and EMACS are manifestations of pure evil. 21/10/98, AV.
  27. *
  28. * Fixed a nasty interaction with sys_umount(). If the accounting
  29. * was suspeneded we failed to stop it on umount(). Messy.
  30. * Another one: remount to readonly didn't stop accounting.
  31. * Question: what should we do if we have CAP_SYS_ADMIN but not
  32. * CAP_SYS_PACCT? Current code does the following: umount returns -EBUSY
  33. * unless we are messing with the root. In that case we are getting a
  34. * real mess with do_remount_sb(). 9/11/98, AV.
  35. *
  36. * Fixed a bunch of races (and pair of leaks). Probably not the best way,
  37. * but this one obviously doesn't introduce deadlocks. Later. BTW, found
  38. * one race (and leak) in BSD implementation.
  39. * OK, that's better. ANOTHER race and leak in BSD variant. There always
  40. * is one more bug... 10/11/98, AV.
  41. *
  42. * Oh, fsck... Oopsable SMP race in do_process_acct() - we must hold
  43. * ->mmap_lock to walk the vma list of current->mm. Nasty, since it leaks
  44. * a struct file opened for write. Fixed. 2/6/2000, AV.
  45. */
  46. #include <linux/mm.h>
  47. #include <linux/slab.h>
  48. #include <linux/acct.h>
  49. #include <linux/capability.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 <linux/mount.h>
  58. #include <linux/uaccess.h>
  59. #include <linux/sched/cputime.h>
  60. #include <asm/div64.h>
  61. #include <linux/blkdev.h> /* sector_div */
  62. #include <linux/pid_namespace.h>
  63. #include <linux/fs_pin.h>
  64. /*
  65. * These constants control the amount of freespace that suspend and
  66. * resume the process accounting system, and the time delay between
  67. * each check.
  68. * Turned into sysctl-controllable parameters. AV, 12/11/98
  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. * External references and all of the globals.
  76. */
  77. struct bsd_acct_struct {
  78. struct fs_pin pin;
  79. atomic_long_t count;
  80. struct rcu_head rcu;
  81. struct mutex lock;
  82. int active;
  83. unsigned long needcheck;
  84. struct file *file;
  85. struct pid_namespace *ns;
  86. struct work_struct work;
  87. struct completion done;
  88. };
  89. static void do_acct_process(struct bsd_acct_struct *acct);
  90. /*
  91. * Check the amount of free space and suspend/resume accordingly.
  92. */
  93. static int check_free_space(struct bsd_acct_struct *acct)
  94. {
  95. struct kstatfs sbuf;
  96. if (time_is_after_jiffies(acct->needcheck))
  97. goto out;
  98. /* May block */
  99. if (vfs_statfs(&acct->file->f_path, &sbuf))
  100. goto out;
  101. if (acct->active) {
  102. u64 suspend = sbuf.f_blocks * SUSPEND;
  103. do_div(suspend, 100);
  104. if (sbuf.f_bavail <= suspend) {
  105. acct->active = 0;
  106. pr_info("Process accounting paused\n");
  107. }
  108. } else {
  109. u64 resume = sbuf.f_blocks * RESUME;
  110. do_div(resume, 100);
  111. if (sbuf.f_bavail >= resume) {
  112. acct->active = 1;
  113. pr_info("Process accounting resumed\n");
  114. }
  115. }
  116. acct->needcheck = jiffies + ACCT_TIMEOUT*HZ;
  117. out:
  118. return acct->active;
  119. }
  120. static void acct_put(struct bsd_acct_struct *p)
  121. {
  122. if (atomic_long_dec_and_test(&p->count))
  123. kfree_rcu(p, rcu);
  124. }
  125. static inline struct bsd_acct_struct *to_acct(struct fs_pin *p)
  126. {
  127. return p ? container_of(p, struct bsd_acct_struct, pin) : NULL;
  128. }
  129. static struct bsd_acct_struct *acct_get(struct pid_namespace *ns)
  130. {
  131. struct bsd_acct_struct *res;
  132. again:
  133. smp_rmb();
  134. rcu_read_lock();
  135. res = to_acct(READ_ONCE(ns->bacct));
  136. if (!res) {
  137. rcu_read_unlock();
  138. return NULL;
  139. }
  140. if (!atomic_long_inc_not_zero(&res->count)) {
  141. rcu_read_unlock();
  142. cpu_relax();
  143. goto again;
  144. }
  145. rcu_read_unlock();
  146. mutex_lock(&res->lock);
  147. if (res != to_acct(READ_ONCE(ns->bacct))) {
  148. mutex_unlock(&res->lock);
  149. acct_put(res);
  150. goto again;
  151. }
  152. return res;
  153. }
  154. static void acct_pin_kill(struct fs_pin *pin)
  155. {
  156. struct bsd_acct_struct *acct = to_acct(pin);
  157. mutex_lock(&acct->lock);
  158. do_acct_process(acct);
  159. schedule_work(&acct->work);
  160. wait_for_completion(&acct->done);
  161. cmpxchg(&acct->ns->bacct, pin, NULL);
  162. mutex_unlock(&acct->lock);
  163. pin_remove(pin);
  164. acct_put(acct);
  165. }
  166. static void close_work(struct work_struct *work)
  167. {
  168. struct bsd_acct_struct *acct = container_of(work, struct bsd_acct_struct, work);
  169. struct file *file = acct->file;
  170. if (file->f_op->flush)
  171. file->f_op->flush(file, NULL);
  172. __fput_sync(file);
  173. complete(&acct->done);
  174. }
  175. static int acct_on(struct filename *pathname)
  176. {
  177. struct file *file;
  178. struct vfsmount *mnt, *internal;
  179. struct pid_namespace *ns = task_active_pid_ns(current);
  180. struct bsd_acct_struct *acct;
  181. struct fs_pin *old;
  182. int err;
  183. acct = kzalloc(sizeof(struct bsd_acct_struct), GFP_KERNEL);
  184. if (!acct)
  185. return -ENOMEM;
  186. /* Difference from BSD - they don't do O_APPEND */
  187. file = file_open_name(pathname, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
  188. if (IS_ERR(file)) {
  189. kfree(acct);
  190. return PTR_ERR(file);
  191. }
  192. if (!S_ISREG(file_inode(file)->i_mode)) {
  193. kfree(acct);
  194. filp_close(file, NULL);
  195. return -EACCES;
  196. }
  197. if (!(file->f_mode & FMODE_CAN_WRITE)) {
  198. kfree(acct);
  199. filp_close(file, NULL);
  200. return -EIO;
  201. }
  202. internal = mnt_clone_internal(&file->f_path);
  203. if (IS_ERR(internal)) {
  204. kfree(acct);
  205. filp_close(file, NULL);
  206. return PTR_ERR(internal);
  207. }
  208. err = __mnt_want_write(internal);
  209. if (err) {
  210. mntput(internal);
  211. kfree(acct);
  212. filp_close(file, NULL);
  213. return err;
  214. }
  215. mnt = file->f_path.mnt;
  216. file->f_path.mnt = internal;
  217. atomic_long_set(&acct->count, 1);
  218. init_fs_pin(&acct->pin, acct_pin_kill);
  219. acct->file = file;
  220. acct->needcheck = jiffies;
  221. acct->ns = ns;
  222. mutex_init(&acct->lock);
  223. INIT_WORK(&acct->work, close_work);
  224. init_completion(&acct->done);
  225. mutex_lock_nested(&acct->lock, 1); /* nobody has seen it yet */
  226. pin_insert(&acct->pin, mnt);
  227. rcu_read_lock();
  228. old = xchg(&ns->bacct, &acct->pin);
  229. mutex_unlock(&acct->lock);
  230. pin_kill(old);
  231. __mnt_drop_write(mnt);
  232. mntput(mnt);
  233. return 0;
  234. }
  235. static DEFINE_MUTEX(acct_on_mutex);
  236. /**
  237. * sys_acct - enable/disable process accounting
  238. * @name: file name for accounting records or NULL to shutdown accounting
  239. *
  240. * sys_acct() is the only system call needed to implement process
  241. * accounting. It takes the name of the file where accounting records
  242. * should be written. If the filename is NULL, accounting will be
  243. * shutdown.
  244. *
  245. * Returns: 0 for success or negative errno values for failure.
  246. */
  247. SYSCALL_DEFINE1(acct, const char __user *, name)
  248. {
  249. int error = 0;
  250. if (!capable(CAP_SYS_PACCT))
  251. return -EPERM;
  252. if (name) {
  253. struct filename *tmp = getname(name);
  254. if (IS_ERR(tmp))
  255. return PTR_ERR(tmp);
  256. mutex_lock(&acct_on_mutex);
  257. error = acct_on(tmp);
  258. mutex_unlock(&acct_on_mutex);
  259. putname(tmp);
  260. } else {
  261. rcu_read_lock();
  262. pin_kill(task_active_pid_ns(current)->bacct);
  263. }
  264. return error;
  265. }
  266. void acct_exit_ns(struct pid_namespace *ns)
  267. {
  268. rcu_read_lock();
  269. pin_kill(ns->bacct);
  270. }
  271. /*
  272. * encode an unsigned long into a comp_t
  273. *
  274. * This routine has been adopted from the encode_comp_t() function in
  275. * the kern_acct.c file of the FreeBSD operating system. The encoding
  276. * is a 13-bit fraction with a 3-bit (base 8) exponent.
  277. */
  278. #define MANTSIZE 13 /* 13 bit mantissa. */
  279. #define EXPSIZE 3 /* Base 8 (3 bit) exponent. */
  280. #define MAXFRACT ((1 << MANTSIZE) - 1) /* Maximum fractional value. */
  281. static comp_t encode_comp_t(unsigned long value)
  282. {
  283. int exp, rnd;
  284. exp = rnd = 0;
  285. while (value > MAXFRACT) {
  286. rnd = value & (1 << (EXPSIZE - 1)); /* Round up? */
  287. value >>= EXPSIZE; /* Base 8 exponent == 3 bit shift. */
  288. exp++;
  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. * Clean it up and polish it off.
  299. */
  300. exp <<= MANTSIZE; /* Shift the exponent into place */
  301. exp += value; /* and add on the mantissa. */
  302. return exp;
  303. }
  304. #if ACCT_VERSION == 1 || ACCT_VERSION == 2
  305. /*
  306. * encode an u64 into a comp2_t (24 bits)
  307. *
  308. * Format: 5 bit base 2 exponent, 20 bits mantissa.
  309. * The leading bit of the mantissa is not stored, but implied for
  310. * non-zero exponents.
  311. * Largest encodable value is 50 bits.
  312. */
  313. #define MANTSIZE2 20 /* 20 bit mantissa. */
  314. #define EXPSIZE2 5 /* 5 bit base 2 exponent. */
  315. #define MAXFRACT2 ((1ul << MANTSIZE2) - 1) /* Maximum fractional value. */
  316. #define MAXEXP2 ((1 << EXPSIZE2) - 1) /* Maximum exponent. */
  317. static comp2_t encode_comp2_t(u64 value)
  318. {
  319. int exp, rnd;
  320. exp = (value > (MAXFRACT2>>1));
  321. rnd = 0;
  322. while (value > MAXFRACT2) {
  323. rnd = value & 1;
  324. value >>= 1;
  325. exp++;
  326. }
  327. /*
  328. * If we need to round up, do it (and handle overflow correctly).
  329. */
  330. if (rnd && (++value > MAXFRACT2)) {
  331. value >>= 1;
  332. exp++;
  333. }
  334. if (exp > MAXEXP2) {
  335. /* Overflow. Return largest representable number instead. */
  336. return (1ul << (MANTSIZE2+EXPSIZE2-1)) - 1;
  337. } else {
  338. return (value & (MAXFRACT2>>1)) | (exp << (MANTSIZE2-1));
  339. }
  340. }
  341. #endif
  342. #if ACCT_VERSION == 3
  343. /*
  344. * encode an u64 into a 32 bit IEEE float
  345. */
  346. static u32 encode_float(u64 value)
  347. {
  348. unsigned exp = 190;
  349. unsigned u;
  350. if (value == 0)
  351. return 0;
  352. while ((s64)value > 0) {
  353. value <<= 1;
  354. exp--;
  355. }
  356. u = (u32)(value >> 40) & 0x7fffffu;
  357. return u | (exp << 23);
  358. }
  359. #endif
  360. /*
  361. * Write an accounting entry for an exiting process
  362. *
  363. * The acct_process() call is the workhorse of the process
  364. * accounting system. The struct acct is built here and then written
  365. * into the accounting file. This function should only be called from
  366. * do_exit() or when switching to a different output file.
  367. */
  368. static void fill_ac(acct_t *ac)
  369. {
  370. struct pacct_struct *pacct = &current->signal->pacct;
  371. u64 elapsed, run_time;
  372. time64_t btime;
  373. struct tty_struct *tty;
  374. /*
  375. * Fill the accounting struct with the needed info as recorded
  376. * by the different kernel functions.
  377. */
  378. memset(ac, 0, sizeof(acct_t));
  379. ac->ac_version = ACCT_VERSION | ACCT_BYTEORDER;
  380. strlcpy(ac->ac_comm, current->comm, sizeof(ac->ac_comm));
  381. /* calculate run_time in nsec*/
  382. run_time = ktime_get_ns();
  383. run_time -= current->group_leader->start_time;
  384. /* convert nsec -> AHZ */
  385. elapsed = nsec_to_AHZ(run_time);
  386. #if ACCT_VERSION == 3
  387. ac->ac_etime = encode_float(elapsed);
  388. #else
  389. ac->ac_etime = encode_comp_t(elapsed < (unsigned long) -1l ?
  390. (unsigned long) elapsed : (unsigned long) -1l);
  391. #endif
  392. #if ACCT_VERSION == 1 || ACCT_VERSION == 2
  393. {
  394. /* new enlarged etime field */
  395. comp2_t etime = encode_comp2_t(elapsed);
  396. ac->ac_etime_hi = etime >> 16;
  397. ac->ac_etime_lo = (u16) etime;
  398. }
  399. #endif
  400. do_div(elapsed, AHZ);
  401. btime = ktime_get_real_seconds() - elapsed;
  402. ac->ac_btime = clamp_t(time64_t, btime, 0, U32_MAX);
  403. #if ACCT_VERSION==2
  404. ac->ac_ahz = AHZ;
  405. #endif
  406. spin_lock_irq(&current->sighand->siglock);
  407. tty = current->signal->tty; /* Safe as we hold the siglock */
  408. ac->ac_tty = tty ? old_encode_dev(tty_devnum(tty)) : 0;
  409. ac->ac_utime = encode_comp_t(nsec_to_AHZ(pacct->ac_utime));
  410. ac->ac_stime = encode_comp_t(nsec_to_AHZ(pacct->ac_stime));
  411. ac->ac_flag = pacct->ac_flag;
  412. ac->ac_mem = encode_comp_t(pacct->ac_mem);
  413. ac->ac_minflt = encode_comp_t(pacct->ac_minflt);
  414. ac->ac_majflt = encode_comp_t(pacct->ac_majflt);
  415. ac->ac_exitcode = pacct->ac_exitcode;
  416. spin_unlock_irq(&current->sighand->siglock);
  417. }
  418. /*
  419. * do_acct_process does all actual work. Caller holds the reference to file.
  420. */
  421. static void do_acct_process(struct bsd_acct_struct *acct)
  422. {
  423. acct_t ac;
  424. unsigned long flim;
  425. const struct cred *orig_cred;
  426. struct file *file = acct->file;
  427. /*
  428. * Accounting records are not subject to resource limits.
  429. */
  430. flim = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
  431. current->signal->rlim[RLIMIT_FSIZE].rlim_cur = RLIM_INFINITY;
  432. /* Perform file operations on behalf of whoever enabled accounting */
  433. orig_cred = override_creds(file->f_cred);
  434. /*
  435. * First check to see if there is enough free_space to continue
  436. * the process accounting system.
  437. */
  438. if (!check_free_space(acct))
  439. goto out;
  440. fill_ac(&ac);
  441. /* we really need to bite the bullet and change layout */
  442. ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid);
  443. ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid);
  444. #if ACCT_VERSION == 1 || ACCT_VERSION == 2
  445. /* backward-compatible 16 bit fields */
  446. ac.ac_uid16 = ac.ac_uid;
  447. ac.ac_gid16 = ac.ac_gid;
  448. #endif
  449. #if ACCT_VERSION == 3
  450. {
  451. struct pid_namespace *ns = acct->ns;
  452. ac.ac_pid = task_tgid_nr_ns(current, ns);
  453. rcu_read_lock();
  454. ac.ac_ppid = task_tgid_nr_ns(rcu_dereference(current->real_parent),
  455. ns);
  456. rcu_read_unlock();
  457. }
  458. #endif
  459. /*
  460. * Get freeze protection. If the fs is frozen, just skip the write
  461. * as we could deadlock the system otherwise.
  462. */
  463. if (file_start_write_trylock(file)) {
  464. /* it's been opened O_APPEND, so position is irrelevant */
  465. loff_t pos = 0;
  466. __kernel_write(file, &ac, sizeof(acct_t), &pos);
  467. file_end_write(file);
  468. }
  469. out:
  470. current->signal->rlim[RLIMIT_FSIZE].rlim_cur = flim;
  471. revert_creds(orig_cred);
  472. }
  473. /**
  474. * acct_collect - collect accounting information into pacct_struct
  475. * @exitcode: task exit code
  476. * @group_dead: not 0, if this thread is the last one in the process.
  477. */
  478. void acct_collect(long exitcode, int group_dead)
  479. {
  480. struct pacct_struct *pacct = &current->signal->pacct;
  481. u64 utime, stime;
  482. unsigned long vsize = 0;
  483. if (group_dead && current->mm) {
  484. struct vm_area_struct *vma;
  485. mmap_read_lock(current->mm);
  486. vma = current->mm->mmap;
  487. while (vma) {
  488. vsize += vma->vm_end - vma->vm_start;
  489. vma = vma->vm_next;
  490. }
  491. mmap_read_unlock(current->mm);
  492. }
  493. spin_lock_irq(&current->sighand->siglock);
  494. if (group_dead)
  495. pacct->ac_mem = vsize / 1024;
  496. if (thread_group_leader(current)) {
  497. pacct->ac_exitcode = exitcode;
  498. if (current->flags & PF_FORKNOEXEC)
  499. pacct->ac_flag |= AFORK;
  500. }
  501. if (current->flags & PF_SUPERPRIV)
  502. pacct->ac_flag |= ASU;
  503. if (current->flags & PF_DUMPCORE)
  504. pacct->ac_flag |= ACORE;
  505. if (current->flags & PF_SIGNALED)
  506. pacct->ac_flag |= AXSIG;
  507. task_cputime(current, &utime, &stime);
  508. pacct->ac_utime += utime;
  509. pacct->ac_stime += stime;
  510. pacct->ac_minflt += current->min_flt;
  511. pacct->ac_majflt += current->maj_flt;
  512. spin_unlock_irq(&current->sighand->siglock);
  513. }
  514. static void slow_acct_process(struct pid_namespace *ns)
  515. {
  516. for ( ; ns; ns = ns->parent) {
  517. struct bsd_acct_struct *acct = acct_get(ns);
  518. if (acct) {
  519. do_acct_process(acct);
  520. mutex_unlock(&acct->lock);
  521. acct_put(acct);
  522. }
  523. }
  524. }
  525. /**
  526. * acct_process - handles process accounting for an exiting task
  527. */
  528. void acct_process(void)
  529. {
  530. struct pid_namespace *ns;
  531. /*
  532. * This loop is safe lockless, since current is still
  533. * alive and holds its namespace, which in turn holds
  534. * its parent.
  535. */
  536. for (ns = task_active_pid_ns(current); ns != NULL; ns = ns->parent) {
  537. if (ns->bacct)
  538. break;
  539. }
  540. if (unlikely(ns))
  541. slow_acct_process(ns);
  542. }