1 /* 2 * linux/fs/proc/array.c 3 * 4 * Copyright (C) 1992 by Linus Torvalds 5 * based on ideas by Darren Senn 6 * 7 * Fixes: 8 * Michael. K. Johnson: stat,statm extensions. 9 * <johnsonm@stolaf.edu> 10 * 11 * Pauline Middelink : Made cmdline,envline only break at '\0's, to 12 * make sure SET_PROCTITLE works. Also removed 13 * bad '!' which forced address recalculation for 14 * EVERY character on the current page. 15 * <middelin@polyware.iaf.nl> 16 * 17 * Danny ter Haar : added cpuinfo 18 * <dth@cistron.nl> 19 * 20 * Alessandro Rubini : profile extension. 21 * <rubini@ipvvis.unipv.it> 22 * 23 * Jeff Tranter : added BogoMips field to cpuinfo 24 * <Jeff_Tranter@Mitel.COM> 25 * 26 * Bruno Haible : remove 4K limit for the maps file 27 * <haible@ma2s2.mathematik.uni-karlsruhe.de> 28 * 29 * Yves Arrouye : remove removal of trailing spaces in get_array. 30 * <Yves.Arrouye@marin.fdn.fr> 31 * 32 * Jerome Forissier : added per-CPU time information to /proc/stat 33 * and /proc/<pid>/cpu extension 34 * <forissier@isia.cma.fr> 35 * - Incorporation and non-SMP safe operation 36 * of forissier patch in 2.1.78 by 37 * Hans Marcus <crowbar@concepts.nl> 38 * 39 * aeb@cwi.nl : /proc/partitions 40 * 41 * 42 * Alan Cox : security fixes. 43 * <alan@lxorguk.ukuu.org.uk> 44 * 45 * Al Viro : safe handling of mm_struct 46 * 47 * Gerhard Wichert : added BIGMEM support 48 * Siemens AG <Gerhard.Wichert@pdb.siemens.de> 49 * 50 * Al Viro & Jeff Garzik : moved most of the thing into base.c and 51 * : proc_misc.c. The rest may eventually go into 52 * : base.c too. 53 */ 54 55 #include <linux/types.h> 56 #include <linux/errno.h> 57 #include <linux/time.h> 58 #include <linux/kernel.h> 59 #include <linux/kernel_stat.h> 60 #include <linux/tty.h> 61 #include <linux/string.h> 62 #include <linux/mman.h> 63 #include <linux/proc_fs.h> 64 #include <linux/ioport.h> 65 #include <linux/uaccess.h> 66 #include <linux/io.h> 67 #include <linux/mm.h> 68 #include <linux/hugetlb.h> 69 #include <linux/pagemap.h> 70 #include <linux/swap.h> 71 #include <linux/smp.h> 72 #include <linux/signal.h> 73 #include <linux/highmem.h> 74 #include <linux/file.h> 75 #include <linux/fdtable.h> 76 #include <linux/times.h> 77 #include <linux/cpuset.h> 78 #include <linux/rcupdate.h> 79 #include <linux/delayacct.h> 80 #include <linux/seq_file.h> 81 #include <linux/pid_namespace.h> 82 #include <linux/ptrace.h> 83 #include <linux/tracehook.h> 84 #include <linux/string_helpers.h> 85 #include <linux/user_namespace.h> 86 #include <linux/fs_struct.h> 87 88 #include <asm/pgtable.h> 89 #include <asm/processor.h> 90 #include "internal.h" 91 92 static inline void task_name(struct seq_file *m, struct task_struct *p) 93 { 94 char *buf; 95 size_t size; 96 char tcomm[sizeof(p->comm)]; 97 int ret; 98 99 get_task_comm(tcomm, p); 100 101 seq_puts(m, "Name:\t"); 102 103 size = seq_get_buf(m, &buf); 104 ret = string_escape_str(tcomm, buf, size, ESCAPE_SPACE | ESCAPE_SPECIAL, "\n\\"); 105 seq_commit(m, ret < size ? ret : -1); 106 107 seq_putc(m, '\n'); 108 } 109 110 /* 111 * The task state array is a strange "bitmap" of 112 * reasons to sleep. Thus "running" is zero, and 113 * you can test for combinations of others with 114 * simple bit tests. 115 */ 116 static const char * const task_state_array[] = { 117 "R (running)", /* 0 */ 118 "S (sleeping)", /* 1 */ 119 "D (disk sleep)", /* 2 */ 120 "T (stopped)", /* 4 */ 121 "t (tracing stop)", /* 8 */ 122 "X (dead)", /* 16 */ 123 "Z (zombie)", /* 32 */ 124 }; 125 126 static inline const char *get_task_state(struct task_struct *tsk) 127 { 128 unsigned int state = (tsk->state | tsk->exit_state) & TASK_REPORT; 129 130 /* 131 * Parked tasks do not run; they sit in __kthread_parkme(). 132 * Without this check, we would report them as running, which is 133 * clearly wrong, so we report them as sleeping instead. 134 */ 135 if (tsk->state == TASK_PARKED) 136 state = TASK_INTERRUPTIBLE; 137 138 BUILD_BUG_ON(1 + ilog2(TASK_REPORT) != ARRAY_SIZE(task_state_array)-1); 139 140 return task_state_array[fls(state)]; 141 } 142 143 static inline int get_task_umask(struct task_struct *tsk) 144 { 145 struct fs_struct *fs; 146 int umask = -ENOENT; 147 148 task_lock(tsk); 149 fs = tsk->fs; 150 if (fs) 151 umask = fs->umask; 152 task_unlock(tsk); 153 return umask; 154 } 155 156 static inline void task_state(struct seq_file *m, struct pid_namespace *ns, 157 struct pid *pid, struct task_struct *p) 158 { 159 struct user_namespace *user_ns = seq_user_ns(m); 160 struct group_info *group_info; 161 int g, umask; 162 struct task_struct *tracer; 163 const struct cred *cred; 164 pid_t ppid, tpid = 0, tgid, ngid; 165 unsigned int max_fds = 0; 166 167 rcu_read_lock(); 168 ppid = pid_alive(p) ? 169 task_tgid_nr_ns(rcu_dereference(p->real_parent), ns) : 0; 170 171 tracer = ptrace_parent(p); 172 if (tracer) 173 tpid = task_pid_nr_ns(tracer, ns); 174 175 tgid = task_tgid_nr_ns(p, ns); 176 ngid = task_numa_group_id(p); 177 cred = get_task_cred(p); 178 179 umask = get_task_umask(p); 180 if (umask >= 0) 181 seq_printf(m, "Umask:\t%#04o\n", umask); 182 183 task_lock(p); 184 if (p->files) 185 max_fds = files_fdtable(p->files)->max_fds; 186 task_unlock(p); 187 rcu_read_unlock(); 188 189 seq_printf(m, "State:\t%s", get_task_state(p)); 190 191 seq_put_decimal_ull(m, "\nTgid:\t", tgid); 192 seq_put_decimal_ull(m, "\nNgid:\t", ngid); 193 seq_put_decimal_ull(m, "\nPid:\t", pid_nr_ns(pid, ns)); 194 seq_put_decimal_ull(m, "\nPPid:\t", ppid); 195 seq_put_decimal_ull(m, "\nTracerPid:\t", tpid); 196 seq_put_decimal_ull(m, "\nUid:\t", from_kuid_munged(user_ns, cred->uid)); 197 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->euid)); 198 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->suid)); 199 seq_put_decimal_ull(m, "\t", from_kuid_munged(user_ns, cred->fsuid)); 200 seq_put_decimal_ull(m, "\nGid:\t", from_kgid_munged(user_ns, cred->gid)); 201 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->egid)); 202 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->sgid)); 203 seq_put_decimal_ull(m, "\t", from_kgid_munged(user_ns, cred->fsgid)); 204 seq_put_decimal_ull(m, "\nFDSize:\t", max_fds); 205 206 seq_puts(m, "\nGroups:\t"); 207 group_info = cred->group_info; 208 for (g = 0; g < group_info->ngroups; g++) 209 seq_put_decimal_ull(m, g ? " " : "", 210 from_kgid_munged(user_ns, group_info->gid[g])); 211 put_cred(cred); 212 /* Trailing space shouldn't have been added in the first place. */ 213 seq_putc(m, ' '); 214 215 #ifdef CONFIG_PID_NS 216 seq_puts(m, "\nNStgid:"); 217 for (g = ns->level; g <= pid->level; g++) 218 seq_put_decimal_ull(m, "\t", task_tgid_nr_ns(p, pid->numbers[g].ns)); 219 seq_puts(m, "\nNSpid:"); 220 for (g = ns->level; g <= pid->level; g++) 221 seq_put_decimal_ull(m, "\t", task_pid_nr_ns(p, pid->numbers[g].ns)); 222 seq_puts(m, "\nNSpgid:"); 223 for (g = ns->level; g <= pid->level; g++) 224 seq_put_decimal_ull(m, "\t", task_pgrp_nr_ns(p, pid->numbers[g].ns)); 225 seq_puts(m, "\nNSsid:"); 226 for (g = ns->level; g <= pid->level; g++) 227 seq_put_decimal_ull(m, "\t", task_session_nr_ns(p, pid->numbers[g].ns)); 228 #endif 229 seq_putc(m, '\n'); 230 } 231 232 void render_sigset_t(struct seq_file *m, const char *header, 233 sigset_t *set) 234 { 235 int i; 236 237 seq_puts(m, header); 238 239 i = _NSIG; 240 do { 241 int x = 0; 242 243 i -= 4; 244 if (sigismember(set, i+1)) x |= 1; 245 if (sigismember(set, i+2)) x |= 2; 246 if (sigismember(set, i+3)) x |= 4; 247 if (sigismember(set, i+4)) x |= 8; 248 seq_printf(m, "%x", x); 249 } while (i >= 4); 250 251 seq_putc(m, '\n'); 252 } 253 254 static void collect_sigign_sigcatch(struct task_struct *p, sigset_t *ign, 255 sigset_t *catch) 256 { 257 struct k_sigaction *k; 258 int i; 259 260 k = p->sighand->action; 261 for (i = 1; i <= _NSIG; ++i, ++k) { 262 if (k->sa.sa_handler == SIG_IGN) 263 sigaddset(ign, i); 264 else if (k->sa.sa_handler != SIG_DFL) 265 sigaddset(catch, i); 266 } 267 } 268 269 static inline void task_sig(struct seq_file *m, struct task_struct *p) 270 { 271 unsigned long flags; 272 sigset_t pending, shpending, blocked, ignored, caught; 273 int num_threads = 0; 274 unsigned long qsize = 0; 275 unsigned long qlim = 0; 276 277 sigemptyset(&pending); 278 sigemptyset(&shpending); 279 sigemptyset(&blocked); 280 sigemptyset(&ignored); 281 sigemptyset(&caught); 282 283 if (lock_task_sighand(p, &flags)) { 284 pending = p->pending.signal; 285 shpending = p->signal->shared_pending.signal; 286 blocked = p->blocked; 287 collect_sigign_sigcatch(p, &ignored, &caught); 288 num_threads = get_nr_threads(p); 289 rcu_read_lock(); /* FIXME: is this correct? */ 290 qsize = atomic_read(&__task_cred(p)->user->sigpending); 291 rcu_read_unlock(); 292 qlim = task_rlimit(p, RLIMIT_SIGPENDING); 293 unlock_task_sighand(p, &flags); 294 } 295 296 seq_put_decimal_ull(m, "Threads:\t", num_threads); 297 seq_put_decimal_ull(m, "\nSigQ:\t", qsize); 298 seq_put_decimal_ull(m, "/", qlim); 299 300 /* render them all */ 301 render_sigset_t(m, "\nSigPnd:\t", &pending); 302 render_sigset_t(m, "ShdPnd:\t", &shpending); 303 render_sigset_t(m, "SigBlk:\t", &blocked); 304 render_sigset_t(m, "SigIgn:\t", &ignored); 305 render_sigset_t(m, "SigCgt:\t", &caught); 306 } 307 308 static void render_cap_t(struct seq_file *m, const char *header, 309 kernel_cap_t *a) 310 { 311 unsigned __capi; 312 313 seq_puts(m, header); 314 CAP_FOR_EACH_U32(__capi) { 315 seq_printf(m, "%08x", 316 a->cap[CAP_LAST_U32 - __capi]); 317 } 318 seq_putc(m, '\n'); 319 } 320 321 static inline void task_cap(struct seq_file *m, struct task_struct *p) 322 { 323 const struct cred *cred; 324 kernel_cap_t cap_inheritable, cap_permitted, cap_effective, 325 cap_bset, cap_ambient; 326 327 rcu_read_lock(); 328 cred = __task_cred(p); 329 cap_inheritable = cred->cap_inheritable; 330 cap_permitted = cred->cap_permitted; 331 cap_effective = cred->cap_effective; 332 cap_bset = cred->cap_bset; 333 cap_ambient = cred->cap_ambient; 334 rcu_read_unlock(); 335 336 render_cap_t(m, "CapInh:\t", &cap_inheritable); 337 render_cap_t(m, "CapPrm:\t", &cap_permitted); 338 render_cap_t(m, "CapEff:\t", &cap_effective); 339 render_cap_t(m, "CapBnd:\t", &cap_bset); 340 render_cap_t(m, "CapAmb:\t", &cap_ambient); 341 } 342 343 static inline void task_seccomp(struct seq_file *m, struct task_struct *p) 344 { 345 #ifdef CONFIG_SECCOMP 346 seq_put_decimal_ull(m, "Seccomp:\t", p->seccomp.mode); 347 seq_putc(m, '\n'); 348 #endif 349 } 350 351 static inline void task_context_switch_counts(struct seq_file *m, 352 struct task_struct *p) 353 { 354 seq_put_decimal_ull(m, "voluntary_ctxt_switches:\t", p->nvcsw); 355 seq_put_decimal_ull(m, "\nnonvoluntary_ctxt_switches:\t", p->nivcsw); 356 seq_putc(m, '\n'); 357 } 358 359 static void task_cpus_allowed(struct seq_file *m, struct task_struct *task) 360 { 361 seq_printf(m, "Cpus_allowed:\t%*pb\n", 362 cpumask_pr_args(&task->cpus_allowed)); 363 seq_printf(m, "Cpus_allowed_list:\t%*pbl\n", 364 cpumask_pr_args(&task->cpus_allowed)); 365 } 366 367 int proc_pid_status(struct seq_file *m, struct pid_namespace *ns, 368 struct pid *pid, struct task_struct *task) 369 { 370 struct mm_struct *mm = get_task_mm(task); 371 372 task_name(m, task); 373 task_state(m, ns, pid, task); 374 375 if (mm) { 376 task_mem(m, mm); 377 mmput(mm); 378 } 379 task_sig(m, task); 380 task_cap(m, task); 381 task_seccomp(m, task); 382 task_cpus_allowed(m, task); 383 cpuset_task_status_allowed(m, task); 384 task_context_switch_counts(m, task); 385 return 0; 386 } 387 388 static int do_task_stat(struct seq_file *m, struct pid_namespace *ns, 389 struct pid *pid, struct task_struct *task, int whole) 390 { 391 unsigned long vsize, eip, esp, wchan = 0; 392 int priority, nice; 393 int tty_pgrp = -1, tty_nr = 0; 394 sigset_t sigign, sigcatch; 395 char state; 396 pid_t ppid = 0, pgid = -1, sid = -1; 397 int num_threads = 0; 398 int permitted; 399 struct mm_struct *mm; 400 unsigned long long start_time; 401 unsigned long cmin_flt = 0, cmaj_flt = 0; 402 unsigned long min_flt = 0, maj_flt = 0; 403 cputime_t cutime, cstime, utime, stime; 404 cputime_t cgtime, gtime; 405 unsigned long rsslim = 0; 406 char tcomm[sizeof(task->comm)]; 407 unsigned long flags; 408 409 state = *get_task_state(task); 410 vsize = eip = esp = 0; 411 permitted = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS | PTRACE_MODE_NOAUDIT); 412 mm = get_task_mm(task); 413 if (mm) { 414 vsize = task_vsize(mm); 415 if (permitted) { 416 eip = KSTK_EIP(task); 417 esp = KSTK_ESP(task); 418 } 419 } 420 421 get_task_comm(tcomm, task); 422 423 sigemptyset(&sigign); 424 sigemptyset(&sigcatch); 425 cutime = cstime = utime = stime = 0; 426 cgtime = gtime = 0; 427 428 if (lock_task_sighand(task, &flags)) { 429 struct signal_struct *sig = task->signal; 430 431 if (sig->tty) { 432 struct pid *pgrp = tty_get_pgrp(sig->tty); 433 tty_pgrp = pid_nr_ns(pgrp, ns); 434 put_pid(pgrp); 435 tty_nr = new_encode_dev(tty_devnum(sig->tty)); 436 } 437 438 num_threads = get_nr_threads(task); 439 collect_sigign_sigcatch(task, &sigign, &sigcatch); 440 441 cmin_flt = sig->cmin_flt; 442 cmaj_flt = sig->cmaj_flt; 443 cutime = sig->cutime; 444 cstime = sig->cstime; 445 cgtime = sig->cgtime; 446 rsslim = ACCESS_ONCE(sig->rlim[RLIMIT_RSS].rlim_cur); 447 448 /* add up live thread stats at the group level */ 449 if (whole) { 450 struct task_struct *t = task; 451 do { 452 min_flt += t->min_flt; 453 maj_flt += t->maj_flt; 454 gtime += task_gtime(t); 455 } while_each_thread(task, t); 456 457 min_flt += sig->min_flt; 458 maj_flt += sig->maj_flt; 459 thread_group_cputime_adjusted(task, &utime, &stime); 460 gtime += sig->gtime; 461 } 462 463 sid = task_session_nr_ns(task, ns); 464 ppid = task_tgid_nr_ns(task->real_parent, ns); 465 pgid = task_pgrp_nr_ns(task, ns); 466 467 unlock_task_sighand(task, &flags); 468 } 469 470 if (permitted && (!whole || num_threads < 2)) 471 wchan = get_wchan(task); 472 if (!whole) { 473 min_flt = task->min_flt; 474 maj_flt = task->maj_flt; 475 task_cputime_adjusted(task, &utime, &stime); 476 gtime = task_gtime(task); 477 } 478 479 /* scale priority and nice values from timeslices to -20..20 */ 480 /* to make it look like a "normal" Unix priority/nice value */ 481 priority = task_prio(task); 482 nice = task_nice(task); 483 484 /* convert nsec -> ticks */ 485 start_time = nsec_to_clock_t(task->real_start_time); 486 487 seq_printf(m, "%d (%s) %c", pid_nr_ns(pid, ns), tcomm, state); 488 seq_put_decimal_ll(m, " ", ppid); 489 seq_put_decimal_ll(m, " ", pgid); 490 seq_put_decimal_ll(m, " ", sid); 491 seq_put_decimal_ll(m, " ", tty_nr); 492 seq_put_decimal_ll(m, " ", tty_pgrp); 493 seq_put_decimal_ull(m, " ", task->flags); 494 seq_put_decimal_ull(m, " ", min_flt); 495 seq_put_decimal_ull(m, " ", cmin_flt); 496 seq_put_decimal_ull(m, " ", maj_flt); 497 seq_put_decimal_ull(m, " ", cmaj_flt); 498 seq_put_decimal_ull(m, " ", cputime_to_clock_t(utime)); 499 seq_put_decimal_ull(m, " ", cputime_to_clock_t(stime)); 500 seq_put_decimal_ll(m, " ", cputime_to_clock_t(cutime)); 501 seq_put_decimal_ll(m, " ", cputime_to_clock_t(cstime)); 502 seq_put_decimal_ll(m, " ", priority); 503 seq_put_decimal_ll(m, " ", nice); 504 seq_put_decimal_ll(m, " ", num_threads); 505 seq_put_decimal_ull(m, " ", 0); 506 seq_put_decimal_ull(m, " ", start_time); 507 seq_put_decimal_ull(m, " ", vsize); 508 seq_put_decimal_ull(m, " ", mm ? get_mm_rss(mm) : 0); 509 seq_put_decimal_ull(m, " ", rsslim); 510 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->start_code : 1) : 0); 511 seq_put_decimal_ull(m, " ", mm ? (permitted ? mm->end_code : 1) : 0); 512 seq_put_decimal_ull(m, " ", (permitted && mm) ? mm->start_stack : 0); 513 seq_put_decimal_ull(m, " ", esp); 514 seq_put_decimal_ull(m, " ", eip); 515 /* The signal information here is obsolete. 516 * It must be decimal for Linux 2.0 compatibility. 517 * Use /proc/#/status for real-time signals. 518 */ 519 seq_put_decimal_ull(m, " ", task->pending.signal.sig[0] & 0x7fffffffUL); 520 seq_put_decimal_ull(m, " ", task->blocked.sig[0] & 0x7fffffffUL); 521 seq_put_decimal_ull(m, " ", sigign.sig[0] & 0x7fffffffUL); 522 seq_put_decimal_ull(m, " ", sigcatch.sig[0] & 0x7fffffffUL); 523 524 /* 525 * We used to output the absolute kernel address, but that's an 526 * information leak - so instead we show a 0/1 flag here, to signal 527 * to user-space whether there's a wchan field in /proc/PID/wchan. 528 * 529 * This works with older implementations of procps as well. 530 */ 531 if (wchan) 532 seq_puts(m, " 1"); 533 else 534 seq_puts(m, " 0"); 535 536 seq_put_decimal_ull(m, " ", 0); 537 seq_put_decimal_ull(m, " ", 0); 538 seq_put_decimal_ll(m, " ", task->exit_signal); 539 seq_put_decimal_ll(m, " ", task_cpu(task)); 540 seq_put_decimal_ull(m, " ", task->rt_priority); 541 seq_put_decimal_ull(m, " ", task->policy); 542 seq_put_decimal_ull(m, " ", delayacct_blkio_ticks(task)); 543 seq_put_decimal_ull(m, " ", cputime_to_clock_t(gtime)); 544 seq_put_decimal_ll(m, " ", cputime_to_clock_t(cgtime)); 545 546 if (mm && permitted) { 547 seq_put_decimal_ull(m, " ", mm->start_data); 548 seq_put_decimal_ull(m, " ", mm->end_data); 549 seq_put_decimal_ull(m, " ", mm->start_brk); 550 seq_put_decimal_ull(m, " ", mm->arg_start); 551 seq_put_decimal_ull(m, " ", mm->arg_end); 552 seq_put_decimal_ull(m, " ", mm->env_start); 553 seq_put_decimal_ull(m, " ", mm->env_end); 554 } else 555 seq_puts(m, " 0 0 0 0 0 0 0"); 556 557 if (permitted) 558 seq_put_decimal_ll(m, " ", task->exit_code); 559 else 560 seq_puts(m, " 0"); 561 562 seq_putc(m, '\n'); 563 if (mm) 564 mmput(mm); 565 return 0; 566 } 567 568 int proc_tid_stat(struct seq_file *m, struct pid_namespace *ns, 569 struct pid *pid, struct task_struct *task) 570 { 571 return do_task_stat(m, ns, pid, task, 0); 572 } 573 574 int proc_tgid_stat(struct seq_file *m, struct pid_namespace *ns, 575 struct pid *pid, struct task_struct *task) 576 { 577 return do_task_stat(m, ns, pid, task, 1); 578 } 579 580 int proc_pid_statm(struct seq_file *m, struct pid_namespace *ns, 581 struct pid *pid, struct task_struct *task) 582 { 583 unsigned long size = 0, resident = 0, shared = 0, text = 0, data = 0; 584 struct mm_struct *mm = get_task_mm(task); 585 586 if (mm) { 587 size = task_statm(mm, &shared, &text, &data, &resident); 588 mmput(mm); 589 } 590 /* 591 * For quick read, open code by putting numbers directly 592 * expected format is 593 * seq_printf(m, "%lu %lu %lu %lu 0 %lu 0\n", 594 * size, resident, shared, text, data); 595 */ 596 seq_put_decimal_ull(m, "", size); 597 seq_put_decimal_ull(m, " ", resident); 598 seq_put_decimal_ull(m, " ", shared); 599 seq_put_decimal_ull(m, " ", text); 600 seq_put_decimal_ull(m, " ", 0); 601 seq_put_decimal_ull(m, " ", data); 602 seq_put_decimal_ull(m, " ", 0); 603 seq_putc(m, '\n'); 604 605 return 0; 606 } 607 608 #ifdef CONFIG_PROC_CHILDREN 609 static struct pid * 610 get_children_pid(struct inode *inode, struct pid *pid_prev, loff_t pos) 611 { 612 struct task_struct *start, *task; 613 struct pid *pid = NULL; 614 615 read_lock(&tasklist_lock); 616 617 start = pid_task(proc_pid(inode), PIDTYPE_PID); 618 if (!start) 619 goto out; 620 621 /* 622 * Lets try to continue searching first, this gives 623 * us significant speedup on children-rich processes. 624 */ 625 if (pid_prev) { 626 task = pid_task(pid_prev, PIDTYPE_PID); 627 if (task && task->real_parent == start && 628 !(list_empty(&task->sibling))) { 629 if (list_is_last(&task->sibling, &start->children)) 630 goto out; 631 task = list_first_entry(&task->sibling, 632 struct task_struct, sibling); 633 pid = get_pid(task_pid(task)); 634 goto out; 635 } 636 } 637 638 /* 639 * Slow search case. 640 * 641 * We might miss some children here if children 642 * are exited while we were not holding the lock, 643 * but it was never promised to be accurate that 644 * much. 645 * 646 * "Just suppose that the parent sleeps, but N children 647 * exit after we printed their tids. Now the slow paths 648 * skips N extra children, we miss N tasks." (c) 649 * 650 * So one need to stop or freeze the leader and all 651 * its children to get a precise result. 652 */ 653 list_for_each_entry(task, &start->children, sibling) { 654 if (pos-- == 0) { 655 pid = get_pid(task_pid(task)); 656 break; 657 } 658 } 659 660 out: 661 read_unlock(&tasklist_lock); 662 return pid; 663 } 664 665 static int children_seq_show(struct seq_file *seq, void *v) 666 { 667 struct inode *inode = seq->private; 668 pid_t pid; 669 670 pid = pid_nr_ns(v, inode->i_sb->s_fs_info); 671 seq_printf(seq, "%d ", pid); 672 673 return 0; 674 } 675 676 static void *children_seq_start(struct seq_file *seq, loff_t *pos) 677 { 678 return get_children_pid(seq->private, NULL, *pos); 679 } 680 681 static void *children_seq_next(struct seq_file *seq, void *v, loff_t *pos) 682 { 683 struct pid *pid; 684 685 pid = get_children_pid(seq->private, v, *pos + 1); 686 put_pid(v); 687 688 ++*pos; 689 return pid; 690 } 691 692 static void children_seq_stop(struct seq_file *seq, void *v) 693 { 694 put_pid(v); 695 } 696 697 static const struct seq_operations children_seq_ops = { 698 .start = children_seq_start, 699 .next = children_seq_next, 700 .stop = children_seq_stop, 701 .show = children_seq_show, 702 }; 703 704 static int children_seq_open(struct inode *inode, struct file *file) 705 { 706 struct seq_file *m; 707 int ret; 708 709 ret = seq_open(file, &children_seq_ops); 710 if (ret) 711 return ret; 712 713 m = file->private_data; 714 m->private = inode; 715 716 return ret; 717 } 718 719 int children_seq_release(struct inode *inode, struct file *file) 720 { 721 seq_release(inode, file); 722 return 0; 723 } 724 725 const struct file_operations proc_tid_children_operations = { 726 .open = children_seq_open, 727 .read = seq_read, 728 .llseek = seq_lseek, 729 .release = children_seq_release, 730 }; 731 #endif /* CONFIG_PROC_CHILDREN */ 732