1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/proc/base.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 * 7 * proc base directory handling functions 8 * 9 * 1999, Al Viro. Rewritten. Now it covers the whole per-process part. 10 * Instead of using magical inumbers to determine the kind of object 11 * we allocate and fill in-core inodes upon lookup. They don't even 12 * go into icache. We cache the reference to task_struct upon lookup too. 13 * Eventually it should become a filesystem in its own. We don't use the 14 * rest of procfs anymore. 15 * 16 * 17 * Changelog: 18 * 17-Jan-2005 19 * Allan Bezerra 20 * Bruna Moreira <bruna.moreira@indt.org.br> 21 * Edjard Mota <edjard.mota@indt.org.br> 22 * Ilias Biris <ilias.biris@indt.org.br> 23 * Mauricio Lin <mauricio.lin@indt.org.br> 24 * 25 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT 26 * 27 * A new process specific entry (smaps) included in /proc. It shows the 28 * size of rss for each memory area. The maps entry lacks information 29 * about physical memory size (rss) for each mapped file, i.e., 30 * rss information for executables and library files. 31 * This additional information is useful for any tools that need to know 32 * about physical memory consumption for a process specific library. 33 * 34 * Changelog: 35 * 21-Feb-2005 36 * Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT 37 * Pud inclusion in the page table walking. 38 * 39 * ChangeLog: 40 * 10-Mar-2005 41 * 10LE Instituto Nokia de Tecnologia - INdT: 42 * A better way to walks through the page table as suggested by Hugh Dickins. 43 * 44 * Simo Piiroinen <simo.piiroinen@nokia.com>: 45 * Smaps information related to shared, private, clean and dirty pages. 46 * 47 * Paul Mundt <paul.mundt@nokia.com>: 48 * Overall revision about smaps. 49 */ 50 51 #include <linux/uaccess.h> 52 53 #include <linux/errno.h> 54 #include <linux/time.h> 55 #include <linux/proc_fs.h> 56 #include <linux/stat.h> 57 #include <linux/task_io_accounting_ops.h> 58 #include <linux/init.h> 59 #include <linux/capability.h> 60 #include <linux/file.h> 61 #include <linux/fdtable.h> 62 #include <linux/string.h> 63 #include <linux/seq_file.h> 64 #include <linux/namei.h> 65 #include <linux/mnt_namespace.h> 66 #include <linux/mm.h> 67 #include <linux/swap.h> 68 #include <linux/rcupdate.h> 69 #include <linux/kallsyms.h> 70 #include <linux/stacktrace.h> 71 #include <linux/resource.h> 72 #include <linux/module.h> 73 #include <linux/mount.h> 74 #include <linux/security.h> 75 #include <linux/ptrace.h> 76 #include <linux/tracehook.h> 77 #include <linux/printk.h> 78 #include <linux/cache.h> 79 #include <linux/cgroup.h> 80 #include <linux/cpuset.h> 81 #include <linux/audit.h> 82 #include <linux/poll.h> 83 #include <linux/nsproxy.h> 84 #include <linux/oom.h> 85 #include <linux/elf.h> 86 #include <linux/pid_namespace.h> 87 #include <linux/user_namespace.h> 88 #include <linux/fs_struct.h> 89 #include <linux/slab.h> 90 #include <linux/sched/autogroup.h> 91 #include <linux/sched/mm.h> 92 #include <linux/sched/coredump.h> 93 #include <linux/sched/debug.h> 94 #include <linux/sched/stat.h> 95 #include <linux/flex_array.h> 96 #include <linux/posix-timers.h> 97 #include <trace/events/oom.h> 98 #include "internal.h" 99 #include "fd.h" 100 101 #include "../../lib/kstrtox.h" 102 103 /* NOTE: 104 * Implementing inode permission operations in /proc is almost 105 * certainly an error. Permission checks need to happen during 106 * each system call not at open time. The reason is that most of 107 * what we wish to check for permissions in /proc varies at runtime. 108 * 109 * The classic example of a problem is opening file descriptors 110 * in /proc for a task before it execs a suid executable. 111 */ 112 113 static u8 nlink_tid __ro_after_init; 114 static u8 nlink_tgid __ro_after_init; 115 116 struct pid_entry { 117 const char *name; 118 unsigned int len; 119 umode_t mode; 120 const struct inode_operations *iop; 121 const struct file_operations *fop; 122 union proc_op op; 123 }; 124 125 #define NOD(NAME, MODE, IOP, FOP, OP) { \ 126 .name = (NAME), \ 127 .len = sizeof(NAME) - 1, \ 128 .mode = MODE, \ 129 .iop = IOP, \ 130 .fop = FOP, \ 131 .op = OP, \ 132 } 133 134 #define DIR(NAME, MODE, iops, fops) \ 135 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} ) 136 #define LNK(NAME, get_link) \ 137 NOD(NAME, (S_IFLNK|S_IRWXUGO), \ 138 &proc_pid_link_inode_operations, NULL, \ 139 { .proc_get_link = get_link } ) 140 #define REG(NAME, MODE, fops) \ 141 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {}) 142 #define ONE(NAME, MODE, show) \ 143 NOD(NAME, (S_IFREG|(MODE)), \ 144 NULL, &proc_single_file_operations, \ 145 { .proc_show = show } ) 146 147 /* 148 * Count the number of hardlinks for the pid_entry table, excluding the . 149 * and .. links. 150 */ 151 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries, 152 unsigned int n) 153 { 154 unsigned int i; 155 unsigned int count; 156 157 count = 2; 158 for (i = 0; i < n; ++i) { 159 if (S_ISDIR(entries[i].mode)) 160 ++count; 161 } 162 163 return count; 164 } 165 166 static int get_task_root(struct task_struct *task, struct path *root) 167 { 168 int result = -ENOENT; 169 170 task_lock(task); 171 if (task->fs) { 172 get_fs_root(task->fs, root); 173 result = 0; 174 } 175 task_unlock(task); 176 return result; 177 } 178 179 static int proc_cwd_link(struct dentry *dentry, struct path *path) 180 { 181 struct task_struct *task = get_proc_task(d_inode(dentry)); 182 int result = -ENOENT; 183 184 if (task) { 185 task_lock(task); 186 if (task->fs) { 187 get_fs_pwd(task->fs, path); 188 result = 0; 189 } 190 task_unlock(task); 191 put_task_struct(task); 192 } 193 return result; 194 } 195 196 static int proc_root_link(struct dentry *dentry, struct path *path) 197 { 198 struct task_struct *task = get_proc_task(d_inode(dentry)); 199 int result = -ENOENT; 200 201 if (task) { 202 result = get_task_root(task, path); 203 put_task_struct(task); 204 } 205 return result; 206 } 207 208 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf, 209 size_t _count, loff_t *pos) 210 { 211 struct task_struct *tsk; 212 struct mm_struct *mm; 213 char *page; 214 unsigned long count = _count; 215 unsigned long arg_start, arg_end, env_start, env_end; 216 unsigned long len1, len2, len; 217 unsigned long p; 218 char c; 219 ssize_t rv; 220 221 BUG_ON(*pos < 0); 222 223 tsk = get_proc_task(file_inode(file)); 224 if (!tsk) 225 return -ESRCH; 226 mm = get_task_mm(tsk); 227 put_task_struct(tsk); 228 if (!mm) 229 return 0; 230 /* Check if process spawned far enough to have cmdline. */ 231 if (!mm->env_end) { 232 rv = 0; 233 goto out_mmput; 234 } 235 236 page = (char *)__get_free_page(GFP_KERNEL); 237 if (!page) { 238 rv = -ENOMEM; 239 goto out_mmput; 240 } 241 242 down_read(&mm->mmap_sem); 243 arg_start = mm->arg_start; 244 arg_end = mm->arg_end; 245 env_start = mm->env_start; 246 env_end = mm->env_end; 247 up_read(&mm->mmap_sem); 248 249 BUG_ON(arg_start > arg_end); 250 BUG_ON(env_start > env_end); 251 252 len1 = arg_end - arg_start; 253 len2 = env_end - env_start; 254 255 /* Empty ARGV. */ 256 if (len1 == 0) { 257 rv = 0; 258 goto out_free_page; 259 } 260 /* 261 * Inherently racy -- command line shares address space 262 * with code and data. 263 */ 264 rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0); 265 if (rv <= 0) 266 goto out_free_page; 267 268 rv = 0; 269 270 if (c == '\0') { 271 /* Command line (set of strings) occupies whole ARGV. */ 272 if (len1 <= *pos) 273 goto out_free_page; 274 275 p = arg_start + *pos; 276 len = len1 - *pos; 277 while (count > 0 && len > 0) { 278 unsigned int _count; 279 int nr_read; 280 281 _count = min3(count, len, PAGE_SIZE); 282 nr_read = access_remote_vm(mm, p, page, _count, 0); 283 if (nr_read < 0) 284 rv = nr_read; 285 if (nr_read <= 0) 286 goto out_free_page; 287 288 if (copy_to_user(buf, page, nr_read)) { 289 rv = -EFAULT; 290 goto out_free_page; 291 } 292 293 p += nr_read; 294 len -= nr_read; 295 buf += nr_read; 296 count -= nr_read; 297 rv += nr_read; 298 } 299 } else { 300 /* 301 * Command line (1 string) occupies ARGV and 302 * extends into ENVP. 303 */ 304 struct { 305 unsigned long p; 306 unsigned long len; 307 } cmdline[2] = { 308 { .p = arg_start, .len = len1 }, 309 { .p = env_start, .len = len2 }, 310 }; 311 loff_t pos1 = *pos; 312 unsigned int i; 313 314 i = 0; 315 while (i < 2 && pos1 >= cmdline[i].len) { 316 pos1 -= cmdline[i].len; 317 i++; 318 } 319 while (i < 2) { 320 p = cmdline[i].p + pos1; 321 len = cmdline[i].len - pos1; 322 while (count > 0 && len > 0) { 323 unsigned int _count, l; 324 int nr_read; 325 bool final; 326 327 _count = min3(count, len, PAGE_SIZE); 328 nr_read = access_remote_vm(mm, p, page, _count, 0); 329 if (nr_read < 0) 330 rv = nr_read; 331 if (nr_read <= 0) 332 goto out_free_page; 333 334 /* 335 * Command line can be shorter than whole ARGV 336 * even if last "marker" byte says it is not. 337 */ 338 final = false; 339 l = strnlen(page, nr_read); 340 if (l < nr_read) { 341 nr_read = l; 342 final = true; 343 } 344 345 if (copy_to_user(buf, page, nr_read)) { 346 rv = -EFAULT; 347 goto out_free_page; 348 } 349 350 p += nr_read; 351 len -= nr_read; 352 buf += nr_read; 353 count -= nr_read; 354 rv += nr_read; 355 356 if (final) 357 goto out_free_page; 358 } 359 360 /* Only first chunk can be read partially. */ 361 pos1 = 0; 362 i++; 363 } 364 } 365 366 out_free_page: 367 free_page((unsigned long)page); 368 out_mmput: 369 mmput(mm); 370 if (rv > 0) 371 *pos += rv; 372 return rv; 373 } 374 375 static const struct file_operations proc_pid_cmdline_ops = { 376 .read = proc_pid_cmdline_read, 377 .llseek = generic_file_llseek, 378 }; 379 380 #ifdef CONFIG_KALLSYMS 381 /* 382 * Provides a wchan file via kallsyms in a proper one-value-per-file format. 383 * Returns the resolved symbol. If that fails, simply return the address. 384 */ 385 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns, 386 struct pid *pid, struct task_struct *task) 387 { 388 unsigned long wchan; 389 char symname[KSYM_NAME_LEN]; 390 391 wchan = get_wchan(task); 392 393 if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS) 394 && !lookup_symbol_name(wchan, symname)) 395 seq_printf(m, "%s", symname); 396 else 397 seq_putc(m, '0'); 398 399 return 0; 400 } 401 #endif /* CONFIG_KALLSYMS */ 402 403 static int lock_trace(struct task_struct *task) 404 { 405 int err = mutex_lock_killable(&task->signal->cred_guard_mutex); 406 if (err) 407 return err; 408 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) { 409 mutex_unlock(&task->signal->cred_guard_mutex); 410 return -EPERM; 411 } 412 return 0; 413 } 414 415 static void unlock_trace(struct task_struct *task) 416 { 417 mutex_unlock(&task->signal->cred_guard_mutex); 418 } 419 420 #ifdef CONFIG_STACKTRACE 421 422 #define MAX_STACK_TRACE_DEPTH 64 423 424 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns, 425 struct pid *pid, struct task_struct *task) 426 { 427 struct stack_trace trace; 428 unsigned long *entries; 429 int err; 430 int i; 431 432 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL); 433 if (!entries) 434 return -ENOMEM; 435 436 trace.nr_entries = 0; 437 trace.max_entries = MAX_STACK_TRACE_DEPTH; 438 trace.entries = entries; 439 trace.skip = 0; 440 441 err = lock_trace(task); 442 if (!err) { 443 save_stack_trace_tsk(task, &trace); 444 445 for (i = 0; i < trace.nr_entries; i++) { 446 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]); 447 } 448 unlock_trace(task); 449 } 450 kfree(entries); 451 452 return err; 453 } 454 #endif 455 456 #ifdef CONFIG_SCHED_INFO 457 /* 458 * Provides /proc/PID/schedstat 459 */ 460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns, 461 struct pid *pid, struct task_struct *task) 462 { 463 if (unlikely(!sched_info_on())) 464 seq_printf(m, "0 0 0\n"); 465 else 466 seq_printf(m, "%llu %llu %lu\n", 467 (unsigned long long)task->se.sum_exec_runtime, 468 (unsigned long long)task->sched_info.run_delay, 469 task->sched_info.pcount); 470 471 return 0; 472 } 473 #endif 474 475 #ifdef CONFIG_LATENCYTOP 476 static int lstats_show_proc(struct seq_file *m, void *v) 477 { 478 int i; 479 struct inode *inode = m->private; 480 struct task_struct *task = get_proc_task(inode); 481 482 if (!task) 483 return -ESRCH; 484 seq_puts(m, "Latency Top version : v0.1\n"); 485 for (i = 0; i < 32; i++) { 486 struct latency_record *lr = &task->latency_record[i]; 487 if (lr->backtrace[0]) { 488 int q; 489 seq_printf(m, "%i %li %li", 490 lr->count, lr->time, lr->max); 491 for (q = 0; q < LT_BACKTRACEDEPTH; q++) { 492 unsigned long bt = lr->backtrace[q]; 493 if (!bt) 494 break; 495 if (bt == ULONG_MAX) 496 break; 497 seq_printf(m, " %ps", (void *)bt); 498 } 499 seq_putc(m, '\n'); 500 } 501 502 } 503 put_task_struct(task); 504 return 0; 505 } 506 507 static int lstats_open(struct inode *inode, struct file *file) 508 { 509 return single_open(file, lstats_show_proc, inode); 510 } 511 512 static ssize_t lstats_write(struct file *file, const char __user *buf, 513 size_t count, loff_t *offs) 514 { 515 struct task_struct *task = get_proc_task(file_inode(file)); 516 517 if (!task) 518 return -ESRCH; 519 clear_all_latency_tracing(task); 520 put_task_struct(task); 521 522 return count; 523 } 524 525 static const struct file_operations proc_lstats_operations = { 526 .open = lstats_open, 527 .read = seq_read, 528 .write = lstats_write, 529 .llseek = seq_lseek, 530 .release = single_release, 531 }; 532 533 #endif 534 535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns, 536 struct pid *pid, struct task_struct *task) 537 { 538 unsigned long totalpages = totalram_pages + total_swap_pages; 539 unsigned long points = 0; 540 541 points = oom_badness(task, NULL, NULL, totalpages) * 542 1000 / totalpages; 543 seq_printf(m, "%lu\n", points); 544 545 return 0; 546 } 547 548 struct limit_names { 549 const char *name; 550 const char *unit; 551 }; 552 553 static const struct limit_names lnames[RLIM_NLIMITS] = { 554 [RLIMIT_CPU] = {"Max cpu time", "seconds"}, 555 [RLIMIT_FSIZE] = {"Max file size", "bytes"}, 556 [RLIMIT_DATA] = {"Max data size", "bytes"}, 557 [RLIMIT_STACK] = {"Max stack size", "bytes"}, 558 [RLIMIT_CORE] = {"Max core file size", "bytes"}, 559 [RLIMIT_RSS] = {"Max resident set", "bytes"}, 560 [RLIMIT_NPROC] = {"Max processes", "processes"}, 561 [RLIMIT_NOFILE] = {"Max open files", "files"}, 562 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"}, 563 [RLIMIT_AS] = {"Max address space", "bytes"}, 564 [RLIMIT_LOCKS] = {"Max file locks", "locks"}, 565 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"}, 566 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"}, 567 [RLIMIT_NICE] = {"Max nice priority", NULL}, 568 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL}, 569 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"}, 570 }; 571 572 /* Display limits for a process */ 573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns, 574 struct pid *pid, struct task_struct *task) 575 { 576 unsigned int i; 577 unsigned long flags; 578 579 struct rlimit rlim[RLIM_NLIMITS]; 580 581 if (!lock_task_sighand(task, &flags)) 582 return 0; 583 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS); 584 unlock_task_sighand(task, &flags); 585 586 /* 587 * print the file header 588 */ 589 seq_printf(m, "%-25s %-20s %-20s %-10s\n", 590 "Limit", "Soft Limit", "Hard Limit", "Units"); 591 592 for (i = 0; i < RLIM_NLIMITS; i++) { 593 if (rlim[i].rlim_cur == RLIM_INFINITY) 594 seq_printf(m, "%-25s %-20s ", 595 lnames[i].name, "unlimited"); 596 else 597 seq_printf(m, "%-25s %-20lu ", 598 lnames[i].name, rlim[i].rlim_cur); 599 600 if (rlim[i].rlim_max == RLIM_INFINITY) 601 seq_printf(m, "%-20s ", "unlimited"); 602 else 603 seq_printf(m, "%-20lu ", rlim[i].rlim_max); 604 605 if (lnames[i].unit) 606 seq_printf(m, "%-10s\n", lnames[i].unit); 607 else 608 seq_putc(m, '\n'); 609 } 610 611 return 0; 612 } 613 614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns, 616 struct pid *pid, struct task_struct *task) 617 { 618 long nr; 619 unsigned long args[6], sp, pc; 620 int res; 621 622 res = lock_trace(task); 623 if (res) 624 return res; 625 626 if (task_current_syscall(task, &nr, args, 6, &sp, &pc)) 627 seq_puts(m, "running\n"); 628 else if (nr < 0) 629 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc); 630 else 631 seq_printf(m, 632 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n", 633 nr, 634 args[0], args[1], args[2], args[3], args[4], args[5], 635 sp, pc); 636 unlock_trace(task); 637 638 return 0; 639 } 640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */ 641 642 /************************************************************************/ 643 /* Here the fs part begins */ 644 /************************************************************************/ 645 646 /* permission checks */ 647 static int proc_fd_access_allowed(struct inode *inode) 648 { 649 struct task_struct *task; 650 int allowed = 0; 651 /* Allow access to a task's file descriptors if it is us or we 652 * may use ptrace attach to the process and find out that 653 * information. 654 */ 655 task = get_proc_task(inode); 656 if (task) { 657 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS); 658 put_task_struct(task); 659 } 660 return allowed; 661 } 662 663 int proc_setattr(struct dentry *dentry, struct iattr *attr) 664 { 665 int error; 666 struct inode *inode = d_inode(dentry); 667 668 if (attr->ia_valid & ATTR_MODE) 669 return -EPERM; 670 671 error = setattr_prepare(dentry, attr); 672 if (error) 673 return error; 674 675 setattr_copy(inode, attr); 676 mark_inode_dirty(inode); 677 return 0; 678 } 679 680 /* 681 * May current process learn task's sched/cmdline info (for hide_pid_min=1) 682 * or euid/egid (for hide_pid_min=2)? 683 */ 684 static bool has_pid_permissions(struct pid_namespace *pid, 685 struct task_struct *task, 686 int hide_pid_min) 687 { 688 if (pid->hide_pid < hide_pid_min) 689 return true; 690 if (in_group_p(pid->pid_gid)) 691 return true; 692 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS); 693 } 694 695 696 static int proc_pid_permission(struct inode *inode, int mask) 697 { 698 struct pid_namespace *pid = inode->i_sb->s_fs_info; 699 struct task_struct *task; 700 bool has_perms; 701 702 task = get_proc_task(inode); 703 if (!task) 704 return -ESRCH; 705 has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS); 706 put_task_struct(task); 707 708 if (!has_perms) { 709 if (pid->hide_pid == HIDEPID_INVISIBLE) { 710 /* 711 * Let's make getdents(), stat(), and open() 712 * consistent with each other. If a process 713 * may not stat() a file, it shouldn't be seen 714 * in procfs at all. 715 */ 716 return -ENOENT; 717 } 718 719 return -EPERM; 720 } 721 return generic_permission(inode, mask); 722 } 723 724 725 726 static const struct inode_operations proc_def_inode_operations = { 727 .setattr = proc_setattr, 728 }; 729 730 static int proc_single_show(struct seq_file *m, void *v) 731 { 732 struct inode *inode = m->private; 733 struct pid_namespace *ns; 734 struct pid *pid; 735 struct task_struct *task; 736 int ret; 737 738 ns = inode->i_sb->s_fs_info; 739 pid = proc_pid(inode); 740 task = get_pid_task(pid, PIDTYPE_PID); 741 if (!task) 742 return -ESRCH; 743 744 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task); 745 746 put_task_struct(task); 747 return ret; 748 } 749 750 static int proc_single_open(struct inode *inode, struct file *filp) 751 { 752 return single_open(filp, proc_single_show, inode); 753 } 754 755 static const struct file_operations proc_single_file_operations = { 756 .open = proc_single_open, 757 .read = seq_read, 758 .llseek = seq_lseek, 759 .release = single_release, 760 }; 761 762 763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode) 764 { 765 struct task_struct *task = get_proc_task(inode); 766 struct mm_struct *mm = ERR_PTR(-ESRCH); 767 768 if (task) { 769 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS); 770 put_task_struct(task); 771 772 if (!IS_ERR_OR_NULL(mm)) { 773 /* ensure this mm_struct can't be freed */ 774 mmgrab(mm); 775 /* but do not pin its memory */ 776 mmput(mm); 777 } 778 } 779 780 return mm; 781 } 782 783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode) 784 { 785 struct mm_struct *mm = proc_mem_open(inode, mode); 786 787 if (IS_ERR(mm)) 788 return PTR_ERR(mm); 789 790 file->private_data = mm; 791 return 0; 792 } 793 794 static int mem_open(struct inode *inode, struct file *file) 795 { 796 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH); 797 798 /* OK to pass negative loff_t, we can catch out-of-range */ 799 file->f_mode |= FMODE_UNSIGNED_OFFSET; 800 801 return ret; 802 } 803 804 static ssize_t mem_rw(struct file *file, char __user *buf, 805 size_t count, loff_t *ppos, int write) 806 { 807 struct mm_struct *mm = file->private_data; 808 unsigned long addr = *ppos; 809 ssize_t copied; 810 char *page; 811 unsigned int flags; 812 813 if (!mm) 814 return 0; 815 816 page = (char *)__get_free_page(GFP_KERNEL); 817 if (!page) 818 return -ENOMEM; 819 820 copied = 0; 821 if (!mmget_not_zero(mm)) 822 goto free; 823 824 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0); 825 826 while (count > 0) { 827 int this_len = min_t(int, count, PAGE_SIZE); 828 829 if (write && copy_from_user(page, buf, this_len)) { 830 copied = -EFAULT; 831 break; 832 } 833 834 this_len = access_remote_vm(mm, addr, page, this_len, flags); 835 if (!this_len) { 836 if (!copied) 837 copied = -EIO; 838 break; 839 } 840 841 if (!write && copy_to_user(buf, page, this_len)) { 842 copied = -EFAULT; 843 break; 844 } 845 846 buf += this_len; 847 addr += this_len; 848 copied += this_len; 849 count -= this_len; 850 } 851 *ppos = addr; 852 853 mmput(mm); 854 free: 855 free_page((unsigned long) page); 856 return copied; 857 } 858 859 static ssize_t mem_read(struct file *file, char __user *buf, 860 size_t count, loff_t *ppos) 861 { 862 return mem_rw(file, buf, count, ppos, 0); 863 } 864 865 static ssize_t mem_write(struct file *file, const char __user *buf, 866 size_t count, loff_t *ppos) 867 { 868 return mem_rw(file, (char __user*)buf, count, ppos, 1); 869 } 870 871 loff_t mem_lseek(struct file *file, loff_t offset, int orig) 872 { 873 switch (orig) { 874 case 0: 875 file->f_pos = offset; 876 break; 877 case 1: 878 file->f_pos += offset; 879 break; 880 default: 881 return -EINVAL; 882 } 883 force_successful_syscall_return(); 884 return file->f_pos; 885 } 886 887 static int mem_release(struct inode *inode, struct file *file) 888 { 889 struct mm_struct *mm = file->private_data; 890 if (mm) 891 mmdrop(mm); 892 return 0; 893 } 894 895 static const struct file_operations proc_mem_operations = { 896 .llseek = mem_lseek, 897 .read = mem_read, 898 .write = mem_write, 899 .open = mem_open, 900 .release = mem_release, 901 }; 902 903 static int environ_open(struct inode *inode, struct file *file) 904 { 905 return __mem_open(inode, file, PTRACE_MODE_READ); 906 } 907 908 static ssize_t environ_read(struct file *file, char __user *buf, 909 size_t count, loff_t *ppos) 910 { 911 char *page; 912 unsigned long src = *ppos; 913 int ret = 0; 914 struct mm_struct *mm = file->private_data; 915 unsigned long env_start, env_end; 916 917 /* Ensure the process spawned far enough to have an environment. */ 918 if (!mm || !mm->env_end) 919 return 0; 920 921 page = (char *)__get_free_page(GFP_KERNEL); 922 if (!page) 923 return -ENOMEM; 924 925 ret = 0; 926 if (!mmget_not_zero(mm)) 927 goto free; 928 929 down_read(&mm->mmap_sem); 930 env_start = mm->env_start; 931 env_end = mm->env_end; 932 up_read(&mm->mmap_sem); 933 934 while (count > 0) { 935 size_t this_len, max_len; 936 int retval; 937 938 if (src >= (env_end - env_start)) 939 break; 940 941 this_len = env_end - (env_start + src); 942 943 max_len = min_t(size_t, PAGE_SIZE, count); 944 this_len = min(max_len, this_len); 945 946 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0); 947 948 if (retval <= 0) { 949 ret = retval; 950 break; 951 } 952 953 if (copy_to_user(buf, page, retval)) { 954 ret = -EFAULT; 955 break; 956 } 957 958 ret += retval; 959 src += retval; 960 buf += retval; 961 count -= retval; 962 } 963 *ppos = src; 964 mmput(mm); 965 966 free: 967 free_page((unsigned long) page); 968 return ret; 969 } 970 971 static const struct file_operations proc_environ_operations = { 972 .open = environ_open, 973 .read = environ_read, 974 .llseek = generic_file_llseek, 975 .release = mem_release, 976 }; 977 978 static int auxv_open(struct inode *inode, struct file *file) 979 { 980 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS); 981 } 982 983 static ssize_t auxv_read(struct file *file, char __user *buf, 984 size_t count, loff_t *ppos) 985 { 986 struct mm_struct *mm = file->private_data; 987 unsigned int nwords = 0; 988 989 if (!mm) 990 return 0; 991 do { 992 nwords += 2; 993 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */ 994 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv, 995 nwords * sizeof(mm->saved_auxv[0])); 996 } 997 998 static const struct file_operations proc_auxv_operations = { 999 .open = auxv_open, 1000 .read = auxv_read, 1001 .llseek = generic_file_llseek, 1002 .release = mem_release, 1003 }; 1004 1005 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count, 1006 loff_t *ppos) 1007 { 1008 struct task_struct *task = get_proc_task(file_inode(file)); 1009 char buffer[PROC_NUMBUF]; 1010 int oom_adj = OOM_ADJUST_MIN; 1011 size_t len; 1012 1013 if (!task) 1014 return -ESRCH; 1015 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX) 1016 oom_adj = OOM_ADJUST_MAX; 1017 else 1018 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) / 1019 OOM_SCORE_ADJ_MAX; 1020 put_task_struct(task); 1021 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj); 1022 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1023 } 1024 1025 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy) 1026 { 1027 static DEFINE_MUTEX(oom_adj_mutex); 1028 struct mm_struct *mm = NULL; 1029 struct task_struct *task; 1030 int err = 0; 1031 1032 task = get_proc_task(file_inode(file)); 1033 if (!task) 1034 return -ESRCH; 1035 1036 mutex_lock(&oom_adj_mutex); 1037 if (legacy) { 1038 if (oom_adj < task->signal->oom_score_adj && 1039 !capable(CAP_SYS_RESOURCE)) { 1040 err = -EACCES; 1041 goto err_unlock; 1042 } 1043 /* 1044 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use 1045 * /proc/pid/oom_score_adj instead. 1046 */ 1047 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n", 1048 current->comm, task_pid_nr(current), task_pid_nr(task), 1049 task_pid_nr(task)); 1050 } else { 1051 if ((short)oom_adj < task->signal->oom_score_adj_min && 1052 !capable(CAP_SYS_RESOURCE)) { 1053 err = -EACCES; 1054 goto err_unlock; 1055 } 1056 } 1057 1058 /* 1059 * Make sure we will check other processes sharing the mm if this is 1060 * not vfrok which wants its own oom_score_adj. 1061 * pin the mm so it doesn't go away and get reused after task_unlock 1062 */ 1063 if (!task->vfork_done) { 1064 struct task_struct *p = find_lock_task_mm(task); 1065 1066 if (p) { 1067 if (atomic_read(&p->mm->mm_users) > 1) { 1068 mm = p->mm; 1069 mmgrab(mm); 1070 } 1071 task_unlock(p); 1072 } 1073 } 1074 1075 task->signal->oom_score_adj = oom_adj; 1076 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE)) 1077 task->signal->oom_score_adj_min = (short)oom_adj; 1078 trace_oom_score_adj_update(task); 1079 1080 if (mm) { 1081 struct task_struct *p; 1082 1083 rcu_read_lock(); 1084 for_each_process(p) { 1085 if (same_thread_group(task, p)) 1086 continue; 1087 1088 /* do not touch kernel threads or the global init */ 1089 if (p->flags & PF_KTHREAD || is_global_init(p)) 1090 continue; 1091 1092 task_lock(p); 1093 if (!p->vfork_done && process_shares_mm(p, mm)) { 1094 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n", 1095 task_pid_nr(p), p->comm, 1096 p->signal->oom_score_adj, oom_adj, 1097 task_pid_nr(task), task->comm); 1098 p->signal->oom_score_adj = oom_adj; 1099 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE)) 1100 p->signal->oom_score_adj_min = (short)oom_adj; 1101 } 1102 task_unlock(p); 1103 } 1104 rcu_read_unlock(); 1105 mmdrop(mm); 1106 } 1107 err_unlock: 1108 mutex_unlock(&oom_adj_mutex); 1109 put_task_struct(task); 1110 return err; 1111 } 1112 1113 /* 1114 * /proc/pid/oom_adj exists solely for backwards compatibility with previous 1115 * kernels. The effective policy is defined by oom_score_adj, which has a 1116 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly. 1117 * Values written to oom_adj are simply mapped linearly to oom_score_adj. 1118 * Processes that become oom disabled via oom_adj will still be oom disabled 1119 * with this implementation. 1120 * 1121 * oom_adj cannot be removed since existing userspace binaries use it. 1122 */ 1123 static ssize_t oom_adj_write(struct file *file, const char __user *buf, 1124 size_t count, loff_t *ppos) 1125 { 1126 char buffer[PROC_NUMBUF]; 1127 int oom_adj; 1128 int err; 1129 1130 memset(buffer, 0, sizeof(buffer)); 1131 if (count > sizeof(buffer) - 1) 1132 count = sizeof(buffer) - 1; 1133 if (copy_from_user(buffer, buf, count)) { 1134 err = -EFAULT; 1135 goto out; 1136 } 1137 1138 err = kstrtoint(strstrip(buffer), 0, &oom_adj); 1139 if (err) 1140 goto out; 1141 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) && 1142 oom_adj != OOM_DISABLE) { 1143 err = -EINVAL; 1144 goto out; 1145 } 1146 1147 /* 1148 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum 1149 * value is always attainable. 1150 */ 1151 if (oom_adj == OOM_ADJUST_MAX) 1152 oom_adj = OOM_SCORE_ADJ_MAX; 1153 else 1154 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE; 1155 1156 err = __set_oom_adj(file, oom_adj, true); 1157 out: 1158 return err < 0 ? err : count; 1159 } 1160 1161 static const struct file_operations proc_oom_adj_operations = { 1162 .read = oom_adj_read, 1163 .write = oom_adj_write, 1164 .llseek = generic_file_llseek, 1165 }; 1166 1167 static ssize_t oom_score_adj_read(struct file *file, char __user *buf, 1168 size_t count, loff_t *ppos) 1169 { 1170 struct task_struct *task = get_proc_task(file_inode(file)); 1171 char buffer[PROC_NUMBUF]; 1172 short oom_score_adj = OOM_SCORE_ADJ_MIN; 1173 size_t len; 1174 1175 if (!task) 1176 return -ESRCH; 1177 oom_score_adj = task->signal->oom_score_adj; 1178 put_task_struct(task); 1179 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj); 1180 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1181 } 1182 1183 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf, 1184 size_t count, loff_t *ppos) 1185 { 1186 char buffer[PROC_NUMBUF]; 1187 int oom_score_adj; 1188 int err; 1189 1190 memset(buffer, 0, sizeof(buffer)); 1191 if (count > sizeof(buffer) - 1) 1192 count = sizeof(buffer) - 1; 1193 if (copy_from_user(buffer, buf, count)) { 1194 err = -EFAULT; 1195 goto out; 1196 } 1197 1198 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj); 1199 if (err) 1200 goto out; 1201 if (oom_score_adj < OOM_SCORE_ADJ_MIN || 1202 oom_score_adj > OOM_SCORE_ADJ_MAX) { 1203 err = -EINVAL; 1204 goto out; 1205 } 1206 1207 err = __set_oom_adj(file, oom_score_adj, false); 1208 out: 1209 return err < 0 ? err : count; 1210 } 1211 1212 static const struct file_operations proc_oom_score_adj_operations = { 1213 .read = oom_score_adj_read, 1214 .write = oom_score_adj_write, 1215 .llseek = default_llseek, 1216 }; 1217 1218 #ifdef CONFIG_AUDITSYSCALL 1219 #define TMPBUFLEN 11 1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf, 1221 size_t count, loff_t *ppos) 1222 { 1223 struct inode * inode = file_inode(file); 1224 struct task_struct *task = get_proc_task(inode); 1225 ssize_t length; 1226 char tmpbuf[TMPBUFLEN]; 1227 1228 if (!task) 1229 return -ESRCH; 1230 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 1231 from_kuid(file->f_cred->user_ns, 1232 audit_get_loginuid(task))); 1233 put_task_struct(task); 1234 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 1235 } 1236 1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf, 1238 size_t count, loff_t *ppos) 1239 { 1240 struct inode * inode = file_inode(file); 1241 uid_t loginuid; 1242 kuid_t kloginuid; 1243 int rv; 1244 1245 rcu_read_lock(); 1246 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) { 1247 rcu_read_unlock(); 1248 return -EPERM; 1249 } 1250 rcu_read_unlock(); 1251 1252 if (*ppos != 0) { 1253 /* No partial writes. */ 1254 return -EINVAL; 1255 } 1256 1257 rv = kstrtou32_from_user(buf, count, 10, &loginuid); 1258 if (rv < 0) 1259 return rv; 1260 1261 /* is userspace tring to explicitly UNSET the loginuid? */ 1262 if (loginuid == AUDIT_UID_UNSET) { 1263 kloginuid = INVALID_UID; 1264 } else { 1265 kloginuid = make_kuid(file->f_cred->user_ns, loginuid); 1266 if (!uid_valid(kloginuid)) 1267 return -EINVAL; 1268 } 1269 1270 rv = audit_set_loginuid(kloginuid); 1271 if (rv < 0) 1272 return rv; 1273 return count; 1274 } 1275 1276 static const struct file_operations proc_loginuid_operations = { 1277 .read = proc_loginuid_read, 1278 .write = proc_loginuid_write, 1279 .llseek = generic_file_llseek, 1280 }; 1281 1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf, 1283 size_t count, loff_t *ppos) 1284 { 1285 struct inode * inode = file_inode(file); 1286 struct task_struct *task = get_proc_task(inode); 1287 ssize_t length; 1288 char tmpbuf[TMPBUFLEN]; 1289 1290 if (!task) 1291 return -ESRCH; 1292 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 1293 audit_get_sessionid(task)); 1294 put_task_struct(task); 1295 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 1296 } 1297 1298 static const struct file_operations proc_sessionid_operations = { 1299 .read = proc_sessionid_read, 1300 .llseek = generic_file_llseek, 1301 }; 1302 #endif 1303 1304 #ifdef CONFIG_FAULT_INJECTION 1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf, 1306 size_t count, loff_t *ppos) 1307 { 1308 struct task_struct *task = get_proc_task(file_inode(file)); 1309 char buffer[PROC_NUMBUF]; 1310 size_t len; 1311 int make_it_fail; 1312 1313 if (!task) 1314 return -ESRCH; 1315 make_it_fail = task->make_it_fail; 1316 put_task_struct(task); 1317 1318 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail); 1319 1320 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1321 } 1322 1323 static ssize_t proc_fault_inject_write(struct file * file, 1324 const char __user * buf, size_t count, loff_t *ppos) 1325 { 1326 struct task_struct *task; 1327 char buffer[PROC_NUMBUF]; 1328 int make_it_fail; 1329 int rv; 1330 1331 if (!capable(CAP_SYS_RESOURCE)) 1332 return -EPERM; 1333 memset(buffer, 0, sizeof(buffer)); 1334 if (count > sizeof(buffer) - 1) 1335 count = sizeof(buffer) - 1; 1336 if (copy_from_user(buffer, buf, count)) 1337 return -EFAULT; 1338 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail); 1339 if (rv < 0) 1340 return rv; 1341 if (make_it_fail < 0 || make_it_fail > 1) 1342 return -EINVAL; 1343 1344 task = get_proc_task(file_inode(file)); 1345 if (!task) 1346 return -ESRCH; 1347 task->make_it_fail = make_it_fail; 1348 put_task_struct(task); 1349 1350 return count; 1351 } 1352 1353 static const struct file_operations proc_fault_inject_operations = { 1354 .read = proc_fault_inject_read, 1355 .write = proc_fault_inject_write, 1356 .llseek = generic_file_llseek, 1357 }; 1358 1359 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf, 1360 size_t count, loff_t *ppos) 1361 { 1362 struct task_struct *task; 1363 int err; 1364 unsigned int n; 1365 1366 err = kstrtouint_from_user(buf, count, 0, &n); 1367 if (err) 1368 return err; 1369 1370 task = get_proc_task(file_inode(file)); 1371 if (!task) 1372 return -ESRCH; 1373 task->fail_nth = n; 1374 put_task_struct(task); 1375 1376 return count; 1377 } 1378 1379 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf, 1380 size_t count, loff_t *ppos) 1381 { 1382 struct task_struct *task; 1383 char numbuf[PROC_NUMBUF]; 1384 ssize_t len; 1385 1386 task = get_proc_task(file_inode(file)); 1387 if (!task) 1388 return -ESRCH; 1389 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth); 1390 len = simple_read_from_buffer(buf, count, ppos, numbuf, len); 1391 put_task_struct(task); 1392 1393 return len; 1394 } 1395 1396 static const struct file_operations proc_fail_nth_operations = { 1397 .read = proc_fail_nth_read, 1398 .write = proc_fail_nth_write, 1399 }; 1400 #endif 1401 1402 1403 #ifdef CONFIG_SCHED_DEBUG 1404 /* 1405 * Print out various scheduling related per-task fields: 1406 */ 1407 static int sched_show(struct seq_file *m, void *v) 1408 { 1409 struct inode *inode = m->private; 1410 struct pid_namespace *ns = inode->i_sb->s_fs_info; 1411 struct task_struct *p; 1412 1413 p = get_proc_task(inode); 1414 if (!p) 1415 return -ESRCH; 1416 proc_sched_show_task(p, ns, m); 1417 1418 put_task_struct(p); 1419 1420 return 0; 1421 } 1422 1423 static ssize_t 1424 sched_write(struct file *file, const char __user *buf, 1425 size_t count, loff_t *offset) 1426 { 1427 struct inode *inode = file_inode(file); 1428 struct task_struct *p; 1429 1430 p = get_proc_task(inode); 1431 if (!p) 1432 return -ESRCH; 1433 proc_sched_set_task(p); 1434 1435 put_task_struct(p); 1436 1437 return count; 1438 } 1439 1440 static int sched_open(struct inode *inode, struct file *filp) 1441 { 1442 return single_open(filp, sched_show, inode); 1443 } 1444 1445 static const struct file_operations proc_pid_sched_operations = { 1446 .open = sched_open, 1447 .read = seq_read, 1448 .write = sched_write, 1449 .llseek = seq_lseek, 1450 .release = single_release, 1451 }; 1452 1453 #endif 1454 1455 #ifdef CONFIG_SCHED_AUTOGROUP 1456 /* 1457 * Print out autogroup related information: 1458 */ 1459 static int sched_autogroup_show(struct seq_file *m, void *v) 1460 { 1461 struct inode *inode = m->private; 1462 struct task_struct *p; 1463 1464 p = get_proc_task(inode); 1465 if (!p) 1466 return -ESRCH; 1467 proc_sched_autogroup_show_task(p, m); 1468 1469 put_task_struct(p); 1470 1471 return 0; 1472 } 1473 1474 static ssize_t 1475 sched_autogroup_write(struct file *file, const char __user *buf, 1476 size_t count, loff_t *offset) 1477 { 1478 struct inode *inode = file_inode(file); 1479 struct task_struct *p; 1480 char buffer[PROC_NUMBUF]; 1481 int nice; 1482 int err; 1483 1484 memset(buffer, 0, sizeof(buffer)); 1485 if (count > sizeof(buffer) - 1) 1486 count = sizeof(buffer) - 1; 1487 if (copy_from_user(buffer, buf, count)) 1488 return -EFAULT; 1489 1490 err = kstrtoint(strstrip(buffer), 0, &nice); 1491 if (err < 0) 1492 return err; 1493 1494 p = get_proc_task(inode); 1495 if (!p) 1496 return -ESRCH; 1497 1498 err = proc_sched_autogroup_set_nice(p, nice); 1499 if (err) 1500 count = err; 1501 1502 put_task_struct(p); 1503 1504 return count; 1505 } 1506 1507 static int sched_autogroup_open(struct inode *inode, struct file *filp) 1508 { 1509 int ret; 1510 1511 ret = single_open(filp, sched_autogroup_show, NULL); 1512 if (!ret) { 1513 struct seq_file *m = filp->private_data; 1514 1515 m->private = inode; 1516 } 1517 return ret; 1518 } 1519 1520 static const struct file_operations proc_pid_sched_autogroup_operations = { 1521 .open = sched_autogroup_open, 1522 .read = seq_read, 1523 .write = sched_autogroup_write, 1524 .llseek = seq_lseek, 1525 .release = single_release, 1526 }; 1527 1528 #endif /* CONFIG_SCHED_AUTOGROUP */ 1529 1530 static ssize_t comm_write(struct file *file, const char __user *buf, 1531 size_t count, loff_t *offset) 1532 { 1533 struct inode *inode = file_inode(file); 1534 struct task_struct *p; 1535 char buffer[TASK_COMM_LEN]; 1536 const size_t maxlen = sizeof(buffer) - 1; 1537 1538 memset(buffer, 0, sizeof(buffer)); 1539 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count)) 1540 return -EFAULT; 1541 1542 p = get_proc_task(inode); 1543 if (!p) 1544 return -ESRCH; 1545 1546 if (same_thread_group(current, p)) 1547 set_task_comm(p, buffer); 1548 else 1549 count = -EINVAL; 1550 1551 put_task_struct(p); 1552 1553 return count; 1554 } 1555 1556 static int comm_show(struct seq_file *m, void *v) 1557 { 1558 struct inode *inode = m->private; 1559 struct task_struct *p; 1560 1561 p = get_proc_task(inode); 1562 if (!p) 1563 return -ESRCH; 1564 1565 task_lock(p); 1566 seq_printf(m, "%s\n", p->comm); 1567 task_unlock(p); 1568 1569 put_task_struct(p); 1570 1571 return 0; 1572 } 1573 1574 static int comm_open(struct inode *inode, struct file *filp) 1575 { 1576 return single_open(filp, comm_show, inode); 1577 } 1578 1579 static const struct file_operations proc_pid_set_comm_operations = { 1580 .open = comm_open, 1581 .read = seq_read, 1582 .write = comm_write, 1583 .llseek = seq_lseek, 1584 .release = single_release, 1585 }; 1586 1587 static int proc_exe_link(struct dentry *dentry, struct path *exe_path) 1588 { 1589 struct task_struct *task; 1590 struct file *exe_file; 1591 1592 task = get_proc_task(d_inode(dentry)); 1593 if (!task) 1594 return -ENOENT; 1595 exe_file = get_task_exe_file(task); 1596 put_task_struct(task); 1597 if (exe_file) { 1598 *exe_path = exe_file->f_path; 1599 path_get(&exe_file->f_path); 1600 fput(exe_file); 1601 return 0; 1602 } else 1603 return -ENOENT; 1604 } 1605 1606 static const char *proc_pid_get_link(struct dentry *dentry, 1607 struct inode *inode, 1608 struct delayed_call *done) 1609 { 1610 struct path path; 1611 int error = -EACCES; 1612 1613 if (!dentry) 1614 return ERR_PTR(-ECHILD); 1615 1616 /* Are we allowed to snoop on the tasks file descriptors? */ 1617 if (!proc_fd_access_allowed(inode)) 1618 goto out; 1619 1620 error = PROC_I(inode)->op.proc_get_link(dentry, &path); 1621 if (error) 1622 goto out; 1623 1624 nd_jump_link(&path); 1625 return NULL; 1626 out: 1627 return ERR_PTR(error); 1628 } 1629 1630 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen) 1631 { 1632 char *tmp = (char *)__get_free_page(GFP_KERNEL); 1633 char *pathname; 1634 int len; 1635 1636 if (!tmp) 1637 return -ENOMEM; 1638 1639 pathname = d_path(path, tmp, PAGE_SIZE); 1640 len = PTR_ERR(pathname); 1641 if (IS_ERR(pathname)) 1642 goto out; 1643 len = tmp + PAGE_SIZE - 1 - pathname; 1644 1645 if (len > buflen) 1646 len = buflen; 1647 if (copy_to_user(buffer, pathname, len)) 1648 len = -EFAULT; 1649 out: 1650 free_page((unsigned long)tmp); 1651 return len; 1652 } 1653 1654 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen) 1655 { 1656 int error = -EACCES; 1657 struct inode *inode = d_inode(dentry); 1658 struct path path; 1659 1660 /* Are we allowed to snoop on the tasks file descriptors? */ 1661 if (!proc_fd_access_allowed(inode)) 1662 goto out; 1663 1664 error = PROC_I(inode)->op.proc_get_link(dentry, &path); 1665 if (error) 1666 goto out; 1667 1668 error = do_proc_readlink(&path, buffer, buflen); 1669 path_put(&path); 1670 out: 1671 return error; 1672 } 1673 1674 const struct inode_operations proc_pid_link_inode_operations = { 1675 .readlink = proc_pid_readlink, 1676 .get_link = proc_pid_get_link, 1677 .setattr = proc_setattr, 1678 }; 1679 1680 1681 /* building an inode */ 1682 1683 void task_dump_owner(struct task_struct *task, umode_t mode, 1684 kuid_t *ruid, kgid_t *rgid) 1685 { 1686 /* Depending on the state of dumpable compute who should own a 1687 * proc file for a task. 1688 */ 1689 const struct cred *cred; 1690 kuid_t uid; 1691 kgid_t gid; 1692 1693 /* Default to the tasks effective ownership */ 1694 rcu_read_lock(); 1695 cred = __task_cred(task); 1696 uid = cred->euid; 1697 gid = cred->egid; 1698 rcu_read_unlock(); 1699 1700 /* 1701 * Before the /proc/pid/status file was created the only way to read 1702 * the effective uid of a /process was to stat /proc/pid. Reading 1703 * /proc/pid/status is slow enough that procps and other packages 1704 * kept stating /proc/pid. To keep the rules in /proc simple I have 1705 * made this apply to all per process world readable and executable 1706 * directories. 1707 */ 1708 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) { 1709 struct mm_struct *mm; 1710 task_lock(task); 1711 mm = task->mm; 1712 /* Make non-dumpable tasks owned by some root */ 1713 if (mm) { 1714 if (get_dumpable(mm) != SUID_DUMP_USER) { 1715 struct user_namespace *user_ns = mm->user_ns; 1716 1717 uid = make_kuid(user_ns, 0); 1718 if (!uid_valid(uid)) 1719 uid = GLOBAL_ROOT_UID; 1720 1721 gid = make_kgid(user_ns, 0); 1722 if (!gid_valid(gid)) 1723 gid = GLOBAL_ROOT_GID; 1724 } 1725 } else { 1726 uid = GLOBAL_ROOT_UID; 1727 gid = GLOBAL_ROOT_GID; 1728 } 1729 task_unlock(task); 1730 } 1731 *ruid = uid; 1732 *rgid = gid; 1733 } 1734 1735 struct inode *proc_pid_make_inode(struct super_block * sb, 1736 struct task_struct *task, umode_t mode) 1737 { 1738 struct inode * inode; 1739 struct proc_inode *ei; 1740 1741 /* We need a new inode */ 1742 1743 inode = new_inode(sb); 1744 if (!inode) 1745 goto out; 1746 1747 /* Common stuff */ 1748 ei = PROC_I(inode); 1749 inode->i_mode = mode; 1750 inode->i_ino = get_next_ino(); 1751 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); 1752 inode->i_op = &proc_def_inode_operations; 1753 1754 /* 1755 * grab the reference to task. 1756 */ 1757 ei->pid = get_task_pid(task, PIDTYPE_PID); 1758 if (!ei->pid) 1759 goto out_unlock; 1760 1761 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid); 1762 security_task_to_inode(task, inode); 1763 1764 out: 1765 return inode; 1766 1767 out_unlock: 1768 iput(inode); 1769 return NULL; 1770 } 1771 1772 int pid_getattr(const struct path *path, struct kstat *stat, 1773 u32 request_mask, unsigned int query_flags) 1774 { 1775 struct inode *inode = d_inode(path->dentry); 1776 struct task_struct *task; 1777 struct pid_namespace *pid = path->dentry->d_sb->s_fs_info; 1778 1779 generic_fillattr(inode, stat); 1780 1781 rcu_read_lock(); 1782 stat->uid = GLOBAL_ROOT_UID; 1783 stat->gid = GLOBAL_ROOT_GID; 1784 task = pid_task(proc_pid(inode), PIDTYPE_PID); 1785 if (task) { 1786 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) { 1787 rcu_read_unlock(); 1788 /* 1789 * This doesn't prevent learning whether PID exists, 1790 * it only makes getattr() consistent with readdir(). 1791 */ 1792 return -ENOENT; 1793 } 1794 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid); 1795 } 1796 rcu_read_unlock(); 1797 return 0; 1798 } 1799 1800 /* dentry stuff */ 1801 1802 /* 1803 * Exceptional case: normally we are not allowed to unhash a busy 1804 * directory. In this case, however, we can do it - no aliasing problems 1805 * due to the way we treat inodes. 1806 * 1807 * Rewrite the inode's ownerships here because the owning task may have 1808 * performed a setuid(), etc. 1809 * 1810 */ 1811 int pid_revalidate(struct dentry *dentry, unsigned int flags) 1812 { 1813 struct inode *inode; 1814 struct task_struct *task; 1815 1816 if (flags & LOOKUP_RCU) 1817 return -ECHILD; 1818 1819 inode = d_inode(dentry); 1820 task = get_proc_task(inode); 1821 1822 if (task) { 1823 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid); 1824 1825 inode->i_mode &= ~(S_ISUID | S_ISGID); 1826 security_task_to_inode(task, inode); 1827 put_task_struct(task); 1828 return 1; 1829 } 1830 return 0; 1831 } 1832 1833 static inline bool proc_inode_is_dead(struct inode *inode) 1834 { 1835 return !proc_pid(inode)->tasks[PIDTYPE_PID].first; 1836 } 1837 1838 int pid_delete_dentry(const struct dentry *dentry) 1839 { 1840 /* Is the task we represent dead? 1841 * If so, then don't put the dentry on the lru list, 1842 * kill it immediately. 1843 */ 1844 return proc_inode_is_dead(d_inode(dentry)); 1845 } 1846 1847 const struct dentry_operations pid_dentry_operations = 1848 { 1849 .d_revalidate = pid_revalidate, 1850 .d_delete = pid_delete_dentry, 1851 }; 1852 1853 /* Lookups */ 1854 1855 /* 1856 * Fill a directory entry. 1857 * 1858 * If possible create the dcache entry and derive our inode number and 1859 * file type from dcache entry. 1860 * 1861 * Since all of the proc inode numbers are dynamically generated, the inode 1862 * numbers do not exist until the inode is cache. This means creating the 1863 * the dcache entry in readdir is necessary to keep the inode numbers 1864 * reported by readdir in sync with the inode numbers reported 1865 * by stat. 1866 */ 1867 bool proc_fill_cache(struct file *file, struct dir_context *ctx, 1868 const char *name, int len, 1869 instantiate_t instantiate, struct task_struct *task, const void *ptr) 1870 { 1871 struct dentry *child, *dir = file->f_path.dentry; 1872 struct qstr qname = QSTR_INIT(name, len); 1873 struct inode *inode; 1874 unsigned type; 1875 ino_t ino; 1876 1877 child = d_hash_and_lookup(dir, &qname); 1878 if (!child) { 1879 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1880 child = d_alloc_parallel(dir, &qname, &wq); 1881 if (IS_ERR(child)) 1882 goto end_instantiate; 1883 if (d_in_lookup(child)) { 1884 int err = instantiate(d_inode(dir), child, task, ptr); 1885 d_lookup_done(child); 1886 if (err < 0) { 1887 dput(child); 1888 goto end_instantiate; 1889 } 1890 } 1891 } 1892 inode = d_inode(child); 1893 ino = inode->i_ino; 1894 type = inode->i_mode >> 12; 1895 dput(child); 1896 return dir_emit(ctx, name, len, ino, type); 1897 1898 end_instantiate: 1899 return dir_emit(ctx, name, len, 1, DT_UNKNOWN); 1900 } 1901 1902 /* 1903 * dname_to_vma_addr - maps a dentry name into two unsigned longs 1904 * which represent vma start and end addresses. 1905 */ 1906 static int dname_to_vma_addr(struct dentry *dentry, 1907 unsigned long *start, unsigned long *end) 1908 { 1909 const char *str = dentry->d_name.name; 1910 unsigned long long sval, eval; 1911 unsigned int len; 1912 1913 len = _parse_integer(str, 16, &sval); 1914 if (len & KSTRTOX_OVERFLOW) 1915 return -EINVAL; 1916 if (sval != (unsigned long)sval) 1917 return -EINVAL; 1918 str += len; 1919 1920 if (*str != '-') 1921 return -EINVAL; 1922 str++; 1923 1924 len = _parse_integer(str, 16, &eval); 1925 if (len & KSTRTOX_OVERFLOW) 1926 return -EINVAL; 1927 if (eval != (unsigned long)eval) 1928 return -EINVAL; 1929 str += len; 1930 1931 if (*str != '\0') 1932 return -EINVAL; 1933 1934 *start = sval; 1935 *end = eval; 1936 1937 return 0; 1938 } 1939 1940 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags) 1941 { 1942 unsigned long vm_start, vm_end; 1943 bool exact_vma_exists = false; 1944 struct mm_struct *mm = NULL; 1945 struct task_struct *task; 1946 struct inode *inode; 1947 int status = 0; 1948 1949 if (flags & LOOKUP_RCU) 1950 return -ECHILD; 1951 1952 inode = d_inode(dentry); 1953 task = get_proc_task(inode); 1954 if (!task) 1955 goto out_notask; 1956 1957 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS); 1958 if (IS_ERR_OR_NULL(mm)) 1959 goto out; 1960 1961 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) { 1962 down_read(&mm->mmap_sem); 1963 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end); 1964 up_read(&mm->mmap_sem); 1965 } 1966 1967 mmput(mm); 1968 1969 if (exact_vma_exists) { 1970 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid); 1971 1972 security_task_to_inode(task, inode); 1973 status = 1; 1974 } 1975 1976 out: 1977 put_task_struct(task); 1978 1979 out_notask: 1980 return status; 1981 } 1982 1983 static const struct dentry_operations tid_map_files_dentry_operations = { 1984 .d_revalidate = map_files_d_revalidate, 1985 .d_delete = pid_delete_dentry, 1986 }; 1987 1988 static int map_files_get_link(struct dentry *dentry, struct path *path) 1989 { 1990 unsigned long vm_start, vm_end; 1991 struct vm_area_struct *vma; 1992 struct task_struct *task; 1993 struct mm_struct *mm; 1994 int rc; 1995 1996 rc = -ENOENT; 1997 task = get_proc_task(d_inode(dentry)); 1998 if (!task) 1999 goto out; 2000 2001 mm = get_task_mm(task); 2002 put_task_struct(task); 2003 if (!mm) 2004 goto out; 2005 2006 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end); 2007 if (rc) 2008 goto out_mmput; 2009 2010 rc = -ENOENT; 2011 down_read(&mm->mmap_sem); 2012 vma = find_exact_vma(mm, vm_start, vm_end); 2013 if (vma && vma->vm_file) { 2014 *path = vma->vm_file->f_path; 2015 path_get(path); 2016 rc = 0; 2017 } 2018 up_read(&mm->mmap_sem); 2019 2020 out_mmput: 2021 mmput(mm); 2022 out: 2023 return rc; 2024 } 2025 2026 struct map_files_info { 2027 unsigned long start; 2028 unsigned long end; 2029 fmode_t mode; 2030 }; 2031 2032 /* 2033 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the 2034 * symlinks may be used to bypass permissions on ancestor directories in the 2035 * path to the file in question. 2036 */ 2037 static const char * 2038 proc_map_files_get_link(struct dentry *dentry, 2039 struct inode *inode, 2040 struct delayed_call *done) 2041 { 2042 if (!capable(CAP_SYS_ADMIN)) 2043 return ERR_PTR(-EPERM); 2044 2045 return proc_pid_get_link(dentry, inode, done); 2046 } 2047 2048 /* 2049 * Identical to proc_pid_link_inode_operations except for get_link() 2050 */ 2051 static const struct inode_operations proc_map_files_link_inode_operations = { 2052 .readlink = proc_pid_readlink, 2053 .get_link = proc_map_files_get_link, 2054 .setattr = proc_setattr, 2055 }; 2056 2057 static int 2058 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry, 2059 struct task_struct *task, const void *ptr) 2060 { 2061 fmode_t mode = (fmode_t)(unsigned long)ptr; 2062 struct proc_inode *ei; 2063 struct inode *inode; 2064 2065 inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK | 2066 ((mode & FMODE_READ ) ? S_IRUSR : 0) | 2067 ((mode & FMODE_WRITE) ? S_IWUSR : 0)); 2068 if (!inode) 2069 return -ENOENT; 2070 2071 ei = PROC_I(inode); 2072 ei->op.proc_get_link = map_files_get_link; 2073 2074 inode->i_op = &proc_map_files_link_inode_operations; 2075 inode->i_size = 64; 2076 2077 d_set_d_op(dentry, &tid_map_files_dentry_operations); 2078 d_add(dentry, inode); 2079 2080 return 0; 2081 } 2082 2083 static struct dentry *proc_map_files_lookup(struct inode *dir, 2084 struct dentry *dentry, unsigned int flags) 2085 { 2086 unsigned long vm_start, vm_end; 2087 struct vm_area_struct *vma; 2088 struct task_struct *task; 2089 int result; 2090 struct mm_struct *mm; 2091 2092 result = -ENOENT; 2093 task = get_proc_task(dir); 2094 if (!task) 2095 goto out; 2096 2097 result = -EACCES; 2098 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) 2099 goto out_put_task; 2100 2101 result = -ENOENT; 2102 if (dname_to_vma_addr(dentry, &vm_start, &vm_end)) 2103 goto out_put_task; 2104 2105 mm = get_task_mm(task); 2106 if (!mm) 2107 goto out_put_task; 2108 2109 down_read(&mm->mmap_sem); 2110 vma = find_exact_vma(mm, vm_start, vm_end); 2111 if (!vma) 2112 goto out_no_vma; 2113 2114 if (vma->vm_file) 2115 result = proc_map_files_instantiate(dir, dentry, task, 2116 (void *)(unsigned long)vma->vm_file->f_mode); 2117 2118 out_no_vma: 2119 up_read(&mm->mmap_sem); 2120 mmput(mm); 2121 out_put_task: 2122 put_task_struct(task); 2123 out: 2124 return ERR_PTR(result); 2125 } 2126 2127 static const struct inode_operations proc_map_files_inode_operations = { 2128 .lookup = proc_map_files_lookup, 2129 .permission = proc_fd_permission, 2130 .setattr = proc_setattr, 2131 }; 2132 2133 static int 2134 proc_map_files_readdir(struct file *file, struct dir_context *ctx) 2135 { 2136 struct vm_area_struct *vma; 2137 struct task_struct *task; 2138 struct mm_struct *mm; 2139 unsigned long nr_files, pos, i; 2140 struct flex_array *fa = NULL; 2141 struct map_files_info info; 2142 struct map_files_info *p; 2143 int ret; 2144 2145 ret = -ENOENT; 2146 task = get_proc_task(file_inode(file)); 2147 if (!task) 2148 goto out; 2149 2150 ret = -EACCES; 2151 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) 2152 goto out_put_task; 2153 2154 ret = 0; 2155 if (!dir_emit_dots(file, ctx)) 2156 goto out_put_task; 2157 2158 mm = get_task_mm(task); 2159 if (!mm) 2160 goto out_put_task; 2161 down_read(&mm->mmap_sem); 2162 2163 nr_files = 0; 2164 2165 /* 2166 * We need two passes here: 2167 * 2168 * 1) Collect vmas of mapped files with mmap_sem taken 2169 * 2) Release mmap_sem and instantiate entries 2170 * 2171 * otherwise we get lockdep complained, since filldir() 2172 * routine might require mmap_sem taken in might_fault(). 2173 */ 2174 2175 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) { 2176 if (vma->vm_file && ++pos > ctx->pos) 2177 nr_files++; 2178 } 2179 2180 if (nr_files) { 2181 fa = flex_array_alloc(sizeof(info), nr_files, 2182 GFP_KERNEL); 2183 if (!fa || flex_array_prealloc(fa, 0, nr_files, 2184 GFP_KERNEL)) { 2185 ret = -ENOMEM; 2186 if (fa) 2187 flex_array_free(fa); 2188 up_read(&mm->mmap_sem); 2189 mmput(mm); 2190 goto out_put_task; 2191 } 2192 for (i = 0, vma = mm->mmap, pos = 2; vma; 2193 vma = vma->vm_next) { 2194 if (!vma->vm_file) 2195 continue; 2196 if (++pos <= ctx->pos) 2197 continue; 2198 2199 info.start = vma->vm_start; 2200 info.end = vma->vm_end; 2201 info.mode = vma->vm_file->f_mode; 2202 if (flex_array_put(fa, i++, &info, GFP_KERNEL)) 2203 BUG(); 2204 } 2205 } 2206 up_read(&mm->mmap_sem); 2207 2208 for (i = 0; i < nr_files; i++) { 2209 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */ 2210 unsigned int len; 2211 2212 p = flex_array_get(fa, i); 2213 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end); 2214 if (!proc_fill_cache(file, ctx, 2215 buf, len, 2216 proc_map_files_instantiate, 2217 task, 2218 (void *)(unsigned long)p->mode)) 2219 break; 2220 ctx->pos++; 2221 } 2222 if (fa) 2223 flex_array_free(fa); 2224 mmput(mm); 2225 2226 out_put_task: 2227 put_task_struct(task); 2228 out: 2229 return ret; 2230 } 2231 2232 static const struct file_operations proc_map_files_operations = { 2233 .read = generic_read_dir, 2234 .iterate_shared = proc_map_files_readdir, 2235 .llseek = generic_file_llseek, 2236 }; 2237 2238 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS) 2239 struct timers_private { 2240 struct pid *pid; 2241 struct task_struct *task; 2242 struct sighand_struct *sighand; 2243 struct pid_namespace *ns; 2244 unsigned long flags; 2245 }; 2246 2247 static void *timers_start(struct seq_file *m, loff_t *pos) 2248 { 2249 struct timers_private *tp = m->private; 2250 2251 tp->task = get_pid_task(tp->pid, PIDTYPE_PID); 2252 if (!tp->task) 2253 return ERR_PTR(-ESRCH); 2254 2255 tp->sighand = lock_task_sighand(tp->task, &tp->flags); 2256 if (!tp->sighand) 2257 return ERR_PTR(-ESRCH); 2258 2259 return seq_list_start(&tp->task->signal->posix_timers, *pos); 2260 } 2261 2262 static void *timers_next(struct seq_file *m, void *v, loff_t *pos) 2263 { 2264 struct timers_private *tp = m->private; 2265 return seq_list_next(v, &tp->task->signal->posix_timers, pos); 2266 } 2267 2268 static void timers_stop(struct seq_file *m, void *v) 2269 { 2270 struct timers_private *tp = m->private; 2271 2272 if (tp->sighand) { 2273 unlock_task_sighand(tp->task, &tp->flags); 2274 tp->sighand = NULL; 2275 } 2276 2277 if (tp->task) { 2278 put_task_struct(tp->task); 2279 tp->task = NULL; 2280 } 2281 } 2282 2283 static int show_timer(struct seq_file *m, void *v) 2284 { 2285 struct k_itimer *timer; 2286 struct timers_private *tp = m->private; 2287 int notify; 2288 static const char * const nstr[] = { 2289 [SIGEV_SIGNAL] = "signal", 2290 [SIGEV_NONE] = "none", 2291 [SIGEV_THREAD] = "thread", 2292 }; 2293 2294 timer = list_entry((struct list_head *)v, struct k_itimer, list); 2295 notify = timer->it_sigev_notify; 2296 2297 seq_printf(m, "ID: %d\n", timer->it_id); 2298 seq_printf(m, "signal: %d/%px\n", 2299 timer->sigq->info.si_signo, 2300 timer->sigq->info.si_value.sival_ptr); 2301 seq_printf(m, "notify: %s/%s.%d\n", 2302 nstr[notify & ~SIGEV_THREAD_ID], 2303 (notify & SIGEV_THREAD_ID) ? "tid" : "pid", 2304 pid_nr_ns(timer->it_pid, tp->ns)); 2305 seq_printf(m, "ClockID: %d\n", timer->it_clock); 2306 2307 return 0; 2308 } 2309 2310 static const struct seq_operations proc_timers_seq_ops = { 2311 .start = timers_start, 2312 .next = timers_next, 2313 .stop = timers_stop, 2314 .show = show_timer, 2315 }; 2316 2317 static int proc_timers_open(struct inode *inode, struct file *file) 2318 { 2319 struct timers_private *tp; 2320 2321 tp = __seq_open_private(file, &proc_timers_seq_ops, 2322 sizeof(struct timers_private)); 2323 if (!tp) 2324 return -ENOMEM; 2325 2326 tp->pid = proc_pid(inode); 2327 tp->ns = inode->i_sb->s_fs_info; 2328 return 0; 2329 } 2330 2331 static const struct file_operations proc_timers_operations = { 2332 .open = proc_timers_open, 2333 .read = seq_read, 2334 .llseek = seq_lseek, 2335 .release = seq_release_private, 2336 }; 2337 #endif 2338 2339 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf, 2340 size_t count, loff_t *offset) 2341 { 2342 struct inode *inode = file_inode(file); 2343 struct task_struct *p; 2344 u64 slack_ns; 2345 int err; 2346 2347 err = kstrtoull_from_user(buf, count, 10, &slack_ns); 2348 if (err < 0) 2349 return err; 2350 2351 p = get_proc_task(inode); 2352 if (!p) 2353 return -ESRCH; 2354 2355 if (p != current) { 2356 if (!capable(CAP_SYS_NICE)) { 2357 count = -EPERM; 2358 goto out; 2359 } 2360 2361 err = security_task_setscheduler(p); 2362 if (err) { 2363 count = err; 2364 goto out; 2365 } 2366 } 2367 2368 task_lock(p); 2369 if (slack_ns == 0) 2370 p->timer_slack_ns = p->default_timer_slack_ns; 2371 else 2372 p->timer_slack_ns = slack_ns; 2373 task_unlock(p); 2374 2375 out: 2376 put_task_struct(p); 2377 2378 return count; 2379 } 2380 2381 static int timerslack_ns_show(struct seq_file *m, void *v) 2382 { 2383 struct inode *inode = m->private; 2384 struct task_struct *p; 2385 int err = 0; 2386 2387 p = get_proc_task(inode); 2388 if (!p) 2389 return -ESRCH; 2390 2391 if (p != current) { 2392 2393 if (!capable(CAP_SYS_NICE)) { 2394 err = -EPERM; 2395 goto out; 2396 } 2397 err = security_task_getscheduler(p); 2398 if (err) 2399 goto out; 2400 } 2401 2402 task_lock(p); 2403 seq_printf(m, "%llu\n", p->timer_slack_ns); 2404 task_unlock(p); 2405 2406 out: 2407 put_task_struct(p); 2408 2409 return err; 2410 } 2411 2412 static int timerslack_ns_open(struct inode *inode, struct file *filp) 2413 { 2414 return single_open(filp, timerslack_ns_show, inode); 2415 } 2416 2417 static const struct file_operations proc_pid_set_timerslack_ns_operations = { 2418 .open = timerslack_ns_open, 2419 .read = seq_read, 2420 .write = timerslack_ns_write, 2421 .llseek = seq_lseek, 2422 .release = single_release, 2423 }; 2424 2425 static int proc_pident_instantiate(struct inode *dir, 2426 struct dentry *dentry, struct task_struct *task, const void *ptr) 2427 { 2428 const struct pid_entry *p = ptr; 2429 struct inode *inode; 2430 struct proc_inode *ei; 2431 2432 inode = proc_pid_make_inode(dir->i_sb, task, p->mode); 2433 if (!inode) 2434 goto out; 2435 2436 ei = PROC_I(inode); 2437 if (S_ISDIR(inode->i_mode)) 2438 set_nlink(inode, 2); /* Use getattr to fix if necessary */ 2439 if (p->iop) 2440 inode->i_op = p->iop; 2441 if (p->fop) 2442 inode->i_fop = p->fop; 2443 ei->op = p->op; 2444 d_set_d_op(dentry, &pid_dentry_operations); 2445 d_add(dentry, inode); 2446 /* Close the race of the process dying before we return the dentry */ 2447 if (pid_revalidate(dentry, 0)) 2448 return 0; 2449 out: 2450 return -ENOENT; 2451 } 2452 2453 static struct dentry *proc_pident_lookup(struct inode *dir, 2454 struct dentry *dentry, 2455 const struct pid_entry *ents, 2456 unsigned int nents) 2457 { 2458 int error; 2459 struct task_struct *task = get_proc_task(dir); 2460 const struct pid_entry *p, *last; 2461 2462 error = -ENOENT; 2463 2464 if (!task) 2465 goto out_no_task; 2466 2467 /* 2468 * Yes, it does not scale. And it should not. Don't add 2469 * new entries into /proc/<tgid>/ without very good reasons. 2470 */ 2471 last = &ents[nents]; 2472 for (p = ents; p < last; p++) { 2473 if (p->len != dentry->d_name.len) 2474 continue; 2475 if (!memcmp(dentry->d_name.name, p->name, p->len)) 2476 break; 2477 } 2478 if (p >= last) 2479 goto out; 2480 2481 error = proc_pident_instantiate(dir, dentry, task, p); 2482 out: 2483 put_task_struct(task); 2484 out_no_task: 2485 return ERR_PTR(error); 2486 } 2487 2488 static int proc_pident_readdir(struct file *file, struct dir_context *ctx, 2489 const struct pid_entry *ents, unsigned int nents) 2490 { 2491 struct task_struct *task = get_proc_task(file_inode(file)); 2492 const struct pid_entry *p; 2493 2494 if (!task) 2495 return -ENOENT; 2496 2497 if (!dir_emit_dots(file, ctx)) 2498 goto out; 2499 2500 if (ctx->pos >= nents + 2) 2501 goto out; 2502 2503 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) { 2504 if (!proc_fill_cache(file, ctx, p->name, p->len, 2505 proc_pident_instantiate, task, p)) 2506 break; 2507 ctx->pos++; 2508 } 2509 out: 2510 put_task_struct(task); 2511 return 0; 2512 } 2513 2514 #ifdef CONFIG_SECURITY 2515 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf, 2516 size_t count, loff_t *ppos) 2517 { 2518 struct inode * inode = file_inode(file); 2519 char *p = NULL; 2520 ssize_t length; 2521 struct task_struct *task = get_proc_task(inode); 2522 2523 if (!task) 2524 return -ESRCH; 2525 2526 length = security_getprocattr(task, 2527 (char*)file->f_path.dentry->d_name.name, 2528 &p); 2529 put_task_struct(task); 2530 if (length > 0) 2531 length = simple_read_from_buffer(buf, count, ppos, p, length); 2532 kfree(p); 2533 return length; 2534 } 2535 2536 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf, 2537 size_t count, loff_t *ppos) 2538 { 2539 struct inode * inode = file_inode(file); 2540 void *page; 2541 ssize_t length; 2542 struct task_struct *task = get_proc_task(inode); 2543 2544 length = -ESRCH; 2545 if (!task) 2546 goto out_no_task; 2547 2548 /* A task may only write its own attributes. */ 2549 length = -EACCES; 2550 if (current != task) 2551 goto out; 2552 2553 if (count > PAGE_SIZE) 2554 count = PAGE_SIZE; 2555 2556 /* No partial writes. */ 2557 length = -EINVAL; 2558 if (*ppos != 0) 2559 goto out; 2560 2561 page = memdup_user(buf, count); 2562 if (IS_ERR(page)) { 2563 length = PTR_ERR(page); 2564 goto out; 2565 } 2566 2567 /* Guard against adverse ptrace interaction */ 2568 length = mutex_lock_interruptible(¤t->signal->cred_guard_mutex); 2569 if (length < 0) 2570 goto out_free; 2571 2572 length = security_setprocattr(file->f_path.dentry->d_name.name, 2573 page, count); 2574 mutex_unlock(¤t->signal->cred_guard_mutex); 2575 out_free: 2576 kfree(page); 2577 out: 2578 put_task_struct(task); 2579 out_no_task: 2580 return length; 2581 } 2582 2583 static const struct file_operations proc_pid_attr_operations = { 2584 .read = proc_pid_attr_read, 2585 .write = proc_pid_attr_write, 2586 .llseek = generic_file_llseek, 2587 }; 2588 2589 static const struct pid_entry attr_dir_stuff[] = { 2590 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2591 REG("prev", S_IRUGO, proc_pid_attr_operations), 2592 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2593 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2594 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2595 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2596 }; 2597 2598 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx) 2599 { 2600 return proc_pident_readdir(file, ctx, 2601 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff)); 2602 } 2603 2604 static const struct file_operations proc_attr_dir_operations = { 2605 .read = generic_read_dir, 2606 .iterate_shared = proc_attr_dir_readdir, 2607 .llseek = generic_file_llseek, 2608 }; 2609 2610 static struct dentry *proc_attr_dir_lookup(struct inode *dir, 2611 struct dentry *dentry, unsigned int flags) 2612 { 2613 return proc_pident_lookup(dir, dentry, 2614 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff)); 2615 } 2616 2617 static const struct inode_operations proc_attr_dir_inode_operations = { 2618 .lookup = proc_attr_dir_lookup, 2619 .getattr = pid_getattr, 2620 .setattr = proc_setattr, 2621 }; 2622 2623 #endif 2624 2625 #ifdef CONFIG_ELF_CORE 2626 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf, 2627 size_t count, loff_t *ppos) 2628 { 2629 struct task_struct *task = get_proc_task(file_inode(file)); 2630 struct mm_struct *mm; 2631 char buffer[PROC_NUMBUF]; 2632 size_t len; 2633 int ret; 2634 2635 if (!task) 2636 return -ESRCH; 2637 2638 ret = 0; 2639 mm = get_task_mm(task); 2640 if (mm) { 2641 len = snprintf(buffer, sizeof(buffer), "%08lx\n", 2642 ((mm->flags & MMF_DUMP_FILTER_MASK) >> 2643 MMF_DUMP_FILTER_SHIFT)); 2644 mmput(mm); 2645 ret = simple_read_from_buffer(buf, count, ppos, buffer, len); 2646 } 2647 2648 put_task_struct(task); 2649 2650 return ret; 2651 } 2652 2653 static ssize_t proc_coredump_filter_write(struct file *file, 2654 const char __user *buf, 2655 size_t count, 2656 loff_t *ppos) 2657 { 2658 struct task_struct *task; 2659 struct mm_struct *mm; 2660 unsigned int val; 2661 int ret; 2662 int i; 2663 unsigned long mask; 2664 2665 ret = kstrtouint_from_user(buf, count, 0, &val); 2666 if (ret < 0) 2667 return ret; 2668 2669 ret = -ESRCH; 2670 task = get_proc_task(file_inode(file)); 2671 if (!task) 2672 goto out_no_task; 2673 2674 mm = get_task_mm(task); 2675 if (!mm) 2676 goto out_no_mm; 2677 ret = 0; 2678 2679 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) { 2680 if (val & mask) 2681 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2682 else 2683 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2684 } 2685 2686 mmput(mm); 2687 out_no_mm: 2688 put_task_struct(task); 2689 out_no_task: 2690 if (ret < 0) 2691 return ret; 2692 return count; 2693 } 2694 2695 static const struct file_operations proc_coredump_filter_operations = { 2696 .read = proc_coredump_filter_read, 2697 .write = proc_coredump_filter_write, 2698 .llseek = generic_file_llseek, 2699 }; 2700 #endif 2701 2702 #ifdef CONFIG_TASK_IO_ACCOUNTING 2703 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole) 2704 { 2705 struct task_io_accounting acct = task->ioac; 2706 unsigned long flags; 2707 int result; 2708 2709 result = mutex_lock_killable(&task->signal->cred_guard_mutex); 2710 if (result) 2711 return result; 2712 2713 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) { 2714 result = -EACCES; 2715 goto out_unlock; 2716 } 2717 2718 if (whole && lock_task_sighand(task, &flags)) { 2719 struct task_struct *t = task; 2720 2721 task_io_accounting_add(&acct, &task->signal->ioac); 2722 while_each_thread(task, t) 2723 task_io_accounting_add(&acct, &t->ioac); 2724 2725 unlock_task_sighand(task, &flags); 2726 } 2727 seq_printf(m, 2728 "rchar: %llu\n" 2729 "wchar: %llu\n" 2730 "syscr: %llu\n" 2731 "syscw: %llu\n" 2732 "read_bytes: %llu\n" 2733 "write_bytes: %llu\n" 2734 "cancelled_write_bytes: %llu\n", 2735 (unsigned long long)acct.rchar, 2736 (unsigned long long)acct.wchar, 2737 (unsigned long long)acct.syscr, 2738 (unsigned long long)acct.syscw, 2739 (unsigned long long)acct.read_bytes, 2740 (unsigned long long)acct.write_bytes, 2741 (unsigned long long)acct.cancelled_write_bytes); 2742 result = 0; 2743 2744 out_unlock: 2745 mutex_unlock(&task->signal->cred_guard_mutex); 2746 return result; 2747 } 2748 2749 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns, 2750 struct pid *pid, struct task_struct *task) 2751 { 2752 return do_io_accounting(task, m, 0); 2753 } 2754 2755 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns, 2756 struct pid *pid, struct task_struct *task) 2757 { 2758 return do_io_accounting(task, m, 1); 2759 } 2760 #endif /* CONFIG_TASK_IO_ACCOUNTING */ 2761 2762 #ifdef CONFIG_USER_NS 2763 static int proc_id_map_open(struct inode *inode, struct file *file, 2764 const struct seq_operations *seq_ops) 2765 { 2766 struct user_namespace *ns = NULL; 2767 struct task_struct *task; 2768 struct seq_file *seq; 2769 int ret = -EINVAL; 2770 2771 task = get_proc_task(inode); 2772 if (task) { 2773 rcu_read_lock(); 2774 ns = get_user_ns(task_cred_xxx(task, user_ns)); 2775 rcu_read_unlock(); 2776 put_task_struct(task); 2777 } 2778 if (!ns) 2779 goto err; 2780 2781 ret = seq_open(file, seq_ops); 2782 if (ret) 2783 goto err_put_ns; 2784 2785 seq = file->private_data; 2786 seq->private = ns; 2787 2788 return 0; 2789 err_put_ns: 2790 put_user_ns(ns); 2791 err: 2792 return ret; 2793 } 2794 2795 static int proc_id_map_release(struct inode *inode, struct file *file) 2796 { 2797 struct seq_file *seq = file->private_data; 2798 struct user_namespace *ns = seq->private; 2799 put_user_ns(ns); 2800 return seq_release(inode, file); 2801 } 2802 2803 static int proc_uid_map_open(struct inode *inode, struct file *file) 2804 { 2805 return proc_id_map_open(inode, file, &proc_uid_seq_operations); 2806 } 2807 2808 static int proc_gid_map_open(struct inode *inode, struct file *file) 2809 { 2810 return proc_id_map_open(inode, file, &proc_gid_seq_operations); 2811 } 2812 2813 static int proc_projid_map_open(struct inode *inode, struct file *file) 2814 { 2815 return proc_id_map_open(inode, file, &proc_projid_seq_operations); 2816 } 2817 2818 static const struct file_operations proc_uid_map_operations = { 2819 .open = proc_uid_map_open, 2820 .write = proc_uid_map_write, 2821 .read = seq_read, 2822 .llseek = seq_lseek, 2823 .release = proc_id_map_release, 2824 }; 2825 2826 static const struct file_operations proc_gid_map_operations = { 2827 .open = proc_gid_map_open, 2828 .write = proc_gid_map_write, 2829 .read = seq_read, 2830 .llseek = seq_lseek, 2831 .release = proc_id_map_release, 2832 }; 2833 2834 static const struct file_operations proc_projid_map_operations = { 2835 .open = proc_projid_map_open, 2836 .write = proc_projid_map_write, 2837 .read = seq_read, 2838 .llseek = seq_lseek, 2839 .release = proc_id_map_release, 2840 }; 2841 2842 static int proc_setgroups_open(struct inode *inode, struct file *file) 2843 { 2844 struct user_namespace *ns = NULL; 2845 struct task_struct *task; 2846 int ret; 2847 2848 ret = -ESRCH; 2849 task = get_proc_task(inode); 2850 if (task) { 2851 rcu_read_lock(); 2852 ns = get_user_ns(task_cred_xxx(task, user_ns)); 2853 rcu_read_unlock(); 2854 put_task_struct(task); 2855 } 2856 if (!ns) 2857 goto err; 2858 2859 if (file->f_mode & FMODE_WRITE) { 2860 ret = -EACCES; 2861 if (!ns_capable(ns, CAP_SYS_ADMIN)) 2862 goto err_put_ns; 2863 } 2864 2865 ret = single_open(file, &proc_setgroups_show, ns); 2866 if (ret) 2867 goto err_put_ns; 2868 2869 return 0; 2870 err_put_ns: 2871 put_user_ns(ns); 2872 err: 2873 return ret; 2874 } 2875 2876 static int proc_setgroups_release(struct inode *inode, struct file *file) 2877 { 2878 struct seq_file *seq = file->private_data; 2879 struct user_namespace *ns = seq->private; 2880 int ret = single_release(inode, file); 2881 put_user_ns(ns); 2882 return ret; 2883 } 2884 2885 static const struct file_operations proc_setgroups_operations = { 2886 .open = proc_setgroups_open, 2887 .write = proc_setgroups_write, 2888 .read = seq_read, 2889 .llseek = seq_lseek, 2890 .release = proc_setgroups_release, 2891 }; 2892 #endif /* CONFIG_USER_NS */ 2893 2894 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns, 2895 struct pid *pid, struct task_struct *task) 2896 { 2897 int err = lock_trace(task); 2898 if (!err) { 2899 seq_printf(m, "%08x\n", task->personality); 2900 unlock_trace(task); 2901 } 2902 return err; 2903 } 2904 2905 #ifdef CONFIG_LIVEPATCH 2906 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns, 2907 struct pid *pid, struct task_struct *task) 2908 { 2909 seq_printf(m, "%d\n", task->patch_state); 2910 return 0; 2911 } 2912 #endif /* CONFIG_LIVEPATCH */ 2913 2914 /* 2915 * Thread groups 2916 */ 2917 static const struct file_operations proc_task_operations; 2918 static const struct inode_operations proc_task_inode_operations; 2919 2920 static const struct pid_entry tgid_base_stuff[] = { 2921 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations), 2922 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations), 2923 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations), 2924 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations), 2925 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations), 2926 #ifdef CONFIG_NET 2927 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations), 2928 #endif 2929 REG("environ", S_IRUSR, proc_environ_operations), 2930 REG("auxv", S_IRUSR, proc_auxv_operations), 2931 ONE("status", S_IRUGO, proc_pid_status), 2932 ONE("personality", S_IRUSR, proc_pid_personality), 2933 ONE("limits", S_IRUGO, proc_pid_limits), 2934 #ifdef CONFIG_SCHED_DEBUG 2935 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations), 2936 #endif 2937 #ifdef CONFIG_SCHED_AUTOGROUP 2938 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations), 2939 #endif 2940 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations), 2941 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 2942 ONE("syscall", S_IRUSR, proc_pid_syscall), 2943 #endif 2944 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops), 2945 ONE("stat", S_IRUGO, proc_tgid_stat), 2946 ONE("statm", S_IRUGO, proc_pid_statm), 2947 REG("maps", S_IRUGO, proc_pid_maps_operations), 2948 #ifdef CONFIG_NUMA 2949 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations), 2950 #endif 2951 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations), 2952 LNK("cwd", proc_cwd_link), 2953 LNK("root", proc_root_link), 2954 LNK("exe", proc_exe_link), 2955 REG("mounts", S_IRUGO, proc_mounts_operations), 2956 REG("mountinfo", S_IRUGO, proc_mountinfo_operations), 2957 REG("mountstats", S_IRUSR, proc_mountstats_operations), 2958 #ifdef CONFIG_PROC_PAGE_MONITOR 2959 REG("clear_refs", S_IWUSR, proc_clear_refs_operations), 2960 REG("smaps", S_IRUGO, proc_pid_smaps_operations), 2961 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations), 2962 REG("pagemap", S_IRUSR, proc_pagemap_operations), 2963 #endif 2964 #ifdef CONFIG_SECURITY 2965 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), 2966 #endif 2967 #ifdef CONFIG_KALLSYMS 2968 ONE("wchan", S_IRUGO, proc_pid_wchan), 2969 #endif 2970 #ifdef CONFIG_STACKTRACE 2971 ONE("stack", S_IRUSR, proc_pid_stack), 2972 #endif 2973 #ifdef CONFIG_SCHED_INFO 2974 ONE("schedstat", S_IRUGO, proc_pid_schedstat), 2975 #endif 2976 #ifdef CONFIG_LATENCYTOP 2977 REG("latency", S_IRUGO, proc_lstats_operations), 2978 #endif 2979 #ifdef CONFIG_PROC_PID_CPUSET 2980 ONE("cpuset", S_IRUGO, proc_cpuset_show), 2981 #endif 2982 #ifdef CONFIG_CGROUPS 2983 ONE("cgroup", S_IRUGO, proc_cgroup_show), 2984 #endif 2985 ONE("oom_score", S_IRUGO, proc_oom_score), 2986 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations), 2987 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations), 2988 #ifdef CONFIG_AUDITSYSCALL 2989 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations), 2990 REG("sessionid", S_IRUGO, proc_sessionid_operations), 2991 #endif 2992 #ifdef CONFIG_FAULT_INJECTION 2993 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations), 2994 REG("fail-nth", 0644, proc_fail_nth_operations), 2995 #endif 2996 #ifdef CONFIG_ELF_CORE 2997 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations), 2998 #endif 2999 #ifdef CONFIG_TASK_IO_ACCOUNTING 3000 ONE("io", S_IRUSR, proc_tgid_io_accounting), 3001 #endif 3002 #ifdef CONFIG_USER_NS 3003 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations), 3004 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations), 3005 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations), 3006 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations), 3007 #endif 3008 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS) 3009 REG("timers", S_IRUGO, proc_timers_operations), 3010 #endif 3011 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations), 3012 #ifdef CONFIG_LIVEPATCH 3013 ONE("patch_state", S_IRUSR, proc_pid_patch_state), 3014 #endif 3015 }; 3016 3017 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx) 3018 { 3019 return proc_pident_readdir(file, ctx, 3020 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 3021 } 3022 3023 static const struct file_operations proc_tgid_base_operations = { 3024 .read = generic_read_dir, 3025 .iterate_shared = proc_tgid_base_readdir, 3026 .llseek = generic_file_llseek, 3027 }; 3028 3029 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 3030 { 3031 return proc_pident_lookup(dir, dentry, 3032 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 3033 } 3034 3035 static const struct inode_operations proc_tgid_base_inode_operations = { 3036 .lookup = proc_tgid_base_lookup, 3037 .getattr = pid_getattr, 3038 .setattr = proc_setattr, 3039 .permission = proc_pid_permission, 3040 }; 3041 3042 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid) 3043 { 3044 struct dentry *dentry, *leader, *dir; 3045 char buf[10 + 1]; 3046 struct qstr name; 3047 3048 name.name = buf; 3049 name.len = snprintf(buf, sizeof(buf), "%u", pid); 3050 /* no ->d_hash() rejects on procfs */ 3051 dentry = d_hash_and_lookup(mnt->mnt_root, &name); 3052 if (dentry) { 3053 d_invalidate(dentry); 3054 dput(dentry); 3055 } 3056 3057 if (pid == tgid) 3058 return; 3059 3060 name.name = buf; 3061 name.len = snprintf(buf, sizeof(buf), "%u", tgid); 3062 leader = d_hash_and_lookup(mnt->mnt_root, &name); 3063 if (!leader) 3064 goto out; 3065 3066 name.name = "task"; 3067 name.len = strlen(name.name); 3068 dir = d_hash_and_lookup(leader, &name); 3069 if (!dir) 3070 goto out_put_leader; 3071 3072 name.name = buf; 3073 name.len = snprintf(buf, sizeof(buf), "%u", pid); 3074 dentry = d_hash_and_lookup(dir, &name); 3075 if (dentry) { 3076 d_invalidate(dentry); 3077 dput(dentry); 3078 } 3079 3080 dput(dir); 3081 out_put_leader: 3082 dput(leader); 3083 out: 3084 return; 3085 } 3086 3087 /** 3088 * proc_flush_task - Remove dcache entries for @task from the /proc dcache. 3089 * @task: task that should be flushed. 3090 * 3091 * When flushing dentries from proc, one needs to flush them from global 3092 * proc (proc_mnt) and from all the namespaces' procs this task was seen 3093 * in. This call is supposed to do all of this job. 3094 * 3095 * Looks in the dcache for 3096 * /proc/@pid 3097 * /proc/@tgid/task/@pid 3098 * if either directory is present flushes it and all of it'ts children 3099 * from the dcache. 3100 * 3101 * It is safe and reasonable to cache /proc entries for a task until 3102 * that task exits. After that they just clog up the dcache with 3103 * useless entries, possibly causing useful dcache entries to be 3104 * flushed instead. This routine is proved to flush those useless 3105 * dcache entries at process exit time. 3106 * 3107 * NOTE: This routine is just an optimization so it does not guarantee 3108 * that no dcache entries will exist at process exit time it 3109 * just makes it very unlikely that any will persist. 3110 */ 3111 3112 void proc_flush_task(struct task_struct *task) 3113 { 3114 int i; 3115 struct pid *pid, *tgid; 3116 struct upid *upid; 3117 3118 pid = task_pid(task); 3119 tgid = task_tgid(task); 3120 3121 for (i = 0; i <= pid->level; i++) { 3122 upid = &pid->numbers[i]; 3123 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr, 3124 tgid->numbers[i].nr); 3125 } 3126 } 3127 3128 static int proc_pid_instantiate(struct inode *dir, 3129 struct dentry * dentry, 3130 struct task_struct *task, const void *ptr) 3131 { 3132 struct inode *inode; 3133 3134 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO); 3135 if (!inode) 3136 goto out; 3137 3138 inode->i_op = &proc_tgid_base_inode_operations; 3139 inode->i_fop = &proc_tgid_base_operations; 3140 inode->i_flags|=S_IMMUTABLE; 3141 3142 set_nlink(inode, nlink_tgid); 3143 3144 d_set_d_op(dentry, &pid_dentry_operations); 3145 3146 d_add(dentry, inode); 3147 /* Close the race of the process dying before we return the dentry */ 3148 if (pid_revalidate(dentry, 0)) 3149 return 0; 3150 out: 3151 return -ENOENT; 3152 } 3153 3154 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 3155 { 3156 int result = -ENOENT; 3157 struct task_struct *task; 3158 unsigned tgid; 3159 struct pid_namespace *ns; 3160 3161 tgid = name_to_int(&dentry->d_name); 3162 if (tgid == ~0U) 3163 goto out; 3164 3165 ns = dentry->d_sb->s_fs_info; 3166 rcu_read_lock(); 3167 task = find_task_by_pid_ns(tgid, ns); 3168 if (task) 3169 get_task_struct(task); 3170 rcu_read_unlock(); 3171 if (!task) 3172 goto out; 3173 3174 result = proc_pid_instantiate(dir, dentry, task, NULL); 3175 put_task_struct(task); 3176 out: 3177 return ERR_PTR(result); 3178 } 3179 3180 /* 3181 * Find the first task with tgid >= tgid 3182 * 3183 */ 3184 struct tgid_iter { 3185 unsigned int tgid; 3186 struct task_struct *task; 3187 }; 3188 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter) 3189 { 3190 struct pid *pid; 3191 3192 if (iter.task) 3193 put_task_struct(iter.task); 3194 rcu_read_lock(); 3195 retry: 3196 iter.task = NULL; 3197 pid = find_ge_pid(iter.tgid, ns); 3198 if (pid) { 3199 iter.tgid = pid_nr_ns(pid, ns); 3200 iter.task = pid_task(pid, PIDTYPE_PID); 3201 /* What we to know is if the pid we have find is the 3202 * pid of a thread_group_leader. Testing for task 3203 * being a thread_group_leader is the obvious thing 3204 * todo but there is a window when it fails, due to 3205 * the pid transfer logic in de_thread. 3206 * 3207 * So we perform the straight forward test of seeing 3208 * if the pid we have found is the pid of a thread 3209 * group leader, and don't worry if the task we have 3210 * found doesn't happen to be a thread group leader. 3211 * As we don't care in the case of readdir. 3212 */ 3213 if (!iter.task || !has_group_leader_pid(iter.task)) { 3214 iter.tgid += 1; 3215 goto retry; 3216 } 3217 get_task_struct(iter.task); 3218 } 3219 rcu_read_unlock(); 3220 return iter; 3221 } 3222 3223 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2) 3224 3225 /* for the /proc/ directory itself, after non-process stuff has been done */ 3226 int proc_pid_readdir(struct file *file, struct dir_context *ctx) 3227 { 3228 struct tgid_iter iter; 3229 struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info; 3230 loff_t pos = ctx->pos; 3231 3232 if (pos >= PID_MAX_LIMIT + TGID_OFFSET) 3233 return 0; 3234 3235 if (pos == TGID_OFFSET - 2) { 3236 struct inode *inode = d_inode(ns->proc_self); 3237 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK)) 3238 return 0; 3239 ctx->pos = pos = pos + 1; 3240 } 3241 if (pos == TGID_OFFSET - 1) { 3242 struct inode *inode = d_inode(ns->proc_thread_self); 3243 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK)) 3244 return 0; 3245 ctx->pos = pos = pos + 1; 3246 } 3247 iter.tgid = pos - TGID_OFFSET; 3248 iter.task = NULL; 3249 for (iter = next_tgid(ns, iter); 3250 iter.task; 3251 iter.tgid += 1, iter = next_tgid(ns, iter)) { 3252 char name[10 + 1]; 3253 int len; 3254 3255 cond_resched(); 3256 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE)) 3257 continue; 3258 3259 len = snprintf(name, sizeof(name), "%u", iter.tgid); 3260 ctx->pos = iter.tgid + TGID_OFFSET; 3261 if (!proc_fill_cache(file, ctx, name, len, 3262 proc_pid_instantiate, iter.task, NULL)) { 3263 put_task_struct(iter.task); 3264 return 0; 3265 } 3266 } 3267 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET; 3268 return 0; 3269 } 3270 3271 /* 3272 * proc_tid_comm_permission is a special permission function exclusively 3273 * used for the node /proc/<pid>/task/<tid>/comm. 3274 * It bypasses generic permission checks in the case where a task of the same 3275 * task group attempts to access the node. 3276 * The rationale behind this is that glibc and bionic access this node for 3277 * cross thread naming (pthread_set/getname_np(!self)). However, if 3278 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0, 3279 * which locks out the cross thread naming implementation. 3280 * This function makes sure that the node is always accessible for members of 3281 * same thread group. 3282 */ 3283 static int proc_tid_comm_permission(struct inode *inode, int mask) 3284 { 3285 bool is_same_tgroup; 3286 struct task_struct *task; 3287 3288 task = get_proc_task(inode); 3289 if (!task) 3290 return -ESRCH; 3291 is_same_tgroup = same_thread_group(current, task); 3292 put_task_struct(task); 3293 3294 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) { 3295 /* This file (/proc/<pid>/task/<tid>/comm) can always be 3296 * read or written by the members of the corresponding 3297 * thread group. 3298 */ 3299 return 0; 3300 } 3301 3302 return generic_permission(inode, mask); 3303 } 3304 3305 static const struct inode_operations proc_tid_comm_inode_operations = { 3306 .permission = proc_tid_comm_permission, 3307 }; 3308 3309 /* 3310 * Tasks 3311 */ 3312 static const struct pid_entry tid_base_stuff[] = { 3313 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations), 3314 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations), 3315 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations), 3316 #ifdef CONFIG_NET 3317 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations), 3318 #endif 3319 REG("environ", S_IRUSR, proc_environ_operations), 3320 REG("auxv", S_IRUSR, proc_auxv_operations), 3321 ONE("status", S_IRUGO, proc_pid_status), 3322 ONE("personality", S_IRUSR, proc_pid_personality), 3323 ONE("limits", S_IRUGO, proc_pid_limits), 3324 #ifdef CONFIG_SCHED_DEBUG 3325 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations), 3326 #endif 3327 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR, 3328 &proc_tid_comm_inode_operations, 3329 &proc_pid_set_comm_operations, {}), 3330 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 3331 ONE("syscall", S_IRUSR, proc_pid_syscall), 3332 #endif 3333 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops), 3334 ONE("stat", S_IRUGO, proc_tid_stat), 3335 ONE("statm", S_IRUGO, proc_pid_statm), 3336 REG("maps", S_IRUGO, proc_tid_maps_operations), 3337 #ifdef CONFIG_PROC_CHILDREN 3338 REG("children", S_IRUGO, proc_tid_children_operations), 3339 #endif 3340 #ifdef CONFIG_NUMA 3341 REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations), 3342 #endif 3343 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations), 3344 LNK("cwd", proc_cwd_link), 3345 LNK("root", proc_root_link), 3346 LNK("exe", proc_exe_link), 3347 REG("mounts", S_IRUGO, proc_mounts_operations), 3348 REG("mountinfo", S_IRUGO, proc_mountinfo_operations), 3349 #ifdef CONFIG_PROC_PAGE_MONITOR 3350 REG("clear_refs", S_IWUSR, proc_clear_refs_operations), 3351 REG("smaps", S_IRUGO, proc_tid_smaps_operations), 3352 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations), 3353 REG("pagemap", S_IRUSR, proc_pagemap_operations), 3354 #endif 3355 #ifdef CONFIG_SECURITY 3356 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), 3357 #endif 3358 #ifdef CONFIG_KALLSYMS 3359 ONE("wchan", S_IRUGO, proc_pid_wchan), 3360 #endif 3361 #ifdef CONFIG_STACKTRACE 3362 ONE("stack", S_IRUSR, proc_pid_stack), 3363 #endif 3364 #ifdef CONFIG_SCHED_INFO 3365 ONE("schedstat", S_IRUGO, proc_pid_schedstat), 3366 #endif 3367 #ifdef CONFIG_LATENCYTOP 3368 REG("latency", S_IRUGO, proc_lstats_operations), 3369 #endif 3370 #ifdef CONFIG_PROC_PID_CPUSET 3371 ONE("cpuset", S_IRUGO, proc_cpuset_show), 3372 #endif 3373 #ifdef CONFIG_CGROUPS 3374 ONE("cgroup", S_IRUGO, proc_cgroup_show), 3375 #endif 3376 ONE("oom_score", S_IRUGO, proc_oom_score), 3377 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations), 3378 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations), 3379 #ifdef CONFIG_AUDITSYSCALL 3380 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations), 3381 REG("sessionid", S_IRUGO, proc_sessionid_operations), 3382 #endif 3383 #ifdef CONFIG_FAULT_INJECTION 3384 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations), 3385 REG("fail-nth", 0644, proc_fail_nth_operations), 3386 #endif 3387 #ifdef CONFIG_TASK_IO_ACCOUNTING 3388 ONE("io", S_IRUSR, proc_tid_io_accounting), 3389 #endif 3390 #ifdef CONFIG_USER_NS 3391 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations), 3392 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations), 3393 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations), 3394 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations), 3395 #endif 3396 #ifdef CONFIG_LIVEPATCH 3397 ONE("patch_state", S_IRUSR, proc_pid_patch_state), 3398 #endif 3399 }; 3400 3401 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx) 3402 { 3403 return proc_pident_readdir(file, ctx, 3404 tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 3405 } 3406 3407 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 3408 { 3409 return proc_pident_lookup(dir, dentry, 3410 tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 3411 } 3412 3413 static const struct file_operations proc_tid_base_operations = { 3414 .read = generic_read_dir, 3415 .iterate_shared = proc_tid_base_readdir, 3416 .llseek = generic_file_llseek, 3417 }; 3418 3419 static const struct inode_operations proc_tid_base_inode_operations = { 3420 .lookup = proc_tid_base_lookup, 3421 .getattr = pid_getattr, 3422 .setattr = proc_setattr, 3423 }; 3424 3425 static int proc_task_instantiate(struct inode *dir, 3426 struct dentry *dentry, struct task_struct *task, const void *ptr) 3427 { 3428 struct inode *inode; 3429 inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO); 3430 3431 if (!inode) 3432 goto out; 3433 inode->i_op = &proc_tid_base_inode_operations; 3434 inode->i_fop = &proc_tid_base_operations; 3435 inode->i_flags|=S_IMMUTABLE; 3436 3437 set_nlink(inode, nlink_tid); 3438 3439 d_set_d_op(dentry, &pid_dentry_operations); 3440 3441 d_add(dentry, inode); 3442 /* Close the race of the process dying before we return the dentry */ 3443 if (pid_revalidate(dentry, 0)) 3444 return 0; 3445 out: 3446 return -ENOENT; 3447 } 3448 3449 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 3450 { 3451 int result = -ENOENT; 3452 struct task_struct *task; 3453 struct task_struct *leader = get_proc_task(dir); 3454 unsigned tid; 3455 struct pid_namespace *ns; 3456 3457 if (!leader) 3458 goto out_no_task; 3459 3460 tid = name_to_int(&dentry->d_name); 3461 if (tid == ~0U) 3462 goto out; 3463 3464 ns = dentry->d_sb->s_fs_info; 3465 rcu_read_lock(); 3466 task = find_task_by_pid_ns(tid, ns); 3467 if (task) 3468 get_task_struct(task); 3469 rcu_read_unlock(); 3470 if (!task) 3471 goto out; 3472 if (!same_thread_group(leader, task)) 3473 goto out_drop_task; 3474 3475 result = proc_task_instantiate(dir, dentry, task, NULL); 3476 out_drop_task: 3477 put_task_struct(task); 3478 out: 3479 put_task_struct(leader); 3480 out_no_task: 3481 return ERR_PTR(result); 3482 } 3483 3484 /* 3485 * Find the first tid of a thread group to return to user space. 3486 * 3487 * Usually this is just the thread group leader, but if the users 3488 * buffer was too small or there was a seek into the middle of the 3489 * directory we have more work todo. 3490 * 3491 * In the case of a short read we start with find_task_by_pid. 3492 * 3493 * In the case of a seek we start with the leader and walk nr 3494 * threads past it. 3495 */ 3496 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos, 3497 struct pid_namespace *ns) 3498 { 3499 struct task_struct *pos, *task; 3500 unsigned long nr = f_pos; 3501 3502 if (nr != f_pos) /* 32bit overflow? */ 3503 return NULL; 3504 3505 rcu_read_lock(); 3506 task = pid_task(pid, PIDTYPE_PID); 3507 if (!task) 3508 goto fail; 3509 3510 /* Attempt to start with the tid of a thread */ 3511 if (tid && nr) { 3512 pos = find_task_by_pid_ns(tid, ns); 3513 if (pos && same_thread_group(pos, task)) 3514 goto found; 3515 } 3516 3517 /* If nr exceeds the number of threads there is nothing todo */ 3518 if (nr >= get_nr_threads(task)) 3519 goto fail; 3520 3521 /* If we haven't found our starting place yet start 3522 * with the leader and walk nr threads forward. 3523 */ 3524 pos = task = task->group_leader; 3525 do { 3526 if (!nr--) 3527 goto found; 3528 } while_each_thread(task, pos); 3529 fail: 3530 pos = NULL; 3531 goto out; 3532 found: 3533 get_task_struct(pos); 3534 out: 3535 rcu_read_unlock(); 3536 return pos; 3537 } 3538 3539 /* 3540 * Find the next thread in the thread list. 3541 * Return NULL if there is an error or no next thread. 3542 * 3543 * The reference to the input task_struct is released. 3544 */ 3545 static struct task_struct *next_tid(struct task_struct *start) 3546 { 3547 struct task_struct *pos = NULL; 3548 rcu_read_lock(); 3549 if (pid_alive(start)) { 3550 pos = next_thread(start); 3551 if (thread_group_leader(pos)) 3552 pos = NULL; 3553 else 3554 get_task_struct(pos); 3555 } 3556 rcu_read_unlock(); 3557 put_task_struct(start); 3558 return pos; 3559 } 3560 3561 /* for the /proc/TGID/task/ directories */ 3562 static int proc_task_readdir(struct file *file, struct dir_context *ctx) 3563 { 3564 struct inode *inode = file_inode(file); 3565 struct task_struct *task; 3566 struct pid_namespace *ns; 3567 int tid; 3568 3569 if (proc_inode_is_dead(inode)) 3570 return -ENOENT; 3571 3572 if (!dir_emit_dots(file, ctx)) 3573 return 0; 3574 3575 /* f_version caches the tgid value that the last readdir call couldn't 3576 * return. lseek aka telldir automagically resets f_version to 0. 3577 */ 3578 ns = inode->i_sb->s_fs_info; 3579 tid = (int)file->f_version; 3580 file->f_version = 0; 3581 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns); 3582 task; 3583 task = next_tid(task), ctx->pos++) { 3584 char name[10 + 1]; 3585 int len; 3586 tid = task_pid_nr_ns(task, ns); 3587 len = snprintf(name, sizeof(name), "%u", tid); 3588 if (!proc_fill_cache(file, ctx, name, len, 3589 proc_task_instantiate, task, NULL)) { 3590 /* returning this tgid failed, save it as the first 3591 * pid for the next readir call */ 3592 file->f_version = (u64)tid; 3593 put_task_struct(task); 3594 break; 3595 } 3596 } 3597 3598 return 0; 3599 } 3600 3601 static int proc_task_getattr(const struct path *path, struct kstat *stat, 3602 u32 request_mask, unsigned int query_flags) 3603 { 3604 struct inode *inode = d_inode(path->dentry); 3605 struct task_struct *p = get_proc_task(inode); 3606 generic_fillattr(inode, stat); 3607 3608 if (p) { 3609 stat->nlink += get_nr_threads(p); 3610 put_task_struct(p); 3611 } 3612 3613 return 0; 3614 } 3615 3616 static const struct inode_operations proc_task_inode_operations = { 3617 .lookup = proc_task_lookup, 3618 .getattr = proc_task_getattr, 3619 .setattr = proc_setattr, 3620 .permission = proc_pid_permission, 3621 }; 3622 3623 static const struct file_operations proc_task_operations = { 3624 .read = generic_read_dir, 3625 .iterate_shared = proc_task_readdir, 3626 .llseek = generic_file_llseek, 3627 }; 3628 3629 void __init set_proc_pid_nlink(void) 3630 { 3631 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 3632 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 3633 } 3634