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/generic-radix-tree.h> 63 #include <linux/string.h> 64 #include <linux/seq_file.h> 65 #include <linux/namei.h> 66 #include <linux/mnt_namespace.h> 67 #include <linux/mm.h> 68 #include <linux/swap.h> 69 #include <linux/rcupdate.h> 70 #include <linux/kallsyms.h> 71 #include <linux/stacktrace.h> 72 #include <linux/resource.h> 73 #include <linux/module.h> 74 #include <linux/mount.h> 75 #include <linux/security.h> 76 #include <linux/ptrace.h> 77 #include <linux/tracehook.h> 78 #include <linux/printk.h> 79 #include <linux/cache.h> 80 #include <linux/cgroup.h> 81 #include <linux/cpuset.h> 82 #include <linux/audit.h> 83 #include <linux/poll.h> 84 #include <linux/nsproxy.h> 85 #include <linux/oom.h> 86 #include <linux/elf.h> 87 #include <linux/pid_namespace.h> 88 #include <linux/user_namespace.h> 89 #include <linux/fs_struct.h> 90 #include <linux/slab.h> 91 #include <linux/sched/autogroup.h> 92 #include <linux/sched/mm.h> 93 #include <linux/sched/coredump.h> 94 #include <linux/sched/debug.h> 95 #include <linux/sched/stat.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 #define ATTR(LSM, NAME, MODE) \ 147 NOD(NAME, (S_IFREG|(MODE)), \ 148 NULL, &proc_pid_attr_operations, \ 149 { .lsm = LSM }) 150 151 /* 152 * Count the number of hardlinks for the pid_entry table, excluding the . 153 * and .. links. 154 */ 155 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries, 156 unsigned int n) 157 { 158 unsigned int i; 159 unsigned int count; 160 161 count = 2; 162 for (i = 0; i < n; ++i) { 163 if (S_ISDIR(entries[i].mode)) 164 ++count; 165 } 166 167 return count; 168 } 169 170 static int get_task_root(struct task_struct *task, struct path *root) 171 { 172 int result = -ENOENT; 173 174 task_lock(task); 175 if (task->fs) { 176 get_fs_root(task->fs, root); 177 result = 0; 178 } 179 task_unlock(task); 180 return result; 181 } 182 183 static int proc_cwd_link(struct dentry *dentry, struct path *path) 184 { 185 struct task_struct *task = get_proc_task(d_inode(dentry)); 186 int result = -ENOENT; 187 188 if (task) { 189 task_lock(task); 190 if (task->fs) { 191 get_fs_pwd(task->fs, path); 192 result = 0; 193 } 194 task_unlock(task); 195 put_task_struct(task); 196 } 197 return result; 198 } 199 200 static int proc_root_link(struct dentry *dentry, struct path *path) 201 { 202 struct task_struct *task = get_proc_task(d_inode(dentry)); 203 int result = -ENOENT; 204 205 if (task) { 206 result = get_task_root(task, path); 207 put_task_struct(task); 208 } 209 return result; 210 } 211 212 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf, 213 size_t count, loff_t *ppos) 214 { 215 unsigned long arg_start, arg_end; 216 unsigned long pos, len; 217 char *page; 218 219 /* Check if process spawned far enough to have cmdline. */ 220 if (!mm->env_end) 221 return 0; 222 223 spin_lock(&mm->arg_lock); 224 arg_start = mm->arg_start; 225 arg_end = mm->arg_end; 226 spin_unlock(&mm->arg_lock); 227 228 if (arg_start >= arg_end) 229 return 0; 230 231 /* We're not going to care if "*ppos" has high bits set */ 232 /* .. but we do check the result is in the proper range */ 233 pos = arg_start + *ppos; 234 if (pos < arg_start || pos >= arg_end) 235 return 0; 236 if (count > arg_end - pos) 237 count = arg_end - pos; 238 239 page = (char *)__get_free_page(GFP_KERNEL); 240 if (!page) 241 return -ENOMEM; 242 243 len = 0; 244 while (count) { 245 int got; 246 size_t size = min_t(size_t, PAGE_SIZE, count); 247 248 got = access_remote_vm(mm, pos, page, size, FOLL_ANON); 249 if (got <= 0) 250 break; 251 got -= copy_to_user(buf, page, got); 252 if (unlikely(!got)) { 253 if (!len) 254 len = -EFAULT; 255 break; 256 } 257 pos += got; 258 buf += got; 259 len += got; 260 count -= got; 261 } 262 263 free_page((unsigned long)page); 264 return len; 265 } 266 267 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf, 268 size_t count, loff_t *pos) 269 { 270 struct mm_struct *mm; 271 ssize_t ret; 272 273 mm = get_task_mm(tsk); 274 if (!mm) 275 return 0; 276 277 ret = get_mm_cmdline(mm, buf, count, pos); 278 mmput(mm); 279 return ret; 280 } 281 282 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf, 283 size_t count, loff_t *pos) 284 { 285 struct task_struct *tsk; 286 ssize_t ret; 287 288 BUG_ON(*pos < 0); 289 290 tsk = get_proc_task(file_inode(file)); 291 if (!tsk) 292 return -ESRCH; 293 ret = get_task_cmdline(tsk, buf, count, pos); 294 put_task_struct(tsk); 295 if (ret > 0) 296 *pos += ret; 297 return ret; 298 } 299 300 static const struct file_operations proc_pid_cmdline_ops = { 301 .read = proc_pid_cmdline_read, 302 .llseek = generic_file_llseek, 303 }; 304 305 #ifdef CONFIG_KALLSYMS 306 /* 307 * Provides a wchan file via kallsyms in a proper one-value-per-file format. 308 * Returns the resolved symbol. If that fails, simply return the address. 309 */ 310 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns, 311 struct pid *pid, struct task_struct *task) 312 { 313 unsigned long wchan; 314 char symname[KSYM_NAME_LEN]; 315 316 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) 317 goto print0; 318 319 wchan = get_wchan(task); 320 if (wchan && !lookup_symbol_name(wchan, symname)) { 321 seq_puts(m, symname); 322 return 0; 323 } 324 325 print0: 326 seq_putc(m, '0'); 327 return 0; 328 } 329 #endif /* CONFIG_KALLSYMS */ 330 331 static int lock_trace(struct task_struct *task) 332 { 333 int err = mutex_lock_killable(&task->signal->cred_guard_mutex); 334 if (err) 335 return err; 336 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) { 337 mutex_unlock(&task->signal->cred_guard_mutex); 338 return -EPERM; 339 } 340 return 0; 341 } 342 343 static void unlock_trace(struct task_struct *task) 344 { 345 mutex_unlock(&task->signal->cred_guard_mutex); 346 } 347 348 #ifdef CONFIG_STACKTRACE 349 350 #define MAX_STACK_TRACE_DEPTH 64 351 352 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns, 353 struct pid *pid, struct task_struct *task) 354 { 355 unsigned long *entries; 356 int err; 357 358 /* 359 * The ability to racily run the kernel stack unwinder on a running task 360 * and then observe the unwinder output is scary; while it is useful for 361 * debugging kernel issues, it can also allow an attacker to leak kernel 362 * stack contents. 363 * Doing this in a manner that is at least safe from races would require 364 * some work to ensure that the remote task can not be scheduled; and 365 * even then, this would still expose the unwinder as local attack 366 * surface. 367 * Therefore, this interface is restricted to root. 368 */ 369 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN)) 370 return -EACCES; 371 372 entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries), 373 GFP_KERNEL); 374 if (!entries) 375 return -ENOMEM; 376 377 err = lock_trace(task); 378 if (!err) { 379 unsigned int i, nr_entries; 380 381 nr_entries = stack_trace_save_tsk(task, entries, 382 MAX_STACK_TRACE_DEPTH, 0); 383 384 for (i = 0; i < nr_entries; i++) { 385 seq_printf(m, "[<0>] %pB\n", (void *)entries[i]); 386 } 387 388 unlock_trace(task); 389 } 390 kfree(entries); 391 392 return err; 393 } 394 #endif 395 396 #ifdef CONFIG_SCHED_INFO 397 /* 398 * Provides /proc/PID/schedstat 399 */ 400 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns, 401 struct pid *pid, struct task_struct *task) 402 { 403 if (unlikely(!sched_info_on())) 404 seq_puts(m, "0 0 0\n"); 405 else 406 seq_printf(m, "%llu %llu %lu\n", 407 (unsigned long long)task->se.sum_exec_runtime, 408 (unsigned long long)task->sched_info.run_delay, 409 task->sched_info.pcount); 410 411 return 0; 412 } 413 #endif 414 415 #ifdef CONFIG_LATENCYTOP 416 static int lstats_show_proc(struct seq_file *m, void *v) 417 { 418 int i; 419 struct inode *inode = m->private; 420 struct task_struct *task = get_proc_task(inode); 421 422 if (!task) 423 return -ESRCH; 424 seq_puts(m, "Latency Top version : v0.1\n"); 425 for (i = 0; i < LT_SAVECOUNT; i++) { 426 struct latency_record *lr = &task->latency_record[i]; 427 if (lr->backtrace[0]) { 428 int q; 429 seq_printf(m, "%i %li %li", 430 lr->count, lr->time, lr->max); 431 for (q = 0; q < LT_BACKTRACEDEPTH; q++) { 432 unsigned long bt = lr->backtrace[q]; 433 434 if (!bt) 435 break; 436 seq_printf(m, " %ps", (void *)bt); 437 } 438 seq_putc(m, '\n'); 439 } 440 441 } 442 put_task_struct(task); 443 return 0; 444 } 445 446 static int lstats_open(struct inode *inode, struct file *file) 447 { 448 return single_open(file, lstats_show_proc, inode); 449 } 450 451 static ssize_t lstats_write(struct file *file, const char __user *buf, 452 size_t count, loff_t *offs) 453 { 454 struct task_struct *task = get_proc_task(file_inode(file)); 455 456 if (!task) 457 return -ESRCH; 458 clear_tsk_latency_tracing(task); 459 put_task_struct(task); 460 461 return count; 462 } 463 464 static const struct file_operations proc_lstats_operations = { 465 .open = lstats_open, 466 .read = seq_read, 467 .write = lstats_write, 468 .llseek = seq_lseek, 469 .release = single_release, 470 }; 471 472 #endif 473 474 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns, 475 struct pid *pid, struct task_struct *task) 476 { 477 unsigned long totalpages = totalram_pages() + total_swap_pages; 478 unsigned long points = 0; 479 480 points = oom_badness(task, NULL, NULL, totalpages) * 481 1000 / totalpages; 482 seq_printf(m, "%lu\n", points); 483 484 return 0; 485 } 486 487 struct limit_names { 488 const char *name; 489 const char *unit; 490 }; 491 492 static const struct limit_names lnames[RLIM_NLIMITS] = { 493 [RLIMIT_CPU] = {"Max cpu time", "seconds"}, 494 [RLIMIT_FSIZE] = {"Max file size", "bytes"}, 495 [RLIMIT_DATA] = {"Max data size", "bytes"}, 496 [RLIMIT_STACK] = {"Max stack size", "bytes"}, 497 [RLIMIT_CORE] = {"Max core file size", "bytes"}, 498 [RLIMIT_RSS] = {"Max resident set", "bytes"}, 499 [RLIMIT_NPROC] = {"Max processes", "processes"}, 500 [RLIMIT_NOFILE] = {"Max open files", "files"}, 501 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"}, 502 [RLIMIT_AS] = {"Max address space", "bytes"}, 503 [RLIMIT_LOCKS] = {"Max file locks", "locks"}, 504 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"}, 505 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"}, 506 [RLIMIT_NICE] = {"Max nice priority", NULL}, 507 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL}, 508 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"}, 509 }; 510 511 /* Display limits for a process */ 512 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns, 513 struct pid *pid, struct task_struct *task) 514 { 515 unsigned int i; 516 unsigned long flags; 517 518 struct rlimit rlim[RLIM_NLIMITS]; 519 520 if (!lock_task_sighand(task, &flags)) 521 return 0; 522 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS); 523 unlock_task_sighand(task, &flags); 524 525 /* 526 * print the file header 527 */ 528 seq_puts(m, "Limit " 529 "Soft Limit " 530 "Hard Limit " 531 "Units \n"); 532 533 for (i = 0; i < RLIM_NLIMITS; i++) { 534 if (rlim[i].rlim_cur == RLIM_INFINITY) 535 seq_printf(m, "%-25s %-20s ", 536 lnames[i].name, "unlimited"); 537 else 538 seq_printf(m, "%-25s %-20lu ", 539 lnames[i].name, rlim[i].rlim_cur); 540 541 if (rlim[i].rlim_max == RLIM_INFINITY) 542 seq_printf(m, "%-20s ", "unlimited"); 543 else 544 seq_printf(m, "%-20lu ", rlim[i].rlim_max); 545 546 if (lnames[i].unit) 547 seq_printf(m, "%-10s\n", lnames[i].unit); 548 else 549 seq_putc(m, '\n'); 550 } 551 552 return 0; 553 } 554 555 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 556 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns, 557 struct pid *pid, struct task_struct *task) 558 { 559 struct syscall_info info; 560 u64 *args = &info.data.args[0]; 561 int res; 562 563 res = lock_trace(task); 564 if (res) 565 return res; 566 567 if (task_current_syscall(task, &info)) 568 seq_puts(m, "running\n"); 569 else if (info.data.nr < 0) 570 seq_printf(m, "%d 0x%llx 0x%llx\n", 571 info.data.nr, info.sp, info.data.instruction_pointer); 572 else 573 seq_printf(m, 574 "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n", 575 info.data.nr, 576 args[0], args[1], args[2], args[3], args[4], args[5], 577 info.sp, info.data.instruction_pointer); 578 unlock_trace(task); 579 580 return 0; 581 } 582 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */ 583 584 /************************************************************************/ 585 /* Here the fs part begins */ 586 /************************************************************************/ 587 588 /* permission checks */ 589 static int proc_fd_access_allowed(struct inode *inode) 590 { 591 struct task_struct *task; 592 int allowed = 0; 593 /* Allow access to a task's file descriptors if it is us or we 594 * may use ptrace attach to the process and find out that 595 * information. 596 */ 597 task = get_proc_task(inode); 598 if (task) { 599 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS); 600 put_task_struct(task); 601 } 602 return allowed; 603 } 604 605 int proc_setattr(struct dentry *dentry, struct iattr *attr) 606 { 607 int error; 608 struct inode *inode = d_inode(dentry); 609 610 if (attr->ia_valid & ATTR_MODE) 611 return -EPERM; 612 613 error = setattr_prepare(dentry, attr); 614 if (error) 615 return error; 616 617 setattr_copy(inode, attr); 618 mark_inode_dirty(inode); 619 return 0; 620 } 621 622 /* 623 * May current process learn task's sched/cmdline info (for hide_pid_min=1) 624 * or euid/egid (for hide_pid_min=2)? 625 */ 626 static bool has_pid_permissions(struct pid_namespace *pid, 627 struct task_struct *task, 628 int hide_pid_min) 629 { 630 if (pid->hide_pid < hide_pid_min) 631 return true; 632 if (in_group_p(pid->pid_gid)) 633 return true; 634 return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS); 635 } 636 637 638 static int proc_pid_permission(struct inode *inode, int mask) 639 { 640 struct pid_namespace *pid = proc_pid_ns(inode); 641 struct task_struct *task; 642 bool has_perms; 643 644 task = get_proc_task(inode); 645 if (!task) 646 return -ESRCH; 647 has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS); 648 put_task_struct(task); 649 650 if (!has_perms) { 651 if (pid->hide_pid == HIDEPID_INVISIBLE) { 652 /* 653 * Let's make getdents(), stat(), and open() 654 * consistent with each other. If a process 655 * may not stat() a file, it shouldn't be seen 656 * in procfs at all. 657 */ 658 return -ENOENT; 659 } 660 661 return -EPERM; 662 } 663 return generic_permission(inode, mask); 664 } 665 666 667 668 static const struct inode_operations proc_def_inode_operations = { 669 .setattr = proc_setattr, 670 }; 671 672 static int proc_single_show(struct seq_file *m, void *v) 673 { 674 struct inode *inode = m->private; 675 struct pid_namespace *ns = proc_pid_ns(inode); 676 struct pid *pid = proc_pid(inode); 677 struct task_struct *task; 678 int ret; 679 680 task = get_pid_task(pid, PIDTYPE_PID); 681 if (!task) 682 return -ESRCH; 683 684 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task); 685 686 put_task_struct(task); 687 return ret; 688 } 689 690 static int proc_single_open(struct inode *inode, struct file *filp) 691 { 692 return single_open(filp, proc_single_show, inode); 693 } 694 695 static const struct file_operations proc_single_file_operations = { 696 .open = proc_single_open, 697 .read = seq_read, 698 .llseek = seq_lseek, 699 .release = single_release, 700 }; 701 702 703 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode) 704 { 705 struct task_struct *task = get_proc_task(inode); 706 struct mm_struct *mm = ERR_PTR(-ESRCH); 707 708 if (task) { 709 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS); 710 put_task_struct(task); 711 712 if (!IS_ERR_OR_NULL(mm)) { 713 /* ensure this mm_struct can't be freed */ 714 mmgrab(mm); 715 /* but do not pin its memory */ 716 mmput(mm); 717 } 718 } 719 720 return mm; 721 } 722 723 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode) 724 { 725 struct mm_struct *mm = proc_mem_open(inode, mode); 726 727 if (IS_ERR(mm)) 728 return PTR_ERR(mm); 729 730 file->private_data = mm; 731 return 0; 732 } 733 734 static int mem_open(struct inode *inode, struct file *file) 735 { 736 int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH); 737 738 /* OK to pass negative loff_t, we can catch out-of-range */ 739 file->f_mode |= FMODE_UNSIGNED_OFFSET; 740 741 return ret; 742 } 743 744 static ssize_t mem_rw(struct file *file, char __user *buf, 745 size_t count, loff_t *ppos, int write) 746 { 747 struct mm_struct *mm = file->private_data; 748 unsigned long addr = *ppos; 749 ssize_t copied; 750 char *page; 751 unsigned int flags; 752 753 if (!mm) 754 return 0; 755 756 page = (char *)__get_free_page(GFP_KERNEL); 757 if (!page) 758 return -ENOMEM; 759 760 copied = 0; 761 if (!mmget_not_zero(mm)) 762 goto free; 763 764 flags = FOLL_FORCE | (write ? FOLL_WRITE : 0); 765 766 while (count > 0) { 767 int this_len = min_t(int, count, PAGE_SIZE); 768 769 if (write && copy_from_user(page, buf, this_len)) { 770 copied = -EFAULT; 771 break; 772 } 773 774 this_len = access_remote_vm(mm, addr, page, this_len, flags); 775 if (!this_len) { 776 if (!copied) 777 copied = -EIO; 778 break; 779 } 780 781 if (!write && copy_to_user(buf, page, this_len)) { 782 copied = -EFAULT; 783 break; 784 } 785 786 buf += this_len; 787 addr += this_len; 788 copied += this_len; 789 count -= this_len; 790 } 791 *ppos = addr; 792 793 mmput(mm); 794 free: 795 free_page((unsigned long) page); 796 return copied; 797 } 798 799 static ssize_t mem_read(struct file *file, char __user *buf, 800 size_t count, loff_t *ppos) 801 { 802 return mem_rw(file, buf, count, ppos, 0); 803 } 804 805 static ssize_t mem_write(struct file *file, const char __user *buf, 806 size_t count, loff_t *ppos) 807 { 808 return mem_rw(file, (char __user*)buf, count, ppos, 1); 809 } 810 811 loff_t mem_lseek(struct file *file, loff_t offset, int orig) 812 { 813 switch (orig) { 814 case 0: 815 file->f_pos = offset; 816 break; 817 case 1: 818 file->f_pos += offset; 819 break; 820 default: 821 return -EINVAL; 822 } 823 force_successful_syscall_return(); 824 return file->f_pos; 825 } 826 827 static int mem_release(struct inode *inode, struct file *file) 828 { 829 struct mm_struct *mm = file->private_data; 830 if (mm) 831 mmdrop(mm); 832 return 0; 833 } 834 835 static const struct file_operations proc_mem_operations = { 836 .llseek = mem_lseek, 837 .read = mem_read, 838 .write = mem_write, 839 .open = mem_open, 840 .release = mem_release, 841 }; 842 843 static int environ_open(struct inode *inode, struct file *file) 844 { 845 return __mem_open(inode, file, PTRACE_MODE_READ); 846 } 847 848 static ssize_t environ_read(struct file *file, char __user *buf, 849 size_t count, loff_t *ppos) 850 { 851 char *page; 852 unsigned long src = *ppos; 853 int ret = 0; 854 struct mm_struct *mm = file->private_data; 855 unsigned long env_start, env_end; 856 857 /* Ensure the process spawned far enough to have an environment. */ 858 if (!mm || !mm->env_end) 859 return 0; 860 861 page = (char *)__get_free_page(GFP_KERNEL); 862 if (!page) 863 return -ENOMEM; 864 865 ret = 0; 866 if (!mmget_not_zero(mm)) 867 goto free; 868 869 spin_lock(&mm->arg_lock); 870 env_start = mm->env_start; 871 env_end = mm->env_end; 872 spin_unlock(&mm->arg_lock); 873 874 while (count > 0) { 875 size_t this_len, max_len; 876 int retval; 877 878 if (src >= (env_end - env_start)) 879 break; 880 881 this_len = env_end - (env_start + src); 882 883 max_len = min_t(size_t, PAGE_SIZE, count); 884 this_len = min(max_len, this_len); 885 886 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON); 887 888 if (retval <= 0) { 889 ret = retval; 890 break; 891 } 892 893 if (copy_to_user(buf, page, retval)) { 894 ret = -EFAULT; 895 break; 896 } 897 898 ret += retval; 899 src += retval; 900 buf += retval; 901 count -= retval; 902 } 903 *ppos = src; 904 mmput(mm); 905 906 free: 907 free_page((unsigned long) page); 908 return ret; 909 } 910 911 static const struct file_operations proc_environ_operations = { 912 .open = environ_open, 913 .read = environ_read, 914 .llseek = generic_file_llseek, 915 .release = mem_release, 916 }; 917 918 static int auxv_open(struct inode *inode, struct file *file) 919 { 920 return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS); 921 } 922 923 static ssize_t auxv_read(struct file *file, char __user *buf, 924 size_t count, loff_t *ppos) 925 { 926 struct mm_struct *mm = file->private_data; 927 unsigned int nwords = 0; 928 929 if (!mm) 930 return 0; 931 do { 932 nwords += 2; 933 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */ 934 return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv, 935 nwords * sizeof(mm->saved_auxv[0])); 936 } 937 938 static const struct file_operations proc_auxv_operations = { 939 .open = auxv_open, 940 .read = auxv_read, 941 .llseek = generic_file_llseek, 942 .release = mem_release, 943 }; 944 945 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count, 946 loff_t *ppos) 947 { 948 struct task_struct *task = get_proc_task(file_inode(file)); 949 char buffer[PROC_NUMBUF]; 950 int oom_adj = OOM_ADJUST_MIN; 951 size_t len; 952 953 if (!task) 954 return -ESRCH; 955 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX) 956 oom_adj = OOM_ADJUST_MAX; 957 else 958 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) / 959 OOM_SCORE_ADJ_MAX; 960 put_task_struct(task); 961 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj); 962 return simple_read_from_buffer(buf, count, ppos, buffer, len); 963 } 964 965 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy) 966 { 967 static DEFINE_MUTEX(oom_adj_mutex); 968 struct mm_struct *mm = NULL; 969 struct task_struct *task; 970 int err = 0; 971 972 task = get_proc_task(file_inode(file)); 973 if (!task) 974 return -ESRCH; 975 976 mutex_lock(&oom_adj_mutex); 977 if (legacy) { 978 if (oom_adj < task->signal->oom_score_adj && 979 !capable(CAP_SYS_RESOURCE)) { 980 err = -EACCES; 981 goto err_unlock; 982 } 983 /* 984 * /proc/pid/oom_adj is provided for legacy purposes, ask users to use 985 * /proc/pid/oom_score_adj instead. 986 */ 987 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n", 988 current->comm, task_pid_nr(current), task_pid_nr(task), 989 task_pid_nr(task)); 990 } else { 991 if ((short)oom_adj < task->signal->oom_score_adj_min && 992 !capable(CAP_SYS_RESOURCE)) { 993 err = -EACCES; 994 goto err_unlock; 995 } 996 } 997 998 /* 999 * Make sure we will check other processes sharing the mm if this is 1000 * not vfrok which wants its own oom_score_adj. 1001 * pin the mm so it doesn't go away and get reused after task_unlock 1002 */ 1003 if (!task->vfork_done) { 1004 struct task_struct *p = find_lock_task_mm(task); 1005 1006 if (p) { 1007 if (atomic_read(&p->mm->mm_users) > 1) { 1008 mm = p->mm; 1009 mmgrab(mm); 1010 } 1011 task_unlock(p); 1012 } 1013 } 1014 1015 task->signal->oom_score_adj = oom_adj; 1016 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE)) 1017 task->signal->oom_score_adj_min = (short)oom_adj; 1018 trace_oom_score_adj_update(task); 1019 1020 if (mm) { 1021 struct task_struct *p; 1022 1023 rcu_read_lock(); 1024 for_each_process(p) { 1025 if (same_thread_group(task, p)) 1026 continue; 1027 1028 /* do not touch kernel threads or the global init */ 1029 if (p->flags & PF_KTHREAD || is_global_init(p)) 1030 continue; 1031 1032 task_lock(p); 1033 if (!p->vfork_done && process_shares_mm(p, mm)) { 1034 p->signal->oom_score_adj = oom_adj; 1035 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE)) 1036 p->signal->oom_score_adj_min = (short)oom_adj; 1037 } 1038 task_unlock(p); 1039 } 1040 rcu_read_unlock(); 1041 mmdrop(mm); 1042 } 1043 err_unlock: 1044 mutex_unlock(&oom_adj_mutex); 1045 put_task_struct(task); 1046 return err; 1047 } 1048 1049 /* 1050 * /proc/pid/oom_adj exists solely for backwards compatibility with previous 1051 * kernels. The effective policy is defined by oom_score_adj, which has a 1052 * different scale: oom_adj grew exponentially and oom_score_adj grows linearly. 1053 * Values written to oom_adj are simply mapped linearly to oom_score_adj. 1054 * Processes that become oom disabled via oom_adj will still be oom disabled 1055 * with this implementation. 1056 * 1057 * oom_adj cannot be removed since existing userspace binaries use it. 1058 */ 1059 static ssize_t oom_adj_write(struct file *file, const char __user *buf, 1060 size_t count, loff_t *ppos) 1061 { 1062 char buffer[PROC_NUMBUF]; 1063 int oom_adj; 1064 int err; 1065 1066 memset(buffer, 0, sizeof(buffer)); 1067 if (count > sizeof(buffer) - 1) 1068 count = sizeof(buffer) - 1; 1069 if (copy_from_user(buffer, buf, count)) { 1070 err = -EFAULT; 1071 goto out; 1072 } 1073 1074 err = kstrtoint(strstrip(buffer), 0, &oom_adj); 1075 if (err) 1076 goto out; 1077 if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) && 1078 oom_adj != OOM_DISABLE) { 1079 err = -EINVAL; 1080 goto out; 1081 } 1082 1083 /* 1084 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum 1085 * value is always attainable. 1086 */ 1087 if (oom_adj == OOM_ADJUST_MAX) 1088 oom_adj = OOM_SCORE_ADJ_MAX; 1089 else 1090 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE; 1091 1092 err = __set_oom_adj(file, oom_adj, true); 1093 out: 1094 return err < 0 ? err : count; 1095 } 1096 1097 static const struct file_operations proc_oom_adj_operations = { 1098 .read = oom_adj_read, 1099 .write = oom_adj_write, 1100 .llseek = generic_file_llseek, 1101 }; 1102 1103 static ssize_t oom_score_adj_read(struct file *file, char __user *buf, 1104 size_t count, loff_t *ppos) 1105 { 1106 struct task_struct *task = get_proc_task(file_inode(file)); 1107 char buffer[PROC_NUMBUF]; 1108 short oom_score_adj = OOM_SCORE_ADJ_MIN; 1109 size_t len; 1110 1111 if (!task) 1112 return -ESRCH; 1113 oom_score_adj = task->signal->oom_score_adj; 1114 put_task_struct(task); 1115 len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj); 1116 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1117 } 1118 1119 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf, 1120 size_t count, loff_t *ppos) 1121 { 1122 char buffer[PROC_NUMBUF]; 1123 int oom_score_adj; 1124 int err; 1125 1126 memset(buffer, 0, sizeof(buffer)); 1127 if (count > sizeof(buffer) - 1) 1128 count = sizeof(buffer) - 1; 1129 if (copy_from_user(buffer, buf, count)) { 1130 err = -EFAULT; 1131 goto out; 1132 } 1133 1134 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj); 1135 if (err) 1136 goto out; 1137 if (oom_score_adj < OOM_SCORE_ADJ_MIN || 1138 oom_score_adj > OOM_SCORE_ADJ_MAX) { 1139 err = -EINVAL; 1140 goto out; 1141 } 1142 1143 err = __set_oom_adj(file, oom_score_adj, false); 1144 out: 1145 return err < 0 ? err : count; 1146 } 1147 1148 static const struct file_operations proc_oom_score_adj_operations = { 1149 .read = oom_score_adj_read, 1150 .write = oom_score_adj_write, 1151 .llseek = default_llseek, 1152 }; 1153 1154 #ifdef CONFIG_AUDIT 1155 #define TMPBUFLEN 11 1156 static ssize_t proc_loginuid_read(struct file * file, char __user * buf, 1157 size_t count, loff_t *ppos) 1158 { 1159 struct inode * inode = file_inode(file); 1160 struct task_struct *task = get_proc_task(inode); 1161 ssize_t length; 1162 char tmpbuf[TMPBUFLEN]; 1163 1164 if (!task) 1165 return -ESRCH; 1166 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 1167 from_kuid(file->f_cred->user_ns, 1168 audit_get_loginuid(task))); 1169 put_task_struct(task); 1170 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 1171 } 1172 1173 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf, 1174 size_t count, loff_t *ppos) 1175 { 1176 struct inode * inode = file_inode(file); 1177 uid_t loginuid; 1178 kuid_t kloginuid; 1179 int rv; 1180 1181 rcu_read_lock(); 1182 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) { 1183 rcu_read_unlock(); 1184 return -EPERM; 1185 } 1186 rcu_read_unlock(); 1187 1188 if (*ppos != 0) { 1189 /* No partial writes. */ 1190 return -EINVAL; 1191 } 1192 1193 rv = kstrtou32_from_user(buf, count, 10, &loginuid); 1194 if (rv < 0) 1195 return rv; 1196 1197 /* is userspace tring to explicitly UNSET the loginuid? */ 1198 if (loginuid == AUDIT_UID_UNSET) { 1199 kloginuid = INVALID_UID; 1200 } else { 1201 kloginuid = make_kuid(file->f_cred->user_ns, loginuid); 1202 if (!uid_valid(kloginuid)) 1203 return -EINVAL; 1204 } 1205 1206 rv = audit_set_loginuid(kloginuid); 1207 if (rv < 0) 1208 return rv; 1209 return count; 1210 } 1211 1212 static const struct file_operations proc_loginuid_operations = { 1213 .read = proc_loginuid_read, 1214 .write = proc_loginuid_write, 1215 .llseek = generic_file_llseek, 1216 }; 1217 1218 static ssize_t proc_sessionid_read(struct file * file, char __user * buf, 1219 size_t count, loff_t *ppos) 1220 { 1221 struct inode * inode = file_inode(file); 1222 struct task_struct *task = get_proc_task(inode); 1223 ssize_t length; 1224 char tmpbuf[TMPBUFLEN]; 1225 1226 if (!task) 1227 return -ESRCH; 1228 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 1229 audit_get_sessionid(task)); 1230 put_task_struct(task); 1231 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 1232 } 1233 1234 static const struct file_operations proc_sessionid_operations = { 1235 .read = proc_sessionid_read, 1236 .llseek = generic_file_llseek, 1237 }; 1238 #endif 1239 1240 #ifdef CONFIG_FAULT_INJECTION 1241 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf, 1242 size_t count, loff_t *ppos) 1243 { 1244 struct task_struct *task = get_proc_task(file_inode(file)); 1245 char buffer[PROC_NUMBUF]; 1246 size_t len; 1247 int make_it_fail; 1248 1249 if (!task) 1250 return -ESRCH; 1251 make_it_fail = task->make_it_fail; 1252 put_task_struct(task); 1253 1254 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail); 1255 1256 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1257 } 1258 1259 static ssize_t proc_fault_inject_write(struct file * file, 1260 const char __user * buf, size_t count, loff_t *ppos) 1261 { 1262 struct task_struct *task; 1263 char buffer[PROC_NUMBUF]; 1264 int make_it_fail; 1265 int rv; 1266 1267 if (!capable(CAP_SYS_RESOURCE)) 1268 return -EPERM; 1269 memset(buffer, 0, sizeof(buffer)); 1270 if (count > sizeof(buffer) - 1) 1271 count = sizeof(buffer) - 1; 1272 if (copy_from_user(buffer, buf, count)) 1273 return -EFAULT; 1274 rv = kstrtoint(strstrip(buffer), 0, &make_it_fail); 1275 if (rv < 0) 1276 return rv; 1277 if (make_it_fail < 0 || make_it_fail > 1) 1278 return -EINVAL; 1279 1280 task = get_proc_task(file_inode(file)); 1281 if (!task) 1282 return -ESRCH; 1283 task->make_it_fail = make_it_fail; 1284 put_task_struct(task); 1285 1286 return count; 1287 } 1288 1289 static const struct file_operations proc_fault_inject_operations = { 1290 .read = proc_fault_inject_read, 1291 .write = proc_fault_inject_write, 1292 .llseek = generic_file_llseek, 1293 }; 1294 1295 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf, 1296 size_t count, loff_t *ppos) 1297 { 1298 struct task_struct *task; 1299 int err; 1300 unsigned int n; 1301 1302 err = kstrtouint_from_user(buf, count, 0, &n); 1303 if (err) 1304 return err; 1305 1306 task = get_proc_task(file_inode(file)); 1307 if (!task) 1308 return -ESRCH; 1309 task->fail_nth = n; 1310 put_task_struct(task); 1311 1312 return count; 1313 } 1314 1315 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf, 1316 size_t count, loff_t *ppos) 1317 { 1318 struct task_struct *task; 1319 char numbuf[PROC_NUMBUF]; 1320 ssize_t len; 1321 1322 task = get_proc_task(file_inode(file)); 1323 if (!task) 1324 return -ESRCH; 1325 len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth); 1326 put_task_struct(task); 1327 return simple_read_from_buffer(buf, count, ppos, numbuf, len); 1328 } 1329 1330 static const struct file_operations proc_fail_nth_operations = { 1331 .read = proc_fail_nth_read, 1332 .write = proc_fail_nth_write, 1333 }; 1334 #endif 1335 1336 1337 #ifdef CONFIG_SCHED_DEBUG 1338 /* 1339 * Print out various scheduling related per-task fields: 1340 */ 1341 static int sched_show(struct seq_file *m, void *v) 1342 { 1343 struct inode *inode = m->private; 1344 struct pid_namespace *ns = proc_pid_ns(inode); 1345 struct task_struct *p; 1346 1347 p = get_proc_task(inode); 1348 if (!p) 1349 return -ESRCH; 1350 proc_sched_show_task(p, ns, m); 1351 1352 put_task_struct(p); 1353 1354 return 0; 1355 } 1356 1357 static ssize_t 1358 sched_write(struct file *file, const char __user *buf, 1359 size_t count, loff_t *offset) 1360 { 1361 struct inode *inode = file_inode(file); 1362 struct task_struct *p; 1363 1364 p = get_proc_task(inode); 1365 if (!p) 1366 return -ESRCH; 1367 proc_sched_set_task(p); 1368 1369 put_task_struct(p); 1370 1371 return count; 1372 } 1373 1374 static int sched_open(struct inode *inode, struct file *filp) 1375 { 1376 return single_open(filp, sched_show, inode); 1377 } 1378 1379 static const struct file_operations proc_pid_sched_operations = { 1380 .open = sched_open, 1381 .read = seq_read, 1382 .write = sched_write, 1383 .llseek = seq_lseek, 1384 .release = single_release, 1385 }; 1386 1387 #endif 1388 1389 #ifdef CONFIG_SCHED_AUTOGROUP 1390 /* 1391 * Print out autogroup related information: 1392 */ 1393 static int sched_autogroup_show(struct seq_file *m, void *v) 1394 { 1395 struct inode *inode = m->private; 1396 struct task_struct *p; 1397 1398 p = get_proc_task(inode); 1399 if (!p) 1400 return -ESRCH; 1401 proc_sched_autogroup_show_task(p, m); 1402 1403 put_task_struct(p); 1404 1405 return 0; 1406 } 1407 1408 static ssize_t 1409 sched_autogroup_write(struct file *file, const char __user *buf, 1410 size_t count, loff_t *offset) 1411 { 1412 struct inode *inode = file_inode(file); 1413 struct task_struct *p; 1414 char buffer[PROC_NUMBUF]; 1415 int nice; 1416 int err; 1417 1418 memset(buffer, 0, sizeof(buffer)); 1419 if (count > sizeof(buffer) - 1) 1420 count = sizeof(buffer) - 1; 1421 if (copy_from_user(buffer, buf, count)) 1422 return -EFAULT; 1423 1424 err = kstrtoint(strstrip(buffer), 0, &nice); 1425 if (err < 0) 1426 return err; 1427 1428 p = get_proc_task(inode); 1429 if (!p) 1430 return -ESRCH; 1431 1432 err = proc_sched_autogroup_set_nice(p, nice); 1433 if (err) 1434 count = err; 1435 1436 put_task_struct(p); 1437 1438 return count; 1439 } 1440 1441 static int sched_autogroup_open(struct inode *inode, struct file *filp) 1442 { 1443 int ret; 1444 1445 ret = single_open(filp, sched_autogroup_show, NULL); 1446 if (!ret) { 1447 struct seq_file *m = filp->private_data; 1448 1449 m->private = inode; 1450 } 1451 return ret; 1452 } 1453 1454 static const struct file_operations proc_pid_sched_autogroup_operations = { 1455 .open = sched_autogroup_open, 1456 .read = seq_read, 1457 .write = sched_autogroup_write, 1458 .llseek = seq_lseek, 1459 .release = single_release, 1460 }; 1461 1462 #endif /* CONFIG_SCHED_AUTOGROUP */ 1463 1464 static ssize_t comm_write(struct file *file, const char __user *buf, 1465 size_t count, loff_t *offset) 1466 { 1467 struct inode *inode = file_inode(file); 1468 struct task_struct *p; 1469 char buffer[TASK_COMM_LEN]; 1470 const size_t maxlen = sizeof(buffer) - 1; 1471 1472 memset(buffer, 0, sizeof(buffer)); 1473 if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count)) 1474 return -EFAULT; 1475 1476 p = get_proc_task(inode); 1477 if (!p) 1478 return -ESRCH; 1479 1480 if (same_thread_group(current, p)) 1481 set_task_comm(p, buffer); 1482 else 1483 count = -EINVAL; 1484 1485 put_task_struct(p); 1486 1487 return count; 1488 } 1489 1490 static int comm_show(struct seq_file *m, void *v) 1491 { 1492 struct inode *inode = m->private; 1493 struct task_struct *p; 1494 1495 p = get_proc_task(inode); 1496 if (!p) 1497 return -ESRCH; 1498 1499 proc_task_name(m, p, false); 1500 seq_putc(m, '\n'); 1501 1502 put_task_struct(p); 1503 1504 return 0; 1505 } 1506 1507 static int comm_open(struct inode *inode, struct file *filp) 1508 { 1509 return single_open(filp, comm_show, inode); 1510 } 1511 1512 static const struct file_operations proc_pid_set_comm_operations = { 1513 .open = comm_open, 1514 .read = seq_read, 1515 .write = comm_write, 1516 .llseek = seq_lseek, 1517 .release = single_release, 1518 }; 1519 1520 static int proc_exe_link(struct dentry *dentry, struct path *exe_path) 1521 { 1522 struct task_struct *task; 1523 struct file *exe_file; 1524 1525 task = get_proc_task(d_inode(dentry)); 1526 if (!task) 1527 return -ENOENT; 1528 exe_file = get_task_exe_file(task); 1529 put_task_struct(task); 1530 if (exe_file) { 1531 *exe_path = exe_file->f_path; 1532 path_get(&exe_file->f_path); 1533 fput(exe_file); 1534 return 0; 1535 } else 1536 return -ENOENT; 1537 } 1538 1539 static const char *proc_pid_get_link(struct dentry *dentry, 1540 struct inode *inode, 1541 struct delayed_call *done) 1542 { 1543 struct path path; 1544 int error = -EACCES; 1545 1546 if (!dentry) 1547 return ERR_PTR(-ECHILD); 1548 1549 /* Are we allowed to snoop on the tasks file descriptors? */ 1550 if (!proc_fd_access_allowed(inode)) 1551 goto out; 1552 1553 error = PROC_I(inode)->op.proc_get_link(dentry, &path); 1554 if (error) 1555 goto out; 1556 1557 nd_jump_link(&path); 1558 return NULL; 1559 out: 1560 return ERR_PTR(error); 1561 } 1562 1563 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen) 1564 { 1565 char *tmp = (char *)__get_free_page(GFP_KERNEL); 1566 char *pathname; 1567 int len; 1568 1569 if (!tmp) 1570 return -ENOMEM; 1571 1572 pathname = d_path(path, tmp, PAGE_SIZE); 1573 len = PTR_ERR(pathname); 1574 if (IS_ERR(pathname)) 1575 goto out; 1576 len = tmp + PAGE_SIZE - 1 - pathname; 1577 1578 if (len > buflen) 1579 len = buflen; 1580 if (copy_to_user(buffer, pathname, len)) 1581 len = -EFAULT; 1582 out: 1583 free_page((unsigned long)tmp); 1584 return len; 1585 } 1586 1587 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen) 1588 { 1589 int error = -EACCES; 1590 struct inode *inode = d_inode(dentry); 1591 struct path path; 1592 1593 /* Are we allowed to snoop on the tasks file descriptors? */ 1594 if (!proc_fd_access_allowed(inode)) 1595 goto out; 1596 1597 error = PROC_I(inode)->op.proc_get_link(dentry, &path); 1598 if (error) 1599 goto out; 1600 1601 error = do_proc_readlink(&path, buffer, buflen); 1602 path_put(&path); 1603 out: 1604 return error; 1605 } 1606 1607 const struct inode_operations proc_pid_link_inode_operations = { 1608 .readlink = proc_pid_readlink, 1609 .get_link = proc_pid_get_link, 1610 .setattr = proc_setattr, 1611 }; 1612 1613 1614 /* building an inode */ 1615 1616 void task_dump_owner(struct task_struct *task, umode_t mode, 1617 kuid_t *ruid, kgid_t *rgid) 1618 { 1619 /* Depending on the state of dumpable compute who should own a 1620 * proc file for a task. 1621 */ 1622 const struct cred *cred; 1623 kuid_t uid; 1624 kgid_t gid; 1625 1626 if (unlikely(task->flags & PF_KTHREAD)) { 1627 *ruid = GLOBAL_ROOT_UID; 1628 *rgid = GLOBAL_ROOT_GID; 1629 return; 1630 } 1631 1632 /* Default to the tasks effective ownership */ 1633 rcu_read_lock(); 1634 cred = __task_cred(task); 1635 uid = cred->euid; 1636 gid = cred->egid; 1637 rcu_read_unlock(); 1638 1639 /* 1640 * Before the /proc/pid/status file was created the only way to read 1641 * the effective uid of a /process was to stat /proc/pid. Reading 1642 * /proc/pid/status is slow enough that procps and other packages 1643 * kept stating /proc/pid. To keep the rules in /proc simple I have 1644 * made this apply to all per process world readable and executable 1645 * directories. 1646 */ 1647 if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) { 1648 struct mm_struct *mm; 1649 task_lock(task); 1650 mm = task->mm; 1651 /* Make non-dumpable tasks owned by some root */ 1652 if (mm) { 1653 if (get_dumpable(mm) != SUID_DUMP_USER) { 1654 struct user_namespace *user_ns = mm->user_ns; 1655 1656 uid = make_kuid(user_ns, 0); 1657 if (!uid_valid(uid)) 1658 uid = GLOBAL_ROOT_UID; 1659 1660 gid = make_kgid(user_ns, 0); 1661 if (!gid_valid(gid)) 1662 gid = GLOBAL_ROOT_GID; 1663 } 1664 } else { 1665 uid = GLOBAL_ROOT_UID; 1666 gid = GLOBAL_ROOT_GID; 1667 } 1668 task_unlock(task); 1669 } 1670 *ruid = uid; 1671 *rgid = gid; 1672 } 1673 1674 struct inode *proc_pid_make_inode(struct super_block * sb, 1675 struct task_struct *task, umode_t mode) 1676 { 1677 struct inode * inode; 1678 struct proc_inode *ei; 1679 1680 /* We need a new inode */ 1681 1682 inode = new_inode(sb); 1683 if (!inode) 1684 goto out; 1685 1686 /* Common stuff */ 1687 ei = PROC_I(inode); 1688 inode->i_mode = mode; 1689 inode->i_ino = get_next_ino(); 1690 inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); 1691 inode->i_op = &proc_def_inode_operations; 1692 1693 /* 1694 * grab the reference to task. 1695 */ 1696 ei->pid = get_task_pid(task, PIDTYPE_PID); 1697 if (!ei->pid) 1698 goto out_unlock; 1699 1700 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid); 1701 security_task_to_inode(task, inode); 1702 1703 out: 1704 return inode; 1705 1706 out_unlock: 1707 iput(inode); 1708 return NULL; 1709 } 1710 1711 int pid_getattr(const struct path *path, struct kstat *stat, 1712 u32 request_mask, unsigned int query_flags) 1713 { 1714 struct inode *inode = d_inode(path->dentry); 1715 struct pid_namespace *pid = proc_pid_ns(inode); 1716 struct task_struct *task; 1717 1718 generic_fillattr(inode, stat); 1719 1720 stat->uid = GLOBAL_ROOT_UID; 1721 stat->gid = GLOBAL_ROOT_GID; 1722 rcu_read_lock(); 1723 task = pid_task(proc_pid(inode), PIDTYPE_PID); 1724 if (task) { 1725 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) { 1726 rcu_read_unlock(); 1727 /* 1728 * This doesn't prevent learning whether PID exists, 1729 * it only makes getattr() consistent with readdir(). 1730 */ 1731 return -ENOENT; 1732 } 1733 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid); 1734 } 1735 rcu_read_unlock(); 1736 return 0; 1737 } 1738 1739 /* dentry stuff */ 1740 1741 /* 1742 * Set <pid>/... inode ownership (can change due to setuid(), etc.) 1743 */ 1744 void pid_update_inode(struct task_struct *task, struct inode *inode) 1745 { 1746 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid); 1747 1748 inode->i_mode &= ~(S_ISUID | S_ISGID); 1749 security_task_to_inode(task, inode); 1750 } 1751 1752 /* 1753 * Rewrite the inode's ownerships here because the owning task may have 1754 * performed a setuid(), etc. 1755 * 1756 */ 1757 static int pid_revalidate(struct dentry *dentry, unsigned int flags) 1758 { 1759 struct inode *inode; 1760 struct task_struct *task; 1761 1762 if (flags & LOOKUP_RCU) 1763 return -ECHILD; 1764 1765 inode = d_inode(dentry); 1766 task = get_proc_task(inode); 1767 1768 if (task) { 1769 pid_update_inode(task, inode); 1770 put_task_struct(task); 1771 return 1; 1772 } 1773 return 0; 1774 } 1775 1776 static inline bool proc_inode_is_dead(struct inode *inode) 1777 { 1778 return !proc_pid(inode)->tasks[PIDTYPE_PID].first; 1779 } 1780 1781 int pid_delete_dentry(const struct dentry *dentry) 1782 { 1783 /* Is the task we represent dead? 1784 * If so, then don't put the dentry on the lru list, 1785 * kill it immediately. 1786 */ 1787 return proc_inode_is_dead(d_inode(dentry)); 1788 } 1789 1790 const struct dentry_operations pid_dentry_operations = 1791 { 1792 .d_revalidate = pid_revalidate, 1793 .d_delete = pid_delete_dentry, 1794 }; 1795 1796 /* Lookups */ 1797 1798 /* 1799 * Fill a directory entry. 1800 * 1801 * If possible create the dcache entry and derive our inode number and 1802 * file type from dcache entry. 1803 * 1804 * Since all of the proc inode numbers are dynamically generated, the inode 1805 * numbers do not exist until the inode is cache. This means creating the 1806 * the dcache entry in readdir is necessary to keep the inode numbers 1807 * reported by readdir in sync with the inode numbers reported 1808 * by stat. 1809 */ 1810 bool proc_fill_cache(struct file *file, struct dir_context *ctx, 1811 const char *name, unsigned int len, 1812 instantiate_t instantiate, struct task_struct *task, const void *ptr) 1813 { 1814 struct dentry *child, *dir = file->f_path.dentry; 1815 struct qstr qname = QSTR_INIT(name, len); 1816 struct inode *inode; 1817 unsigned type = DT_UNKNOWN; 1818 ino_t ino = 1; 1819 1820 child = d_hash_and_lookup(dir, &qname); 1821 if (!child) { 1822 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 1823 child = d_alloc_parallel(dir, &qname, &wq); 1824 if (IS_ERR(child)) 1825 goto end_instantiate; 1826 if (d_in_lookup(child)) { 1827 struct dentry *res; 1828 res = instantiate(child, task, ptr); 1829 d_lookup_done(child); 1830 if (unlikely(res)) { 1831 dput(child); 1832 child = res; 1833 if (IS_ERR(child)) 1834 goto end_instantiate; 1835 } 1836 } 1837 } 1838 inode = d_inode(child); 1839 ino = inode->i_ino; 1840 type = inode->i_mode >> 12; 1841 dput(child); 1842 end_instantiate: 1843 return dir_emit(ctx, name, len, ino, type); 1844 } 1845 1846 /* 1847 * dname_to_vma_addr - maps a dentry name into two unsigned longs 1848 * which represent vma start and end addresses. 1849 */ 1850 static int dname_to_vma_addr(struct dentry *dentry, 1851 unsigned long *start, unsigned long *end) 1852 { 1853 const char *str = dentry->d_name.name; 1854 unsigned long long sval, eval; 1855 unsigned int len; 1856 1857 if (str[0] == '0' && str[1] != '-') 1858 return -EINVAL; 1859 len = _parse_integer(str, 16, &sval); 1860 if (len & KSTRTOX_OVERFLOW) 1861 return -EINVAL; 1862 if (sval != (unsigned long)sval) 1863 return -EINVAL; 1864 str += len; 1865 1866 if (*str != '-') 1867 return -EINVAL; 1868 str++; 1869 1870 if (str[0] == '0' && str[1]) 1871 return -EINVAL; 1872 len = _parse_integer(str, 16, &eval); 1873 if (len & KSTRTOX_OVERFLOW) 1874 return -EINVAL; 1875 if (eval != (unsigned long)eval) 1876 return -EINVAL; 1877 str += len; 1878 1879 if (*str != '\0') 1880 return -EINVAL; 1881 1882 *start = sval; 1883 *end = eval; 1884 1885 return 0; 1886 } 1887 1888 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags) 1889 { 1890 unsigned long vm_start, vm_end; 1891 bool exact_vma_exists = false; 1892 struct mm_struct *mm = NULL; 1893 struct task_struct *task; 1894 struct inode *inode; 1895 int status = 0; 1896 1897 if (flags & LOOKUP_RCU) 1898 return -ECHILD; 1899 1900 inode = d_inode(dentry); 1901 task = get_proc_task(inode); 1902 if (!task) 1903 goto out_notask; 1904 1905 mm = mm_access(task, PTRACE_MODE_READ_FSCREDS); 1906 if (IS_ERR_OR_NULL(mm)) 1907 goto out; 1908 1909 if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) { 1910 down_read(&mm->mmap_sem); 1911 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end); 1912 up_read(&mm->mmap_sem); 1913 } 1914 1915 mmput(mm); 1916 1917 if (exact_vma_exists) { 1918 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid); 1919 1920 security_task_to_inode(task, inode); 1921 status = 1; 1922 } 1923 1924 out: 1925 put_task_struct(task); 1926 1927 out_notask: 1928 return status; 1929 } 1930 1931 static const struct dentry_operations tid_map_files_dentry_operations = { 1932 .d_revalidate = map_files_d_revalidate, 1933 .d_delete = pid_delete_dentry, 1934 }; 1935 1936 static int map_files_get_link(struct dentry *dentry, struct path *path) 1937 { 1938 unsigned long vm_start, vm_end; 1939 struct vm_area_struct *vma; 1940 struct task_struct *task; 1941 struct mm_struct *mm; 1942 int rc; 1943 1944 rc = -ENOENT; 1945 task = get_proc_task(d_inode(dentry)); 1946 if (!task) 1947 goto out; 1948 1949 mm = get_task_mm(task); 1950 put_task_struct(task); 1951 if (!mm) 1952 goto out; 1953 1954 rc = dname_to_vma_addr(dentry, &vm_start, &vm_end); 1955 if (rc) 1956 goto out_mmput; 1957 1958 rc = -ENOENT; 1959 down_read(&mm->mmap_sem); 1960 vma = find_exact_vma(mm, vm_start, vm_end); 1961 if (vma && vma->vm_file) { 1962 *path = vma->vm_file->f_path; 1963 path_get(path); 1964 rc = 0; 1965 } 1966 up_read(&mm->mmap_sem); 1967 1968 out_mmput: 1969 mmput(mm); 1970 out: 1971 return rc; 1972 } 1973 1974 struct map_files_info { 1975 unsigned long start; 1976 unsigned long end; 1977 fmode_t mode; 1978 }; 1979 1980 /* 1981 * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the 1982 * symlinks may be used to bypass permissions on ancestor directories in the 1983 * path to the file in question. 1984 */ 1985 static const char * 1986 proc_map_files_get_link(struct dentry *dentry, 1987 struct inode *inode, 1988 struct delayed_call *done) 1989 { 1990 if (!capable(CAP_SYS_ADMIN)) 1991 return ERR_PTR(-EPERM); 1992 1993 return proc_pid_get_link(dentry, inode, done); 1994 } 1995 1996 /* 1997 * Identical to proc_pid_link_inode_operations except for get_link() 1998 */ 1999 static const struct inode_operations proc_map_files_link_inode_operations = { 2000 .readlink = proc_pid_readlink, 2001 .get_link = proc_map_files_get_link, 2002 .setattr = proc_setattr, 2003 }; 2004 2005 static struct dentry * 2006 proc_map_files_instantiate(struct dentry *dentry, 2007 struct task_struct *task, const void *ptr) 2008 { 2009 fmode_t mode = (fmode_t)(unsigned long)ptr; 2010 struct proc_inode *ei; 2011 struct inode *inode; 2012 2013 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK | 2014 ((mode & FMODE_READ ) ? S_IRUSR : 0) | 2015 ((mode & FMODE_WRITE) ? S_IWUSR : 0)); 2016 if (!inode) 2017 return ERR_PTR(-ENOENT); 2018 2019 ei = PROC_I(inode); 2020 ei->op.proc_get_link = map_files_get_link; 2021 2022 inode->i_op = &proc_map_files_link_inode_operations; 2023 inode->i_size = 64; 2024 2025 d_set_d_op(dentry, &tid_map_files_dentry_operations); 2026 return d_splice_alias(inode, dentry); 2027 } 2028 2029 static struct dentry *proc_map_files_lookup(struct inode *dir, 2030 struct dentry *dentry, unsigned int flags) 2031 { 2032 unsigned long vm_start, vm_end; 2033 struct vm_area_struct *vma; 2034 struct task_struct *task; 2035 struct dentry *result; 2036 struct mm_struct *mm; 2037 2038 result = ERR_PTR(-ENOENT); 2039 task = get_proc_task(dir); 2040 if (!task) 2041 goto out; 2042 2043 result = ERR_PTR(-EACCES); 2044 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) 2045 goto out_put_task; 2046 2047 result = ERR_PTR(-ENOENT); 2048 if (dname_to_vma_addr(dentry, &vm_start, &vm_end)) 2049 goto out_put_task; 2050 2051 mm = get_task_mm(task); 2052 if (!mm) 2053 goto out_put_task; 2054 2055 down_read(&mm->mmap_sem); 2056 vma = find_exact_vma(mm, vm_start, vm_end); 2057 if (!vma) 2058 goto out_no_vma; 2059 2060 if (vma->vm_file) 2061 result = proc_map_files_instantiate(dentry, task, 2062 (void *)(unsigned long)vma->vm_file->f_mode); 2063 2064 out_no_vma: 2065 up_read(&mm->mmap_sem); 2066 mmput(mm); 2067 out_put_task: 2068 put_task_struct(task); 2069 out: 2070 return result; 2071 } 2072 2073 static const struct inode_operations proc_map_files_inode_operations = { 2074 .lookup = proc_map_files_lookup, 2075 .permission = proc_fd_permission, 2076 .setattr = proc_setattr, 2077 }; 2078 2079 static int 2080 proc_map_files_readdir(struct file *file, struct dir_context *ctx) 2081 { 2082 struct vm_area_struct *vma; 2083 struct task_struct *task; 2084 struct mm_struct *mm; 2085 unsigned long nr_files, pos, i; 2086 GENRADIX(struct map_files_info) fa; 2087 struct map_files_info *p; 2088 int ret; 2089 2090 genradix_init(&fa); 2091 2092 ret = -ENOENT; 2093 task = get_proc_task(file_inode(file)); 2094 if (!task) 2095 goto out; 2096 2097 ret = -EACCES; 2098 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) 2099 goto out_put_task; 2100 2101 ret = 0; 2102 if (!dir_emit_dots(file, ctx)) 2103 goto out_put_task; 2104 2105 mm = get_task_mm(task); 2106 if (!mm) 2107 goto out_put_task; 2108 down_read(&mm->mmap_sem); 2109 2110 nr_files = 0; 2111 2112 /* 2113 * We need two passes here: 2114 * 2115 * 1) Collect vmas of mapped files with mmap_sem taken 2116 * 2) Release mmap_sem and instantiate entries 2117 * 2118 * otherwise we get lockdep complained, since filldir() 2119 * routine might require mmap_sem taken in might_fault(). 2120 */ 2121 2122 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) { 2123 if (!vma->vm_file) 2124 continue; 2125 if (++pos <= ctx->pos) 2126 continue; 2127 2128 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL); 2129 if (!p) { 2130 ret = -ENOMEM; 2131 up_read(&mm->mmap_sem); 2132 mmput(mm); 2133 goto out_put_task; 2134 } 2135 2136 p->start = vma->vm_start; 2137 p->end = vma->vm_end; 2138 p->mode = vma->vm_file->f_mode; 2139 } 2140 up_read(&mm->mmap_sem); 2141 mmput(mm); 2142 2143 for (i = 0; i < nr_files; i++) { 2144 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */ 2145 unsigned int len; 2146 2147 p = genradix_ptr(&fa, i); 2148 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end); 2149 if (!proc_fill_cache(file, ctx, 2150 buf, len, 2151 proc_map_files_instantiate, 2152 task, 2153 (void *)(unsigned long)p->mode)) 2154 break; 2155 ctx->pos++; 2156 } 2157 2158 out_put_task: 2159 put_task_struct(task); 2160 out: 2161 genradix_free(&fa); 2162 return ret; 2163 } 2164 2165 static const struct file_operations proc_map_files_operations = { 2166 .read = generic_read_dir, 2167 .iterate_shared = proc_map_files_readdir, 2168 .llseek = generic_file_llseek, 2169 }; 2170 2171 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS) 2172 struct timers_private { 2173 struct pid *pid; 2174 struct task_struct *task; 2175 struct sighand_struct *sighand; 2176 struct pid_namespace *ns; 2177 unsigned long flags; 2178 }; 2179 2180 static void *timers_start(struct seq_file *m, loff_t *pos) 2181 { 2182 struct timers_private *tp = m->private; 2183 2184 tp->task = get_pid_task(tp->pid, PIDTYPE_PID); 2185 if (!tp->task) 2186 return ERR_PTR(-ESRCH); 2187 2188 tp->sighand = lock_task_sighand(tp->task, &tp->flags); 2189 if (!tp->sighand) 2190 return ERR_PTR(-ESRCH); 2191 2192 return seq_list_start(&tp->task->signal->posix_timers, *pos); 2193 } 2194 2195 static void *timers_next(struct seq_file *m, void *v, loff_t *pos) 2196 { 2197 struct timers_private *tp = m->private; 2198 return seq_list_next(v, &tp->task->signal->posix_timers, pos); 2199 } 2200 2201 static void timers_stop(struct seq_file *m, void *v) 2202 { 2203 struct timers_private *tp = m->private; 2204 2205 if (tp->sighand) { 2206 unlock_task_sighand(tp->task, &tp->flags); 2207 tp->sighand = NULL; 2208 } 2209 2210 if (tp->task) { 2211 put_task_struct(tp->task); 2212 tp->task = NULL; 2213 } 2214 } 2215 2216 static int show_timer(struct seq_file *m, void *v) 2217 { 2218 struct k_itimer *timer; 2219 struct timers_private *tp = m->private; 2220 int notify; 2221 static const char * const nstr[] = { 2222 [SIGEV_SIGNAL] = "signal", 2223 [SIGEV_NONE] = "none", 2224 [SIGEV_THREAD] = "thread", 2225 }; 2226 2227 timer = list_entry((struct list_head *)v, struct k_itimer, list); 2228 notify = timer->it_sigev_notify; 2229 2230 seq_printf(m, "ID: %d\n", timer->it_id); 2231 seq_printf(m, "signal: %d/%px\n", 2232 timer->sigq->info.si_signo, 2233 timer->sigq->info.si_value.sival_ptr); 2234 seq_printf(m, "notify: %s/%s.%d\n", 2235 nstr[notify & ~SIGEV_THREAD_ID], 2236 (notify & SIGEV_THREAD_ID) ? "tid" : "pid", 2237 pid_nr_ns(timer->it_pid, tp->ns)); 2238 seq_printf(m, "ClockID: %d\n", timer->it_clock); 2239 2240 return 0; 2241 } 2242 2243 static const struct seq_operations proc_timers_seq_ops = { 2244 .start = timers_start, 2245 .next = timers_next, 2246 .stop = timers_stop, 2247 .show = show_timer, 2248 }; 2249 2250 static int proc_timers_open(struct inode *inode, struct file *file) 2251 { 2252 struct timers_private *tp; 2253 2254 tp = __seq_open_private(file, &proc_timers_seq_ops, 2255 sizeof(struct timers_private)); 2256 if (!tp) 2257 return -ENOMEM; 2258 2259 tp->pid = proc_pid(inode); 2260 tp->ns = proc_pid_ns(inode); 2261 return 0; 2262 } 2263 2264 static const struct file_operations proc_timers_operations = { 2265 .open = proc_timers_open, 2266 .read = seq_read, 2267 .llseek = seq_lseek, 2268 .release = seq_release_private, 2269 }; 2270 #endif 2271 2272 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf, 2273 size_t count, loff_t *offset) 2274 { 2275 struct inode *inode = file_inode(file); 2276 struct task_struct *p; 2277 u64 slack_ns; 2278 int err; 2279 2280 err = kstrtoull_from_user(buf, count, 10, &slack_ns); 2281 if (err < 0) 2282 return err; 2283 2284 p = get_proc_task(inode); 2285 if (!p) 2286 return -ESRCH; 2287 2288 if (p != current) { 2289 rcu_read_lock(); 2290 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { 2291 rcu_read_unlock(); 2292 count = -EPERM; 2293 goto out; 2294 } 2295 rcu_read_unlock(); 2296 2297 err = security_task_setscheduler(p); 2298 if (err) { 2299 count = err; 2300 goto out; 2301 } 2302 } 2303 2304 task_lock(p); 2305 if (slack_ns == 0) 2306 p->timer_slack_ns = p->default_timer_slack_ns; 2307 else 2308 p->timer_slack_ns = slack_ns; 2309 task_unlock(p); 2310 2311 out: 2312 put_task_struct(p); 2313 2314 return count; 2315 } 2316 2317 static int timerslack_ns_show(struct seq_file *m, void *v) 2318 { 2319 struct inode *inode = m->private; 2320 struct task_struct *p; 2321 int err = 0; 2322 2323 p = get_proc_task(inode); 2324 if (!p) 2325 return -ESRCH; 2326 2327 if (p != current) { 2328 rcu_read_lock(); 2329 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) { 2330 rcu_read_unlock(); 2331 err = -EPERM; 2332 goto out; 2333 } 2334 rcu_read_unlock(); 2335 2336 err = security_task_getscheduler(p); 2337 if (err) 2338 goto out; 2339 } 2340 2341 task_lock(p); 2342 seq_printf(m, "%llu\n", p->timer_slack_ns); 2343 task_unlock(p); 2344 2345 out: 2346 put_task_struct(p); 2347 2348 return err; 2349 } 2350 2351 static int timerslack_ns_open(struct inode *inode, struct file *filp) 2352 { 2353 return single_open(filp, timerslack_ns_show, inode); 2354 } 2355 2356 static const struct file_operations proc_pid_set_timerslack_ns_operations = { 2357 .open = timerslack_ns_open, 2358 .read = seq_read, 2359 .write = timerslack_ns_write, 2360 .llseek = seq_lseek, 2361 .release = single_release, 2362 }; 2363 2364 static struct dentry *proc_pident_instantiate(struct dentry *dentry, 2365 struct task_struct *task, const void *ptr) 2366 { 2367 const struct pid_entry *p = ptr; 2368 struct inode *inode; 2369 struct proc_inode *ei; 2370 2371 inode = proc_pid_make_inode(dentry->d_sb, task, p->mode); 2372 if (!inode) 2373 return ERR_PTR(-ENOENT); 2374 2375 ei = PROC_I(inode); 2376 if (S_ISDIR(inode->i_mode)) 2377 set_nlink(inode, 2); /* Use getattr to fix if necessary */ 2378 if (p->iop) 2379 inode->i_op = p->iop; 2380 if (p->fop) 2381 inode->i_fop = p->fop; 2382 ei->op = p->op; 2383 pid_update_inode(task, inode); 2384 d_set_d_op(dentry, &pid_dentry_operations); 2385 return d_splice_alias(inode, dentry); 2386 } 2387 2388 static struct dentry *proc_pident_lookup(struct inode *dir, 2389 struct dentry *dentry, 2390 const struct pid_entry *p, 2391 const struct pid_entry *end) 2392 { 2393 struct task_struct *task = get_proc_task(dir); 2394 struct dentry *res = ERR_PTR(-ENOENT); 2395 2396 if (!task) 2397 goto out_no_task; 2398 2399 /* 2400 * Yes, it does not scale. And it should not. Don't add 2401 * new entries into /proc/<tgid>/ without very good reasons. 2402 */ 2403 for (; p < end; p++) { 2404 if (p->len != dentry->d_name.len) 2405 continue; 2406 if (!memcmp(dentry->d_name.name, p->name, p->len)) { 2407 res = proc_pident_instantiate(dentry, task, p); 2408 break; 2409 } 2410 } 2411 put_task_struct(task); 2412 out_no_task: 2413 return res; 2414 } 2415 2416 static int proc_pident_readdir(struct file *file, struct dir_context *ctx, 2417 const struct pid_entry *ents, unsigned int nents) 2418 { 2419 struct task_struct *task = get_proc_task(file_inode(file)); 2420 const struct pid_entry *p; 2421 2422 if (!task) 2423 return -ENOENT; 2424 2425 if (!dir_emit_dots(file, ctx)) 2426 goto out; 2427 2428 if (ctx->pos >= nents + 2) 2429 goto out; 2430 2431 for (p = ents + (ctx->pos - 2); p < ents + nents; p++) { 2432 if (!proc_fill_cache(file, ctx, p->name, p->len, 2433 proc_pident_instantiate, task, p)) 2434 break; 2435 ctx->pos++; 2436 } 2437 out: 2438 put_task_struct(task); 2439 return 0; 2440 } 2441 2442 #ifdef CONFIG_SECURITY 2443 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf, 2444 size_t count, loff_t *ppos) 2445 { 2446 struct inode * inode = file_inode(file); 2447 char *p = NULL; 2448 ssize_t length; 2449 struct task_struct *task = get_proc_task(inode); 2450 2451 if (!task) 2452 return -ESRCH; 2453 2454 length = security_getprocattr(task, PROC_I(inode)->op.lsm, 2455 (char*)file->f_path.dentry->d_name.name, 2456 &p); 2457 put_task_struct(task); 2458 if (length > 0) 2459 length = simple_read_from_buffer(buf, count, ppos, p, length); 2460 kfree(p); 2461 return length; 2462 } 2463 2464 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf, 2465 size_t count, loff_t *ppos) 2466 { 2467 struct inode * inode = file_inode(file); 2468 struct task_struct *task; 2469 void *page; 2470 int rv; 2471 2472 rcu_read_lock(); 2473 task = pid_task(proc_pid(inode), PIDTYPE_PID); 2474 if (!task) { 2475 rcu_read_unlock(); 2476 return -ESRCH; 2477 } 2478 /* A task may only write its own attributes. */ 2479 if (current != task) { 2480 rcu_read_unlock(); 2481 return -EACCES; 2482 } 2483 /* Prevent changes to overridden credentials. */ 2484 if (current_cred() != current_real_cred()) { 2485 rcu_read_unlock(); 2486 return -EBUSY; 2487 } 2488 rcu_read_unlock(); 2489 2490 if (count > PAGE_SIZE) 2491 count = PAGE_SIZE; 2492 2493 /* No partial writes. */ 2494 if (*ppos != 0) 2495 return -EINVAL; 2496 2497 page = memdup_user(buf, count); 2498 if (IS_ERR(page)) { 2499 rv = PTR_ERR(page); 2500 goto out; 2501 } 2502 2503 /* Guard against adverse ptrace interaction */ 2504 rv = mutex_lock_interruptible(¤t->signal->cred_guard_mutex); 2505 if (rv < 0) 2506 goto out_free; 2507 2508 rv = security_setprocattr(PROC_I(inode)->op.lsm, 2509 file->f_path.dentry->d_name.name, page, 2510 count); 2511 mutex_unlock(¤t->signal->cred_guard_mutex); 2512 out_free: 2513 kfree(page); 2514 out: 2515 return rv; 2516 } 2517 2518 static const struct file_operations proc_pid_attr_operations = { 2519 .read = proc_pid_attr_read, 2520 .write = proc_pid_attr_write, 2521 .llseek = generic_file_llseek, 2522 }; 2523 2524 #define LSM_DIR_OPS(LSM) \ 2525 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \ 2526 struct dir_context *ctx) \ 2527 { \ 2528 return proc_pident_readdir(filp, ctx, \ 2529 LSM##_attr_dir_stuff, \ 2530 ARRAY_SIZE(LSM##_attr_dir_stuff)); \ 2531 } \ 2532 \ 2533 static const struct file_operations proc_##LSM##_attr_dir_ops = { \ 2534 .read = generic_read_dir, \ 2535 .iterate = proc_##LSM##_attr_dir_iterate, \ 2536 .llseek = default_llseek, \ 2537 }; \ 2538 \ 2539 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \ 2540 struct dentry *dentry, unsigned int flags) \ 2541 { \ 2542 return proc_pident_lookup(dir, dentry, \ 2543 LSM##_attr_dir_stuff, \ 2544 LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \ 2545 } \ 2546 \ 2547 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \ 2548 .lookup = proc_##LSM##_attr_dir_lookup, \ 2549 .getattr = pid_getattr, \ 2550 .setattr = proc_setattr, \ 2551 } 2552 2553 #ifdef CONFIG_SECURITY_SMACK 2554 static const struct pid_entry smack_attr_dir_stuff[] = { 2555 ATTR("smack", "current", 0666), 2556 }; 2557 LSM_DIR_OPS(smack); 2558 #endif 2559 2560 static const struct pid_entry attr_dir_stuff[] = { 2561 ATTR(NULL, "current", 0666), 2562 ATTR(NULL, "prev", 0444), 2563 ATTR(NULL, "exec", 0666), 2564 ATTR(NULL, "fscreate", 0666), 2565 ATTR(NULL, "keycreate", 0666), 2566 ATTR(NULL, "sockcreate", 0666), 2567 #ifdef CONFIG_SECURITY_SMACK 2568 DIR("smack", 0555, 2569 proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops), 2570 #endif 2571 }; 2572 2573 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx) 2574 { 2575 return proc_pident_readdir(file, ctx, 2576 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff)); 2577 } 2578 2579 static const struct file_operations proc_attr_dir_operations = { 2580 .read = generic_read_dir, 2581 .iterate_shared = proc_attr_dir_readdir, 2582 .llseek = generic_file_llseek, 2583 }; 2584 2585 static struct dentry *proc_attr_dir_lookup(struct inode *dir, 2586 struct dentry *dentry, unsigned int flags) 2587 { 2588 return proc_pident_lookup(dir, dentry, 2589 attr_dir_stuff, 2590 attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff)); 2591 } 2592 2593 static const struct inode_operations proc_attr_dir_inode_operations = { 2594 .lookup = proc_attr_dir_lookup, 2595 .getattr = pid_getattr, 2596 .setattr = proc_setattr, 2597 }; 2598 2599 #endif 2600 2601 #ifdef CONFIG_ELF_CORE 2602 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf, 2603 size_t count, loff_t *ppos) 2604 { 2605 struct task_struct *task = get_proc_task(file_inode(file)); 2606 struct mm_struct *mm; 2607 char buffer[PROC_NUMBUF]; 2608 size_t len; 2609 int ret; 2610 2611 if (!task) 2612 return -ESRCH; 2613 2614 ret = 0; 2615 mm = get_task_mm(task); 2616 if (mm) { 2617 len = snprintf(buffer, sizeof(buffer), "%08lx\n", 2618 ((mm->flags & MMF_DUMP_FILTER_MASK) >> 2619 MMF_DUMP_FILTER_SHIFT)); 2620 mmput(mm); 2621 ret = simple_read_from_buffer(buf, count, ppos, buffer, len); 2622 } 2623 2624 put_task_struct(task); 2625 2626 return ret; 2627 } 2628 2629 static ssize_t proc_coredump_filter_write(struct file *file, 2630 const char __user *buf, 2631 size_t count, 2632 loff_t *ppos) 2633 { 2634 struct task_struct *task; 2635 struct mm_struct *mm; 2636 unsigned int val; 2637 int ret; 2638 int i; 2639 unsigned long mask; 2640 2641 ret = kstrtouint_from_user(buf, count, 0, &val); 2642 if (ret < 0) 2643 return ret; 2644 2645 ret = -ESRCH; 2646 task = get_proc_task(file_inode(file)); 2647 if (!task) 2648 goto out_no_task; 2649 2650 mm = get_task_mm(task); 2651 if (!mm) 2652 goto out_no_mm; 2653 ret = 0; 2654 2655 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) { 2656 if (val & mask) 2657 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2658 else 2659 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2660 } 2661 2662 mmput(mm); 2663 out_no_mm: 2664 put_task_struct(task); 2665 out_no_task: 2666 if (ret < 0) 2667 return ret; 2668 return count; 2669 } 2670 2671 static const struct file_operations proc_coredump_filter_operations = { 2672 .read = proc_coredump_filter_read, 2673 .write = proc_coredump_filter_write, 2674 .llseek = generic_file_llseek, 2675 }; 2676 #endif 2677 2678 #ifdef CONFIG_TASK_IO_ACCOUNTING 2679 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole) 2680 { 2681 struct task_io_accounting acct = task->ioac; 2682 unsigned long flags; 2683 int result; 2684 2685 result = mutex_lock_killable(&task->signal->cred_guard_mutex); 2686 if (result) 2687 return result; 2688 2689 if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) { 2690 result = -EACCES; 2691 goto out_unlock; 2692 } 2693 2694 if (whole && lock_task_sighand(task, &flags)) { 2695 struct task_struct *t = task; 2696 2697 task_io_accounting_add(&acct, &task->signal->ioac); 2698 while_each_thread(task, t) 2699 task_io_accounting_add(&acct, &t->ioac); 2700 2701 unlock_task_sighand(task, &flags); 2702 } 2703 seq_printf(m, 2704 "rchar: %llu\n" 2705 "wchar: %llu\n" 2706 "syscr: %llu\n" 2707 "syscw: %llu\n" 2708 "read_bytes: %llu\n" 2709 "write_bytes: %llu\n" 2710 "cancelled_write_bytes: %llu\n", 2711 (unsigned long long)acct.rchar, 2712 (unsigned long long)acct.wchar, 2713 (unsigned long long)acct.syscr, 2714 (unsigned long long)acct.syscw, 2715 (unsigned long long)acct.read_bytes, 2716 (unsigned long long)acct.write_bytes, 2717 (unsigned long long)acct.cancelled_write_bytes); 2718 result = 0; 2719 2720 out_unlock: 2721 mutex_unlock(&task->signal->cred_guard_mutex); 2722 return result; 2723 } 2724 2725 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns, 2726 struct pid *pid, struct task_struct *task) 2727 { 2728 return do_io_accounting(task, m, 0); 2729 } 2730 2731 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns, 2732 struct pid *pid, struct task_struct *task) 2733 { 2734 return do_io_accounting(task, m, 1); 2735 } 2736 #endif /* CONFIG_TASK_IO_ACCOUNTING */ 2737 2738 #ifdef CONFIG_USER_NS 2739 static int proc_id_map_open(struct inode *inode, struct file *file, 2740 const struct seq_operations *seq_ops) 2741 { 2742 struct user_namespace *ns = NULL; 2743 struct task_struct *task; 2744 struct seq_file *seq; 2745 int ret = -EINVAL; 2746 2747 task = get_proc_task(inode); 2748 if (task) { 2749 rcu_read_lock(); 2750 ns = get_user_ns(task_cred_xxx(task, user_ns)); 2751 rcu_read_unlock(); 2752 put_task_struct(task); 2753 } 2754 if (!ns) 2755 goto err; 2756 2757 ret = seq_open(file, seq_ops); 2758 if (ret) 2759 goto err_put_ns; 2760 2761 seq = file->private_data; 2762 seq->private = ns; 2763 2764 return 0; 2765 err_put_ns: 2766 put_user_ns(ns); 2767 err: 2768 return ret; 2769 } 2770 2771 static int proc_id_map_release(struct inode *inode, struct file *file) 2772 { 2773 struct seq_file *seq = file->private_data; 2774 struct user_namespace *ns = seq->private; 2775 put_user_ns(ns); 2776 return seq_release(inode, file); 2777 } 2778 2779 static int proc_uid_map_open(struct inode *inode, struct file *file) 2780 { 2781 return proc_id_map_open(inode, file, &proc_uid_seq_operations); 2782 } 2783 2784 static int proc_gid_map_open(struct inode *inode, struct file *file) 2785 { 2786 return proc_id_map_open(inode, file, &proc_gid_seq_operations); 2787 } 2788 2789 static int proc_projid_map_open(struct inode *inode, struct file *file) 2790 { 2791 return proc_id_map_open(inode, file, &proc_projid_seq_operations); 2792 } 2793 2794 static const struct file_operations proc_uid_map_operations = { 2795 .open = proc_uid_map_open, 2796 .write = proc_uid_map_write, 2797 .read = seq_read, 2798 .llseek = seq_lseek, 2799 .release = proc_id_map_release, 2800 }; 2801 2802 static const struct file_operations proc_gid_map_operations = { 2803 .open = proc_gid_map_open, 2804 .write = proc_gid_map_write, 2805 .read = seq_read, 2806 .llseek = seq_lseek, 2807 .release = proc_id_map_release, 2808 }; 2809 2810 static const struct file_operations proc_projid_map_operations = { 2811 .open = proc_projid_map_open, 2812 .write = proc_projid_map_write, 2813 .read = seq_read, 2814 .llseek = seq_lseek, 2815 .release = proc_id_map_release, 2816 }; 2817 2818 static int proc_setgroups_open(struct inode *inode, struct file *file) 2819 { 2820 struct user_namespace *ns = NULL; 2821 struct task_struct *task; 2822 int ret; 2823 2824 ret = -ESRCH; 2825 task = get_proc_task(inode); 2826 if (task) { 2827 rcu_read_lock(); 2828 ns = get_user_ns(task_cred_xxx(task, user_ns)); 2829 rcu_read_unlock(); 2830 put_task_struct(task); 2831 } 2832 if (!ns) 2833 goto err; 2834 2835 if (file->f_mode & FMODE_WRITE) { 2836 ret = -EACCES; 2837 if (!ns_capable(ns, CAP_SYS_ADMIN)) 2838 goto err_put_ns; 2839 } 2840 2841 ret = single_open(file, &proc_setgroups_show, ns); 2842 if (ret) 2843 goto err_put_ns; 2844 2845 return 0; 2846 err_put_ns: 2847 put_user_ns(ns); 2848 err: 2849 return ret; 2850 } 2851 2852 static int proc_setgroups_release(struct inode *inode, struct file *file) 2853 { 2854 struct seq_file *seq = file->private_data; 2855 struct user_namespace *ns = seq->private; 2856 int ret = single_release(inode, file); 2857 put_user_ns(ns); 2858 return ret; 2859 } 2860 2861 static const struct file_operations proc_setgroups_operations = { 2862 .open = proc_setgroups_open, 2863 .write = proc_setgroups_write, 2864 .read = seq_read, 2865 .llseek = seq_lseek, 2866 .release = proc_setgroups_release, 2867 }; 2868 #endif /* CONFIG_USER_NS */ 2869 2870 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns, 2871 struct pid *pid, struct task_struct *task) 2872 { 2873 int err = lock_trace(task); 2874 if (!err) { 2875 seq_printf(m, "%08x\n", task->personality); 2876 unlock_trace(task); 2877 } 2878 return err; 2879 } 2880 2881 #ifdef CONFIG_LIVEPATCH 2882 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns, 2883 struct pid *pid, struct task_struct *task) 2884 { 2885 seq_printf(m, "%d\n", task->patch_state); 2886 return 0; 2887 } 2888 #endif /* CONFIG_LIVEPATCH */ 2889 2890 #ifdef CONFIG_STACKLEAK_METRICS 2891 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns, 2892 struct pid *pid, struct task_struct *task) 2893 { 2894 unsigned long prev_depth = THREAD_SIZE - 2895 (task->prev_lowest_stack & (THREAD_SIZE - 1)); 2896 unsigned long depth = THREAD_SIZE - 2897 (task->lowest_stack & (THREAD_SIZE - 1)); 2898 2899 seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n", 2900 prev_depth, depth); 2901 return 0; 2902 } 2903 #endif /* CONFIG_STACKLEAK_METRICS */ 2904 2905 /* 2906 * Thread groups 2907 */ 2908 static const struct file_operations proc_task_operations; 2909 static const struct inode_operations proc_task_inode_operations; 2910 2911 static const struct pid_entry tgid_base_stuff[] = { 2912 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations), 2913 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations), 2914 DIR("map_files", S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations), 2915 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations), 2916 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations), 2917 #ifdef CONFIG_NET 2918 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations), 2919 #endif 2920 REG("environ", S_IRUSR, proc_environ_operations), 2921 REG("auxv", S_IRUSR, proc_auxv_operations), 2922 ONE("status", S_IRUGO, proc_pid_status), 2923 ONE("personality", S_IRUSR, proc_pid_personality), 2924 ONE("limits", S_IRUGO, proc_pid_limits), 2925 #ifdef CONFIG_SCHED_DEBUG 2926 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations), 2927 #endif 2928 #ifdef CONFIG_SCHED_AUTOGROUP 2929 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations), 2930 #endif 2931 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations), 2932 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 2933 ONE("syscall", S_IRUSR, proc_pid_syscall), 2934 #endif 2935 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops), 2936 ONE("stat", S_IRUGO, proc_tgid_stat), 2937 ONE("statm", S_IRUGO, proc_pid_statm), 2938 REG("maps", S_IRUGO, proc_pid_maps_operations), 2939 #ifdef CONFIG_NUMA 2940 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations), 2941 #endif 2942 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations), 2943 LNK("cwd", proc_cwd_link), 2944 LNK("root", proc_root_link), 2945 LNK("exe", proc_exe_link), 2946 REG("mounts", S_IRUGO, proc_mounts_operations), 2947 REG("mountinfo", S_IRUGO, proc_mountinfo_operations), 2948 REG("mountstats", S_IRUSR, proc_mountstats_operations), 2949 #ifdef CONFIG_PROC_PAGE_MONITOR 2950 REG("clear_refs", S_IWUSR, proc_clear_refs_operations), 2951 REG("smaps", S_IRUGO, proc_pid_smaps_operations), 2952 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations), 2953 REG("pagemap", S_IRUSR, proc_pagemap_operations), 2954 #endif 2955 #ifdef CONFIG_SECURITY 2956 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), 2957 #endif 2958 #ifdef CONFIG_KALLSYMS 2959 ONE("wchan", S_IRUGO, proc_pid_wchan), 2960 #endif 2961 #ifdef CONFIG_STACKTRACE 2962 ONE("stack", S_IRUSR, proc_pid_stack), 2963 #endif 2964 #ifdef CONFIG_SCHED_INFO 2965 ONE("schedstat", S_IRUGO, proc_pid_schedstat), 2966 #endif 2967 #ifdef CONFIG_LATENCYTOP 2968 REG("latency", S_IRUGO, proc_lstats_operations), 2969 #endif 2970 #ifdef CONFIG_PROC_PID_CPUSET 2971 ONE("cpuset", S_IRUGO, proc_cpuset_show), 2972 #endif 2973 #ifdef CONFIG_CGROUPS 2974 ONE("cgroup", S_IRUGO, proc_cgroup_show), 2975 #endif 2976 ONE("oom_score", S_IRUGO, proc_oom_score), 2977 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations), 2978 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations), 2979 #ifdef CONFIG_AUDIT 2980 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations), 2981 REG("sessionid", S_IRUGO, proc_sessionid_operations), 2982 #endif 2983 #ifdef CONFIG_FAULT_INJECTION 2984 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations), 2985 REG("fail-nth", 0644, proc_fail_nth_operations), 2986 #endif 2987 #ifdef CONFIG_ELF_CORE 2988 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations), 2989 #endif 2990 #ifdef CONFIG_TASK_IO_ACCOUNTING 2991 ONE("io", S_IRUSR, proc_tgid_io_accounting), 2992 #endif 2993 #ifdef CONFIG_USER_NS 2994 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations), 2995 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations), 2996 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations), 2997 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations), 2998 #endif 2999 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS) 3000 REG("timers", S_IRUGO, proc_timers_operations), 3001 #endif 3002 REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations), 3003 #ifdef CONFIG_LIVEPATCH 3004 ONE("patch_state", S_IRUSR, proc_pid_patch_state), 3005 #endif 3006 #ifdef CONFIG_STACKLEAK_METRICS 3007 ONE("stack_depth", S_IRUGO, proc_stack_depth), 3008 #endif 3009 }; 3010 3011 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx) 3012 { 3013 return proc_pident_readdir(file, ctx, 3014 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 3015 } 3016 3017 static const struct file_operations proc_tgid_base_operations = { 3018 .read = generic_read_dir, 3019 .iterate_shared = proc_tgid_base_readdir, 3020 .llseek = generic_file_llseek, 3021 }; 3022 3023 struct pid *tgid_pidfd_to_pid(const struct file *file) 3024 { 3025 if (file->f_op != &proc_tgid_base_operations) 3026 return ERR_PTR(-EBADF); 3027 3028 return proc_pid(file_inode(file)); 3029 } 3030 3031 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 3032 { 3033 return proc_pident_lookup(dir, dentry, 3034 tgid_base_stuff, 3035 tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff)); 3036 } 3037 3038 static const struct inode_operations proc_tgid_base_inode_operations = { 3039 .lookup = proc_tgid_base_lookup, 3040 .getattr = pid_getattr, 3041 .setattr = proc_setattr, 3042 .permission = proc_pid_permission, 3043 }; 3044 3045 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid) 3046 { 3047 struct dentry *dentry, *leader, *dir; 3048 char buf[10 + 1]; 3049 struct qstr name; 3050 3051 name.name = buf; 3052 name.len = snprintf(buf, sizeof(buf), "%u", pid); 3053 /* no ->d_hash() rejects on procfs */ 3054 dentry = d_hash_and_lookup(mnt->mnt_root, &name); 3055 if (dentry) { 3056 d_invalidate(dentry); 3057 dput(dentry); 3058 } 3059 3060 if (pid == tgid) 3061 return; 3062 3063 name.name = buf; 3064 name.len = snprintf(buf, sizeof(buf), "%u", tgid); 3065 leader = d_hash_and_lookup(mnt->mnt_root, &name); 3066 if (!leader) 3067 goto out; 3068 3069 name.name = "task"; 3070 name.len = strlen(name.name); 3071 dir = d_hash_and_lookup(leader, &name); 3072 if (!dir) 3073 goto out_put_leader; 3074 3075 name.name = buf; 3076 name.len = snprintf(buf, sizeof(buf), "%u", pid); 3077 dentry = d_hash_and_lookup(dir, &name); 3078 if (dentry) { 3079 d_invalidate(dentry); 3080 dput(dentry); 3081 } 3082 3083 dput(dir); 3084 out_put_leader: 3085 dput(leader); 3086 out: 3087 return; 3088 } 3089 3090 /** 3091 * proc_flush_task - Remove dcache entries for @task from the /proc dcache. 3092 * @task: task that should be flushed. 3093 * 3094 * When flushing dentries from proc, one needs to flush them from global 3095 * proc (proc_mnt) and from all the namespaces' procs this task was seen 3096 * in. This call is supposed to do all of this job. 3097 * 3098 * Looks in the dcache for 3099 * /proc/@pid 3100 * /proc/@tgid/task/@pid 3101 * if either directory is present flushes it and all of it'ts children 3102 * from the dcache. 3103 * 3104 * It is safe and reasonable to cache /proc entries for a task until 3105 * that task exits. After that they just clog up the dcache with 3106 * useless entries, possibly causing useful dcache entries to be 3107 * flushed instead. This routine is proved to flush those useless 3108 * dcache entries at process exit time. 3109 * 3110 * NOTE: This routine is just an optimization so it does not guarantee 3111 * that no dcache entries will exist at process exit time it 3112 * just makes it very unlikely that any will persist. 3113 */ 3114 3115 void proc_flush_task(struct task_struct *task) 3116 { 3117 int i; 3118 struct pid *pid, *tgid; 3119 struct upid *upid; 3120 3121 pid = task_pid(task); 3122 tgid = task_tgid(task); 3123 3124 for (i = 0; i <= pid->level; i++) { 3125 upid = &pid->numbers[i]; 3126 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr, 3127 tgid->numbers[i].nr); 3128 } 3129 } 3130 3131 static struct dentry *proc_pid_instantiate(struct dentry * dentry, 3132 struct task_struct *task, const void *ptr) 3133 { 3134 struct inode *inode; 3135 3136 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO); 3137 if (!inode) 3138 return ERR_PTR(-ENOENT); 3139 3140 inode->i_op = &proc_tgid_base_inode_operations; 3141 inode->i_fop = &proc_tgid_base_operations; 3142 inode->i_flags|=S_IMMUTABLE; 3143 3144 set_nlink(inode, nlink_tgid); 3145 pid_update_inode(task, inode); 3146 3147 d_set_d_op(dentry, &pid_dentry_operations); 3148 return d_splice_alias(inode, dentry); 3149 } 3150 3151 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags) 3152 { 3153 struct task_struct *task; 3154 unsigned tgid; 3155 struct pid_namespace *ns; 3156 struct dentry *result = ERR_PTR(-ENOENT); 3157 3158 tgid = name_to_int(&dentry->d_name); 3159 if (tgid == ~0U) 3160 goto out; 3161 3162 ns = dentry->d_sb->s_fs_info; 3163 rcu_read_lock(); 3164 task = find_task_by_pid_ns(tgid, ns); 3165 if (task) 3166 get_task_struct(task); 3167 rcu_read_unlock(); 3168 if (!task) 3169 goto out; 3170 3171 result = proc_pid_instantiate(dentry, task, NULL); 3172 put_task_struct(task); 3173 out: 3174 return result; 3175 } 3176 3177 /* 3178 * Find the first task with tgid >= tgid 3179 * 3180 */ 3181 struct tgid_iter { 3182 unsigned int tgid; 3183 struct task_struct *task; 3184 }; 3185 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter) 3186 { 3187 struct pid *pid; 3188 3189 if (iter.task) 3190 put_task_struct(iter.task); 3191 rcu_read_lock(); 3192 retry: 3193 iter.task = NULL; 3194 pid = find_ge_pid(iter.tgid, ns); 3195 if (pid) { 3196 iter.tgid = pid_nr_ns(pid, ns); 3197 iter.task = pid_task(pid, PIDTYPE_PID); 3198 /* What we to know is if the pid we have find is the 3199 * pid of a thread_group_leader. Testing for task 3200 * being a thread_group_leader is the obvious thing 3201 * todo but there is a window when it fails, due to 3202 * the pid transfer logic in de_thread. 3203 * 3204 * So we perform the straight forward test of seeing 3205 * if the pid we have found is the pid of a thread 3206 * group leader, and don't worry if the task we have 3207 * found doesn't happen to be a thread group leader. 3208 * As we don't care in the case of readdir. 3209 */ 3210 if (!iter.task || !has_group_leader_pid(iter.task)) { 3211 iter.tgid += 1; 3212 goto retry; 3213 } 3214 get_task_struct(iter.task); 3215 } 3216 rcu_read_unlock(); 3217 return iter; 3218 } 3219 3220 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2) 3221 3222 /* for the /proc/ directory itself, after non-process stuff has been done */ 3223 int proc_pid_readdir(struct file *file, struct dir_context *ctx) 3224 { 3225 struct tgid_iter iter; 3226 struct pid_namespace *ns = proc_pid_ns(file_inode(file)); 3227 loff_t pos = ctx->pos; 3228 3229 if (pos >= PID_MAX_LIMIT + TGID_OFFSET) 3230 return 0; 3231 3232 if (pos == TGID_OFFSET - 2) { 3233 struct inode *inode = d_inode(ns->proc_self); 3234 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK)) 3235 return 0; 3236 ctx->pos = pos = pos + 1; 3237 } 3238 if (pos == TGID_OFFSET - 1) { 3239 struct inode *inode = d_inode(ns->proc_thread_self); 3240 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK)) 3241 return 0; 3242 ctx->pos = pos = pos + 1; 3243 } 3244 iter.tgid = pos - TGID_OFFSET; 3245 iter.task = NULL; 3246 for (iter = next_tgid(ns, iter); 3247 iter.task; 3248 iter.tgid += 1, iter = next_tgid(ns, iter)) { 3249 char name[10 + 1]; 3250 unsigned int len; 3251 3252 cond_resched(); 3253 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE)) 3254 continue; 3255 3256 len = snprintf(name, sizeof(name), "%u", iter.tgid); 3257 ctx->pos = iter.tgid + TGID_OFFSET; 3258 if (!proc_fill_cache(file, ctx, name, len, 3259 proc_pid_instantiate, iter.task, NULL)) { 3260 put_task_struct(iter.task); 3261 return 0; 3262 } 3263 } 3264 ctx->pos = PID_MAX_LIMIT + TGID_OFFSET; 3265 return 0; 3266 } 3267 3268 /* 3269 * proc_tid_comm_permission is a special permission function exclusively 3270 * used for the node /proc/<pid>/task/<tid>/comm. 3271 * It bypasses generic permission checks in the case where a task of the same 3272 * task group attempts to access the node. 3273 * The rationale behind this is that glibc and bionic access this node for 3274 * cross thread naming (pthread_set/getname_np(!self)). However, if 3275 * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0, 3276 * which locks out the cross thread naming implementation. 3277 * This function makes sure that the node is always accessible for members of 3278 * same thread group. 3279 */ 3280 static int proc_tid_comm_permission(struct inode *inode, int mask) 3281 { 3282 bool is_same_tgroup; 3283 struct task_struct *task; 3284 3285 task = get_proc_task(inode); 3286 if (!task) 3287 return -ESRCH; 3288 is_same_tgroup = same_thread_group(current, task); 3289 put_task_struct(task); 3290 3291 if (likely(is_same_tgroup && !(mask & MAY_EXEC))) { 3292 /* This file (/proc/<pid>/task/<tid>/comm) can always be 3293 * read or written by the members of the corresponding 3294 * thread group. 3295 */ 3296 return 0; 3297 } 3298 3299 return generic_permission(inode, mask); 3300 } 3301 3302 static const struct inode_operations proc_tid_comm_inode_operations = { 3303 .permission = proc_tid_comm_permission, 3304 }; 3305 3306 /* 3307 * Tasks 3308 */ 3309 static const struct pid_entry tid_base_stuff[] = { 3310 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations), 3311 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations), 3312 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations), 3313 #ifdef CONFIG_NET 3314 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations), 3315 #endif 3316 REG("environ", S_IRUSR, proc_environ_operations), 3317 REG("auxv", S_IRUSR, proc_auxv_operations), 3318 ONE("status", S_IRUGO, proc_pid_status), 3319 ONE("personality", S_IRUSR, proc_pid_personality), 3320 ONE("limits", S_IRUGO, proc_pid_limits), 3321 #ifdef CONFIG_SCHED_DEBUG 3322 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations), 3323 #endif 3324 NOD("comm", S_IFREG|S_IRUGO|S_IWUSR, 3325 &proc_tid_comm_inode_operations, 3326 &proc_pid_set_comm_operations, {}), 3327 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 3328 ONE("syscall", S_IRUSR, proc_pid_syscall), 3329 #endif 3330 REG("cmdline", S_IRUGO, proc_pid_cmdline_ops), 3331 ONE("stat", S_IRUGO, proc_tid_stat), 3332 ONE("statm", S_IRUGO, proc_pid_statm), 3333 REG("maps", S_IRUGO, proc_pid_maps_operations), 3334 #ifdef CONFIG_PROC_CHILDREN 3335 REG("children", S_IRUGO, proc_tid_children_operations), 3336 #endif 3337 #ifdef CONFIG_NUMA 3338 REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations), 3339 #endif 3340 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations), 3341 LNK("cwd", proc_cwd_link), 3342 LNK("root", proc_root_link), 3343 LNK("exe", proc_exe_link), 3344 REG("mounts", S_IRUGO, proc_mounts_operations), 3345 REG("mountinfo", S_IRUGO, proc_mountinfo_operations), 3346 #ifdef CONFIG_PROC_PAGE_MONITOR 3347 REG("clear_refs", S_IWUSR, proc_clear_refs_operations), 3348 REG("smaps", S_IRUGO, proc_pid_smaps_operations), 3349 REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations), 3350 REG("pagemap", S_IRUSR, proc_pagemap_operations), 3351 #endif 3352 #ifdef CONFIG_SECURITY 3353 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), 3354 #endif 3355 #ifdef CONFIG_KALLSYMS 3356 ONE("wchan", S_IRUGO, proc_pid_wchan), 3357 #endif 3358 #ifdef CONFIG_STACKTRACE 3359 ONE("stack", S_IRUSR, proc_pid_stack), 3360 #endif 3361 #ifdef CONFIG_SCHED_INFO 3362 ONE("schedstat", S_IRUGO, proc_pid_schedstat), 3363 #endif 3364 #ifdef CONFIG_LATENCYTOP 3365 REG("latency", S_IRUGO, proc_lstats_operations), 3366 #endif 3367 #ifdef CONFIG_PROC_PID_CPUSET 3368 ONE("cpuset", S_IRUGO, proc_cpuset_show), 3369 #endif 3370 #ifdef CONFIG_CGROUPS 3371 ONE("cgroup", S_IRUGO, proc_cgroup_show), 3372 #endif 3373 ONE("oom_score", S_IRUGO, proc_oom_score), 3374 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adj_operations), 3375 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations), 3376 #ifdef CONFIG_AUDIT 3377 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations), 3378 REG("sessionid", S_IRUGO, proc_sessionid_operations), 3379 #endif 3380 #ifdef CONFIG_FAULT_INJECTION 3381 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations), 3382 REG("fail-nth", 0644, proc_fail_nth_operations), 3383 #endif 3384 #ifdef CONFIG_TASK_IO_ACCOUNTING 3385 ONE("io", S_IRUSR, proc_tid_io_accounting), 3386 #endif 3387 #ifdef CONFIG_USER_NS 3388 REG("uid_map", S_IRUGO|S_IWUSR, proc_uid_map_operations), 3389 REG("gid_map", S_IRUGO|S_IWUSR, proc_gid_map_operations), 3390 REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations), 3391 REG("setgroups", S_IRUGO|S_IWUSR, proc_setgroups_operations), 3392 #endif 3393 #ifdef CONFIG_LIVEPATCH 3394 ONE("patch_state", S_IRUSR, proc_pid_patch_state), 3395 #endif 3396 }; 3397 3398 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx) 3399 { 3400 return proc_pident_readdir(file, ctx, 3401 tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 3402 } 3403 3404 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 3405 { 3406 return proc_pident_lookup(dir, dentry, 3407 tid_base_stuff, 3408 tid_base_stuff + ARRAY_SIZE(tid_base_stuff)); 3409 } 3410 3411 static const struct file_operations proc_tid_base_operations = { 3412 .read = generic_read_dir, 3413 .iterate_shared = proc_tid_base_readdir, 3414 .llseek = generic_file_llseek, 3415 }; 3416 3417 static const struct inode_operations proc_tid_base_inode_operations = { 3418 .lookup = proc_tid_base_lookup, 3419 .getattr = pid_getattr, 3420 .setattr = proc_setattr, 3421 }; 3422 3423 static struct dentry *proc_task_instantiate(struct dentry *dentry, 3424 struct task_struct *task, const void *ptr) 3425 { 3426 struct inode *inode; 3427 inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO); 3428 if (!inode) 3429 return ERR_PTR(-ENOENT); 3430 3431 inode->i_op = &proc_tid_base_inode_operations; 3432 inode->i_fop = &proc_tid_base_operations; 3433 inode->i_flags |= S_IMMUTABLE; 3434 3435 set_nlink(inode, nlink_tid); 3436 pid_update_inode(task, inode); 3437 3438 d_set_d_op(dentry, &pid_dentry_operations); 3439 return d_splice_alias(inode, dentry); 3440 } 3441 3442 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags) 3443 { 3444 struct task_struct *task; 3445 struct task_struct *leader = get_proc_task(dir); 3446 unsigned tid; 3447 struct pid_namespace *ns; 3448 struct dentry *result = ERR_PTR(-ENOENT); 3449 3450 if (!leader) 3451 goto out_no_task; 3452 3453 tid = name_to_int(&dentry->d_name); 3454 if (tid == ~0U) 3455 goto out; 3456 3457 ns = dentry->d_sb->s_fs_info; 3458 rcu_read_lock(); 3459 task = find_task_by_pid_ns(tid, ns); 3460 if (task) 3461 get_task_struct(task); 3462 rcu_read_unlock(); 3463 if (!task) 3464 goto out; 3465 if (!same_thread_group(leader, task)) 3466 goto out_drop_task; 3467 3468 result = proc_task_instantiate(dentry, task, NULL); 3469 out_drop_task: 3470 put_task_struct(task); 3471 out: 3472 put_task_struct(leader); 3473 out_no_task: 3474 return result; 3475 } 3476 3477 /* 3478 * Find the first tid of a thread group to return to user space. 3479 * 3480 * Usually this is just the thread group leader, but if the users 3481 * buffer was too small or there was a seek into the middle of the 3482 * directory we have more work todo. 3483 * 3484 * In the case of a short read we start with find_task_by_pid. 3485 * 3486 * In the case of a seek we start with the leader and walk nr 3487 * threads past it. 3488 */ 3489 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos, 3490 struct pid_namespace *ns) 3491 { 3492 struct task_struct *pos, *task; 3493 unsigned long nr = f_pos; 3494 3495 if (nr != f_pos) /* 32bit overflow? */ 3496 return NULL; 3497 3498 rcu_read_lock(); 3499 task = pid_task(pid, PIDTYPE_PID); 3500 if (!task) 3501 goto fail; 3502 3503 /* Attempt to start with the tid of a thread */ 3504 if (tid && nr) { 3505 pos = find_task_by_pid_ns(tid, ns); 3506 if (pos && same_thread_group(pos, task)) 3507 goto found; 3508 } 3509 3510 /* If nr exceeds the number of threads there is nothing todo */ 3511 if (nr >= get_nr_threads(task)) 3512 goto fail; 3513 3514 /* If we haven't found our starting place yet start 3515 * with the leader and walk nr threads forward. 3516 */ 3517 pos = task = task->group_leader; 3518 do { 3519 if (!nr--) 3520 goto found; 3521 } while_each_thread(task, pos); 3522 fail: 3523 pos = NULL; 3524 goto out; 3525 found: 3526 get_task_struct(pos); 3527 out: 3528 rcu_read_unlock(); 3529 return pos; 3530 } 3531 3532 /* 3533 * Find the next thread in the thread list. 3534 * Return NULL if there is an error or no next thread. 3535 * 3536 * The reference to the input task_struct is released. 3537 */ 3538 static struct task_struct *next_tid(struct task_struct *start) 3539 { 3540 struct task_struct *pos = NULL; 3541 rcu_read_lock(); 3542 if (pid_alive(start)) { 3543 pos = next_thread(start); 3544 if (thread_group_leader(pos)) 3545 pos = NULL; 3546 else 3547 get_task_struct(pos); 3548 } 3549 rcu_read_unlock(); 3550 put_task_struct(start); 3551 return pos; 3552 } 3553 3554 /* for the /proc/TGID/task/ directories */ 3555 static int proc_task_readdir(struct file *file, struct dir_context *ctx) 3556 { 3557 struct inode *inode = file_inode(file); 3558 struct task_struct *task; 3559 struct pid_namespace *ns; 3560 int tid; 3561 3562 if (proc_inode_is_dead(inode)) 3563 return -ENOENT; 3564 3565 if (!dir_emit_dots(file, ctx)) 3566 return 0; 3567 3568 /* f_version caches the tgid value that the last readdir call couldn't 3569 * return. lseek aka telldir automagically resets f_version to 0. 3570 */ 3571 ns = proc_pid_ns(inode); 3572 tid = (int)file->f_version; 3573 file->f_version = 0; 3574 for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns); 3575 task; 3576 task = next_tid(task), ctx->pos++) { 3577 char name[10 + 1]; 3578 unsigned int len; 3579 tid = task_pid_nr_ns(task, ns); 3580 len = snprintf(name, sizeof(name), "%u", tid); 3581 if (!proc_fill_cache(file, ctx, name, len, 3582 proc_task_instantiate, task, NULL)) { 3583 /* returning this tgid failed, save it as the first 3584 * pid for the next readir call */ 3585 file->f_version = (u64)tid; 3586 put_task_struct(task); 3587 break; 3588 } 3589 } 3590 3591 return 0; 3592 } 3593 3594 static int proc_task_getattr(const struct path *path, struct kstat *stat, 3595 u32 request_mask, unsigned int query_flags) 3596 { 3597 struct inode *inode = d_inode(path->dentry); 3598 struct task_struct *p = get_proc_task(inode); 3599 generic_fillattr(inode, stat); 3600 3601 if (p) { 3602 stat->nlink += get_nr_threads(p); 3603 put_task_struct(p); 3604 } 3605 3606 return 0; 3607 } 3608 3609 static const struct inode_operations proc_task_inode_operations = { 3610 .lookup = proc_task_lookup, 3611 .getattr = proc_task_getattr, 3612 .setattr = proc_setattr, 3613 .permission = proc_pid_permission, 3614 }; 3615 3616 static const struct file_operations proc_task_operations = { 3617 .read = generic_read_dir, 3618 .iterate_shared = proc_task_readdir, 3619 .llseek = generic_file_llseek, 3620 }; 3621 3622 void __init set_proc_pid_nlink(void) 3623 { 3624 nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 3625 nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 3626 } 3627