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