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/cgroup.h> 77 #include <linux/cpuset.h> 78 #include <linux/audit.h> 79 #include <linux/poll.h> 80 #include <linux/nsproxy.h> 81 #include <linux/oom.h> 82 #include <linux/elf.h> 83 #include <linux/pid_namespace.h> 84 #include <linux/fs_struct.h> 85 #include <linux/slab.h> 86 #ifdef CONFIG_HARDWALL 87 #include <asm/hardwall.h> 88 #endif 89 #include "internal.h" 90 91 /* NOTE: 92 * Implementing inode permission operations in /proc is almost 93 * certainly an error. Permission checks need to happen during 94 * each system call not at open time. The reason is that most of 95 * what we wish to check for permissions in /proc varies at runtime. 96 * 97 * The classic example of a problem is opening file descriptors 98 * in /proc for a task before it execs a suid executable. 99 */ 100 101 struct pid_entry { 102 char *name; 103 int len; 104 mode_t mode; 105 const struct inode_operations *iop; 106 const struct file_operations *fop; 107 union proc_op op; 108 }; 109 110 #define NOD(NAME, MODE, IOP, FOP, OP) { \ 111 .name = (NAME), \ 112 .len = sizeof(NAME) - 1, \ 113 .mode = MODE, \ 114 .iop = IOP, \ 115 .fop = FOP, \ 116 .op = OP, \ 117 } 118 119 #define DIR(NAME, MODE, iops, fops) \ 120 NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} ) 121 #define LNK(NAME, get_link) \ 122 NOD(NAME, (S_IFLNK|S_IRWXUGO), \ 123 &proc_pid_link_inode_operations, NULL, \ 124 { .proc_get_link = get_link } ) 125 #define REG(NAME, MODE, fops) \ 126 NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {}) 127 #define INF(NAME, MODE, read) \ 128 NOD(NAME, (S_IFREG|(MODE)), \ 129 NULL, &proc_info_file_operations, \ 130 { .proc_read = read } ) 131 #define ONE(NAME, MODE, show) \ 132 NOD(NAME, (S_IFREG|(MODE)), \ 133 NULL, &proc_single_file_operations, \ 134 { .proc_show = show } ) 135 136 /* 137 * Count the number of hardlinks for the pid_entry table, excluding the . 138 * and .. links. 139 */ 140 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries, 141 unsigned int n) 142 { 143 unsigned int i; 144 unsigned int count; 145 146 count = 0; 147 for (i = 0; i < n; ++i) { 148 if (S_ISDIR(entries[i].mode)) 149 ++count; 150 } 151 152 return count; 153 } 154 155 static int get_task_root(struct task_struct *task, struct path *root) 156 { 157 int result = -ENOENT; 158 159 task_lock(task); 160 if (task->fs) { 161 get_fs_root(task->fs, root); 162 result = 0; 163 } 164 task_unlock(task); 165 return result; 166 } 167 168 static int proc_cwd_link(struct inode *inode, struct path *path) 169 { 170 struct task_struct *task = get_proc_task(inode); 171 int result = -ENOENT; 172 173 if (task) { 174 task_lock(task); 175 if (task->fs) { 176 get_fs_pwd(task->fs, path); 177 result = 0; 178 } 179 task_unlock(task); 180 put_task_struct(task); 181 } 182 return result; 183 } 184 185 static int proc_root_link(struct inode *inode, struct path *path) 186 { 187 struct task_struct *task = get_proc_task(inode); 188 int result = -ENOENT; 189 190 if (task) { 191 result = get_task_root(task, path); 192 put_task_struct(task); 193 } 194 return result; 195 } 196 197 static struct mm_struct *__check_mem_permission(struct task_struct *task) 198 { 199 struct mm_struct *mm; 200 201 mm = get_task_mm(task); 202 if (!mm) 203 return ERR_PTR(-EINVAL); 204 205 /* 206 * A task can always look at itself, in case it chooses 207 * to use system calls instead of load instructions. 208 */ 209 if (task == current) 210 return mm; 211 212 /* 213 * If current is actively ptrace'ing, and would also be 214 * permitted to freshly attach with ptrace now, permit it. 215 */ 216 if (task_is_stopped_or_traced(task)) { 217 int match; 218 rcu_read_lock(); 219 match = (ptrace_parent(task) == current); 220 rcu_read_unlock(); 221 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH)) 222 return mm; 223 } 224 225 /* 226 * No one else is allowed. 227 */ 228 mmput(mm); 229 return ERR_PTR(-EPERM); 230 } 231 232 /* 233 * If current may access user memory in @task return a reference to the 234 * corresponding mm, otherwise ERR_PTR. 235 */ 236 static struct mm_struct *check_mem_permission(struct task_struct *task) 237 { 238 struct mm_struct *mm; 239 int err; 240 241 /* 242 * Avoid racing if task exec's as we might get a new mm but validate 243 * against old credentials. 244 */ 245 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 246 if (err) 247 return ERR_PTR(err); 248 249 mm = __check_mem_permission(task); 250 mutex_unlock(&task->signal->cred_guard_mutex); 251 252 return mm; 253 } 254 255 struct mm_struct *mm_for_maps(struct task_struct *task) 256 { 257 struct mm_struct *mm; 258 int err; 259 260 err = mutex_lock_killable(&task->signal->cred_guard_mutex); 261 if (err) 262 return ERR_PTR(err); 263 264 mm = get_task_mm(task); 265 if (mm && mm != current->mm && 266 !ptrace_may_access(task, PTRACE_MODE_READ)) { 267 mmput(mm); 268 mm = ERR_PTR(-EACCES); 269 } 270 mutex_unlock(&task->signal->cred_guard_mutex); 271 272 return mm; 273 } 274 275 static int proc_pid_cmdline(struct task_struct *task, char * buffer) 276 { 277 int res = 0; 278 unsigned int len; 279 struct mm_struct *mm = get_task_mm(task); 280 if (!mm) 281 goto out; 282 if (!mm->arg_end) 283 goto out_mm; /* Shh! No looking before we're done */ 284 285 len = mm->arg_end - mm->arg_start; 286 287 if (len > PAGE_SIZE) 288 len = PAGE_SIZE; 289 290 res = access_process_vm(task, mm->arg_start, buffer, len, 0); 291 292 // If the nul at the end of args has been overwritten, then 293 // assume application is using setproctitle(3). 294 if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) { 295 len = strnlen(buffer, res); 296 if (len < res) { 297 res = len; 298 } else { 299 len = mm->env_end - mm->env_start; 300 if (len > PAGE_SIZE - res) 301 len = PAGE_SIZE - res; 302 res += access_process_vm(task, mm->env_start, buffer+res, len, 0); 303 res = strnlen(buffer, res); 304 } 305 } 306 out_mm: 307 mmput(mm); 308 out: 309 return res; 310 } 311 312 static int proc_pid_auxv(struct task_struct *task, char *buffer) 313 { 314 struct mm_struct *mm = mm_for_maps(task); 315 int res = PTR_ERR(mm); 316 if (mm && !IS_ERR(mm)) { 317 unsigned int nwords = 0; 318 do { 319 nwords += 2; 320 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */ 321 res = nwords * sizeof(mm->saved_auxv[0]); 322 if (res > PAGE_SIZE) 323 res = PAGE_SIZE; 324 memcpy(buffer, mm->saved_auxv, res); 325 mmput(mm); 326 } 327 return res; 328 } 329 330 331 #ifdef CONFIG_KALLSYMS 332 /* 333 * Provides a wchan file via kallsyms in a proper one-value-per-file format. 334 * Returns the resolved symbol. If that fails, simply return the address. 335 */ 336 static int proc_pid_wchan(struct task_struct *task, char *buffer) 337 { 338 unsigned long wchan; 339 char symname[KSYM_NAME_LEN]; 340 341 wchan = get_wchan(task); 342 343 if (lookup_symbol_name(wchan, symname) < 0) 344 if (!ptrace_may_access(task, PTRACE_MODE_READ)) 345 return 0; 346 else 347 return sprintf(buffer, "%lu", wchan); 348 else 349 return sprintf(buffer, "%s", symname); 350 } 351 #endif /* CONFIG_KALLSYMS */ 352 353 static int lock_trace(struct task_struct *task) 354 { 355 int err = mutex_lock_killable(&task->signal->cred_guard_mutex); 356 if (err) 357 return err; 358 if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) { 359 mutex_unlock(&task->signal->cred_guard_mutex); 360 return -EPERM; 361 } 362 return 0; 363 } 364 365 static void unlock_trace(struct task_struct *task) 366 { 367 mutex_unlock(&task->signal->cred_guard_mutex); 368 } 369 370 #ifdef CONFIG_STACKTRACE 371 372 #define MAX_STACK_TRACE_DEPTH 64 373 374 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns, 375 struct pid *pid, struct task_struct *task) 376 { 377 struct stack_trace trace; 378 unsigned long *entries; 379 int err; 380 int i; 381 382 entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL); 383 if (!entries) 384 return -ENOMEM; 385 386 trace.nr_entries = 0; 387 trace.max_entries = MAX_STACK_TRACE_DEPTH; 388 trace.entries = entries; 389 trace.skip = 0; 390 391 err = lock_trace(task); 392 if (!err) { 393 save_stack_trace_tsk(task, &trace); 394 395 for (i = 0; i < trace.nr_entries; i++) { 396 seq_printf(m, "[<%pK>] %pS\n", 397 (void *)entries[i], (void *)entries[i]); 398 } 399 unlock_trace(task); 400 } 401 kfree(entries); 402 403 return err; 404 } 405 #endif 406 407 #ifdef CONFIG_SCHEDSTATS 408 /* 409 * Provides /proc/PID/schedstat 410 */ 411 static int proc_pid_schedstat(struct task_struct *task, char *buffer) 412 { 413 return sprintf(buffer, "%llu %llu %lu\n", 414 (unsigned long long)task->se.sum_exec_runtime, 415 (unsigned long long)task->sched_info.run_delay, 416 task->sched_info.pcount); 417 } 418 #endif 419 420 #ifdef CONFIG_LATENCYTOP 421 static int lstats_show_proc(struct seq_file *m, void *v) 422 { 423 int i; 424 struct inode *inode = m->private; 425 struct task_struct *task = get_proc_task(inode); 426 427 if (!task) 428 return -ESRCH; 429 seq_puts(m, "Latency Top version : v0.1\n"); 430 for (i = 0; i < 32; i++) { 431 struct latency_record *lr = &task->latency_record[i]; 432 if (lr->backtrace[0]) { 433 int q; 434 seq_printf(m, "%i %li %li", 435 lr->count, lr->time, lr->max); 436 for (q = 0; q < LT_BACKTRACEDEPTH; q++) { 437 unsigned long bt = lr->backtrace[q]; 438 if (!bt) 439 break; 440 if (bt == ULONG_MAX) 441 break; 442 seq_printf(m, " %ps", (void *)bt); 443 } 444 seq_putc(m, '\n'); 445 } 446 447 } 448 put_task_struct(task); 449 return 0; 450 } 451 452 static int lstats_open(struct inode *inode, struct file *file) 453 { 454 return single_open(file, lstats_show_proc, inode); 455 } 456 457 static ssize_t lstats_write(struct file *file, const char __user *buf, 458 size_t count, loff_t *offs) 459 { 460 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 461 462 if (!task) 463 return -ESRCH; 464 clear_all_latency_tracing(task); 465 put_task_struct(task); 466 467 return count; 468 } 469 470 static const struct file_operations proc_lstats_operations = { 471 .open = lstats_open, 472 .read = seq_read, 473 .write = lstats_write, 474 .llseek = seq_lseek, 475 .release = single_release, 476 }; 477 478 #endif 479 480 static int proc_oom_score(struct task_struct *task, char *buffer) 481 { 482 unsigned long points = 0; 483 484 read_lock(&tasklist_lock); 485 if (pid_alive(task)) 486 points = oom_badness(task, NULL, NULL, 487 totalram_pages + total_swap_pages); 488 read_unlock(&tasklist_lock); 489 return sprintf(buffer, "%lu\n", points); 490 } 491 492 struct limit_names { 493 char *name; 494 char *unit; 495 }; 496 497 static const struct limit_names lnames[RLIM_NLIMITS] = { 498 [RLIMIT_CPU] = {"Max cpu time", "seconds"}, 499 [RLIMIT_FSIZE] = {"Max file size", "bytes"}, 500 [RLIMIT_DATA] = {"Max data size", "bytes"}, 501 [RLIMIT_STACK] = {"Max stack size", "bytes"}, 502 [RLIMIT_CORE] = {"Max core file size", "bytes"}, 503 [RLIMIT_RSS] = {"Max resident set", "bytes"}, 504 [RLIMIT_NPROC] = {"Max processes", "processes"}, 505 [RLIMIT_NOFILE] = {"Max open files", "files"}, 506 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"}, 507 [RLIMIT_AS] = {"Max address space", "bytes"}, 508 [RLIMIT_LOCKS] = {"Max file locks", "locks"}, 509 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"}, 510 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"}, 511 [RLIMIT_NICE] = {"Max nice priority", NULL}, 512 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL}, 513 [RLIMIT_RTTIME] = {"Max realtime timeout", "us"}, 514 }; 515 516 /* Display limits for a process */ 517 static int proc_pid_limits(struct task_struct *task, char *buffer) 518 { 519 unsigned int i; 520 int count = 0; 521 unsigned long flags; 522 char *bufptr = buffer; 523 524 struct rlimit rlim[RLIM_NLIMITS]; 525 526 if (!lock_task_sighand(task, &flags)) 527 return 0; 528 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS); 529 unlock_task_sighand(task, &flags); 530 531 /* 532 * print the file header 533 */ 534 count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n", 535 "Limit", "Soft Limit", "Hard Limit", "Units"); 536 537 for (i = 0; i < RLIM_NLIMITS; i++) { 538 if (rlim[i].rlim_cur == RLIM_INFINITY) 539 count += sprintf(&bufptr[count], "%-25s %-20s ", 540 lnames[i].name, "unlimited"); 541 else 542 count += sprintf(&bufptr[count], "%-25s %-20lu ", 543 lnames[i].name, rlim[i].rlim_cur); 544 545 if (rlim[i].rlim_max == RLIM_INFINITY) 546 count += sprintf(&bufptr[count], "%-20s ", "unlimited"); 547 else 548 count += sprintf(&bufptr[count], "%-20lu ", 549 rlim[i].rlim_max); 550 551 if (lnames[i].unit) 552 count += sprintf(&bufptr[count], "%-10s\n", 553 lnames[i].unit); 554 else 555 count += sprintf(&bufptr[count], "\n"); 556 } 557 558 return count; 559 } 560 561 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 562 static int proc_pid_syscall(struct task_struct *task, char *buffer) 563 { 564 long nr; 565 unsigned long args[6], sp, pc; 566 int res = lock_trace(task); 567 if (res) 568 return res; 569 570 if (task_current_syscall(task, &nr, args, 6, &sp, &pc)) 571 res = sprintf(buffer, "running\n"); 572 else if (nr < 0) 573 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc); 574 else 575 res = sprintf(buffer, 576 "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n", 577 nr, 578 args[0], args[1], args[2], args[3], args[4], args[5], 579 sp, pc); 580 unlock_trace(task); 581 return res; 582 } 583 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */ 584 585 /************************************************************************/ 586 /* Here the fs part begins */ 587 /************************************************************************/ 588 589 /* permission checks */ 590 static int proc_fd_access_allowed(struct inode *inode) 591 { 592 struct task_struct *task; 593 int allowed = 0; 594 /* Allow access to a task's file descriptors if it is us or we 595 * may use ptrace attach to the process and find out that 596 * information. 597 */ 598 task = get_proc_task(inode); 599 if (task) { 600 allowed = ptrace_may_access(task, PTRACE_MODE_READ); 601 put_task_struct(task); 602 } 603 return allowed; 604 } 605 606 int proc_setattr(struct dentry *dentry, struct iattr *attr) 607 { 608 int error; 609 struct inode *inode = dentry->d_inode; 610 611 if (attr->ia_valid & ATTR_MODE) 612 return -EPERM; 613 614 error = inode_change_ok(inode, attr); 615 if (error) 616 return error; 617 618 if ((attr->ia_valid & ATTR_SIZE) && 619 attr->ia_size != i_size_read(inode)) { 620 error = vmtruncate(inode, attr->ia_size); 621 if (error) 622 return error; 623 } 624 625 setattr_copy(inode, attr); 626 mark_inode_dirty(inode); 627 return 0; 628 } 629 630 static const struct inode_operations proc_def_inode_operations = { 631 .setattr = proc_setattr, 632 }; 633 634 static int mounts_open_common(struct inode *inode, struct file *file, 635 const struct seq_operations *op) 636 { 637 struct task_struct *task = get_proc_task(inode); 638 struct nsproxy *nsp; 639 struct mnt_namespace *ns = NULL; 640 struct path root; 641 struct proc_mounts *p; 642 int ret = -EINVAL; 643 644 if (task) { 645 rcu_read_lock(); 646 nsp = task_nsproxy(task); 647 if (nsp) { 648 ns = nsp->mnt_ns; 649 if (ns) 650 get_mnt_ns(ns); 651 } 652 rcu_read_unlock(); 653 if (ns && get_task_root(task, &root) == 0) 654 ret = 0; 655 put_task_struct(task); 656 } 657 658 if (!ns) 659 goto err; 660 if (ret) 661 goto err_put_ns; 662 663 ret = -ENOMEM; 664 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL); 665 if (!p) 666 goto err_put_path; 667 668 file->private_data = &p->m; 669 ret = seq_open(file, op); 670 if (ret) 671 goto err_free; 672 673 p->m.private = p; 674 p->ns = ns; 675 p->root = root; 676 p->m.poll_event = ns->event; 677 678 return 0; 679 680 err_free: 681 kfree(p); 682 err_put_path: 683 path_put(&root); 684 err_put_ns: 685 put_mnt_ns(ns); 686 err: 687 return ret; 688 } 689 690 static int mounts_release(struct inode *inode, struct file *file) 691 { 692 struct proc_mounts *p = file->private_data; 693 path_put(&p->root); 694 put_mnt_ns(p->ns); 695 return seq_release(inode, file); 696 } 697 698 static unsigned mounts_poll(struct file *file, poll_table *wait) 699 { 700 struct proc_mounts *p = file->private_data; 701 unsigned res = POLLIN | POLLRDNORM; 702 703 poll_wait(file, &p->ns->poll, wait); 704 if (mnt_had_events(p)) 705 res |= POLLERR | POLLPRI; 706 707 return res; 708 } 709 710 static int mounts_open(struct inode *inode, struct file *file) 711 { 712 return mounts_open_common(inode, file, &mounts_op); 713 } 714 715 static const struct file_operations proc_mounts_operations = { 716 .open = mounts_open, 717 .read = seq_read, 718 .llseek = seq_lseek, 719 .release = mounts_release, 720 .poll = mounts_poll, 721 }; 722 723 static int mountinfo_open(struct inode *inode, struct file *file) 724 { 725 return mounts_open_common(inode, file, &mountinfo_op); 726 } 727 728 static const struct file_operations proc_mountinfo_operations = { 729 .open = mountinfo_open, 730 .read = seq_read, 731 .llseek = seq_lseek, 732 .release = mounts_release, 733 .poll = mounts_poll, 734 }; 735 736 static int mountstats_open(struct inode *inode, struct file *file) 737 { 738 return mounts_open_common(inode, file, &mountstats_op); 739 } 740 741 static const struct file_operations proc_mountstats_operations = { 742 .open = mountstats_open, 743 .read = seq_read, 744 .llseek = seq_lseek, 745 .release = mounts_release, 746 }; 747 748 #define PROC_BLOCK_SIZE (3*1024) /* 4K page size but our output routines use some slack for overruns */ 749 750 static ssize_t proc_info_read(struct file * file, char __user * buf, 751 size_t count, loff_t *ppos) 752 { 753 struct inode * inode = file->f_path.dentry->d_inode; 754 unsigned long page; 755 ssize_t length; 756 struct task_struct *task = get_proc_task(inode); 757 758 length = -ESRCH; 759 if (!task) 760 goto out_no_task; 761 762 if (count > PROC_BLOCK_SIZE) 763 count = PROC_BLOCK_SIZE; 764 765 length = -ENOMEM; 766 if (!(page = __get_free_page(GFP_TEMPORARY))) 767 goto out; 768 769 length = PROC_I(inode)->op.proc_read(task, (char*)page); 770 771 if (length >= 0) 772 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length); 773 free_page(page); 774 out: 775 put_task_struct(task); 776 out_no_task: 777 return length; 778 } 779 780 static const struct file_operations proc_info_file_operations = { 781 .read = proc_info_read, 782 .llseek = generic_file_llseek, 783 }; 784 785 static int proc_single_show(struct seq_file *m, void *v) 786 { 787 struct inode *inode = m->private; 788 struct pid_namespace *ns; 789 struct pid *pid; 790 struct task_struct *task; 791 int ret; 792 793 ns = inode->i_sb->s_fs_info; 794 pid = proc_pid(inode); 795 task = get_pid_task(pid, PIDTYPE_PID); 796 if (!task) 797 return -ESRCH; 798 799 ret = PROC_I(inode)->op.proc_show(m, ns, pid, task); 800 801 put_task_struct(task); 802 return ret; 803 } 804 805 static int proc_single_open(struct inode *inode, struct file *filp) 806 { 807 return single_open(filp, proc_single_show, inode); 808 } 809 810 static const struct file_operations proc_single_file_operations = { 811 .open = proc_single_open, 812 .read = seq_read, 813 .llseek = seq_lseek, 814 .release = single_release, 815 }; 816 817 static int mem_open(struct inode* inode, struct file* file) 818 { 819 file->private_data = (void*)((long)current->self_exec_id); 820 /* OK to pass negative loff_t, we can catch out-of-range */ 821 file->f_mode |= FMODE_UNSIGNED_OFFSET; 822 return 0; 823 } 824 825 static ssize_t mem_read(struct file * file, char __user * buf, 826 size_t count, loff_t *ppos) 827 { 828 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 829 char *page; 830 unsigned long src = *ppos; 831 int ret = -ESRCH; 832 struct mm_struct *mm; 833 834 if (!task) 835 goto out_no_task; 836 837 ret = -ENOMEM; 838 page = (char *)__get_free_page(GFP_TEMPORARY); 839 if (!page) 840 goto out; 841 842 mm = check_mem_permission(task); 843 ret = PTR_ERR(mm); 844 if (IS_ERR(mm)) 845 goto out_free; 846 847 ret = -EIO; 848 849 if (file->private_data != (void*)((long)current->self_exec_id)) 850 goto out_put; 851 852 ret = 0; 853 854 while (count > 0) { 855 int this_len, retval; 856 857 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count; 858 retval = access_remote_vm(mm, src, page, this_len, 0); 859 if (!retval) { 860 if (!ret) 861 ret = -EIO; 862 break; 863 } 864 865 if (copy_to_user(buf, page, retval)) { 866 ret = -EFAULT; 867 break; 868 } 869 870 ret += retval; 871 src += retval; 872 buf += retval; 873 count -= retval; 874 } 875 *ppos = src; 876 877 out_put: 878 mmput(mm); 879 out_free: 880 free_page((unsigned long) page); 881 out: 882 put_task_struct(task); 883 out_no_task: 884 return ret; 885 } 886 887 static ssize_t mem_write(struct file * file, const char __user *buf, 888 size_t count, loff_t *ppos) 889 { 890 int copied; 891 char *page; 892 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 893 unsigned long dst = *ppos; 894 struct mm_struct *mm; 895 896 copied = -ESRCH; 897 if (!task) 898 goto out_no_task; 899 900 copied = -ENOMEM; 901 page = (char *)__get_free_page(GFP_TEMPORARY); 902 if (!page) 903 goto out_task; 904 905 mm = check_mem_permission(task); 906 copied = PTR_ERR(mm); 907 if (IS_ERR(mm)) 908 goto out_free; 909 910 copied = -EIO; 911 if (file->private_data != (void *)((long)current->self_exec_id)) 912 goto out_mm; 913 914 copied = 0; 915 while (count > 0) { 916 int this_len, retval; 917 918 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count; 919 if (copy_from_user(page, buf, this_len)) { 920 copied = -EFAULT; 921 break; 922 } 923 retval = access_remote_vm(mm, dst, page, this_len, 1); 924 if (!retval) { 925 if (!copied) 926 copied = -EIO; 927 break; 928 } 929 copied += retval; 930 buf += retval; 931 dst += retval; 932 count -= retval; 933 } 934 *ppos = dst; 935 936 out_mm: 937 mmput(mm); 938 out_free: 939 free_page((unsigned long) page); 940 out_task: 941 put_task_struct(task); 942 out_no_task: 943 return copied; 944 } 945 946 loff_t mem_lseek(struct file *file, loff_t offset, int orig) 947 { 948 switch (orig) { 949 case 0: 950 file->f_pos = offset; 951 break; 952 case 1: 953 file->f_pos += offset; 954 break; 955 default: 956 return -EINVAL; 957 } 958 force_successful_syscall_return(); 959 return file->f_pos; 960 } 961 962 static const struct file_operations proc_mem_operations = { 963 .llseek = mem_lseek, 964 .read = mem_read, 965 .write = mem_write, 966 .open = mem_open, 967 }; 968 969 static ssize_t environ_read(struct file *file, char __user *buf, 970 size_t count, loff_t *ppos) 971 { 972 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 973 char *page; 974 unsigned long src = *ppos; 975 int ret = -ESRCH; 976 struct mm_struct *mm; 977 978 if (!task) 979 goto out_no_task; 980 981 ret = -ENOMEM; 982 page = (char *)__get_free_page(GFP_TEMPORARY); 983 if (!page) 984 goto out; 985 986 987 mm = mm_for_maps(task); 988 ret = PTR_ERR(mm); 989 if (!mm || IS_ERR(mm)) 990 goto out_free; 991 992 ret = 0; 993 while (count > 0) { 994 int this_len, retval, max_len; 995 996 this_len = mm->env_end - (mm->env_start + src); 997 998 if (this_len <= 0) 999 break; 1000 1001 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count; 1002 this_len = (this_len > max_len) ? max_len : this_len; 1003 1004 retval = access_process_vm(task, (mm->env_start + src), 1005 page, this_len, 0); 1006 1007 if (retval <= 0) { 1008 ret = retval; 1009 break; 1010 } 1011 1012 if (copy_to_user(buf, page, retval)) { 1013 ret = -EFAULT; 1014 break; 1015 } 1016 1017 ret += retval; 1018 src += retval; 1019 buf += retval; 1020 count -= retval; 1021 } 1022 *ppos = src; 1023 1024 mmput(mm); 1025 out_free: 1026 free_page((unsigned long) page); 1027 out: 1028 put_task_struct(task); 1029 out_no_task: 1030 return ret; 1031 } 1032 1033 static const struct file_operations proc_environ_operations = { 1034 .read = environ_read, 1035 .llseek = generic_file_llseek, 1036 }; 1037 1038 static ssize_t oom_adjust_read(struct file *file, char __user *buf, 1039 size_t count, loff_t *ppos) 1040 { 1041 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 1042 char buffer[PROC_NUMBUF]; 1043 size_t len; 1044 int oom_adjust = OOM_DISABLE; 1045 unsigned long flags; 1046 1047 if (!task) 1048 return -ESRCH; 1049 1050 if (lock_task_sighand(task, &flags)) { 1051 oom_adjust = task->signal->oom_adj; 1052 unlock_task_sighand(task, &flags); 1053 } 1054 1055 put_task_struct(task); 1056 1057 len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust); 1058 1059 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1060 } 1061 1062 static ssize_t oom_adjust_write(struct file *file, const char __user *buf, 1063 size_t count, loff_t *ppos) 1064 { 1065 struct task_struct *task; 1066 char buffer[PROC_NUMBUF]; 1067 int oom_adjust; 1068 unsigned long flags; 1069 int err; 1070 1071 memset(buffer, 0, sizeof(buffer)); 1072 if (count > sizeof(buffer) - 1) 1073 count = sizeof(buffer) - 1; 1074 if (copy_from_user(buffer, buf, count)) { 1075 err = -EFAULT; 1076 goto out; 1077 } 1078 1079 err = kstrtoint(strstrip(buffer), 0, &oom_adjust); 1080 if (err) 1081 goto out; 1082 if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) && 1083 oom_adjust != OOM_DISABLE) { 1084 err = -EINVAL; 1085 goto out; 1086 } 1087 1088 task = get_proc_task(file->f_path.dentry->d_inode); 1089 if (!task) { 1090 err = -ESRCH; 1091 goto out; 1092 } 1093 1094 task_lock(task); 1095 if (!task->mm) { 1096 err = -EINVAL; 1097 goto err_task_lock; 1098 } 1099 1100 if (!lock_task_sighand(task, &flags)) { 1101 err = -ESRCH; 1102 goto err_task_lock; 1103 } 1104 1105 if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) { 1106 err = -EACCES; 1107 goto err_sighand; 1108 } 1109 1110 if (oom_adjust != task->signal->oom_adj) { 1111 if (oom_adjust == OOM_DISABLE) 1112 atomic_inc(&task->mm->oom_disable_count); 1113 if (task->signal->oom_adj == OOM_DISABLE) 1114 atomic_dec(&task->mm->oom_disable_count); 1115 } 1116 1117 /* 1118 * Warn that /proc/pid/oom_adj is deprecated, see 1119 * Documentation/feature-removal-schedule.txt. 1120 */ 1121 WARN_ONCE(1, "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n", 1122 current->comm, task_pid_nr(current), task_pid_nr(task), 1123 task_pid_nr(task)); 1124 task->signal->oom_adj = oom_adjust; 1125 /* 1126 * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum 1127 * value is always attainable. 1128 */ 1129 if (task->signal->oom_adj == OOM_ADJUST_MAX) 1130 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX; 1131 else 1132 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) / 1133 -OOM_DISABLE; 1134 err_sighand: 1135 unlock_task_sighand(task, &flags); 1136 err_task_lock: 1137 task_unlock(task); 1138 put_task_struct(task); 1139 out: 1140 return err < 0 ? err : count; 1141 } 1142 1143 static const struct file_operations proc_oom_adjust_operations = { 1144 .read = oom_adjust_read, 1145 .write = oom_adjust_write, 1146 .llseek = generic_file_llseek, 1147 }; 1148 1149 static ssize_t oom_score_adj_read(struct file *file, char __user *buf, 1150 size_t count, loff_t *ppos) 1151 { 1152 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 1153 char buffer[PROC_NUMBUF]; 1154 int oom_score_adj = OOM_SCORE_ADJ_MIN; 1155 unsigned long flags; 1156 size_t len; 1157 1158 if (!task) 1159 return -ESRCH; 1160 if (lock_task_sighand(task, &flags)) { 1161 oom_score_adj = task->signal->oom_score_adj; 1162 unlock_task_sighand(task, &flags); 1163 } 1164 put_task_struct(task); 1165 len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj); 1166 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1167 } 1168 1169 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf, 1170 size_t count, loff_t *ppos) 1171 { 1172 struct task_struct *task; 1173 char buffer[PROC_NUMBUF]; 1174 unsigned long flags; 1175 int oom_score_adj; 1176 int err; 1177 1178 memset(buffer, 0, sizeof(buffer)); 1179 if (count > sizeof(buffer) - 1) 1180 count = sizeof(buffer) - 1; 1181 if (copy_from_user(buffer, buf, count)) { 1182 err = -EFAULT; 1183 goto out; 1184 } 1185 1186 err = kstrtoint(strstrip(buffer), 0, &oom_score_adj); 1187 if (err) 1188 goto out; 1189 if (oom_score_adj < OOM_SCORE_ADJ_MIN || 1190 oom_score_adj > OOM_SCORE_ADJ_MAX) { 1191 err = -EINVAL; 1192 goto out; 1193 } 1194 1195 task = get_proc_task(file->f_path.dentry->d_inode); 1196 if (!task) { 1197 err = -ESRCH; 1198 goto out; 1199 } 1200 1201 task_lock(task); 1202 if (!task->mm) { 1203 err = -EINVAL; 1204 goto err_task_lock; 1205 } 1206 1207 if (!lock_task_sighand(task, &flags)) { 1208 err = -ESRCH; 1209 goto err_task_lock; 1210 } 1211 1212 if (oom_score_adj < task->signal->oom_score_adj_min && 1213 !capable(CAP_SYS_RESOURCE)) { 1214 err = -EACCES; 1215 goto err_sighand; 1216 } 1217 1218 if (oom_score_adj != task->signal->oom_score_adj) { 1219 if (oom_score_adj == OOM_SCORE_ADJ_MIN) 1220 atomic_inc(&task->mm->oom_disable_count); 1221 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) 1222 atomic_dec(&task->mm->oom_disable_count); 1223 } 1224 task->signal->oom_score_adj = oom_score_adj; 1225 if (has_capability_noaudit(current, CAP_SYS_RESOURCE)) 1226 task->signal->oom_score_adj_min = oom_score_adj; 1227 /* 1228 * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is 1229 * always attainable. 1230 */ 1231 if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) 1232 task->signal->oom_adj = OOM_DISABLE; 1233 else 1234 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) / 1235 OOM_SCORE_ADJ_MAX; 1236 err_sighand: 1237 unlock_task_sighand(task, &flags); 1238 err_task_lock: 1239 task_unlock(task); 1240 put_task_struct(task); 1241 out: 1242 return err < 0 ? err : count; 1243 } 1244 1245 static const struct file_operations proc_oom_score_adj_operations = { 1246 .read = oom_score_adj_read, 1247 .write = oom_score_adj_write, 1248 .llseek = default_llseek, 1249 }; 1250 1251 #ifdef CONFIG_AUDITSYSCALL 1252 #define TMPBUFLEN 21 1253 static ssize_t proc_loginuid_read(struct file * file, char __user * buf, 1254 size_t count, loff_t *ppos) 1255 { 1256 struct inode * inode = file->f_path.dentry->d_inode; 1257 struct task_struct *task = get_proc_task(inode); 1258 ssize_t length; 1259 char tmpbuf[TMPBUFLEN]; 1260 1261 if (!task) 1262 return -ESRCH; 1263 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 1264 audit_get_loginuid(task)); 1265 put_task_struct(task); 1266 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 1267 } 1268 1269 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf, 1270 size_t count, loff_t *ppos) 1271 { 1272 struct inode * inode = file->f_path.dentry->d_inode; 1273 char *page, *tmp; 1274 ssize_t length; 1275 uid_t loginuid; 1276 1277 if (!capable(CAP_AUDIT_CONTROL)) 1278 return -EPERM; 1279 1280 rcu_read_lock(); 1281 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) { 1282 rcu_read_unlock(); 1283 return -EPERM; 1284 } 1285 rcu_read_unlock(); 1286 1287 if (count >= PAGE_SIZE) 1288 count = PAGE_SIZE - 1; 1289 1290 if (*ppos != 0) { 1291 /* No partial writes. */ 1292 return -EINVAL; 1293 } 1294 page = (char*)__get_free_page(GFP_TEMPORARY); 1295 if (!page) 1296 return -ENOMEM; 1297 length = -EFAULT; 1298 if (copy_from_user(page, buf, count)) 1299 goto out_free_page; 1300 1301 page[count] = '\0'; 1302 loginuid = simple_strtoul(page, &tmp, 10); 1303 if (tmp == page) { 1304 length = -EINVAL; 1305 goto out_free_page; 1306 1307 } 1308 length = audit_set_loginuid(current, loginuid); 1309 if (likely(length == 0)) 1310 length = count; 1311 1312 out_free_page: 1313 free_page((unsigned long) page); 1314 return length; 1315 } 1316 1317 static const struct file_operations proc_loginuid_operations = { 1318 .read = proc_loginuid_read, 1319 .write = proc_loginuid_write, 1320 .llseek = generic_file_llseek, 1321 }; 1322 1323 static ssize_t proc_sessionid_read(struct file * file, char __user * buf, 1324 size_t count, loff_t *ppos) 1325 { 1326 struct inode * inode = file->f_path.dentry->d_inode; 1327 struct task_struct *task = get_proc_task(inode); 1328 ssize_t length; 1329 char tmpbuf[TMPBUFLEN]; 1330 1331 if (!task) 1332 return -ESRCH; 1333 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 1334 audit_get_sessionid(task)); 1335 put_task_struct(task); 1336 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 1337 } 1338 1339 static const struct file_operations proc_sessionid_operations = { 1340 .read = proc_sessionid_read, 1341 .llseek = generic_file_llseek, 1342 }; 1343 #endif 1344 1345 #ifdef CONFIG_FAULT_INJECTION 1346 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf, 1347 size_t count, loff_t *ppos) 1348 { 1349 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 1350 char buffer[PROC_NUMBUF]; 1351 size_t len; 1352 int make_it_fail; 1353 1354 if (!task) 1355 return -ESRCH; 1356 make_it_fail = task->make_it_fail; 1357 put_task_struct(task); 1358 1359 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail); 1360 1361 return simple_read_from_buffer(buf, count, ppos, buffer, len); 1362 } 1363 1364 static ssize_t proc_fault_inject_write(struct file * file, 1365 const char __user * buf, size_t count, loff_t *ppos) 1366 { 1367 struct task_struct *task; 1368 char buffer[PROC_NUMBUF], *end; 1369 int make_it_fail; 1370 1371 if (!capable(CAP_SYS_RESOURCE)) 1372 return -EPERM; 1373 memset(buffer, 0, sizeof(buffer)); 1374 if (count > sizeof(buffer) - 1) 1375 count = sizeof(buffer) - 1; 1376 if (copy_from_user(buffer, buf, count)) 1377 return -EFAULT; 1378 make_it_fail = simple_strtol(strstrip(buffer), &end, 0); 1379 if (*end) 1380 return -EINVAL; 1381 task = get_proc_task(file->f_dentry->d_inode); 1382 if (!task) 1383 return -ESRCH; 1384 task->make_it_fail = make_it_fail; 1385 put_task_struct(task); 1386 1387 return count; 1388 } 1389 1390 static const struct file_operations proc_fault_inject_operations = { 1391 .read = proc_fault_inject_read, 1392 .write = proc_fault_inject_write, 1393 .llseek = generic_file_llseek, 1394 }; 1395 #endif 1396 1397 1398 #ifdef CONFIG_SCHED_DEBUG 1399 /* 1400 * Print out various scheduling related per-task fields: 1401 */ 1402 static int sched_show(struct seq_file *m, void *v) 1403 { 1404 struct inode *inode = m->private; 1405 struct task_struct *p; 1406 1407 p = get_proc_task(inode); 1408 if (!p) 1409 return -ESRCH; 1410 proc_sched_show_task(p, m); 1411 1412 put_task_struct(p); 1413 1414 return 0; 1415 } 1416 1417 static ssize_t 1418 sched_write(struct file *file, const char __user *buf, 1419 size_t count, loff_t *offset) 1420 { 1421 struct inode *inode = file->f_path.dentry->d_inode; 1422 struct task_struct *p; 1423 1424 p = get_proc_task(inode); 1425 if (!p) 1426 return -ESRCH; 1427 proc_sched_set_task(p); 1428 1429 put_task_struct(p); 1430 1431 return count; 1432 } 1433 1434 static int sched_open(struct inode *inode, struct file *filp) 1435 { 1436 return single_open(filp, sched_show, inode); 1437 } 1438 1439 static const struct file_operations proc_pid_sched_operations = { 1440 .open = sched_open, 1441 .read = seq_read, 1442 .write = sched_write, 1443 .llseek = seq_lseek, 1444 .release = single_release, 1445 }; 1446 1447 #endif 1448 1449 #ifdef CONFIG_SCHED_AUTOGROUP 1450 /* 1451 * Print out autogroup related information: 1452 */ 1453 static int sched_autogroup_show(struct seq_file *m, void *v) 1454 { 1455 struct inode *inode = m->private; 1456 struct task_struct *p; 1457 1458 p = get_proc_task(inode); 1459 if (!p) 1460 return -ESRCH; 1461 proc_sched_autogroup_show_task(p, m); 1462 1463 put_task_struct(p); 1464 1465 return 0; 1466 } 1467 1468 static ssize_t 1469 sched_autogroup_write(struct file *file, const char __user *buf, 1470 size_t count, loff_t *offset) 1471 { 1472 struct inode *inode = file->f_path.dentry->d_inode; 1473 struct task_struct *p; 1474 char buffer[PROC_NUMBUF]; 1475 int nice; 1476 int err; 1477 1478 memset(buffer, 0, sizeof(buffer)); 1479 if (count > sizeof(buffer) - 1) 1480 count = sizeof(buffer) - 1; 1481 if (copy_from_user(buffer, buf, count)) 1482 return -EFAULT; 1483 1484 err = kstrtoint(strstrip(buffer), 0, &nice); 1485 if (err < 0) 1486 return err; 1487 1488 p = get_proc_task(inode); 1489 if (!p) 1490 return -ESRCH; 1491 1492 err = nice; 1493 err = proc_sched_autogroup_set_nice(p, &err); 1494 if (err) 1495 count = err; 1496 1497 put_task_struct(p); 1498 1499 return count; 1500 } 1501 1502 static int sched_autogroup_open(struct inode *inode, struct file *filp) 1503 { 1504 int ret; 1505 1506 ret = single_open(filp, sched_autogroup_show, NULL); 1507 if (!ret) { 1508 struct seq_file *m = filp->private_data; 1509 1510 m->private = inode; 1511 } 1512 return ret; 1513 } 1514 1515 static const struct file_operations proc_pid_sched_autogroup_operations = { 1516 .open = sched_autogroup_open, 1517 .read = seq_read, 1518 .write = sched_autogroup_write, 1519 .llseek = seq_lseek, 1520 .release = single_release, 1521 }; 1522 1523 #endif /* CONFIG_SCHED_AUTOGROUP */ 1524 1525 static ssize_t comm_write(struct file *file, const char __user *buf, 1526 size_t count, loff_t *offset) 1527 { 1528 struct inode *inode = file->f_path.dentry->d_inode; 1529 struct task_struct *p; 1530 char buffer[TASK_COMM_LEN]; 1531 1532 memset(buffer, 0, sizeof(buffer)); 1533 if (count > sizeof(buffer) - 1) 1534 count = sizeof(buffer) - 1; 1535 if (copy_from_user(buffer, buf, count)) 1536 return -EFAULT; 1537 1538 p = get_proc_task(inode); 1539 if (!p) 1540 return -ESRCH; 1541 1542 if (same_thread_group(current, p)) 1543 set_task_comm(p, buffer); 1544 else 1545 count = -EINVAL; 1546 1547 put_task_struct(p); 1548 1549 return count; 1550 } 1551 1552 static int comm_show(struct seq_file *m, void *v) 1553 { 1554 struct inode *inode = m->private; 1555 struct task_struct *p; 1556 1557 p = get_proc_task(inode); 1558 if (!p) 1559 return -ESRCH; 1560 1561 task_lock(p); 1562 seq_printf(m, "%s\n", p->comm); 1563 task_unlock(p); 1564 1565 put_task_struct(p); 1566 1567 return 0; 1568 } 1569 1570 static int comm_open(struct inode *inode, struct file *filp) 1571 { 1572 return single_open(filp, comm_show, inode); 1573 } 1574 1575 static const struct file_operations proc_pid_set_comm_operations = { 1576 .open = comm_open, 1577 .read = seq_read, 1578 .write = comm_write, 1579 .llseek = seq_lseek, 1580 .release = single_release, 1581 }; 1582 1583 static int proc_exe_link(struct inode *inode, struct path *exe_path) 1584 { 1585 struct task_struct *task; 1586 struct mm_struct *mm; 1587 struct file *exe_file; 1588 1589 task = get_proc_task(inode); 1590 if (!task) 1591 return -ENOENT; 1592 mm = get_task_mm(task); 1593 put_task_struct(task); 1594 if (!mm) 1595 return -ENOENT; 1596 exe_file = get_mm_exe_file(mm); 1597 mmput(mm); 1598 if (exe_file) { 1599 *exe_path = exe_file->f_path; 1600 path_get(&exe_file->f_path); 1601 fput(exe_file); 1602 return 0; 1603 } else 1604 return -ENOENT; 1605 } 1606 1607 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd) 1608 { 1609 struct inode *inode = dentry->d_inode; 1610 int error = -EACCES; 1611 1612 /* We don't need a base pointer in the /proc filesystem */ 1613 path_put(&nd->path); 1614 1615 /* Are we allowed to snoop on the tasks file descriptors? */ 1616 if (!proc_fd_access_allowed(inode)) 1617 goto out; 1618 1619 error = PROC_I(inode)->op.proc_get_link(inode, &nd->path); 1620 out: 1621 return ERR_PTR(error); 1622 } 1623 1624 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen) 1625 { 1626 char *tmp = (char*)__get_free_page(GFP_TEMPORARY); 1627 char *pathname; 1628 int len; 1629 1630 if (!tmp) 1631 return -ENOMEM; 1632 1633 pathname = d_path(path, tmp, PAGE_SIZE); 1634 len = PTR_ERR(pathname); 1635 if (IS_ERR(pathname)) 1636 goto out; 1637 len = tmp + PAGE_SIZE - 1 - pathname; 1638 1639 if (len > buflen) 1640 len = buflen; 1641 if (copy_to_user(buffer, pathname, len)) 1642 len = -EFAULT; 1643 out: 1644 free_page((unsigned long)tmp); 1645 return len; 1646 } 1647 1648 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen) 1649 { 1650 int error = -EACCES; 1651 struct inode *inode = dentry->d_inode; 1652 struct path path; 1653 1654 /* Are we allowed to snoop on the tasks file descriptors? */ 1655 if (!proc_fd_access_allowed(inode)) 1656 goto out; 1657 1658 error = PROC_I(inode)->op.proc_get_link(inode, &path); 1659 if (error) 1660 goto out; 1661 1662 error = do_proc_readlink(&path, buffer, buflen); 1663 path_put(&path); 1664 out: 1665 return error; 1666 } 1667 1668 static const struct inode_operations proc_pid_link_inode_operations = { 1669 .readlink = proc_pid_readlink, 1670 .follow_link = proc_pid_follow_link, 1671 .setattr = proc_setattr, 1672 }; 1673 1674 1675 /* building an inode */ 1676 1677 static int task_dumpable(struct task_struct *task) 1678 { 1679 int dumpable = 0; 1680 struct mm_struct *mm; 1681 1682 task_lock(task); 1683 mm = task->mm; 1684 if (mm) 1685 dumpable = get_dumpable(mm); 1686 task_unlock(task); 1687 if(dumpable == 1) 1688 return 1; 1689 return 0; 1690 } 1691 1692 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task) 1693 { 1694 struct inode * inode; 1695 struct proc_inode *ei; 1696 const struct cred *cred; 1697 1698 /* We need a new inode */ 1699 1700 inode = new_inode(sb); 1701 if (!inode) 1702 goto out; 1703 1704 /* Common stuff */ 1705 ei = PROC_I(inode); 1706 inode->i_ino = get_next_ino(); 1707 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 1708 inode->i_op = &proc_def_inode_operations; 1709 1710 /* 1711 * grab the reference to task. 1712 */ 1713 ei->pid = get_task_pid(task, PIDTYPE_PID); 1714 if (!ei->pid) 1715 goto out_unlock; 1716 1717 if (task_dumpable(task)) { 1718 rcu_read_lock(); 1719 cred = __task_cred(task); 1720 inode->i_uid = cred->euid; 1721 inode->i_gid = cred->egid; 1722 rcu_read_unlock(); 1723 } 1724 security_task_to_inode(task, inode); 1725 1726 out: 1727 return inode; 1728 1729 out_unlock: 1730 iput(inode); 1731 return NULL; 1732 } 1733 1734 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 1735 { 1736 struct inode *inode = dentry->d_inode; 1737 struct task_struct *task; 1738 const struct cred *cred; 1739 1740 generic_fillattr(inode, stat); 1741 1742 rcu_read_lock(); 1743 stat->uid = 0; 1744 stat->gid = 0; 1745 task = pid_task(proc_pid(inode), PIDTYPE_PID); 1746 if (task) { 1747 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) || 1748 task_dumpable(task)) { 1749 cred = __task_cred(task); 1750 stat->uid = cred->euid; 1751 stat->gid = cred->egid; 1752 } 1753 } 1754 rcu_read_unlock(); 1755 return 0; 1756 } 1757 1758 /* dentry stuff */ 1759 1760 /* 1761 * Exceptional case: normally we are not allowed to unhash a busy 1762 * directory. In this case, however, we can do it - no aliasing problems 1763 * due to the way we treat inodes. 1764 * 1765 * Rewrite the inode's ownerships here because the owning task may have 1766 * performed a setuid(), etc. 1767 * 1768 * Before the /proc/pid/status file was created the only way to read 1769 * the effective uid of a /process was to stat /proc/pid. Reading 1770 * /proc/pid/status is slow enough that procps and other packages 1771 * kept stating /proc/pid. To keep the rules in /proc simple I have 1772 * made this apply to all per process world readable and executable 1773 * directories. 1774 */ 1775 int pid_revalidate(struct dentry *dentry, struct nameidata *nd) 1776 { 1777 struct inode *inode; 1778 struct task_struct *task; 1779 const struct cred *cred; 1780 1781 if (nd && nd->flags & LOOKUP_RCU) 1782 return -ECHILD; 1783 1784 inode = dentry->d_inode; 1785 task = get_proc_task(inode); 1786 1787 if (task) { 1788 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) || 1789 task_dumpable(task)) { 1790 rcu_read_lock(); 1791 cred = __task_cred(task); 1792 inode->i_uid = cred->euid; 1793 inode->i_gid = cred->egid; 1794 rcu_read_unlock(); 1795 } else { 1796 inode->i_uid = 0; 1797 inode->i_gid = 0; 1798 } 1799 inode->i_mode &= ~(S_ISUID | S_ISGID); 1800 security_task_to_inode(task, inode); 1801 put_task_struct(task); 1802 return 1; 1803 } 1804 d_drop(dentry); 1805 return 0; 1806 } 1807 1808 static int pid_delete_dentry(const struct dentry * dentry) 1809 { 1810 /* Is the task we represent dead? 1811 * If so, then don't put the dentry on the lru list, 1812 * kill it immediately. 1813 */ 1814 return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first; 1815 } 1816 1817 const struct dentry_operations pid_dentry_operations = 1818 { 1819 .d_revalidate = pid_revalidate, 1820 .d_delete = pid_delete_dentry, 1821 }; 1822 1823 /* Lookups */ 1824 1825 /* 1826 * Fill a directory entry. 1827 * 1828 * If possible create the dcache entry and derive our inode number and 1829 * file type from dcache entry. 1830 * 1831 * Since all of the proc inode numbers are dynamically generated, the inode 1832 * numbers do not exist until the inode is cache. This means creating the 1833 * the dcache entry in readdir is necessary to keep the inode numbers 1834 * reported by readdir in sync with the inode numbers reported 1835 * by stat. 1836 */ 1837 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir, 1838 const char *name, int len, 1839 instantiate_t instantiate, struct task_struct *task, const void *ptr) 1840 { 1841 struct dentry *child, *dir = filp->f_path.dentry; 1842 struct inode *inode; 1843 struct qstr qname; 1844 ino_t ino = 0; 1845 unsigned type = DT_UNKNOWN; 1846 1847 qname.name = name; 1848 qname.len = len; 1849 qname.hash = full_name_hash(name, len); 1850 1851 child = d_lookup(dir, &qname); 1852 if (!child) { 1853 struct dentry *new; 1854 new = d_alloc(dir, &qname); 1855 if (new) { 1856 child = instantiate(dir->d_inode, new, task, ptr); 1857 if (child) 1858 dput(new); 1859 else 1860 child = new; 1861 } 1862 } 1863 if (!child || IS_ERR(child) || !child->d_inode) 1864 goto end_instantiate; 1865 inode = child->d_inode; 1866 if (inode) { 1867 ino = inode->i_ino; 1868 type = inode->i_mode >> 12; 1869 } 1870 dput(child); 1871 end_instantiate: 1872 if (!ino) 1873 ino = find_inode_number(dir, &qname); 1874 if (!ino) 1875 ino = 1; 1876 return filldir(dirent, name, len, filp->f_pos, ino, type); 1877 } 1878 1879 static unsigned name_to_int(struct dentry *dentry) 1880 { 1881 const char *name = dentry->d_name.name; 1882 int len = dentry->d_name.len; 1883 unsigned n = 0; 1884 1885 if (len > 1 && *name == '0') 1886 goto out; 1887 while (len-- > 0) { 1888 unsigned c = *name++ - '0'; 1889 if (c > 9) 1890 goto out; 1891 if (n >= (~0U-9)/10) 1892 goto out; 1893 n *= 10; 1894 n += c; 1895 } 1896 return n; 1897 out: 1898 return ~0U; 1899 } 1900 1901 #define PROC_FDINFO_MAX 64 1902 1903 static int proc_fd_info(struct inode *inode, struct path *path, char *info) 1904 { 1905 struct task_struct *task = get_proc_task(inode); 1906 struct files_struct *files = NULL; 1907 struct file *file; 1908 int fd = proc_fd(inode); 1909 1910 if (task) { 1911 files = get_files_struct(task); 1912 put_task_struct(task); 1913 } 1914 if (files) { 1915 /* 1916 * We are not taking a ref to the file structure, so we must 1917 * hold ->file_lock. 1918 */ 1919 spin_lock(&files->file_lock); 1920 file = fcheck_files(files, fd); 1921 if (file) { 1922 if (path) { 1923 *path = file->f_path; 1924 path_get(&file->f_path); 1925 } 1926 if (info) 1927 snprintf(info, PROC_FDINFO_MAX, 1928 "pos:\t%lli\n" 1929 "flags:\t0%o\n", 1930 (long long) file->f_pos, 1931 file->f_flags); 1932 spin_unlock(&files->file_lock); 1933 put_files_struct(files); 1934 return 0; 1935 } 1936 spin_unlock(&files->file_lock); 1937 put_files_struct(files); 1938 } 1939 return -ENOENT; 1940 } 1941 1942 static int proc_fd_link(struct inode *inode, struct path *path) 1943 { 1944 return proc_fd_info(inode, path, NULL); 1945 } 1946 1947 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd) 1948 { 1949 struct inode *inode; 1950 struct task_struct *task; 1951 int fd; 1952 struct files_struct *files; 1953 const struct cred *cred; 1954 1955 if (nd && nd->flags & LOOKUP_RCU) 1956 return -ECHILD; 1957 1958 inode = dentry->d_inode; 1959 task = get_proc_task(inode); 1960 fd = proc_fd(inode); 1961 1962 if (task) { 1963 files = get_files_struct(task); 1964 if (files) { 1965 rcu_read_lock(); 1966 if (fcheck_files(files, fd)) { 1967 rcu_read_unlock(); 1968 put_files_struct(files); 1969 if (task_dumpable(task)) { 1970 rcu_read_lock(); 1971 cred = __task_cred(task); 1972 inode->i_uid = cred->euid; 1973 inode->i_gid = cred->egid; 1974 rcu_read_unlock(); 1975 } else { 1976 inode->i_uid = 0; 1977 inode->i_gid = 0; 1978 } 1979 inode->i_mode &= ~(S_ISUID | S_ISGID); 1980 security_task_to_inode(task, inode); 1981 put_task_struct(task); 1982 return 1; 1983 } 1984 rcu_read_unlock(); 1985 put_files_struct(files); 1986 } 1987 put_task_struct(task); 1988 } 1989 d_drop(dentry); 1990 return 0; 1991 } 1992 1993 static const struct dentry_operations tid_fd_dentry_operations = 1994 { 1995 .d_revalidate = tid_fd_revalidate, 1996 .d_delete = pid_delete_dentry, 1997 }; 1998 1999 static struct dentry *proc_fd_instantiate(struct inode *dir, 2000 struct dentry *dentry, struct task_struct *task, const void *ptr) 2001 { 2002 unsigned fd = *(const unsigned *)ptr; 2003 struct file *file; 2004 struct files_struct *files; 2005 struct inode *inode; 2006 struct proc_inode *ei; 2007 struct dentry *error = ERR_PTR(-ENOENT); 2008 2009 inode = proc_pid_make_inode(dir->i_sb, task); 2010 if (!inode) 2011 goto out; 2012 ei = PROC_I(inode); 2013 ei->fd = fd; 2014 files = get_files_struct(task); 2015 if (!files) 2016 goto out_iput; 2017 inode->i_mode = S_IFLNK; 2018 2019 /* 2020 * We are not taking a ref to the file structure, so we must 2021 * hold ->file_lock. 2022 */ 2023 spin_lock(&files->file_lock); 2024 file = fcheck_files(files, fd); 2025 if (!file) 2026 goto out_unlock; 2027 if (file->f_mode & FMODE_READ) 2028 inode->i_mode |= S_IRUSR | S_IXUSR; 2029 if (file->f_mode & FMODE_WRITE) 2030 inode->i_mode |= S_IWUSR | S_IXUSR; 2031 spin_unlock(&files->file_lock); 2032 put_files_struct(files); 2033 2034 inode->i_op = &proc_pid_link_inode_operations; 2035 inode->i_size = 64; 2036 ei->op.proc_get_link = proc_fd_link; 2037 d_set_d_op(dentry, &tid_fd_dentry_operations); 2038 d_add(dentry, inode); 2039 /* Close the race of the process dying before we return the dentry */ 2040 if (tid_fd_revalidate(dentry, NULL)) 2041 error = NULL; 2042 2043 out: 2044 return error; 2045 out_unlock: 2046 spin_unlock(&files->file_lock); 2047 put_files_struct(files); 2048 out_iput: 2049 iput(inode); 2050 goto out; 2051 } 2052 2053 static struct dentry *proc_lookupfd_common(struct inode *dir, 2054 struct dentry *dentry, 2055 instantiate_t instantiate) 2056 { 2057 struct task_struct *task = get_proc_task(dir); 2058 unsigned fd = name_to_int(dentry); 2059 struct dentry *result = ERR_PTR(-ENOENT); 2060 2061 if (!task) 2062 goto out_no_task; 2063 if (fd == ~0U) 2064 goto out; 2065 2066 result = instantiate(dir, dentry, task, &fd); 2067 out: 2068 put_task_struct(task); 2069 out_no_task: 2070 return result; 2071 } 2072 2073 static int proc_readfd_common(struct file * filp, void * dirent, 2074 filldir_t filldir, instantiate_t instantiate) 2075 { 2076 struct dentry *dentry = filp->f_path.dentry; 2077 struct inode *inode = dentry->d_inode; 2078 struct task_struct *p = get_proc_task(inode); 2079 unsigned int fd, ino; 2080 int retval; 2081 struct files_struct * files; 2082 2083 retval = -ENOENT; 2084 if (!p) 2085 goto out_no_task; 2086 retval = 0; 2087 2088 fd = filp->f_pos; 2089 switch (fd) { 2090 case 0: 2091 if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0) 2092 goto out; 2093 filp->f_pos++; 2094 case 1: 2095 ino = parent_ino(dentry); 2096 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0) 2097 goto out; 2098 filp->f_pos++; 2099 default: 2100 files = get_files_struct(p); 2101 if (!files) 2102 goto out; 2103 rcu_read_lock(); 2104 for (fd = filp->f_pos-2; 2105 fd < files_fdtable(files)->max_fds; 2106 fd++, filp->f_pos++) { 2107 char name[PROC_NUMBUF]; 2108 int len; 2109 2110 if (!fcheck_files(files, fd)) 2111 continue; 2112 rcu_read_unlock(); 2113 2114 len = snprintf(name, sizeof(name), "%d", fd); 2115 if (proc_fill_cache(filp, dirent, filldir, 2116 name, len, instantiate, 2117 p, &fd) < 0) { 2118 rcu_read_lock(); 2119 break; 2120 } 2121 rcu_read_lock(); 2122 } 2123 rcu_read_unlock(); 2124 put_files_struct(files); 2125 } 2126 out: 2127 put_task_struct(p); 2128 out_no_task: 2129 return retval; 2130 } 2131 2132 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry, 2133 struct nameidata *nd) 2134 { 2135 return proc_lookupfd_common(dir, dentry, proc_fd_instantiate); 2136 } 2137 2138 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir) 2139 { 2140 return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate); 2141 } 2142 2143 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf, 2144 size_t len, loff_t *ppos) 2145 { 2146 char tmp[PROC_FDINFO_MAX]; 2147 int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp); 2148 if (!err) 2149 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp)); 2150 return err; 2151 } 2152 2153 static const struct file_operations proc_fdinfo_file_operations = { 2154 .open = nonseekable_open, 2155 .read = proc_fdinfo_read, 2156 .llseek = no_llseek, 2157 }; 2158 2159 static const struct file_operations proc_fd_operations = { 2160 .read = generic_read_dir, 2161 .readdir = proc_readfd, 2162 .llseek = default_llseek, 2163 }; 2164 2165 /* 2166 * /proc/pid/fd needs a special permission handler so that a process can still 2167 * access /proc/self/fd after it has executed a setuid(). 2168 */ 2169 static int proc_fd_permission(struct inode *inode, int mask) 2170 { 2171 int rv = generic_permission(inode, mask); 2172 if (rv == 0) 2173 return 0; 2174 if (task_pid(current) == proc_pid(inode)) 2175 rv = 0; 2176 return rv; 2177 } 2178 2179 /* 2180 * proc directories can do almost nothing.. 2181 */ 2182 static const struct inode_operations proc_fd_inode_operations = { 2183 .lookup = proc_lookupfd, 2184 .permission = proc_fd_permission, 2185 .setattr = proc_setattr, 2186 }; 2187 2188 static struct dentry *proc_fdinfo_instantiate(struct inode *dir, 2189 struct dentry *dentry, struct task_struct *task, const void *ptr) 2190 { 2191 unsigned fd = *(unsigned *)ptr; 2192 struct inode *inode; 2193 struct proc_inode *ei; 2194 struct dentry *error = ERR_PTR(-ENOENT); 2195 2196 inode = proc_pid_make_inode(dir->i_sb, task); 2197 if (!inode) 2198 goto out; 2199 ei = PROC_I(inode); 2200 ei->fd = fd; 2201 inode->i_mode = S_IFREG | S_IRUSR; 2202 inode->i_fop = &proc_fdinfo_file_operations; 2203 d_set_d_op(dentry, &tid_fd_dentry_operations); 2204 d_add(dentry, inode); 2205 /* Close the race of the process dying before we return the dentry */ 2206 if (tid_fd_revalidate(dentry, NULL)) 2207 error = NULL; 2208 2209 out: 2210 return error; 2211 } 2212 2213 static struct dentry *proc_lookupfdinfo(struct inode *dir, 2214 struct dentry *dentry, 2215 struct nameidata *nd) 2216 { 2217 return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate); 2218 } 2219 2220 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir) 2221 { 2222 return proc_readfd_common(filp, dirent, filldir, 2223 proc_fdinfo_instantiate); 2224 } 2225 2226 static const struct file_operations proc_fdinfo_operations = { 2227 .read = generic_read_dir, 2228 .readdir = proc_readfdinfo, 2229 .llseek = default_llseek, 2230 }; 2231 2232 /* 2233 * proc directories can do almost nothing.. 2234 */ 2235 static const struct inode_operations proc_fdinfo_inode_operations = { 2236 .lookup = proc_lookupfdinfo, 2237 .setattr = proc_setattr, 2238 }; 2239 2240 2241 static struct dentry *proc_pident_instantiate(struct inode *dir, 2242 struct dentry *dentry, struct task_struct *task, const void *ptr) 2243 { 2244 const struct pid_entry *p = ptr; 2245 struct inode *inode; 2246 struct proc_inode *ei; 2247 struct dentry *error = ERR_PTR(-ENOENT); 2248 2249 inode = proc_pid_make_inode(dir->i_sb, task); 2250 if (!inode) 2251 goto out; 2252 2253 ei = PROC_I(inode); 2254 inode->i_mode = p->mode; 2255 if (S_ISDIR(inode->i_mode)) 2256 inode->i_nlink = 2; /* Use getattr to fix if necessary */ 2257 if (p->iop) 2258 inode->i_op = p->iop; 2259 if (p->fop) 2260 inode->i_fop = p->fop; 2261 ei->op = p->op; 2262 d_set_d_op(dentry, &pid_dentry_operations); 2263 d_add(dentry, inode); 2264 /* Close the race of the process dying before we return the dentry */ 2265 if (pid_revalidate(dentry, NULL)) 2266 error = NULL; 2267 out: 2268 return error; 2269 } 2270 2271 static struct dentry *proc_pident_lookup(struct inode *dir, 2272 struct dentry *dentry, 2273 const struct pid_entry *ents, 2274 unsigned int nents) 2275 { 2276 struct dentry *error; 2277 struct task_struct *task = get_proc_task(dir); 2278 const struct pid_entry *p, *last; 2279 2280 error = ERR_PTR(-ENOENT); 2281 2282 if (!task) 2283 goto out_no_task; 2284 2285 /* 2286 * Yes, it does not scale. And it should not. Don't add 2287 * new entries into /proc/<tgid>/ without very good reasons. 2288 */ 2289 last = &ents[nents - 1]; 2290 for (p = ents; p <= last; p++) { 2291 if (p->len != dentry->d_name.len) 2292 continue; 2293 if (!memcmp(dentry->d_name.name, p->name, p->len)) 2294 break; 2295 } 2296 if (p > last) 2297 goto out; 2298 2299 error = proc_pident_instantiate(dir, dentry, task, p); 2300 out: 2301 put_task_struct(task); 2302 out_no_task: 2303 return error; 2304 } 2305 2306 static int proc_pident_fill_cache(struct file *filp, void *dirent, 2307 filldir_t filldir, struct task_struct *task, const struct pid_entry *p) 2308 { 2309 return proc_fill_cache(filp, dirent, filldir, p->name, p->len, 2310 proc_pident_instantiate, task, p); 2311 } 2312 2313 static int proc_pident_readdir(struct file *filp, 2314 void *dirent, filldir_t filldir, 2315 const struct pid_entry *ents, unsigned int nents) 2316 { 2317 int i; 2318 struct dentry *dentry = filp->f_path.dentry; 2319 struct inode *inode = dentry->d_inode; 2320 struct task_struct *task = get_proc_task(inode); 2321 const struct pid_entry *p, *last; 2322 ino_t ino; 2323 int ret; 2324 2325 ret = -ENOENT; 2326 if (!task) 2327 goto out_no_task; 2328 2329 ret = 0; 2330 i = filp->f_pos; 2331 switch (i) { 2332 case 0: 2333 ino = inode->i_ino; 2334 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0) 2335 goto out; 2336 i++; 2337 filp->f_pos++; 2338 /* fall through */ 2339 case 1: 2340 ino = parent_ino(dentry); 2341 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0) 2342 goto out; 2343 i++; 2344 filp->f_pos++; 2345 /* fall through */ 2346 default: 2347 i -= 2; 2348 if (i >= nents) { 2349 ret = 1; 2350 goto out; 2351 } 2352 p = ents + i; 2353 last = &ents[nents - 1]; 2354 while (p <= last) { 2355 if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0) 2356 goto out; 2357 filp->f_pos++; 2358 p++; 2359 } 2360 } 2361 2362 ret = 1; 2363 out: 2364 put_task_struct(task); 2365 out_no_task: 2366 return ret; 2367 } 2368 2369 #ifdef CONFIG_SECURITY 2370 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf, 2371 size_t count, loff_t *ppos) 2372 { 2373 struct inode * inode = file->f_path.dentry->d_inode; 2374 char *p = NULL; 2375 ssize_t length; 2376 struct task_struct *task = get_proc_task(inode); 2377 2378 if (!task) 2379 return -ESRCH; 2380 2381 length = security_getprocattr(task, 2382 (char*)file->f_path.dentry->d_name.name, 2383 &p); 2384 put_task_struct(task); 2385 if (length > 0) 2386 length = simple_read_from_buffer(buf, count, ppos, p, length); 2387 kfree(p); 2388 return length; 2389 } 2390 2391 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf, 2392 size_t count, loff_t *ppos) 2393 { 2394 struct inode * inode = file->f_path.dentry->d_inode; 2395 char *page; 2396 ssize_t length; 2397 struct task_struct *task = get_proc_task(inode); 2398 2399 length = -ESRCH; 2400 if (!task) 2401 goto out_no_task; 2402 if (count > PAGE_SIZE) 2403 count = PAGE_SIZE; 2404 2405 /* No partial writes. */ 2406 length = -EINVAL; 2407 if (*ppos != 0) 2408 goto out; 2409 2410 length = -ENOMEM; 2411 page = (char*)__get_free_page(GFP_TEMPORARY); 2412 if (!page) 2413 goto out; 2414 2415 length = -EFAULT; 2416 if (copy_from_user(page, buf, count)) 2417 goto out_free; 2418 2419 /* Guard against adverse ptrace interaction */ 2420 length = mutex_lock_interruptible(&task->signal->cred_guard_mutex); 2421 if (length < 0) 2422 goto out_free; 2423 2424 length = security_setprocattr(task, 2425 (char*)file->f_path.dentry->d_name.name, 2426 (void*)page, count); 2427 mutex_unlock(&task->signal->cred_guard_mutex); 2428 out_free: 2429 free_page((unsigned long) page); 2430 out: 2431 put_task_struct(task); 2432 out_no_task: 2433 return length; 2434 } 2435 2436 static const struct file_operations proc_pid_attr_operations = { 2437 .read = proc_pid_attr_read, 2438 .write = proc_pid_attr_write, 2439 .llseek = generic_file_llseek, 2440 }; 2441 2442 static const struct pid_entry attr_dir_stuff[] = { 2443 REG("current", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2444 REG("prev", S_IRUGO, proc_pid_attr_operations), 2445 REG("exec", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2446 REG("fscreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2447 REG("keycreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2448 REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations), 2449 }; 2450 2451 static int proc_attr_dir_readdir(struct file * filp, 2452 void * dirent, filldir_t filldir) 2453 { 2454 return proc_pident_readdir(filp,dirent,filldir, 2455 attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff)); 2456 } 2457 2458 static const struct file_operations proc_attr_dir_operations = { 2459 .read = generic_read_dir, 2460 .readdir = proc_attr_dir_readdir, 2461 .llseek = default_llseek, 2462 }; 2463 2464 static struct dentry *proc_attr_dir_lookup(struct inode *dir, 2465 struct dentry *dentry, struct nameidata *nd) 2466 { 2467 return proc_pident_lookup(dir, dentry, 2468 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff)); 2469 } 2470 2471 static const struct inode_operations proc_attr_dir_inode_operations = { 2472 .lookup = proc_attr_dir_lookup, 2473 .getattr = pid_getattr, 2474 .setattr = proc_setattr, 2475 }; 2476 2477 #endif 2478 2479 #ifdef CONFIG_ELF_CORE 2480 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf, 2481 size_t count, loff_t *ppos) 2482 { 2483 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 2484 struct mm_struct *mm; 2485 char buffer[PROC_NUMBUF]; 2486 size_t len; 2487 int ret; 2488 2489 if (!task) 2490 return -ESRCH; 2491 2492 ret = 0; 2493 mm = get_task_mm(task); 2494 if (mm) { 2495 len = snprintf(buffer, sizeof(buffer), "%08lx\n", 2496 ((mm->flags & MMF_DUMP_FILTER_MASK) >> 2497 MMF_DUMP_FILTER_SHIFT)); 2498 mmput(mm); 2499 ret = simple_read_from_buffer(buf, count, ppos, buffer, len); 2500 } 2501 2502 put_task_struct(task); 2503 2504 return ret; 2505 } 2506 2507 static ssize_t proc_coredump_filter_write(struct file *file, 2508 const char __user *buf, 2509 size_t count, 2510 loff_t *ppos) 2511 { 2512 struct task_struct *task; 2513 struct mm_struct *mm; 2514 char buffer[PROC_NUMBUF], *end; 2515 unsigned int val; 2516 int ret; 2517 int i; 2518 unsigned long mask; 2519 2520 ret = -EFAULT; 2521 memset(buffer, 0, sizeof(buffer)); 2522 if (count > sizeof(buffer) - 1) 2523 count = sizeof(buffer) - 1; 2524 if (copy_from_user(buffer, buf, count)) 2525 goto out_no_task; 2526 2527 ret = -EINVAL; 2528 val = (unsigned int)simple_strtoul(buffer, &end, 0); 2529 if (*end == '\n') 2530 end++; 2531 if (end - buffer == 0) 2532 goto out_no_task; 2533 2534 ret = -ESRCH; 2535 task = get_proc_task(file->f_dentry->d_inode); 2536 if (!task) 2537 goto out_no_task; 2538 2539 ret = end - buffer; 2540 mm = get_task_mm(task); 2541 if (!mm) 2542 goto out_no_mm; 2543 2544 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) { 2545 if (val & mask) 2546 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2547 else 2548 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2549 } 2550 2551 mmput(mm); 2552 out_no_mm: 2553 put_task_struct(task); 2554 out_no_task: 2555 return ret; 2556 } 2557 2558 static const struct file_operations proc_coredump_filter_operations = { 2559 .read = proc_coredump_filter_read, 2560 .write = proc_coredump_filter_write, 2561 .llseek = generic_file_llseek, 2562 }; 2563 #endif 2564 2565 /* 2566 * /proc/self: 2567 */ 2568 static int proc_self_readlink(struct dentry *dentry, char __user *buffer, 2569 int buflen) 2570 { 2571 struct pid_namespace *ns = dentry->d_sb->s_fs_info; 2572 pid_t tgid = task_tgid_nr_ns(current, ns); 2573 char tmp[PROC_NUMBUF]; 2574 if (!tgid) 2575 return -ENOENT; 2576 sprintf(tmp, "%d", tgid); 2577 return vfs_readlink(dentry,buffer,buflen,tmp); 2578 } 2579 2580 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd) 2581 { 2582 struct pid_namespace *ns = dentry->d_sb->s_fs_info; 2583 pid_t tgid = task_tgid_nr_ns(current, ns); 2584 char *name = ERR_PTR(-ENOENT); 2585 if (tgid) { 2586 name = __getname(); 2587 if (!name) 2588 name = ERR_PTR(-ENOMEM); 2589 else 2590 sprintf(name, "%d", tgid); 2591 } 2592 nd_set_link(nd, name); 2593 return NULL; 2594 } 2595 2596 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd, 2597 void *cookie) 2598 { 2599 char *s = nd_get_link(nd); 2600 if (!IS_ERR(s)) 2601 __putname(s); 2602 } 2603 2604 static const struct inode_operations proc_self_inode_operations = { 2605 .readlink = proc_self_readlink, 2606 .follow_link = proc_self_follow_link, 2607 .put_link = proc_self_put_link, 2608 }; 2609 2610 /* 2611 * proc base 2612 * 2613 * These are the directory entries in the root directory of /proc 2614 * that properly belong to the /proc filesystem, as they describe 2615 * describe something that is process related. 2616 */ 2617 static const struct pid_entry proc_base_stuff[] = { 2618 NOD("self", S_IFLNK|S_IRWXUGO, 2619 &proc_self_inode_operations, NULL, {}), 2620 }; 2621 2622 static struct dentry *proc_base_instantiate(struct inode *dir, 2623 struct dentry *dentry, struct task_struct *task, const void *ptr) 2624 { 2625 const struct pid_entry *p = ptr; 2626 struct inode *inode; 2627 struct proc_inode *ei; 2628 struct dentry *error; 2629 2630 /* Allocate the inode */ 2631 error = ERR_PTR(-ENOMEM); 2632 inode = new_inode(dir->i_sb); 2633 if (!inode) 2634 goto out; 2635 2636 /* Initialize the inode */ 2637 ei = PROC_I(inode); 2638 inode->i_ino = get_next_ino(); 2639 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 2640 2641 /* 2642 * grab the reference to the task. 2643 */ 2644 ei->pid = get_task_pid(task, PIDTYPE_PID); 2645 if (!ei->pid) 2646 goto out_iput; 2647 2648 inode->i_mode = p->mode; 2649 if (S_ISDIR(inode->i_mode)) 2650 inode->i_nlink = 2; 2651 if (S_ISLNK(inode->i_mode)) 2652 inode->i_size = 64; 2653 if (p->iop) 2654 inode->i_op = p->iop; 2655 if (p->fop) 2656 inode->i_fop = p->fop; 2657 ei->op = p->op; 2658 d_add(dentry, inode); 2659 error = NULL; 2660 out: 2661 return error; 2662 out_iput: 2663 iput(inode); 2664 goto out; 2665 } 2666 2667 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry) 2668 { 2669 struct dentry *error; 2670 struct task_struct *task = get_proc_task(dir); 2671 const struct pid_entry *p, *last; 2672 2673 error = ERR_PTR(-ENOENT); 2674 2675 if (!task) 2676 goto out_no_task; 2677 2678 /* Lookup the directory entry */ 2679 last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1]; 2680 for (p = proc_base_stuff; p <= last; p++) { 2681 if (p->len != dentry->d_name.len) 2682 continue; 2683 if (!memcmp(dentry->d_name.name, p->name, p->len)) 2684 break; 2685 } 2686 if (p > last) 2687 goto out; 2688 2689 error = proc_base_instantiate(dir, dentry, task, p); 2690 2691 out: 2692 put_task_struct(task); 2693 out_no_task: 2694 return error; 2695 } 2696 2697 static int proc_base_fill_cache(struct file *filp, void *dirent, 2698 filldir_t filldir, struct task_struct *task, const struct pid_entry *p) 2699 { 2700 return proc_fill_cache(filp, dirent, filldir, p->name, p->len, 2701 proc_base_instantiate, task, p); 2702 } 2703 2704 #ifdef CONFIG_TASK_IO_ACCOUNTING 2705 static int do_io_accounting(struct task_struct *task, char *buffer, int whole) 2706 { 2707 struct task_io_accounting acct = task->ioac; 2708 unsigned long flags; 2709 int result; 2710 2711 result = mutex_lock_killable(&task->signal->cred_guard_mutex); 2712 if (result) 2713 return result; 2714 2715 if (!ptrace_may_access(task, PTRACE_MODE_READ)) { 2716 result = -EACCES; 2717 goto out_unlock; 2718 } 2719 2720 if (whole && lock_task_sighand(task, &flags)) { 2721 struct task_struct *t = task; 2722 2723 task_io_accounting_add(&acct, &task->signal->ioac); 2724 while_each_thread(task, t) 2725 task_io_accounting_add(&acct, &t->ioac); 2726 2727 unlock_task_sighand(task, &flags); 2728 } 2729 result = sprintf(buffer, 2730 "rchar: %llu\n" 2731 "wchar: %llu\n" 2732 "syscr: %llu\n" 2733 "syscw: %llu\n" 2734 "read_bytes: %llu\n" 2735 "write_bytes: %llu\n" 2736 "cancelled_write_bytes: %llu\n", 2737 (unsigned long long)acct.rchar, 2738 (unsigned long long)acct.wchar, 2739 (unsigned long long)acct.syscr, 2740 (unsigned long long)acct.syscw, 2741 (unsigned long long)acct.read_bytes, 2742 (unsigned long long)acct.write_bytes, 2743 (unsigned long long)acct.cancelled_write_bytes); 2744 out_unlock: 2745 mutex_unlock(&task->signal->cred_guard_mutex); 2746 return result; 2747 } 2748 2749 static int proc_tid_io_accounting(struct task_struct *task, char *buffer) 2750 { 2751 return do_io_accounting(task, buffer, 0); 2752 } 2753 2754 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer) 2755 { 2756 return do_io_accounting(task, buffer, 1); 2757 } 2758 #endif /* CONFIG_TASK_IO_ACCOUNTING */ 2759 2760 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns, 2761 struct pid *pid, struct task_struct *task) 2762 { 2763 int err = lock_trace(task); 2764 if (!err) { 2765 seq_printf(m, "%08x\n", task->personality); 2766 unlock_trace(task); 2767 } 2768 return err; 2769 } 2770 2771 /* 2772 * Thread groups 2773 */ 2774 static const struct file_operations proc_task_operations; 2775 static const struct inode_operations proc_task_inode_operations; 2776 2777 static const struct pid_entry tgid_base_stuff[] = { 2778 DIR("task", S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations), 2779 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations), 2780 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations), 2781 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations), 2782 #ifdef CONFIG_NET 2783 DIR("net", S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations), 2784 #endif 2785 REG("environ", S_IRUSR, proc_environ_operations), 2786 INF("auxv", S_IRUSR, proc_pid_auxv), 2787 ONE("status", S_IRUGO, proc_pid_status), 2788 ONE("personality", S_IRUGO, proc_pid_personality), 2789 INF("limits", S_IRUGO, proc_pid_limits), 2790 #ifdef CONFIG_SCHED_DEBUG 2791 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations), 2792 #endif 2793 #ifdef CONFIG_SCHED_AUTOGROUP 2794 REG("autogroup", S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations), 2795 #endif 2796 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations), 2797 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 2798 INF("syscall", S_IRUGO, proc_pid_syscall), 2799 #endif 2800 INF("cmdline", S_IRUGO, proc_pid_cmdline), 2801 ONE("stat", S_IRUGO, proc_tgid_stat), 2802 ONE("statm", S_IRUGO, proc_pid_statm), 2803 REG("maps", S_IRUGO, proc_maps_operations), 2804 #ifdef CONFIG_NUMA 2805 REG("numa_maps", S_IRUGO, proc_numa_maps_operations), 2806 #endif 2807 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations), 2808 LNK("cwd", proc_cwd_link), 2809 LNK("root", proc_root_link), 2810 LNK("exe", proc_exe_link), 2811 REG("mounts", S_IRUGO, proc_mounts_operations), 2812 REG("mountinfo", S_IRUGO, proc_mountinfo_operations), 2813 REG("mountstats", S_IRUSR, proc_mountstats_operations), 2814 #ifdef CONFIG_PROC_PAGE_MONITOR 2815 REG("clear_refs", S_IWUSR, proc_clear_refs_operations), 2816 REG("smaps", S_IRUGO, proc_smaps_operations), 2817 REG("pagemap", S_IRUGO, proc_pagemap_operations), 2818 #endif 2819 #ifdef CONFIG_SECURITY 2820 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), 2821 #endif 2822 #ifdef CONFIG_KALLSYMS 2823 INF("wchan", S_IRUGO, proc_pid_wchan), 2824 #endif 2825 #ifdef CONFIG_STACKTRACE 2826 ONE("stack", S_IRUGO, proc_pid_stack), 2827 #endif 2828 #ifdef CONFIG_SCHEDSTATS 2829 INF("schedstat", S_IRUGO, proc_pid_schedstat), 2830 #endif 2831 #ifdef CONFIG_LATENCYTOP 2832 REG("latency", S_IRUGO, proc_lstats_operations), 2833 #endif 2834 #ifdef CONFIG_PROC_PID_CPUSET 2835 REG("cpuset", S_IRUGO, proc_cpuset_operations), 2836 #endif 2837 #ifdef CONFIG_CGROUPS 2838 REG("cgroup", S_IRUGO, proc_cgroup_operations), 2839 #endif 2840 INF("oom_score", S_IRUGO, proc_oom_score), 2841 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations), 2842 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations), 2843 #ifdef CONFIG_AUDITSYSCALL 2844 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations), 2845 REG("sessionid", S_IRUGO, proc_sessionid_operations), 2846 #endif 2847 #ifdef CONFIG_FAULT_INJECTION 2848 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations), 2849 #endif 2850 #ifdef CONFIG_ELF_CORE 2851 REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations), 2852 #endif 2853 #ifdef CONFIG_TASK_IO_ACCOUNTING 2854 INF("io", S_IRUSR, proc_tgid_io_accounting), 2855 #endif 2856 #ifdef CONFIG_HARDWALL 2857 INF("hardwall", S_IRUGO, proc_pid_hardwall), 2858 #endif 2859 }; 2860 2861 static int proc_tgid_base_readdir(struct file * filp, 2862 void * dirent, filldir_t filldir) 2863 { 2864 return proc_pident_readdir(filp,dirent,filldir, 2865 tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff)); 2866 } 2867 2868 static const struct file_operations proc_tgid_base_operations = { 2869 .read = generic_read_dir, 2870 .readdir = proc_tgid_base_readdir, 2871 .llseek = default_llseek, 2872 }; 2873 2874 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){ 2875 return proc_pident_lookup(dir, dentry, 2876 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 2877 } 2878 2879 static const struct inode_operations proc_tgid_base_inode_operations = { 2880 .lookup = proc_tgid_base_lookup, 2881 .getattr = pid_getattr, 2882 .setattr = proc_setattr, 2883 }; 2884 2885 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid) 2886 { 2887 struct dentry *dentry, *leader, *dir; 2888 char buf[PROC_NUMBUF]; 2889 struct qstr name; 2890 2891 name.name = buf; 2892 name.len = snprintf(buf, sizeof(buf), "%d", pid); 2893 dentry = d_hash_and_lookup(mnt->mnt_root, &name); 2894 if (dentry) { 2895 shrink_dcache_parent(dentry); 2896 d_drop(dentry); 2897 dput(dentry); 2898 } 2899 2900 name.name = buf; 2901 name.len = snprintf(buf, sizeof(buf), "%d", tgid); 2902 leader = d_hash_and_lookup(mnt->mnt_root, &name); 2903 if (!leader) 2904 goto out; 2905 2906 name.name = "task"; 2907 name.len = strlen(name.name); 2908 dir = d_hash_and_lookup(leader, &name); 2909 if (!dir) 2910 goto out_put_leader; 2911 2912 name.name = buf; 2913 name.len = snprintf(buf, sizeof(buf), "%d", pid); 2914 dentry = d_hash_and_lookup(dir, &name); 2915 if (dentry) { 2916 shrink_dcache_parent(dentry); 2917 d_drop(dentry); 2918 dput(dentry); 2919 } 2920 2921 dput(dir); 2922 out_put_leader: 2923 dput(leader); 2924 out: 2925 return; 2926 } 2927 2928 /** 2929 * proc_flush_task - Remove dcache entries for @task from the /proc dcache. 2930 * @task: task that should be flushed. 2931 * 2932 * When flushing dentries from proc, one needs to flush them from global 2933 * proc (proc_mnt) and from all the namespaces' procs this task was seen 2934 * in. This call is supposed to do all of this job. 2935 * 2936 * Looks in the dcache for 2937 * /proc/@pid 2938 * /proc/@tgid/task/@pid 2939 * if either directory is present flushes it and all of it'ts children 2940 * from the dcache. 2941 * 2942 * It is safe and reasonable to cache /proc entries for a task until 2943 * that task exits. After that they just clog up the dcache with 2944 * useless entries, possibly causing useful dcache entries to be 2945 * flushed instead. This routine is proved to flush those useless 2946 * dcache entries at process exit time. 2947 * 2948 * NOTE: This routine is just an optimization so it does not guarantee 2949 * that no dcache entries will exist at process exit time it 2950 * just makes it very unlikely that any will persist. 2951 */ 2952 2953 void proc_flush_task(struct task_struct *task) 2954 { 2955 int i; 2956 struct pid *pid, *tgid; 2957 struct upid *upid; 2958 2959 pid = task_pid(task); 2960 tgid = task_tgid(task); 2961 2962 for (i = 0; i <= pid->level; i++) { 2963 upid = &pid->numbers[i]; 2964 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr, 2965 tgid->numbers[i].nr); 2966 } 2967 2968 upid = &pid->numbers[pid->level]; 2969 if (upid->nr == 1) 2970 pid_ns_release_proc(upid->ns); 2971 } 2972 2973 static struct dentry *proc_pid_instantiate(struct inode *dir, 2974 struct dentry * dentry, 2975 struct task_struct *task, const void *ptr) 2976 { 2977 struct dentry *error = ERR_PTR(-ENOENT); 2978 struct inode *inode; 2979 2980 inode = proc_pid_make_inode(dir->i_sb, task); 2981 if (!inode) 2982 goto out; 2983 2984 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO; 2985 inode->i_op = &proc_tgid_base_inode_operations; 2986 inode->i_fop = &proc_tgid_base_operations; 2987 inode->i_flags|=S_IMMUTABLE; 2988 2989 inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff, 2990 ARRAY_SIZE(tgid_base_stuff)); 2991 2992 d_set_d_op(dentry, &pid_dentry_operations); 2993 2994 d_add(dentry, inode); 2995 /* Close the race of the process dying before we return the dentry */ 2996 if (pid_revalidate(dentry, NULL)) 2997 error = NULL; 2998 out: 2999 return error; 3000 } 3001 3002 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 3003 { 3004 struct dentry *result; 3005 struct task_struct *task; 3006 unsigned tgid; 3007 struct pid_namespace *ns; 3008 3009 result = proc_base_lookup(dir, dentry); 3010 if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT) 3011 goto out; 3012 3013 tgid = name_to_int(dentry); 3014 if (tgid == ~0U) 3015 goto out; 3016 3017 ns = dentry->d_sb->s_fs_info; 3018 rcu_read_lock(); 3019 task = find_task_by_pid_ns(tgid, ns); 3020 if (task) 3021 get_task_struct(task); 3022 rcu_read_unlock(); 3023 if (!task) 3024 goto out; 3025 3026 result = proc_pid_instantiate(dir, dentry, task, NULL); 3027 put_task_struct(task); 3028 out: 3029 return result; 3030 } 3031 3032 /* 3033 * Find the first task with tgid >= tgid 3034 * 3035 */ 3036 struct tgid_iter { 3037 unsigned int tgid; 3038 struct task_struct *task; 3039 }; 3040 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter) 3041 { 3042 struct pid *pid; 3043 3044 if (iter.task) 3045 put_task_struct(iter.task); 3046 rcu_read_lock(); 3047 retry: 3048 iter.task = NULL; 3049 pid = find_ge_pid(iter.tgid, ns); 3050 if (pid) { 3051 iter.tgid = pid_nr_ns(pid, ns); 3052 iter.task = pid_task(pid, PIDTYPE_PID); 3053 /* What we to know is if the pid we have find is the 3054 * pid of a thread_group_leader. Testing for task 3055 * being a thread_group_leader is the obvious thing 3056 * todo but there is a window when it fails, due to 3057 * the pid transfer logic in de_thread. 3058 * 3059 * So we perform the straight forward test of seeing 3060 * if the pid we have found is the pid of a thread 3061 * group leader, and don't worry if the task we have 3062 * found doesn't happen to be a thread group leader. 3063 * As we don't care in the case of readdir. 3064 */ 3065 if (!iter.task || !has_group_leader_pid(iter.task)) { 3066 iter.tgid += 1; 3067 goto retry; 3068 } 3069 get_task_struct(iter.task); 3070 } 3071 rcu_read_unlock(); 3072 return iter; 3073 } 3074 3075 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff)) 3076 3077 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir, 3078 struct tgid_iter iter) 3079 { 3080 char name[PROC_NUMBUF]; 3081 int len = snprintf(name, sizeof(name), "%d", iter.tgid); 3082 return proc_fill_cache(filp, dirent, filldir, name, len, 3083 proc_pid_instantiate, iter.task, NULL); 3084 } 3085 3086 /* for the /proc/ directory itself, after non-process stuff has been done */ 3087 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir) 3088 { 3089 unsigned int nr; 3090 struct task_struct *reaper; 3091 struct tgid_iter iter; 3092 struct pid_namespace *ns; 3093 3094 if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET) 3095 goto out_no_task; 3096 nr = filp->f_pos - FIRST_PROCESS_ENTRY; 3097 3098 reaper = get_proc_task(filp->f_path.dentry->d_inode); 3099 if (!reaper) 3100 goto out_no_task; 3101 3102 for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) { 3103 const struct pid_entry *p = &proc_base_stuff[nr]; 3104 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0) 3105 goto out; 3106 } 3107 3108 ns = filp->f_dentry->d_sb->s_fs_info; 3109 iter.task = NULL; 3110 iter.tgid = filp->f_pos - TGID_OFFSET; 3111 for (iter = next_tgid(ns, iter); 3112 iter.task; 3113 iter.tgid += 1, iter = next_tgid(ns, iter)) { 3114 filp->f_pos = iter.tgid + TGID_OFFSET; 3115 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) { 3116 put_task_struct(iter.task); 3117 goto out; 3118 } 3119 } 3120 filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET; 3121 out: 3122 put_task_struct(reaper); 3123 out_no_task: 3124 return 0; 3125 } 3126 3127 /* 3128 * Tasks 3129 */ 3130 static const struct pid_entry tid_base_stuff[] = { 3131 DIR("fd", S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations), 3132 DIR("fdinfo", S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations), 3133 DIR("ns", S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations), 3134 REG("environ", S_IRUSR, proc_environ_operations), 3135 INF("auxv", S_IRUSR, proc_pid_auxv), 3136 ONE("status", S_IRUGO, proc_pid_status), 3137 ONE("personality", S_IRUGO, proc_pid_personality), 3138 INF("limits", S_IRUGO, proc_pid_limits), 3139 #ifdef CONFIG_SCHED_DEBUG 3140 REG("sched", S_IRUGO|S_IWUSR, proc_pid_sched_operations), 3141 #endif 3142 REG("comm", S_IRUGO|S_IWUSR, proc_pid_set_comm_operations), 3143 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK 3144 INF("syscall", S_IRUGO, proc_pid_syscall), 3145 #endif 3146 INF("cmdline", S_IRUGO, proc_pid_cmdline), 3147 ONE("stat", S_IRUGO, proc_tid_stat), 3148 ONE("statm", S_IRUGO, proc_pid_statm), 3149 REG("maps", S_IRUGO, proc_maps_operations), 3150 #ifdef CONFIG_NUMA 3151 REG("numa_maps", S_IRUGO, proc_numa_maps_operations), 3152 #endif 3153 REG("mem", S_IRUSR|S_IWUSR, proc_mem_operations), 3154 LNK("cwd", proc_cwd_link), 3155 LNK("root", proc_root_link), 3156 LNK("exe", proc_exe_link), 3157 REG("mounts", S_IRUGO, proc_mounts_operations), 3158 REG("mountinfo", S_IRUGO, proc_mountinfo_operations), 3159 #ifdef CONFIG_PROC_PAGE_MONITOR 3160 REG("clear_refs", S_IWUSR, proc_clear_refs_operations), 3161 REG("smaps", S_IRUGO, proc_smaps_operations), 3162 REG("pagemap", S_IRUGO, proc_pagemap_operations), 3163 #endif 3164 #ifdef CONFIG_SECURITY 3165 DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), 3166 #endif 3167 #ifdef CONFIG_KALLSYMS 3168 INF("wchan", S_IRUGO, proc_pid_wchan), 3169 #endif 3170 #ifdef CONFIG_STACKTRACE 3171 ONE("stack", S_IRUGO, proc_pid_stack), 3172 #endif 3173 #ifdef CONFIG_SCHEDSTATS 3174 INF("schedstat", S_IRUGO, proc_pid_schedstat), 3175 #endif 3176 #ifdef CONFIG_LATENCYTOP 3177 REG("latency", S_IRUGO, proc_lstats_operations), 3178 #endif 3179 #ifdef CONFIG_PROC_PID_CPUSET 3180 REG("cpuset", S_IRUGO, proc_cpuset_operations), 3181 #endif 3182 #ifdef CONFIG_CGROUPS 3183 REG("cgroup", S_IRUGO, proc_cgroup_operations), 3184 #endif 3185 INF("oom_score", S_IRUGO, proc_oom_score), 3186 REG("oom_adj", S_IRUGO|S_IWUSR, proc_oom_adjust_operations), 3187 REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations), 3188 #ifdef CONFIG_AUDITSYSCALL 3189 REG("loginuid", S_IWUSR|S_IRUGO, proc_loginuid_operations), 3190 REG("sessionid", S_IRUGO, proc_sessionid_operations), 3191 #endif 3192 #ifdef CONFIG_FAULT_INJECTION 3193 REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations), 3194 #endif 3195 #ifdef CONFIG_TASK_IO_ACCOUNTING 3196 INF("io", S_IRUSR, proc_tid_io_accounting), 3197 #endif 3198 #ifdef CONFIG_HARDWALL 3199 INF("hardwall", S_IRUGO, proc_pid_hardwall), 3200 #endif 3201 }; 3202 3203 static int proc_tid_base_readdir(struct file * filp, 3204 void * dirent, filldir_t filldir) 3205 { 3206 return proc_pident_readdir(filp,dirent,filldir, 3207 tid_base_stuff,ARRAY_SIZE(tid_base_stuff)); 3208 } 3209 3210 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){ 3211 return proc_pident_lookup(dir, dentry, 3212 tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 3213 } 3214 3215 static const struct file_operations proc_tid_base_operations = { 3216 .read = generic_read_dir, 3217 .readdir = proc_tid_base_readdir, 3218 .llseek = default_llseek, 3219 }; 3220 3221 static const struct inode_operations proc_tid_base_inode_operations = { 3222 .lookup = proc_tid_base_lookup, 3223 .getattr = pid_getattr, 3224 .setattr = proc_setattr, 3225 }; 3226 3227 static struct dentry *proc_task_instantiate(struct inode *dir, 3228 struct dentry *dentry, struct task_struct *task, const void *ptr) 3229 { 3230 struct dentry *error = ERR_PTR(-ENOENT); 3231 struct inode *inode; 3232 inode = proc_pid_make_inode(dir->i_sb, task); 3233 3234 if (!inode) 3235 goto out; 3236 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO; 3237 inode->i_op = &proc_tid_base_inode_operations; 3238 inode->i_fop = &proc_tid_base_operations; 3239 inode->i_flags|=S_IMMUTABLE; 3240 3241 inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff, 3242 ARRAY_SIZE(tid_base_stuff)); 3243 3244 d_set_d_op(dentry, &pid_dentry_operations); 3245 3246 d_add(dentry, inode); 3247 /* Close the race of the process dying before we return the dentry */ 3248 if (pid_revalidate(dentry, NULL)) 3249 error = NULL; 3250 out: 3251 return error; 3252 } 3253 3254 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 3255 { 3256 struct dentry *result = ERR_PTR(-ENOENT); 3257 struct task_struct *task; 3258 struct task_struct *leader = get_proc_task(dir); 3259 unsigned tid; 3260 struct pid_namespace *ns; 3261 3262 if (!leader) 3263 goto out_no_task; 3264 3265 tid = name_to_int(dentry); 3266 if (tid == ~0U) 3267 goto out; 3268 3269 ns = dentry->d_sb->s_fs_info; 3270 rcu_read_lock(); 3271 task = find_task_by_pid_ns(tid, ns); 3272 if (task) 3273 get_task_struct(task); 3274 rcu_read_unlock(); 3275 if (!task) 3276 goto out; 3277 if (!same_thread_group(leader, task)) 3278 goto out_drop_task; 3279 3280 result = proc_task_instantiate(dir, dentry, task, NULL); 3281 out_drop_task: 3282 put_task_struct(task); 3283 out: 3284 put_task_struct(leader); 3285 out_no_task: 3286 return result; 3287 } 3288 3289 /* 3290 * Find the first tid of a thread group to return to user space. 3291 * 3292 * Usually this is just the thread group leader, but if the users 3293 * buffer was too small or there was a seek into the middle of the 3294 * directory we have more work todo. 3295 * 3296 * In the case of a short read we start with find_task_by_pid. 3297 * 3298 * In the case of a seek we start with the leader and walk nr 3299 * threads past it. 3300 */ 3301 static struct task_struct *first_tid(struct task_struct *leader, 3302 int tid, int nr, struct pid_namespace *ns) 3303 { 3304 struct task_struct *pos; 3305 3306 rcu_read_lock(); 3307 /* Attempt to start with the pid of a thread */ 3308 if (tid && (nr > 0)) { 3309 pos = find_task_by_pid_ns(tid, ns); 3310 if (pos && (pos->group_leader == leader)) 3311 goto found; 3312 } 3313 3314 /* If nr exceeds the number of threads there is nothing todo */ 3315 pos = NULL; 3316 if (nr && nr >= get_nr_threads(leader)) 3317 goto out; 3318 3319 /* If we haven't found our starting place yet start 3320 * with the leader and walk nr threads forward. 3321 */ 3322 for (pos = leader; nr > 0; --nr) { 3323 pos = next_thread(pos); 3324 if (pos == leader) { 3325 pos = NULL; 3326 goto out; 3327 } 3328 } 3329 found: 3330 get_task_struct(pos); 3331 out: 3332 rcu_read_unlock(); 3333 return pos; 3334 } 3335 3336 /* 3337 * Find the next thread in the thread list. 3338 * Return NULL if there is an error or no next thread. 3339 * 3340 * The reference to the input task_struct is released. 3341 */ 3342 static struct task_struct *next_tid(struct task_struct *start) 3343 { 3344 struct task_struct *pos = NULL; 3345 rcu_read_lock(); 3346 if (pid_alive(start)) { 3347 pos = next_thread(start); 3348 if (thread_group_leader(pos)) 3349 pos = NULL; 3350 else 3351 get_task_struct(pos); 3352 } 3353 rcu_read_unlock(); 3354 put_task_struct(start); 3355 return pos; 3356 } 3357 3358 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir, 3359 struct task_struct *task, int tid) 3360 { 3361 char name[PROC_NUMBUF]; 3362 int len = snprintf(name, sizeof(name), "%d", tid); 3363 return proc_fill_cache(filp, dirent, filldir, name, len, 3364 proc_task_instantiate, task, NULL); 3365 } 3366 3367 /* for the /proc/TGID/task/ directories */ 3368 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir) 3369 { 3370 struct dentry *dentry = filp->f_path.dentry; 3371 struct inode *inode = dentry->d_inode; 3372 struct task_struct *leader = NULL; 3373 struct task_struct *task; 3374 int retval = -ENOENT; 3375 ino_t ino; 3376 int tid; 3377 struct pid_namespace *ns; 3378 3379 task = get_proc_task(inode); 3380 if (!task) 3381 goto out_no_task; 3382 rcu_read_lock(); 3383 if (pid_alive(task)) { 3384 leader = task->group_leader; 3385 get_task_struct(leader); 3386 } 3387 rcu_read_unlock(); 3388 put_task_struct(task); 3389 if (!leader) 3390 goto out_no_task; 3391 retval = 0; 3392 3393 switch ((unsigned long)filp->f_pos) { 3394 case 0: 3395 ino = inode->i_ino; 3396 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0) 3397 goto out; 3398 filp->f_pos++; 3399 /* fall through */ 3400 case 1: 3401 ino = parent_ino(dentry); 3402 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0) 3403 goto out; 3404 filp->f_pos++; 3405 /* fall through */ 3406 } 3407 3408 /* f_version caches the tgid value that the last readdir call couldn't 3409 * return. lseek aka telldir automagically resets f_version to 0. 3410 */ 3411 ns = filp->f_dentry->d_sb->s_fs_info; 3412 tid = (int)filp->f_version; 3413 filp->f_version = 0; 3414 for (task = first_tid(leader, tid, filp->f_pos - 2, ns); 3415 task; 3416 task = next_tid(task), filp->f_pos++) { 3417 tid = task_pid_nr_ns(task, ns); 3418 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) { 3419 /* returning this tgid failed, save it as the first 3420 * pid for the next readir call */ 3421 filp->f_version = (u64)tid; 3422 put_task_struct(task); 3423 break; 3424 } 3425 } 3426 out: 3427 put_task_struct(leader); 3428 out_no_task: 3429 return retval; 3430 } 3431 3432 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 3433 { 3434 struct inode *inode = dentry->d_inode; 3435 struct task_struct *p = get_proc_task(inode); 3436 generic_fillattr(inode, stat); 3437 3438 if (p) { 3439 stat->nlink += get_nr_threads(p); 3440 put_task_struct(p); 3441 } 3442 3443 return 0; 3444 } 3445 3446 static const struct inode_operations proc_task_inode_operations = { 3447 .lookup = proc_task_lookup, 3448 .getattr = proc_task_getattr, 3449 .setattr = proc_setattr, 3450 }; 3451 3452 static const struct file_operations proc_task_operations = { 3453 .read = generic_read_dir, 3454 .readdir = proc_task_readdir, 3455 .llseek = default_llseek, 3456 }; 3457