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/init.h> 57 #include <linux/capability.h> 58 #include <linux/file.h> 59 #include <linux/string.h> 60 #include <linux/seq_file.h> 61 #include <linux/namei.h> 62 #include <linux/mnt_namespace.h> 63 #include <linux/mm.h> 64 #include <linux/rcupdate.h> 65 #include <linux/kallsyms.h> 66 #include <linux/resource.h> 67 #include <linux/module.h> 68 #include <linux/mount.h> 69 #include <linux/security.h> 70 #include <linux/ptrace.h> 71 #include <linux/cgroup.h> 72 #include <linux/cpuset.h> 73 #include <linux/audit.h> 74 #include <linux/poll.h> 75 #include <linux/nsproxy.h> 76 #include <linux/oom.h> 77 #include <linux/elf.h> 78 #include <linux/pid_namespace.h> 79 #include "internal.h" 80 81 /* NOTE: 82 * Implementing inode permission operations in /proc is almost 83 * certainly an error. Permission checks need to happen during 84 * each system call not at open time. The reason is that most of 85 * what we wish to check for permissions in /proc varies at runtime. 86 * 87 * The classic example of a problem is opening file descriptors 88 * in /proc for a task before it execs a suid executable. 89 */ 90 91 92 /* Worst case buffer size needed for holding an integer. */ 93 #define PROC_NUMBUF 13 94 95 struct pid_entry { 96 char *name; 97 int len; 98 mode_t mode; 99 const struct inode_operations *iop; 100 const struct file_operations *fop; 101 union proc_op op; 102 }; 103 104 #define NOD(NAME, MODE, IOP, FOP, OP) { \ 105 .name = (NAME), \ 106 .len = sizeof(NAME) - 1, \ 107 .mode = MODE, \ 108 .iop = IOP, \ 109 .fop = FOP, \ 110 .op = OP, \ 111 } 112 113 #define DIR(NAME, MODE, OTYPE) \ 114 NOD(NAME, (S_IFDIR|(MODE)), \ 115 &proc_##OTYPE##_inode_operations, &proc_##OTYPE##_operations, \ 116 {} ) 117 #define LNK(NAME, OTYPE) \ 118 NOD(NAME, (S_IFLNK|S_IRWXUGO), \ 119 &proc_pid_link_inode_operations, NULL, \ 120 { .proc_get_link = &proc_##OTYPE##_link } ) 121 #define REG(NAME, MODE, OTYPE) \ 122 NOD(NAME, (S_IFREG|(MODE)), NULL, \ 123 &proc_##OTYPE##_operations, {}) 124 #define INF(NAME, MODE, OTYPE) \ 125 NOD(NAME, (S_IFREG|(MODE)), \ 126 NULL, &proc_info_file_operations, \ 127 { .proc_read = &proc_##OTYPE } ) 128 129 int maps_protect; 130 EXPORT_SYMBOL(maps_protect); 131 132 static struct fs_struct *get_fs_struct(struct task_struct *task) 133 { 134 struct fs_struct *fs; 135 task_lock(task); 136 fs = task->fs; 137 if(fs) 138 atomic_inc(&fs->count); 139 task_unlock(task); 140 return fs; 141 } 142 143 static int get_nr_threads(struct task_struct *tsk) 144 { 145 /* Must be called with the rcu_read_lock held */ 146 unsigned long flags; 147 int count = 0; 148 149 if (lock_task_sighand(tsk, &flags)) { 150 count = atomic_read(&tsk->signal->count); 151 unlock_task_sighand(tsk, &flags); 152 } 153 return count; 154 } 155 156 static int proc_cwd_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt) 157 { 158 struct task_struct *task = get_proc_task(inode); 159 struct fs_struct *fs = NULL; 160 int result = -ENOENT; 161 162 if (task) { 163 fs = get_fs_struct(task); 164 put_task_struct(task); 165 } 166 if (fs) { 167 read_lock(&fs->lock); 168 *mnt = mntget(fs->pwdmnt); 169 *dentry = dget(fs->pwd); 170 read_unlock(&fs->lock); 171 result = 0; 172 put_fs_struct(fs); 173 } 174 return result; 175 } 176 177 static int proc_root_link(struct inode *inode, struct dentry **dentry, struct vfsmount **mnt) 178 { 179 struct task_struct *task = get_proc_task(inode); 180 struct fs_struct *fs = NULL; 181 int result = -ENOENT; 182 183 if (task) { 184 fs = get_fs_struct(task); 185 put_task_struct(task); 186 } 187 if (fs) { 188 read_lock(&fs->lock); 189 *mnt = mntget(fs->rootmnt); 190 *dentry = dget(fs->root); 191 read_unlock(&fs->lock); 192 result = 0; 193 put_fs_struct(fs); 194 } 195 return result; 196 } 197 198 #define MAY_PTRACE(task) \ 199 (task == current || \ 200 (task->parent == current && \ 201 (task->ptrace & PT_PTRACED) && \ 202 (task->state == TASK_STOPPED || task->state == TASK_TRACED) && \ 203 security_ptrace(current,task) == 0)) 204 205 struct mm_struct *mm_for_maps(struct task_struct *task) 206 { 207 struct mm_struct *mm = get_task_mm(task); 208 if (!mm) 209 return NULL; 210 down_read(&mm->mmap_sem); 211 task_lock(task); 212 if (task->mm != mm) 213 goto out; 214 if (task->mm != current->mm && __ptrace_may_attach(task) < 0) 215 goto out; 216 task_unlock(task); 217 return mm; 218 out: 219 task_unlock(task); 220 up_read(&mm->mmap_sem); 221 mmput(mm); 222 return NULL; 223 } 224 225 static int proc_pid_cmdline(struct task_struct *task, char * buffer) 226 { 227 int res = 0; 228 unsigned int len; 229 struct mm_struct *mm = get_task_mm(task); 230 if (!mm) 231 goto out; 232 if (!mm->arg_end) 233 goto out_mm; /* Shh! No looking before we're done */ 234 235 len = mm->arg_end - mm->arg_start; 236 237 if (len > PAGE_SIZE) 238 len = PAGE_SIZE; 239 240 res = access_process_vm(task, mm->arg_start, buffer, len, 0); 241 242 // If the nul at the end of args has been overwritten, then 243 // assume application is using setproctitle(3). 244 if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) { 245 len = strnlen(buffer, res); 246 if (len < res) { 247 res = len; 248 } else { 249 len = mm->env_end - mm->env_start; 250 if (len > PAGE_SIZE - res) 251 len = PAGE_SIZE - res; 252 res += access_process_vm(task, mm->env_start, buffer+res, len, 0); 253 res = strnlen(buffer, res); 254 } 255 } 256 out_mm: 257 mmput(mm); 258 out: 259 return res; 260 } 261 262 static int proc_pid_auxv(struct task_struct *task, char *buffer) 263 { 264 int res = 0; 265 struct mm_struct *mm = get_task_mm(task); 266 if (mm) { 267 unsigned int nwords = 0; 268 do 269 nwords += 2; 270 while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */ 271 res = nwords * sizeof(mm->saved_auxv[0]); 272 if (res > PAGE_SIZE) 273 res = PAGE_SIZE; 274 memcpy(buffer, mm->saved_auxv, res); 275 mmput(mm); 276 } 277 return res; 278 } 279 280 281 #ifdef CONFIG_KALLSYMS 282 /* 283 * Provides a wchan file via kallsyms in a proper one-value-per-file format. 284 * Returns the resolved symbol. If that fails, simply return the address. 285 */ 286 static int proc_pid_wchan(struct task_struct *task, char *buffer) 287 { 288 unsigned long wchan; 289 char symname[KSYM_NAME_LEN]; 290 291 wchan = get_wchan(task); 292 293 if (lookup_symbol_name(wchan, symname) < 0) 294 return sprintf(buffer, "%lu", wchan); 295 else 296 return sprintf(buffer, "%s", symname); 297 } 298 #endif /* CONFIG_KALLSYMS */ 299 300 #ifdef CONFIG_SCHEDSTATS 301 /* 302 * Provides /proc/PID/schedstat 303 */ 304 static int proc_pid_schedstat(struct task_struct *task, char *buffer) 305 { 306 return sprintf(buffer, "%llu %llu %lu\n", 307 task->sched_info.cpu_time, 308 task->sched_info.run_delay, 309 task->sched_info.pcount); 310 } 311 #endif 312 313 /* The badness from the OOM killer */ 314 unsigned long badness(struct task_struct *p, unsigned long uptime); 315 static int proc_oom_score(struct task_struct *task, char *buffer) 316 { 317 unsigned long points; 318 struct timespec uptime; 319 320 do_posix_clock_monotonic_gettime(&uptime); 321 read_lock(&tasklist_lock); 322 points = badness(task, uptime.tv_sec); 323 read_unlock(&tasklist_lock); 324 return sprintf(buffer, "%lu\n", points); 325 } 326 327 struct limit_names { 328 char *name; 329 char *unit; 330 }; 331 332 static const struct limit_names lnames[RLIM_NLIMITS] = { 333 [RLIMIT_CPU] = {"Max cpu time", "ms"}, 334 [RLIMIT_FSIZE] = {"Max file size", "bytes"}, 335 [RLIMIT_DATA] = {"Max data size", "bytes"}, 336 [RLIMIT_STACK] = {"Max stack size", "bytes"}, 337 [RLIMIT_CORE] = {"Max core file size", "bytes"}, 338 [RLIMIT_RSS] = {"Max resident set", "bytes"}, 339 [RLIMIT_NPROC] = {"Max processes", "processes"}, 340 [RLIMIT_NOFILE] = {"Max open files", "files"}, 341 [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"}, 342 [RLIMIT_AS] = {"Max address space", "bytes"}, 343 [RLIMIT_LOCKS] = {"Max file locks", "locks"}, 344 [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"}, 345 [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"}, 346 [RLIMIT_NICE] = {"Max nice priority", NULL}, 347 [RLIMIT_RTPRIO] = {"Max realtime priority", NULL}, 348 }; 349 350 /* Display limits for a process */ 351 static int proc_pid_limits(struct task_struct *task, char *buffer) 352 { 353 unsigned int i; 354 int count = 0; 355 unsigned long flags; 356 char *bufptr = buffer; 357 358 struct rlimit rlim[RLIM_NLIMITS]; 359 360 rcu_read_lock(); 361 if (!lock_task_sighand(task,&flags)) { 362 rcu_read_unlock(); 363 return 0; 364 } 365 memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS); 366 unlock_task_sighand(task, &flags); 367 rcu_read_unlock(); 368 369 /* 370 * print the file header 371 */ 372 count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n", 373 "Limit", "Soft Limit", "Hard Limit", "Units"); 374 375 for (i = 0; i < RLIM_NLIMITS; i++) { 376 if (rlim[i].rlim_cur == RLIM_INFINITY) 377 count += sprintf(&bufptr[count], "%-25s %-20s ", 378 lnames[i].name, "unlimited"); 379 else 380 count += sprintf(&bufptr[count], "%-25s %-20lu ", 381 lnames[i].name, rlim[i].rlim_cur); 382 383 if (rlim[i].rlim_max == RLIM_INFINITY) 384 count += sprintf(&bufptr[count], "%-20s ", "unlimited"); 385 else 386 count += sprintf(&bufptr[count], "%-20lu ", 387 rlim[i].rlim_max); 388 389 if (lnames[i].unit) 390 count += sprintf(&bufptr[count], "%-10s\n", 391 lnames[i].unit); 392 else 393 count += sprintf(&bufptr[count], "\n"); 394 } 395 396 return count; 397 } 398 399 /************************************************************************/ 400 /* Here the fs part begins */ 401 /************************************************************************/ 402 403 /* permission checks */ 404 static int proc_fd_access_allowed(struct inode *inode) 405 { 406 struct task_struct *task; 407 int allowed = 0; 408 /* Allow access to a task's file descriptors if it is us or we 409 * may use ptrace attach to the process and find out that 410 * information. 411 */ 412 task = get_proc_task(inode); 413 if (task) { 414 allowed = ptrace_may_attach(task); 415 put_task_struct(task); 416 } 417 return allowed; 418 } 419 420 static int proc_setattr(struct dentry *dentry, struct iattr *attr) 421 { 422 int error; 423 struct inode *inode = dentry->d_inode; 424 425 if (attr->ia_valid & ATTR_MODE) 426 return -EPERM; 427 428 error = inode_change_ok(inode, attr); 429 if (!error) 430 error = inode_setattr(inode, attr); 431 return error; 432 } 433 434 static const struct inode_operations proc_def_inode_operations = { 435 .setattr = proc_setattr, 436 }; 437 438 extern struct seq_operations mounts_op; 439 struct proc_mounts { 440 struct seq_file m; 441 int event; 442 }; 443 444 static int mounts_open(struct inode *inode, struct file *file) 445 { 446 struct task_struct *task = get_proc_task(inode); 447 struct nsproxy *nsp; 448 struct mnt_namespace *ns = NULL; 449 struct proc_mounts *p; 450 int ret = -EINVAL; 451 452 if (task) { 453 rcu_read_lock(); 454 nsp = task_nsproxy(task); 455 if (nsp) { 456 ns = nsp->mnt_ns; 457 if (ns) 458 get_mnt_ns(ns); 459 } 460 rcu_read_unlock(); 461 462 put_task_struct(task); 463 } 464 465 if (ns) { 466 ret = -ENOMEM; 467 p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL); 468 if (p) { 469 file->private_data = &p->m; 470 ret = seq_open(file, &mounts_op); 471 if (!ret) { 472 p->m.private = ns; 473 p->event = ns->event; 474 return 0; 475 } 476 kfree(p); 477 } 478 put_mnt_ns(ns); 479 } 480 return ret; 481 } 482 483 static int mounts_release(struct inode *inode, struct file *file) 484 { 485 struct seq_file *m = file->private_data; 486 struct mnt_namespace *ns = m->private; 487 put_mnt_ns(ns); 488 return seq_release(inode, file); 489 } 490 491 static unsigned mounts_poll(struct file *file, poll_table *wait) 492 { 493 struct proc_mounts *p = file->private_data; 494 struct mnt_namespace *ns = p->m.private; 495 unsigned res = 0; 496 497 poll_wait(file, &ns->poll, wait); 498 499 spin_lock(&vfsmount_lock); 500 if (p->event != ns->event) { 501 p->event = ns->event; 502 res = POLLERR; 503 } 504 spin_unlock(&vfsmount_lock); 505 506 return res; 507 } 508 509 static const struct file_operations proc_mounts_operations = { 510 .open = mounts_open, 511 .read = seq_read, 512 .llseek = seq_lseek, 513 .release = mounts_release, 514 .poll = mounts_poll, 515 }; 516 517 extern struct seq_operations mountstats_op; 518 static int mountstats_open(struct inode *inode, struct file *file) 519 { 520 int ret = seq_open(file, &mountstats_op); 521 522 if (!ret) { 523 struct seq_file *m = file->private_data; 524 struct nsproxy *nsp; 525 struct mnt_namespace *mnt_ns = NULL; 526 struct task_struct *task = get_proc_task(inode); 527 528 if (task) { 529 rcu_read_lock(); 530 nsp = task_nsproxy(task); 531 if (nsp) { 532 mnt_ns = nsp->mnt_ns; 533 if (mnt_ns) 534 get_mnt_ns(mnt_ns); 535 } 536 rcu_read_unlock(); 537 538 put_task_struct(task); 539 } 540 541 if (mnt_ns) 542 m->private = mnt_ns; 543 else { 544 seq_release(inode, file); 545 ret = -EINVAL; 546 } 547 } 548 return ret; 549 } 550 551 static const struct file_operations proc_mountstats_operations = { 552 .open = mountstats_open, 553 .read = seq_read, 554 .llseek = seq_lseek, 555 .release = mounts_release, 556 }; 557 558 #define PROC_BLOCK_SIZE (3*1024) /* 4K page size but our output routines use some slack for overruns */ 559 560 static ssize_t proc_info_read(struct file * file, char __user * buf, 561 size_t count, loff_t *ppos) 562 { 563 struct inode * inode = file->f_path.dentry->d_inode; 564 unsigned long page; 565 ssize_t length; 566 struct task_struct *task = get_proc_task(inode); 567 568 length = -ESRCH; 569 if (!task) 570 goto out_no_task; 571 572 if (count > PROC_BLOCK_SIZE) 573 count = PROC_BLOCK_SIZE; 574 575 length = -ENOMEM; 576 if (!(page = __get_free_page(GFP_TEMPORARY))) 577 goto out; 578 579 length = PROC_I(inode)->op.proc_read(task, (char*)page); 580 581 if (length >= 0) 582 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length); 583 free_page(page); 584 out: 585 put_task_struct(task); 586 out_no_task: 587 return length; 588 } 589 590 static const struct file_operations proc_info_file_operations = { 591 .read = proc_info_read, 592 }; 593 594 static int mem_open(struct inode* inode, struct file* file) 595 { 596 file->private_data = (void*)((long)current->self_exec_id); 597 return 0; 598 } 599 600 static ssize_t mem_read(struct file * file, char __user * buf, 601 size_t count, loff_t *ppos) 602 { 603 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 604 char *page; 605 unsigned long src = *ppos; 606 int ret = -ESRCH; 607 struct mm_struct *mm; 608 609 if (!task) 610 goto out_no_task; 611 612 if (!MAY_PTRACE(task) || !ptrace_may_attach(task)) 613 goto out; 614 615 ret = -ENOMEM; 616 page = (char *)__get_free_page(GFP_TEMPORARY); 617 if (!page) 618 goto out; 619 620 ret = 0; 621 622 mm = get_task_mm(task); 623 if (!mm) 624 goto out_free; 625 626 ret = -EIO; 627 628 if (file->private_data != (void*)((long)current->self_exec_id)) 629 goto out_put; 630 631 ret = 0; 632 633 while (count > 0) { 634 int this_len, retval; 635 636 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count; 637 retval = access_process_vm(task, src, page, this_len, 0); 638 if (!retval || !MAY_PTRACE(task) || !ptrace_may_attach(task)) { 639 if (!ret) 640 ret = -EIO; 641 break; 642 } 643 644 if (copy_to_user(buf, page, retval)) { 645 ret = -EFAULT; 646 break; 647 } 648 649 ret += retval; 650 src += retval; 651 buf += retval; 652 count -= retval; 653 } 654 *ppos = src; 655 656 out_put: 657 mmput(mm); 658 out_free: 659 free_page((unsigned long) page); 660 out: 661 put_task_struct(task); 662 out_no_task: 663 return ret; 664 } 665 666 #define mem_write NULL 667 668 #ifndef mem_write 669 /* This is a security hazard */ 670 static ssize_t mem_write(struct file * file, const char __user *buf, 671 size_t count, loff_t *ppos) 672 { 673 int copied; 674 char *page; 675 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 676 unsigned long dst = *ppos; 677 678 copied = -ESRCH; 679 if (!task) 680 goto out_no_task; 681 682 if (!MAY_PTRACE(task) || !ptrace_may_attach(task)) 683 goto out; 684 685 copied = -ENOMEM; 686 page = (char *)__get_free_page(GFP_TEMPORARY); 687 if (!page) 688 goto out; 689 690 copied = 0; 691 while (count > 0) { 692 int this_len, retval; 693 694 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count; 695 if (copy_from_user(page, buf, this_len)) { 696 copied = -EFAULT; 697 break; 698 } 699 retval = access_process_vm(task, dst, page, this_len, 1); 700 if (!retval) { 701 if (!copied) 702 copied = -EIO; 703 break; 704 } 705 copied += retval; 706 buf += retval; 707 dst += retval; 708 count -= retval; 709 } 710 *ppos = dst; 711 free_page((unsigned long) page); 712 out: 713 put_task_struct(task); 714 out_no_task: 715 return copied; 716 } 717 #endif 718 719 static loff_t mem_lseek(struct file * file, loff_t offset, int orig) 720 { 721 switch (orig) { 722 case 0: 723 file->f_pos = offset; 724 break; 725 case 1: 726 file->f_pos += offset; 727 break; 728 default: 729 return -EINVAL; 730 } 731 force_successful_syscall_return(); 732 return file->f_pos; 733 } 734 735 static const struct file_operations proc_mem_operations = { 736 .llseek = mem_lseek, 737 .read = mem_read, 738 .write = mem_write, 739 .open = mem_open, 740 }; 741 742 static ssize_t environ_read(struct file *file, char __user *buf, 743 size_t count, loff_t *ppos) 744 { 745 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 746 char *page; 747 unsigned long src = *ppos; 748 int ret = -ESRCH; 749 struct mm_struct *mm; 750 751 if (!task) 752 goto out_no_task; 753 754 if (!ptrace_may_attach(task)) 755 goto out; 756 757 ret = -ENOMEM; 758 page = (char *)__get_free_page(GFP_TEMPORARY); 759 if (!page) 760 goto out; 761 762 ret = 0; 763 764 mm = get_task_mm(task); 765 if (!mm) 766 goto out_free; 767 768 while (count > 0) { 769 int this_len, retval, max_len; 770 771 this_len = mm->env_end - (mm->env_start + src); 772 773 if (this_len <= 0) 774 break; 775 776 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count; 777 this_len = (this_len > max_len) ? max_len : this_len; 778 779 retval = access_process_vm(task, (mm->env_start + src), 780 page, this_len, 0); 781 782 if (retval <= 0) { 783 ret = retval; 784 break; 785 } 786 787 if (copy_to_user(buf, page, retval)) { 788 ret = -EFAULT; 789 break; 790 } 791 792 ret += retval; 793 src += retval; 794 buf += retval; 795 count -= retval; 796 } 797 *ppos = src; 798 799 mmput(mm); 800 out_free: 801 free_page((unsigned long) page); 802 out: 803 put_task_struct(task); 804 out_no_task: 805 return ret; 806 } 807 808 static const struct file_operations proc_environ_operations = { 809 .read = environ_read, 810 }; 811 812 static ssize_t oom_adjust_read(struct file *file, char __user *buf, 813 size_t count, loff_t *ppos) 814 { 815 struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode); 816 char buffer[PROC_NUMBUF]; 817 size_t len; 818 int oom_adjust; 819 820 if (!task) 821 return -ESRCH; 822 oom_adjust = task->oomkilladj; 823 put_task_struct(task); 824 825 len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust); 826 827 return simple_read_from_buffer(buf, count, ppos, buffer, len); 828 } 829 830 static ssize_t oom_adjust_write(struct file *file, const char __user *buf, 831 size_t count, loff_t *ppos) 832 { 833 struct task_struct *task; 834 char buffer[PROC_NUMBUF], *end; 835 int oom_adjust; 836 837 memset(buffer, 0, sizeof(buffer)); 838 if (count > sizeof(buffer) - 1) 839 count = sizeof(buffer) - 1; 840 if (copy_from_user(buffer, buf, count)) 841 return -EFAULT; 842 oom_adjust = simple_strtol(buffer, &end, 0); 843 if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) && 844 oom_adjust != OOM_DISABLE) 845 return -EINVAL; 846 if (*end == '\n') 847 end++; 848 task = get_proc_task(file->f_path.dentry->d_inode); 849 if (!task) 850 return -ESRCH; 851 if (oom_adjust < task->oomkilladj && !capable(CAP_SYS_RESOURCE)) { 852 put_task_struct(task); 853 return -EACCES; 854 } 855 task->oomkilladj = oom_adjust; 856 put_task_struct(task); 857 if (end - buffer == 0) 858 return -EIO; 859 return end - buffer; 860 } 861 862 static const struct file_operations proc_oom_adjust_operations = { 863 .read = oom_adjust_read, 864 .write = oom_adjust_write, 865 }; 866 867 #ifdef CONFIG_MMU 868 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 869 size_t count, loff_t *ppos) 870 { 871 struct task_struct *task; 872 char buffer[PROC_NUMBUF], *end; 873 struct mm_struct *mm; 874 875 memset(buffer, 0, sizeof(buffer)); 876 if (count > sizeof(buffer) - 1) 877 count = sizeof(buffer) - 1; 878 if (copy_from_user(buffer, buf, count)) 879 return -EFAULT; 880 if (!simple_strtol(buffer, &end, 0)) 881 return -EINVAL; 882 if (*end == '\n') 883 end++; 884 task = get_proc_task(file->f_path.dentry->d_inode); 885 if (!task) 886 return -ESRCH; 887 mm = get_task_mm(task); 888 if (mm) { 889 clear_refs_smap(mm); 890 mmput(mm); 891 } 892 put_task_struct(task); 893 if (end - buffer == 0) 894 return -EIO; 895 return end - buffer; 896 } 897 898 static struct file_operations proc_clear_refs_operations = { 899 .write = clear_refs_write, 900 }; 901 #endif 902 903 #ifdef CONFIG_AUDITSYSCALL 904 #define TMPBUFLEN 21 905 static ssize_t proc_loginuid_read(struct file * file, char __user * buf, 906 size_t count, loff_t *ppos) 907 { 908 struct inode * inode = file->f_path.dentry->d_inode; 909 struct task_struct *task = get_proc_task(inode); 910 ssize_t length; 911 char tmpbuf[TMPBUFLEN]; 912 913 if (!task) 914 return -ESRCH; 915 length = scnprintf(tmpbuf, TMPBUFLEN, "%u", 916 audit_get_loginuid(task->audit_context)); 917 put_task_struct(task); 918 return simple_read_from_buffer(buf, count, ppos, tmpbuf, length); 919 } 920 921 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf, 922 size_t count, loff_t *ppos) 923 { 924 struct inode * inode = file->f_path.dentry->d_inode; 925 char *page, *tmp; 926 ssize_t length; 927 uid_t loginuid; 928 929 if (!capable(CAP_AUDIT_CONTROL)) 930 return -EPERM; 931 932 if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) 933 return -EPERM; 934 935 if (count >= PAGE_SIZE) 936 count = PAGE_SIZE - 1; 937 938 if (*ppos != 0) { 939 /* No partial writes. */ 940 return -EINVAL; 941 } 942 page = (char*)__get_free_page(GFP_TEMPORARY); 943 if (!page) 944 return -ENOMEM; 945 length = -EFAULT; 946 if (copy_from_user(page, buf, count)) 947 goto out_free_page; 948 949 page[count] = '\0'; 950 loginuid = simple_strtoul(page, &tmp, 10); 951 if (tmp == page) { 952 length = -EINVAL; 953 goto out_free_page; 954 955 } 956 length = audit_set_loginuid(current, loginuid); 957 if (likely(length == 0)) 958 length = count; 959 960 out_free_page: 961 free_page((unsigned long) page); 962 return length; 963 } 964 965 static const struct file_operations proc_loginuid_operations = { 966 .read = proc_loginuid_read, 967 .write = proc_loginuid_write, 968 }; 969 #endif 970 971 #ifdef CONFIG_FAULT_INJECTION 972 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf, 973 size_t count, loff_t *ppos) 974 { 975 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 976 char buffer[PROC_NUMBUF]; 977 size_t len; 978 int make_it_fail; 979 980 if (!task) 981 return -ESRCH; 982 make_it_fail = task->make_it_fail; 983 put_task_struct(task); 984 985 len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail); 986 987 return simple_read_from_buffer(buf, count, ppos, buffer, len); 988 } 989 990 static ssize_t proc_fault_inject_write(struct file * file, 991 const char __user * buf, size_t count, loff_t *ppos) 992 { 993 struct task_struct *task; 994 char buffer[PROC_NUMBUF], *end; 995 int make_it_fail; 996 997 if (!capable(CAP_SYS_RESOURCE)) 998 return -EPERM; 999 memset(buffer, 0, sizeof(buffer)); 1000 if (count > sizeof(buffer) - 1) 1001 count = sizeof(buffer) - 1; 1002 if (copy_from_user(buffer, buf, count)) 1003 return -EFAULT; 1004 make_it_fail = simple_strtol(buffer, &end, 0); 1005 if (*end == '\n') 1006 end++; 1007 task = get_proc_task(file->f_dentry->d_inode); 1008 if (!task) 1009 return -ESRCH; 1010 task->make_it_fail = make_it_fail; 1011 put_task_struct(task); 1012 if (end - buffer == 0) 1013 return -EIO; 1014 return end - buffer; 1015 } 1016 1017 static const struct file_operations proc_fault_inject_operations = { 1018 .read = proc_fault_inject_read, 1019 .write = proc_fault_inject_write, 1020 }; 1021 #endif 1022 1023 #ifdef CONFIG_SCHED_DEBUG 1024 /* 1025 * Print out various scheduling related per-task fields: 1026 */ 1027 static int sched_show(struct seq_file *m, void *v) 1028 { 1029 struct inode *inode = m->private; 1030 struct task_struct *p; 1031 1032 WARN_ON(!inode); 1033 1034 p = get_proc_task(inode); 1035 if (!p) 1036 return -ESRCH; 1037 proc_sched_show_task(p, m); 1038 1039 put_task_struct(p); 1040 1041 return 0; 1042 } 1043 1044 static ssize_t 1045 sched_write(struct file *file, const char __user *buf, 1046 size_t count, loff_t *offset) 1047 { 1048 struct inode *inode = file->f_path.dentry->d_inode; 1049 struct task_struct *p; 1050 1051 WARN_ON(!inode); 1052 1053 p = get_proc_task(inode); 1054 if (!p) 1055 return -ESRCH; 1056 proc_sched_set_task(p); 1057 1058 put_task_struct(p); 1059 1060 return count; 1061 } 1062 1063 static int sched_open(struct inode *inode, struct file *filp) 1064 { 1065 int ret; 1066 1067 ret = single_open(filp, sched_show, NULL); 1068 if (!ret) { 1069 struct seq_file *m = filp->private_data; 1070 1071 m->private = inode; 1072 } 1073 return ret; 1074 } 1075 1076 static const struct file_operations proc_pid_sched_operations = { 1077 .open = sched_open, 1078 .read = seq_read, 1079 .write = sched_write, 1080 .llseek = seq_lseek, 1081 .release = single_release, 1082 }; 1083 1084 #endif 1085 1086 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd) 1087 { 1088 struct inode *inode = dentry->d_inode; 1089 int error = -EACCES; 1090 1091 /* We don't need a base pointer in the /proc filesystem */ 1092 path_release(nd); 1093 1094 /* Are we allowed to snoop on the tasks file descriptors? */ 1095 if (!proc_fd_access_allowed(inode)) 1096 goto out; 1097 1098 error = PROC_I(inode)->op.proc_get_link(inode, &nd->dentry, &nd->mnt); 1099 nd->last_type = LAST_BIND; 1100 out: 1101 return ERR_PTR(error); 1102 } 1103 1104 static int do_proc_readlink(struct dentry *dentry, struct vfsmount *mnt, 1105 char __user *buffer, int buflen) 1106 { 1107 struct inode * inode; 1108 char *tmp = (char*)__get_free_page(GFP_TEMPORARY); 1109 char *path; 1110 int len; 1111 1112 if (!tmp) 1113 return -ENOMEM; 1114 1115 inode = dentry->d_inode; 1116 path = d_path(dentry, mnt, tmp, PAGE_SIZE); 1117 len = PTR_ERR(path); 1118 if (IS_ERR(path)) 1119 goto out; 1120 len = tmp + PAGE_SIZE - 1 - path; 1121 1122 if (len > buflen) 1123 len = buflen; 1124 if (copy_to_user(buffer, path, len)) 1125 len = -EFAULT; 1126 out: 1127 free_page((unsigned long)tmp); 1128 return len; 1129 } 1130 1131 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen) 1132 { 1133 int error = -EACCES; 1134 struct inode *inode = dentry->d_inode; 1135 struct dentry *de; 1136 struct vfsmount *mnt = NULL; 1137 1138 /* Are we allowed to snoop on the tasks file descriptors? */ 1139 if (!proc_fd_access_allowed(inode)) 1140 goto out; 1141 1142 error = PROC_I(inode)->op.proc_get_link(inode, &de, &mnt); 1143 if (error) 1144 goto out; 1145 1146 error = do_proc_readlink(de, mnt, buffer, buflen); 1147 dput(de); 1148 mntput(mnt); 1149 out: 1150 return error; 1151 } 1152 1153 static const struct inode_operations proc_pid_link_inode_operations = { 1154 .readlink = proc_pid_readlink, 1155 .follow_link = proc_pid_follow_link, 1156 .setattr = proc_setattr, 1157 }; 1158 1159 1160 /* building an inode */ 1161 1162 static int task_dumpable(struct task_struct *task) 1163 { 1164 int dumpable = 0; 1165 struct mm_struct *mm; 1166 1167 task_lock(task); 1168 mm = task->mm; 1169 if (mm) 1170 dumpable = get_dumpable(mm); 1171 task_unlock(task); 1172 if(dumpable == 1) 1173 return 1; 1174 return 0; 1175 } 1176 1177 1178 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task) 1179 { 1180 struct inode * inode; 1181 struct proc_inode *ei; 1182 1183 /* We need a new inode */ 1184 1185 inode = new_inode(sb); 1186 if (!inode) 1187 goto out; 1188 1189 /* Common stuff */ 1190 ei = PROC_I(inode); 1191 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 1192 inode->i_op = &proc_def_inode_operations; 1193 1194 /* 1195 * grab the reference to task. 1196 */ 1197 ei->pid = get_task_pid(task, PIDTYPE_PID); 1198 if (!ei->pid) 1199 goto out_unlock; 1200 1201 inode->i_uid = 0; 1202 inode->i_gid = 0; 1203 if (task_dumpable(task)) { 1204 inode->i_uid = task->euid; 1205 inode->i_gid = task->egid; 1206 } 1207 security_task_to_inode(task, inode); 1208 1209 out: 1210 return inode; 1211 1212 out_unlock: 1213 iput(inode); 1214 return NULL; 1215 } 1216 1217 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 1218 { 1219 struct inode *inode = dentry->d_inode; 1220 struct task_struct *task; 1221 generic_fillattr(inode, stat); 1222 1223 rcu_read_lock(); 1224 stat->uid = 0; 1225 stat->gid = 0; 1226 task = pid_task(proc_pid(inode), PIDTYPE_PID); 1227 if (task) { 1228 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) || 1229 task_dumpable(task)) { 1230 stat->uid = task->euid; 1231 stat->gid = task->egid; 1232 } 1233 } 1234 rcu_read_unlock(); 1235 return 0; 1236 } 1237 1238 /* dentry stuff */ 1239 1240 /* 1241 * Exceptional case: normally we are not allowed to unhash a busy 1242 * directory. In this case, however, we can do it - no aliasing problems 1243 * due to the way we treat inodes. 1244 * 1245 * Rewrite the inode's ownerships here because the owning task may have 1246 * performed a setuid(), etc. 1247 * 1248 * Before the /proc/pid/status file was created the only way to read 1249 * the effective uid of a /process was to stat /proc/pid. Reading 1250 * /proc/pid/status is slow enough that procps and other packages 1251 * kept stating /proc/pid. To keep the rules in /proc simple I have 1252 * made this apply to all per process world readable and executable 1253 * directories. 1254 */ 1255 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd) 1256 { 1257 struct inode *inode = dentry->d_inode; 1258 struct task_struct *task = get_proc_task(inode); 1259 if (task) { 1260 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) || 1261 task_dumpable(task)) { 1262 inode->i_uid = task->euid; 1263 inode->i_gid = task->egid; 1264 } else { 1265 inode->i_uid = 0; 1266 inode->i_gid = 0; 1267 } 1268 inode->i_mode &= ~(S_ISUID | S_ISGID); 1269 security_task_to_inode(task, inode); 1270 put_task_struct(task); 1271 return 1; 1272 } 1273 d_drop(dentry); 1274 return 0; 1275 } 1276 1277 static int pid_delete_dentry(struct dentry * dentry) 1278 { 1279 /* Is the task we represent dead? 1280 * If so, then don't put the dentry on the lru list, 1281 * kill it immediately. 1282 */ 1283 return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first; 1284 } 1285 1286 static struct dentry_operations pid_dentry_operations = 1287 { 1288 .d_revalidate = pid_revalidate, 1289 .d_delete = pid_delete_dentry, 1290 }; 1291 1292 /* Lookups */ 1293 1294 typedef struct dentry *instantiate_t(struct inode *, struct dentry *, 1295 struct task_struct *, const void *); 1296 1297 /* 1298 * Fill a directory entry. 1299 * 1300 * If possible create the dcache entry and derive our inode number and 1301 * file type from dcache entry. 1302 * 1303 * Since all of the proc inode numbers are dynamically generated, the inode 1304 * numbers do not exist until the inode is cache. This means creating the 1305 * the dcache entry in readdir is necessary to keep the inode numbers 1306 * reported by readdir in sync with the inode numbers reported 1307 * by stat. 1308 */ 1309 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir, 1310 char *name, int len, 1311 instantiate_t instantiate, struct task_struct *task, const void *ptr) 1312 { 1313 struct dentry *child, *dir = filp->f_path.dentry; 1314 struct inode *inode; 1315 struct qstr qname; 1316 ino_t ino = 0; 1317 unsigned type = DT_UNKNOWN; 1318 1319 qname.name = name; 1320 qname.len = len; 1321 qname.hash = full_name_hash(name, len); 1322 1323 child = d_lookup(dir, &qname); 1324 if (!child) { 1325 struct dentry *new; 1326 new = d_alloc(dir, &qname); 1327 if (new) { 1328 child = instantiate(dir->d_inode, new, task, ptr); 1329 if (child) 1330 dput(new); 1331 else 1332 child = new; 1333 } 1334 } 1335 if (!child || IS_ERR(child) || !child->d_inode) 1336 goto end_instantiate; 1337 inode = child->d_inode; 1338 if (inode) { 1339 ino = inode->i_ino; 1340 type = inode->i_mode >> 12; 1341 } 1342 dput(child); 1343 end_instantiate: 1344 if (!ino) 1345 ino = find_inode_number(dir, &qname); 1346 if (!ino) 1347 ino = 1; 1348 return filldir(dirent, name, len, filp->f_pos, ino, type); 1349 } 1350 1351 static unsigned name_to_int(struct dentry *dentry) 1352 { 1353 const char *name = dentry->d_name.name; 1354 int len = dentry->d_name.len; 1355 unsigned n = 0; 1356 1357 if (len > 1 && *name == '0') 1358 goto out; 1359 while (len-- > 0) { 1360 unsigned c = *name++ - '0'; 1361 if (c > 9) 1362 goto out; 1363 if (n >= (~0U-9)/10) 1364 goto out; 1365 n *= 10; 1366 n += c; 1367 } 1368 return n; 1369 out: 1370 return ~0U; 1371 } 1372 1373 #define PROC_FDINFO_MAX 64 1374 1375 static int proc_fd_info(struct inode *inode, struct dentry **dentry, 1376 struct vfsmount **mnt, char *info) 1377 { 1378 struct task_struct *task = get_proc_task(inode); 1379 struct files_struct *files = NULL; 1380 struct file *file; 1381 int fd = proc_fd(inode); 1382 1383 if (task) { 1384 files = get_files_struct(task); 1385 put_task_struct(task); 1386 } 1387 if (files) { 1388 /* 1389 * We are not taking a ref to the file structure, so we must 1390 * hold ->file_lock. 1391 */ 1392 spin_lock(&files->file_lock); 1393 file = fcheck_files(files, fd); 1394 if (file) { 1395 if (mnt) 1396 *mnt = mntget(file->f_path.mnt); 1397 if (dentry) 1398 *dentry = dget(file->f_path.dentry); 1399 if (info) 1400 snprintf(info, PROC_FDINFO_MAX, 1401 "pos:\t%lli\n" 1402 "flags:\t0%o\n", 1403 (long long) file->f_pos, 1404 file->f_flags); 1405 spin_unlock(&files->file_lock); 1406 put_files_struct(files); 1407 return 0; 1408 } 1409 spin_unlock(&files->file_lock); 1410 put_files_struct(files); 1411 } 1412 return -ENOENT; 1413 } 1414 1415 static int proc_fd_link(struct inode *inode, struct dentry **dentry, 1416 struct vfsmount **mnt) 1417 { 1418 return proc_fd_info(inode, dentry, mnt, NULL); 1419 } 1420 1421 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd) 1422 { 1423 struct inode *inode = dentry->d_inode; 1424 struct task_struct *task = get_proc_task(inode); 1425 int fd = proc_fd(inode); 1426 struct files_struct *files; 1427 1428 if (task) { 1429 files = get_files_struct(task); 1430 if (files) { 1431 rcu_read_lock(); 1432 if (fcheck_files(files, fd)) { 1433 rcu_read_unlock(); 1434 put_files_struct(files); 1435 if (task_dumpable(task)) { 1436 inode->i_uid = task->euid; 1437 inode->i_gid = task->egid; 1438 } else { 1439 inode->i_uid = 0; 1440 inode->i_gid = 0; 1441 } 1442 inode->i_mode &= ~(S_ISUID | S_ISGID); 1443 security_task_to_inode(task, inode); 1444 put_task_struct(task); 1445 return 1; 1446 } 1447 rcu_read_unlock(); 1448 put_files_struct(files); 1449 } 1450 put_task_struct(task); 1451 } 1452 d_drop(dentry); 1453 return 0; 1454 } 1455 1456 static struct dentry_operations tid_fd_dentry_operations = 1457 { 1458 .d_revalidate = tid_fd_revalidate, 1459 .d_delete = pid_delete_dentry, 1460 }; 1461 1462 static struct dentry *proc_fd_instantiate(struct inode *dir, 1463 struct dentry *dentry, struct task_struct *task, const void *ptr) 1464 { 1465 unsigned fd = *(const unsigned *)ptr; 1466 struct file *file; 1467 struct files_struct *files; 1468 struct inode *inode; 1469 struct proc_inode *ei; 1470 struct dentry *error = ERR_PTR(-ENOENT); 1471 1472 inode = proc_pid_make_inode(dir->i_sb, task); 1473 if (!inode) 1474 goto out; 1475 ei = PROC_I(inode); 1476 ei->fd = fd; 1477 files = get_files_struct(task); 1478 if (!files) 1479 goto out_iput; 1480 inode->i_mode = S_IFLNK; 1481 1482 /* 1483 * We are not taking a ref to the file structure, so we must 1484 * hold ->file_lock. 1485 */ 1486 spin_lock(&files->file_lock); 1487 file = fcheck_files(files, fd); 1488 if (!file) 1489 goto out_unlock; 1490 if (file->f_mode & 1) 1491 inode->i_mode |= S_IRUSR | S_IXUSR; 1492 if (file->f_mode & 2) 1493 inode->i_mode |= S_IWUSR | S_IXUSR; 1494 spin_unlock(&files->file_lock); 1495 put_files_struct(files); 1496 1497 inode->i_op = &proc_pid_link_inode_operations; 1498 inode->i_size = 64; 1499 ei->op.proc_get_link = proc_fd_link; 1500 dentry->d_op = &tid_fd_dentry_operations; 1501 d_add(dentry, inode); 1502 /* Close the race of the process dying before we return the dentry */ 1503 if (tid_fd_revalidate(dentry, NULL)) 1504 error = NULL; 1505 1506 out: 1507 return error; 1508 out_unlock: 1509 spin_unlock(&files->file_lock); 1510 put_files_struct(files); 1511 out_iput: 1512 iput(inode); 1513 goto out; 1514 } 1515 1516 static struct dentry *proc_lookupfd_common(struct inode *dir, 1517 struct dentry *dentry, 1518 instantiate_t instantiate) 1519 { 1520 struct task_struct *task = get_proc_task(dir); 1521 unsigned fd = name_to_int(dentry); 1522 struct dentry *result = ERR_PTR(-ENOENT); 1523 1524 if (!task) 1525 goto out_no_task; 1526 if (fd == ~0U) 1527 goto out; 1528 1529 result = instantiate(dir, dentry, task, &fd); 1530 out: 1531 put_task_struct(task); 1532 out_no_task: 1533 return result; 1534 } 1535 1536 static int proc_readfd_common(struct file * filp, void * dirent, 1537 filldir_t filldir, instantiate_t instantiate) 1538 { 1539 struct dentry *dentry = filp->f_path.dentry; 1540 struct inode *inode = dentry->d_inode; 1541 struct task_struct *p = get_proc_task(inode); 1542 unsigned int fd, ino; 1543 int retval; 1544 struct files_struct * files; 1545 struct fdtable *fdt; 1546 1547 retval = -ENOENT; 1548 if (!p) 1549 goto out_no_task; 1550 retval = 0; 1551 1552 fd = filp->f_pos; 1553 switch (fd) { 1554 case 0: 1555 if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0) 1556 goto out; 1557 filp->f_pos++; 1558 case 1: 1559 ino = parent_ino(dentry); 1560 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0) 1561 goto out; 1562 filp->f_pos++; 1563 default: 1564 files = get_files_struct(p); 1565 if (!files) 1566 goto out; 1567 rcu_read_lock(); 1568 fdt = files_fdtable(files); 1569 for (fd = filp->f_pos-2; 1570 fd < fdt->max_fds; 1571 fd++, filp->f_pos++) { 1572 char name[PROC_NUMBUF]; 1573 int len; 1574 1575 if (!fcheck_files(files, fd)) 1576 continue; 1577 rcu_read_unlock(); 1578 1579 len = snprintf(name, sizeof(name), "%d", fd); 1580 if (proc_fill_cache(filp, dirent, filldir, 1581 name, len, instantiate, 1582 p, &fd) < 0) { 1583 rcu_read_lock(); 1584 break; 1585 } 1586 rcu_read_lock(); 1587 } 1588 rcu_read_unlock(); 1589 put_files_struct(files); 1590 } 1591 out: 1592 put_task_struct(p); 1593 out_no_task: 1594 return retval; 1595 } 1596 1597 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry, 1598 struct nameidata *nd) 1599 { 1600 return proc_lookupfd_common(dir, dentry, proc_fd_instantiate); 1601 } 1602 1603 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir) 1604 { 1605 return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate); 1606 } 1607 1608 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf, 1609 size_t len, loff_t *ppos) 1610 { 1611 char tmp[PROC_FDINFO_MAX]; 1612 int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, NULL, tmp); 1613 if (!err) 1614 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp)); 1615 return err; 1616 } 1617 1618 static const struct file_operations proc_fdinfo_file_operations = { 1619 .open = nonseekable_open, 1620 .read = proc_fdinfo_read, 1621 }; 1622 1623 static const struct file_operations proc_fd_operations = { 1624 .read = generic_read_dir, 1625 .readdir = proc_readfd, 1626 }; 1627 1628 /* 1629 * /proc/pid/fd needs a special permission handler so that a process can still 1630 * access /proc/self/fd after it has executed a setuid(). 1631 */ 1632 static int proc_fd_permission(struct inode *inode, int mask, 1633 struct nameidata *nd) 1634 { 1635 int rv; 1636 1637 rv = generic_permission(inode, mask, NULL); 1638 if (rv == 0) 1639 return 0; 1640 if (task_pid(current) == proc_pid(inode)) 1641 rv = 0; 1642 return rv; 1643 } 1644 1645 /* 1646 * proc directories can do almost nothing.. 1647 */ 1648 static const struct inode_operations proc_fd_inode_operations = { 1649 .lookup = proc_lookupfd, 1650 .permission = proc_fd_permission, 1651 .setattr = proc_setattr, 1652 }; 1653 1654 static struct dentry *proc_fdinfo_instantiate(struct inode *dir, 1655 struct dentry *dentry, struct task_struct *task, const void *ptr) 1656 { 1657 unsigned fd = *(unsigned *)ptr; 1658 struct inode *inode; 1659 struct proc_inode *ei; 1660 struct dentry *error = ERR_PTR(-ENOENT); 1661 1662 inode = proc_pid_make_inode(dir->i_sb, task); 1663 if (!inode) 1664 goto out; 1665 ei = PROC_I(inode); 1666 ei->fd = fd; 1667 inode->i_mode = S_IFREG | S_IRUSR; 1668 inode->i_fop = &proc_fdinfo_file_operations; 1669 dentry->d_op = &tid_fd_dentry_operations; 1670 d_add(dentry, inode); 1671 /* Close the race of the process dying before we return the dentry */ 1672 if (tid_fd_revalidate(dentry, NULL)) 1673 error = NULL; 1674 1675 out: 1676 return error; 1677 } 1678 1679 static struct dentry *proc_lookupfdinfo(struct inode *dir, 1680 struct dentry *dentry, 1681 struct nameidata *nd) 1682 { 1683 return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate); 1684 } 1685 1686 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir) 1687 { 1688 return proc_readfd_common(filp, dirent, filldir, 1689 proc_fdinfo_instantiate); 1690 } 1691 1692 static const struct file_operations proc_fdinfo_operations = { 1693 .read = generic_read_dir, 1694 .readdir = proc_readfdinfo, 1695 }; 1696 1697 /* 1698 * proc directories can do almost nothing.. 1699 */ 1700 static const struct inode_operations proc_fdinfo_inode_operations = { 1701 .lookup = proc_lookupfdinfo, 1702 .setattr = proc_setattr, 1703 }; 1704 1705 1706 static struct dentry *proc_pident_instantiate(struct inode *dir, 1707 struct dentry *dentry, struct task_struct *task, const void *ptr) 1708 { 1709 const struct pid_entry *p = ptr; 1710 struct inode *inode; 1711 struct proc_inode *ei; 1712 struct dentry *error = ERR_PTR(-EINVAL); 1713 1714 inode = proc_pid_make_inode(dir->i_sb, task); 1715 if (!inode) 1716 goto out; 1717 1718 ei = PROC_I(inode); 1719 inode->i_mode = p->mode; 1720 if (S_ISDIR(inode->i_mode)) 1721 inode->i_nlink = 2; /* Use getattr to fix if necessary */ 1722 if (p->iop) 1723 inode->i_op = p->iop; 1724 if (p->fop) 1725 inode->i_fop = p->fop; 1726 ei->op = p->op; 1727 dentry->d_op = &pid_dentry_operations; 1728 d_add(dentry, inode); 1729 /* Close the race of the process dying before we return the dentry */ 1730 if (pid_revalidate(dentry, NULL)) 1731 error = NULL; 1732 out: 1733 return error; 1734 } 1735 1736 static struct dentry *proc_pident_lookup(struct inode *dir, 1737 struct dentry *dentry, 1738 const struct pid_entry *ents, 1739 unsigned int nents) 1740 { 1741 struct inode *inode; 1742 struct dentry *error; 1743 struct task_struct *task = get_proc_task(dir); 1744 const struct pid_entry *p, *last; 1745 1746 error = ERR_PTR(-ENOENT); 1747 inode = NULL; 1748 1749 if (!task) 1750 goto out_no_task; 1751 1752 /* 1753 * Yes, it does not scale. And it should not. Don't add 1754 * new entries into /proc/<tgid>/ without very good reasons. 1755 */ 1756 last = &ents[nents - 1]; 1757 for (p = ents; p <= last; p++) { 1758 if (p->len != dentry->d_name.len) 1759 continue; 1760 if (!memcmp(dentry->d_name.name, p->name, p->len)) 1761 break; 1762 } 1763 if (p > last) 1764 goto out; 1765 1766 error = proc_pident_instantiate(dir, dentry, task, p); 1767 out: 1768 put_task_struct(task); 1769 out_no_task: 1770 return error; 1771 } 1772 1773 static int proc_pident_fill_cache(struct file *filp, void *dirent, 1774 filldir_t filldir, struct task_struct *task, const struct pid_entry *p) 1775 { 1776 return proc_fill_cache(filp, dirent, filldir, p->name, p->len, 1777 proc_pident_instantiate, task, p); 1778 } 1779 1780 static int proc_pident_readdir(struct file *filp, 1781 void *dirent, filldir_t filldir, 1782 const struct pid_entry *ents, unsigned int nents) 1783 { 1784 int i; 1785 struct dentry *dentry = filp->f_path.dentry; 1786 struct inode *inode = dentry->d_inode; 1787 struct task_struct *task = get_proc_task(inode); 1788 const struct pid_entry *p, *last; 1789 ino_t ino; 1790 int ret; 1791 1792 ret = -ENOENT; 1793 if (!task) 1794 goto out_no_task; 1795 1796 ret = 0; 1797 i = filp->f_pos; 1798 switch (i) { 1799 case 0: 1800 ino = inode->i_ino; 1801 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0) 1802 goto out; 1803 i++; 1804 filp->f_pos++; 1805 /* fall through */ 1806 case 1: 1807 ino = parent_ino(dentry); 1808 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0) 1809 goto out; 1810 i++; 1811 filp->f_pos++; 1812 /* fall through */ 1813 default: 1814 i -= 2; 1815 if (i >= nents) { 1816 ret = 1; 1817 goto out; 1818 } 1819 p = ents + i; 1820 last = &ents[nents - 1]; 1821 while (p <= last) { 1822 if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0) 1823 goto out; 1824 filp->f_pos++; 1825 p++; 1826 } 1827 } 1828 1829 ret = 1; 1830 out: 1831 put_task_struct(task); 1832 out_no_task: 1833 return ret; 1834 } 1835 1836 #ifdef CONFIG_SECURITY 1837 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf, 1838 size_t count, loff_t *ppos) 1839 { 1840 struct inode * inode = file->f_path.dentry->d_inode; 1841 char *p = NULL; 1842 ssize_t length; 1843 struct task_struct *task = get_proc_task(inode); 1844 1845 if (!task) 1846 return -ESRCH; 1847 1848 length = security_getprocattr(task, 1849 (char*)file->f_path.dentry->d_name.name, 1850 &p); 1851 put_task_struct(task); 1852 if (length > 0) 1853 length = simple_read_from_buffer(buf, count, ppos, p, length); 1854 kfree(p); 1855 return length; 1856 } 1857 1858 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf, 1859 size_t count, loff_t *ppos) 1860 { 1861 struct inode * inode = file->f_path.dentry->d_inode; 1862 char *page; 1863 ssize_t length; 1864 struct task_struct *task = get_proc_task(inode); 1865 1866 length = -ESRCH; 1867 if (!task) 1868 goto out_no_task; 1869 if (count > PAGE_SIZE) 1870 count = PAGE_SIZE; 1871 1872 /* No partial writes. */ 1873 length = -EINVAL; 1874 if (*ppos != 0) 1875 goto out; 1876 1877 length = -ENOMEM; 1878 page = (char*)__get_free_page(GFP_TEMPORARY); 1879 if (!page) 1880 goto out; 1881 1882 length = -EFAULT; 1883 if (copy_from_user(page, buf, count)) 1884 goto out_free; 1885 1886 length = security_setprocattr(task, 1887 (char*)file->f_path.dentry->d_name.name, 1888 (void*)page, count); 1889 out_free: 1890 free_page((unsigned long) page); 1891 out: 1892 put_task_struct(task); 1893 out_no_task: 1894 return length; 1895 } 1896 1897 static const struct file_operations proc_pid_attr_operations = { 1898 .read = proc_pid_attr_read, 1899 .write = proc_pid_attr_write, 1900 }; 1901 1902 static const struct pid_entry attr_dir_stuff[] = { 1903 REG("current", S_IRUGO|S_IWUGO, pid_attr), 1904 REG("prev", S_IRUGO, pid_attr), 1905 REG("exec", S_IRUGO|S_IWUGO, pid_attr), 1906 REG("fscreate", S_IRUGO|S_IWUGO, pid_attr), 1907 REG("keycreate", S_IRUGO|S_IWUGO, pid_attr), 1908 REG("sockcreate", S_IRUGO|S_IWUGO, pid_attr), 1909 }; 1910 1911 static int proc_attr_dir_readdir(struct file * filp, 1912 void * dirent, filldir_t filldir) 1913 { 1914 return proc_pident_readdir(filp,dirent,filldir, 1915 attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff)); 1916 } 1917 1918 static const struct file_operations proc_attr_dir_operations = { 1919 .read = generic_read_dir, 1920 .readdir = proc_attr_dir_readdir, 1921 }; 1922 1923 static struct dentry *proc_attr_dir_lookup(struct inode *dir, 1924 struct dentry *dentry, struct nameidata *nd) 1925 { 1926 return proc_pident_lookup(dir, dentry, 1927 attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff)); 1928 } 1929 1930 static const struct inode_operations proc_attr_dir_inode_operations = { 1931 .lookup = proc_attr_dir_lookup, 1932 .getattr = pid_getattr, 1933 .setattr = proc_setattr, 1934 }; 1935 1936 #endif 1937 1938 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 1939 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf, 1940 size_t count, loff_t *ppos) 1941 { 1942 struct task_struct *task = get_proc_task(file->f_dentry->d_inode); 1943 struct mm_struct *mm; 1944 char buffer[PROC_NUMBUF]; 1945 size_t len; 1946 int ret; 1947 1948 if (!task) 1949 return -ESRCH; 1950 1951 ret = 0; 1952 mm = get_task_mm(task); 1953 if (mm) { 1954 len = snprintf(buffer, sizeof(buffer), "%08lx\n", 1955 ((mm->flags & MMF_DUMP_FILTER_MASK) >> 1956 MMF_DUMP_FILTER_SHIFT)); 1957 mmput(mm); 1958 ret = simple_read_from_buffer(buf, count, ppos, buffer, len); 1959 } 1960 1961 put_task_struct(task); 1962 1963 return ret; 1964 } 1965 1966 static ssize_t proc_coredump_filter_write(struct file *file, 1967 const char __user *buf, 1968 size_t count, 1969 loff_t *ppos) 1970 { 1971 struct task_struct *task; 1972 struct mm_struct *mm; 1973 char buffer[PROC_NUMBUF], *end; 1974 unsigned int val; 1975 int ret; 1976 int i; 1977 unsigned long mask; 1978 1979 ret = -EFAULT; 1980 memset(buffer, 0, sizeof(buffer)); 1981 if (count > sizeof(buffer) - 1) 1982 count = sizeof(buffer) - 1; 1983 if (copy_from_user(buffer, buf, count)) 1984 goto out_no_task; 1985 1986 ret = -EINVAL; 1987 val = (unsigned int)simple_strtoul(buffer, &end, 0); 1988 if (*end == '\n') 1989 end++; 1990 if (end - buffer == 0) 1991 goto out_no_task; 1992 1993 ret = -ESRCH; 1994 task = get_proc_task(file->f_dentry->d_inode); 1995 if (!task) 1996 goto out_no_task; 1997 1998 ret = end - buffer; 1999 mm = get_task_mm(task); 2000 if (!mm) 2001 goto out_no_mm; 2002 2003 for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) { 2004 if (val & mask) 2005 set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2006 else 2007 clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags); 2008 } 2009 2010 mmput(mm); 2011 out_no_mm: 2012 put_task_struct(task); 2013 out_no_task: 2014 return ret; 2015 } 2016 2017 static const struct file_operations proc_coredump_filter_operations = { 2018 .read = proc_coredump_filter_read, 2019 .write = proc_coredump_filter_write, 2020 }; 2021 #endif 2022 2023 /* 2024 * /proc/self: 2025 */ 2026 static int proc_self_readlink(struct dentry *dentry, char __user *buffer, 2027 int buflen) 2028 { 2029 char tmp[PROC_NUMBUF]; 2030 sprintf(tmp, "%d", task_tgid_vnr(current)); 2031 return vfs_readlink(dentry,buffer,buflen,tmp); 2032 } 2033 2034 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd) 2035 { 2036 char tmp[PROC_NUMBUF]; 2037 sprintf(tmp, "%d", task_tgid_vnr(current)); 2038 return ERR_PTR(vfs_follow_link(nd,tmp)); 2039 } 2040 2041 static const struct inode_operations proc_self_inode_operations = { 2042 .readlink = proc_self_readlink, 2043 .follow_link = proc_self_follow_link, 2044 }; 2045 2046 /* 2047 * proc base 2048 * 2049 * These are the directory entries in the root directory of /proc 2050 * that properly belong to the /proc filesystem, as they describe 2051 * describe something that is process related. 2052 */ 2053 static const struct pid_entry proc_base_stuff[] = { 2054 NOD("self", S_IFLNK|S_IRWXUGO, 2055 &proc_self_inode_operations, NULL, {}), 2056 }; 2057 2058 /* 2059 * Exceptional case: normally we are not allowed to unhash a busy 2060 * directory. In this case, however, we can do it - no aliasing problems 2061 * due to the way we treat inodes. 2062 */ 2063 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd) 2064 { 2065 struct inode *inode = dentry->d_inode; 2066 struct task_struct *task = get_proc_task(inode); 2067 if (task) { 2068 put_task_struct(task); 2069 return 1; 2070 } 2071 d_drop(dentry); 2072 return 0; 2073 } 2074 2075 static struct dentry_operations proc_base_dentry_operations = 2076 { 2077 .d_revalidate = proc_base_revalidate, 2078 .d_delete = pid_delete_dentry, 2079 }; 2080 2081 static struct dentry *proc_base_instantiate(struct inode *dir, 2082 struct dentry *dentry, struct task_struct *task, const void *ptr) 2083 { 2084 const struct pid_entry *p = ptr; 2085 struct inode *inode; 2086 struct proc_inode *ei; 2087 struct dentry *error = ERR_PTR(-EINVAL); 2088 2089 /* Allocate the inode */ 2090 error = ERR_PTR(-ENOMEM); 2091 inode = new_inode(dir->i_sb); 2092 if (!inode) 2093 goto out; 2094 2095 /* Initialize the inode */ 2096 ei = PROC_I(inode); 2097 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 2098 2099 /* 2100 * grab the reference to the task. 2101 */ 2102 ei->pid = get_task_pid(task, PIDTYPE_PID); 2103 if (!ei->pid) 2104 goto out_iput; 2105 2106 inode->i_uid = 0; 2107 inode->i_gid = 0; 2108 inode->i_mode = p->mode; 2109 if (S_ISDIR(inode->i_mode)) 2110 inode->i_nlink = 2; 2111 if (S_ISLNK(inode->i_mode)) 2112 inode->i_size = 64; 2113 if (p->iop) 2114 inode->i_op = p->iop; 2115 if (p->fop) 2116 inode->i_fop = p->fop; 2117 ei->op = p->op; 2118 dentry->d_op = &proc_base_dentry_operations; 2119 d_add(dentry, inode); 2120 error = NULL; 2121 out: 2122 return error; 2123 out_iput: 2124 iput(inode); 2125 goto out; 2126 } 2127 2128 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry) 2129 { 2130 struct dentry *error; 2131 struct task_struct *task = get_proc_task(dir); 2132 const struct pid_entry *p, *last; 2133 2134 error = ERR_PTR(-ENOENT); 2135 2136 if (!task) 2137 goto out_no_task; 2138 2139 /* Lookup the directory entry */ 2140 last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1]; 2141 for (p = proc_base_stuff; p <= last; p++) { 2142 if (p->len != dentry->d_name.len) 2143 continue; 2144 if (!memcmp(dentry->d_name.name, p->name, p->len)) 2145 break; 2146 } 2147 if (p > last) 2148 goto out; 2149 2150 error = proc_base_instantiate(dir, dentry, task, p); 2151 2152 out: 2153 put_task_struct(task); 2154 out_no_task: 2155 return error; 2156 } 2157 2158 static int proc_base_fill_cache(struct file *filp, void *dirent, 2159 filldir_t filldir, struct task_struct *task, const struct pid_entry *p) 2160 { 2161 return proc_fill_cache(filp, dirent, filldir, p->name, p->len, 2162 proc_base_instantiate, task, p); 2163 } 2164 2165 #ifdef CONFIG_TASK_IO_ACCOUNTING 2166 static int proc_pid_io_accounting(struct task_struct *task, char *buffer) 2167 { 2168 return sprintf(buffer, 2169 #ifdef CONFIG_TASK_XACCT 2170 "rchar: %llu\n" 2171 "wchar: %llu\n" 2172 "syscr: %llu\n" 2173 "syscw: %llu\n" 2174 #endif 2175 "read_bytes: %llu\n" 2176 "write_bytes: %llu\n" 2177 "cancelled_write_bytes: %llu\n", 2178 #ifdef CONFIG_TASK_XACCT 2179 (unsigned long long)task->rchar, 2180 (unsigned long long)task->wchar, 2181 (unsigned long long)task->syscr, 2182 (unsigned long long)task->syscw, 2183 #endif 2184 (unsigned long long)task->ioac.read_bytes, 2185 (unsigned long long)task->ioac.write_bytes, 2186 (unsigned long long)task->ioac.cancelled_write_bytes); 2187 } 2188 #endif 2189 2190 /* 2191 * Thread groups 2192 */ 2193 static const struct file_operations proc_task_operations; 2194 static const struct inode_operations proc_task_inode_operations; 2195 2196 static const struct pid_entry tgid_base_stuff[] = { 2197 DIR("task", S_IRUGO|S_IXUGO, task), 2198 DIR("fd", S_IRUSR|S_IXUSR, fd), 2199 DIR("fdinfo", S_IRUSR|S_IXUSR, fdinfo), 2200 REG("environ", S_IRUSR, environ), 2201 INF("auxv", S_IRUSR, pid_auxv), 2202 INF("status", S_IRUGO, pid_status), 2203 INF("limits", S_IRUSR, pid_limits), 2204 #ifdef CONFIG_SCHED_DEBUG 2205 REG("sched", S_IRUGO|S_IWUSR, pid_sched), 2206 #endif 2207 INF("cmdline", S_IRUGO, pid_cmdline), 2208 INF("stat", S_IRUGO, tgid_stat), 2209 INF("statm", S_IRUGO, pid_statm), 2210 REG("maps", S_IRUGO, maps), 2211 #ifdef CONFIG_NUMA 2212 REG("numa_maps", S_IRUGO, numa_maps), 2213 #endif 2214 REG("mem", S_IRUSR|S_IWUSR, mem), 2215 LNK("cwd", cwd), 2216 LNK("root", root), 2217 LNK("exe", exe), 2218 REG("mounts", S_IRUGO, mounts), 2219 REG("mountstats", S_IRUSR, mountstats), 2220 #ifdef CONFIG_MMU 2221 REG("clear_refs", S_IWUSR, clear_refs), 2222 REG("smaps", S_IRUGO, smaps), 2223 #endif 2224 #ifdef CONFIG_SECURITY 2225 DIR("attr", S_IRUGO|S_IXUGO, attr_dir), 2226 #endif 2227 #ifdef CONFIG_KALLSYMS 2228 INF("wchan", S_IRUGO, pid_wchan), 2229 #endif 2230 #ifdef CONFIG_SCHEDSTATS 2231 INF("schedstat", S_IRUGO, pid_schedstat), 2232 #endif 2233 #ifdef CONFIG_PROC_PID_CPUSET 2234 REG("cpuset", S_IRUGO, cpuset), 2235 #endif 2236 #ifdef CONFIG_CGROUPS 2237 REG("cgroup", S_IRUGO, cgroup), 2238 #endif 2239 INF("oom_score", S_IRUGO, oom_score), 2240 REG("oom_adj", S_IRUGO|S_IWUSR, oom_adjust), 2241 #ifdef CONFIG_AUDITSYSCALL 2242 REG("loginuid", S_IWUSR|S_IRUGO, loginuid), 2243 #endif 2244 #ifdef CONFIG_FAULT_INJECTION 2245 REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject), 2246 #endif 2247 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE) 2248 REG("coredump_filter", S_IRUGO|S_IWUSR, coredump_filter), 2249 #endif 2250 #ifdef CONFIG_TASK_IO_ACCOUNTING 2251 INF("io", S_IRUGO, pid_io_accounting), 2252 #endif 2253 }; 2254 2255 static int proc_tgid_base_readdir(struct file * filp, 2256 void * dirent, filldir_t filldir) 2257 { 2258 return proc_pident_readdir(filp,dirent,filldir, 2259 tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff)); 2260 } 2261 2262 static const struct file_operations proc_tgid_base_operations = { 2263 .read = generic_read_dir, 2264 .readdir = proc_tgid_base_readdir, 2265 }; 2266 2267 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){ 2268 return proc_pident_lookup(dir, dentry, 2269 tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff)); 2270 } 2271 2272 static const struct inode_operations proc_tgid_base_inode_operations = { 2273 .lookup = proc_tgid_base_lookup, 2274 .getattr = pid_getattr, 2275 .setattr = proc_setattr, 2276 }; 2277 2278 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid) 2279 { 2280 struct dentry *dentry, *leader, *dir; 2281 char buf[PROC_NUMBUF]; 2282 struct qstr name; 2283 2284 name.name = buf; 2285 name.len = snprintf(buf, sizeof(buf), "%d", pid); 2286 dentry = d_hash_and_lookup(mnt->mnt_root, &name); 2287 if (dentry) { 2288 shrink_dcache_parent(dentry); 2289 d_drop(dentry); 2290 dput(dentry); 2291 } 2292 2293 if (tgid == 0) 2294 goto out; 2295 2296 name.name = buf; 2297 name.len = snprintf(buf, sizeof(buf), "%d", tgid); 2298 leader = d_hash_and_lookup(mnt->mnt_root, &name); 2299 if (!leader) 2300 goto out; 2301 2302 name.name = "task"; 2303 name.len = strlen(name.name); 2304 dir = d_hash_and_lookup(leader, &name); 2305 if (!dir) 2306 goto out_put_leader; 2307 2308 name.name = buf; 2309 name.len = snprintf(buf, sizeof(buf), "%d", pid); 2310 dentry = d_hash_and_lookup(dir, &name); 2311 if (dentry) { 2312 shrink_dcache_parent(dentry); 2313 d_drop(dentry); 2314 dput(dentry); 2315 } 2316 2317 dput(dir); 2318 out_put_leader: 2319 dput(leader); 2320 out: 2321 return; 2322 } 2323 2324 /** 2325 * proc_flush_task - Remove dcache entries for @task from the /proc dcache. 2326 * @task: task that should be flushed. 2327 * 2328 * When flushing dentries from proc, one needs to flush them from global 2329 * proc (proc_mnt) and from all the namespaces' procs this task was seen 2330 * in. This call is supposed to do all of this job. 2331 * 2332 * Looks in the dcache for 2333 * /proc/@pid 2334 * /proc/@tgid/task/@pid 2335 * if either directory is present flushes it and all of it'ts children 2336 * from the dcache. 2337 * 2338 * It is safe and reasonable to cache /proc entries for a task until 2339 * that task exits. After that they just clog up the dcache with 2340 * useless entries, possibly causing useful dcache entries to be 2341 * flushed instead. This routine is proved to flush those useless 2342 * dcache entries at process exit time. 2343 * 2344 * NOTE: This routine is just an optimization so it does not guarantee 2345 * that no dcache entries will exist at process exit time it 2346 * just makes it very unlikely that any will persist. 2347 */ 2348 2349 void proc_flush_task(struct task_struct *task) 2350 { 2351 int i; 2352 struct pid *pid, *tgid = NULL; 2353 struct upid *upid; 2354 2355 pid = task_pid(task); 2356 if (thread_group_leader(task)) 2357 tgid = task_tgid(task); 2358 2359 for (i = 0; i <= pid->level; i++) { 2360 upid = &pid->numbers[i]; 2361 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr, 2362 tgid ? tgid->numbers[i].nr : 0); 2363 } 2364 2365 upid = &pid->numbers[pid->level]; 2366 if (upid->nr == 1) 2367 pid_ns_release_proc(upid->ns); 2368 } 2369 2370 static struct dentry *proc_pid_instantiate(struct inode *dir, 2371 struct dentry * dentry, 2372 struct task_struct *task, const void *ptr) 2373 { 2374 struct dentry *error = ERR_PTR(-ENOENT); 2375 struct inode *inode; 2376 2377 inode = proc_pid_make_inode(dir->i_sb, task); 2378 if (!inode) 2379 goto out; 2380 2381 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO; 2382 inode->i_op = &proc_tgid_base_inode_operations; 2383 inode->i_fop = &proc_tgid_base_operations; 2384 inode->i_flags|=S_IMMUTABLE; 2385 inode->i_nlink = 5; 2386 #ifdef CONFIG_SECURITY 2387 inode->i_nlink += 1; 2388 #endif 2389 2390 dentry->d_op = &pid_dentry_operations; 2391 2392 d_add(dentry, inode); 2393 /* Close the race of the process dying before we return the dentry */ 2394 if (pid_revalidate(dentry, NULL)) 2395 error = NULL; 2396 out: 2397 return error; 2398 } 2399 2400 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 2401 { 2402 struct dentry *result = ERR_PTR(-ENOENT); 2403 struct task_struct *task; 2404 unsigned tgid; 2405 struct pid_namespace *ns; 2406 2407 result = proc_base_lookup(dir, dentry); 2408 if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT) 2409 goto out; 2410 2411 tgid = name_to_int(dentry); 2412 if (tgid == ~0U) 2413 goto out; 2414 2415 ns = dentry->d_sb->s_fs_info; 2416 rcu_read_lock(); 2417 task = find_task_by_pid_ns(tgid, ns); 2418 if (task) 2419 get_task_struct(task); 2420 rcu_read_unlock(); 2421 if (!task) 2422 goto out; 2423 2424 result = proc_pid_instantiate(dir, dentry, task, NULL); 2425 put_task_struct(task); 2426 out: 2427 return result; 2428 } 2429 2430 /* 2431 * Find the first task with tgid >= tgid 2432 * 2433 */ 2434 struct tgid_iter { 2435 unsigned int tgid; 2436 struct task_struct *task; 2437 }; 2438 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter) 2439 { 2440 struct pid *pid; 2441 2442 if (iter.task) 2443 put_task_struct(iter.task); 2444 rcu_read_lock(); 2445 retry: 2446 iter.task = NULL; 2447 pid = find_ge_pid(iter.tgid, ns); 2448 if (pid) { 2449 iter.tgid = pid_nr_ns(pid, ns); 2450 iter.task = pid_task(pid, PIDTYPE_PID); 2451 /* What we to know is if the pid we have find is the 2452 * pid of a thread_group_leader. Testing for task 2453 * being a thread_group_leader is the obvious thing 2454 * todo but there is a window when it fails, due to 2455 * the pid transfer logic in de_thread. 2456 * 2457 * So we perform the straight forward test of seeing 2458 * if the pid we have found is the pid of a thread 2459 * group leader, and don't worry if the task we have 2460 * found doesn't happen to be a thread group leader. 2461 * As we don't care in the case of readdir. 2462 */ 2463 if (!iter.task || !has_group_leader_pid(iter.task)) { 2464 iter.tgid += 1; 2465 goto retry; 2466 } 2467 get_task_struct(iter.task); 2468 } 2469 rcu_read_unlock(); 2470 return iter; 2471 } 2472 2473 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff)) 2474 2475 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir, 2476 struct tgid_iter iter) 2477 { 2478 char name[PROC_NUMBUF]; 2479 int len = snprintf(name, sizeof(name), "%d", iter.tgid); 2480 return proc_fill_cache(filp, dirent, filldir, name, len, 2481 proc_pid_instantiate, iter.task, NULL); 2482 } 2483 2484 /* for the /proc/ directory itself, after non-process stuff has been done */ 2485 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir) 2486 { 2487 unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY; 2488 struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode); 2489 struct tgid_iter iter; 2490 struct pid_namespace *ns; 2491 2492 if (!reaper) 2493 goto out_no_task; 2494 2495 for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) { 2496 const struct pid_entry *p = &proc_base_stuff[nr]; 2497 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0) 2498 goto out; 2499 } 2500 2501 ns = filp->f_dentry->d_sb->s_fs_info; 2502 iter.task = NULL; 2503 iter.tgid = filp->f_pos - TGID_OFFSET; 2504 for (iter = next_tgid(ns, iter); 2505 iter.task; 2506 iter.tgid += 1, iter = next_tgid(ns, iter)) { 2507 filp->f_pos = iter.tgid + TGID_OFFSET; 2508 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) { 2509 put_task_struct(iter.task); 2510 goto out; 2511 } 2512 } 2513 filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET; 2514 out: 2515 put_task_struct(reaper); 2516 out_no_task: 2517 return 0; 2518 } 2519 2520 /* 2521 * Tasks 2522 */ 2523 static const struct pid_entry tid_base_stuff[] = { 2524 DIR("fd", S_IRUSR|S_IXUSR, fd), 2525 DIR("fdinfo", S_IRUSR|S_IXUSR, fdinfo), 2526 REG("environ", S_IRUSR, environ), 2527 INF("auxv", S_IRUSR, pid_auxv), 2528 INF("status", S_IRUGO, pid_status), 2529 INF("limits", S_IRUSR, pid_limits), 2530 #ifdef CONFIG_SCHED_DEBUG 2531 REG("sched", S_IRUGO|S_IWUSR, pid_sched), 2532 #endif 2533 INF("cmdline", S_IRUGO, pid_cmdline), 2534 INF("stat", S_IRUGO, tid_stat), 2535 INF("statm", S_IRUGO, pid_statm), 2536 REG("maps", S_IRUGO, maps), 2537 #ifdef CONFIG_NUMA 2538 REG("numa_maps", S_IRUGO, numa_maps), 2539 #endif 2540 REG("mem", S_IRUSR|S_IWUSR, mem), 2541 LNK("cwd", cwd), 2542 LNK("root", root), 2543 LNK("exe", exe), 2544 REG("mounts", S_IRUGO, mounts), 2545 #ifdef CONFIG_MMU 2546 REG("clear_refs", S_IWUSR, clear_refs), 2547 REG("smaps", S_IRUGO, smaps), 2548 #endif 2549 #ifdef CONFIG_SECURITY 2550 DIR("attr", S_IRUGO|S_IXUGO, attr_dir), 2551 #endif 2552 #ifdef CONFIG_KALLSYMS 2553 INF("wchan", S_IRUGO, pid_wchan), 2554 #endif 2555 #ifdef CONFIG_SCHEDSTATS 2556 INF("schedstat", S_IRUGO, pid_schedstat), 2557 #endif 2558 #ifdef CONFIG_PROC_PID_CPUSET 2559 REG("cpuset", S_IRUGO, cpuset), 2560 #endif 2561 #ifdef CONFIG_CGROUPS 2562 REG("cgroup", S_IRUGO, cgroup), 2563 #endif 2564 INF("oom_score", S_IRUGO, oom_score), 2565 REG("oom_adj", S_IRUGO|S_IWUSR, oom_adjust), 2566 #ifdef CONFIG_AUDITSYSCALL 2567 REG("loginuid", S_IWUSR|S_IRUGO, loginuid), 2568 #endif 2569 #ifdef CONFIG_FAULT_INJECTION 2570 REG("make-it-fail", S_IRUGO|S_IWUSR, fault_inject), 2571 #endif 2572 }; 2573 2574 static int proc_tid_base_readdir(struct file * filp, 2575 void * dirent, filldir_t filldir) 2576 { 2577 return proc_pident_readdir(filp,dirent,filldir, 2578 tid_base_stuff,ARRAY_SIZE(tid_base_stuff)); 2579 } 2580 2581 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){ 2582 return proc_pident_lookup(dir, dentry, 2583 tid_base_stuff, ARRAY_SIZE(tid_base_stuff)); 2584 } 2585 2586 static const struct file_operations proc_tid_base_operations = { 2587 .read = generic_read_dir, 2588 .readdir = proc_tid_base_readdir, 2589 }; 2590 2591 static const struct inode_operations proc_tid_base_inode_operations = { 2592 .lookup = proc_tid_base_lookup, 2593 .getattr = pid_getattr, 2594 .setattr = proc_setattr, 2595 }; 2596 2597 static struct dentry *proc_task_instantiate(struct inode *dir, 2598 struct dentry *dentry, struct task_struct *task, const void *ptr) 2599 { 2600 struct dentry *error = ERR_PTR(-ENOENT); 2601 struct inode *inode; 2602 inode = proc_pid_make_inode(dir->i_sb, task); 2603 2604 if (!inode) 2605 goto out; 2606 inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO; 2607 inode->i_op = &proc_tid_base_inode_operations; 2608 inode->i_fop = &proc_tid_base_operations; 2609 inode->i_flags|=S_IMMUTABLE; 2610 inode->i_nlink = 4; 2611 #ifdef CONFIG_SECURITY 2612 inode->i_nlink += 1; 2613 #endif 2614 2615 dentry->d_op = &pid_dentry_operations; 2616 2617 d_add(dentry, inode); 2618 /* Close the race of the process dying before we return the dentry */ 2619 if (pid_revalidate(dentry, NULL)) 2620 error = NULL; 2621 out: 2622 return error; 2623 } 2624 2625 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) 2626 { 2627 struct dentry *result = ERR_PTR(-ENOENT); 2628 struct task_struct *task; 2629 struct task_struct *leader = get_proc_task(dir); 2630 unsigned tid; 2631 struct pid_namespace *ns; 2632 2633 if (!leader) 2634 goto out_no_task; 2635 2636 tid = name_to_int(dentry); 2637 if (tid == ~0U) 2638 goto out; 2639 2640 ns = dentry->d_sb->s_fs_info; 2641 rcu_read_lock(); 2642 task = find_task_by_pid_ns(tid, ns); 2643 if (task) 2644 get_task_struct(task); 2645 rcu_read_unlock(); 2646 if (!task) 2647 goto out; 2648 if (!same_thread_group(leader, task)) 2649 goto out_drop_task; 2650 2651 result = proc_task_instantiate(dir, dentry, task, NULL); 2652 out_drop_task: 2653 put_task_struct(task); 2654 out: 2655 put_task_struct(leader); 2656 out_no_task: 2657 return result; 2658 } 2659 2660 /* 2661 * Find the first tid of a thread group to return to user space. 2662 * 2663 * Usually this is just the thread group leader, but if the users 2664 * buffer was too small or there was a seek into the middle of the 2665 * directory we have more work todo. 2666 * 2667 * In the case of a short read we start with find_task_by_pid. 2668 * 2669 * In the case of a seek we start with the leader and walk nr 2670 * threads past it. 2671 */ 2672 static struct task_struct *first_tid(struct task_struct *leader, 2673 int tid, int nr, struct pid_namespace *ns) 2674 { 2675 struct task_struct *pos; 2676 2677 rcu_read_lock(); 2678 /* Attempt to start with the pid of a thread */ 2679 if (tid && (nr > 0)) { 2680 pos = find_task_by_pid_ns(tid, ns); 2681 if (pos && (pos->group_leader == leader)) 2682 goto found; 2683 } 2684 2685 /* If nr exceeds the number of threads there is nothing todo */ 2686 pos = NULL; 2687 if (nr && nr >= get_nr_threads(leader)) 2688 goto out; 2689 2690 /* If we haven't found our starting place yet start 2691 * with the leader and walk nr threads forward. 2692 */ 2693 for (pos = leader; nr > 0; --nr) { 2694 pos = next_thread(pos); 2695 if (pos == leader) { 2696 pos = NULL; 2697 goto out; 2698 } 2699 } 2700 found: 2701 get_task_struct(pos); 2702 out: 2703 rcu_read_unlock(); 2704 return pos; 2705 } 2706 2707 /* 2708 * Find the next thread in the thread list. 2709 * Return NULL if there is an error or no next thread. 2710 * 2711 * The reference to the input task_struct is released. 2712 */ 2713 static struct task_struct *next_tid(struct task_struct *start) 2714 { 2715 struct task_struct *pos = NULL; 2716 rcu_read_lock(); 2717 if (pid_alive(start)) { 2718 pos = next_thread(start); 2719 if (thread_group_leader(pos)) 2720 pos = NULL; 2721 else 2722 get_task_struct(pos); 2723 } 2724 rcu_read_unlock(); 2725 put_task_struct(start); 2726 return pos; 2727 } 2728 2729 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir, 2730 struct task_struct *task, int tid) 2731 { 2732 char name[PROC_NUMBUF]; 2733 int len = snprintf(name, sizeof(name), "%d", tid); 2734 return proc_fill_cache(filp, dirent, filldir, name, len, 2735 proc_task_instantiate, task, NULL); 2736 } 2737 2738 /* for the /proc/TGID/task/ directories */ 2739 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir) 2740 { 2741 struct dentry *dentry = filp->f_path.dentry; 2742 struct inode *inode = dentry->d_inode; 2743 struct task_struct *leader = NULL; 2744 struct task_struct *task; 2745 int retval = -ENOENT; 2746 ino_t ino; 2747 int tid; 2748 unsigned long pos = filp->f_pos; /* avoiding "long long" filp->f_pos */ 2749 struct pid_namespace *ns; 2750 2751 task = get_proc_task(inode); 2752 if (!task) 2753 goto out_no_task; 2754 rcu_read_lock(); 2755 if (pid_alive(task)) { 2756 leader = task->group_leader; 2757 get_task_struct(leader); 2758 } 2759 rcu_read_unlock(); 2760 put_task_struct(task); 2761 if (!leader) 2762 goto out_no_task; 2763 retval = 0; 2764 2765 switch (pos) { 2766 case 0: 2767 ino = inode->i_ino; 2768 if (filldir(dirent, ".", 1, pos, ino, DT_DIR) < 0) 2769 goto out; 2770 pos++; 2771 /* fall through */ 2772 case 1: 2773 ino = parent_ino(dentry); 2774 if (filldir(dirent, "..", 2, pos, ino, DT_DIR) < 0) 2775 goto out; 2776 pos++; 2777 /* fall through */ 2778 } 2779 2780 /* f_version caches the tgid value that the last readdir call couldn't 2781 * return. lseek aka telldir automagically resets f_version to 0. 2782 */ 2783 ns = filp->f_dentry->d_sb->s_fs_info; 2784 tid = (int)filp->f_version; 2785 filp->f_version = 0; 2786 for (task = first_tid(leader, tid, pos - 2, ns); 2787 task; 2788 task = next_tid(task), pos++) { 2789 tid = task_pid_nr_ns(task, ns); 2790 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) { 2791 /* returning this tgid failed, save it as the first 2792 * pid for the next readir call */ 2793 filp->f_version = (u64)tid; 2794 put_task_struct(task); 2795 break; 2796 } 2797 } 2798 out: 2799 filp->f_pos = pos; 2800 put_task_struct(leader); 2801 out_no_task: 2802 return retval; 2803 } 2804 2805 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) 2806 { 2807 struct inode *inode = dentry->d_inode; 2808 struct task_struct *p = get_proc_task(inode); 2809 generic_fillattr(inode, stat); 2810 2811 if (p) { 2812 rcu_read_lock(); 2813 stat->nlink += get_nr_threads(p); 2814 rcu_read_unlock(); 2815 put_task_struct(p); 2816 } 2817 2818 return 0; 2819 } 2820 2821 static const struct inode_operations proc_task_inode_operations = { 2822 .lookup = proc_task_lookup, 2823 .getattr = proc_task_getattr, 2824 .setattr = proc_setattr, 2825 }; 2826 2827 static const struct file_operations proc_task_operations = { 2828 .read = generic_read_dir, 2829 .readdir = proc_task_readdir, 2830 }; 2831