fs/proc/base.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  *  linux/fs/proc/base.c
   4  *
   5  *  Copyright (C) 1991, 1992 Linus Torvalds
   6  *
   7  *  proc base directory handling functions
   8  *
   9  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
  10  *  Instead of using magical inumbers to determine the kind of object
  11  *  we allocate and fill in-core inodes upon lookup. They don't even
  12  *  go into icache. We cache the reference to task_struct upon lookup too.
  13  *  Eventually it should become a filesystem in its own. We don't use the
  14  *  rest of procfs anymore.
  15  *
  16  *
  17  *  Changelog:
  18  *  17-Jan-2005
  19  *  Allan Bezerra
  20  *  Bruna Moreira <bruna.moreira@indt.org.br>
  21  *  Edjard Mota <edjard.mota@indt.org.br>
  22  *  Ilias Biris <ilias.biris@indt.org.br>
  23  *  Mauricio Lin <mauricio.lin@indt.org.br>
  24  *
  25  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  26  *
  27  *  A new process specific entry (smaps) included in /proc. It shows the
  28  *  size of rss for each memory area. The maps entry lacks information
  29  *  about physical memory size (rss) for each mapped file, i.e.,
  30  *  rss information for executables and library files.
  31  *  This additional information is useful for any tools that need to know
  32  *  about physical memory consumption for a process specific library.
  33  *
  34  *  Changelog:
  35  *  21-Feb-2005
  36  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  37  *  Pud inclusion in the page table walking.
  38  *
  39  *  ChangeLog:
  40  *  10-Mar-2005
  41  *  10LE Instituto Nokia de Tecnologia - INdT:
  42  *  A better way to walks through the page table as suggested by Hugh Dickins.
  43  *
  44  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  45  *  Smaps information related to shared, private, clean and dirty pages.
  46  *
  47  *  Paul Mundt <paul.mundt@nokia.com>:
  48  *  Overall revision about smaps.
  49  */
  50
  51 #include <linux/uaccess.h>
  52
  53 #include <linux/errno.h>
  54 #include <linux/time.h>
  55 #include <linux/proc_fs.h>
  56 #include <linux/stat.h>
  57 #include <linux/task_io_accounting_ops.h>
  58 #include <linux/init.h>
  59 #include <linux/capability.h>
  60 #include <linux/file.h>
  61 #include <linux/fdtable.h>
  62 #include <linux/generic-radix-tree.h>
  63 #include <linux/string.h>
  64 #include <linux/seq_file.h>
  65 #include <linux/namei.h>
  66 #include <linux/mnt_namespace.h>
  67 #include <linux/mm.h>
  68 #include <linux/swap.h>
  69 #include <linux/rcupdate.h>
  70 #include <linux/kallsyms.h>
  71 #include <linux/stacktrace.h>
  72 #include <linux/resource.h>
  73 #include <linux/module.h>
  74 #include <linux/mount.h>
  75 #include <linux/security.h>
  76 #include <linux/ptrace.h>
  77 #include <linux/tracehook.h>
  78 #include <linux/printk.h>
  79 #include <linux/cache.h>
  80 #include <linux/cgroup.h>
  81 #include <linux/cpuset.h>
  82 #include <linux/audit.h>
  83 #include <linux/poll.h>
  84 #include <linux/nsproxy.h>
  85 #include <linux/oom.h>
  86 #include <linux/elf.h>
  87 #include <linux/pid_namespace.h>
  88 #include <linux/user_namespace.h>
  89 #include <linux/fs_struct.h>
  90 #include <linux/slab.h>
  91 #include <linux/sched/autogroup.h>
  92 #include <linux/sched/mm.h>
  93 #include <linux/sched/coredump.h>
  94 #include <linux/sched/debug.h>
  95 #include <linux/sched/stat.h>
  96 #include <linux/posix-timers.h>
  97 #include <trace/events/oom.h>
  98 #include "internal.h"
  99 #include "fd.h"
 100
 101 #include "../../lib/kstrtox.h"
 102
 103 /* NOTE:
 104  *      Implementing inode permission operations in /proc is almost
 105  *      certainly an error.  Permission checks need to happen during
 106  *      each system call not at open time.  The reason is that most of
 107  *      what we wish to check for permissions in /proc varies at runtime.
 108  *
 109  *      The classic example of a problem is opening file descriptors
 110  *      in /proc for a task before it execs a suid executable.
 111  */
 112
 113 static u8 nlink_tid __ro_after_init;
 114 static u8 nlink_tgid __ro_after_init;
 115
 116 struct pid_entry {
 117         const char *name;
 118         unsigned int len;
 119         umode_t mode;
 120         const struct inode_operations *iop;
 121         const struct file_operations *fop;
 122         union proc_op op;
 123 };
 124
 125 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 126         .name = (NAME),                                 \
 127         .len  = sizeof(NAME) - 1,                       \
 128         .mode = MODE,                                   \
 129         .iop  = IOP,                                    \
 130         .fop  = FOP,                                    \
 131         .op   = OP,                                     \
 132 }
 133
 134 #define DIR(NAME, MODE, iops, fops)     \
 135         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 136 #define LNK(NAME, get_link)                                     \
 137         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 138                 &proc_pid_link_inode_operations, NULL,          \
 139                 { .proc_get_link = get_link } )
 140 #define REG(NAME, MODE, fops)                           \
 141         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 142 #define ONE(NAME, MODE, show)                           \
 143         NOD(NAME, (S_IFREG|(MODE)),                     \
 144                 NULL, &proc_single_file_operations,     \
 145                 { .proc_show = show } )
 146 #define ATTR(LSM, NAME, MODE)                           \
 147         NOD(NAME, (S_IFREG|(MODE)),                     \
 148                 NULL, &proc_pid_attr_operations,        \
 149                 { .lsm = LSM })
 150
 151 /*
 152  * Count the number of hardlinks for the pid_entry table, excluding the .
 153  * and .. links.
 154  */
 155 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 156         unsigned int n)
 157 {
 158         unsigned int i;
 159         unsigned int count;
 160
 161         count = 2;
 162         for (i = 0; i < n; ++i) {
 163                 if (S_ISDIR(entries[i].mode))
 164                         ++count;
 165         }
 166
 167         return count;
 168 }
 169
 170 static int get_task_root(struct task_struct *task, struct path *root)
 171 {
 172         int result = -ENOENT;
 173
 174         task_lock(task);
 175         if (task->fs) {
 176                 get_fs_root(task->fs, root);
 177                 result = 0;
 178         }
 179         task_unlock(task);
 180         return result;
 181 }
 182
 183 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 184 {
 185         struct task_struct *task = get_proc_task(d_inode(dentry));
 186         int result = -ENOENT;
 187
 188         if (task) {
 189                 task_lock(task);
 190                 if (task->fs) {
 191                         get_fs_pwd(task->fs, path);
 192                         result = 0;
 193                 }
 194                 task_unlock(task);
 195                 put_task_struct(task);
 196         }
 197         return result;
 198 }
 199
 200 static int proc_root_link(struct dentry *dentry, struct path *path)
 201 {
 202         struct task_struct *task = get_proc_task(d_inode(dentry));
 203         int result = -ENOENT;
 204
 205         if (task) {
 206                 result = get_task_root(task, path);
 207                 put_task_struct(task);
 208         }
 209         return result;
 210 }
 211
 212 /*
 213  * If the user used setproctitle(), we just get the string from
 214  * user space at arg_start, and limit it to a maximum of one page.
 215  */
 216 static ssize_t get_mm_proctitle(struct mm_struct *mm, char __user *buf,
 217                                 size_t count, unsigned long pos,
 218                                 unsigned long arg_start)
 219 {
 220         char *page;
 221         int ret, got;
 222
 223         if (pos >= PAGE_SIZE)
 224                 return 0;
 225
 226         page = (char *)__get_free_page(GFP_KERNEL);
 227         if (!page)
 228                 return -ENOMEM;
 229
 230         ret = 0;
 231         got = access_remote_vm(mm, arg_start, page, PAGE_SIZE, FOLL_ANON);
 232         if (got > 0) {
 233                 int len = strnlen(page, got);
 234
 235                 /* Include the NUL character if it was found */
 236                 if (len < got)
 237                         len++;
 238
 239                 if (len > pos) {
 240                         len -= pos;
 241                         if (len > count)
 242                                 len = count;
 243                         len -= copy_to_user(buf, page+pos, len);
 244                         if (!len)
 245                                 len = -EFAULT;
 246                         ret = len;
 247                 }
 248         }
 249         free_page((unsigned long)page);
 250         return ret;
 251 }
 252
 253 static ssize_t get_mm_cmdline(struct mm_struct *mm, char __user *buf,
 254                               size_t count, loff_t *ppos)
 255 {
 256         unsigned long arg_start, arg_end, env_start, env_end;
 257         unsigned long pos, len;
 258         char *page, c;
 259
 260         /* Check if process spawned far enough to have cmdline. */
 261         if (!mm->env_end)
 262                 return 0;
 263
 264         spin_lock(&mm->arg_lock);
 265         arg_start = mm->arg_start;
 266         arg_end = mm->arg_end;
 267         env_start = mm->env_start;
 268         env_end = mm->env_end;
 269         spin_unlock(&mm->arg_lock);
 270
 271         if (arg_start >= arg_end)
 272                 return 0;
 273
 274         /*
 275          * We allow setproctitle() to overwrite the argument
 276          * strings, and overflow past the original end. But
 277          * only when it overflows into the environment area.
 278          */
 279         if (env_start != arg_end || env_end < env_start)
 280                 env_start = env_end = arg_end;
 281         len = env_end - arg_start;
 282
 283         /* We're not going to care if "*ppos" has high bits set */
 284         pos = *ppos;
 285         if (pos >= len)
 286                 return 0;
 287         if (count > len - pos)
 288                 count = len - pos;
 289         if (!count)
 290                 return 0;
 291
 292         /*
 293          * Magical special case: if the argv[] end byte is not
 294          * zero, the user has overwritten it with setproctitle(3).
 295          *
 296          * Possible future enhancement: do this only once when
 297          * pos is 0, and set a flag in the 'struct file'.
 298          */
 299         if (access_remote_vm(mm, arg_end-1, &c, 1, FOLL_ANON) == 1 && c)
 300                 return get_mm_proctitle(mm, buf, count, pos, arg_start);
 301
 302         /*
 303          * For the non-setproctitle() case we limit things strictly
 304          * to the [arg_start, arg_end[ range.
 305          */
 306         pos += arg_start;
 307         if (pos < arg_start || pos >= arg_end)
 308                 return 0;
 309         if (count > arg_end - pos)
 310                 count = arg_end - pos;
 311
 312         page = (char *)__get_free_page(GFP_KERNEL);
 313         if (!page)
 314                 return -ENOMEM;
 315
 316         len = 0;
 317         while (count) {
 318                 int got;
 319                 size_t size = min_t(size_t, PAGE_SIZE, count);
 320
 321                 got = access_remote_vm(mm, pos, page, size, FOLL_ANON);
 322                 if (got <= 0)
 323                         break;
 324                 got -= copy_to_user(buf, page, got);
 325                 if (unlikely(!got)) {
 326                         if (!len)
 327                                 len = -EFAULT;
 328                         break;
 329                 }
 330                 pos += got;
 331                 buf += got;
 332                 len += got;
 333                 count -= got;
 334         }
 335
 336         free_page((unsigned long)page);
 337         return len;
 338 }
 339
 340 static ssize_t get_task_cmdline(struct task_struct *tsk, char __user *buf,
 341                                 size_t count, loff_t *pos)
 342 {
 343         struct mm_struct *mm;
 344         ssize_t ret;
 345
 346         mm = get_task_mm(tsk);
 347         if (!mm)
 348                 return 0;
 349
 350         ret = get_mm_cmdline(mm, buf, count, pos);
 351         mmput(mm);
 352         return ret;
 353 }
 354
 355 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 356                                      size_t count, loff_t *pos)
 357 {
 358         struct task_struct *tsk;
 359         ssize_t ret;
 360
 361         BUG_ON(*pos < 0);
 362
 363         tsk = get_proc_task(file_inode(file));
 364         if (!tsk)
 365                 return -ESRCH;
 366         ret = get_task_cmdline(tsk, buf, count, pos);
 367         put_task_struct(tsk);
 368         if (ret > 0)
 369                 *pos += ret;
 370         return ret;
 371 }
 372
 373 static const struct file_operations proc_pid_cmdline_ops = {
 374         .read   = proc_pid_cmdline_read,
 375         .llseek = generic_file_llseek,
 376 };
 377
 378 #ifdef CONFIG_KALLSYMS
 379 /*
 380  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 381  * Returns the resolved symbol.  If that fails, simply return the address.
 382  */
 383 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 384                           struct pid *pid, struct task_struct *task)
 385 {
 386         unsigned long wchan;
 387         char symname[KSYM_NAME_LEN];
 388
 389         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
 390                 goto print0;
 391
 392         wchan = get_wchan(task);
 393         if (wchan && !lookup_symbol_name(wchan, symname)) {
 394                 seq_puts(m, symname);
 395                 return 0;
 396         }
 397
 398 print0:
 399         seq_putc(m, '0');
 400         return 0;
 401 }
 402 #endif /* CONFIG_KALLSYMS */
 403
 404 static int lock_trace(struct task_struct *task)
 405 {
 406         int err = down_read_killable(&task->signal->exec_update_lock);
 407         if (err)
 408                 return err;
 409         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 410                 up_read(&task->signal->exec_update_lock);
 411                 return -EPERM;
 412         }
 413         return 0;
 414 }
 415
 416 static void unlock_trace(struct task_struct *task)
 417 {
 418         up_read(&task->signal->exec_update_lock);
 419 }
 420
 421 #ifdef CONFIG_STACKTRACE
 422
 423 #define MAX_STACK_TRACE_DEPTH   64
 424
 425 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 426                           struct pid *pid, struct task_struct *task)
 427 {
 428         unsigned long *entries;
 429         int err;
 430
 431         /*
 432          * The ability to racily run the kernel stack unwinder on a running task
 433          * and then observe the unwinder output is scary; while it is useful for
 434          * debugging kernel issues, it can also allow an attacker to leak kernel
 435          * stack contents.
 436          * Doing this in a manner that is at least safe from races would require
 437          * some work to ensure that the remote task can not be scheduled; and
 438          * even then, this would still expose the unwinder as local attack
 439          * surface.
 440          * Therefore, this interface is restricted to root.
 441          */
 442         if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN))
 443                 return -EACCES;
 444
 445         entries = kmalloc_array(MAX_STACK_TRACE_DEPTH, sizeof(*entries),
 446                                 GFP_KERNEL);
 447         if (!entries)
 448                 return -ENOMEM;
 449
 450         err = lock_trace(task);
 451         if (!err) {
 452                 unsigned int i, nr_entries;
 453
 454                 nr_entries = stack_trace_save_tsk(task, entries,
 455                                                   MAX_STACK_TRACE_DEPTH, 0);
 456
 457                 for (i = 0; i < nr_entries; i++) {
 458                         seq_printf(m, "[<0>] %pB\n", (void *)entries[i]);
 459                 }
 460
 461                 unlock_trace(task);
 462         }
 463         kfree(entries);
 464
 465         return err;
 466 }
 467 #endif
 468
 469 #ifdef CONFIG_SCHED_INFO
 470 /*
 471  * Provides /proc/PID/schedstat
 472  */
 473 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 474                               struct pid *pid, struct task_struct *task)
 475 {
 476         if (unlikely(!sched_info_on()))
 477                 seq_puts(m, "0 0 0\n");
 478         else
 479                 seq_printf(m, "%llu %llu %lu\n",
 480                    (unsigned long long)task->se.sum_exec_runtime,
 481                    (unsigned long long)task->sched_info.run_delay,
 482                    task->sched_info.pcount);
 483
 484         return 0;
 485 }
 486 #endif
 487
 488 #ifdef CONFIG_LATENCYTOP
 489 static int lstats_show_proc(struct seq_file *m, void *v)
 490 {
 491         int i;
 492         struct inode *inode = m->private;
 493         struct task_struct *task = get_proc_task(inode);
 494
 495         if (!task)
 496                 return -ESRCH;
 497         seq_puts(m, "Latency Top version : v0.1\n");
 498         for (i = 0; i < LT_SAVECOUNT; i++) {
 499                 struct latency_record *lr = &task->latency_record[i];
 500                 if (lr->backtrace[0]) {
 501                         int q;
 502                         seq_printf(m, "%i %li %li",
 503                                    lr->count, lr->time, lr->max);
 504                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 505                                 unsigned long bt = lr->backtrace[q];
 506
 507                                 if (!bt)
 508                                         break;
 509                                 seq_printf(m, " %ps", (void *)bt);
 510                         }
 511                         seq_putc(m, '\n');
 512                 }
 513
 514         }
 515         put_task_struct(task);
 516         return 0;
 517 }
 518
 519 static int lstats_open(struct inode *inode, struct file *file)
 520 {
 521         return single_open(file, lstats_show_proc, inode);
 522 }
 523
 524 static ssize_t lstats_write(struct file *file, const char __user *buf,
 525                             size_t count, loff_t *offs)
 526 {
 527         struct task_struct *task = get_proc_task(file_inode(file));
 528
 529         if (!task)
 530                 return -ESRCH;
 531         clear_tsk_latency_tracing(task);
 532         put_task_struct(task);
 533
 534         return count;
 535 }
 536
 537 static const struct file_operations proc_lstats_operations = {
 538         .open           = lstats_open,
 539         .read           = seq_read,
 540         .write          = lstats_write,
 541         .llseek         = seq_lseek,
 542         .release        = single_release,
 543 };
 544
 545 #endif
 546
 547 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 548                           struct pid *pid, struct task_struct *task)
 549 {
 550         unsigned long totalpages = totalram_pages() + total_swap_pages;
 551         unsigned long points = 0;
 552         long badness;
 553
 554         badness = oom_badness(task, totalpages);
 555         /*
 556          * Special case OOM_SCORE_ADJ_MIN for all others scale the
 557          * badness value into [0, 2000] range which we have been
 558          * exporting for a long time so userspace might depend on it.
 559          */
 560         if (badness != LONG_MIN)
 561                 points = (1000 + badness * 1000 / (long)totalpages) * 2 / 3;
 562
 563         seq_printf(m, "%lu\n", points);
 564
 565         return 0;
 566 }
 567
 568 struct limit_names {
 569         const char *name;
 570         const char *unit;
 571 };
 572
 573 static const struct limit_names lnames[RLIM_NLIMITS] = {
 574         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 575         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 576         [RLIMIT_DATA] = {"Max data size", "bytes"},
 577         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 578         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 579         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 580         [RLIMIT_NPROC] = {"Max processes", "processes"},
 581         [RLIMIT_NOFILE] = {"Max open files", "files"},
 582         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 583         [RLIMIT_AS] = {"Max address space", "bytes"},
 584         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 585         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 586         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 587         [RLIMIT_NICE] = {"Max nice priority", NULL},
 588         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 589         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 590 };
 591
 592 /* Display limits for a process */
 593 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 594                            struct pid *pid, struct task_struct *task)
 595 {
 596         unsigned int i;
 597         unsigned long flags;
 598
 599         struct rlimit rlim[RLIM_NLIMITS];
 600
 601         if (!lock_task_sighand(task, &flags))
 602                 return 0;
 603         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 604         unlock_task_sighand(task, &flags);
 605
 606         /*
 607          * print the file header
 608          */
 609         seq_puts(m, "Limit                     "
 610                 "Soft Limit           "
 611                 "Hard Limit           "
 612                 "Units     \n");
 613
 614         for (i = 0; i < RLIM_NLIMITS; i++) {
 615                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 616                         seq_printf(m, "%-25s %-20s ",
 617                                    lnames[i].name, "unlimited");
 618                 else
 619                         seq_printf(m, "%-25s %-20lu ",
 620                                    lnames[i].name, rlim[i].rlim_cur);
 621
 622                 if (rlim[i].rlim_max == RLIM_INFINITY)
 623                         seq_printf(m, "%-20s ", "unlimited");
 624                 else
 625                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 626
 627                 if (lnames[i].unit)
 628                         seq_printf(m, "%-10s\n", lnames[i].unit);
 629                 else
 630                         seq_putc(m, '\n');
 631         }
 632
 633         return 0;
 634 }
 635
 636 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 637 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 638                             struct pid *pid, struct task_struct *task)
 639 {
 640         struct syscall_info info;
 641         u64 *args = &info.data.args[0];
 642         int res;
 643
 644         res = lock_trace(task);
 645         if (res)
 646                 return res;
 647
 648         if (task_current_syscall(task, &info))
 649                 seq_puts(m, "running\n");
 650         else if (info.data.nr < 0)
 651                 seq_printf(m, "%d 0x%llx 0x%llx\n",
 652                            info.data.nr, info.sp, info.data.instruction_pointer);
 653         else
 654                 seq_printf(m,
 655                        "%d 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx 0x%llx\n",
 656                        info.data.nr,
 657                        args[0], args[1], args[2], args[3], args[4], args[5],
 658                        info.sp, info.data.instruction_pointer);
 659         unlock_trace(task);
 660
 661         return 0;
 662 }
 663 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 664
 665 /************************************************************************/
 666 /*                       Here the fs part begins                        */
 667 /************************************************************************/
 668
 669 /* permission checks */
 670 static int proc_fd_access_allowed(struct inode *inode)
 671 {
 672         struct task_struct *task;
 673         int allowed = 0;
 674         /* Allow access to a task's file descriptors if it is us or we
 675          * may use ptrace attach to the process and find out that
 676          * information.
 677          */
 678         task = get_proc_task(inode);
 679         if (task) {
 680                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 681                 put_task_struct(task);
 682         }
 683         return allowed;
 684 }
 685
 686 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 687 {
 688         int error;
 689         struct inode *inode = d_inode(dentry);
 690
 691         if (attr->ia_valid & ATTR_MODE)
 692                 return -EPERM;
 693
 694         error = setattr_prepare(dentry, attr);
 695         if (error)
 696                 return error;
 697
 698         setattr_copy(inode, attr);
 699         mark_inode_dirty(inode);
 700         return 0;
 701 }
 702
 703 /*
 704  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 705  * or euid/egid (for hide_pid_min=2)?
 706  */
 707 static bool has_pid_permissions(struct pid_namespace *pid,
 708                                  struct task_struct *task,
 709                                  int hide_pid_min)
 710 {
 711         if (pid->hide_pid < hide_pid_min)
 712                 return true;
 713         if (in_group_p(pid->pid_gid))
 714                 return true;
 715         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 716 }
 717
 718
 719 static int proc_pid_permission(struct inode *inode, int mask)
 720 {
 721         struct pid_namespace *pid = proc_pid_ns(inode);
 722         struct task_struct *task;
 723         bool has_perms;
 724
 725         task = get_proc_task(inode);
 726         if (!task)
 727                 return -ESRCH;
 728         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 729         put_task_struct(task);
 730
 731         if (!has_perms) {
 732                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 733                         /*
 734                          * Let's make getdents(), stat(), and open()
 735                          * consistent with each other.  If a process
 736                          * may not stat() a file, it shouldn't be seen
 737                          * in procfs at all.
 738                          */
 739                         return -ENOENT;
 740                 }
 741
 742                 return -EPERM;
 743         }
 744         return generic_permission(inode, mask);
 745 }
 746
 747
 748
 749 static const struct inode_operations proc_def_inode_operations = {
 750         .setattr        = proc_setattr,
 751 };
 752
 753 static int proc_single_show(struct seq_file *m, void *v)
 754 {
 755         struct inode *inode = m->private;
 756         struct pid_namespace *ns = proc_pid_ns(inode);
 757         struct pid *pid = proc_pid(inode);
 758         struct task_struct *task;
 759         int ret;
 760
 761         task = get_pid_task(pid, PIDTYPE_PID);
 762         if (!task)
 763                 return -ESRCH;
 764
 765         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 766
 767         put_task_struct(task);
 768         return ret;
 769 }
 770
 771 static int proc_single_open(struct inode *inode, struct file *filp)
 772 {
 773         return single_open(filp, proc_single_show, inode);
 774 }
 775
 776 static const struct file_operations proc_single_file_operations = {
 777         .open           = proc_single_open,
 778         .read           = seq_read,
 779         .llseek         = seq_lseek,
 780         .release        = single_release,
 781 };
 782
 783
 784 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 785 {
 786         struct task_struct *task = get_proc_task(inode);
 787         struct mm_struct *mm = ERR_PTR(-ESRCH);
 788
 789         if (task) {
 790                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 791                 put_task_struct(task);
 792
 793                 if (!IS_ERR_OR_NULL(mm)) {
 794                         /* ensure this mm_struct can't be freed */
 795                         mmgrab(mm);
 796                         /* but do not pin its memory */
 797                         mmput(mm);
 798                 }
 799         }
 800
 801         return mm;
 802 }
 803
 804 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 805 {
 806         struct mm_struct *mm = proc_mem_open(inode, mode);
 807
 808         if (IS_ERR(mm))
 809                 return PTR_ERR(mm);
 810
 811         file->private_data = mm;
 812         return 0;
 813 }
 814
 815 static int mem_open(struct inode *inode, struct file *file)
 816 {
 817         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 818
 819         /* OK to pass negative loff_t, we can catch out-of-range */
 820         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 821
 822         return ret;
 823 }
 824
 825 static ssize_t mem_rw(struct file *file, char __user *buf,
 826                         size_t count, loff_t *ppos, int write)
 827 {
 828         struct mm_struct *mm = file->private_data;
 829         unsigned long addr = *ppos;
 830         ssize_t copied;
 831         char *page;
 832         unsigned int flags;
 833
 834         if (!mm)
 835                 return 0;
 836
 837         page = (char *)__get_free_page(GFP_KERNEL);
 838         if (!page)
 839                 return -ENOMEM;
 840
 841         copied = 0;
 842         if (!mmget_not_zero(mm))
 843                 goto free;
 844
 845         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 846
 847         while (count > 0) {
 848                 size_t this_len = min_t(size_t, count, PAGE_SIZE);
 849
 850                 if (write && copy_from_user(page, buf, this_len)) {
 851                         copied = -EFAULT;
 852                         break;
 853                 }
 854
 855                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 856                 if (!this_len) {
 857                         if (!copied)
 858                                 copied = -EIO;
 859                         break;
 860                 }
 861
 862                 if (!write && copy_to_user(buf, page, this_len)) {
 863                         copied = -EFAULT;
 864                         break;
 865                 }
 866
 867                 buf += this_len;
 868                 addr += this_len;
 869                 copied += this_len;
 870                 count -= this_len;
 871         }
 872         *ppos = addr;
 873
 874         mmput(mm);
 875 free:
 876         free_page((unsigned long) page);
 877         return copied;
 878 }
 879
 880 static ssize_t mem_read(struct file *file, char __user *buf,
 881                         size_t count, loff_t *ppos)
 882 {
 883         return mem_rw(file, buf, count, ppos, 0);
 884 }
 885
 886 static ssize_t mem_write(struct file *file, const char __user *buf,
 887                          size_t count, loff_t *ppos)
 888 {
 889         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 890 }
 891
 892 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 893 {
 894         switch (orig) {
 895         case 0:
 896                 file->f_pos = offset;
 897                 break;
 898         case 1:
 899                 file->f_pos += offset;
 900                 break;
 901         default:
 902                 return -EINVAL;
 903         }
 904         force_successful_syscall_return();
 905         return file->f_pos;
 906 }
 907
 908 static int mem_release(struct inode *inode, struct file *file)
 909 {
 910         struct mm_struct *mm = file->private_data;
 911         if (mm)
 912                 mmdrop(mm);
 913         return 0;
 914 }
 915
 916 static const struct file_operations proc_mem_operations = {
 917         .llseek         = mem_lseek,
 918         .read           = mem_read,
 919         .write          = mem_write,
 920         .open           = mem_open,
 921         .release        = mem_release,
 922 };
 923
 924 static int environ_open(struct inode *inode, struct file *file)
 925 {
 926         return __mem_open(inode, file, PTRACE_MODE_READ);
 927 }
 928
 929 static ssize_t environ_read(struct file *file, char __user *buf,
 930                         size_t count, loff_t *ppos)
 931 {
 932         char *page;
 933         unsigned long src = *ppos;
 934         int ret = 0;
 935         struct mm_struct *mm = file->private_data;
 936         unsigned long env_start, env_end;
 937
 938         /* Ensure the process spawned far enough to have an environment. */
 939         if (!mm || !mm->env_end)
 940                 return 0;
 941
 942         page = (char *)__get_free_page(GFP_KERNEL);
 943         if (!page)
 944                 return -ENOMEM;
 945
 946         ret = 0;
 947         if (!mmget_not_zero(mm))
 948                 goto free;
 949
 950         spin_lock(&mm->arg_lock);
 951         env_start = mm->env_start;
 952         env_end = mm->env_end;
 953         spin_unlock(&mm->arg_lock);
 954
 955         while (count > 0) {
 956                 size_t this_len, max_len;
 957                 int retval;
 958
 959                 if (src >= (env_end - env_start))
 960                         break;
 961
 962                 this_len = env_end - (env_start + src);
 963
 964                 max_len = min_t(size_t, PAGE_SIZE, count);
 965                 this_len = min(max_len, this_len);
 966
 967                 retval = access_remote_vm(mm, (env_start + src), page, this_len, FOLL_ANON);
 968
 969                 if (retval <= 0) {
 970                         ret = retval;
 971                         break;
 972                 }
 973
 974                 if (copy_to_user(buf, page, retval)) {
 975                         ret = -EFAULT;
 976                         break;
 977                 }
 978
 979                 ret += retval;
 980                 src += retval;
 981                 buf += retval;
 982                 count -= retval;
 983         }
 984         *ppos = src;
 985         mmput(mm);
 986
 987 free:
 988         free_page((unsigned long) page);
 989         return ret;
 990 }
 991
 992 static const struct file_operations proc_environ_operations = {
 993         .open           = environ_open,
 994         .read           = environ_read,
 995         .llseek         = generic_file_llseek,
 996         .release        = mem_release,
 997 };
 998
 999 static int auxv_open(struct inode *inode, struct file *file)
1000 {
1001         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1002 }
1003
1004 static ssize_t auxv_read(struct file *file, char __user *buf,
1005                         size_t count, loff_t *ppos)
1006 {
1007         struct mm_struct *mm = file->private_data;
1008         unsigned int nwords = 0;
1009
1010         if (!mm)
1011                 return 0;
1012         do {
1013                 nwords += 2;
1014         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1015         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1016                                        nwords * sizeof(mm->saved_auxv[0]));
1017 }
1018
1019 static const struct file_operations proc_auxv_operations = {
1020         .open           = auxv_open,
1021         .read           = auxv_read,
1022         .llseek         = generic_file_llseek,
1023         .release        = mem_release,
1024 };
1025
1026 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1027                             loff_t *ppos)
1028 {
1029         struct task_struct *task = get_proc_task(file_inode(file));
1030         char buffer[PROC_NUMBUF];
1031         int oom_adj = OOM_ADJUST_MIN;
1032         size_t len;
1033
1034         if (!task)
1035                 return -ESRCH;
1036         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1037                 oom_adj = OOM_ADJUST_MAX;
1038         else
1039                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1040                           OOM_SCORE_ADJ_MAX;
1041         put_task_struct(task);
1042         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1043         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1044 }
1045
1046 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1047 {
1048         struct mm_struct *mm = NULL;
1049         struct task_struct *task;
1050         int err = 0;
1051
1052         task = get_proc_task(file_inode(file));
1053         if (!task)
1054                 return -ESRCH;
1055
1056         mutex_lock(&oom_adj_mutex);
1057         if (legacy) {
1058                 if (oom_adj < task->signal->oom_score_adj &&
1059                                 !capable(CAP_SYS_RESOURCE)) {
1060                         err = -EACCES;
1061                         goto err_unlock;
1062                 }
1063                 /*
1064                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1065                  * /proc/pid/oom_score_adj instead.
1066                  */
1067                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1068                           current->comm, task_pid_nr(current), task_pid_nr(task),
1069                           task_pid_nr(task));
1070         } else {
1071                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1072                                 !capable(CAP_SYS_RESOURCE)) {
1073                         err = -EACCES;
1074                         goto err_unlock;
1075                 }
1076         }
1077
1078         /*
1079          * Make sure we will check other processes sharing the mm if this is
1080          * not vfrok which wants its own oom_score_adj.
1081          * pin the mm so it doesn't go away and get reused after task_unlock
1082          */
1083         if (!task->vfork_done) {
1084                 struct task_struct *p = find_lock_task_mm(task);
1085
1086                 if (p) {
1087                         if (test_bit(MMF_MULTIPROCESS, &p->mm->flags)) {
1088                                 mm = p->mm;
1089                                 mmgrab(mm);
1090                         }
1091                         task_unlock(p);
1092                 }
1093         }
1094
1095         task->signal->oom_score_adj = oom_adj;
1096         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1097                 task->signal->oom_score_adj_min = (short)oom_adj;
1098         trace_oom_score_adj_update(task);
1099
1100         if (mm) {
1101                 struct task_struct *p;
1102
1103                 rcu_read_lock();
1104                 for_each_process(p) {
1105                         if (same_thread_group(task, p))
1106                                 continue;
1107
1108                         /* do not touch kernel threads or the global init */
1109                         if (p->flags & PF_KTHREAD || is_global_init(p))
1110                                 continue;
1111
1112                         task_lock(p);
1113                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1114                                 p->signal->oom_score_adj = oom_adj;
1115                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1116                                         p->signal->oom_score_adj_min = (short)oom_adj;
1117                         }
1118                         task_unlock(p);
1119                 }
1120                 rcu_read_unlock();
1121                 mmdrop(mm);
1122         }
1123 err_unlock:
1124         mutex_unlock(&oom_adj_mutex);
1125         put_task_struct(task);
1126         return err;
1127 }
1128
1129 /*
1130  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1131  * kernels.  The effective policy is defined by oom_score_adj, which has a
1132  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1133  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1134  * Processes that become oom disabled via oom_adj will still be oom disabled
1135  * with this implementation.
1136  *
1137  * oom_adj cannot be removed since existing userspace binaries use it.
1138  */
1139 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1140                              size_t count, loff_t *ppos)
1141 {
1142         char buffer[PROC_NUMBUF];
1143         int oom_adj;
1144         int err;
1145
1146         memset(buffer, 0, sizeof(buffer));
1147         if (count > sizeof(buffer) - 1)
1148                 count = sizeof(buffer) - 1;
1149         if (copy_from_user(buffer, buf, count)) {
1150                 err = -EFAULT;
1151                 goto out;
1152         }
1153
1154         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1155         if (err)
1156                 goto out;
1157         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1158              oom_adj != OOM_DISABLE) {
1159                 err = -EINVAL;
1160                 goto out;
1161         }
1162
1163         /*
1164          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1165          * value is always attainable.
1166          */
1167         if (oom_adj == OOM_ADJUST_MAX)
1168                 oom_adj = OOM_SCORE_ADJ_MAX;
1169         else
1170                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1171
1172         err = __set_oom_adj(file, oom_adj, true);
1173 out:
1174         return err < 0 ? err : count;
1175 }
1176
1177 static const struct file_operations proc_oom_adj_operations = {
1178         .read           = oom_adj_read,
1179         .write          = oom_adj_write,
1180         .llseek         = generic_file_llseek,
1181 };
1182
1183 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1184                                         size_t count, loff_t *ppos)
1185 {
1186         struct task_struct *task = get_proc_task(file_inode(file));
1187         char buffer[PROC_NUMBUF];
1188         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1189         size_t len;
1190
1191         if (!task)
1192                 return -ESRCH;
1193         oom_score_adj = task->signal->oom_score_adj;
1194         put_task_struct(task);
1195         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1196         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1197 }
1198
1199 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1200                                         size_t count, loff_t *ppos)
1201 {
1202         char buffer[PROC_NUMBUF];
1203         int oom_score_adj;
1204         int err;
1205
1206         memset(buffer, 0, sizeof(buffer));
1207         if (count > sizeof(buffer) - 1)
1208                 count = sizeof(buffer) - 1;
1209         if (copy_from_user(buffer, buf, count)) {
1210                 err = -EFAULT;
1211                 goto out;
1212         }
1213
1214         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1215         if (err)
1216                 goto out;
1217         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1218                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1219                 err = -EINVAL;
1220                 goto out;
1221         }
1222
1223         err = __set_oom_adj(file, oom_score_adj, false);
1224 out:
1225         return err < 0 ? err : count;
1226 }
1227
1228 static const struct file_operations proc_oom_score_adj_operations = {
1229         .read           = oom_score_adj_read,
1230         .write          = oom_score_adj_write,
1231         .llseek         = default_llseek,
1232 };
1233
1234 #ifdef CONFIG_AUDIT
1235 #define TMPBUFLEN 11
1236 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1237                                   size_t count, loff_t *ppos)
1238 {
1239         struct inode * inode = file_inode(file);
1240         struct task_struct *task = get_proc_task(inode);
1241         ssize_t length;
1242         char tmpbuf[TMPBUFLEN];
1243
1244         if (!task)
1245                 return -ESRCH;
1246         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1247                            from_kuid(file->f_cred->user_ns,
1248                                      audit_get_loginuid(task)));
1249         put_task_struct(task);
1250         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1251 }
1252
1253 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1254                                    size_t count, loff_t *ppos)
1255 {
1256         struct inode * inode = file_inode(file);
1257         uid_t loginuid;
1258         kuid_t kloginuid;
1259         int rv;
1260
1261         rcu_read_lock();
1262         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1263                 rcu_read_unlock();
1264                 return -EPERM;
1265         }
1266         rcu_read_unlock();
1267
1268         if (*ppos != 0) {
1269                 /* No partial writes. */
1270                 return -EINVAL;
1271         }
1272
1273         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1274         if (rv < 0)
1275                 return rv;
1276
1277         /* is userspace tring to explicitly UNSET the loginuid? */
1278         if (loginuid == AUDIT_UID_UNSET) {
1279                 kloginuid = INVALID_UID;
1280         } else {
1281                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1282                 if (!uid_valid(kloginuid))
1283                         return -EINVAL;
1284         }
1285
1286         rv = audit_set_loginuid(kloginuid);
1287         if (rv < 0)
1288                 return rv;
1289         return count;
1290 }
1291
1292 static const struct file_operations proc_loginuid_operations = {
1293         .read           = proc_loginuid_read,
1294         .write          = proc_loginuid_write,
1295         .llseek         = generic_file_llseek,
1296 };
1297
1298 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1299                                   size_t count, loff_t *ppos)
1300 {
1301         struct inode * inode = file_inode(file);
1302         struct task_struct *task = get_proc_task(inode);
1303         ssize_t length;
1304         char tmpbuf[TMPBUFLEN];
1305
1306         if (!task)
1307                 return -ESRCH;
1308         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1309                                 audit_get_sessionid(task));
1310         put_task_struct(task);
1311         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1312 }
1313
1314 static const struct file_operations proc_sessionid_operations = {
1315         .read           = proc_sessionid_read,
1316         .llseek         = generic_file_llseek,
1317 };
1318 #endif
1319
1320 #ifdef CONFIG_FAULT_INJECTION
1321 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1322                                       size_t count, loff_t *ppos)
1323 {
1324         struct task_struct *task = get_proc_task(file_inode(file));
1325         char buffer[PROC_NUMBUF];
1326         size_t len;
1327         int make_it_fail;
1328
1329         if (!task)
1330                 return -ESRCH;
1331         make_it_fail = task->make_it_fail;
1332         put_task_struct(task);
1333
1334         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1335
1336         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1337 }
1338
1339 static ssize_t proc_fault_inject_write(struct file * file,
1340                         const char __user * buf, size_t count, loff_t *ppos)
1341 {
1342         struct task_struct *task;
1343         char buffer[PROC_NUMBUF];
1344         int make_it_fail;
1345         int rv;
1346
1347         if (!capable(CAP_SYS_RESOURCE))
1348                 return -EPERM;
1349         memset(buffer, 0, sizeof(buffer));
1350         if (count > sizeof(buffer) - 1)
1351                 count = sizeof(buffer) - 1;
1352         if (copy_from_user(buffer, buf, count))
1353                 return -EFAULT;
1354         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1355         if (rv < 0)
1356                 return rv;
1357         if (make_it_fail < 0 || make_it_fail > 1)
1358                 return -EINVAL;
1359
1360         task = get_proc_task(file_inode(file));
1361         if (!task)
1362                 return -ESRCH;
1363         task->make_it_fail = make_it_fail;
1364         put_task_struct(task);
1365
1366         return count;
1367 }
1368
1369 static const struct file_operations proc_fault_inject_operations = {
1370         .read           = proc_fault_inject_read,
1371         .write          = proc_fault_inject_write,
1372         .llseek         = generic_file_llseek,
1373 };
1374
1375 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1376                                    size_t count, loff_t *ppos)
1377 {
1378         struct task_struct *task;
1379         int err;
1380         unsigned int n;
1381
1382         err = kstrtouint_from_user(buf, count, 0, &n);
1383         if (err)
1384                 return err;
1385
1386         task = get_proc_task(file_inode(file));
1387         if (!task)
1388                 return -ESRCH;
1389         task->fail_nth = n;
1390         put_task_struct(task);
1391
1392         return count;
1393 }
1394
1395 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1396                                   size_t count, loff_t *ppos)
1397 {
1398         struct task_struct *task;
1399         char numbuf[PROC_NUMBUF];
1400         ssize_t len;
1401
1402         task = get_proc_task(file_inode(file));
1403         if (!task)
1404                 return -ESRCH;
1405         len = snprintf(numbuf, sizeof(numbuf), "%u\n", task->fail_nth);
1406         put_task_struct(task);
1407         return simple_read_from_buffer(buf, count, ppos, numbuf, len);
1408 }
1409
1410 static const struct file_operations proc_fail_nth_operations = {
1411         .read           = proc_fail_nth_read,
1412         .write          = proc_fail_nth_write,
1413 };
1414 #endif
1415
1416
1417 #ifdef CONFIG_SCHED_DEBUG
1418 /*
1419  * Print out various scheduling related per-task fields:
1420  */
1421 static int sched_show(struct seq_file *m, void *v)
1422 {
1423         struct inode *inode = m->private;
1424         struct pid_namespace *ns = proc_pid_ns(inode);
1425         struct task_struct *p;
1426
1427         p = get_proc_task(inode);
1428         if (!p)
1429                 return -ESRCH;
1430         proc_sched_show_task(p, ns, m);
1431
1432         put_task_struct(p);
1433
1434         return 0;
1435 }
1436
1437 static ssize_t
1438 sched_write(struct file *file, const char __user *buf,
1439             size_t count, loff_t *offset)
1440 {
1441         struct inode *inode = file_inode(file);
1442         struct task_struct *p;
1443
1444         p = get_proc_task(inode);
1445         if (!p)
1446                 return -ESRCH;
1447         proc_sched_set_task(p);
1448
1449         put_task_struct(p);
1450
1451         return count;
1452 }
1453
1454 static int sched_open(struct inode *inode, struct file *filp)
1455 {
1456         return single_open(filp, sched_show, inode);
1457 }
1458
1459 static const struct file_operations proc_pid_sched_operations = {
1460         .open           = sched_open,
1461         .read           = seq_read,
1462         .write          = sched_write,
1463         .llseek         = seq_lseek,
1464         .release        = single_release,
1465 };
1466
1467 #endif
1468
1469 #ifdef CONFIG_SCHED_AUTOGROUP
1470 /*
1471  * Print out autogroup related information:
1472  */
1473 static int sched_autogroup_show(struct seq_file *m, void *v)
1474 {
1475         struct inode *inode = m->private;
1476         struct task_struct *p;
1477
1478         p = get_proc_task(inode);
1479         if (!p)
1480                 return -ESRCH;
1481         proc_sched_autogroup_show_task(p, m);
1482
1483         put_task_struct(p);
1484
1485         return 0;
1486 }
1487
1488 static ssize_t
1489 sched_autogroup_write(struct file *file, const char __user *buf,
1490             size_t count, loff_t *offset)
1491 {
1492         struct inode *inode = file_inode(file);
1493         struct task_struct *p;
1494         char buffer[PROC_NUMBUF];
1495         int nice;
1496         int err;
1497
1498         memset(buffer, 0, sizeof(buffer));
1499         if (count > sizeof(buffer) - 1)
1500                 count = sizeof(buffer) - 1;
1501         if (copy_from_user(buffer, buf, count))
1502                 return -EFAULT;
1503
1504         err = kstrtoint(strstrip(buffer), 0, &nice);
1505         if (err < 0)
1506                 return err;
1507
1508         p = get_proc_task(inode);
1509         if (!p)
1510                 return -ESRCH;
1511
1512         err = proc_sched_autogroup_set_nice(p, nice);
1513         if (err)
1514                 count = err;
1515
1516         put_task_struct(p);
1517
1518         return count;
1519 }
1520
1521 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1522 {
1523         int ret;
1524
1525         ret = single_open(filp, sched_autogroup_show, NULL);
1526         if (!ret) {
1527                 struct seq_file *m = filp->private_data;
1528
1529                 m->private = inode;
1530         }
1531         return ret;
1532 }
1533
1534 static const struct file_operations proc_pid_sched_autogroup_operations = {
1535         .open           = sched_autogroup_open,
1536         .read           = seq_read,
1537         .write          = sched_autogroup_write,
1538         .llseek         = seq_lseek,
1539         .release        = single_release,
1540 };
1541
1542 #endif /* CONFIG_SCHED_AUTOGROUP */
1543
1544 static ssize_t comm_write(struct file *file, const char __user *buf,
1545                                 size_t count, loff_t *offset)
1546 {
1547         struct inode *inode = file_inode(file);
1548         struct task_struct *p;
1549         char buffer[TASK_COMM_LEN];
1550         const size_t maxlen = sizeof(buffer) - 1;
1551
1552         memset(buffer, 0, sizeof(buffer));
1553         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1554                 return -EFAULT;
1555
1556         p = get_proc_task(inode);
1557         if (!p)
1558                 return -ESRCH;
1559
1560         if (same_thread_group(current, p))
1561                 set_task_comm(p, buffer);
1562         else
1563                 count = -EINVAL;
1564
1565         put_task_struct(p);
1566
1567         return count;
1568 }
1569
1570 static int comm_show(struct seq_file *m, void *v)
1571 {
1572         struct inode *inode = m->private;
1573         struct task_struct *p;
1574
1575         p = get_proc_task(inode);
1576         if (!p)
1577                 return -ESRCH;
1578
1579         proc_task_name(m, p, false);
1580         seq_putc(m, '\n');
1581
1582         put_task_struct(p);
1583
1584         return 0;
1585 }
1586
1587 static int comm_open(struct inode *inode, struct file *filp)
1588 {
1589         return single_open(filp, comm_show, inode);
1590 }
1591
1592 static const struct file_operations proc_pid_set_comm_operations = {
1593         .open           = comm_open,
1594         .read           = seq_read,
1595         .write          = comm_write,
1596         .llseek         = seq_lseek,
1597         .release        = single_release,
1598 };
1599
1600 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1601 {
1602         struct task_struct *task;
1603         struct file *exe_file;
1604
1605         task = get_proc_task(d_inode(dentry));
1606         if (!task)
1607                 return -ENOENT;
1608         exe_file = get_task_exe_file(task);
1609         put_task_struct(task);
1610         if (exe_file) {
1611                 *exe_path = exe_file->f_path;
1612                 path_get(&exe_file->f_path);
1613                 fput(exe_file);
1614                 return 0;
1615         } else
1616                 return -ENOENT;
1617 }
1618
1619 static const char *proc_pid_get_link(struct dentry *dentry,
1620                                      struct inode *inode,
1621                                      struct delayed_call *done)
1622 {
1623         struct path path;
1624         int error = -EACCES;
1625
1626         if (!dentry)
1627                 return ERR_PTR(-ECHILD);
1628
1629         /* Are we allowed to snoop on the tasks file descriptors? */
1630         if (!proc_fd_access_allowed(inode))
1631                 goto out;
1632
1633         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1634         if (error)
1635                 goto out;
1636
1637         nd_jump_link(&path);
1638         return NULL;
1639 out:
1640         return ERR_PTR(error);
1641 }
1642
1643 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1644 {
1645         char *tmp = (char *)__get_free_page(GFP_KERNEL);
1646         char *pathname;
1647         int len;
1648
1649         if (!tmp)
1650                 return -ENOMEM;
1651
1652         pathname = d_path(path, tmp, PAGE_SIZE);
1653         len = PTR_ERR(pathname);
1654         if (IS_ERR(pathname))
1655                 goto out;
1656         len = tmp + PAGE_SIZE - 1 - pathname;
1657
1658         if (len > buflen)
1659                 len = buflen;
1660         if (copy_to_user(buffer, pathname, len))
1661                 len = -EFAULT;
1662  out:
1663         free_page((unsigned long)tmp);
1664         return len;
1665 }
1666
1667 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1668 {
1669         int error = -EACCES;
1670         struct inode *inode = d_inode(dentry);
1671         struct path path;
1672
1673         /* Are we allowed to snoop on the tasks file descriptors? */
1674         if (!proc_fd_access_allowed(inode))
1675                 goto out;
1676
1677         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1678         if (error)
1679                 goto out;
1680
1681         error = do_proc_readlink(&path, buffer, buflen);
1682         path_put(&path);
1683 out:
1684         return error;
1685 }
1686
1687 const struct inode_operations proc_pid_link_inode_operations = {
1688         .readlink       = proc_pid_readlink,
1689         .get_link       = proc_pid_get_link,
1690         .setattr        = proc_setattr,
1691 };
1692
1693
1694 /* building an inode */
1695
1696 void task_dump_owner(struct task_struct *task, umode_t mode,
1697                      kuid_t *ruid, kgid_t *rgid)
1698 {
1699         /* Depending on the state of dumpable compute who should own a
1700          * proc file for a task.
1701          */
1702         const struct cred *cred;
1703         kuid_t uid;
1704         kgid_t gid;
1705
1706         if (unlikely(task->flags & PF_KTHREAD)) {
1707                 *ruid = GLOBAL_ROOT_UID;
1708                 *rgid = GLOBAL_ROOT_GID;
1709                 return;
1710         }
1711
1712         /* Default to the tasks effective ownership */
1713         rcu_read_lock();
1714         cred = __task_cred(task);
1715         uid = cred->euid;
1716         gid = cred->egid;
1717         rcu_read_unlock();
1718
1719         /*
1720          * Before the /proc/pid/status file was created the only way to read
1721          * the effective uid of a /process was to stat /proc/pid.  Reading
1722          * /proc/pid/status is slow enough that procps and other packages
1723          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1724          * made this apply to all per process world readable and executable
1725          * directories.
1726          */
1727         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1728                 struct mm_struct *mm;
1729                 task_lock(task);
1730                 mm = task->mm;
1731                 /* Make non-dumpable tasks owned by some root */
1732                 if (mm) {
1733                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1734                                 struct user_namespace *user_ns = mm->user_ns;
1735
1736                                 uid = make_kuid(user_ns, 0);
1737                                 if (!uid_valid(uid))
1738                                         uid = GLOBAL_ROOT_UID;
1739
1740                                 gid = make_kgid(user_ns, 0);
1741                                 if (!gid_valid(gid))
1742                                         gid = GLOBAL_ROOT_GID;
1743                         }
1744                 } else {
1745                         uid = GLOBAL_ROOT_UID;
1746                         gid = GLOBAL_ROOT_GID;
1747                 }
1748                 task_unlock(task);
1749         }
1750         *ruid = uid;
1751         *rgid = gid;
1752 }
1753
1754 struct inode *proc_pid_make_inode(struct super_block * sb,
1755                                   struct task_struct *task, umode_t mode)
1756 {
1757         struct inode * inode;
1758         struct proc_inode *ei;
1759
1760         /* We need a new inode */
1761
1762         inode = new_inode(sb);
1763         if (!inode)
1764                 goto out;
1765
1766         /* Common stuff */
1767         ei = PROC_I(inode);
1768         inode->i_mode = mode;
1769         inode->i_ino = get_next_ino();
1770         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1771         inode->i_op = &proc_def_inode_operations;
1772
1773         /*
1774          * grab the reference to task.
1775          */
1776         ei->pid = get_task_pid(task, PIDTYPE_PID);
1777         if (!ei->pid)
1778                 goto out_unlock;
1779
1780         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1781         security_task_to_inode(task, inode);
1782
1783 out:
1784         return inode;
1785
1786 out_unlock:
1787         iput(inode);
1788         return NULL;
1789 }
1790
1791 int pid_getattr(const struct path *path, struct kstat *stat,
1792                 u32 request_mask, unsigned int query_flags)
1793 {
1794         struct inode *inode = d_inode(path->dentry);
1795         struct pid_namespace *pid = proc_pid_ns(inode);
1796         struct task_struct *task;
1797
1798         generic_fillattr(inode, stat);
1799
1800         stat->uid = GLOBAL_ROOT_UID;
1801         stat->gid = GLOBAL_ROOT_GID;
1802         rcu_read_lock();
1803         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1804         if (task) {
1805                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1806                         rcu_read_unlock();
1807                         /*
1808                          * This doesn't prevent learning whether PID exists,
1809                          * it only makes getattr() consistent with readdir().
1810                          */
1811                         return -ENOENT;
1812                 }
1813                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1814         }
1815         rcu_read_unlock();
1816         return 0;
1817 }
1818
1819 /* dentry stuff */
1820
1821 /*
1822  * Set <pid>/... inode ownership (can change due to setuid(), etc.)
1823  */
1824 void pid_update_inode(struct task_struct *task, struct inode *inode)
1825 {
1826         task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1827
1828         inode->i_mode &= ~(S_ISUID | S_ISGID);
1829         security_task_to_inode(task, inode);
1830 }
1831
1832 /*
1833  * Rewrite the inode's ownerships here because the owning task may have
1834  * performed a setuid(), etc.
1835  *
1836  */
1837 static int pid_revalidate(struct dentry *dentry, unsigned int flags)
1838 {
1839         struct inode *inode;
1840         struct task_struct *task;
1841
1842         if (flags & LOOKUP_RCU)
1843                 return -ECHILD;
1844
1845         inode = d_inode(dentry);
1846         task = get_proc_task(inode);
1847
1848         if (task) {
1849                 pid_update_inode(task, inode);
1850                 put_task_struct(task);
1851                 return 1;
1852         }
1853         return 0;
1854 }
1855
1856 static inline bool proc_inode_is_dead(struct inode *inode)
1857 {
1858         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1859 }
1860
1861 int pid_delete_dentry(const struct dentry *dentry)
1862 {
1863         /* Is the task we represent dead?
1864          * If so, then don't put the dentry on the lru list,
1865          * kill it immediately.
1866          */
1867         return proc_inode_is_dead(d_inode(dentry));
1868 }
1869
1870 const struct dentry_operations pid_dentry_operations =
1871 {
1872         .d_revalidate   = pid_revalidate,
1873         .d_delete       = pid_delete_dentry,
1874 };
1875
1876 /* Lookups */
1877
1878 /*
1879  * Fill a directory entry.
1880  *
1881  * If possible create the dcache entry and derive our inode number and
1882  * file type from dcache entry.
1883  *
1884  * Since all of the proc inode numbers are dynamically generated, the inode
1885  * numbers do not exist until the inode is cache.  This means creating the
1886  * the dcache entry in readdir is necessary to keep the inode numbers
1887  * reported by readdir in sync with the inode numbers reported
1888  * by stat.
1889  */
1890 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1891         const char *name, unsigned int len,
1892         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1893 {
1894         struct dentry *child, *dir = file->f_path.dentry;
1895         struct qstr qname = QSTR_INIT(name, len);
1896         struct inode *inode;
1897         unsigned type = DT_UNKNOWN;
1898         ino_t ino = 1;
1899
1900         child = d_hash_and_lookup(dir, &qname);
1901         if (!child) {
1902                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1903                 child = d_alloc_parallel(dir, &qname, &wq);
1904                 if (IS_ERR(child))
1905                         goto end_instantiate;
1906                 if (d_in_lookup(child)) {
1907                         struct dentry *res;
1908                         res = instantiate(child, task, ptr);
1909                         d_lookup_done(child);
1910                         if (unlikely(res)) {
1911                                 dput(child);
1912                                 child = res;
1913                                 if (IS_ERR(child))
1914                                         goto end_instantiate;
1915                         }
1916                 }
1917         }
1918         inode = d_inode(child);
1919         ino = inode->i_ino;
1920         type = inode->i_mode >> 12;
1921         dput(child);
1922 end_instantiate:
1923         return dir_emit(ctx, name, len, ino, type);
1924 }
1925
1926 /*
1927  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1928  * which represent vma start and end addresses.
1929  */
1930 static int dname_to_vma_addr(struct dentry *dentry,
1931                              unsigned long *start, unsigned long *end)
1932 {
1933         const char *str = dentry->d_name.name;
1934         unsigned long long sval, eval;
1935         unsigned int len;
1936
1937         if (str[0] == '0' && str[1] != '-')
1938                 return -EINVAL;
1939         len = _parse_integer(str, 16, &sval);
1940         if (len & KSTRTOX_OVERFLOW)
1941                 return -EINVAL;
1942         if (sval != (unsigned long)sval)
1943                 return -EINVAL;
1944         str += len;
1945
1946         if (*str != '-')
1947                 return -EINVAL;
1948         str++;
1949
1950         if (str[0] == '0' && str[1])
1951                 return -EINVAL;
1952         len = _parse_integer(str, 16, &eval);
1953         if (len & KSTRTOX_OVERFLOW)
1954                 return -EINVAL;
1955         if (eval != (unsigned long)eval)
1956                 return -EINVAL;
1957         str += len;
1958
1959         if (*str != '\0')
1960                 return -EINVAL;
1961
1962         *start = sval;
1963         *end = eval;
1964
1965         return 0;
1966 }
1967
1968 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1969 {
1970         unsigned long vm_start, vm_end;
1971         bool exact_vma_exists = false;
1972         struct mm_struct *mm = NULL;
1973         struct task_struct *task;
1974         struct inode *inode;
1975         int status = 0;
1976
1977         if (flags & LOOKUP_RCU)
1978                 return -ECHILD;
1979
1980         inode = d_inode(dentry);
1981         task = get_proc_task(inode);
1982         if (!task)
1983                 goto out_notask;
1984
1985         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1986         if (IS_ERR_OR_NULL(mm))
1987                 goto out;
1988
1989         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1990                 status = down_read_killable(&mm->mmap_sem);
1991                 if (!status) {
1992                         exact_vma_exists = !!find_exact_vma(mm, vm_start,
1993                                                             vm_end);
1994                         up_read(&mm->mmap_sem);
1995                 }
1996         }
1997
1998         mmput(mm);
1999
2000         if (exact_vma_exists) {
2001                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
2002
2003                 security_task_to_inode(task, inode);
2004                 status = 1;
2005         }
2006
2007 out:
2008         put_task_struct(task);
2009
2010 out_notask:
2011         return status;
2012 }
2013
2014 static const struct dentry_operations tid_map_files_dentry_operations = {
2015         .d_revalidate   = map_files_d_revalidate,
2016         .d_delete       = pid_delete_dentry,
2017 };
2018
2019 static int map_files_get_link(struct dentry *dentry, struct path *path)
2020 {
2021         unsigned long vm_start, vm_end;
2022         struct vm_area_struct *vma;
2023         struct task_struct *task;
2024         struct mm_struct *mm;
2025         int rc;
2026
2027         rc = -ENOENT;
2028         task = get_proc_task(d_inode(dentry));
2029         if (!task)
2030                 goto out;
2031
2032         mm = get_task_mm(task);
2033         put_task_struct(task);
2034         if (!mm)
2035                 goto out;
2036
2037         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2038         if (rc)
2039                 goto out_mmput;
2040
2041         rc = down_read_killable(&mm->mmap_sem);
2042         if (rc)
2043                 goto out_mmput;
2044
2045         rc = -ENOENT;
2046         vma = find_exact_vma(mm, vm_start, vm_end);
2047         if (vma && vma->vm_file) {
2048                 *path = vma->vm_file->f_path;
2049                 path_get(path);
2050                 rc = 0;
2051         }
2052         up_read(&mm->mmap_sem);
2053
2054 out_mmput:
2055         mmput(mm);
2056 out:
2057         return rc;
2058 }
2059
2060 struct map_files_info {
2061         unsigned long   start;
2062         unsigned long   end;
2063         fmode_t         mode;
2064 };
2065
2066 /*
2067  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2068  * symlinks may be used to bypass permissions on ancestor directories in the
2069  * path to the file in question.
2070  */
2071 static const char *
2072 proc_map_files_get_link(struct dentry *dentry,
2073                         struct inode *inode,
2074                         struct delayed_call *done)
2075 {
2076         if (!capable(CAP_SYS_ADMIN))
2077                 return ERR_PTR(-EPERM);
2078
2079         return proc_pid_get_link(dentry, inode, done);
2080 }
2081
2082 /*
2083  * Identical to proc_pid_link_inode_operations except for get_link()
2084  */
2085 static const struct inode_operations proc_map_files_link_inode_operations = {
2086         .readlink       = proc_pid_readlink,
2087         .get_link       = proc_map_files_get_link,
2088         .setattr        = proc_setattr,
2089 };
2090
2091 static struct dentry *
2092 proc_map_files_instantiate(struct dentry *dentry,
2093                            struct task_struct *task, const void *ptr)
2094 {
2095         fmode_t mode = (fmode_t)(unsigned long)ptr;
2096         struct proc_inode *ei;
2097         struct inode *inode;
2098
2099         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFLNK |
2100                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2101                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2102         if (!inode)
2103                 return ERR_PTR(-ENOENT);
2104
2105         ei = PROC_I(inode);
2106         ei->op.proc_get_link = map_files_get_link;
2107
2108         inode->i_op = &proc_map_files_link_inode_operations;
2109         inode->i_size = 64;
2110
2111         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2112         return d_splice_alias(inode, dentry);
2113 }
2114
2115 static struct dentry *proc_map_files_lookup(struct inode *dir,
2116                 struct dentry *dentry, unsigned int flags)
2117 {
2118         unsigned long vm_start, vm_end;
2119         struct vm_area_struct *vma;
2120         struct task_struct *task;
2121         struct dentry *result;
2122         struct mm_struct *mm;
2123
2124         result = ERR_PTR(-ENOENT);
2125         task = get_proc_task(dir);
2126         if (!task)
2127                 goto out;
2128
2129         result = ERR_PTR(-EACCES);
2130         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2131                 goto out_put_task;
2132
2133         result = ERR_PTR(-ENOENT);
2134         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2135                 goto out_put_task;
2136
2137         mm = get_task_mm(task);
2138         if (!mm)
2139                 goto out_put_task;
2140
2141         result = ERR_PTR(-EINTR);
2142         if (down_read_killable(&mm->mmap_sem))
2143                 goto out_put_mm;
2144
2145         result = ERR_PTR(-ENOENT);
2146         vma = find_exact_vma(mm, vm_start, vm_end);
2147         if (!vma)
2148                 goto out_no_vma;
2149
2150         if (vma->vm_file)
2151                 result = proc_map_files_instantiate(dentry, task,
2152                                 (void *)(unsigned long)vma->vm_file->f_mode);
2153
2154 out_no_vma:
2155         up_read(&mm->mmap_sem);
2156 out_put_mm:
2157         mmput(mm);
2158 out_put_task:
2159         put_task_struct(task);
2160 out:
2161         return result;
2162 }
2163
2164 static const struct inode_operations proc_map_files_inode_operations = {
2165         .lookup         = proc_map_files_lookup,
2166         .permission     = proc_fd_permission,
2167         .setattr        = proc_setattr,
2168 };
2169
2170 static int
2171 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2172 {
2173         struct vm_area_struct *vma;
2174         struct task_struct *task;
2175         struct mm_struct *mm;
2176         unsigned long nr_files, pos, i;
2177         GENRADIX(struct map_files_info) fa;
2178         struct map_files_info *p;
2179         int ret;
2180
2181         genradix_init(&fa);
2182
2183         ret = -ENOENT;
2184         task = get_proc_task(file_inode(file));
2185         if (!task)
2186                 goto out;
2187
2188         ret = -EACCES;
2189         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2190                 goto out_put_task;
2191
2192         ret = 0;
2193         if (!dir_emit_dots(file, ctx))
2194                 goto out_put_task;
2195
2196         mm = get_task_mm(task);
2197         if (!mm)
2198                 goto out_put_task;
2199
2200         ret = down_read_killable(&mm->mmap_sem);
2201         if (ret) {
2202                 mmput(mm);
2203                 goto out_put_task;
2204         }
2205
2206         nr_files = 0;
2207
2208         /*
2209          * We need two passes here:
2210          *
2211          *  1) Collect vmas of mapped files with mmap_sem taken
2212          *  2) Release mmap_sem and instantiate entries
2213          *
2214          * otherwise we get lockdep complained, since filldir()
2215          * routine might require mmap_sem taken in might_fault().
2216          */
2217
2218         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2219                 if (!vma->vm_file)
2220                         continue;
2221                 if (++pos <= ctx->pos)
2222                         continue;
2223
2224                 p = genradix_ptr_alloc(&fa, nr_files++, GFP_KERNEL);
2225                 if (!p) {
2226                         ret = -ENOMEM;
2227                         up_read(&mm->mmap_sem);
2228                         mmput(mm);
2229                         goto out_put_task;
2230                 }
2231
2232                 p->start = vma->vm_start;
2233                 p->end = vma->vm_end;
2234                 p->mode = vma->vm_file->f_mode;
2235         }
2236         up_read(&mm->mmap_sem);
2237         mmput(mm);
2238
2239         for (i = 0; i < nr_files; i++) {
2240                 char buf[4 * sizeof(long) + 2]; /* max: %lx-%lx\0 */
2241                 unsigned int len;
2242
2243                 p = genradix_ptr(&fa, i);
2244                 len = snprintf(buf, sizeof(buf), "%lx-%lx", p->start, p->end);
2245                 if (!proc_fill_cache(file, ctx,
2246                                       buf, len,
2247                                       proc_map_files_instantiate,
2248                                       task,
2249                                       (void *)(unsigned long)p->mode))
2250                         break;
2251                 ctx->pos++;
2252         }
2253
2254 out_put_task:
2255         put_task_struct(task);
2256 out:
2257         genradix_free(&fa);
2258         return ret;
2259 }
2260
2261 static const struct file_operations proc_map_files_operations = {
2262         .read           = generic_read_dir,
2263         .iterate_shared = proc_map_files_readdir,
2264         .llseek         = generic_file_llseek,
2265 };
2266
2267 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2268 struct timers_private {
2269         struct pid *pid;
2270         struct task_struct *task;
2271         struct sighand_struct *sighand;
2272         struct pid_namespace *ns;
2273         unsigned long flags;
2274 };
2275
2276 static void *timers_start(struct seq_file *m, loff_t *pos)
2277 {
2278         struct timers_private *tp = m->private;
2279
2280         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2281         if (!tp->task)
2282                 return ERR_PTR(-ESRCH);
2283
2284         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2285         if (!tp->sighand)
2286                 return ERR_PTR(-ESRCH);
2287
2288         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2289 }
2290
2291 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2292 {
2293         struct timers_private *tp = m->private;
2294         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2295 }
2296
2297 static void timers_stop(struct seq_file *m, void *v)
2298 {
2299         struct timers_private *tp = m->private;
2300
2301         if (tp->sighand) {
2302                 unlock_task_sighand(tp->task, &tp->flags);
2303                 tp->sighand = NULL;
2304         }
2305
2306         if (tp->task) {
2307                 put_task_struct(tp->task);
2308                 tp->task = NULL;
2309         }
2310 }
2311
2312 static int show_timer(struct seq_file *m, void *v)
2313 {
2314         struct k_itimer *timer;
2315         struct timers_private *tp = m->private;
2316         int notify;
2317         static const char * const nstr[] = {
2318                 [SIGEV_SIGNAL] = "signal",
2319                 [SIGEV_NONE] = "none",
2320                 [SIGEV_THREAD] = "thread",
2321         };
2322
2323         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2324         notify = timer->it_sigev_notify;
2325
2326         seq_printf(m, "ID: %d\n", timer->it_id);
2327         seq_printf(m, "signal: %d/%px\n",
2328                    timer->sigq->info.si_signo,
2329                    timer->sigq->info.si_value.sival_ptr);
2330         seq_printf(m, "notify: %s/%s.%d\n",
2331                    nstr[notify & ~SIGEV_THREAD_ID],
2332                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2333                    pid_nr_ns(timer->it_pid, tp->ns));
2334         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2335
2336         return 0;
2337 }
2338
2339 static const struct seq_operations proc_timers_seq_ops = {
2340         .start  = timers_start,
2341         .next   = timers_next,
2342         .stop   = timers_stop,
2343         .show   = show_timer,
2344 };
2345
2346 static int proc_timers_open(struct inode *inode, struct file *file)
2347 {
2348         struct timers_private *tp;
2349
2350         tp = __seq_open_private(file, &proc_timers_seq_ops,
2351                         sizeof(struct timers_private));
2352         if (!tp)
2353                 return -ENOMEM;
2354
2355         tp->pid = proc_pid(inode);
2356         tp->ns = proc_pid_ns(inode);
2357         return 0;
2358 }
2359
2360 static const struct file_operations proc_timers_operations = {
2361         .open           = proc_timers_open,
2362         .read           = seq_read,
2363         .llseek         = seq_lseek,
2364         .release        = seq_release_private,
2365 };
2366 #endif
2367
2368 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2369                                         size_t count, loff_t *offset)
2370 {
2371         struct inode *inode = file_inode(file);
2372         struct task_struct *p;
2373         u64 slack_ns;
2374         int err;
2375
2376         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2377         if (err < 0)
2378                 return err;
2379
2380         p = get_proc_task(inode);
2381         if (!p)
2382                 return -ESRCH;
2383
2384         if (p != current) {
2385                 rcu_read_lock();
2386                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2387                         rcu_read_unlock();
2388                         count = -EPERM;
2389                         goto out;
2390                 }
2391                 rcu_read_unlock();
2392
2393                 err = security_task_setscheduler(p);
2394                 if (err) {
2395                         count = err;
2396                         goto out;
2397                 }
2398         }
2399
2400         task_lock(p);
2401         if (slack_ns == 0)
2402                 p->timer_slack_ns = p->default_timer_slack_ns;
2403         else
2404                 p->timer_slack_ns = slack_ns;
2405         task_unlock(p);
2406
2407 out:
2408         put_task_struct(p);
2409
2410         return count;
2411 }
2412
2413 static int timerslack_ns_show(struct seq_file *m, void *v)
2414 {
2415         struct inode *inode = m->private;
2416         struct task_struct *p;
2417         int err = 0;
2418
2419         p = get_proc_task(inode);
2420         if (!p)
2421                 return -ESRCH;
2422
2423         if (p != current) {
2424                 rcu_read_lock();
2425                 if (!ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) {
2426                         rcu_read_unlock();
2427                         err = -EPERM;
2428                         goto out;
2429                 }
2430                 rcu_read_unlock();
2431
2432                 err = security_task_getscheduler(p);
2433                 if (err)
2434                         goto out;
2435         }
2436
2437         task_lock(p);
2438         seq_printf(m, "%llu\n", p->timer_slack_ns);
2439         task_unlock(p);
2440
2441 out:
2442         put_task_struct(p);
2443
2444         return err;
2445 }
2446
2447 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2448 {
2449         return single_open(filp, timerslack_ns_show, inode);
2450 }
2451
2452 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2453         .open           = timerslack_ns_open,
2454         .read           = seq_read,
2455         .write          = timerslack_ns_write,
2456         .llseek         = seq_lseek,
2457         .release        = single_release,
2458 };
2459
2460 static struct dentry *proc_pident_instantiate(struct dentry *dentry,
2461         struct task_struct *task, const void *ptr)
2462 {
2463         const struct pid_entry *p = ptr;
2464         struct inode *inode;
2465         struct proc_inode *ei;
2466
2467         inode = proc_pid_make_inode(dentry->d_sb, task, p->mode);
2468         if (!inode)
2469                 return ERR_PTR(-ENOENT);
2470
2471         ei = PROC_I(inode);
2472         if (S_ISDIR(inode->i_mode))
2473                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2474         if (p->iop)
2475                 inode->i_op = p->iop;
2476         if (p->fop)
2477                 inode->i_fop = p->fop;
2478         ei->op = p->op;
2479         pid_update_inode(task, inode);
2480         d_set_d_op(dentry, &pid_dentry_operations);
2481         return d_splice_alias(inode, dentry);
2482 }
2483
2484 static struct dentry *proc_pident_lookup(struct inode *dir,
2485                                          struct dentry *dentry,
2486                                          const struct pid_entry *p,
2487                                          const struct pid_entry *end)
2488 {
2489         struct task_struct *task = get_proc_task(dir);
2490         struct dentry *res = ERR_PTR(-ENOENT);
2491
2492         if (!task)
2493                 goto out_no_task;
2494
2495         /*
2496          * Yes, it does not scale. And it should not. Don't add
2497          * new entries into /proc/<tgid>/ without very good reasons.
2498          */
2499         for (; p < end; p++) {
2500                 if (p->len != dentry->d_name.len)
2501                         continue;
2502                 if (!memcmp(dentry->d_name.name, p->name, p->len)) {
2503                         res = proc_pident_instantiate(dentry, task, p);
2504                         break;
2505                 }
2506         }
2507         put_task_struct(task);
2508 out_no_task:
2509         return res;
2510 }
2511
2512 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2513                 const struct pid_entry *ents, unsigned int nents)
2514 {
2515         struct task_struct *task = get_proc_task(file_inode(file));
2516         const struct pid_entry *p;
2517
2518         if (!task)
2519                 return -ENOENT;
2520
2521         if (!dir_emit_dots(file, ctx))
2522                 goto out;
2523
2524         if (ctx->pos >= nents + 2)
2525                 goto out;
2526
2527         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2528                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2529                                 proc_pident_instantiate, task, p))
2530                         break;
2531                 ctx->pos++;
2532         }
2533 out:
2534         put_task_struct(task);
2535         return 0;
2536 }
2537
2538 #ifdef CONFIG_SECURITY
2539 static int proc_pid_attr_open(struct inode *inode, struct file *file)
2540 {
2541         file->private_data = NULL;
2542         __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
2543         return 0;
2544 }
2545
2546 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2547                                   size_t count, loff_t *ppos)
2548 {
2549         struct inode * inode = file_inode(file);
2550         char *p = NULL;
2551         ssize_t length;
2552         struct task_struct *task = get_proc_task(inode);
2553
2554         if (!task)
2555                 return -ESRCH;
2556
2557         length = security_getprocattr(task, PROC_I(inode)->op.lsm,
2558                                       (char*)file->f_path.dentry->d_name.name,
2559                                       &p);
2560         put_task_struct(task);
2561         if (length > 0)
2562                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2563         kfree(p);
2564         return length;
2565 }
2566
2567 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2568                                    size_t count, loff_t *ppos)
2569 {
2570         struct inode * inode = file_inode(file);
2571         struct task_struct *task;
2572         void *page;
2573         int rv;
2574
2575         /* A task may only write when it was the opener. */
2576         if (file->private_data != current->mm)
2577                 return -EPERM;
2578
2579         rcu_read_lock();
2580         task = pid_task(proc_pid(inode), PIDTYPE_PID);
2581         if (!task) {
2582                 rcu_read_unlock();
2583                 return -ESRCH;
2584         }
2585         /* A task may only write its own attributes. */
2586         if (current != task) {
2587                 rcu_read_unlock();
2588                 return -EACCES;
2589         }
2590         /* Prevent changes to overridden credentials. */
2591         if (current_cred() != current_real_cred()) {
2592                 rcu_read_unlock();
2593                 return -EBUSY;
2594         }
2595         rcu_read_unlock();
2596
2597         if (count > PAGE_SIZE)
2598                 count = PAGE_SIZE;
2599
2600         /* No partial writes. */
2601         if (*ppos != 0)
2602                 return -EINVAL;
2603
2604         page = memdup_user(buf, count);
2605         if (IS_ERR(page)) {
2606                 rv = PTR_ERR(page);
2607                 goto out;
2608         }
2609
2610         /* Guard against adverse ptrace interaction */
2611         rv = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2612         if (rv < 0)
2613                 goto out_free;
2614
2615         rv = security_setprocattr(PROC_I(inode)->op.lsm,
2616                                   file->f_path.dentry->d_name.name, page,
2617                                   count);
2618         mutex_unlock(&current->signal->cred_guard_mutex);
2619 out_free:
2620         kfree(page);
2621 out:
2622         return rv;
2623 }
2624
2625 static const struct file_operations proc_pid_attr_operations = {
2626         .open           = proc_pid_attr_open,
2627         .read           = proc_pid_attr_read,
2628         .write          = proc_pid_attr_write,
2629         .llseek         = generic_file_llseek,
2630         .release        = mem_release,
2631 };
2632
2633 #define LSM_DIR_OPS(LSM) \
2634 static int proc_##LSM##_attr_dir_iterate(struct file *filp, \
2635                              struct dir_context *ctx) \
2636 { \
2637         return proc_pident_readdir(filp, ctx, \
2638                                    LSM##_attr_dir_stuff, \
2639                                    ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2640 } \
2641 \
2642 static const struct file_operations proc_##LSM##_attr_dir_ops = { \
2643         .read           = generic_read_dir, \
2644         .iterate        = proc_##LSM##_attr_dir_iterate, \
2645         .llseek         = default_llseek, \
2646 }; \
2647 \
2648 static struct dentry *proc_##LSM##_attr_dir_lookup(struct inode *dir, \
2649                                 struct dentry *dentry, unsigned int flags) \
2650 { \
2651         return proc_pident_lookup(dir, dentry, \
2652                                   LSM##_attr_dir_stuff, \
2653                                   LSM##_attr_dir_stuff + ARRAY_SIZE(LSM##_attr_dir_stuff)); \
2654 } \
2655 \
2656 static const struct inode_operations proc_##LSM##_attr_dir_inode_ops = { \
2657         .lookup         = proc_##LSM##_attr_dir_lookup, \
2658         .getattr        = pid_getattr, \
2659         .setattr        = proc_setattr, \
2660 }
2661
2662 #ifdef CONFIG_SECURITY_SMACK
2663 static const struct pid_entry smack_attr_dir_stuff[] = {
2664         ATTR("smack", "current",        0666),
2665 };
2666 LSM_DIR_OPS(smack);
2667 #endif
2668
2669 static const struct pid_entry attr_dir_stuff[] = {
2670         ATTR(NULL, "current",           0666),
2671         ATTR(NULL, "prev",              0444),
2672         ATTR(NULL, "exec",              0666),
2673         ATTR(NULL, "fscreate",          0666),
2674         ATTR(NULL, "keycreate",         0666),
2675         ATTR(NULL, "sockcreate",        0666),
2676 #ifdef CONFIG_SECURITY_SMACK
2677         DIR("smack",                    0555,
2678             proc_smack_attr_dir_inode_ops, proc_smack_attr_dir_ops),
2679 #endif
2680 };
2681
2682 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2683 {
2684         return proc_pident_readdir(file, ctx,
2685                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2686 }
2687
2688 static const struct file_operations proc_attr_dir_operations = {
2689         .read           = generic_read_dir,
2690         .iterate_shared = proc_attr_dir_readdir,
2691         .llseek         = generic_file_llseek,
2692 };
2693
2694 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2695                                 struct dentry *dentry, unsigned int flags)
2696 {
2697         return proc_pident_lookup(dir, dentry,
2698                                   attr_dir_stuff,
2699                                   attr_dir_stuff + ARRAY_SIZE(attr_dir_stuff));
2700 }
2701
2702 static const struct inode_operations proc_attr_dir_inode_operations = {
2703         .lookup         = proc_attr_dir_lookup,
2704         .getattr        = pid_getattr,
2705         .setattr        = proc_setattr,
2706 };
2707
2708 #endif
2709
2710 #ifdef CONFIG_ELF_CORE
2711 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2712                                          size_t count, loff_t *ppos)
2713 {
2714         struct task_struct *task = get_proc_task(file_inode(file));
2715         struct mm_struct *mm;
2716         char buffer[PROC_NUMBUF];
2717         size_t len;
2718         int ret;
2719
2720         if (!task)
2721                 return -ESRCH;
2722
2723         ret = 0;
2724         mm = get_task_mm(task);
2725         if (mm) {
2726                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2727                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2728                                 MMF_DUMP_FILTER_SHIFT));
2729                 mmput(mm);
2730                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2731         }
2732
2733         put_task_struct(task);
2734
2735         return ret;
2736 }
2737
2738 static ssize_t proc_coredump_filter_write(struct file *file,
2739                                           const char __user *buf,
2740                                           size_t count,
2741                                           loff_t *ppos)
2742 {
2743         struct task_struct *task;
2744         struct mm_struct *mm;
2745         unsigned int val;
2746         int ret;
2747         int i;
2748         unsigned long mask;
2749
2750         ret = kstrtouint_from_user(buf, count, 0, &val);
2751         if (ret < 0)
2752                 return ret;
2753
2754         ret = -ESRCH;
2755         task = get_proc_task(file_inode(file));
2756         if (!task)
2757                 goto out_no_task;
2758
2759         mm = get_task_mm(task);
2760         if (!mm)
2761                 goto out_no_mm;
2762         ret = 0;
2763
2764         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2765                 if (val & mask)
2766                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2767                 else
2768                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2769         }
2770
2771         mmput(mm);
2772  out_no_mm:
2773         put_task_struct(task);
2774  out_no_task:
2775         if (ret < 0)
2776                 return ret;
2777         return count;
2778 }
2779
2780 static const struct file_operations proc_coredump_filter_operations = {
2781         .read           = proc_coredump_filter_read,
2782         .write          = proc_coredump_filter_write,
2783         .llseek         = generic_file_llseek,
2784 };
2785 #endif
2786
2787 #ifdef CONFIG_TASK_IO_ACCOUNTING
2788 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2789 {
2790         struct task_io_accounting acct = task->ioac;
2791         unsigned long flags;
2792         int result;
2793
2794         result = down_read_killable(&task->signal->exec_update_lock);
2795         if (result)
2796                 return result;
2797
2798         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2799                 result = -EACCES;
2800                 goto out_unlock;
2801         }
2802
2803         if (whole && lock_task_sighand(task, &flags)) {
2804                 struct task_struct *t = task;
2805
2806                 task_io_accounting_add(&acct, &task->signal->ioac);
2807                 while_each_thread(task, t)
2808                         task_io_accounting_add(&acct, &t->ioac);
2809
2810                 unlock_task_sighand(task, &flags);
2811         }
2812         seq_printf(m,
2813                    "rchar: %llu\n"
2814                    "wchar: %llu\n"
2815                    "syscr: %llu\n"
2816                    "syscw: %llu\n"
2817                    "read_bytes: %llu\n"
2818                    "write_bytes: %llu\n"
2819                    "cancelled_write_bytes: %llu\n",
2820                    (unsigned long long)acct.rchar,
2821                    (unsigned long long)acct.wchar,
2822                    (unsigned long long)acct.syscr,
2823                    (unsigned long long)acct.syscw,
2824                    (unsigned long long)acct.read_bytes,
2825                    (unsigned long long)acct.write_bytes,
2826                    (unsigned long long)acct.cancelled_write_bytes);
2827         result = 0;
2828
2829 out_unlock:
2830         up_read(&task->signal->exec_update_lock);
2831         return result;
2832 }
2833
2834 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2835                                   struct pid *pid, struct task_struct *task)
2836 {
2837         return do_io_accounting(task, m, 0);
2838 }
2839
2840 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2841                                    struct pid *pid, struct task_struct *task)
2842 {
2843         return do_io_accounting(task, m, 1);
2844 }
2845 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2846
2847 #ifdef CONFIG_USER_NS
2848 static int proc_id_map_open(struct inode *inode, struct file *file,
2849         const struct seq_operations *seq_ops)
2850 {
2851         struct user_namespace *ns = NULL;
2852         struct task_struct *task;
2853         struct seq_file *seq;
2854         int ret = -EINVAL;
2855
2856         task = get_proc_task(inode);
2857         if (task) {
2858                 rcu_read_lock();
2859                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2860                 rcu_read_unlock();
2861                 put_task_struct(task);
2862         }
2863         if (!ns)
2864                 goto err;
2865
2866         ret = seq_open(file, seq_ops);
2867         if (ret)
2868                 goto err_put_ns;
2869
2870         seq = file->private_data;
2871         seq->private = ns;
2872
2873         return 0;
2874 err_put_ns:
2875         put_user_ns(ns);
2876 err:
2877         return ret;
2878 }
2879
2880 static int proc_id_map_release(struct inode *inode, struct file *file)
2881 {
2882         struct seq_file *seq = file->private_data;
2883         struct user_namespace *ns = seq->private;
2884         put_user_ns(ns);
2885         return seq_release(inode, file);
2886 }
2887
2888 static int proc_uid_map_open(struct inode *inode, struct file *file)
2889 {
2890         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2891 }
2892
2893 static int proc_gid_map_open(struct inode *inode, struct file *file)
2894 {
2895         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2896 }
2897
2898 static int proc_projid_map_open(struct inode *inode, struct file *file)
2899 {
2900         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2901 }
2902
2903 static const struct file_operations proc_uid_map_operations = {
2904         .open           = proc_uid_map_open,
2905         .write          = proc_uid_map_write,
2906         .read           = seq_read,
2907         .llseek         = seq_lseek,
2908         .release        = proc_id_map_release,
2909 };
2910
2911 static const struct file_operations proc_gid_map_operations = {
2912         .open           = proc_gid_map_open,
2913         .write          = proc_gid_map_write,
2914         .read           = seq_read,
2915         .llseek         = seq_lseek,
2916         .release        = proc_id_map_release,
2917 };
2918
2919 static const struct file_operations proc_projid_map_operations = {
2920         .open           = proc_projid_map_open,
2921         .write          = proc_projid_map_write,
2922         .read           = seq_read,
2923         .llseek         = seq_lseek,
2924         .release        = proc_id_map_release,
2925 };
2926
2927 static int proc_setgroups_open(struct inode *inode, struct file *file)
2928 {
2929         struct user_namespace *ns = NULL;
2930         struct task_struct *task;
2931         int ret;
2932
2933         ret = -ESRCH;
2934         task = get_proc_task(inode);
2935         if (task) {
2936                 rcu_read_lock();
2937                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2938                 rcu_read_unlock();
2939                 put_task_struct(task);
2940         }
2941         if (!ns)
2942                 goto err;
2943
2944         if (file->f_mode & FMODE_WRITE) {
2945                 ret = -EACCES;
2946                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2947                         goto err_put_ns;
2948         }
2949
2950         ret = single_open(file, &proc_setgroups_show, ns);
2951         if (ret)
2952                 goto err_put_ns;
2953
2954         return 0;
2955 err_put_ns:
2956         put_user_ns(ns);
2957 err:
2958         return ret;
2959 }
2960
2961 static int proc_setgroups_release(struct inode *inode, struct file *file)
2962 {
2963         struct seq_file *seq = file->private_data;
2964         struct user_namespace *ns = seq->private;
2965         int ret = single_release(inode, file);
2966         put_user_ns(ns);
2967         return ret;
2968 }
2969
2970 static const struct file_operations proc_setgroups_operations = {
2971         .open           = proc_setgroups_open,
2972         .write          = proc_setgroups_write,
2973         .read           = seq_read,
2974         .llseek         = seq_lseek,
2975         .release        = proc_setgroups_release,
2976 };
2977 #endif /* CONFIG_USER_NS */
2978
2979 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2980                                 struct pid *pid, struct task_struct *task)
2981 {
2982         int err = lock_trace(task);
2983         if (!err) {
2984                 seq_printf(m, "%08x\n", task->personality);
2985                 unlock_trace(task);
2986         }
2987         return err;
2988 }
2989
2990 #ifdef CONFIG_LIVEPATCH
2991 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2992                                 struct pid *pid, struct task_struct *task)
2993 {
2994         seq_printf(m, "%d\n", task->patch_state);
2995         return 0;
2996 }
2997 #endif /* CONFIG_LIVEPATCH */
2998
2999 #ifdef CONFIG_STACKLEAK_METRICS
3000 static int proc_stack_depth(struct seq_file *m, struct pid_namespace *ns,
3001                                 struct pid *pid, struct task_struct *task)
3002 {
3003         unsigned long prev_depth = THREAD_SIZE -
3004                                 (task->prev_lowest_stack & (THREAD_SIZE - 1));
3005         unsigned long depth = THREAD_SIZE -
3006                                 (task->lowest_stack & (THREAD_SIZE - 1));
3007
3008         seq_printf(m, "previous stack depth: %lu\nstack depth: %lu\n",
3009                                                         prev_depth, depth);
3010         return 0;
3011 }
3012 #endif /* CONFIG_STACKLEAK_METRICS */
3013
3014 /*
3015  * Thread groups
3016  */
3017 static const struct file_operations proc_task_operations;
3018 static const struct inode_operations proc_task_inode_operations;
3019
3020 static const struct pid_entry tgid_base_stuff[] = {
3021         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3022         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3023         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3024         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3025         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3026 #ifdef CONFIG_NET
3027         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3028 #endif
3029         REG("environ",    S_IRUSR, proc_environ_operations),
3030         REG("auxv",       S_IRUSR, proc_auxv_operations),
3031         ONE("status",     S_IRUGO, proc_pid_status),
3032         ONE("personality", S_IRUSR, proc_pid_personality),
3033         ONE("limits",     S_IRUGO, proc_pid_limits),
3034 #ifdef CONFIG_SCHED_DEBUG
3035         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3036 #endif
3037 #ifdef CONFIG_SCHED_AUTOGROUP
3038         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3039 #endif
3040         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3041 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3042         ONE("syscall",    S_IRUSR, proc_pid_syscall),
3043 #endif
3044         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
3045         ONE("stat",       S_IRUGO, proc_tgid_stat),
3046         ONE("statm",      S_IRUGO, proc_pid_statm),
3047         REG("maps",       S_IRUGO, proc_pid_maps_operations),
3048 #ifdef CONFIG_NUMA
3049         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
3050 #endif
3051         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3052         LNK("cwd",        proc_cwd_link),
3053         LNK("root",       proc_root_link),
3054         LNK("exe",        proc_exe_link),
3055         REG("mounts",     S_IRUGO, proc_mounts_operations),
3056         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3057         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3058 #ifdef CONFIG_PROC_PAGE_MONITOR
3059         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3060         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
3061         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3062         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3063 #endif
3064 #ifdef CONFIG_SECURITY
3065         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3066 #endif
3067 #ifdef CONFIG_KALLSYMS
3068         ONE("wchan",      S_IRUGO, proc_pid_wchan),
3069 #endif
3070 #ifdef CONFIG_STACKTRACE
3071         ONE("stack",      S_IRUSR, proc_pid_stack),
3072 #endif
3073 #ifdef CONFIG_SCHED_INFO
3074         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
3075 #endif
3076 #ifdef CONFIG_LATENCYTOP
3077         REG("latency",  S_IRUGO, proc_lstats_operations),
3078 #endif
3079 #ifdef CONFIG_PROC_PID_CPUSET
3080         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
3081 #endif
3082 #ifdef CONFIG_CGROUPS
3083         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3084 #endif
3085         ONE("oom_score",  S_IRUGO, proc_oom_score),
3086         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3087         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3088 #ifdef CONFIG_AUDIT
3089         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3090         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3091 #endif
3092 #ifdef CONFIG_FAULT_INJECTION
3093         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3094         REG("fail-nth", 0644, proc_fail_nth_operations),
3095 #endif
3096 #ifdef CONFIG_ELF_CORE
3097         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3098 #endif
3099 #ifdef CONFIG_TASK_IO_ACCOUNTING
3100         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
3101 #endif
3102 #ifdef CONFIG_USER_NS
3103         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3104         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3105         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3106         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3107 #endif
3108 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
3109         REG("timers",     S_IRUGO, proc_timers_operations),
3110 #endif
3111         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
3112 #ifdef CONFIG_LIVEPATCH
3113         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3114 #endif
3115 #ifdef CONFIG_STACKLEAK_METRICS
3116         ONE("stack_depth", S_IRUGO, proc_stack_depth),
3117 #endif
3118 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3119         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3120 #endif
3121 };
3122
3123 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
3124 {
3125         return proc_pident_readdir(file, ctx,
3126                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3127 }
3128
3129 static const struct file_operations proc_tgid_base_operations = {
3130         .read           = generic_read_dir,
3131         .iterate_shared = proc_tgid_base_readdir,
3132         .llseek         = generic_file_llseek,
3133 };
3134
3135 struct pid *tgid_pidfd_to_pid(const struct file *file)
3136 {
3137         if (file->f_op != &proc_tgid_base_operations)
3138                 return ERR_PTR(-EBADF);
3139
3140         return proc_pid(file_inode(file));
3141 }
3142
3143 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3144 {
3145         return proc_pident_lookup(dir, dentry,
3146                                   tgid_base_stuff,
3147                                   tgid_base_stuff + ARRAY_SIZE(tgid_base_stuff));
3148 }
3149
3150 static const struct inode_operations proc_tgid_base_inode_operations = {
3151         .lookup         = proc_tgid_base_lookup,
3152         .getattr        = pid_getattr,
3153         .setattr        = proc_setattr,
3154         .permission     = proc_pid_permission,
3155 };
3156
3157 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3158 {
3159         struct dentry *dentry, *leader, *dir;
3160         char buf[10 + 1];
3161         struct qstr name;
3162
3163         name.name = buf;
3164         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3165         /* no ->d_hash() rejects on procfs */
3166         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3167         if (dentry) {
3168                 d_invalidate(dentry);
3169                 dput(dentry);
3170         }
3171
3172         if (pid == tgid)
3173                 return;
3174
3175         name.name = buf;
3176         name.len = snprintf(buf, sizeof(buf), "%u", tgid);
3177         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3178         if (!leader)
3179                 goto out;
3180
3181         name.name = "task";
3182         name.len = strlen(name.name);
3183         dir = d_hash_and_lookup(leader, &name);
3184         if (!dir)
3185                 goto out_put_leader;
3186
3187         name.name = buf;
3188         name.len = snprintf(buf, sizeof(buf), "%u", pid);
3189         dentry = d_hash_and_lookup(dir, &name);
3190         if (dentry) {
3191                 d_invalidate(dentry);
3192                 dput(dentry);
3193         }
3194
3195         dput(dir);
3196 out_put_leader:
3197         dput(leader);
3198 out:
3199         return;
3200 }
3201
3202 /**
3203  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3204  * @task: task that should be flushed.
3205  *
3206  * When flushing dentries from proc, one needs to flush them from global
3207  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3208  * in. This call is supposed to do all of this job.
3209  *
3210  * Looks in the dcache for
3211  * /proc/@pid
3212  * /proc/@tgid/task/@pid
3213  * if either directory is present flushes it and all of it'ts children
3214  * from the dcache.
3215  *
3216  * It is safe and reasonable to cache /proc entries for a task until
3217  * that task exits.  After that they just clog up the dcache with
3218  * useless entries, possibly causing useful dcache entries to be
3219  * flushed instead.  This routine is proved to flush those useless
3220  * dcache entries at process exit time.
3221  *
3222  * NOTE: This routine is just an optimization so it does not guarantee
3223  *       that no dcache entries will exist at process exit time it
3224  *       just makes it very unlikely that any will persist.
3225  */
3226
3227 void proc_flush_task(struct task_struct *task)
3228 {
3229         int i;
3230         struct pid *pid, *tgid;
3231         struct upid *upid;
3232
3233         pid = task_pid(task);
3234         tgid = task_tgid(task);
3235
3236         for (i = 0; i <= pid->level; i++) {
3237                 upid = &pid->numbers[i];
3238                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3239                                         tgid->numbers[i].nr);
3240         }
3241 }
3242
3243 static struct dentry *proc_pid_instantiate(struct dentry * dentry,
3244                                    struct task_struct *task, const void *ptr)
3245 {
3246         struct inode *inode;
3247
3248         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3249         if (!inode)
3250                 return ERR_PTR(-ENOENT);
3251
3252         inode->i_op = &proc_tgid_base_inode_operations;
3253         inode->i_fop = &proc_tgid_base_operations;
3254         inode->i_flags|=S_IMMUTABLE;
3255
3256         set_nlink(inode, nlink_tgid);
3257         pid_update_inode(task, inode);
3258
3259         d_set_d_op(dentry, &pid_dentry_operations);
3260         return d_splice_alias(inode, dentry);
3261 }
3262
3263 struct dentry *proc_pid_lookup(struct dentry *dentry, unsigned int flags)
3264 {
3265         struct task_struct *task;
3266         unsigned tgid;
3267         struct pid_namespace *ns;
3268         struct dentry *result = ERR_PTR(-ENOENT);
3269
3270         tgid = name_to_int(&dentry->d_name);
3271         if (tgid == ~0U)
3272                 goto out;
3273
3274         ns = dentry->d_sb->s_fs_info;
3275         rcu_read_lock();
3276         task = find_task_by_pid_ns(tgid, ns);
3277         if (task)
3278                 get_task_struct(task);
3279         rcu_read_unlock();
3280         if (!task)
3281                 goto out;
3282
3283         result = proc_pid_instantiate(dentry, task, NULL);
3284         put_task_struct(task);
3285 out:
3286         return result;
3287 }
3288
3289 /*
3290  * Find the first task with tgid >= tgid
3291  *
3292  */
3293 struct tgid_iter {
3294         unsigned int tgid;
3295         struct task_struct *task;
3296 };
3297 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3298 {
3299         struct pid *pid;
3300
3301         if (iter.task)
3302                 put_task_struct(iter.task);
3303         rcu_read_lock();
3304 retry:
3305         iter.task = NULL;
3306         pid = find_ge_pid(iter.tgid, ns);
3307         if (pid) {
3308                 iter.tgid = pid_nr_ns(pid, ns);
3309                 iter.task = pid_task(pid, PIDTYPE_PID);
3310                 /* What we to know is if the pid we have find is the
3311                  * pid of a thread_group_leader.  Testing for task
3312                  * being a thread_group_leader is the obvious thing
3313                  * todo but there is a window when it fails, due to
3314                  * the pid transfer logic in de_thread.
3315                  *
3316                  * So we perform the straight forward test of seeing
3317                  * if the pid we have found is the pid of a thread
3318                  * group leader, and don't worry if the task we have
3319                  * found doesn't happen to be a thread group leader.
3320                  * As we don't care in the case of readdir.
3321                  */
3322                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3323                         iter.tgid += 1;
3324                         goto retry;
3325                 }
3326                 get_task_struct(iter.task);
3327         }
3328         rcu_read_unlock();
3329         return iter;
3330 }
3331
3332 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3333
3334 /* for the /proc/ directory itself, after non-process stuff has been done */
3335 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3336 {
3337         struct tgid_iter iter;
3338         struct pid_namespace *ns = proc_pid_ns(file_inode(file));
3339         loff_t pos = ctx->pos;
3340
3341         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3342                 return 0;
3343
3344         if (pos == TGID_OFFSET - 2) {
3345                 struct inode *inode = d_inode(ns->proc_self);
3346                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3347                         return 0;
3348                 ctx->pos = pos = pos + 1;
3349         }
3350         if (pos == TGID_OFFSET - 1) {
3351                 struct inode *inode = d_inode(ns->proc_thread_self);
3352                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3353                         return 0;
3354                 ctx->pos = pos = pos + 1;
3355         }
3356         iter.tgid = pos - TGID_OFFSET;
3357         iter.task = NULL;
3358         for (iter = next_tgid(ns, iter);
3359              iter.task;
3360              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3361                 char name[10 + 1];
3362                 unsigned int len;
3363
3364                 cond_resched();
3365                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3366                         continue;
3367
3368                 len = snprintf(name, sizeof(name), "%u", iter.tgid);
3369                 ctx->pos = iter.tgid + TGID_OFFSET;
3370                 if (!proc_fill_cache(file, ctx, name, len,
3371                                      proc_pid_instantiate, iter.task, NULL)) {
3372                         put_task_struct(iter.task);
3373                         return 0;
3374                 }
3375         }
3376         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3377         return 0;
3378 }
3379
3380 /*
3381  * proc_tid_comm_permission is a special permission function exclusively
3382  * used for the node /proc/<pid>/task/<tid>/comm.
3383  * It bypasses generic permission checks in the case where a task of the same
3384  * task group attempts to access the node.
3385  * The rationale behind this is that glibc and bionic access this node for
3386  * cross thread naming (pthread_set/getname_np(!self)). However, if
3387  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3388  * which locks out the cross thread naming implementation.
3389  * This function makes sure that the node is always accessible for members of
3390  * same thread group.
3391  */
3392 static int proc_tid_comm_permission(struct inode *inode, int mask)
3393 {
3394         bool is_same_tgroup;
3395         struct task_struct *task;
3396
3397         task = get_proc_task(inode);
3398         if (!task)
3399                 return -ESRCH;
3400         is_same_tgroup = same_thread_group(current, task);
3401         put_task_struct(task);
3402
3403         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3404                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3405                  * read or written by the members of the corresponding
3406                  * thread group.
3407                  */
3408                 return 0;
3409         }
3410
3411         return generic_permission(inode, mask);
3412 }
3413
3414 static const struct inode_operations proc_tid_comm_inode_operations = {
3415                 .setattr        = proc_setattr,
3416                 .permission     = proc_tid_comm_permission,
3417 };
3418
3419 /*
3420  * Tasks
3421  */
3422 static const struct pid_entry tid_base_stuff[] = {
3423         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3424         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3425         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3426 #ifdef CONFIG_NET
3427         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3428 #endif
3429         REG("environ",   S_IRUSR, proc_environ_operations),
3430         REG("auxv",      S_IRUSR, proc_auxv_operations),
3431         ONE("status",    S_IRUGO, proc_pid_status),
3432         ONE("personality", S_IRUSR, proc_pid_personality),
3433         ONE("limits",    S_IRUGO, proc_pid_limits),
3434 #ifdef CONFIG_SCHED_DEBUG
3435         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3436 #endif
3437         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3438                          &proc_tid_comm_inode_operations,
3439                          &proc_pid_set_comm_operations, {}),
3440 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3441         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3442 #endif
3443         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3444         ONE("stat",      S_IRUGO, proc_tid_stat),
3445         ONE("statm",     S_IRUGO, proc_pid_statm),
3446         REG("maps",      S_IRUGO, proc_pid_maps_operations),
3447 #ifdef CONFIG_PROC_CHILDREN
3448         REG("children",  S_IRUGO, proc_tid_children_operations),
3449 #endif
3450 #ifdef CONFIG_NUMA
3451         REG("numa_maps", S_IRUGO, proc_pid_numa_maps_operations),
3452 #endif
3453         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3454         LNK("cwd",       proc_cwd_link),
3455         LNK("root",      proc_root_link),
3456         LNK("exe",       proc_exe_link),
3457         REG("mounts",    S_IRUGO, proc_mounts_operations),
3458         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3459 #ifdef CONFIG_PROC_PAGE_MONITOR
3460         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3461         REG("smaps",     S_IRUGO, proc_pid_smaps_operations),
3462         REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations),
3463         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3464 #endif
3465 #ifdef CONFIG_SECURITY
3466         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3467 #endif
3468 #ifdef CONFIG_KALLSYMS
3469         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3470 #endif
3471 #ifdef CONFIG_STACKTRACE
3472         ONE("stack",      S_IRUSR, proc_pid_stack),
3473 #endif
3474 #ifdef CONFIG_SCHED_INFO
3475         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3476 #endif
3477 #ifdef CONFIG_LATENCYTOP
3478         REG("latency",  S_IRUGO, proc_lstats_operations),
3479 #endif
3480 #ifdef CONFIG_PROC_PID_CPUSET
3481         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3482 #endif
3483 #ifdef CONFIG_CGROUPS
3484         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3485 #endif
3486         ONE("oom_score", S_IRUGO, proc_oom_score),
3487         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3488         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3489 #ifdef CONFIG_AUDIT
3490         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3491         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3492 #endif
3493 #ifdef CONFIG_FAULT_INJECTION
3494         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3495         REG("fail-nth", 0644, proc_fail_nth_operations),
3496 #endif
3497 #ifdef CONFIG_TASK_IO_ACCOUNTING
3498         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3499 #endif
3500 #ifdef CONFIG_USER_NS
3501         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3502         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3503         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3504         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3505 #endif
3506 #ifdef CONFIG_LIVEPATCH
3507         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3508 #endif
3509 #ifdef CONFIG_PROC_PID_ARCH_STATUS
3510         ONE("arch_status", S_IRUGO, proc_pid_arch_status),
3511 #endif
3512 };
3513
3514 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3515 {
3516         return proc_pident_readdir(file, ctx,
3517                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3518 }
3519
3520 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3521 {
3522         return proc_pident_lookup(dir, dentry,
3523                                   tid_base_stuff,
3524                                   tid_base_stuff + ARRAY_SIZE(tid_base_stuff));
3525 }
3526
3527 static const struct file_operations proc_tid_base_operations = {
3528         .read           = generic_read_dir,
3529         .iterate_shared = proc_tid_base_readdir,
3530         .llseek         = generic_file_llseek,
3531 };
3532
3533 static const struct inode_operations proc_tid_base_inode_operations = {
3534         .lookup         = proc_tid_base_lookup,
3535         .getattr        = pid_getattr,
3536         .setattr        = proc_setattr,
3537 };
3538
3539 static struct dentry *proc_task_instantiate(struct dentry *dentry,
3540         struct task_struct *task, const void *ptr)
3541 {
3542         struct inode *inode;
3543         inode = proc_pid_make_inode(dentry->d_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3544         if (!inode)
3545                 return ERR_PTR(-ENOENT);
3546
3547         inode->i_op = &proc_tid_base_inode_operations;
3548         inode->i_fop = &proc_tid_base_operations;
3549         inode->i_flags |= S_IMMUTABLE;
3550
3551         set_nlink(inode, nlink_tid);
3552         pid_update_inode(task, inode);
3553
3554         d_set_d_op(dentry, &pid_dentry_operations);
3555         return d_splice_alias(inode, dentry);
3556 }
3557
3558 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3559 {
3560         struct task_struct *task;
3561         struct task_struct *leader = get_proc_task(dir);
3562         unsigned tid;
3563         struct pid_namespace *ns;
3564         struct dentry *result = ERR_PTR(-ENOENT);
3565
3566         if (!leader)
3567                 goto out_no_task;
3568
3569         tid = name_to_int(&dentry->d_name);
3570         if (tid == ~0U)
3571                 goto out;
3572
3573         ns = dentry->d_sb->s_fs_info;
3574         rcu_read_lock();
3575         task = find_task_by_pid_ns(tid, ns);
3576         if (task)
3577                 get_task_struct(task);
3578         rcu_read_unlock();
3579         if (!task)
3580                 goto out;
3581         if (!same_thread_group(leader, task))
3582                 goto out_drop_task;
3583
3584         result = proc_task_instantiate(dentry, task, NULL);
3585 out_drop_task:
3586         put_task_struct(task);
3587 out:
3588         put_task_struct(leader);
3589 out_no_task:
3590         return result;
3591 }
3592
3593 /*
3594  * Find the first tid of a thread group to return to user space.
3595  *
3596  * Usually this is just the thread group leader, but if the users
3597  * buffer was too small or there was a seek into the middle of the
3598  * directory we have more work todo.
3599  *
3600  * In the case of a short read we start with find_task_by_pid.
3601  *
3602  * In the case of a seek we start with the leader and walk nr
3603  * threads past it.
3604  */
3605 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3606                                         struct pid_namespace *ns)
3607 {
3608         struct task_struct *pos, *task;
3609         unsigned long nr = f_pos;
3610
3611         if (nr != f_pos)        /* 32bit overflow? */
3612                 return NULL;
3613
3614         rcu_read_lock();
3615         task = pid_task(pid, PIDTYPE_PID);
3616         if (!task)
3617                 goto fail;
3618
3619         /* Attempt to start with the tid of a thread */
3620         if (tid && nr) {
3621                 pos = find_task_by_pid_ns(tid, ns);
3622                 if (pos && same_thread_group(pos, task))
3623                         goto found;
3624         }
3625
3626         /* If nr exceeds the number of threads there is nothing todo */
3627         if (nr >= get_nr_threads(task))
3628                 goto fail;
3629
3630         /* If we haven't found our starting place yet start
3631          * with the leader and walk nr threads forward.
3632          */
3633         pos = task = task->group_leader;
3634         do {
3635                 if (!nr--)
3636                         goto found;
3637         } while_each_thread(task, pos);
3638 fail:
3639         pos = NULL;
3640         goto out;
3641 found:
3642         get_task_struct(pos);
3643 out:
3644         rcu_read_unlock();
3645         return pos;
3646 }
3647
3648 /*
3649  * Find the next thread in the thread list.
3650  * Return NULL if there is an error or no next thread.
3651  *
3652  * The reference to the input task_struct is released.
3653  */
3654 static struct task_struct *next_tid(struct task_struct *start)
3655 {
3656         struct task_struct *pos = NULL;
3657         rcu_read_lock();
3658         if (pid_alive(start)) {
3659                 pos = next_thread(start);
3660                 if (thread_group_leader(pos))
3661                         pos = NULL;
3662                 else
3663                         get_task_struct(pos);
3664         }
3665         rcu_read_unlock();
3666         put_task_struct(start);
3667         return pos;
3668 }
3669
3670 /* for the /proc/TGID/task/ directories */
3671 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3672 {
3673         struct inode *inode = file_inode(file);
3674         struct task_struct *task;
3675         struct pid_namespace *ns;
3676         int tid;
3677
3678         if (proc_inode_is_dead(inode))
3679                 return -ENOENT;
3680
3681         if (!dir_emit_dots(file, ctx))
3682                 return 0;
3683
3684         /* f_version caches the tgid value that the last readdir call couldn't
3685          * return. lseek aka telldir automagically resets f_version to 0.
3686          */
3687         ns = proc_pid_ns(inode);
3688         tid = (int)file->f_version;
3689         file->f_version = 0;
3690         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3691              task;
3692              task = next_tid(task), ctx->pos++) {
3693                 char name[10 + 1];
3694                 unsigned int len;
3695                 tid = task_pid_nr_ns(task, ns);
3696                 len = snprintf(name, sizeof(name), "%u", tid);
3697                 if (!proc_fill_cache(file, ctx, name, len,
3698                                 proc_task_instantiate, task, NULL)) {
3699                         /* returning this tgid failed, save it as the first
3700                          * pid for the next readir call */
3701                         file->f_version = (u64)tid;
3702                         put_task_struct(task);
3703                         break;
3704                 }
3705         }
3706
3707         return 0;
3708 }
3709
3710 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3711                              u32 request_mask, unsigned int query_flags)
3712 {
3713         struct inode *inode = d_inode(path->dentry);
3714         struct task_struct *p = get_proc_task(inode);
3715         generic_fillattr(inode, stat);
3716
3717         if (p) {
3718                 stat->nlink += get_nr_threads(p);
3719                 put_task_struct(p);
3720         }
3721
3722         return 0;
3723 }
3724
3725 static const struct inode_operations proc_task_inode_operations = {
3726         .lookup         = proc_task_lookup,
3727         .getattr        = proc_task_getattr,
3728         .setattr        = proc_setattr,
3729         .permission     = proc_pid_permission,
3730 };
3731
3732 static const struct file_operations proc_task_operations = {
3733         .read           = generic_read_dir,
3734         .iterate_shared = proc_task_readdir,
3735         .llseek         = generic_file_llseek,
3736 };
3737
3738 void __init set_proc_pid_nlink(void)
3739 {
3740         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3741         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3742 }