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