fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <linux/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/sched/autogroup.h>
  89 #include <linux/sched/mm.h>
  90 #include <linux/sched/coredump.h>
  91 #include <linux/sched/debug.h>
  92 #include <linux/sched/stat.h>
  93 #include <linux/flex_array.h>
  94 #include <linux/posix-timers.h>
  95 #ifdef CONFIG_HARDWALL
  96 #include <asm/hardwall.h>
  97 #endif
  98 #include <trace/events/oom.h>
  99 #include "internal.h"
 100 #include "fd.h"
 101
 102 /* NOTE:
 103  *      Implementing inode permission operations in /proc is almost
 104  *      certainly an error.  Permission checks need to happen during
 105  *      each system call not at open time.  The reason is that most of
 106  *      what we wish to check for permissions in /proc varies at runtime.
 107  *
 108  *      The classic example of a problem is opening file descriptors
 109  *      in /proc for a task before it execs a suid executable.
 110  */
 111
 112 static u8 nlink_tid;
 113 static u8 nlink_tgid;
 114
 115 struct pid_entry {
 116         const char *name;
 117         unsigned int len;
 118         umode_t mode;
 119         const struct inode_operations *iop;
 120         const struct file_operations *fop;
 121         union proc_op op;
 122 };
 123
 124 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 125         .name = (NAME),                                 \
 126         .len  = sizeof(NAME) - 1,                       \
 127         .mode = MODE,                                   \
 128         .iop  = IOP,                                    \
 129         .fop  = FOP,                                    \
 130         .op   = OP,                                     \
 131 }
 132
 133 #define DIR(NAME, MODE, iops, fops)     \
 134         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 135 #define LNK(NAME, get_link)                                     \
 136         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 137                 &proc_pid_link_inode_operations, NULL,          \
 138                 { .proc_get_link = get_link } )
 139 #define REG(NAME, MODE, fops)                           \
 140         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 141 #define ONE(NAME, MODE, show)                           \
 142         NOD(NAME, (S_IFREG|(MODE)),                     \
 143                 NULL, &proc_single_file_operations,     \
 144                 { .proc_show = show } )
 145
 146 /*
 147  * Count the number of hardlinks for the pid_entry table, excluding the .
 148  * and .. links.
 149  */
 150 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 151         unsigned int n)
 152 {
 153         unsigned int i;
 154         unsigned int count;
 155
 156         count = 2;
 157         for (i = 0; i < n; ++i) {
 158                 if (S_ISDIR(entries[i].mode))
 159                         ++count;
 160         }
 161
 162         return count;
 163 }
 164
 165 static int get_task_root(struct task_struct *task, struct path *root)
 166 {
 167         int result = -ENOENT;
 168
 169         task_lock(task);
 170         if (task->fs) {
 171                 get_fs_root(task->fs, root);
 172                 result = 0;
 173         }
 174         task_unlock(task);
 175         return result;
 176 }
 177
 178 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 179 {
 180         struct task_struct *task = get_proc_task(d_inode(dentry));
 181         int result = -ENOENT;
 182
 183         if (task) {
 184                 task_lock(task);
 185                 if (task->fs) {
 186                         get_fs_pwd(task->fs, path);
 187                         result = 0;
 188                 }
 189                 task_unlock(task);
 190                 put_task_struct(task);
 191         }
 192         return result;
 193 }
 194
 195 static int proc_root_link(struct dentry *dentry, struct path *path)
 196 {
 197         struct task_struct *task = get_proc_task(d_inode(dentry));
 198         int result = -ENOENT;
 199
 200         if (task) {
 201                 result = get_task_root(task, path);
 202                 put_task_struct(task);
 203         }
 204         return result;
 205 }
 206
 207 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 208                                      size_t _count, loff_t *pos)
 209 {
 210         struct task_struct *tsk;
 211         struct mm_struct *mm;
 212         char *page;
 213         unsigned long count = _count;
 214         unsigned long arg_start, arg_end, env_start, env_end;
 215         unsigned long len1, len2, len;
 216         unsigned long p;
 217         char c;
 218         ssize_t rv;
 219
 220         BUG_ON(*pos < 0);
 221
 222         tsk = get_proc_task(file_inode(file));
 223         if (!tsk)
 224                 return -ESRCH;
 225         mm = get_task_mm(tsk);
 226         put_task_struct(tsk);
 227         if (!mm)
 228                 return 0;
 229         /* Check if process spawned far enough to have cmdline. */
 230         if (!mm->env_end) {
 231                 rv = 0;
 232                 goto out_mmput;
 233         }
 234
 235         page = (char *)__get_free_page(GFP_TEMPORARY);
 236         if (!page) {
 237                 rv = -ENOMEM;
 238                 goto out_mmput;
 239         }
 240
 241         down_read(&mm->mmap_sem);
 242         arg_start = mm->arg_start;
 243         arg_end = mm->arg_end;
 244         env_start = mm->env_start;
 245         env_end = mm->env_end;
 246         up_read(&mm->mmap_sem);
 247
 248         BUG_ON(arg_start > arg_end);
 249         BUG_ON(env_start > env_end);
 250
 251         len1 = arg_end - arg_start;
 252         len2 = env_end - env_start;
 253
 254         /* Empty ARGV. */
 255         if (len1 == 0) {
 256                 rv = 0;
 257                 goto out_free_page;
 258         }
 259         /*
 260          * Inherently racy -- command line shares address space
 261          * with code and data.
 262          */
 263         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 264         if (rv <= 0)
 265                 goto out_free_page;
 266
 267         rv = 0;
 268
 269         if (c == '\0') {
 270                 /* Command line (set of strings) occupies whole ARGV. */
 271                 if (len1 <= *pos)
 272                         goto out_free_page;
 273
 274                 p = arg_start + *pos;
 275                 len = len1 - *pos;
 276                 while (count > 0 && len > 0) {
 277                         unsigned int _count;
 278                         int nr_read;
 279
 280                         _count = min3(count, len, PAGE_SIZE);
 281                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 282                         if (nr_read < 0)
 283                                 rv = nr_read;
 284                         if (nr_read <= 0)
 285                                 goto out_free_page;
 286
 287                         if (copy_to_user(buf, page, nr_read)) {
 288                                 rv = -EFAULT;
 289                                 goto out_free_page;
 290                         }
 291
 292                         p       += nr_read;
 293                         len     -= nr_read;
 294                         buf     += nr_read;
 295                         count   -= nr_read;
 296                         rv      += nr_read;
 297                 }
 298         } else {
 299                 /*
 300                  * Command line (1 string) occupies ARGV and
 301                  * extends into ENVP.
 302                  */
 303                 struct {
 304                         unsigned long p;
 305                         unsigned long len;
 306                 } cmdline[2] = {
 307                         { .p = arg_start, .len = len1 },
 308                         { .p = env_start, .len = len2 },
 309                 };
 310                 loff_t pos1 = *pos;
 311                 unsigned int i;
 312
 313                 i = 0;
 314                 while (i < 2 && pos1 >= cmdline[i].len) {
 315                         pos1 -= cmdline[i].len;
 316                         i++;
 317                 }
 318                 while (i < 2) {
 319                         p = cmdline[i].p + pos1;
 320                         len = cmdline[i].len - pos1;
 321                         while (count > 0 && len > 0) {
 322                                 unsigned int _count, l;
 323                                 int nr_read;
 324                                 bool final;
 325
 326                                 _count = min3(count, len, PAGE_SIZE);
 327                                 nr_read = access_remote_vm(mm, p, page, _count, 0);
 328                                 if (nr_read < 0)
 329                                         rv = nr_read;
 330                                 if (nr_read <= 0)
 331                                         goto out_free_page;
 332
 333                                 /*
 334                                  * Command line can be shorter than whole ARGV
 335                                  * even if last "marker" byte says it is not.
 336                                  */
 337                                 final = false;
 338                                 l = strnlen(page, nr_read);
 339                                 if (l < nr_read) {
 340                                         nr_read = l;
 341                                         final = true;
 342                                 }
 343
 344                                 if (copy_to_user(buf, page, nr_read)) {
 345                                         rv = -EFAULT;
 346                                         goto out_free_page;
 347                                 }
 348
 349                                 p       += nr_read;
 350                                 len     -= nr_read;
 351                                 buf     += nr_read;
 352                                 count   -= nr_read;
 353                                 rv      += nr_read;
 354
 355                                 if (final)
 356                                         goto out_free_page;
 357                         }
 358
 359                         /* Only first chunk can be read partially. */
 360                         pos1 = 0;
 361                         i++;
 362                 }
 363         }
 364
 365 out_free_page:
 366         free_page((unsigned long)page);
 367 out_mmput:
 368         mmput(mm);
 369         if (rv > 0)
 370                 *pos += rv;
 371         return rv;
 372 }
 373
 374 static const struct file_operations proc_pid_cmdline_ops = {
 375         .read   = proc_pid_cmdline_read,
 376         .llseek = generic_file_llseek,
 377 };
 378
 379 #ifdef CONFIG_KALLSYMS
 380 /*
 381  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 382  * Returns the resolved symbol.  If that fails, simply return the address.
 383  */
 384 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 385                           struct pid *pid, struct task_struct *task)
 386 {
 387         unsigned long wchan;
 388         char symname[KSYM_NAME_LEN];
 389
 390         wchan = get_wchan(task);
 391
 392         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 393                         && !lookup_symbol_name(wchan, symname))
 394                 seq_printf(m, "%s", symname);
 395         else
 396                 seq_putc(m, '0');
 397
 398         return 0;
 399 }
 400 #endif /* CONFIG_KALLSYMS */
 401
 402 static int lock_trace(struct task_struct *task)
 403 {
 404         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 405         if (err)
 406                 return err;
 407         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 408                 mutex_unlock(&task->signal->cred_guard_mutex);
 409                 return -EPERM;
 410         }
 411         return 0;
 412 }
 413
 414 static void unlock_trace(struct task_struct *task)
 415 {
 416         mutex_unlock(&task->signal->cred_guard_mutex);
 417 }
 418
 419 #ifdef CONFIG_STACKTRACE
 420
 421 #define MAX_STACK_TRACE_DEPTH   64
 422
 423 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 424                           struct pid *pid, struct task_struct *task)
 425 {
 426         struct stack_trace trace;
 427         unsigned long *entries;
 428         int err;
 429         int i;
 430
 431         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 432         if (!entries)
 433                 return -ENOMEM;
 434
 435         trace.nr_entries        = 0;
 436         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 437         trace.entries           = entries;
 438         trace.skip              = 0;
 439
 440         err = lock_trace(task);
 441         if (!err) {
 442                 save_stack_trace_tsk(task, &trace);
 443
 444                 for (i = 0; i < trace.nr_entries; i++) {
 445                         seq_printf(m, "[<%pK>] %pB\n",
 446                                    (void *)entries[i], (void *)entries[i]);
 447                 }
 448                 unlock_trace(task);
 449         }
 450         kfree(entries);
 451
 452         return err;
 453 }
 454 #endif
 455
 456 #ifdef CONFIG_SCHED_INFO
 457 /*
 458  * Provides /proc/PID/schedstat
 459  */
 460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 461                               struct pid *pid, struct task_struct *task)
 462 {
 463         if (unlikely(!sched_info_on()))
 464                 seq_printf(m, "0 0 0\n");
 465         else
 466                 seq_printf(m, "%llu %llu %lu\n",
 467                    (unsigned long long)task->se.sum_exec_runtime,
 468                    (unsigned long long)task->sched_info.run_delay,
 469                    task->sched_info.pcount);
 470
 471         return 0;
 472 }
 473 #endif
 474
 475 #ifdef CONFIG_LATENCYTOP
 476 static int lstats_show_proc(struct seq_file *m, void *v)
 477 {
 478         int i;
 479         struct inode *inode = m->private;
 480         struct task_struct *task = get_proc_task(inode);
 481
 482         if (!task)
 483                 return -ESRCH;
 484         seq_puts(m, "Latency Top version : v0.1\n");
 485         for (i = 0; i < 32; i++) {
 486                 struct latency_record *lr = &task->latency_record[i];
 487                 if (lr->backtrace[0]) {
 488                         int q;
 489                         seq_printf(m, "%i %li %li",
 490                                    lr->count, lr->time, lr->max);
 491                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 492                                 unsigned long bt = lr->backtrace[q];
 493                                 if (!bt)
 494                                         break;
 495                                 if (bt == ULONG_MAX)
 496                                         break;
 497                                 seq_printf(m, " %ps", (void *)bt);
 498                         }
 499                         seq_putc(m, '\n');
 500                 }
 501
 502         }
 503         put_task_struct(task);
 504         return 0;
 505 }
 506
 507 static int lstats_open(struct inode *inode, struct file *file)
 508 {
 509         return single_open(file, lstats_show_proc, inode);
 510 }
 511
 512 static ssize_t lstats_write(struct file *file, const char __user *buf,
 513                             size_t count, loff_t *offs)
 514 {
 515         struct task_struct *task = get_proc_task(file_inode(file));
 516
 517         if (!task)
 518                 return -ESRCH;
 519         clear_all_latency_tracing(task);
 520         put_task_struct(task);
 521
 522         return count;
 523 }
 524
 525 static const struct file_operations proc_lstats_operations = {
 526         .open           = lstats_open,
 527         .read           = seq_read,
 528         .write          = lstats_write,
 529         .llseek         = seq_lseek,
 530         .release        = single_release,
 531 };
 532
 533 #endif
 534
 535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 536                           struct pid *pid, struct task_struct *task)
 537 {
 538         unsigned long totalpages = totalram_pages + total_swap_pages;
 539         unsigned long points = 0;
 540
 541         points = oom_badness(task, NULL, NULL, totalpages) *
 542                                         1000 / totalpages;
 543         seq_printf(m, "%lu\n", points);
 544
 545         return 0;
 546 }
 547
 548 struct limit_names {
 549         const char *name;
 550         const char *unit;
 551 };
 552
 553 static const struct limit_names lnames[RLIM_NLIMITS] = {
 554         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 555         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 556         [RLIMIT_DATA] = {"Max data size", "bytes"},
 557         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 558         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 559         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 560         [RLIMIT_NPROC] = {"Max processes", "processes"},
 561         [RLIMIT_NOFILE] = {"Max open files", "files"},
 562         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 563         [RLIMIT_AS] = {"Max address space", "bytes"},
 564         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 565         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 566         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 567         [RLIMIT_NICE] = {"Max nice priority", NULL},
 568         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 569         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 570 };
 571
 572 /* Display limits for a process */
 573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 574                            struct pid *pid, struct task_struct *task)
 575 {
 576         unsigned int i;
 577         unsigned long flags;
 578
 579         struct rlimit rlim[RLIM_NLIMITS];
 580
 581         if (!lock_task_sighand(task, &flags))
 582                 return 0;
 583         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 584         unlock_task_sighand(task, &flags);
 585
 586         /*
 587          * print the file header
 588          */
 589        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 590                   "Limit", "Soft Limit", "Hard Limit", "Units");
 591
 592         for (i = 0; i < RLIM_NLIMITS; i++) {
 593                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 594                         seq_printf(m, "%-25s %-20s ",
 595                                    lnames[i].name, "unlimited");
 596                 else
 597                         seq_printf(m, "%-25s %-20lu ",
 598                                    lnames[i].name, rlim[i].rlim_cur);
 599
 600                 if (rlim[i].rlim_max == RLIM_INFINITY)
 601                         seq_printf(m, "%-20s ", "unlimited");
 602                 else
 603                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 604
 605                 if (lnames[i].unit)
 606                         seq_printf(m, "%-10s\n", lnames[i].unit);
 607                 else
 608                         seq_putc(m, '\n');
 609         }
 610
 611         return 0;
 612 }
 613
 614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 616                             struct pid *pid, struct task_struct *task)
 617 {
 618         long nr;
 619         unsigned long args[6], sp, pc;
 620         int res;
 621
 622         res = lock_trace(task);
 623         if (res)
 624                 return res;
 625
 626         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 627                 seq_puts(m, "running\n");
 628         else if (nr < 0)
 629                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 630         else
 631                 seq_printf(m,
 632                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 633                        nr,
 634                        args[0], args[1], args[2], args[3], args[4], args[5],
 635                        sp, pc);
 636         unlock_trace(task);
 637
 638         return 0;
 639 }
 640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 641
 642 /************************************************************************/
 643 /*                       Here the fs part begins                        */
 644 /************************************************************************/
 645
 646 /* permission checks */
 647 static int proc_fd_access_allowed(struct inode *inode)
 648 {
 649         struct task_struct *task;
 650         int allowed = 0;
 651         /* Allow access to a task's file descriptors if it is us or we
 652          * may use ptrace attach to the process and find out that
 653          * information.
 654          */
 655         task = get_proc_task(inode);
 656         if (task) {
 657                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 658                 put_task_struct(task);
 659         }
 660         return allowed;
 661 }
 662
 663 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 664 {
 665         int error;
 666         struct inode *inode = d_inode(dentry);
 667
 668         if (attr->ia_valid & ATTR_MODE)
 669                 return -EPERM;
 670
 671         error = setattr_prepare(dentry, attr);
 672         if (error)
 673                 return error;
 674
 675         setattr_copy(inode, attr);
 676         mark_inode_dirty(inode);
 677         return 0;
 678 }
 679
 680 /*
 681  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 682  * or euid/egid (for hide_pid_min=2)?
 683  */
 684 static bool has_pid_permissions(struct pid_namespace *pid,
 685                                  struct task_struct *task,
 686                                  int hide_pid_min)
 687 {
 688         if (pid->hide_pid < hide_pid_min)
 689                 return true;
 690         if (in_group_p(pid->pid_gid))
 691                 return true;
 692         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 693 }
 694
 695
 696 static int proc_pid_permission(struct inode *inode, int mask)
 697 {
 698         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 699         struct task_struct *task;
 700         bool has_perms;
 701
 702         task = get_proc_task(inode);
 703         if (!task)
 704                 return -ESRCH;
 705         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 706         put_task_struct(task);
 707
 708         if (!has_perms) {
 709                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 710                         /*
 711                          * Let's make getdents(), stat(), and open()
 712                          * consistent with each other.  If a process
 713                          * may not stat() a file, it shouldn't be seen
 714                          * in procfs at all.
 715                          */
 716                         return -ENOENT;
 717                 }
 718
 719                 return -EPERM;
 720         }
 721         return generic_permission(inode, mask);
 722 }
 723
 724
 725
 726 static const struct inode_operations proc_def_inode_operations = {
 727         .setattr        = proc_setattr,
 728 };
 729
 730 static int proc_single_show(struct seq_file *m, void *v)
 731 {
 732         struct inode *inode = m->private;
 733         struct pid_namespace *ns;
 734         struct pid *pid;
 735         struct task_struct *task;
 736         int ret;
 737
 738         ns = inode->i_sb->s_fs_info;
 739         pid = proc_pid(inode);
 740         task = get_pid_task(pid, PIDTYPE_PID);
 741         if (!task)
 742                 return -ESRCH;
 743
 744         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 745
 746         put_task_struct(task);
 747         return ret;
 748 }
 749
 750 static int proc_single_open(struct inode *inode, struct file *filp)
 751 {
 752         return single_open(filp, proc_single_show, inode);
 753 }
 754
 755 static const struct file_operations proc_single_file_operations = {
 756         .open           = proc_single_open,
 757         .read           = seq_read,
 758         .llseek         = seq_lseek,
 759         .release        = single_release,
 760 };
 761
 762
 763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 764 {
 765         struct task_struct *task = get_proc_task(inode);
 766         struct mm_struct *mm = ERR_PTR(-ESRCH);
 767
 768         if (task) {
 769                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 770                 put_task_struct(task);
 771
 772                 if (!IS_ERR_OR_NULL(mm)) {
 773                         /* ensure this mm_struct can't be freed */
 774                         mmgrab(mm);
 775                         /* but do not pin its memory */
 776                         mmput(mm);
 777                 }
 778         }
 779
 780         return mm;
 781 }
 782
 783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 784 {
 785         struct mm_struct *mm = proc_mem_open(inode, mode);
 786
 787         if (IS_ERR(mm))
 788                 return PTR_ERR(mm);
 789
 790         file->private_data = mm;
 791         return 0;
 792 }
 793
 794 static int mem_open(struct inode *inode, struct file *file)
 795 {
 796         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 797
 798         /* OK to pass negative loff_t, we can catch out-of-range */
 799         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 800
 801         return ret;
 802 }
 803
 804 static ssize_t mem_rw(struct file *file, char __user *buf,
 805                         size_t count, loff_t *ppos, int write)
 806 {
 807         struct mm_struct *mm = file->private_data;
 808         unsigned long addr = *ppos;
 809         ssize_t copied;
 810         char *page;
 811         unsigned int flags;
 812
 813         if (!mm)
 814                 return 0;
 815
 816         page = (char *)__get_free_page(GFP_TEMPORARY);
 817         if (!page)
 818                 return -ENOMEM;
 819
 820         copied = 0;
 821         if (!mmget_not_zero(mm))
 822                 goto free;
 823
 824         /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
 825         flags = FOLL_FORCE;
 826         if (write)
 827                 flags |= FOLL_WRITE;
 828
 829         while (count > 0) {
 830                 int this_len = min_t(int, count, PAGE_SIZE);
 831
 832                 if (write && copy_from_user(page, buf, this_len)) {
 833                         copied = -EFAULT;
 834                         break;
 835                 }
 836
 837                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 838                 if (!this_len) {
 839                         if (!copied)
 840                                 copied = -EIO;
 841                         break;
 842                 }
 843
 844                 if (!write && copy_to_user(buf, page, this_len)) {
 845                         copied = -EFAULT;
 846                         break;
 847                 }
 848
 849                 buf += this_len;
 850                 addr += this_len;
 851                 copied += this_len;
 852                 count -= this_len;
 853         }
 854         *ppos = addr;
 855
 856         mmput(mm);
 857 free:
 858         free_page((unsigned long) page);
 859         return copied;
 860 }
 861
 862 static ssize_t mem_read(struct file *file, char __user *buf,
 863                         size_t count, loff_t *ppos)
 864 {
 865         return mem_rw(file, buf, count, ppos, 0);
 866 }
 867
 868 static ssize_t mem_write(struct file *file, const char __user *buf,
 869                          size_t count, loff_t *ppos)
 870 {
 871         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 872 }
 873
 874 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 875 {
 876         switch (orig) {
 877         case 0:
 878                 file->f_pos = offset;
 879                 break;
 880         case 1:
 881                 file->f_pos += offset;
 882                 break;
 883         default:
 884                 return -EINVAL;
 885         }
 886         force_successful_syscall_return();
 887         return file->f_pos;
 888 }
 889
 890 static int mem_release(struct inode *inode, struct file *file)
 891 {
 892         struct mm_struct *mm = file->private_data;
 893         if (mm)
 894                 mmdrop(mm);
 895         return 0;
 896 }
 897
 898 static const struct file_operations proc_mem_operations = {
 899         .llseek         = mem_lseek,
 900         .read           = mem_read,
 901         .write          = mem_write,
 902         .open           = mem_open,
 903         .release        = mem_release,
 904 };
 905
 906 static int environ_open(struct inode *inode, struct file *file)
 907 {
 908         return __mem_open(inode, file, PTRACE_MODE_READ);
 909 }
 910
 911 static ssize_t environ_read(struct file *file, char __user *buf,
 912                         size_t count, loff_t *ppos)
 913 {
 914         char *page;
 915         unsigned long src = *ppos;
 916         int ret = 0;
 917         struct mm_struct *mm = file->private_data;
 918         unsigned long env_start, env_end;
 919
 920         /* Ensure the process spawned far enough to have an environment. */
 921         if (!mm || !mm->env_end)
 922                 return 0;
 923
 924         page = (char *)__get_free_page(GFP_TEMPORARY);
 925         if (!page)
 926                 return -ENOMEM;
 927
 928         ret = 0;
 929         if (!mmget_not_zero(mm))
 930                 goto free;
 931
 932         down_read(&mm->mmap_sem);
 933         env_start = mm->env_start;
 934         env_end = mm->env_end;
 935         up_read(&mm->mmap_sem);
 936
 937         while (count > 0) {
 938                 size_t this_len, max_len;
 939                 int retval;
 940
 941                 if (src >= (env_end - env_start))
 942                         break;
 943
 944                 this_len = env_end - (env_start + src);
 945
 946                 max_len = min_t(size_t, PAGE_SIZE, count);
 947                 this_len = min(max_len, this_len);
 948
 949                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 950
 951                 if (retval <= 0) {
 952                         ret = retval;
 953                         break;
 954                 }
 955
 956                 if (copy_to_user(buf, page, retval)) {
 957                         ret = -EFAULT;
 958                         break;
 959                 }
 960
 961                 ret += retval;
 962                 src += retval;
 963                 buf += retval;
 964                 count -= retval;
 965         }
 966         *ppos = src;
 967         mmput(mm);
 968
 969 free:
 970         free_page((unsigned long) page);
 971         return ret;
 972 }
 973
 974 static const struct file_operations proc_environ_operations = {
 975         .open           = environ_open,
 976         .read           = environ_read,
 977         .llseek         = generic_file_llseek,
 978         .release        = mem_release,
 979 };
 980
 981 static int auxv_open(struct inode *inode, struct file *file)
 982 {
 983         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 984 }
 985
 986 static ssize_t auxv_read(struct file *file, char __user *buf,
 987                         size_t count, loff_t *ppos)
 988 {
 989         struct mm_struct *mm = file->private_data;
 990         unsigned int nwords = 0;
 991
 992         if (!mm)
 993                 return 0;
 994         do {
 995                 nwords += 2;
 996         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 997         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 998                                        nwords * sizeof(mm->saved_auxv[0]));
 999 }
1000
1001 static const struct file_operations proc_auxv_operations = {
1002         .open           = auxv_open,
1003         .read           = auxv_read,
1004         .llseek         = generic_file_llseek,
1005         .release        = mem_release,
1006 };
1007
1008 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1009                             loff_t *ppos)
1010 {
1011         struct task_struct *task = get_proc_task(file_inode(file));
1012         char buffer[PROC_NUMBUF];
1013         int oom_adj = OOM_ADJUST_MIN;
1014         size_t len;
1015
1016         if (!task)
1017                 return -ESRCH;
1018         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1019                 oom_adj = OOM_ADJUST_MAX;
1020         else
1021                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1022                           OOM_SCORE_ADJ_MAX;
1023         put_task_struct(task);
1024         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1025         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1026 }
1027
1028 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1029 {
1030         static DEFINE_MUTEX(oom_adj_mutex);
1031         struct mm_struct *mm = NULL;
1032         struct task_struct *task;
1033         int err = 0;
1034
1035         task = get_proc_task(file_inode(file));
1036         if (!task)
1037                 return -ESRCH;
1038
1039         mutex_lock(&oom_adj_mutex);
1040         if (legacy) {
1041                 if (oom_adj < task->signal->oom_score_adj &&
1042                                 !capable(CAP_SYS_RESOURCE)) {
1043                         err = -EACCES;
1044                         goto err_unlock;
1045                 }
1046                 /*
1047                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1048                  * /proc/pid/oom_score_adj instead.
1049                  */
1050                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1051                           current->comm, task_pid_nr(current), task_pid_nr(task),
1052                           task_pid_nr(task));
1053         } else {
1054                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1055                                 !capable(CAP_SYS_RESOURCE)) {
1056                         err = -EACCES;
1057                         goto err_unlock;
1058                 }
1059         }
1060
1061         /*
1062          * Make sure we will check other processes sharing the mm if this is
1063          * not vfrok which wants its own oom_score_adj.
1064          * pin the mm so it doesn't go away and get reused after task_unlock
1065          */
1066         if (!task->vfork_done) {
1067                 struct task_struct *p = find_lock_task_mm(task);
1068
1069                 if (p) {
1070                         if (atomic_read(&p->mm->mm_users) > 1) {
1071                                 mm = p->mm;
1072                                 mmgrab(mm);
1073                         }
1074                         task_unlock(p);
1075                 }
1076         }
1077
1078         task->signal->oom_score_adj = oom_adj;
1079         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1080                 task->signal->oom_score_adj_min = (short)oom_adj;
1081         trace_oom_score_adj_update(task);
1082
1083         if (mm) {
1084                 struct task_struct *p;
1085
1086                 rcu_read_lock();
1087                 for_each_process(p) {
1088                         if (same_thread_group(task, p))
1089                                 continue;
1090
1091                         /* do not touch kernel threads or the global init */
1092                         if (p->flags & PF_KTHREAD || is_global_init(p))
1093                                 continue;
1094
1095                         task_lock(p);
1096                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1097                                 pr_info("updating oom_score_adj for %d (%s) from %d to %d because it shares mm with %d (%s). Report if this is unexpected.\n",
1098                                                 task_pid_nr(p), p->comm,
1099                                                 p->signal->oom_score_adj, oom_adj,
1100                                                 task_pid_nr(task), task->comm);
1101                                 p->signal->oom_score_adj = oom_adj;
1102                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1103                                         p->signal->oom_score_adj_min = (short)oom_adj;
1104                         }
1105                         task_unlock(p);
1106                 }
1107                 rcu_read_unlock();
1108                 mmdrop(mm);
1109         }
1110 err_unlock:
1111         mutex_unlock(&oom_adj_mutex);
1112         put_task_struct(task);
1113         return err;
1114 }
1115
1116 /*
1117  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1118  * kernels.  The effective policy is defined by oom_score_adj, which has a
1119  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1120  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1121  * Processes that become oom disabled via oom_adj will still be oom disabled
1122  * with this implementation.
1123  *
1124  * oom_adj cannot be removed since existing userspace binaries use it.
1125  */
1126 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1127                              size_t count, loff_t *ppos)
1128 {
1129         char buffer[PROC_NUMBUF];
1130         int oom_adj;
1131         int err;
1132
1133         memset(buffer, 0, sizeof(buffer));
1134         if (count > sizeof(buffer) - 1)
1135                 count = sizeof(buffer) - 1;
1136         if (copy_from_user(buffer, buf, count)) {
1137                 err = -EFAULT;
1138                 goto out;
1139         }
1140
1141         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1142         if (err)
1143                 goto out;
1144         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1145              oom_adj != OOM_DISABLE) {
1146                 err = -EINVAL;
1147                 goto out;
1148         }
1149
1150         /*
1151          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1152          * value is always attainable.
1153          */
1154         if (oom_adj == OOM_ADJUST_MAX)
1155                 oom_adj = OOM_SCORE_ADJ_MAX;
1156         else
1157                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1158
1159         err = __set_oom_adj(file, oom_adj, true);
1160 out:
1161         return err < 0 ? err : count;
1162 }
1163
1164 static const struct file_operations proc_oom_adj_operations = {
1165         .read           = oom_adj_read,
1166         .write          = oom_adj_write,
1167         .llseek         = generic_file_llseek,
1168 };
1169
1170 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1171                                         size_t count, loff_t *ppos)
1172 {
1173         struct task_struct *task = get_proc_task(file_inode(file));
1174         char buffer[PROC_NUMBUF];
1175         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1176         size_t len;
1177
1178         if (!task)
1179                 return -ESRCH;
1180         oom_score_adj = task->signal->oom_score_adj;
1181         put_task_struct(task);
1182         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1183         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1184 }
1185
1186 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1187                                         size_t count, loff_t *ppos)
1188 {
1189         char buffer[PROC_NUMBUF];
1190         int oom_score_adj;
1191         int err;
1192
1193         memset(buffer, 0, sizeof(buffer));
1194         if (count > sizeof(buffer) - 1)
1195                 count = sizeof(buffer) - 1;
1196         if (copy_from_user(buffer, buf, count)) {
1197                 err = -EFAULT;
1198                 goto out;
1199         }
1200
1201         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1202         if (err)
1203                 goto out;
1204         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1205                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1206                 err = -EINVAL;
1207                 goto out;
1208         }
1209
1210         err = __set_oom_adj(file, oom_score_adj, false);
1211 out:
1212         return err < 0 ? err : count;
1213 }
1214
1215 static const struct file_operations proc_oom_score_adj_operations = {
1216         .read           = oom_score_adj_read,
1217         .write          = oom_score_adj_write,
1218         .llseek         = default_llseek,
1219 };
1220
1221 #ifdef CONFIG_AUDITSYSCALL
1222 #define TMPBUFLEN 11
1223 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1224                                   size_t count, loff_t *ppos)
1225 {
1226         struct inode * inode = file_inode(file);
1227         struct task_struct *task = get_proc_task(inode);
1228         ssize_t length;
1229         char tmpbuf[TMPBUFLEN];
1230
1231         if (!task)
1232                 return -ESRCH;
1233         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1234                            from_kuid(file->f_cred->user_ns,
1235                                      audit_get_loginuid(task)));
1236         put_task_struct(task);
1237         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1238 }
1239
1240 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1241                                    size_t count, loff_t *ppos)
1242 {
1243         struct inode * inode = file_inode(file);
1244         uid_t loginuid;
1245         kuid_t kloginuid;
1246         int rv;
1247
1248         rcu_read_lock();
1249         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1250                 rcu_read_unlock();
1251                 return -EPERM;
1252         }
1253         rcu_read_unlock();
1254
1255         if (*ppos != 0) {
1256                 /* No partial writes. */
1257                 return -EINVAL;
1258         }
1259
1260         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1261         if (rv < 0)
1262                 return rv;
1263
1264         /* is userspace tring to explicitly UNSET the loginuid? */
1265         if (loginuid == AUDIT_UID_UNSET) {
1266                 kloginuid = INVALID_UID;
1267         } else {
1268                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1269                 if (!uid_valid(kloginuid))
1270                         return -EINVAL;
1271         }
1272
1273         rv = audit_set_loginuid(kloginuid);
1274         if (rv < 0)
1275                 return rv;
1276         return count;
1277 }
1278
1279 static const struct file_operations proc_loginuid_operations = {
1280         .read           = proc_loginuid_read,
1281         .write          = proc_loginuid_write,
1282         .llseek         = generic_file_llseek,
1283 };
1284
1285 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1286                                   size_t count, loff_t *ppos)
1287 {
1288         struct inode * inode = file_inode(file);
1289         struct task_struct *task = get_proc_task(inode);
1290         ssize_t length;
1291         char tmpbuf[TMPBUFLEN];
1292
1293         if (!task)
1294                 return -ESRCH;
1295         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1296                                 audit_get_sessionid(task));
1297         put_task_struct(task);
1298         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1299 }
1300
1301 static const struct file_operations proc_sessionid_operations = {
1302         .read           = proc_sessionid_read,
1303         .llseek         = generic_file_llseek,
1304 };
1305 #endif
1306
1307 #ifdef CONFIG_FAULT_INJECTION
1308 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1309                                       size_t count, loff_t *ppos)
1310 {
1311         struct task_struct *task = get_proc_task(file_inode(file));
1312         char buffer[PROC_NUMBUF];
1313         size_t len;
1314         int make_it_fail;
1315
1316         if (!task)
1317                 return -ESRCH;
1318         make_it_fail = task->make_it_fail;
1319         put_task_struct(task);
1320
1321         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1322
1323         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1324 }
1325
1326 static ssize_t proc_fault_inject_write(struct file * file,
1327                         const char __user * buf, size_t count, loff_t *ppos)
1328 {
1329         struct task_struct *task;
1330         char buffer[PROC_NUMBUF];
1331         int make_it_fail;
1332         int rv;
1333
1334         if (!capable(CAP_SYS_RESOURCE))
1335                 return -EPERM;
1336         memset(buffer, 0, sizeof(buffer));
1337         if (count > sizeof(buffer) - 1)
1338                 count = sizeof(buffer) - 1;
1339         if (copy_from_user(buffer, buf, count))
1340                 return -EFAULT;
1341         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1342         if (rv < 0)
1343                 return rv;
1344         if (make_it_fail < 0 || make_it_fail > 1)
1345                 return -EINVAL;
1346
1347         task = get_proc_task(file_inode(file));
1348         if (!task)
1349                 return -ESRCH;
1350         task->make_it_fail = make_it_fail;
1351         put_task_struct(task);
1352
1353         return count;
1354 }
1355
1356 static const struct file_operations proc_fault_inject_operations = {
1357         .read           = proc_fault_inject_read,
1358         .write          = proc_fault_inject_write,
1359         .llseek         = generic_file_llseek,
1360 };
1361 #endif
1362
1363
1364 #ifdef CONFIG_SCHED_DEBUG
1365 /*
1366  * Print out various scheduling related per-task fields:
1367  */
1368 static int sched_show(struct seq_file *m, void *v)
1369 {
1370         struct inode *inode = m->private;
1371         struct task_struct *p;
1372
1373         p = get_proc_task(inode);
1374         if (!p)
1375                 return -ESRCH;
1376         proc_sched_show_task(p, m);
1377
1378         put_task_struct(p);
1379
1380         return 0;
1381 }
1382
1383 static ssize_t
1384 sched_write(struct file *file, const char __user *buf,
1385             size_t count, loff_t *offset)
1386 {
1387         struct inode *inode = file_inode(file);
1388         struct task_struct *p;
1389
1390         p = get_proc_task(inode);
1391         if (!p)
1392                 return -ESRCH;
1393         proc_sched_set_task(p);
1394
1395         put_task_struct(p);
1396
1397         return count;
1398 }
1399
1400 static int sched_open(struct inode *inode, struct file *filp)
1401 {
1402         return single_open(filp, sched_show, inode);
1403 }
1404
1405 static const struct file_operations proc_pid_sched_operations = {
1406         .open           = sched_open,
1407         .read           = seq_read,
1408         .write          = sched_write,
1409         .llseek         = seq_lseek,
1410         .release        = single_release,
1411 };
1412
1413 #endif
1414
1415 #ifdef CONFIG_SCHED_AUTOGROUP
1416 /*
1417  * Print out autogroup related information:
1418  */
1419 static int sched_autogroup_show(struct seq_file *m, void *v)
1420 {
1421         struct inode *inode = m->private;
1422         struct task_struct *p;
1423
1424         p = get_proc_task(inode);
1425         if (!p)
1426                 return -ESRCH;
1427         proc_sched_autogroup_show_task(p, m);
1428
1429         put_task_struct(p);
1430
1431         return 0;
1432 }
1433
1434 static ssize_t
1435 sched_autogroup_write(struct file *file, const char __user *buf,
1436             size_t count, loff_t *offset)
1437 {
1438         struct inode *inode = file_inode(file);
1439         struct task_struct *p;
1440         char buffer[PROC_NUMBUF];
1441         int nice;
1442         int err;
1443
1444         memset(buffer, 0, sizeof(buffer));
1445         if (count > sizeof(buffer) - 1)
1446                 count = sizeof(buffer) - 1;
1447         if (copy_from_user(buffer, buf, count))
1448                 return -EFAULT;
1449
1450         err = kstrtoint(strstrip(buffer), 0, &nice);
1451         if (err < 0)
1452                 return err;
1453
1454         p = get_proc_task(inode);
1455         if (!p)
1456                 return -ESRCH;
1457
1458         err = proc_sched_autogroup_set_nice(p, nice);
1459         if (err)
1460                 count = err;
1461
1462         put_task_struct(p);
1463
1464         return count;
1465 }
1466
1467 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1468 {
1469         int ret;
1470
1471         ret = single_open(filp, sched_autogroup_show, NULL);
1472         if (!ret) {
1473                 struct seq_file *m = filp->private_data;
1474
1475                 m->private = inode;
1476         }
1477         return ret;
1478 }
1479
1480 static const struct file_operations proc_pid_sched_autogroup_operations = {
1481         .open           = sched_autogroup_open,
1482         .read           = seq_read,
1483         .write          = sched_autogroup_write,
1484         .llseek         = seq_lseek,
1485         .release        = single_release,
1486 };
1487
1488 #endif /* CONFIG_SCHED_AUTOGROUP */
1489
1490 static ssize_t comm_write(struct file *file, const char __user *buf,
1491                                 size_t count, loff_t *offset)
1492 {
1493         struct inode *inode = file_inode(file);
1494         struct task_struct *p;
1495         char buffer[TASK_COMM_LEN];
1496         const size_t maxlen = sizeof(buffer) - 1;
1497
1498         memset(buffer, 0, sizeof(buffer));
1499         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1500                 return -EFAULT;
1501
1502         p = get_proc_task(inode);
1503         if (!p)
1504                 return -ESRCH;
1505
1506         if (same_thread_group(current, p))
1507                 set_task_comm(p, buffer);
1508         else
1509                 count = -EINVAL;
1510
1511         put_task_struct(p);
1512
1513         return count;
1514 }
1515
1516 static int comm_show(struct seq_file *m, void *v)
1517 {
1518         struct inode *inode = m->private;
1519         struct task_struct *p;
1520
1521         p = get_proc_task(inode);
1522         if (!p)
1523                 return -ESRCH;
1524
1525         task_lock(p);
1526         seq_printf(m, "%s\n", p->comm);
1527         task_unlock(p);
1528
1529         put_task_struct(p);
1530
1531         return 0;
1532 }
1533
1534 static int comm_open(struct inode *inode, struct file *filp)
1535 {
1536         return single_open(filp, comm_show, inode);
1537 }
1538
1539 static const struct file_operations proc_pid_set_comm_operations = {
1540         .open           = comm_open,
1541         .read           = seq_read,
1542         .write          = comm_write,
1543         .llseek         = seq_lseek,
1544         .release        = single_release,
1545 };
1546
1547 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1548 {
1549         struct task_struct *task;
1550         struct file *exe_file;
1551
1552         task = get_proc_task(d_inode(dentry));
1553         if (!task)
1554                 return -ENOENT;
1555         exe_file = get_task_exe_file(task);
1556         put_task_struct(task);
1557         if (exe_file) {
1558                 *exe_path = exe_file->f_path;
1559                 path_get(&exe_file->f_path);
1560                 fput(exe_file);
1561                 return 0;
1562         } else
1563                 return -ENOENT;
1564 }
1565
1566 static const char *proc_pid_get_link(struct dentry *dentry,
1567                                      struct inode *inode,
1568                                      struct delayed_call *done)
1569 {
1570         struct path path;
1571         int error = -EACCES;
1572
1573         if (!dentry)
1574                 return ERR_PTR(-ECHILD);
1575
1576         /* Are we allowed to snoop on the tasks file descriptors? */
1577         if (!proc_fd_access_allowed(inode))
1578                 goto out;
1579
1580         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1581         if (error)
1582                 goto out;
1583
1584         nd_jump_link(&path);
1585         return NULL;
1586 out:
1587         return ERR_PTR(error);
1588 }
1589
1590 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1591 {
1592         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1593         char *pathname;
1594         int len;
1595
1596         if (!tmp)
1597                 return -ENOMEM;
1598
1599         pathname = d_path(path, tmp, PAGE_SIZE);
1600         len = PTR_ERR(pathname);
1601         if (IS_ERR(pathname))
1602                 goto out;
1603         len = tmp + PAGE_SIZE - 1 - pathname;
1604
1605         if (len > buflen)
1606                 len = buflen;
1607         if (copy_to_user(buffer, pathname, len))
1608                 len = -EFAULT;
1609  out:
1610         free_page((unsigned long)tmp);
1611         return len;
1612 }
1613
1614 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1615 {
1616         int error = -EACCES;
1617         struct inode *inode = d_inode(dentry);
1618         struct path path;
1619
1620         /* Are we allowed to snoop on the tasks file descriptors? */
1621         if (!proc_fd_access_allowed(inode))
1622                 goto out;
1623
1624         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1625         if (error)
1626                 goto out;
1627
1628         error = do_proc_readlink(&path, buffer, buflen);
1629         path_put(&path);
1630 out:
1631         return error;
1632 }
1633
1634 const struct inode_operations proc_pid_link_inode_operations = {
1635         .readlink       = proc_pid_readlink,
1636         .get_link       = proc_pid_get_link,
1637         .setattr        = proc_setattr,
1638 };
1639
1640
1641 /* building an inode */
1642
1643 void task_dump_owner(struct task_struct *task, mode_t mode,
1644                      kuid_t *ruid, kgid_t *rgid)
1645 {
1646         /* Depending on the state of dumpable compute who should own a
1647          * proc file for a task.
1648          */
1649         const struct cred *cred;
1650         kuid_t uid;
1651         kgid_t gid;
1652
1653         /* Default to the tasks effective ownership */
1654         rcu_read_lock();
1655         cred = __task_cred(task);
1656         uid = cred->euid;
1657         gid = cred->egid;
1658         rcu_read_unlock();
1659
1660         /*
1661          * Before the /proc/pid/status file was created the only way to read
1662          * the effective uid of a /process was to stat /proc/pid.  Reading
1663          * /proc/pid/status is slow enough that procps and other packages
1664          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1665          * made this apply to all per process world readable and executable
1666          * directories.
1667          */
1668         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1669                 struct mm_struct *mm;
1670                 task_lock(task);
1671                 mm = task->mm;
1672                 /* Make non-dumpable tasks owned by some root */
1673                 if (mm) {
1674                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1675                                 struct user_namespace *user_ns = mm->user_ns;
1676
1677                                 uid = make_kuid(user_ns, 0);
1678                                 if (!uid_valid(uid))
1679                                         uid = GLOBAL_ROOT_UID;
1680
1681                                 gid = make_kgid(user_ns, 0);
1682                                 if (!gid_valid(gid))
1683                                         gid = GLOBAL_ROOT_GID;
1684                         }
1685                 } else {
1686                         uid = GLOBAL_ROOT_UID;
1687                         gid = GLOBAL_ROOT_GID;
1688                 }
1689                 task_unlock(task);
1690         }
1691         *ruid = uid;
1692         *rgid = gid;
1693 }
1694
1695 struct inode *proc_pid_make_inode(struct super_block * sb,
1696                                   struct task_struct *task, umode_t mode)
1697 {
1698         struct inode * inode;
1699         struct proc_inode *ei;
1700
1701         /* We need a new inode */
1702
1703         inode = new_inode(sb);
1704         if (!inode)
1705                 goto out;
1706
1707         /* Common stuff */
1708         ei = PROC_I(inode);
1709         inode->i_mode = mode;
1710         inode->i_ino = get_next_ino();
1711         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1712         inode->i_op = &proc_def_inode_operations;
1713
1714         /*
1715          * grab the reference to task.
1716          */
1717         ei->pid = get_task_pid(task, PIDTYPE_PID);
1718         if (!ei->pid)
1719                 goto out_unlock;
1720
1721         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1722         security_task_to_inode(task, inode);
1723
1724 out:
1725         return inode;
1726
1727 out_unlock:
1728         iput(inode);
1729         return NULL;
1730 }
1731
1732 int pid_getattr(const struct path *path, struct kstat *stat,
1733                 u32 request_mask, unsigned int query_flags)
1734 {
1735         struct inode *inode = d_inode(path->dentry);
1736         struct task_struct *task;
1737         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1738
1739         generic_fillattr(inode, stat);
1740
1741         rcu_read_lock();
1742         stat->uid = GLOBAL_ROOT_UID;
1743         stat->gid = GLOBAL_ROOT_GID;
1744         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1745         if (task) {
1746                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1747                         rcu_read_unlock();
1748                         /*
1749                          * This doesn't prevent learning whether PID exists,
1750                          * it only makes getattr() consistent with readdir().
1751                          */
1752                         return -ENOENT;
1753                 }
1754                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1755         }
1756         rcu_read_unlock();
1757         return 0;
1758 }
1759
1760 /* dentry stuff */
1761
1762 /*
1763  *      Exceptional case: normally we are not allowed to unhash a busy
1764  * directory. In this case, however, we can do it - no aliasing problems
1765  * due to the way we treat inodes.
1766  *
1767  * Rewrite the inode's ownerships here because the owning task may have
1768  * performed a setuid(), etc.
1769  *
1770  */
1771 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1772 {
1773         struct inode *inode;
1774         struct task_struct *task;
1775
1776         if (flags & LOOKUP_RCU)
1777                 return -ECHILD;
1778
1779         inode = d_inode(dentry);
1780         task = get_proc_task(inode);
1781
1782         if (task) {
1783                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1784
1785                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1786                 security_task_to_inode(task, inode);
1787                 put_task_struct(task);
1788                 return 1;
1789         }
1790         return 0;
1791 }
1792
1793 static inline bool proc_inode_is_dead(struct inode *inode)
1794 {
1795         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1796 }
1797
1798 int pid_delete_dentry(const struct dentry *dentry)
1799 {
1800         /* Is the task we represent dead?
1801          * If so, then don't put the dentry on the lru list,
1802          * kill it immediately.
1803          */
1804         return proc_inode_is_dead(d_inode(dentry));
1805 }
1806
1807 const struct dentry_operations pid_dentry_operations =
1808 {
1809         .d_revalidate   = pid_revalidate,
1810         .d_delete       = pid_delete_dentry,
1811 };
1812
1813 /* Lookups */
1814
1815 /*
1816  * Fill a directory entry.
1817  *
1818  * If possible create the dcache entry and derive our inode number and
1819  * file type from dcache entry.
1820  *
1821  * Since all of the proc inode numbers are dynamically generated, the inode
1822  * numbers do not exist until the inode is cache.  This means creating the
1823  * the dcache entry in readdir is necessary to keep the inode numbers
1824  * reported by readdir in sync with the inode numbers reported
1825  * by stat.
1826  */
1827 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1828         const char *name, int len,
1829         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1830 {
1831         struct dentry *child, *dir = file->f_path.dentry;
1832         struct qstr qname = QSTR_INIT(name, len);
1833         struct inode *inode;
1834         unsigned type;
1835         ino_t ino;
1836
1837         child = d_hash_and_lookup(dir, &qname);
1838         if (!child) {
1839                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1840                 child = d_alloc_parallel(dir, &qname, &wq);
1841                 if (IS_ERR(child))
1842                         goto end_instantiate;
1843                 if (d_in_lookup(child)) {
1844                         int err = instantiate(d_inode(dir), child, task, ptr);
1845                         d_lookup_done(child);
1846                         if (err < 0) {
1847                                 dput(child);
1848                                 goto end_instantiate;
1849                         }
1850                 }
1851         }
1852         inode = d_inode(child);
1853         ino = inode->i_ino;
1854         type = inode->i_mode >> 12;
1855         dput(child);
1856         return dir_emit(ctx, name, len, ino, type);
1857
1858 end_instantiate:
1859         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1860 }
1861
1862 /*
1863  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1864  * which represent vma start and end addresses.
1865  */
1866 static int dname_to_vma_addr(struct dentry *dentry,
1867                              unsigned long *start, unsigned long *end)
1868 {
1869         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1870                 return -EINVAL;
1871
1872         return 0;
1873 }
1874
1875 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1876 {
1877         unsigned long vm_start, vm_end;
1878         bool exact_vma_exists = false;
1879         struct mm_struct *mm = NULL;
1880         struct task_struct *task;
1881         struct inode *inode;
1882         int status = 0;
1883
1884         if (flags & LOOKUP_RCU)
1885                 return -ECHILD;
1886
1887         inode = d_inode(dentry);
1888         task = get_proc_task(inode);
1889         if (!task)
1890                 goto out_notask;
1891
1892         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1893         if (IS_ERR_OR_NULL(mm))
1894                 goto out;
1895
1896         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1897                 down_read(&mm->mmap_sem);
1898                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1899                 up_read(&mm->mmap_sem);
1900         }
1901
1902         mmput(mm);
1903
1904         if (exact_vma_exists) {
1905                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1906
1907                 security_task_to_inode(task, inode);
1908                 status = 1;
1909         }
1910
1911 out:
1912         put_task_struct(task);
1913
1914 out_notask:
1915         return status;
1916 }
1917
1918 static const struct dentry_operations tid_map_files_dentry_operations = {
1919         .d_revalidate   = map_files_d_revalidate,
1920         .d_delete       = pid_delete_dentry,
1921 };
1922
1923 static int map_files_get_link(struct dentry *dentry, struct path *path)
1924 {
1925         unsigned long vm_start, vm_end;
1926         struct vm_area_struct *vma;
1927         struct task_struct *task;
1928         struct mm_struct *mm;
1929         int rc;
1930
1931         rc = -ENOENT;
1932         task = get_proc_task(d_inode(dentry));
1933         if (!task)
1934                 goto out;
1935
1936         mm = get_task_mm(task);
1937         put_task_struct(task);
1938         if (!mm)
1939                 goto out;
1940
1941         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1942         if (rc)
1943                 goto out_mmput;
1944
1945         rc = -ENOENT;
1946         down_read(&mm->mmap_sem);
1947         vma = find_exact_vma(mm, vm_start, vm_end);
1948         if (vma && vma->vm_file) {
1949                 *path = vma->vm_file->f_path;
1950                 path_get(path);
1951                 rc = 0;
1952         }
1953         up_read(&mm->mmap_sem);
1954
1955 out_mmput:
1956         mmput(mm);
1957 out:
1958         return rc;
1959 }
1960
1961 struct map_files_info {
1962         fmode_t         mode;
1963         unsigned int    len;
1964         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1965 };
1966
1967 /*
1968  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1969  * symlinks may be used to bypass permissions on ancestor directories in the
1970  * path to the file in question.
1971  */
1972 static const char *
1973 proc_map_files_get_link(struct dentry *dentry,
1974                         struct inode *inode,
1975                         struct delayed_call *done)
1976 {
1977         if (!capable(CAP_SYS_ADMIN))
1978                 return ERR_PTR(-EPERM);
1979
1980         return proc_pid_get_link(dentry, inode, done);
1981 }
1982
1983 /*
1984  * Identical to proc_pid_link_inode_operations except for get_link()
1985  */
1986 static const struct inode_operations proc_map_files_link_inode_operations = {
1987         .readlink       = proc_pid_readlink,
1988         .get_link       = proc_map_files_get_link,
1989         .setattr        = proc_setattr,
1990 };
1991
1992 static int
1993 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1994                            struct task_struct *task, const void *ptr)
1995 {
1996         fmode_t mode = (fmode_t)(unsigned long)ptr;
1997         struct proc_inode *ei;
1998         struct inode *inode;
1999
2000         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2001                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2002                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2003         if (!inode)
2004                 return -ENOENT;
2005
2006         ei = PROC_I(inode);
2007         ei->op.proc_get_link = map_files_get_link;
2008
2009         inode->i_op = &proc_map_files_link_inode_operations;
2010         inode->i_size = 64;
2011
2012         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2013         d_add(dentry, inode);
2014
2015         return 0;
2016 }
2017
2018 static struct dentry *proc_map_files_lookup(struct inode *dir,
2019                 struct dentry *dentry, unsigned int flags)
2020 {
2021         unsigned long vm_start, vm_end;
2022         struct vm_area_struct *vma;
2023         struct task_struct *task;
2024         int result;
2025         struct mm_struct *mm;
2026
2027         result = -ENOENT;
2028         task = get_proc_task(dir);
2029         if (!task)
2030                 goto out;
2031
2032         result = -EACCES;
2033         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2034                 goto out_put_task;
2035
2036         result = -ENOENT;
2037         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2038                 goto out_put_task;
2039
2040         mm = get_task_mm(task);
2041         if (!mm)
2042                 goto out_put_task;
2043
2044         down_read(&mm->mmap_sem);
2045         vma = find_exact_vma(mm, vm_start, vm_end);
2046         if (!vma)
2047                 goto out_no_vma;
2048
2049         if (vma->vm_file)
2050                 result = proc_map_files_instantiate(dir, dentry, task,
2051                                 (void *)(unsigned long)vma->vm_file->f_mode);
2052
2053 out_no_vma:
2054         up_read(&mm->mmap_sem);
2055         mmput(mm);
2056 out_put_task:
2057         put_task_struct(task);
2058 out:
2059         return ERR_PTR(result);
2060 }
2061
2062 static const struct inode_operations proc_map_files_inode_operations = {
2063         .lookup         = proc_map_files_lookup,
2064         .permission     = proc_fd_permission,
2065         .setattr        = proc_setattr,
2066 };
2067
2068 static int
2069 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2070 {
2071         struct vm_area_struct *vma;
2072         struct task_struct *task;
2073         struct mm_struct *mm;
2074         unsigned long nr_files, pos, i;
2075         struct flex_array *fa = NULL;
2076         struct map_files_info info;
2077         struct map_files_info *p;
2078         int ret;
2079
2080         ret = -ENOENT;
2081         task = get_proc_task(file_inode(file));
2082         if (!task)
2083                 goto out;
2084
2085         ret = -EACCES;
2086         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2087                 goto out_put_task;
2088
2089         ret = 0;
2090         if (!dir_emit_dots(file, ctx))
2091                 goto out_put_task;
2092
2093         mm = get_task_mm(task);
2094         if (!mm)
2095                 goto out_put_task;
2096         down_read(&mm->mmap_sem);
2097
2098         nr_files = 0;
2099
2100         /*
2101          * We need two passes here:
2102          *
2103          *  1) Collect vmas of mapped files with mmap_sem taken
2104          *  2) Release mmap_sem and instantiate entries
2105          *
2106          * otherwise we get lockdep complained, since filldir()
2107          * routine might require mmap_sem taken in might_fault().
2108          */
2109
2110         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2111                 if (vma->vm_file && ++pos > ctx->pos)
2112                         nr_files++;
2113         }
2114
2115         if (nr_files) {
2116                 fa = flex_array_alloc(sizeof(info), nr_files,
2117                                         GFP_KERNEL);
2118                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2119                                                 GFP_KERNEL)) {
2120                         ret = -ENOMEM;
2121                         if (fa)
2122                                 flex_array_free(fa);
2123                         up_read(&mm->mmap_sem);
2124                         mmput(mm);
2125                         goto out_put_task;
2126                 }
2127                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2128                                 vma = vma->vm_next) {
2129                         if (!vma->vm_file)
2130                                 continue;
2131                         if (++pos <= ctx->pos)
2132                                 continue;
2133
2134                         info.mode = vma->vm_file->f_mode;
2135                         info.len = snprintf(info.name,
2136                                         sizeof(info.name), "%lx-%lx",
2137                                         vma->vm_start, vma->vm_end);
2138                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2139                                 BUG();
2140                 }
2141         }
2142         up_read(&mm->mmap_sem);
2143
2144         for (i = 0; i < nr_files; i++) {
2145                 p = flex_array_get(fa, i);
2146                 if (!proc_fill_cache(file, ctx,
2147                                       p->name, p->len,
2148                                       proc_map_files_instantiate,
2149                                       task,
2150                                       (void *)(unsigned long)p->mode))
2151                         break;
2152                 ctx->pos++;
2153         }
2154         if (fa)
2155                 flex_array_free(fa);
2156         mmput(mm);
2157
2158 out_put_task:
2159         put_task_struct(task);
2160 out:
2161         return ret;
2162 }
2163
2164 static const struct file_operations proc_map_files_operations = {
2165         .read           = generic_read_dir,
2166         .iterate_shared = proc_map_files_readdir,
2167         .llseek         = generic_file_llseek,
2168 };
2169
2170 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2171 struct timers_private {
2172         struct pid *pid;
2173         struct task_struct *task;
2174         struct sighand_struct *sighand;
2175         struct pid_namespace *ns;
2176         unsigned long flags;
2177 };
2178
2179 static void *timers_start(struct seq_file *m, loff_t *pos)
2180 {
2181         struct timers_private *tp = m->private;
2182
2183         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2184         if (!tp->task)
2185                 return ERR_PTR(-ESRCH);
2186
2187         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2188         if (!tp->sighand)
2189                 return ERR_PTR(-ESRCH);
2190
2191         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2192 }
2193
2194 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2195 {
2196         struct timers_private *tp = m->private;
2197         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2198 }
2199
2200 static void timers_stop(struct seq_file *m, void *v)
2201 {
2202         struct timers_private *tp = m->private;
2203
2204         if (tp->sighand) {
2205                 unlock_task_sighand(tp->task, &tp->flags);
2206                 tp->sighand = NULL;
2207         }
2208
2209         if (tp->task) {
2210                 put_task_struct(tp->task);
2211                 tp->task = NULL;
2212         }
2213 }
2214
2215 static int show_timer(struct seq_file *m, void *v)
2216 {
2217         struct k_itimer *timer;
2218         struct timers_private *tp = m->private;
2219         int notify;
2220         static const char * const nstr[] = {
2221                 [SIGEV_SIGNAL] = "signal",
2222                 [SIGEV_NONE] = "none",
2223                 [SIGEV_THREAD] = "thread",
2224         };
2225
2226         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2227         notify = timer->it_sigev_notify;
2228
2229         seq_printf(m, "ID: %d\n", timer->it_id);
2230         seq_printf(m, "signal: %d/%p\n",
2231                    timer->sigq->info.si_signo,
2232                    timer->sigq->info.si_value.sival_ptr);
2233         seq_printf(m, "notify: %s/%s.%d\n",
2234                    nstr[notify & ~SIGEV_THREAD_ID],
2235                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2236                    pid_nr_ns(timer->it_pid, tp->ns));
2237         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2238
2239         return 0;
2240 }
2241
2242 static const struct seq_operations proc_timers_seq_ops = {
2243         .start  = timers_start,
2244         .next   = timers_next,
2245         .stop   = timers_stop,
2246         .show   = show_timer,
2247 };
2248
2249 static int proc_timers_open(struct inode *inode, struct file *file)
2250 {
2251         struct timers_private *tp;
2252
2253         tp = __seq_open_private(file, &proc_timers_seq_ops,
2254                         sizeof(struct timers_private));
2255         if (!tp)
2256                 return -ENOMEM;
2257
2258         tp->pid = proc_pid(inode);
2259         tp->ns = inode->i_sb->s_fs_info;
2260         return 0;
2261 }
2262
2263 static const struct file_operations proc_timers_operations = {
2264         .open           = proc_timers_open,
2265         .read           = seq_read,
2266         .llseek         = seq_lseek,
2267         .release        = seq_release_private,
2268 };
2269 #endif
2270
2271 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2272                                         size_t count, loff_t *offset)
2273 {
2274         struct inode *inode = file_inode(file);
2275         struct task_struct *p;
2276         u64 slack_ns;
2277         int err;
2278
2279         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2280         if (err < 0)
2281                 return err;
2282
2283         p = get_proc_task(inode);
2284         if (!p)
2285                 return -ESRCH;
2286
2287         if (p != current) {
2288                 if (!capable(CAP_SYS_NICE)) {
2289                         count = -EPERM;
2290                         goto out;
2291                 }
2292
2293                 err = security_task_setscheduler(p);
2294                 if (err) {
2295                         count = err;
2296                         goto out;
2297                 }
2298         }
2299
2300         task_lock(p);
2301         if (slack_ns == 0)
2302                 p->timer_slack_ns = p->default_timer_slack_ns;
2303         else
2304                 p->timer_slack_ns = slack_ns;
2305         task_unlock(p);
2306
2307 out:
2308         put_task_struct(p);
2309
2310         return count;
2311 }
2312
2313 static int timerslack_ns_show(struct seq_file *m, void *v)
2314 {
2315         struct inode *inode = m->private;
2316         struct task_struct *p;
2317         int err = 0;
2318
2319         p = get_proc_task(inode);
2320         if (!p)
2321                 return -ESRCH;
2322
2323         if (p != current) {
2324
2325                 if (!capable(CAP_SYS_NICE)) {
2326                         err = -EPERM;
2327                         goto out;
2328                 }
2329                 err = security_task_getscheduler(p);
2330                 if (err)
2331                         goto out;
2332         }
2333
2334         task_lock(p);
2335         seq_printf(m, "%llu\n", p->timer_slack_ns);
2336         task_unlock(p);
2337
2338 out:
2339         put_task_struct(p);
2340
2341         return err;
2342 }
2343
2344 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2345 {
2346         return single_open(filp, timerslack_ns_show, inode);
2347 }
2348
2349 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2350         .open           = timerslack_ns_open,
2351         .read           = seq_read,
2352         .write          = timerslack_ns_write,
2353         .llseek         = seq_lseek,
2354         .release        = single_release,
2355 };
2356
2357 static int proc_pident_instantiate(struct inode *dir,
2358         struct dentry *dentry, struct task_struct *task, const void *ptr)
2359 {
2360         const struct pid_entry *p = ptr;
2361         struct inode *inode;
2362         struct proc_inode *ei;
2363
2364         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2365         if (!inode)
2366                 goto out;
2367
2368         ei = PROC_I(inode);
2369         if (S_ISDIR(inode->i_mode))
2370                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2371         if (p->iop)
2372                 inode->i_op = p->iop;
2373         if (p->fop)
2374                 inode->i_fop = p->fop;
2375         ei->op = p->op;
2376         d_set_d_op(dentry, &pid_dentry_operations);
2377         d_add(dentry, inode);
2378         /* Close the race of the process dying before we return the dentry */
2379         if (pid_revalidate(dentry, 0))
2380                 return 0;
2381 out:
2382         return -ENOENT;
2383 }
2384
2385 static struct dentry *proc_pident_lookup(struct inode *dir,
2386                                          struct dentry *dentry,
2387                                          const struct pid_entry *ents,
2388                                          unsigned int nents)
2389 {
2390         int error;
2391         struct task_struct *task = get_proc_task(dir);
2392         const struct pid_entry *p, *last;
2393
2394         error = -ENOENT;
2395
2396         if (!task)
2397                 goto out_no_task;
2398
2399         /*
2400          * Yes, it does not scale. And it should not. Don't add
2401          * new entries into /proc/<tgid>/ without very good reasons.
2402          */
2403         last = &ents[nents];
2404         for (p = ents; p < last; p++) {
2405                 if (p->len != dentry->d_name.len)
2406                         continue;
2407                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2408                         break;
2409         }
2410         if (p >= last)
2411                 goto out;
2412
2413         error = proc_pident_instantiate(dir, dentry, task, p);
2414 out:
2415         put_task_struct(task);
2416 out_no_task:
2417         return ERR_PTR(error);
2418 }
2419
2420 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2421                 const struct pid_entry *ents, unsigned int nents)
2422 {
2423         struct task_struct *task = get_proc_task(file_inode(file));
2424         const struct pid_entry *p;
2425
2426         if (!task)
2427                 return -ENOENT;
2428
2429         if (!dir_emit_dots(file, ctx))
2430                 goto out;
2431
2432         if (ctx->pos >= nents + 2)
2433                 goto out;
2434
2435         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2436                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2437                                 proc_pident_instantiate, task, p))
2438                         break;
2439                 ctx->pos++;
2440         }
2441 out:
2442         put_task_struct(task);
2443         return 0;
2444 }
2445
2446 #ifdef CONFIG_SECURITY
2447 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2448                                   size_t count, loff_t *ppos)
2449 {
2450         struct inode * inode = file_inode(file);
2451         char *p = NULL;
2452         ssize_t length;
2453         struct task_struct *task = get_proc_task(inode);
2454
2455         if (!task)
2456                 return -ESRCH;
2457
2458         length = security_getprocattr(task,
2459                                       (char*)file->f_path.dentry->d_name.name,
2460                                       &p);
2461         put_task_struct(task);
2462         if (length > 0)
2463                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2464         kfree(p);
2465         return length;
2466 }
2467
2468 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2469                                    size_t count, loff_t *ppos)
2470 {
2471         struct inode * inode = file_inode(file);
2472         void *page;
2473         ssize_t length;
2474         struct task_struct *task = get_proc_task(inode);
2475
2476         length = -ESRCH;
2477         if (!task)
2478                 goto out_no_task;
2479
2480         /* A task may only write its own attributes. */
2481         length = -EACCES;
2482         if (current != task)
2483                 goto out;
2484
2485         if (count > PAGE_SIZE)
2486                 count = PAGE_SIZE;
2487
2488         /* No partial writes. */
2489         length = -EINVAL;
2490         if (*ppos != 0)
2491                 goto out;
2492
2493         page = memdup_user(buf, count);
2494         if (IS_ERR(page)) {
2495                 length = PTR_ERR(page);
2496                 goto out;
2497         }
2498
2499         /* Guard against adverse ptrace interaction */
2500         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2501         if (length < 0)
2502                 goto out_free;
2503
2504         length = security_setprocattr(file->f_path.dentry->d_name.name,
2505                                       page, count);
2506         mutex_unlock(&current->signal->cred_guard_mutex);
2507 out_free:
2508         kfree(page);
2509 out:
2510         put_task_struct(task);
2511 out_no_task:
2512         return length;
2513 }
2514
2515 static const struct file_operations proc_pid_attr_operations = {
2516         .read           = proc_pid_attr_read,
2517         .write          = proc_pid_attr_write,
2518         .llseek         = generic_file_llseek,
2519 };
2520
2521 static const struct pid_entry attr_dir_stuff[] = {
2522         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2523         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2524         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2525         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2526         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2527         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2528 };
2529
2530 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2531 {
2532         return proc_pident_readdir(file, ctx,
2533                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2534 }
2535
2536 static const struct file_operations proc_attr_dir_operations = {
2537         .read           = generic_read_dir,
2538         .iterate_shared = proc_attr_dir_readdir,
2539         .llseek         = generic_file_llseek,
2540 };
2541
2542 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2543                                 struct dentry *dentry, unsigned int flags)
2544 {
2545         return proc_pident_lookup(dir, dentry,
2546                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2547 }
2548
2549 static const struct inode_operations proc_attr_dir_inode_operations = {
2550         .lookup         = proc_attr_dir_lookup,
2551         .getattr        = pid_getattr,
2552         .setattr        = proc_setattr,
2553 };
2554
2555 #endif
2556
2557 #ifdef CONFIG_ELF_CORE
2558 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2559                                          size_t count, loff_t *ppos)
2560 {
2561         struct task_struct *task = get_proc_task(file_inode(file));
2562         struct mm_struct *mm;
2563         char buffer[PROC_NUMBUF];
2564         size_t len;
2565         int ret;
2566
2567         if (!task)
2568                 return -ESRCH;
2569
2570         ret = 0;
2571         mm = get_task_mm(task);
2572         if (mm) {
2573                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2574                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2575                                 MMF_DUMP_FILTER_SHIFT));
2576                 mmput(mm);
2577                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2578         }
2579
2580         put_task_struct(task);
2581
2582         return ret;
2583 }
2584
2585 static ssize_t proc_coredump_filter_write(struct file *file,
2586                                           const char __user *buf,
2587                                           size_t count,
2588                                           loff_t *ppos)
2589 {
2590         struct task_struct *task;
2591         struct mm_struct *mm;
2592         unsigned int val;
2593         int ret;
2594         int i;
2595         unsigned long mask;
2596
2597         ret = kstrtouint_from_user(buf, count, 0, &val);
2598         if (ret < 0)
2599                 return ret;
2600
2601         ret = -ESRCH;
2602         task = get_proc_task(file_inode(file));
2603         if (!task)
2604                 goto out_no_task;
2605
2606         mm = get_task_mm(task);
2607         if (!mm)
2608                 goto out_no_mm;
2609         ret = 0;
2610
2611         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2612                 if (val & mask)
2613                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2614                 else
2615                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2616         }
2617
2618         mmput(mm);
2619  out_no_mm:
2620         put_task_struct(task);
2621  out_no_task:
2622         if (ret < 0)
2623                 return ret;
2624         return count;
2625 }
2626
2627 static const struct file_operations proc_coredump_filter_operations = {
2628         .read           = proc_coredump_filter_read,
2629         .write          = proc_coredump_filter_write,
2630         .llseek         = generic_file_llseek,
2631 };
2632 #endif
2633
2634 #ifdef CONFIG_TASK_IO_ACCOUNTING
2635 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2636 {
2637         struct task_io_accounting acct = task->ioac;
2638         unsigned long flags;
2639         int result;
2640
2641         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2642         if (result)
2643                 return result;
2644
2645         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2646                 result = -EACCES;
2647                 goto out_unlock;
2648         }
2649
2650         if (whole && lock_task_sighand(task, &flags)) {
2651                 struct task_struct *t = task;
2652
2653                 task_io_accounting_add(&acct, &task->signal->ioac);
2654                 while_each_thread(task, t)
2655                         task_io_accounting_add(&acct, &t->ioac);
2656
2657                 unlock_task_sighand(task, &flags);
2658         }
2659         seq_printf(m,
2660                    "rchar: %llu\n"
2661                    "wchar: %llu\n"
2662                    "syscr: %llu\n"
2663                    "syscw: %llu\n"
2664                    "read_bytes: %llu\n"
2665                    "write_bytes: %llu\n"
2666                    "cancelled_write_bytes: %llu\n",
2667                    (unsigned long long)acct.rchar,
2668                    (unsigned long long)acct.wchar,
2669                    (unsigned long long)acct.syscr,
2670                    (unsigned long long)acct.syscw,
2671                    (unsigned long long)acct.read_bytes,
2672                    (unsigned long long)acct.write_bytes,
2673                    (unsigned long long)acct.cancelled_write_bytes);
2674         result = 0;
2675
2676 out_unlock:
2677         mutex_unlock(&task->signal->cred_guard_mutex);
2678         return result;
2679 }
2680
2681 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2682                                   struct pid *pid, struct task_struct *task)
2683 {
2684         return do_io_accounting(task, m, 0);
2685 }
2686
2687 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2688                                    struct pid *pid, struct task_struct *task)
2689 {
2690         return do_io_accounting(task, m, 1);
2691 }
2692 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2693
2694 #ifdef CONFIG_USER_NS
2695 static int proc_id_map_open(struct inode *inode, struct file *file,
2696         const struct seq_operations *seq_ops)
2697 {
2698         struct user_namespace *ns = NULL;
2699         struct task_struct *task;
2700         struct seq_file *seq;
2701         int ret = -EINVAL;
2702
2703         task = get_proc_task(inode);
2704         if (task) {
2705                 rcu_read_lock();
2706                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2707                 rcu_read_unlock();
2708                 put_task_struct(task);
2709         }
2710         if (!ns)
2711                 goto err;
2712
2713         ret = seq_open(file, seq_ops);
2714         if (ret)
2715                 goto err_put_ns;
2716
2717         seq = file->private_data;
2718         seq->private = ns;
2719
2720         return 0;
2721 err_put_ns:
2722         put_user_ns(ns);
2723 err:
2724         return ret;
2725 }
2726
2727 static int proc_id_map_release(struct inode *inode, struct file *file)
2728 {
2729         struct seq_file *seq = file->private_data;
2730         struct user_namespace *ns = seq->private;
2731         put_user_ns(ns);
2732         return seq_release(inode, file);
2733 }
2734
2735 static int proc_uid_map_open(struct inode *inode, struct file *file)
2736 {
2737         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2738 }
2739
2740 static int proc_gid_map_open(struct inode *inode, struct file *file)
2741 {
2742         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2743 }
2744
2745 static int proc_projid_map_open(struct inode *inode, struct file *file)
2746 {
2747         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2748 }
2749
2750 static const struct file_operations proc_uid_map_operations = {
2751         .open           = proc_uid_map_open,
2752         .write          = proc_uid_map_write,
2753         .read           = seq_read,
2754         .llseek         = seq_lseek,
2755         .release        = proc_id_map_release,
2756 };
2757
2758 static const struct file_operations proc_gid_map_operations = {
2759         .open           = proc_gid_map_open,
2760         .write          = proc_gid_map_write,
2761         .read           = seq_read,
2762         .llseek         = seq_lseek,
2763         .release        = proc_id_map_release,
2764 };
2765
2766 static const struct file_operations proc_projid_map_operations = {
2767         .open           = proc_projid_map_open,
2768         .write          = proc_projid_map_write,
2769         .read           = seq_read,
2770         .llseek         = seq_lseek,
2771         .release        = proc_id_map_release,
2772 };
2773
2774 static int proc_setgroups_open(struct inode *inode, struct file *file)
2775 {
2776         struct user_namespace *ns = NULL;
2777         struct task_struct *task;
2778         int ret;
2779
2780         ret = -ESRCH;
2781         task = get_proc_task(inode);
2782         if (task) {
2783                 rcu_read_lock();
2784                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2785                 rcu_read_unlock();
2786                 put_task_struct(task);
2787         }
2788         if (!ns)
2789                 goto err;
2790
2791         if (file->f_mode & FMODE_WRITE) {
2792                 ret = -EACCES;
2793                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2794                         goto err_put_ns;
2795         }
2796
2797         ret = single_open(file, &proc_setgroups_show, ns);
2798         if (ret)
2799                 goto err_put_ns;
2800
2801         return 0;
2802 err_put_ns:
2803         put_user_ns(ns);
2804 err:
2805         return ret;
2806 }
2807
2808 static int proc_setgroups_release(struct inode *inode, struct file *file)
2809 {
2810         struct seq_file *seq = file->private_data;
2811         struct user_namespace *ns = seq->private;
2812         int ret = single_release(inode, file);
2813         put_user_ns(ns);
2814         return ret;
2815 }
2816
2817 static const struct file_operations proc_setgroups_operations = {
2818         .open           = proc_setgroups_open,
2819         .write          = proc_setgroups_write,
2820         .read           = seq_read,
2821         .llseek         = seq_lseek,
2822         .release        = proc_setgroups_release,
2823 };
2824 #endif /* CONFIG_USER_NS */
2825
2826 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2827                                 struct pid *pid, struct task_struct *task)
2828 {
2829         int err = lock_trace(task);
2830         if (!err) {
2831                 seq_printf(m, "%08x\n", task->personality);
2832                 unlock_trace(task);
2833         }
2834         return err;
2835 }
2836
2837 #ifdef CONFIG_LIVEPATCH
2838 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2839                                 struct pid *pid, struct task_struct *task)
2840 {
2841         seq_printf(m, "%d\n", task->patch_state);
2842         return 0;
2843 }
2844 #endif /* CONFIG_LIVEPATCH */
2845
2846 /*
2847  * Thread groups
2848  */
2849 static const struct file_operations proc_task_operations;
2850 static const struct inode_operations proc_task_inode_operations;
2851
2852 static const struct pid_entry tgid_base_stuff[] = {
2853         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2854         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2855         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2856         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2857         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2858 #ifdef CONFIG_NET
2859         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2860 #endif
2861         REG("environ",    S_IRUSR, proc_environ_operations),
2862         REG("auxv",       S_IRUSR, proc_auxv_operations),
2863         ONE("status",     S_IRUGO, proc_pid_status),
2864         ONE("personality", S_IRUSR, proc_pid_personality),
2865         ONE("limits",     S_IRUGO, proc_pid_limits),
2866 #ifdef CONFIG_SCHED_DEBUG
2867         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2868 #endif
2869 #ifdef CONFIG_SCHED_AUTOGROUP
2870         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2871 #endif
2872         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2873 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2874         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2875 #endif
2876         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2877         ONE("stat",       S_IRUGO, proc_tgid_stat),
2878         ONE("statm",      S_IRUGO, proc_pid_statm),
2879         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2880 #ifdef CONFIG_NUMA
2881         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2882 #endif
2883         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2884         LNK("cwd",        proc_cwd_link),
2885         LNK("root",       proc_root_link),
2886         LNK("exe",        proc_exe_link),
2887         REG("mounts",     S_IRUGO, proc_mounts_operations),
2888         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2889         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2890 #ifdef CONFIG_PROC_PAGE_MONITOR
2891         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2892         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2893         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2894 #endif
2895 #ifdef CONFIG_SECURITY
2896         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2897 #endif
2898 #ifdef CONFIG_KALLSYMS
2899         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2900 #endif
2901 #ifdef CONFIG_STACKTRACE
2902         ONE("stack",      S_IRUSR, proc_pid_stack),
2903 #endif
2904 #ifdef CONFIG_SCHED_INFO
2905         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2906 #endif
2907 #ifdef CONFIG_LATENCYTOP
2908         REG("latency",  S_IRUGO, proc_lstats_operations),
2909 #endif
2910 #ifdef CONFIG_PROC_PID_CPUSET
2911         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2912 #endif
2913 #ifdef CONFIG_CGROUPS
2914         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2915 #endif
2916         ONE("oom_score",  S_IRUGO, proc_oom_score),
2917         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2918         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2919 #ifdef CONFIG_AUDITSYSCALL
2920         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2921         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2922 #endif
2923 #ifdef CONFIG_FAULT_INJECTION
2924         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2925 #endif
2926 #ifdef CONFIG_ELF_CORE
2927         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2928 #endif
2929 #ifdef CONFIG_TASK_IO_ACCOUNTING
2930         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2931 #endif
2932 #ifdef CONFIG_HARDWALL
2933         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2934 #endif
2935 #ifdef CONFIG_USER_NS
2936         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2937         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2938         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2939         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2940 #endif
2941 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2942         REG("timers",     S_IRUGO, proc_timers_operations),
2943 #endif
2944         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2945 #ifdef CONFIG_LIVEPATCH
2946         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2947 #endif
2948 };
2949
2950 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2951 {
2952         return proc_pident_readdir(file, ctx,
2953                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2954 }
2955
2956 static const struct file_operations proc_tgid_base_operations = {
2957         .read           = generic_read_dir,
2958         .iterate_shared = proc_tgid_base_readdir,
2959         .llseek         = generic_file_llseek,
2960 };
2961
2962 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2963 {
2964         return proc_pident_lookup(dir, dentry,
2965                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2966 }
2967
2968 static const struct inode_operations proc_tgid_base_inode_operations = {
2969         .lookup         = proc_tgid_base_lookup,
2970         .getattr        = pid_getattr,
2971         .setattr        = proc_setattr,
2972         .permission     = proc_pid_permission,
2973 };
2974
2975 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2976 {
2977         struct dentry *dentry, *leader, *dir;
2978         char buf[PROC_NUMBUF];
2979         struct qstr name;
2980
2981         name.name = buf;
2982         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2983         /* no ->d_hash() rejects on procfs */
2984         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2985         if (dentry) {
2986                 d_invalidate(dentry);
2987                 dput(dentry);
2988         }
2989
2990         if (pid == tgid)
2991                 return;
2992
2993         name.name = buf;
2994         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2995         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2996         if (!leader)
2997                 goto out;
2998
2999         name.name = "task";
3000         name.len = strlen(name.name);
3001         dir = d_hash_and_lookup(leader, &name);
3002         if (!dir)
3003                 goto out_put_leader;
3004
3005         name.name = buf;
3006         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3007         dentry = d_hash_and_lookup(dir, &name);
3008         if (dentry) {
3009                 d_invalidate(dentry);
3010                 dput(dentry);
3011         }
3012
3013         dput(dir);
3014 out_put_leader:
3015         dput(leader);
3016 out:
3017         return;
3018 }
3019
3020 /**
3021  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3022  * @task: task that should be flushed.
3023  *
3024  * When flushing dentries from proc, one needs to flush them from global
3025  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3026  * in. This call is supposed to do all of this job.
3027  *
3028  * Looks in the dcache for
3029  * /proc/@pid
3030  * /proc/@tgid/task/@pid
3031  * if either directory is present flushes it and all of it'ts children
3032  * from the dcache.
3033  *
3034  * It is safe and reasonable to cache /proc entries for a task until
3035  * that task exits.  After that they just clog up the dcache with
3036  * useless entries, possibly causing useful dcache entries to be
3037  * flushed instead.  This routine is proved to flush those useless
3038  * dcache entries at process exit time.
3039  *
3040  * NOTE: This routine is just an optimization so it does not guarantee
3041  *       that no dcache entries will exist at process exit time it
3042  *       just makes it very unlikely that any will persist.
3043  */
3044
3045 void proc_flush_task(struct task_struct *task)
3046 {
3047         int i;
3048         struct pid *pid, *tgid;
3049         struct upid *upid;
3050
3051         pid = task_pid(task);
3052         tgid = task_tgid(task);
3053
3054         for (i = 0; i <= pid->level; i++) {
3055                 upid = &pid->numbers[i];
3056                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3057                                         tgid->numbers[i].nr);
3058         }
3059 }
3060
3061 static int proc_pid_instantiate(struct inode *dir,
3062                                    struct dentry * dentry,
3063                                    struct task_struct *task, const void *ptr)
3064 {
3065         struct inode *inode;
3066
3067         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3068         if (!inode)
3069                 goto out;
3070
3071         inode->i_op = &proc_tgid_base_inode_operations;
3072         inode->i_fop = &proc_tgid_base_operations;
3073         inode->i_flags|=S_IMMUTABLE;
3074
3075         set_nlink(inode, nlink_tgid);
3076
3077         d_set_d_op(dentry, &pid_dentry_operations);
3078
3079         d_add(dentry, inode);
3080         /* Close the race of the process dying before we return the dentry */
3081         if (pid_revalidate(dentry, 0))
3082                 return 0;
3083 out:
3084         return -ENOENT;
3085 }
3086
3087 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3088 {
3089         int result = -ENOENT;
3090         struct task_struct *task;
3091         unsigned tgid;
3092         struct pid_namespace *ns;
3093
3094         tgid = name_to_int(&dentry->d_name);
3095         if (tgid == ~0U)
3096                 goto out;
3097
3098         ns = dentry->d_sb->s_fs_info;
3099         rcu_read_lock();
3100         task = find_task_by_pid_ns(tgid, ns);
3101         if (task)
3102                 get_task_struct(task);
3103         rcu_read_unlock();
3104         if (!task)
3105                 goto out;
3106
3107         result = proc_pid_instantiate(dir, dentry, task, NULL);
3108         put_task_struct(task);
3109 out:
3110         return ERR_PTR(result);
3111 }
3112
3113 /*
3114  * Find the first task with tgid >= tgid
3115  *
3116  */
3117 struct tgid_iter {
3118         unsigned int tgid;
3119         struct task_struct *task;
3120 };
3121 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3122 {
3123         struct pid *pid;
3124
3125         if (iter.task)
3126                 put_task_struct(iter.task);
3127         rcu_read_lock();
3128 retry:
3129         iter.task = NULL;
3130         pid = find_ge_pid(iter.tgid, ns);
3131         if (pid) {
3132                 iter.tgid = pid_nr_ns(pid, ns);
3133                 iter.task = pid_task(pid, PIDTYPE_PID);
3134                 /* What we to know is if the pid we have find is the
3135                  * pid of a thread_group_leader.  Testing for task
3136                  * being a thread_group_leader is the obvious thing
3137                  * todo but there is a window when it fails, due to
3138                  * the pid transfer logic in de_thread.
3139                  *
3140                  * So we perform the straight forward test of seeing
3141                  * if the pid we have found is the pid of a thread
3142                  * group leader, and don't worry if the task we have
3143                  * found doesn't happen to be a thread group leader.
3144                  * As we don't care in the case of readdir.
3145                  */
3146                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3147                         iter.tgid += 1;
3148                         goto retry;
3149                 }
3150                 get_task_struct(iter.task);
3151         }
3152         rcu_read_unlock();
3153         return iter;
3154 }
3155
3156 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3157
3158 /* for the /proc/ directory itself, after non-process stuff has been done */
3159 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3160 {
3161         struct tgid_iter iter;
3162         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3163         loff_t pos = ctx->pos;
3164
3165         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3166                 return 0;
3167
3168         if (pos == TGID_OFFSET - 2) {
3169                 struct inode *inode = d_inode(ns->proc_self);
3170                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3171                         return 0;
3172                 ctx->pos = pos = pos + 1;
3173         }
3174         if (pos == TGID_OFFSET - 1) {
3175                 struct inode *inode = d_inode(ns->proc_thread_self);
3176                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3177                         return 0;
3178                 ctx->pos = pos = pos + 1;
3179         }
3180         iter.tgid = pos - TGID_OFFSET;
3181         iter.task = NULL;
3182         for (iter = next_tgid(ns, iter);
3183              iter.task;
3184              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3185                 char name[PROC_NUMBUF];
3186                 int len;
3187
3188                 cond_resched();
3189                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3190                         continue;
3191
3192                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3193                 ctx->pos = iter.tgid + TGID_OFFSET;
3194                 if (!proc_fill_cache(file, ctx, name, len,
3195                                      proc_pid_instantiate, iter.task, NULL)) {
3196                         put_task_struct(iter.task);
3197                         return 0;
3198                 }
3199         }
3200         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3201         return 0;
3202 }
3203
3204 /*
3205  * proc_tid_comm_permission is a special permission function exclusively
3206  * used for the node /proc/<pid>/task/<tid>/comm.
3207  * It bypasses generic permission checks in the case where a task of the same
3208  * task group attempts to access the node.
3209  * The rationale behind this is that glibc and bionic access this node for
3210  * cross thread naming (pthread_set/getname_np(!self)). However, if
3211  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3212  * which locks out the cross thread naming implementation.
3213  * This function makes sure that the node is always accessible for members of
3214  * same thread group.
3215  */
3216 static int proc_tid_comm_permission(struct inode *inode, int mask)
3217 {
3218         bool is_same_tgroup;
3219         struct task_struct *task;
3220
3221         task = get_proc_task(inode);
3222         if (!task)
3223                 return -ESRCH;
3224         is_same_tgroup = same_thread_group(current, task);
3225         put_task_struct(task);
3226
3227         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3228                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3229                  * read or written by the members of the corresponding
3230                  * thread group.
3231                  */
3232                 return 0;
3233         }
3234
3235         return generic_permission(inode, mask);
3236 }
3237
3238 static const struct inode_operations proc_tid_comm_inode_operations = {
3239                 .permission = proc_tid_comm_permission,
3240 };
3241
3242 /*
3243  * Tasks
3244  */
3245 static const struct pid_entry tid_base_stuff[] = {
3246         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3247         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3248         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3249 #ifdef CONFIG_NET
3250         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3251 #endif
3252         REG("environ",   S_IRUSR, proc_environ_operations),
3253         REG("auxv",      S_IRUSR, proc_auxv_operations),
3254         ONE("status",    S_IRUGO, proc_pid_status),
3255         ONE("personality", S_IRUSR, proc_pid_personality),
3256         ONE("limits",    S_IRUGO, proc_pid_limits),
3257 #ifdef CONFIG_SCHED_DEBUG
3258         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3259 #endif
3260         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3261                          &proc_tid_comm_inode_operations,
3262                          &proc_pid_set_comm_operations, {}),
3263 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3264         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3265 #endif
3266         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3267         ONE("stat",      S_IRUGO, proc_tid_stat),
3268         ONE("statm",     S_IRUGO, proc_pid_statm),
3269         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3270 #ifdef CONFIG_PROC_CHILDREN
3271         REG("children",  S_IRUGO, proc_tid_children_operations),
3272 #endif
3273 #ifdef CONFIG_NUMA
3274         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3275 #endif
3276         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3277         LNK("cwd",       proc_cwd_link),
3278         LNK("root",      proc_root_link),
3279         LNK("exe",       proc_exe_link),
3280         REG("mounts",    S_IRUGO, proc_mounts_operations),
3281         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3282 #ifdef CONFIG_PROC_PAGE_MONITOR
3283         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3284         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3285         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3286 #endif
3287 #ifdef CONFIG_SECURITY
3288         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3289 #endif
3290 #ifdef CONFIG_KALLSYMS
3291         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3292 #endif
3293 #ifdef CONFIG_STACKTRACE
3294         ONE("stack",      S_IRUSR, proc_pid_stack),
3295 #endif
3296 #ifdef CONFIG_SCHED_INFO
3297         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3298 #endif
3299 #ifdef CONFIG_LATENCYTOP
3300         REG("latency",  S_IRUGO, proc_lstats_operations),
3301 #endif
3302 #ifdef CONFIG_PROC_PID_CPUSET
3303         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3304 #endif
3305 #ifdef CONFIG_CGROUPS
3306         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3307 #endif
3308         ONE("oom_score", S_IRUGO, proc_oom_score),
3309         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3310         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3311 #ifdef CONFIG_AUDITSYSCALL
3312         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3313         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3314 #endif
3315 #ifdef CONFIG_FAULT_INJECTION
3316         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3317 #endif
3318 #ifdef CONFIG_TASK_IO_ACCOUNTING
3319         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3320 #endif
3321 #ifdef CONFIG_HARDWALL
3322         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3323 #endif
3324 #ifdef CONFIG_USER_NS
3325         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3326         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3327         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3328         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3329 #endif
3330 #ifdef CONFIG_LIVEPATCH
3331         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3332 #endif
3333 };
3334
3335 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3336 {
3337         return proc_pident_readdir(file, ctx,
3338                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3339 }
3340
3341 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3342 {
3343         return proc_pident_lookup(dir, dentry,
3344                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3345 }
3346
3347 static const struct file_operations proc_tid_base_operations = {
3348         .read           = generic_read_dir,
3349         .iterate_shared = proc_tid_base_readdir,
3350         .llseek         = generic_file_llseek,
3351 };
3352
3353 static const struct inode_operations proc_tid_base_inode_operations = {
3354         .lookup         = proc_tid_base_lookup,
3355         .getattr        = pid_getattr,
3356         .setattr        = proc_setattr,
3357 };
3358
3359 static int proc_task_instantiate(struct inode *dir,
3360         struct dentry *dentry, struct task_struct *task, const void *ptr)
3361 {
3362         struct inode *inode;
3363         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3364
3365         if (!inode)
3366                 goto out;
3367         inode->i_op = &proc_tid_base_inode_operations;
3368         inode->i_fop = &proc_tid_base_operations;
3369         inode->i_flags|=S_IMMUTABLE;
3370
3371         set_nlink(inode, nlink_tid);
3372
3373         d_set_d_op(dentry, &pid_dentry_operations);
3374
3375         d_add(dentry, inode);
3376         /* Close the race of the process dying before we return the dentry */
3377         if (pid_revalidate(dentry, 0))
3378                 return 0;
3379 out:
3380         return -ENOENT;
3381 }
3382
3383 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3384 {
3385         int result = -ENOENT;
3386         struct task_struct *task;
3387         struct task_struct *leader = get_proc_task(dir);
3388         unsigned tid;
3389         struct pid_namespace *ns;
3390
3391         if (!leader)
3392                 goto out_no_task;
3393
3394         tid = name_to_int(&dentry->d_name);
3395         if (tid == ~0U)
3396                 goto out;
3397
3398         ns = dentry->d_sb->s_fs_info;
3399         rcu_read_lock();
3400         task = find_task_by_pid_ns(tid, ns);
3401         if (task)
3402                 get_task_struct(task);
3403         rcu_read_unlock();
3404         if (!task)
3405                 goto out;
3406         if (!same_thread_group(leader, task))
3407                 goto out_drop_task;
3408
3409         result = proc_task_instantiate(dir, dentry, task, NULL);
3410 out_drop_task:
3411         put_task_struct(task);
3412 out:
3413         put_task_struct(leader);
3414 out_no_task:
3415         return ERR_PTR(result);
3416 }
3417
3418 /*
3419  * Find the first tid of a thread group to return to user space.
3420  *
3421  * Usually this is just the thread group leader, but if the users
3422  * buffer was too small or there was a seek into the middle of the
3423  * directory we have more work todo.
3424  *
3425  * In the case of a short read we start with find_task_by_pid.
3426  *
3427  * In the case of a seek we start with the leader and walk nr
3428  * threads past it.
3429  */
3430 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3431                                         struct pid_namespace *ns)
3432 {
3433         struct task_struct *pos, *task;
3434         unsigned long nr = f_pos;
3435
3436         if (nr != f_pos)        /* 32bit overflow? */
3437                 return NULL;
3438
3439         rcu_read_lock();
3440         task = pid_task(pid, PIDTYPE_PID);
3441         if (!task)
3442                 goto fail;
3443
3444         /* Attempt to start with the tid of a thread */
3445         if (tid && nr) {
3446                 pos = find_task_by_pid_ns(tid, ns);
3447                 if (pos && same_thread_group(pos, task))
3448                         goto found;
3449         }
3450
3451         /* If nr exceeds the number of threads there is nothing todo */
3452         if (nr >= get_nr_threads(task))
3453                 goto fail;
3454
3455         /* If we haven't found our starting place yet start
3456          * with the leader and walk nr threads forward.
3457          */
3458         pos = task = task->group_leader;
3459         do {
3460                 if (!nr--)
3461                         goto found;
3462         } while_each_thread(task, pos);
3463 fail:
3464         pos = NULL;
3465         goto out;
3466 found:
3467         get_task_struct(pos);
3468 out:
3469         rcu_read_unlock();
3470         return pos;
3471 }
3472
3473 /*
3474  * Find the next thread in the thread list.
3475  * Return NULL if there is an error or no next thread.
3476  *
3477  * The reference to the input task_struct is released.
3478  */
3479 static struct task_struct *next_tid(struct task_struct *start)
3480 {
3481         struct task_struct *pos = NULL;
3482         rcu_read_lock();
3483         if (pid_alive(start)) {
3484                 pos = next_thread(start);
3485                 if (thread_group_leader(pos))
3486                         pos = NULL;
3487                 else
3488                         get_task_struct(pos);
3489         }
3490         rcu_read_unlock();
3491         put_task_struct(start);
3492         return pos;
3493 }
3494
3495 /* for the /proc/TGID/task/ directories */
3496 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3497 {
3498         struct inode *inode = file_inode(file);
3499         struct task_struct *task;
3500         struct pid_namespace *ns;
3501         int tid;
3502
3503         if (proc_inode_is_dead(inode))
3504                 return -ENOENT;
3505
3506         if (!dir_emit_dots(file, ctx))
3507                 return 0;
3508
3509         /* f_version caches the tgid value that the last readdir call couldn't
3510          * return. lseek aka telldir automagically resets f_version to 0.
3511          */
3512         ns = inode->i_sb->s_fs_info;
3513         tid = (int)file->f_version;
3514         file->f_version = 0;
3515         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3516              task;
3517              task = next_tid(task), ctx->pos++) {
3518                 char name[PROC_NUMBUF];
3519                 int len;
3520                 tid = task_pid_nr_ns(task, ns);
3521                 len = snprintf(name, sizeof(name), "%d", tid);
3522                 if (!proc_fill_cache(file, ctx, name, len,
3523                                 proc_task_instantiate, task, NULL)) {
3524                         /* returning this tgid failed, save it as the first
3525                          * pid for the next readir call */
3526                         file->f_version = (u64)tid;
3527                         put_task_struct(task);
3528                         break;
3529                 }
3530         }
3531
3532         return 0;
3533 }
3534
3535 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3536                              u32 request_mask, unsigned int query_flags)
3537 {
3538         struct inode *inode = d_inode(path->dentry);
3539         struct task_struct *p = get_proc_task(inode);
3540         generic_fillattr(inode, stat);
3541
3542         if (p) {
3543                 stat->nlink += get_nr_threads(p);
3544                 put_task_struct(p);
3545         }
3546
3547         return 0;
3548 }
3549
3550 static const struct inode_operations proc_task_inode_operations = {
3551         .lookup         = proc_task_lookup,
3552         .getattr        = proc_task_getattr,
3553         .setattr        = proc_setattr,
3554         .permission     = proc_pid_permission,
3555 };
3556
3557 static const struct file_operations proc_task_operations = {
3558         .read           = generic_read_dir,
3559         .iterate_shared = proc_task_readdir,
3560         .llseek         = generic_file_llseek,
3561 };
3562
3563 void __init set_proc_pid_nlink(void)
3564 {
3565         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3566         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3567 }