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