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
 836         if (!mm)
 837                 return 0;
 838
 839         page = (char *)__get_free_page(GFP_TEMPORARY);
 840         if (!page)
 841                 return -ENOMEM;
 842
 843         copied = 0;
 844         if (!atomic_inc_not_zero(&mm->mm_users))
 845                 goto free;
 846
 847         while (count > 0) {
 848                 int this_len = min_t(int, count, PAGE_SIZE);
 849
 850                 if (write && copy_from_user(page, buf, this_len)) {
 851                         copied = -EFAULT;
 852                         break;
 853                 }
 854
 855                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 856                 if (!this_len) {
 857                         if (!copied)
 858                                 copied = -EIO;
 859                         break;
 860                 }
 861
 862                 if (!write && copy_to_user(buf, page, this_len)) {
 863                         copied = -EFAULT;
 864                         break;
 865                 }
 866
 867                 buf += this_len;
 868                 addr += this_len;
 869                 copied += this_len;
 870                 count -= this_len;
 871         }
 872         *ppos = addr;
 873
 874         mmput(mm);
 875 free:
 876         free_page((unsigned long) page);
 877         return copied;
 878 }
 879
 880 static ssize_t mem_read(struct file *file, char __user *buf,
 881                         size_t count, loff_t *ppos)
 882 {
 883         return mem_rw(file, buf, count, ppos, 0);
 884 }
 885
 886 static ssize_t mem_write(struct file *file, const char __user *buf,
 887                          size_t count, loff_t *ppos)
 888 {
 889         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 890 }
 891
 892 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 893 {
 894         switch (orig) {
 895         case 0:
 896                 file->f_pos = offset;
 897                 break;
 898         case 1:
 899                 file->f_pos += offset;
 900                 break;
 901         default:
 902                 return -EINVAL;
 903         }
 904         force_successful_syscall_return();
 905         return file->f_pos;
 906 }
 907
 908 static int mem_release(struct inode *inode, struct file *file)
 909 {
 910         struct mm_struct *mm = file->private_data;
 911         if (mm)
 912                 mmdrop(mm);
 913         return 0;
 914 }
 915
 916 static const struct file_operations proc_mem_operations = {
 917         .llseek         = mem_lseek,
 918         .read           = mem_read,
 919         .write          = mem_write,
 920         .open           = mem_open,
 921         .release        = mem_release,
 922 };
 923
 924 static int environ_open(struct inode *inode, struct file *file)
 925 {
 926         return __mem_open(inode, file, PTRACE_MODE_READ);
 927 }
 928
 929 static ssize_t environ_read(struct file *file, char __user *buf,
 930                         size_t count, loff_t *ppos)
 931 {
 932         char *page;
 933         unsigned long src = *ppos;
 934         int ret = 0;
 935         struct mm_struct *mm = file->private_data;
 936         unsigned long env_start, env_end;
 937
 938         /* Ensure the process spawned far enough to have an environment. */
 939         if (!mm || !mm->env_end)
 940                 return 0;
 941
 942         page = (char *)__get_free_page(GFP_TEMPORARY);
 943         if (!page)
 944                 return -ENOMEM;
 945
 946         ret = 0;
 947         if (!atomic_inc_not_zero(&mm->mm_users))
 948                 goto free;
 949
 950         down_read(&mm->mmap_sem);
 951         env_start = mm->env_start;
 952         env_end = mm->env_end;
 953         up_read(&mm->mmap_sem);
 954
 955         while (count > 0) {
 956                 size_t this_len, max_len;
 957                 int retval;
 958
 959                 if (src >= (env_end - env_start))
 960                         break;
 961
 962                 this_len = env_end - (env_start + src);
 963
 964                 max_len = min_t(size_t, PAGE_SIZE, count);
 965                 this_len = min(max_len, this_len);
 966
 967                 retval = access_remote_vm(mm, (env_start + src),
 968                         page, this_len, 0);
 969
 970                 if (retval <= 0) {
 971                         ret = retval;
 972                         break;
 973                 }
 974
 975                 if (copy_to_user(buf, page, retval)) {
 976                         ret = -EFAULT;
 977                         break;
 978                 }
 979
 980                 ret += retval;
 981                 src += retval;
 982                 buf += retval;
 983                 count -= retval;
 984         }
 985         *ppos = src;
 986         mmput(mm);
 987
 988 free:
 989         free_page((unsigned long) page);
 990         return ret;
 991 }
 992
 993 static const struct file_operations proc_environ_operations = {
 994         .open           = environ_open,
 995         .read           = environ_read,
 996         .llseek         = generic_file_llseek,
 997         .release        = mem_release,
 998 };
 999
1000 static int auxv_open(struct inode *inode, struct file *file)
1001 {
1002         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
1003 }
1004
1005 static ssize_t auxv_read(struct file *file, char __user *buf,
1006                         size_t count, loff_t *ppos)
1007 {
1008         struct mm_struct *mm = file->private_data;
1009         unsigned int nwords = 0;
1010         do {
1011                 nwords += 2;
1012         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
1013         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
1014                                        nwords * sizeof(mm->saved_auxv[0]));
1015 }
1016
1017 static const struct file_operations proc_auxv_operations = {
1018         .open           = auxv_open,
1019         .read           = auxv_read,
1020         .llseek         = generic_file_llseek,
1021         .release        = mem_release,
1022 };
1023
1024 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1025                             loff_t *ppos)
1026 {
1027         struct task_struct *task = get_proc_task(file_inode(file));
1028         char buffer[PROC_NUMBUF];
1029         int oom_adj = OOM_ADJUST_MIN;
1030         size_t len;
1031
1032         if (!task)
1033                 return -ESRCH;
1034         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1035                 oom_adj = OOM_ADJUST_MAX;
1036         else
1037                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1038                           OOM_SCORE_ADJ_MAX;
1039         put_task_struct(task);
1040         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1041         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1042 }
1043
1044 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1045 {
1046         static DEFINE_MUTEX(oom_adj_mutex);
1047         struct mm_struct *mm = NULL;
1048         struct task_struct *task;
1049         int err = 0;
1050
1051         task = get_proc_task(file_inode(file));
1052         if (!task)
1053                 return -ESRCH;
1054
1055         mutex_lock(&oom_adj_mutex);
1056         if (legacy) {
1057                 if (oom_adj < task->signal->oom_score_adj &&
1058                                 !capable(CAP_SYS_RESOURCE)) {
1059                         err = -EACCES;
1060                         goto err_unlock;
1061                 }
1062                 /*
1063                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1064                  * /proc/pid/oom_score_adj instead.
1065                  */
1066                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1067                           current->comm, task_pid_nr(current), task_pid_nr(task),
1068                           task_pid_nr(task));
1069         } else {
1070                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1071                                 !capable(CAP_SYS_RESOURCE)) {
1072                         err = -EACCES;
1073                         goto err_unlock;
1074                 }
1075         }
1076
1077         /*
1078          * Make sure we will check other processes sharing the mm if this is
1079          * not vfrok which wants its own oom_score_adj.
1080          * pin the mm so it doesn't go away and get reused after task_unlock
1081          */
1082         if (!task->vfork_done) {
1083                 struct task_struct *p = find_lock_task_mm(task);
1084
1085                 if (p) {
1086                         if (atomic_read(&p->mm->mm_users) > 1) {
1087                                 mm = p->mm;
1088                                 atomic_inc(&mm->mm_count);
1089                         }
1090                         task_unlock(p);
1091                 }
1092         }
1093
1094         task->signal->oom_score_adj = oom_adj;
1095         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1096                 task->signal->oom_score_adj_min = (short)oom_adj;
1097         trace_oom_score_adj_update(task);
1098
1099         if (mm) {
1100                 struct task_struct *p;
1101
1102                 rcu_read_lock();
1103                 for_each_process(p) {
1104                         if (same_thread_group(task, p))
1105                                 continue;
1106
1107                         /* do not touch kernel threads or the global init */
1108                         if (p->flags & PF_KTHREAD || is_global_init(p))
1109                                 continue;
1110
1111                         task_lock(p);
1112                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1113                                 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",
1114                                                 task_pid_nr(p), p->comm,
1115                                                 p->signal->oom_score_adj, oom_adj,
1116                                                 task_pid_nr(task), task->comm);
1117                                 p->signal->oom_score_adj = oom_adj;
1118                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1119                                         p->signal->oom_score_adj_min = (short)oom_adj;
1120                         }
1121                         task_unlock(p);
1122                 }
1123                 rcu_read_unlock();
1124                 mmdrop(mm);
1125         }
1126 err_unlock:
1127         mutex_unlock(&oom_adj_mutex);
1128         put_task_struct(task);
1129         return err;
1130 }
1131
1132 /*
1133  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1134  * kernels.  The effective policy is defined by oom_score_adj, which has a
1135  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1136  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1137  * Processes that become oom disabled via oom_adj will still be oom disabled
1138  * with this implementation.
1139  *
1140  * oom_adj cannot be removed since existing userspace binaries use it.
1141  */
1142 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1143                              size_t count, loff_t *ppos)
1144 {
1145         char buffer[PROC_NUMBUF];
1146         int oom_adj;
1147         int err;
1148
1149         memset(buffer, 0, sizeof(buffer));
1150         if (count > sizeof(buffer) - 1)
1151                 count = sizeof(buffer) - 1;
1152         if (copy_from_user(buffer, buf, count)) {
1153                 err = -EFAULT;
1154                 goto out;
1155         }
1156
1157         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1158         if (err)
1159                 goto out;
1160         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1161              oom_adj != OOM_DISABLE) {
1162                 err = -EINVAL;
1163                 goto out;
1164         }
1165
1166         /*
1167          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1168          * value is always attainable.
1169          */
1170         if (oom_adj == OOM_ADJUST_MAX)
1171                 oom_adj = OOM_SCORE_ADJ_MAX;
1172         else
1173                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1174
1175         err = __set_oom_adj(file, oom_adj, true);
1176 out:
1177         return err < 0 ? err : count;
1178 }
1179
1180 static const struct file_operations proc_oom_adj_operations = {
1181         .read           = oom_adj_read,
1182         .write          = oom_adj_write,
1183         .llseek         = generic_file_llseek,
1184 };
1185
1186 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1187                                         size_t count, loff_t *ppos)
1188 {
1189         struct task_struct *task = get_proc_task(file_inode(file));
1190         char buffer[PROC_NUMBUF];
1191         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1192         size_t len;
1193
1194         if (!task)
1195                 return -ESRCH;
1196         oom_score_adj = task->signal->oom_score_adj;
1197         put_task_struct(task);
1198         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1199         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1200 }
1201
1202 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1203                                         size_t count, loff_t *ppos)
1204 {
1205         char buffer[PROC_NUMBUF];
1206         int oom_score_adj;
1207         int err;
1208
1209         memset(buffer, 0, sizeof(buffer));
1210         if (count > sizeof(buffer) - 1)
1211                 count = sizeof(buffer) - 1;
1212         if (copy_from_user(buffer, buf, count)) {
1213                 err = -EFAULT;
1214                 goto out;
1215         }
1216
1217         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1218         if (err)
1219                 goto out;
1220         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1221                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1222                 err = -EINVAL;
1223                 goto out;
1224         }
1225
1226         err = __set_oom_adj(file, oom_score_adj, false);
1227 out:
1228         return err < 0 ? err : count;
1229 }
1230
1231 static const struct file_operations proc_oom_score_adj_operations = {
1232         .read           = oom_score_adj_read,
1233         .write          = oom_score_adj_write,
1234         .llseek         = default_llseek,
1235 };
1236
1237 #ifdef CONFIG_AUDITSYSCALL
1238 #define TMPBUFLEN 21
1239 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1240                                   size_t count, loff_t *ppos)
1241 {
1242         struct inode * inode = file_inode(file);
1243         struct task_struct *task = get_proc_task(inode);
1244         ssize_t length;
1245         char tmpbuf[TMPBUFLEN];
1246
1247         if (!task)
1248                 return -ESRCH;
1249         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1250                            from_kuid(file->f_cred->user_ns,
1251                                      audit_get_loginuid(task)));
1252         put_task_struct(task);
1253         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1254 }
1255
1256 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1257                                    size_t count, loff_t *ppos)
1258 {
1259         struct inode * inode = file_inode(file);
1260         uid_t loginuid;
1261         kuid_t kloginuid;
1262         int rv;
1263
1264         rcu_read_lock();
1265         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1266                 rcu_read_unlock();
1267                 return -EPERM;
1268         }
1269         rcu_read_unlock();
1270
1271         if (*ppos != 0) {
1272                 /* No partial writes. */
1273                 return -EINVAL;
1274         }
1275
1276         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1277         if (rv < 0)
1278                 return rv;
1279
1280         /* is userspace tring to explicitly UNSET the loginuid? */
1281         if (loginuid == AUDIT_UID_UNSET) {
1282                 kloginuid = INVALID_UID;
1283         } else {
1284                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1285                 if (!uid_valid(kloginuid))
1286                         return -EINVAL;
1287         }
1288
1289         rv = audit_set_loginuid(kloginuid);
1290         if (rv < 0)
1291                 return rv;
1292         return count;
1293 }
1294
1295 static const struct file_operations proc_loginuid_operations = {
1296         .read           = proc_loginuid_read,
1297         .write          = proc_loginuid_write,
1298         .llseek         = generic_file_llseek,
1299 };
1300
1301 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1302                                   size_t count, loff_t *ppos)
1303 {
1304         struct inode * inode = file_inode(file);
1305         struct task_struct *task = get_proc_task(inode);
1306         ssize_t length;
1307         char tmpbuf[TMPBUFLEN];
1308
1309         if (!task)
1310                 return -ESRCH;
1311         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1312                                 audit_get_sessionid(task));
1313         put_task_struct(task);
1314         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1315 }
1316
1317 static const struct file_operations proc_sessionid_operations = {
1318         .read           = proc_sessionid_read,
1319         .llseek         = generic_file_llseek,
1320 };
1321 #endif
1322
1323 #ifdef CONFIG_FAULT_INJECTION
1324 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1325                                       size_t count, loff_t *ppos)
1326 {
1327         struct task_struct *task = get_proc_task(file_inode(file));
1328         char buffer[PROC_NUMBUF];
1329         size_t len;
1330         int make_it_fail;
1331
1332         if (!task)
1333                 return -ESRCH;
1334         make_it_fail = task->make_it_fail;
1335         put_task_struct(task);
1336
1337         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1338
1339         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1340 }
1341
1342 static ssize_t proc_fault_inject_write(struct file * file,
1343                         const char __user * buf, size_t count, loff_t *ppos)
1344 {
1345         struct task_struct *task;
1346         char buffer[PROC_NUMBUF];
1347         int make_it_fail;
1348         int rv;
1349
1350         if (!capable(CAP_SYS_RESOURCE))
1351                 return -EPERM;
1352         memset(buffer, 0, sizeof(buffer));
1353         if (count > sizeof(buffer) - 1)
1354                 count = sizeof(buffer) - 1;
1355         if (copy_from_user(buffer, buf, count))
1356                 return -EFAULT;
1357         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1358         if (rv < 0)
1359                 return rv;
1360         if (make_it_fail < 0 || make_it_fail > 1)
1361                 return -EINVAL;
1362
1363         task = get_proc_task(file_inode(file));
1364         if (!task)
1365                 return -ESRCH;
1366         task->make_it_fail = make_it_fail;
1367         put_task_struct(task);
1368
1369         return count;
1370 }
1371
1372 static const struct file_operations proc_fault_inject_operations = {
1373         .read           = proc_fault_inject_read,
1374         .write          = proc_fault_inject_write,
1375         .llseek         = generic_file_llseek,
1376 };
1377 #endif
1378
1379
1380 #ifdef CONFIG_SCHED_DEBUG
1381 /*
1382  * Print out various scheduling related per-task fields:
1383  */
1384 static int sched_show(struct seq_file *m, void *v)
1385 {
1386         struct inode *inode = m->private;
1387         struct task_struct *p;
1388
1389         p = get_proc_task(inode);
1390         if (!p)
1391                 return -ESRCH;
1392         proc_sched_show_task(p, m);
1393
1394         put_task_struct(p);
1395
1396         return 0;
1397 }
1398
1399 static ssize_t
1400 sched_write(struct file *file, const char __user *buf,
1401             size_t count, loff_t *offset)
1402 {
1403         struct inode *inode = file_inode(file);
1404         struct task_struct *p;
1405
1406         p = get_proc_task(inode);
1407         if (!p)
1408                 return -ESRCH;
1409         proc_sched_set_task(p);
1410
1411         put_task_struct(p);
1412
1413         return count;
1414 }
1415
1416 static int sched_open(struct inode *inode, struct file *filp)
1417 {
1418         return single_open(filp, sched_show, inode);
1419 }
1420
1421 static const struct file_operations proc_pid_sched_operations = {
1422         .open           = sched_open,
1423         .read           = seq_read,
1424         .write          = sched_write,
1425         .llseek         = seq_lseek,
1426         .release        = single_release,
1427 };
1428
1429 #endif
1430
1431 #ifdef CONFIG_SCHED_AUTOGROUP
1432 /*
1433  * Print out autogroup related information:
1434  */
1435 static int sched_autogroup_show(struct seq_file *m, void *v)
1436 {
1437         struct inode *inode = m->private;
1438         struct task_struct *p;
1439
1440         p = get_proc_task(inode);
1441         if (!p)
1442                 return -ESRCH;
1443         proc_sched_autogroup_show_task(p, m);
1444
1445         put_task_struct(p);
1446
1447         return 0;
1448 }
1449
1450 static ssize_t
1451 sched_autogroup_write(struct file *file, const char __user *buf,
1452             size_t count, loff_t *offset)
1453 {
1454         struct inode *inode = file_inode(file);
1455         struct task_struct *p;
1456         char buffer[PROC_NUMBUF];
1457         int nice;
1458         int err;
1459
1460         memset(buffer, 0, sizeof(buffer));
1461         if (count > sizeof(buffer) - 1)
1462                 count = sizeof(buffer) - 1;
1463         if (copy_from_user(buffer, buf, count))
1464                 return -EFAULT;
1465
1466         err = kstrtoint(strstrip(buffer), 0, &nice);
1467         if (err < 0)
1468                 return err;
1469
1470         p = get_proc_task(inode);
1471         if (!p)
1472                 return -ESRCH;
1473
1474         err = proc_sched_autogroup_set_nice(p, nice);
1475         if (err)
1476                 count = err;
1477
1478         put_task_struct(p);
1479
1480         return count;
1481 }
1482
1483 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1484 {
1485         int ret;
1486
1487         ret = single_open(filp, sched_autogroup_show, NULL);
1488         if (!ret) {
1489                 struct seq_file *m = filp->private_data;
1490
1491                 m->private = inode;
1492         }
1493         return ret;
1494 }
1495
1496 static const struct file_operations proc_pid_sched_autogroup_operations = {
1497         .open           = sched_autogroup_open,
1498         .read           = seq_read,
1499         .write          = sched_autogroup_write,
1500         .llseek         = seq_lseek,
1501         .release        = single_release,
1502 };
1503
1504 #endif /* CONFIG_SCHED_AUTOGROUP */
1505
1506 static ssize_t comm_write(struct file *file, const char __user *buf,
1507                                 size_t count, loff_t *offset)
1508 {
1509         struct inode *inode = file_inode(file);
1510         struct task_struct *p;
1511         char buffer[TASK_COMM_LEN];
1512         const size_t maxlen = sizeof(buffer) - 1;
1513
1514         memset(buffer, 0, sizeof(buffer));
1515         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1516                 return -EFAULT;
1517
1518         p = get_proc_task(inode);
1519         if (!p)
1520                 return -ESRCH;
1521
1522         if (same_thread_group(current, p))
1523                 set_task_comm(p, buffer);
1524         else
1525                 count = -EINVAL;
1526
1527         put_task_struct(p);
1528
1529         return count;
1530 }
1531
1532 static int comm_show(struct seq_file *m, void *v)
1533 {
1534         struct inode *inode = m->private;
1535         struct task_struct *p;
1536
1537         p = get_proc_task(inode);
1538         if (!p)
1539                 return -ESRCH;
1540
1541         task_lock(p);
1542         seq_printf(m, "%s\n", p->comm);
1543         task_unlock(p);
1544
1545         put_task_struct(p);
1546
1547         return 0;
1548 }
1549
1550 static int comm_open(struct inode *inode, struct file *filp)
1551 {
1552         return single_open(filp, comm_show, inode);
1553 }
1554
1555 static const struct file_operations proc_pid_set_comm_operations = {
1556         .open           = comm_open,
1557         .read           = seq_read,
1558         .write          = comm_write,
1559         .llseek         = seq_lseek,
1560         .release        = single_release,
1561 };
1562
1563 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1564 {
1565         struct task_struct *task;
1566         struct file *exe_file;
1567
1568         task = get_proc_task(d_inode(dentry));
1569         if (!task)
1570                 return -ENOENT;
1571         exe_file = get_task_exe_file(task);
1572         put_task_struct(task);
1573         if (exe_file) {
1574                 *exe_path = exe_file->f_path;
1575                 path_get(&exe_file->f_path);
1576                 fput(exe_file);
1577                 return 0;
1578         } else
1579                 return -ENOENT;
1580 }
1581
1582 static const char *proc_pid_get_link(struct dentry *dentry,
1583                                      struct inode *inode,
1584                                      struct delayed_call *done)
1585 {
1586         struct path path;
1587         int error = -EACCES;
1588
1589         if (!dentry)
1590                 return ERR_PTR(-ECHILD);
1591
1592         /* Are we allowed to snoop on the tasks file descriptors? */
1593         if (!proc_fd_access_allowed(inode))
1594                 goto out;
1595
1596         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1597         if (error)
1598                 goto out;
1599
1600         nd_jump_link(&path);
1601         return NULL;
1602 out:
1603         return ERR_PTR(error);
1604 }
1605
1606 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1607 {
1608         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1609         char *pathname;
1610         int len;
1611
1612         if (!tmp)
1613                 return -ENOMEM;
1614
1615         pathname = d_path(path, tmp, PAGE_SIZE);
1616         len = PTR_ERR(pathname);
1617         if (IS_ERR(pathname))
1618                 goto out;
1619         len = tmp + PAGE_SIZE - 1 - pathname;
1620
1621         if (len > buflen)
1622                 len = buflen;
1623         if (copy_to_user(buffer, pathname, len))
1624                 len = -EFAULT;
1625  out:
1626         free_page((unsigned long)tmp);
1627         return len;
1628 }
1629
1630 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1631 {
1632         int error = -EACCES;
1633         struct inode *inode = d_inode(dentry);
1634         struct path path;
1635
1636         /* Are we allowed to snoop on the tasks file descriptors? */
1637         if (!proc_fd_access_allowed(inode))
1638                 goto out;
1639
1640         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1641         if (error)
1642                 goto out;
1643
1644         error = do_proc_readlink(&path, buffer, buflen);
1645         path_put(&path);
1646 out:
1647         return error;
1648 }
1649
1650 const struct inode_operations proc_pid_link_inode_operations = {
1651         .readlink       = proc_pid_readlink,
1652         .get_link       = proc_pid_get_link,
1653         .setattr        = proc_setattr,
1654 };
1655
1656
1657 /* building an inode */
1658
1659 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1660 {
1661         struct inode * inode;
1662         struct proc_inode *ei;
1663         const struct cred *cred;
1664
1665         /* We need a new inode */
1666
1667         inode = new_inode(sb);
1668         if (!inode)
1669                 goto out;
1670
1671         /* Common stuff */
1672         ei = PROC_I(inode);
1673         inode->i_ino = get_next_ino();
1674         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1675         inode->i_op = &proc_def_inode_operations;
1676
1677         /*
1678          * grab the reference to task.
1679          */
1680         ei->pid = get_task_pid(task, PIDTYPE_PID);
1681         if (!ei->pid)
1682                 goto out_unlock;
1683
1684         if (task_dumpable(task)) {
1685                 rcu_read_lock();
1686                 cred = __task_cred(task);
1687                 inode->i_uid = cred->euid;
1688                 inode->i_gid = cred->egid;
1689                 rcu_read_unlock();
1690         }
1691         security_task_to_inode(task, inode);
1692
1693 out:
1694         return inode;
1695
1696 out_unlock:
1697         iput(inode);
1698         return NULL;
1699 }
1700
1701 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1702 {
1703         struct inode *inode = d_inode(dentry);
1704         struct task_struct *task;
1705         const struct cred *cred;
1706         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1707
1708         generic_fillattr(inode, stat);
1709
1710         rcu_read_lock();
1711         stat->uid = GLOBAL_ROOT_UID;
1712         stat->gid = GLOBAL_ROOT_GID;
1713         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1714         if (task) {
1715                 if (!has_pid_permissions(pid, task, 2)) {
1716                         rcu_read_unlock();
1717                         /*
1718                          * This doesn't prevent learning whether PID exists,
1719                          * it only makes getattr() consistent with readdir().
1720                          */
1721                         return -ENOENT;
1722                 }
1723                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1724                     task_dumpable(task)) {
1725                         cred = __task_cred(task);
1726                         stat->uid = cred->euid;
1727                         stat->gid = cred->egid;
1728                 }
1729         }
1730         rcu_read_unlock();
1731         return 0;
1732 }
1733
1734 /* dentry stuff */
1735
1736 /*
1737  *      Exceptional case: normally we are not allowed to unhash a busy
1738  * directory. In this case, however, we can do it - no aliasing problems
1739  * due to the way we treat inodes.
1740  *
1741  * Rewrite the inode's ownerships here because the owning task may have
1742  * performed a setuid(), etc.
1743  *
1744  * Before the /proc/pid/status file was created the only way to read
1745  * the effective uid of a /process was to stat /proc/pid.  Reading
1746  * /proc/pid/status is slow enough that procps and other packages
1747  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1748  * made this apply to all per process world readable and executable
1749  * directories.
1750  */
1751 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1752 {
1753         struct inode *inode;
1754         struct task_struct *task;
1755         const struct cred *cred;
1756
1757         if (flags & LOOKUP_RCU)
1758                 return -ECHILD;
1759
1760         inode = d_inode(dentry);
1761         task = get_proc_task(inode);
1762
1763         if (task) {
1764                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1765                     task_dumpable(task)) {
1766                         rcu_read_lock();
1767                         cred = __task_cred(task);
1768                         inode->i_uid = cred->euid;
1769                         inode->i_gid = cred->egid;
1770                         rcu_read_unlock();
1771                 } else {
1772                         inode->i_uid = GLOBAL_ROOT_UID;
1773                         inode->i_gid = GLOBAL_ROOT_GID;
1774                 }
1775                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1776                 security_task_to_inode(task, inode);
1777                 put_task_struct(task);
1778                 return 1;
1779         }
1780         return 0;
1781 }
1782
1783 static inline bool proc_inode_is_dead(struct inode *inode)
1784 {
1785         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1786 }
1787
1788 int pid_delete_dentry(const struct dentry *dentry)
1789 {
1790         /* Is the task we represent dead?
1791          * If so, then don't put the dentry on the lru list,
1792          * kill it immediately.
1793          */
1794         return proc_inode_is_dead(d_inode(dentry));
1795 }
1796
1797 const struct dentry_operations pid_dentry_operations =
1798 {
1799         .d_revalidate   = pid_revalidate,
1800         .d_delete       = pid_delete_dentry,
1801 };
1802
1803 /* Lookups */
1804
1805 /*
1806  * Fill a directory entry.
1807  *
1808  * If possible create the dcache entry and derive our inode number and
1809  * file type from dcache entry.
1810  *
1811  * Since all of the proc inode numbers are dynamically generated, the inode
1812  * numbers do not exist until the inode is cache.  This means creating the
1813  * the dcache entry in readdir is necessary to keep the inode numbers
1814  * reported by readdir in sync with the inode numbers reported
1815  * by stat.
1816  */
1817 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1818         const char *name, int len,
1819         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1820 {
1821         struct dentry *child, *dir = file->f_path.dentry;
1822         struct qstr qname = QSTR_INIT(name, len);
1823         struct inode *inode;
1824         unsigned type;
1825         ino_t ino;
1826
1827         child = d_hash_and_lookup(dir, &qname);
1828         if (!child) {
1829                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1830                 child = d_alloc_parallel(dir, &qname, &wq);
1831                 if (IS_ERR(child))
1832                         goto end_instantiate;
1833                 if (d_in_lookup(child)) {
1834                         int err = instantiate(d_inode(dir), child, task, ptr);
1835                         d_lookup_done(child);
1836                         if (err < 0) {
1837                                 dput(child);
1838                                 goto end_instantiate;
1839                         }
1840                 }
1841         }
1842         inode = d_inode(child);
1843         ino = inode->i_ino;
1844         type = inode->i_mode >> 12;
1845         dput(child);
1846         return dir_emit(ctx, name, len, ino, type);
1847
1848 end_instantiate:
1849         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1850 }
1851
1852 /*
1853  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1854  * which represent vma start and end addresses.
1855  */
1856 static int dname_to_vma_addr(struct dentry *dentry,
1857                              unsigned long *start, unsigned long *end)
1858 {
1859         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1860                 return -EINVAL;
1861
1862         return 0;
1863 }
1864
1865 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1866 {
1867         unsigned long vm_start, vm_end;
1868         bool exact_vma_exists = false;
1869         struct mm_struct *mm = NULL;
1870         struct task_struct *task;
1871         const struct cred *cred;
1872         struct inode *inode;
1873         int status = 0;
1874
1875         if (flags & LOOKUP_RCU)
1876                 return -ECHILD;
1877
1878         inode = d_inode(dentry);
1879         task = get_proc_task(inode);
1880         if (!task)
1881                 goto out_notask;
1882
1883         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1884         if (IS_ERR_OR_NULL(mm))
1885                 goto out;
1886
1887         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1888                 down_read(&mm->mmap_sem);
1889                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1890                 up_read(&mm->mmap_sem);
1891         }
1892
1893         mmput(mm);
1894
1895         if (exact_vma_exists) {
1896                 if (task_dumpable(task)) {
1897                         rcu_read_lock();
1898                         cred = __task_cred(task);
1899                         inode->i_uid = cred->euid;
1900                         inode->i_gid = cred->egid;
1901                         rcu_read_unlock();
1902                 } else {
1903                         inode->i_uid = GLOBAL_ROOT_UID;
1904                         inode->i_gid = GLOBAL_ROOT_GID;
1905                 }
1906                 security_task_to_inode(task, inode);
1907                 status = 1;
1908         }
1909
1910 out:
1911         put_task_struct(task);
1912
1913 out_notask:
1914         return status;
1915 }
1916
1917 static const struct dentry_operations tid_map_files_dentry_operations = {
1918         .d_revalidate   = map_files_d_revalidate,
1919         .d_delete       = pid_delete_dentry,
1920 };
1921
1922 static int map_files_get_link(struct dentry *dentry, struct path *path)
1923 {
1924         unsigned long vm_start, vm_end;
1925         struct vm_area_struct *vma;
1926         struct task_struct *task;
1927         struct mm_struct *mm;
1928         int rc;
1929
1930         rc = -ENOENT;
1931         task = get_proc_task(d_inode(dentry));
1932         if (!task)
1933                 goto out;
1934
1935         mm = get_task_mm(task);
1936         put_task_struct(task);
1937         if (!mm)
1938                 goto out;
1939
1940         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1941         if (rc)
1942                 goto out_mmput;
1943
1944         rc = -ENOENT;
1945         down_read(&mm->mmap_sem);
1946         vma = find_exact_vma(mm, vm_start, vm_end);
1947         if (vma && vma->vm_file) {
1948                 *path = vma->vm_file->f_path;
1949                 path_get(path);
1950                 rc = 0;
1951         }
1952         up_read(&mm->mmap_sem);
1953
1954 out_mmput:
1955         mmput(mm);
1956 out:
1957         return rc;
1958 }
1959
1960 struct map_files_info {
1961         fmode_t         mode;
1962         unsigned long   len;
1963         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1964 };
1965
1966 /*
1967  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1968  * symlinks may be used to bypass permissions on ancestor directories in the
1969  * path to the file in question.
1970  */
1971 static const char *
1972 proc_map_files_get_link(struct dentry *dentry,
1973                         struct inode *inode,
1974                         struct delayed_call *done)
1975 {
1976         if (!capable(CAP_SYS_ADMIN))
1977                 return ERR_PTR(-EPERM);
1978
1979         return proc_pid_get_link(dentry, inode, done);
1980 }
1981
1982 /*
1983  * Identical to proc_pid_link_inode_operations except for get_link()
1984  */
1985 static const struct inode_operations proc_map_files_link_inode_operations = {
1986         .readlink       = proc_pid_readlink,
1987         .get_link       = proc_map_files_get_link,
1988         .setattr        = proc_setattr,
1989 };
1990
1991 static int
1992 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1993                            struct task_struct *task, const void *ptr)
1994 {
1995         fmode_t mode = (fmode_t)(unsigned long)ptr;
1996         struct proc_inode *ei;
1997         struct inode *inode;
1998
1999         inode = proc_pid_make_inode(dir->i_sb, task);
2000         if (!inode)
2001                 return -ENOENT;
2002
2003         ei = PROC_I(inode);
2004         ei->op.proc_get_link = map_files_get_link;
2005
2006         inode->i_op = &proc_map_files_link_inode_operations;
2007         inode->i_size = 64;
2008         inode->i_mode = S_IFLNK;
2009
2010         if (mode & FMODE_READ)
2011                 inode->i_mode |= S_IRUSR;
2012         if (mode & FMODE_WRITE)
2013                 inode->i_mode |= S_IWUSR;
2014
2015         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2016         d_add(dentry, inode);
2017
2018         return 0;
2019 }
2020
2021 static struct dentry *proc_map_files_lookup(struct inode *dir,
2022                 struct dentry *dentry, unsigned int flags)
2023 {
2024         unsigned long vm_start, vm_end;
2025         struct vm_area_struct *vma;
2026         struct task_struct *task;
2027         int result;
2028         struct mm_struct *mm;
2029
2030         result = -ENOENT;
2031         task = get_proc_task(dir);
2032         if (!task)
2033                 goto out;
2034
2035         result = -EACCES;
2036         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2037                 goto out_put_task;
2038
2039         result = -ENOENT;
2040         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2041                 goto out_put_task;
2042
2043         mm = get_task_mm(task);
2044         if (!mm)
2045                 goto out_put_task;
2046
2047         down_read(&mm->mmap_sem);
2048         vma = find_exact_vma(mm, vm_start, vm_end);
2049         if (!vma)
2050                 goto out_no_vma;
2051
2052         if (vma->vm_file)
2053                 result = proc_map_files_instantiate(dir, dentry, task,
2054                                 (void *)(unsigned long)vma->vm_file->f_mode);
2055
2056 out_no_vma:
2057         up_read(&mm->mmap_sem);
2058         mmput(mm);
2059 out_put_task:
2060         put_task_struct(task);
2061 out:
2062         return ERR_PTR(result);
2063 }
2064
2065 static const struct inode_operations proc_map_files_inode_operations = {
2066         .lookup         = proc_map_files_lookup,
2067         .permission     = proc_fd_permission,
2068         .setattr        = proc_setattr,
2069 };
2070
2071 static int
2072 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2073 {
2074         struct vm_area_struct *vma;
2075         struct task_struct *task;
2076         struct mm_struct *mm;
2077         unsigned long nr_files, pos, i;
2078         struct flex_array *fa = NULL;
2079         struct map_files_info info;
2080         struct map_files_info *p;
2081         int ret;
2082
2083         ret = -ENOENT;
2084         task = get_proc_task(file_inode(file));
2085         if (!task)
2086                 goto out;
2087
2088         ret = -EACCES;
2089         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2090                 goto out_put_task;
2091
2092         ret = 0;
2093         if (!dir_emit_dots(file, ctx))
2094                 goto out_put_task;
2095
2096         mm = get_task_mm(task);
2097         if (!mm)
2098                 goto out_put_task;
2099         down_read(&mm->mmap_sem);
2100
2101         nr_files = 0;
2102
2103         /*
2104          * We need two passes here:
2105          *
2106          *  1) Collect vmas of mapped files with mmap_sem taken
2107          *  2) Release mmap_sem and instantiate entries
2108          *
2109          * otherwise we get lockdep complained, since filldir()
2110          * routine might require mmap_sem taken in might_fault().
2111          */
2112
2113         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2114                 if (vma->vm_file && ++pos > ctx->pos)
2115                         nr_files++;
2116         }
2117
2118         if (nr_files) {
2119                 fa = flex_array_alloc(sizeof(info), nr_files,
2120                                         GFP_KERNEL);
2121                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2122                                                 GFP_KERNEL)) {
2123                         ret = -ENOMEM;
2124                         if (fa)
2125                                 flex_array_free(fa);
2126                         up_read(&mm->mmap_sem);
2127                         mmput(mm);
2128                         goto out_put_task;
2129                 }
2130                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2131                                 vma = vma->vm_next) {
2132                         if (!vma->vm_file)
2133                                 continue;
2134                         if (++pos <= ctx->pos)
2135                                 continue;
2136
2137                         info.mode = vma->vm_file->f_mode;
2138                         info.len = snprintf(info.name,
2139                                         sizeof(info.name), "%lx-%lx",
2140                                         vma->vm_start, vma->vm_end);
2141                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2142                                 BUG();
2143                 }
2144         }
2145         up_read(&mm->mmap_sem);
2146
2147         for (i = 0; i < nr_files; i++) {
2148                 p = flex_array_get(fa, i);
2149                 if (!proc_fill_cache(file, ctx,
2150                                       p->name, p->len,
2151                                       proc_map_files_instantiate,
2152                                       task,
2153                                       (void *)(unsigned long)p->mode))
2154                         break;
2155                 ctx->pos++;
2156         }
2157         if (fa)
2158                 flex_array_free(fa);
2159         mmput(mm);
2160
2161 out_put_task:
2162         put_task_struct(task);
2163 out:
2164         return ret;
2165 }
2166
2167 static const struct file_operations proc_map_files_operations = {
2168         .read           = generic_read_dir,
2169         .iterate_shared = proc_map_files_readdir,
2170         .llseek         = generic_file_llseek,
2171 };
2172
2173 #ifdef CONFIG_CHECKPOINT_RESTORE
2174 struct timers_private {
2175         struct pid *pid;
2176         struct task_struct *task;
2177         struct sighand_struct *sighand;
2178         struct pid_namespace *ns;
2179         unsigned long flags;
2180 };
2181
2182 static void *timers_start(struct seq_file *m, loff_t *pos)
2183 {
2184         struct timers_private *tp = m->private;
2185
2186         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2187         if (!tp->task)
2188                 return ERR_PTR(-ESRCH);
2189
2190         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2191         if (!tp->sighand)
2192                 return ERR_PTR(-ESRCH);
2193
2194         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2195 }
2196
2197 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2198 {
2199         struct timers_private *tp = m->private;
2200         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2201 }
2202
2203 static void timers_stop(struct seq_file *m, void *v)
2204 {
2205         struct timers_private *tp = m->private;
2206
2207         if (tp->sighand) {
2208                 unlock_task_sighand(tp->task, &tp->flags);
2209                 tp->sighand = NULL;
2210         }
2211
2212         if (tp->task) {
2213                 put_task_struct(tp->task);
2214                 tp->task = NULL;
2215         }
2216 }
2217
2218 static int show_timer(struct seq_file *m, void *v)
2219 {
2220         struct k_itimer *timer;
2221         struct timers_private *tp = m->private;
2222         int notify;
2223         static const char * const nstr[] = {
2224                 [SIGEV_SIGNAL] = "signal",
2225                 [SIGEV_NONE] = "none",
2226                 [SIGEV_THREAD] = "thread",
2227         };
2228
2229         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2230         notify = timer->it_sigev_notify;
2231
2232         seq_printf(m, "ID: %d\n", timer->it_id);
2233         seq_printf(m, "signal: %d/%p\n",
2234                    timer->sigq->info.si_signo,
2235                    timer->sigq->info.si_value.sival_ptr);
2236         seq_printf(m, "notify: %s/%s.%d\n",
2237                    nstr[notify & ~SIGEV_THREAD_ID],
2238                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2239                    pid_nr_ns(timer->it_pid, tp->ns));
2240         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2241
2242         return 0;
2243 }
2244
2245 static const struct seq_operations proc_timers_seq_ops = {
2246         .start  = timers_start,
2247         .next   = timers_next,
2248         .stop   = timers_stop,
2249         .show   = show_timer,
2250 };
2251
2252 static int proc_timers_open(struct inode *inode, struct file *file)
2253 {
2254         struct timers_private *tp;
2255
2256         tp = __seq_open_private(file, &proc_timers_seq_ops,
2257                         sizeof(struct timers_private));
2258         if (!tp)
2259                 return -ENOMEM;
2260
2261         tp->pid = proc_pid(inode);
2262         tp->ns = inode->i_sb->s_fs_info;
2263         return 0;
2264 }
2265
2266 static const struct file_operations proc_timers_operations = {
2267         .open           = proc_timers_open,
2268         .read           = seq_read,
2269         .llseek         = seq_lseek,
2270         .release        = seq_release_private,
2271 };
2272 #endif
2273
2274 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2275                                         size_t count, loff_t *offset)
2276 {
2277         struct inode *inode = file_inode(file);
2278         struct task_struct *p;
2279         u64 slack_ns;
2280         int err;
2281
2282         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2283         if (err < 0)
2284                 return err;
2285
2286         p = get_proc_task(inode);
2287         if (!p)
2288                 return -ESRCH;
2289
2290         if (p != current) {
2291                 if (!capable(CAP_SYS_NICE)) {
2292                         count = -EPERM;
2293                         goto out;
2294                 }
2295
2296                 err = security_task_setscheduler(p);
2297                 if (err) {
2298                         count = err;
2299                         goto out;
2300                 }
2301         }
2302
2303         task_lock(p);
2304         if (slack_ns == 0)
2305                 p->timer_slack_ns = p->default_timer_slack_ns;
2306         else
2307                 p->timer_slack_ns = slack_ns;
2308         task_unlock(p);
2309
2310 out:
2311         put_task_struct(p);
2312
2313         return count;
2314 }
2315
2316 static int timerslack_ns_show(struct seq_file *m, void *v)
2317 {
2318         struct inode *inode = m->private;
2319         struct task_struct *p;
2320         int err = 0;
2321
2322         p = get_proc_task(inode);
2323         if (!p)
2324                 return -ESRCH;
2325
2326         if (p != current) {
2327
2328                 if (!capable(CAP_SYS_NICE)) {
2329                         err = -EPERM;
2330                         goto out;
2331                 }
2332                 err = security_task_getscheduler(p);
2333                 if (err)
2334                         goto out;
2335         }
2336
2337         task_lock(p);
2338         seq_printf(m, "%llu\n", p->timer_slack_ns);
2339         task_unlock(p);
2340
2341 out:
2342         put_task_struct(p);
2343
2344         return err;
2345 }
2346
2347 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2348 {
2349         return single_open(filp, timerslack_ns_show, inode);
2350 }
2351
2352 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2353         .open           = timerslack_ns_open,
2354         .read           = seq_read,
2355         .write          = timerslack_ns_write,
2356         .llseek         = seq_lseek,
2357         .release        = single_release,
2358 };
2359
2360 static int proc_pident_instantiate(struct inode *dir,
2361         struct dentry *dentry, struct task_struct *task, const void *ptr)
2362 {
2363         const struct pid_entry *p = ptr;
2364         struct inode *inode;
2365         struct proc_inode *ei;
2366
2367         inode = proc_pid_make_inode(dir->i_sb, task);
2368         if (!inode)
2369                 goto out;
2370
2371         ei = PROC_I(inode);
2372         inode->i_mode = p->mode;
2373         if (S_ISDIR(inode->i_mode))
2374                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2375         if (p->iop)
2376                 inode->i_op = p->iop;
2377         if (p->fop)
2378                 inode->i_fop = p->fop;
2379         ei->op = p->op;
2380         d_set_d_op(dentry, &pid_dentry_operations);
2381         d_add(dentry, inode);
2382         /* Close the race of the process dying before we return the dentry */
2383         if (pid_revalidate(dentry, 0))
2384                 return 0;
2385 out:
2386         return -ENOENT;
2387 }
2388
2389 static struct dentry *proc_pident_lookup(struct inode *dir,
2390                                          struct dentry *dentry,
2391                                          const struct pid_entry *ents,
2392                                          unsigned int nents)
2393 {
2394         int error;
2395         struct task_struct *task = get_proc_task(dir);
2396         const struct pid_entry *p, *last;
2397
2398         error = -ENOENT;
2399
2400         if (!task)
2401                 goto out_no_task;
2402
2403         /*
2404          * Yes, it does not scale. And it should not. Don't add
2405          * new entries into /proc/<tgid>/ without very good reasons.
2406          */
2407         last = &ents[nents - 1];
2408         for (p = ents; p <= last; p++) {
2409                 if (p->len != dentry->d_name.len)
2410                         continue;
2411                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2412                         break;
2413         }
2414         if (p > last)
2415                 goto out;
2416
2417         error = proc_pident_instantiate(dir, dentry, task, p);
2418 out:
2419         put_task_struct(task);
2420 out_no_task:
2421         return ERR_PTR(error);
2422 }
2423
2424 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2425                 const struct pid_entry *ents, unsigned int nents)
2426 {
2427         struct task_struct *task = get_proc_task(file_inode(file));
2428         const struct pid_entry *p;
2429
2430         if (!task)
2431                 return -ENOENT;
2432
2433         if (!dir_emit_dots(file, ctx))
2434                 goto out;
2435
2436         if (ctx->pos >= nents + 2)
2437                 goto out;
2438
2439         for (p = ents + (ctx->pos - 2); p <= ents + nents - 1; p++) {
2440                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2441                                 proc_pident_instantiate, task, p))
2442                         break;
2443                 ctx->pos++;
2444         }
2445 out:
2446         put_task_struct(task);
2447         return 0;
2448 }
2449
2450 #ifdef CONFIG_SECURITY
2451 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2452                                   size_t count, loff_t *ppos)
2453 {
2454         struct inode * inode = file_inode(file);
2455         char *p = NULL;
2456         ssize_t length;
2457         struct task_struct *task = get_proc_task(inode);
2458
2459         if (!task)
2460                 return -ESRCH;
2461
2462         length = security_getprocattr(task,
2463                                       (char*)file->f_path.dentry->d_name.name,
2464                                       &p);
2465         put_task_struct(task);
2466         if (length > 0)
2467                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2468         kfree(p);
2469         return length;
2470 }
2471
2472 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2473                                    size_t count, loff_t *ppos)
2474 {
2475         struct inode * inode = file_inode(file);
2476         void *page;
2477         ssize_t length;
2478         struct task_struct *task = get_proc_task(inode);
2479
2480         length = -ESRCH;
2481         if (!task)
2482                 goto out_no_task;
2483         if (count > PAGE_SIZE)
2484                 count = PAGE_SIZE;
2485
2486         /* No partial writes. */
2487         length = -EINVAL;
2488         if (*ppos != 0)
2489                 goto out;
2490
2491         page = memdup_user(buf, count);
2492         if (IS_ERR(page)) {
2493                 length = PTR_ERR(page);
2494                 goto out;
2495         }
2496
2497         /* Guard against adverse ptrace interaction */
2498         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2499         if (length < 0)
2500                 goto out_free;
2501
2502         length = security_setprocattr(task,
2503                                       (char*)file->f_path.dentry->d_name.name,
2504                                       page, count);
2505         mutex_unlock(&task->signal->cred_guard_mutex);
2506 out_free:
2507         kfree(page);
2508 out:
2509         put_task_struct(task);
2510 out_no_task:
2511         return length;
2512 }
2513
2514 static const struct file_operations proc_pid_attr_operations = {
2515         .read           = proc_pid_attr_read,
2516         .write          = proc_pid_attr_write,
2517         .llseek         = generic_file_llseek,
2518 };
2519
2520 static const struct pid_entry attr_dir_stuff[] = {
2521         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2522         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2523         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2524         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2525         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2526         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2527 };
2528
2529 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2530 {
2531         return proc_pident_readdir(file, ctx,
2532                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2533 }
2534
2535 static const struct file_operations proc_attr_dir_operations = {
2536         .read           = generic_read_dir,
2537         .iterate_shared = proc_attr_dir_readdir,
2538         .llseek         = generic_file_llseek,
2539 };
2540
2541 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2542                                 struct dentry *dentry, unsigned int flags)
2543 {
2544         return proc_pident_lookup(dir, dentry,
2545                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2546 }
2547
2548 static const struct inode_operations proc_attr_dir_inode_operations = {
2549         .lookup         = proc_attr_dir_lookup,
2550         .getattr        = pid_getattr,
2551         .setattr        = proc_setattr,
2552 };
2553
2554 #endif
2555
2556 #ifdef CONFIG_ELF_CORE
2557 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2558                                          size_t count, loff_t *ppos)
2559 {
2560         struct task_struct *task = get_proc_task(file_inode(file));
2561         struct mm_struct *mm;
2562         char buffer[PROC_NUMBUF];
2563         size_t len;
2564         int ret;
2565
2566         if (!task)
2567                 return -ESRCH;
2568
2569         ret = 0;
2570         mm = get_task_mm(task);
2571         if (mm) {
2572                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2573                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2574                                 MMF_DUMP_FILTER_SHIFT));
2575                 mmput(mm);
2576                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2577         }
2578
2579         put_task_struct(task);
2580
2581         return ret;
2582 }
2583
2584 static ssize_t proc_coredump_filter_write(struct file *file,
2585                                           const char __user *buf,
2586                                           size_t count,
2587                                           loff_t *ppos)
2588 {
2589         struct task_struct *task;
2590         struct mm_struct *mm;
2591         unsigned int val;
2592         int ret;
2593         int i;
2594         unsigned long mask;
2595
2596         ret = kstrtouint_from_user(buf, count, 0, &val);
2597         if (ret < 0)
2598                 return ret;
2599
2600         ret = -ESRCH;
2601         task = get_proc_task(file_inode(file));
2602         if (!task)
2603                 goto out_no_task;
2604
2605         mm = get_task_mm(task);
2606         if (!mm)
2607                 goto out_no_mm;
2608         ret = 0;
2609
2610         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2611                 if (val & mask)
2612                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2613                 else
2614                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2615         }
2616
2617         mmput(mm);
2618  out_no_mm:
2619         put_task_struct(task);
2620  out_no_task:
2621         if (ret < 0)
2622                 return ret;
2623         return count;
2624 }
2625
2626 static const struct file_operations proc_coredump_filter_operations = {
2627         .read           = proc_coredump_filter_read,
2628         .write          = proc_coredump_filter_write,
2629         .llseek         = generic_file_llseek,
2630 };
2631 #endif
2632
2633 #ifdef CONFIG_TASK_IO_ACCOUNTING
2634 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2635 {
2636         struct task_io_accounting acct = task->ioac;
2637         unsigned long flags;
2638         int result;
2639
2640         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2641         if (result)
2642                 return result;
2643
2644         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2645                 result = -EACCES;
2646                 goto out_unlock;
2647         }
2648
2649         if (whole && lock_task_sighand(task, &flags)) {
2650                 struct task_struct *t = task;
2651
2652                 task_io_accounting_add(&acct, &task->signal->ioac);
2653                 while_each_thread(task, t)
2654                         task_io_accounting_add(&acct, &t->ioac);
2655
2656                 unlock_task_sighand(task, &flags);
2657         }
2658         seq_printf(m,
2659                    "rchar: %llu\n"
2660                    "wchar: %llu\n"
2661                    "syscr: %llu\n"
2662                    "syscw: %llu\n"
2663                    "read_bytes: %llu\n"
2664                    "write_bytes: %llu\n"
2665                    "cancelled_write_bytes: %llu\n",
2666                    (unsigned long long)acct.rchar,
2667                    (unsigned long long)acct.wchar,
2668                    (unsigned long long)acct.syscr,
2669                    (unsigned long long)acct.syscw,
2670                    (unsigned long long)acct.read_bytes,
2671                    (unsigned long long)acct.write_bytes,
2672                    (unsigned long long)acct.cancelled_write_bytes);
2673         result = 0;
2674
2675 out_unlock:
2676         mutex_unlock(&task->signal->cred_guard_mutex);
2677         return result;
2678 }
2679
2680 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2681                                   struct pid *pid, struct task_struct *task)
2682 {
2683         return do_io_accounting(task, m, 0);
2684 }
2685
2686 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2687                                    struct pid *pid, struct task_struct *task)
2688 {
2689         return do_io_accounting(task, m, 1);
2690 }
2691 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2692
2693 #ifdef CONFIG_USER_NS
2694 static int proc_id_map_open(struct inode *inode, struct file *file,
2695         const struct seq_operations *seq_ops)
2696 {
2697         struct user_namespace *ns = NULL;
2698         struct task_struct *task;
2699         struct seq_file *seq;
2700         int ret = -EINVAL;
2701
2702         task = get_proc_task(inode);
2703         if (task) {
2704                 rcu_read_lock();
2705                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2706                 rcu_read_unlock();
2707                 put_task_struct(task);
2708         }
2709         if (!ns)
2710                 goto err;
2711
2712         ret = seq_open(file, seq_ops);
2713         if (ret)
2714                 goto err_put_ns;
2715
2716         seq = file->private_data;
2717         seq->private = ns;
2718
2719         return 0;
2720 err_put_ns:
2721         put_user_ns(ns);
2722 err:
2723         return ret;
2724 }
2725
2726 static int proc_id_map_release(struct inode *inode, struct file *file)
2727 {
2728         struct seq_file *seq = file->private_data;
2729         struct user_namespace *ns = seq->private;
2730         put_user_ns(ns);
2731         return seq_release(inode, file);
2732 }
2733
2734 static int proc_uid_map_open(struct inode *inode, struct file *file)
2735 {
2736         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2737 }
2738
2739 static int proc_gid_map_open(struct inode *inode, struct file *file)
2740 {
2741         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2742 }
2743
2744 static int proc_projid_map_open(struct inode *inode, struct file *file)
2745 {
2746         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2747 }
2748
2749 static const struct file_operations proc_uid_map_operations = {
2750         .open           = proc_uid_map_open,
2751         .write          = proc_uid_map_write,
2752         .read           = seq_read,
2753         .llseek         = seq_lseek,
2754         .release        = proc_id_map_release,
2755 };
2756
2757 static const struct file_operations proc_gid_map_operations = {
2758         .open           = proc_gid_map_open,
2759         .write          = proc_gid_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_projid_map_operations = {
2766         .open           = proc_projid_map_open,
2767         .write          = proc_projid_map_write,
2768         .read           = seq_read,
2769         .llseek         = seq_lseek,
2770         .release        = proc_id_map_release,
2771 };
2772
2773 static int proc_setgroups_open(struct inode *inode, struct file *file)
2774 {
2775         struct user_namespace *ns = NULL;
2776         struct task_struct *task;
2777         int ret;
2778
2779         ret = -ESRCH;
2780         task = get_proc_task(inode);
2781         if (task) {
2782                 rcu_read_lock();
2783                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2784                 rcu_read_unlock();
2785                 put_task_struct(task);
2786         }
2787         if (!ns)
2788                 goto err;
2789
2790         if (file->f_mode & FMODE_WRITE) {
2791                 ret = -EACCES;
2792                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2793                         goto err_put_ns;
2794         }
2795
2796         ret = single_open(file, &proc_setgroups_show, ns);
2797         if (ret)
2798                 goto err_put_ns;
2799
2800         return 0;
2801 err_put_ns:
2802         put_user_ns(ns);
2803 err:
2804         return ret;
2805 }
2806
2807 static int proc_setgroups_release(struct inode *inode, struct file *file)
2808 {
2809         struct seq_file *seq = file->private_data;
2810         struct user_namespace *ns = seq->private;
2811         int ret = single_release(inode, file);
2812         put_user_ns(ns);
2813         return ret;
2814 }
2815
2816 static const struct file_operations proc_setgroups_operations = {
2817         .open           = proc_setgroups_open,
2818         .write          = proc_setgroups_write,
2819         .read           = seq_read,
2820         .llseek         = seq_lseek,
2821         .release        = proc_setgroups_release,
2822 };
2823 #endif /* CONFIG_USER_NS */
2824
2825 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2826                                 struct pid *pid, struct task_struct *task)
2827 {
2828         int err = lock_trace(task);
2829         if (!err) {
2830                 seq_printf(m, "%08x\n", task->personality);
2831                 unlock_trace(task);
2832         }
2833         return err;
2834 }
2835
2836 /*
2837  * Thread groups
2838  */
2839 static const struct file_operations proc_task_operations;
2840 static const struct inode_operations proc_task_inode_operations;
2841
2842 static const struct pid_entry tgid_base_stuff[] = {
2843         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2844         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2845         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2846         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2847         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2848 #ifdef CONFIG_NET
2849         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2850 #endif
2851         REG("environ",    S_IRUSR, proc_environ_operations),
2852         REG("auxv",       S_IRUSR, proc_auxv_operations),
2853         ONE("status",     S_IRUGO, proc_pid_status),
2854         ONE("personality", S_IRUSR, proc_pid_personality),
2855         ONE("limits",     S_IRUGO, proc_pid_limits),
2856 #ifdef CONFIG_SCHED_DEBUG
2857         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2858 #endif
2859 #ifdef CONFIG_SCHED_AUTOGROUP
2860         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2861 #endif
2862         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2863 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2864         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2865 #endif
2866         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2867         ONE("stat",       S_IRUGO, proc_tgid_stat),
2868         ONE("statm",      S_IRUGO, proc_pid_statm),
2869         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2870 #ifdef CONFIG_NUMA
2871         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2872 #endif
2873         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2874         LNK("cwd",        proc_cwd_link),
2875         LNK("root",       proc_root_link),
2876         LNK("exe",        proc_exe_link),
2877         REG("mounts",     S_IRUGO, proc_mounts_operations),
2878         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2879         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2880 #ifdef CONFIG_PROC_PAGE_MONITOR
2881         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2882         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2883         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2884 #endif
2885 #ifdef CONFIG_SECURITY
2886         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2887 #endif
2888 #ifdef CONFIG_KALLSYMS
2889         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2890 #endif
2891 #ifdef CONFIG_STACKTRACE
2892         ONE("stack",      S_IRUSR, proc_pid_stack),
2893 #endif
2894 #ifdef CONFIG_SCHED_INFO
2895         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2896 #endif
2897 #ifdef CONFIG_LATENCYTOP
2898         REG("latency",  S_IRUGO, proc_lstats_operations),
2899 #endif
2900 #ifdef CONFIG_PROC_PID_CPUSET
2901         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2902 #endif
2903 #ifdef CONFIG_CGROUPS
2904         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2905 #endif
2906         ONE("oom_score",  S_IRUGO, proc_oom_score),
2907         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2908         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2909 #ifdef CONFIG_AUDITSYSCALL
2910         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2911         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2912 #endif
2913 #ifdef CONFIG_FAULT_INJECTION
2914         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2915 #endif
2916 #ifdef CONFIG_ELF_CORE
2917         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2918 #endif
2919 #ifdef CONFIG_TASK_IO_ACCOUNTING
2920         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2921 #endif
2922 #ifdef CONFIG_HARDWALL
2923         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2924 #endif
2925 #ifdef CONFIG_USER_NS
2926         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2927         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2928         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2929         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2930 #endif
2931 #ifdef CONFIG_CHECKPOINT_RESTORE
2932         REG("timers",     S_IRUGO, proc_timers_operations),
2933 #endif
2934         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2935 };
2936
2937 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2938 {
2939         return proc_pident_readdir(file, ctx,
2940                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2941 }
2942
2943 static const struct file_operations proc_tgid_base_operations = {
2944         .read           = generic_read_dir,
2945         .iterate_shared = proc_tgid_base_readdir,
2946         .llseek         = generic_file_llseek,
2947 };
2948
2949 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2950 {
2951         return proc_pident_lookup(dir, dentry,
2952                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2953 }
2954
2955 static const struct inode_operations proc_tgid_base_inode_operations = {
2956         .lookup         = proc_tgid_base_lookup,
2957         .getattr        = pid_getattr,
2958         .setattr        = proc_setattr,
2959         .permission     = proc_pid_permission,
2960 };
2961
2962 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2963 {
2964         struct dentry *dentry, *leader, *dir;
2965         char buf[PROC_NUMBUF];
2966         struct qstr name;
2967
2968         name.name = buf;
2969         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2970         /* no ->d_hash() rejects on procfs */
2971         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2972         if (dentry) {
2973                 d_invalidate(dentry);
2974                 dput(dentry);
2975         }
2976
2977         if (pid == tgid)
2978                 return;
2979
2980         name.name = buf;
2981         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2982         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2983         if (!leader)
2984                 goto out;
2985
2986         name.name = "task";
2987         name.len = strlen(name.name);
2988         dir = d_hash_and_lookup(leader, &name);
2989         if (!dir)
2990                 goto out_put_leader;
2991
2992         name.name = buf;
2993         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2994         dentry = d_hash_and_lookup(dir, &name);
2995         if (dentry) {
2996                 d_invalidate(dentry);
2997                 dput(dentry);
2998         }
2999
3000         dput(dir);
3001 out_put_leader:
3002         dput(leader);
3003 out:
3004         return;
3005 }
3006
3007 /**
3008  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3009  * @task: task that should be flushed.
3010  *
3011  * When flushing dentries from proc, one needs to flush them from global
3012  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3013  * in. This call is supposed to do all of this job.
3014  *
3015  * Looks in the dcache for
3016  * /proc/@pid
3017  * /proc/@tgid/task/@pid
3018  * if either directory is present flushes it and all of it'ts children
3019  * from the dcache.
3020  *
3021  * It is safe and reasonable to cache /proc entries for a task until
3022  * that task exits.  After that they just clog up the dcache with
3023  * useless entries, possibly causing useful dcache entries to be
3024  * flushed instead.  This routine is proved to flush those useless
3025  * dcache entries at process exit time.
3026  *
3027  * NOTE: This routine is just an optimization so it does not guarantee
3028  *       that no dcache entries will exist at process exit time it
3029  *       just makes it very unlikely that any will persist.
3030  */
3031
3032 void proc_flush_task(struct task_struct *task)
3033 {
3034         int i;
3035         struct pid *pid, *tgid;
3036         struct upid *upid;
3037
3038         pid = task_pid(task);
3039         tgid = task_tgid(task);
3040
3041         for (i = 0; i <= pid->level; i++) {
3042                 upid = &pid->numbers[i];
3043                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3044                                         tgid->numbers[i].nr);
3045         }
3046 }
3047
3048 static int proc_pid_instantiate(struct inode *dir,
3049                                    struct dentry * dentry,
3050                                    struct task_struct *task, const void *ptr)
3051 {
3052         struct inode *inode;
3053
3054         inode = proc_pid_make_inode(dir->i_sb, task);
3055         if (!inode)
3056                 goto out;
3057
3058         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3059         inode->i_op = &proc_tgid_base_inode_operations;
3060         inode->i_fop = &proc_tgid_base_operations;
3061         inode->i_flags|=S_IMMUTABLE;
3062
3063         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3064                                                   ARRAY_SIZE(tgid_base_stuff)));
3065
3066         d_set_d_op(dentry, &pid_dentry_operations);
3067
3068         d_add(dentry, inode);
3069         /* Close the race of the process dying before we return the dentry */
3070         if (pid_revalidate(dentry, 0))
3071                 return 0;
3072 out:
3073         return -ENOENT;
3074 }
3075
3076 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3077 {
3078         int result = -ENOENT;
3079         struct task_struct *task;
3080         unsigned tgid;
3081         struct pid_namespace *ns;
3082
3083         tgid = name_to_int(&dentry->d_name);
3084         if (tgid == ~0U)
3085                 goto out;
3086
3087         ns = dentry->d_sb->s_fs_info;
3088         rcu_read_lock();
3089         task = find_task_by_pid_ns(tgid, ns);
3090         if (task)
3091                 get_task_struct(task);
3092         rcu_read_unlock();
3093         if (!task)
3094                 goto out;
3095
3096         result = proc_pid_instantiate(dir, dentry, task, NULL);
3097         put_task_struct(task);
3098 out:
3099         return ERR_PTR(result);
3100 }
3101
3102 /*
3103  * Find the first task with tgid >= tgid
3104  *
3105  */
3106 struct tgid_iter {
3107         unsigned int tgid;
3108         struct task_struct *task;
3109 };
3110 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3111 {
3112         struct pid *pid;
3113
3114         if (iter.task)
3115                 put_task_struct(iter.task);
3116         rcu_read_lock();
3117 retry:
3118         iter.task = NULL;
3119         pid = find_ge_pid(iter.tgid, ns);
3120         if (pid) {
3121                 iter.tgid = pid_nr_ns(pid, ns);
3122                 iter.task = pid_task(pid, PIDTYPE_PID);
3123                 /* What we to know is if the pid we have find is the
3124                  * pid of a thread_group_leader.  Testing for task
3125                  * being a thread_group_leader is the obvious thing
3126                  * todo but there is a window when it fails, due to
3127                  * the pid transfer logic in de_thread.
3128                  *
3129                  * So we perform the straight forward test of seeing
3130                  * if the pid we have found is the pid of a thread
3131                  * group leader, and don't worry if the task we have
3132                  * found doesn't happen to be a thread group leader.
3133                  * As we don't care in the case of readdir.
3134                  */
3135                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3136                         iter.tgid += 1;
3137                         goto retry;
3138                 }
3139                 get_task_struct(iter.task);
3140         }
3141         rcu_read_unlock();
3142         return iter;
3143 }
3144
3145 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3146
3147 /* for the /proc/ directory itself, after non-process stuff has been done */
3148 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3149 {
3150         struct tgid_iter iter;
3151         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3152         loff_t pos = ctx->pos;
3153
3154         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3155                 return 0;
3156
3157         if (pos == TGID_OFFSET - 2) {
3158                 struct inode *inode = d_inode(ns->proc_self);
3159                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3160                         return 0;
3161                 ctx->pos = pos = pos + 1;
3162         }
3163         if (pos == TGID_OFFSET - 1) {
3164                 struct inode *inode = d_inode(ns->proc_thread_self);
3165                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3166                         return 0;
3167                 ctx->pos = pos = pos + 1;
3168         }
3169         iter.tgid = pos - TGID_OFFSET;
3170         iter.task = NULL;
3171         for (iter = next_tgid(ns, iter);
3172              iter.task;
3173              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3174                 char name[PROC_NUMBUF];
3175                 int len;
3176                 if (!has_pid_permissions(ns, iter.task, 2))
3177                         continue;
3178
3179                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3180                 ctx->pos = iter.tgid + TGID_OFFSET;
3181                 if (!proc_fill_cache(file, ctx, name, len,
3182                                      proc_pid_instantiate, iter.task, NULL)) {
3183                         put_task_struct(iter.task);
3184                         return 0;
3185                 }
3186         }
3187         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3188         return 0;
3189 }
3190
3191 /*
3192  * proc_tid_comm_permission is a special permission function exclusively
3193  * used for the node /proc/<pid>/task/<tid>/comm.
3194  * It bypasses generic permission checks in the case where a task of the same
3195  * task group attempts to access the node.
3196  * The rationale behind this is that glibc and bionic access this node for
3197  * cross thread naming (pthread_set/getname_np(!self)). However, if
3198  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3199  * which locks out the cross thread naming implementation.
3200  * This function makes sure that the node is always accessible for members of
3201  * same thread group.
3202  */
3203 static int proc_tid_comm_permission(struct inode *inode, int mask)
3204 {
3205         bool is_same_tgroup;
3206         struct task_struct *task;
3207
3208         task = get_proc_task(inode);
3209         if (!task)
3210                 return -ESRCH;
3211         is_same_tgroup = same_thread_group(current, task);
3212         put_task_struct(task);
3213
3214         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3215                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3216                  * read or written by the members of the corresponding
3217                  * thread group.
3218                  */
3219                 return 0;
3220         }
3221
3222         return generic_permission(inode, mask);
3223 }
3224
3225 static const struct inode_operations proc_tid_comm_inode_operations = {
3226                 .permission = proc_tid_comm_permission,
3227 };
3228
3229 /*
3230  * Tasks
3231  */
3232 static const struct pid_entry tid_base_stuff[] = {
3233         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3234         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3235         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3236 #ifdef CONFIG_NET
3237         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3238 #endif
3239         REG("environ",   S_IRUSR, proc_environ_operations),
3240         REG("auxv",      S_IRUSR, proc_auxv_operations),
3241         ONE("status",    S_IRUGO, proc_pid_status),
3242         ONE("personality", S_IRUSR, proc_pid_personality),
3243         ONE("limits",    S_IRUGO, proc_pid_limits),
3244 #ifdef CONFIG_SCHED_DEBUG
3245         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3246 #endif
3247         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3248                          &proc_tid_comm_inode_operations,
3249                          &proc_pid_set_comm_operations, {}),
3250 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3251         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3252 #endif
3253         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3254         ONE("stat",      S_IRUGO, proc_tid_stat),
3255         ONE("statm",     S_IRUGO, proc_pid_statm),
3256         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3257 #ifdef CONFIG_PROC_CHILDREN
3258         REG("children",  S_IRUGO, proc_tid_children_operations),
3259 #endif
3260 #ifdef CONFIG_NUMA
3261         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3262 #endif
3263         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3264         LNK("cwd",       proc_cwd_link),
3265         LNK("root",      proc_root_link),
3266         LNK("exe",       proc_exe_link),
3267         REG("mounts",    S_IRUGO, proc_mounts_operations),
3268         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3269 #ifdef CONFIG_PROC_PAGE_MONITOR
3270         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3271         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3272         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3273 #endif
3274 #ifdef CONFIG_SECURITY
3275         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3276 #endif
3277 #ifdef CONFIG_KALLSYMS
3278         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3279 #endif
3280 #ifdef CONFIG_STACKTRACE
3281         ONE("stack",      S_IRUSR, proc_pid_stack),
3282 #endif
3283 #ifdef CONFIG_SCHED_INFO
3284         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3285 #endif
3286 #ifdef CONFIG_LATENCYTOP
3287         REG("latency",  S_IRUGO, proc_lstats_operations),
3288 #endif
3289 #ifdef CONFIG_PROC_PID_CPUSET
3290         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3291 #endif
3292 #ifdef CONFIG_CGROUPS
3293         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3294 #endif
3295         ONE("oom_score", S_IRUGO, proc_oom_score),
3296         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3297         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3298 #ifdef CONFIG_AUDITSYSCALL
3299         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3300         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3301 #endif
3302 #ifdef CONFIG_FAULT_INJECTION
3303         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3304 #endif
3305 #ifdef CONFIG_TASK_IO_ACCOUNTING
3306         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3307 #endif
3308 #ifdef CONFIG_HARDWALL
3309         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3310 #endif
3311 #ifdef CONFIG_USER_NS
3312         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3313         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3314         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3315         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3316 #endif
3317 };
3318
3319 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3320 {
3321         return proc_pident_readdir(file, ctx,
3322                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3323 }
3324
3325 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3326 {
3327         return proc_pident_lookup(dir, dentry,
3328                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3329 }
3330
3331 static const struct file_operations proc_tid_base_operations = {
3332         .read           = generic_read_dir,
3333         .iterate_shared = proc_tid_base_readdir,
3334         .llseek         = generic_file_llseek,
3335 };
3336
3337 static const struct inode_operations proc_tid_base_inode_operations = {
3338         .lookup         = proc_tid_base_lookup,
3339         .getattr        = pid_getattr,
3340         .setattr        = proc_setattr,
3341 };
3342
3343 static int proc_task_instantiate(struct inode *dir,
3344         struct dentry *dentry, struct task_struct *task, const void *ptr)
3345 {
3346         struct inode *inode;
3347         inode = proc_pid_make_inode(dir->i_sb, task);
3348
3349         if (!inode)
3350                 goto out;
3351         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3352         inode->i_op = &proc_tid_base_inode_operations;
3353         inode->i_fop = &proc_tid_base_operations;
3354         inode->i_flags|=S_IMMUTABLE;
3355
3356         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3357                                                   ARRAY_SIZE(tid_base_stuff)));
3358
3359         d_set_d_op(dentry, &pid_dentry_operations);
3360
3361         d_add(dentry, inode);
3362         /* Close the race of the process dying before we return the dentry */
3363         if (pid_revalidate(dentry, 0))
3364                 return 0;
3365 out:
3366         return -ENOENT;
3367 }
3368
3369 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3370 {
3371         int result = -ENOENT;
3372         struct task_struct *task;
3373         struct task_struct *leader = get_proc_task(dir);
3374         unsigned tid;
3375         struct pid_namespace *ns;
3376
3377         if (!leader)
3378                 goto out_no_task;
3379
3380         tid = name_to_int(&dentry->d_name);
3381         if (tid == ~0U)
3382                 goto out;
3383
3384         ns = dentry->d_sb->s_fs_info;
3385         rcu_read_lock();
3386         task = find_task_by_pid_ns(tid, ns);
3387         if (task)
3388                 get_task_struct(task);
3389         rcu_read_unlock();
3390         if (!task)
3391                 goto out;
3392         if (!same_thread_group(leader, task))
3393                 goto out_drop_task;
3394
3395         result = proc_task_instantiate(dir, dentry, task, NULL);
3396 out_drop_task:
3397         put_task_struct(task);
3398 out:
3399         put_task_struct(leader);
3400 out_no_task:
3401         return ERR_PTR(result);
3402 }
3403
3404 /*
3405  * Find the first tid of a thread group to return to user space.
3406  *
3407  * Usually this is just the thread group leader, but if the users
3408  * buffer was too small or there was a seek into the middle of the
3409  * directory we have more work todo.
3410  *
3411  * In the case of a short read we start with find_task_by_pid.
3412  *
3413  * In the case of a seek we start with the leader and walk nr
3414  * threads past it.
3415  */
3416 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3417                                         struct pid_namespace *ns)
3418 {
3419         struct task_struct *pos, *task;
3420         unsigned long nr = f_pos;
3421
3422         if (nr != f_pos)        /* 32bit overflow? */
3423                 return NULL;
3424
3425         rcu_read_lock();
3426         task = pid_task(pid, PIDTYPE_PID);
3427         if (!task)
3428                 goto fail;
3429
3430         /* Attempt to start with the tid of a thread */
3431         if (tid && nr) {
3432                 pos = find_task_by_pid_ns(tid, ns);
3433                 if (pos && same_thread_group(pos, task))
3434                         goto found;
3435         }
3436
3437         /* If nr exceeds the number of threads there is nothing todo */
3438         if (nr >= get_nr_threads(task))
3439                 goto fail;
3440
3441         /* If we haven't found our starting place yet start
3442          * with the leader and walk nr threads forward.
3443          */
3444         pos = task = task->group_leader;
3445         do {
3446                 if (!nr--)
3447                         goto found;
3448         } while_each_thread(task, pos);
3449 fail:
3450         pos = NULL;
3451         goto out;
3452 found:
3453         get_task_struct(pos);
3454 out:
3455         rcu_read_unlock();
3456         return pos;
3457 }
3458
3459 /*
3460  * Find the next thread in the thread list.
3461  * Return NULL if there is an error or no next thread.
3462  *
3463  * The reference to the input task_struct is released.
3464  */
3465 static struct task_struct *next_tid(struct task_struct *start)
3466 {
3467         struct task_struct *pos = NULL;
3468         rcu_read_lock();
3469         if (pid_alive(start)) {
3470                 pos = next_thread(start);
3471                 if (thread_group_leader(pos))
3472                         pos = NULL;
3473                 else
3474                         get_task_struct(pos);
3475         }
3476         rcu_read_unlock();
3477         put_task_struct(start);
3478         return pos;
3479 }
3480
3481 /* for the /proc/TGID/task/ directories */
3482 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3483 {
3484         struct inode *inode = file_inode(file);
3485         struct task_struct *task;
3486         struct pid_namespace *ns;
3487         int tid;
3488
3489         if (proc_inode_is_dead(inode))
3490                 return -ENOENT;
3491
3492         if (!dir_emit_dots(file, ctx))
3493                 return 0;
3494
3495         /* f_version caches the tgid value that the last readdir call couldn't
3496          * return. lseek aka telldir automagically resets f_version to 0.
3497          */
3498         ns = inode->i_sb->s_fs_info;
3499         tid = (int)file->f_version;
3500         file->f_version = 0;
3501         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3502              task;
3503              task = next_tid(task), ctx->pos++) {
3504                 char name[PROC_NUMBUF];
3505                 int len;
3506                 tid = task_pid_nr_ns(task, ns);
3507                 len = snprintf(name, sizeof(name), "%d", tid);
3508                 if (!proc_fill_cache(file, ctx, name, len,
3509                                 proc_task_instantiate, task, NULL)) {
3510                         /* returning this tgid failed, save it as the first
3511                          * pid for the next readir call */
3512                         file->f_version = (u64)tid;
3513                         put_task_struct(task);
3514                         break;
3515                 }
3516         }
3517
3518         return 0;
3519 }
3520
3521 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3522 {
3523         struct inode *inode = d_inode(dentry);
3524         struct task_struct *p = get_proc_task(inode);
3525         generic_fillattr(inode, stat);
3526
3527         if (p) {
3528                 stat->nlink += get_nr_threads(p);
3529                 put_task_struct(p);
3530         }
3531
3532         return 0;
3533 }
3534
3535 static const struct inode_operations proc_task_inode_operations = {
3536         .lookup         = proc_task_lookup,
3537         .getattr        = proc_task_getattr,
3538         .setattr        = proc_setattr,
3539         .permission     = proc_pid_permission,
3540 };
3541
3542 static const struct file_operations proc_task_operations = {
3543         .read           = generic_read_dir,
3544         .iterate_shared = proc_task_readdir,
3545         .llseek         = generic_file_llseek,
3546 };