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