fs/proc/base.c

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