fs/proc/base.c

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