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/sched/mm.h>
  90 #include <linux/sched/coredump.h>
  91 #include <linux/flex_array.h>
  92 #include <linux/posix-timers.h>
  93 #ifdef CONFIG_HARDWALL
  94 #include <asm/hardwall.h>
  95 #endif
  96 #include <trace/events/oom.h>
  97 #include "internal.h"
  98 #include "fd.h"
  99
 100 /* NOTE:
 101  *      Implementing inode permission operations in /proc is almost
 102  *      certainly an error.  Permission checks need to happen during
 103  *      each system call not at open time.  The reason is that most of
 104  *      what we wish to check for permissions in /proc varies at runtime.
 105  *
 106  *      The classic example of a problem is opening file descriptors
 107  *      in /proc for a task before it execs a suid executable.
 108  */
 109
 110 static u8 nlink_tid;
 111 static u8 nlink_tgid;
 112
 113 struct pid_entry {
 114         const char *name;
 115         unsigned int len;
 116         umode_t mode;
 117         const struct inode_operations *iop;
 118         const struct file_operations *fop;
 119         union proc_op op;
 120 };
 121
 122 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 123         .name = (NAME),                                 \
 124         .len  = sizeof(NAME) - 1,                       \
 125         .mode = MODE,                                   \
 126         .iop  = IOP,                                    \
 127         .fop  = FOP,                                    \
 128         .op   = OP,                                     \
 129 }
 130
 131 #define DIR(NAME, MODE, iops, fops)     \
 132         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 133 #define LNK(NAME, get_link)                                     \
 134         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 135                 &proc_pid_link_inode_operations, NULL,          \
 136                 { .proc_get_link = get_link } )
 137 #define REG(NAME, MODE, fops)                           \
 138         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 139 #define ONE(NAME, MODE, show)                           \
 140         NOD(NAME, (S_IFREG|(MODE)),                     \
 141                 NULL, &proc_single_file_operations,     \
 142                 { .proc_show = show } )
 143
 144 /*
 145  * Count the number of hardlinks for the pid_entry table, excluding the .
 146  * and .. links.
 147  */
 148 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 149         unsigned int n)
 150 {
 151         unsigned int i;
 152         unsigned int count;
 153
 154         count = 2;
 155         for (i = 0; i < n; ++i) {
 156                 if (S_ISDIR(entries[i].mode))
 157                         ++count;
 158         }
 159
 160         return count;
 161 }
 162
 163 static int get_task_root(struct task_struct *task, struct path *root)
 164 {
 165         int result = -ENOENT;
 166
 167         task_lock(task);
 168         if (task->fs) {
 169                 get_fs_root(task->fs, root);
 170                 result = 0;
 171         }
 172         task_unlock(task);
 173         return result;
 174 }
 175
 176 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 177 {
 178         struct task_struct *task = get_proc_task(d_inode(dentry));
 179         int result = -ENOENT;
 180
 181         if (task) {
 182                 task_lock(task);
 183                 if (task->fs) {
 184                         get_fs_pwd(task->fs, path);
 185                         result = 0;
 186                 }
 187                 task_unlock(task);
 188                 put_task_struct(task);
 189         }
 190         return result;
 191 }
 192
 193 static int proc_root_link(struct dentry *dentry, struct path *path)
 194 {
 195         struct task_struct *task = get_proc_task(d_inode(dentry));
 196         int result = -ENOENT;
 197
 198         if (task) {
 199                 result = get_task_root(task, path);
 200                 put_task_struct(task);
 201         }
 202         return result;
 203 }
 204
 205 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 206                                      size_t _count, loff_t *pos)
 207 {
 208         struct task_struct *tsk;
 209         struct mm_struct *mm;
 210         char *page;
 211         unsigned long count = _count;
 212         unsigned long arg_start, arg_end, env_start, env_end;
 213         unsigned long len1, len2, len;
 214         unsigned long p;
 215         char c;
 216         ssize_t rv;
 217
 218         BUG_ON(*pos < 0);
 219
 220         tsk = get_proc_task(file_inode(file));
 221         if (!tsk)
 222                 return -ESRCH;
 223         mm = get_task_mm(tsk);
 224         put_task_struct(tsk);
 225         if (!mm)
 226                 return 0;
 227         /* Check if process spawned far enough to have cmdline. */
 228         if (!mm->env_end) {
 229                 rv = 0;
 230                 goto out_mmput;
 231         }
 232
 233         page = (char *)__get_free_page(GFP_TEMPORARY);
 234         if (!page) {
 235                 rv = -ENOMEM;
 236                 goto out_mmput;
 237         }
 238
 239         down_read(&mm->mmap_sem);
 240         arg_start = mm->arg_start;
 241         arg_end = mm->arg_end;
 242         env_start = mm->env_start;
 243         env_end = mm->env_end;
 244         up_read(&mm->mmap_sem);
 245
 246         BUG_ON(arg_start > arg_end);
 247         BUG_ON(env_start > env_end);
 248
 249         len1 = arg_end - arg_start;
 250         len2 = env_end - env_start;
 251
 252         /* Empty ARGV. */
 253         if (len1 == 0) {
 254                 rv = 0;
 255                 goto out_free_page;
 256         }
 257         /*
 258          * Inherently racy -- command line shares address space
 259          * with code and data.
 260          */
 261         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 262         if (rv <= 0)
 263                 goto out_free_page;
 264
 265         rv = 0;
 266
 267         if (c == '\0') {
 268                 /* Command line (set of strings) occupies whole ARGV. */
 269                 if (len1 <= *pos)
 270                         goto out_free_page;
 271
 272                 p = arg_start + *pos;
 273                 len = len1 - *pos;
 274                 while (count > 0 && len > 0) {
 275                         unsigned int _count;
 276                         int nr_read;
 277
 278                         _count = min3(count, len, PAGE_SIZE);
 279                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 280                         if (nr_read < 0)
 281                                 rv = nr_read;
 282                         if (nr_read <= 0)
 283                                 goto out_free_page;
 284
 285                         if (copy_to_user(buf, page, nr_read)) {
 286                                 rv = -EFAULT;
 287                                 goto out_free_page;
 288                         }
 289
 290                         p       += nr_read;
 291                         len     -= nr_read;
 292                         buf     += nr_read;
 293                         count   -= nr_read;
 294                         rv      += nr_read;
 295                 }
 296         } else {
 297                 /*
 298                  * Command line (1 string) occupies ARGV and
 299                  * extends into ENVP.
 300                  */
 301                 struct {
 302                         unsigned long p;
 303                         unsigned long len;
 304                 } cmdline[2] = {
 305                         { .p = arg_start, .len = len1 },
 306                         { .p = env_start, .len = len2 },
 307                 };
 308                 loff_t pos1 = *pos;
 309                 unsigned int i;
 310
 311                 i = 0;
 312                 while (i < 2 && pos1 >= cmdline[i].len) {
 313                         pos1 -= cmdline[i].len;
 314                         i++;
 315                 }
 316                 while (i < 2) {
 317                         p = cmdline[i].p + pos1;
 318                         len = cmdline[i].len - pos1;
 319                         while (count > 0 && len > 0) {
 320                                 unsigned int _count, l;
 321                                 int nr_read;
 322                                 bool final;
 323
 324                                 _count = min3(count, len, PAGE_SIZE);
 325                                 nr_read = access_remote_vm(mm, p, page, _count, 0);
 326                                 if (nr_read < 0)
 327                                         rv = nr_read;
 328                                 if (nr_read <= 0)
 329                                         goto out_free_page;
 330
 331                                 /*
 332                                  * Command line can be shorter than whole ARGV
 333                                  * even if last "marker" byte says it is not.
 334                                  */
 335                                 final = false;
 336                                 l = strnlen(page, nr_read);
 337                                 if (l < nr_read) {
 338                                         nr_read = l;
 339                                         final = true;
 340                                 }
 341
 342                                 if (copy_to_user(buf, page, nr_read)) {
 343                                         rv = -EFAULT;
 344                                         goto out_free_page;
 345                                 }
 346
 347                                 p       += nr_read;
 348                                 len     -= nr_read;
 349                                 buf     += nr_read;
 350                                 count   -= nr_read;
 351                                 rv      += nr_read;
 352
 353                                 if (final)
 354                                         goto out_free_page;
 355                         }
 356
 357                         /* Only first chunk can be read partially. */
 358                         pos1 = 0;
 359                         i++;
 360                 }
 361         }
 362
 363 out_free_page:
 364         free_page((unsigned long)page);
 365 out_mmput:
 366         mmput(mm);
 367         if (rv > 0)
 368                 *pos += rv;
 369         return rv;
 370 }
 371
 372 static const struct file_operations proc_pid_cmdline_ops = {
 373         .read   = proc_pid_cmdline_read,
 374         .llseek = generic_file_llseek,
 375 };
 376
 377 #ifdef CONFIG_KALLSYMS
 378 /*
 379  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 380  * Returns the resolved symbol.  If that fails, simply return the address.
 381  */
 382 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 383                           struct pid *pid, struct task_struct *task)
 384 {
 385         unsigned long wchan;
 386         char symname[KSYM_NAME_LEN];
 387
 388         wchan = get_wchan(task);
 389
 390         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 391                         && !lookup_symbol_name(wchan, symname))
 392                 seq_printf(m, "%s", symname);
 393         else
 394                 seq_putc(m, '0');
 395
 396         return 0;
 397 }
 398 #endif /* CONFIG_KALLSYMS */
 399
 400 static int lock_trace(struct task_struct *task)
 401 {
 402         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 403         if (err)
 404                 return err;
 405         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 406                 mutex_unlock(&task->signal->cred_guard_mutex);
 407                 return -EPERM;
 408         }
 409         return 0;
 410 }
 411
 412 static void unlock_trace(struct task_struct *task)
 413 {
 414         mutex_unlock(&task->signal->cred_guard_mutex);
 415 }
 416
 417 #ifdef CONFIG_STACKTRACE
 418
 419 #define MAX_STACK_TRACE_DEPTH   64
 420
 421 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 422                           struct pid *pid, struct task_struct *task)
 423 {
 424         struct stack_trace trace;
 425         unsigned long *entries;
 426         int err;
 427         int i;
 428
 429         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 430         if (!entries)
 431                 return -ENOMEM;
 432
 433         trace.nr_entries        = 0;
 434         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 435         trace.entries           = entries;
 436         trace.skip              = 0;
 437
 438         err = lock_trace(task);
 439         if (!err) {
 440                 save_stack_trace_tsk(task, &trace);
 441
 442                 for (i = 0; i < trace.nr_entries; i++) {
 443                         seq_printf(m, "[<%pK>] %pB\n",
 444                                    (void *)entries[i], (void *)entries[i]);
 445                 }
 446                 unlock_trace(task);
 447         }
 448         kfree(entries);
 449
 450         return err;
 451 }
 452 #endif
 453
 454 #ifdef CONFIG_SCHED_INFO
 455 /*
 456  * Provides /proc/PID/schedstat
 457  */
 458 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 459                               struct pid *pid, struct task_struct *task)
 460 {
 461         if (unlikely(!sched_info_on()))
 462                 seq_printf(m, "0 0 0\n");
 463         else
 464                 seq_printf(m, "%llu %llu %lu\n",
 465                    (unsigned long long)task->se.sum_exec_runtime,
 466                    (unsigned long long)task->sched_info.run_delay,
 467                    task->sched_info.pcount);
 468
 469         return 0;
 470 }
 471 #endif
 472
 473 #ifdef CONFIG_LATENCYTOP
 474 static int lstats_show_proc(struct seq_file *m, void *v)
 475 {
 476         int i;
 477         struct inode *inode = m->private;
 478         struct task_struct *task = get_proc_task(inode);
 479
 480         if (!task)
 481                 return -ESRCH;
 482         seq_puts(m, "Latency Top version : v0.1\n");
 483         for (i = 0; i < 32; i++) {
 484                 struct latency_record *lr = &task->latency_record[i];
 485                 if (lr->backtrace[0]) {
 486                         int q;
 487                         seq_printf(m, "%i %li %li",
 488                                    lr->count, lr->time, lr->max);
 489                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 490                                 unsigned long bt = lr->backtrace[q];
 491                                 if (!bt)
 492                                         break;
 493                                 if (bt == ULONG_MAX)
 494                                         break;
 495                                 seq_printf(m, " %ps", (void *)bt);
 496                         }
 497                         seq_putc(m, '\n');
 498                 }
 499
 500         }
 501         put_task_struct(task);
 502         return 0;
 503 }
 504
 505 static int lstats_open(struct inode *inode, struct file *file)
 506 {
 507         return single_open(file, lstats_show_proc, inode);
 508 }
 509
 510 static ssize_t lstats_write(struct file *file, const char __user *buf,
 511                             size_t count, loff_t *offs)
 512 {
 513         struct task_struct *task = get_proc_task(file_inode(file));
 514
 515         if (!task)
 516                 return -ESRCH;
 517         clear_all_latency_tracing(task);
 518         put_task_struct(task);
 519
 520         return count;
 521 }
 522
 523 static const struct file_operations proc_lstats_operations = {
 524         .open           = lstats_open,
 525         .read           = seq_read,
 526         .write          = lstats_write,
 527         .llseek         = seq_lseek,
 528         .release        = single_release,
 529 };
 530
 531 #endif
 532
 533 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 534                           struct pid *pid, struct task_struct *task)
 535 {
 536         unsigned long totalpages = totalram_pages + total_swap_pages;
 537         unsigned long points = 0;
 538
 539         points = oom_badness(task, NULL, NULL, totalpages) *
 540                                         1000 / totalpages;
 541         seq_printf(m, "%lu\n", points);
 542
 543         return 0;
 544 }
 545
 546 struct limit_names {
 547         const char *name;
 548         const char *unit;
 549 };
 550
 551 static const struct limit_names lnames[RLIM_NLIMITS] = {
 552         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 553         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 554         [RLIMIT_DATA] = {"Max data size", "bytes"},
 555         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 556         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 557         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 558         [RLIMIT_NPROC] = {"Max processes", "processes"},
 559         [RLIMIT_NOFILE] = {"Max open files", "files"},
 560         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 561         [RLIMIT_AS] = {"Max address space", "bytes"},
 562         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 563         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 564         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 565         [RLIMIT_NICE] = {"Max nice priority", NULL},
 566         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 567         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 568 };
 569
 570 /* Display limits for a process */
 571 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 572                            struct pid *pid, struct task_struct *task)
 573 {
 574         unsigned int i;
 575         unsigned long flags;
 576
 577         struct rlimit rlim[RLIM_NLIMITS];
 578
 579         if (!lock_task_sighand(task, &flags))
 580                 return 0;
 581         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 582         unlock_task_sighand(task, &flags);
 583
 584         /*
 585          * print the file header
 586          */
 587        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 588                   "Limit", "Soft Limit", "Hard Limit", "Units");
 589
 590         for (i = 0; i < RLIM_NLIMITS; i++) {
 591                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 592                         seq_printf(m, "%-25s %-20s ",
 593                                    lnames[i].name, "unlimited");
 594                 else
 595                         seq_printf(m, "%-25s %-20lu ",
 596                                    lnames[i].name, rlim[i].rlim_cur);
 597
 598                 if (rlim[i].rlim_max == RLIM_INFINITY)
 599                         seq_printf(m, "%-20s ", "unlimited");
 600                 else
 601                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 602
 603                 if (lnames[i].unit)
 604                         seq_printf(m, "%-10s\n", lnames[i].unit);
 605                 else
 606                         seq_putc(m, '\n');
 607         }
 608
 609         return 0;
 610 }
 611
 612 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 613 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 614                             struct pid *pid, struct task_struct *task)
 615 {
 616         long nr;
 617         unsigned long args[6], sp, pc;
 618         int res;
 619
 620         res = lock_trace(task);
 621         if (res)
 622                 return res;
 623
 624         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 625                 seq_puts(m, "running\n");
 626         else if (nr < 0)
 627                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 628         else
 629                 seq_printf(m,
 630                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 631                        nr,
 632                        args[0], args[1], args[2], args[3], args[4], args[5],
 633                        sp, pc);
 634         unlock_trace(task);
 635
 636         return 0;
 637 }
 638 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 639
 640 /************************************************************************/
 641 /*                       Here the fs part begins                        */
 642 /************************************************************************/
 643
 644 /* permission checks */
 645 static int proc_fd_access_allowed(struct inode *inode)
 646 {
 647         struct task_struct *task;
 648         int allowed = 0;
 649         /* Allow access to a task's file descriptors if it is us or we
 650          * may use ptrace attach to the process and find out that
 651          * information.
 652          */
 653         task = get_proc_task(inode);
 654         if (task) {
 655                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 656                 put_task_struct(task);
 657         }
 658         return allowed;
 659 }
 660
 661 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 662 {
 663         int error;
 664         struct inode *inode = d_inode(dentry);
 665
 666         if (attr->ia_valid & ATTR_MODE)
 667                 return -EPERM;
 668
 669         error = setattr_prepare(dentry, attr);
 670         if (error)
 671                 return error;
 672
 673         setattr_copy(inode, attr);
 674         mark_inode_dirty(inode);
 675         return 0;
 676 }
 677
 678 /*
 679  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 680  * or euid/egid (for hide_pid_min=2)?
 681  */
 682 static bool has_pid_permissions(struct pid_namespace *pid,
 683                                  struct task_struct *task,
 684                                  int hide_pid_min)
 685 {
 686         if (pid->hide_pid < hide_pid_min)
 687                 return true;
 688         if (in_group_p(pid->pid_gid))
 689                 return true;
 690         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 691 }
 692
 693
 694 static int proc_pid_permission(struct inode *inode, int mask)
 695 {
 696         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 697         struct task_struct *task;
 698         bool has_perms;
 699
 700         task = get_proc_task(inode);
 701         if (!task)
 702                 return -ESRCH;
 703         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 704         put_task_struct(task);
 705
 706         if (!has_perms) {
 707                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 708                         /*
 709                          * Let's make getdents(), stat(), and open()
 710                          * consistent with each other.  If a process
 711                          * may not stat() a file, it shouldn't be seen
 712                          * in procfs at all.
 713                          */
 714                         return -ENOENT;
 715                 }
 716
 717                 return -EPERM;
 718         }
 719         return generic_permission(inode, mask);
 720 }
 721
 722
 723
 724 static const struct inode_operations proc_def_inode_operations = {
 725         .setattr        = proc_setattr,
 726 };
 727
 728 static int proc_single_show(struct seq_file *m, void *v)
 729 {
 730         struct inode *inode = m->private;
 731         struct pid_namespace *ns;
 732         struct pid *pid;
 733         struct task_struct *task;
 734         int ret;
 735
 736         ns = inode->i_sb->s_fs_info;
 737         pid = proc_pid(inode);
 738         task = get_pid_task(pid, PIDTYPE_PID);
 739         if (!task)
 740                 return -ESRCH;
 741
 742         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 743
 744         put_task_struct(task);
 745         return ret;
 746 }
 747
 748 static int proc_single_open(struct inode *inode, struct file *filp)
 749 {
 750         return single_open(filp, proc_single_show, inode);
 751 }
 752
 753 static const struct file_operations proc_single_file_operations = {
 754         .open           = proc_single_open,
 755         .read           = seq_read,
 756         .llseek         = seq_lseek,
 757         .release        = single_release,
 758 };
 759
 760
 761 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 762 {
 763         struct task_struct *task = get_proc_task(inode);
 764         struct mm_struct *mm = ERR_PTR(-ESRCH);
 765
 766         if (task) {
 767                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 768                 put_task_struct(task);
 769
 770                 if (!IS_ERR_OR_NULL(mm)) {
 771                         /* ensure this mm_struct can't be freed */
 772                         mmgrab(mm);
 773                         /* but do not pin its memory */
 774                         mmput(mm);
 775                 }
 776         }
 777
 778         return mm;
 779 }
 780
 781 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 782 {
 783         struct mm_struct *mm = proc_mem_open(inode, mode);
 784
 785         if (IS_ERR(mm))
 786                 return PTR_ERR(mm);
 787
 788         file->private_data = mm;
 789         return 0;
 790 }
 791
 792 static int mem_open(struct inode *inode, struct file *file)
 793 {
 794         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 795
 796         /* OK to pass negative loff_t, we can catch out-of-range */
 797         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 798
 799         return ret;
 800 }
 801
 802 static ssize_t mem_rw(struct file *file, char __user *buf,
 803                         size_t count, loff_t *ppos, int write)
 804 {
 805         struct mm_struct *mm = file->private_data;
 806         unsigned long addr = *ppos;
 807         ssize_t copied;
 808         char *page;
 809         unsigned int flags;
 810
 811         if (!mm)
 812                 return 0;
 813
 814         page = (char *)__get_free_page(GFP_TEMPORARY);
 815         if (!page)
 816                 return -ENOMEM;
 817
 818         copied = 0;
 819         if (!mmget_not_zero(mm))
 820                 goto free;
 821
 822         /* Maybe we should limit FOLL_FORCE to actual ptrace users? */
 823         flags = FOLL_FORCE;
 824         if (write)
 825                 flags |= FOLL_WRITE;
 826
 827         while (count > 0) {
 828                 int this_len = min_t(int, count, PAGE_SIZE);
 829
 830                 if (write && copy_from_user(page, buf, this_len)) {
 831                         copied = -EFAULT;
 832                         break;
 833                 }
 834
 835                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 836                 if (!this_len) {
 837                         if (!copied)
 838                                 copied = -EIO;
 839                         break;
 840                 }
 841
 842                 if (!write && copy_to_user(buf, page, this_len)) {
 843                         copied = -EFAULT;
 844                         break;
 845                 }
 846
 847                 buf += this_len;
 848                 addr += this_len;
 849                 copied += this_len;
 850                 count -= this_len;
 851         }
 852         *ppos = addr;
 853
 854         mmput(mm);
 855 free:
 856         free_page((unsigned long) page);
 857         return copied;
 858 }
 859
 860 static ssize_t mem_read(struct file *file, char __user *buf,
 861                         size_t count, loff_t *ppos)
 862 {
 863         return mem_rw(file, buf, count, ppos, 0);
 864 }
 865
 866 static ssize_t mem_write(struct file *file, const char __user *buf,
 867                          size_t count, loff_t *ppos)
 868 {
 869         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 870 }
 871
 872 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 873 {
 874         switch (orig) {
 875         case 0:
 876                 file->f_pos = offset;
 877                 break;
 878         case 1:
 879                 file->f_pos += offset;
 880                 break;
 881         default:
 882                 return -EINVAL;
 883         }
 884         force_successful_syscall_return();
 885         return file->f_pos;
 886 }
 887
 888 static int mem_release(struct inode *inode, struct file *file)
 889 {
 890         struct mm_struct *mm = file->private_data;
 891         if (mm)
 892                 mmdrop(mm);
 893         return 0;
 894 }
 895
 896 static const struct file_operations proc_mem_operations = {
 897         .llseek         = mem_lseek,
 898         .read           = mem_read,
 899         .write          = mem_write,
 900         .open           = mem_open,
 901         .release        = mem_release,
 902 };
 903
 904 static int environ_open(struct inode *inode, struct file *file)
 905 {
 906         return __mem_open(inode, file, PTRACE_MODE_READ);
 907 }
 908
 909 static ssize_t environ_read(struct file *file, char __user *buf,
 910                         size_t count, loff_t *ppos)
 911 {
 912         char *page;
 913         unsigned long src = *ppos;
 914         int ret = 0;
 915         struct mm_struct *mm = file->private_data;
 916         unsigned long env_start, env_end;
 917
 918         /* Ensure the process spawned far enough to have an environment. */
 919         if (!mm || !mm->env_end)
 920                 return 0;
 921
 922         page = (char *)__get_free_page(GFP_TEMPORARY);
 923         if (!page)
 924                 return -ENOMEM;
 925
 926         ret = 0;
 927         if (!mmget_not_zero(mm))
 928                 goto free;
 929
 930         down_read(&mm->mmap_sem);
 931         env_start = mm->env_start;
 932         env_end = mm->env_end;
 933         up_read(&mm->mmap_sem);
 934
 935         while (count > 0) {
 936                 size_t this_len, max_len;
 937                 int retval;
 938
 939                 if (src >= (env_end - env_start))
 940                         break;
 941
 942                 this_len = env_end - (env_start + src);
 943
 944                 max_len = min_t(size_t, PAGE_SIZE, count);
 945                 this_len = min(max_len, this_len);
 946
 947                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 948
 949                 if (retval <= 0) {
 950                         ret = retval;
 951                         break;
 952                 }
 953
 954                 if (copy_to_user(buf, page, retval)) {
 955                         ret = -EFAULT;
 956                         break;
 957                 }
 958
 959                 ret += retval;
 960                 src += retval;
 961                 buf += retval;
 962                 count -= retval;
 963         }
 964         *ppos = src;
 965         mmput(mm);
 966
 967 free:
 968         free_page((unsigned long) page);
 969         return ret;
 970 }
 971
 972 static const struct file_operations proc_environ_operations = {
 973         .open           = environ_open,
 974         .read           = environ_read,
 975         .llseek         = generic_file_llseek,
 976         .release        = mem_release,
 977 };
 978
 979 static int auxv_open(struct inode *inode, struct file *file)
 980 {
 981         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 982 }
 983
 984 static ssize_t auxv_read(struct file *file, char __user *buf,
 985                         size_t count, loff_t *ppos)
 986 {
 987         struct mm_struct *mm = file->private_data;
 988         unsigned int nwords = 0;
 989
 990         if (!mm)
 991                 return 0;
 992         do {
 993                 nwords += 2;
 994         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 995         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 996                                        nwords * sizeof(mm->saved_auxv[0]));
 997 }
 998
 999 static const struct file_operations proc_auxv_operations = {
1000         .open           = auxv_open,
1001         .read           = auxv_read,
1002         .llseek         = generic_file_llseek,
1003         .release        = mem_release,
1004 };
1005
1006 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1007                             loff_t *ppos)
1008 {
1009         struct task_struct *task = get_proc_task(file_inode(file));
1010         char buffer[PROC_NUMBUF];
1011         int oom_adj = OOM_ADJUST_MIN;
1012         size_t len;
1013
1014         if (!task)
1015                 return -ESRCH;
1016         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1017                 oom_adj = OOM_ADJUST_MAX;
1018         else
1019                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1020                           OOM_SCORE_ADJ_MAX;
1021         put_task_struct(task);
1022         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1023         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1024 }
1025
1026 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1027 {
1028         static DEFINE_MUTEX(oom_adj_mutex);
1029         struct mm_struct *mm = NULL;
1030         struct task_struct *task;
1031         int err = 0;
1032
1033         task = get_proc_task(file_inode(file));
1034         if (!task)
1035                 return -ESRCH;
1036
1037         mutex_lock(&oom_adj_mutex);
1038         if (legacy) {
1039                 if (oom_adj < task->signal->oom_score_adj &&
1040                                 !capable(CAP_SYS_RESOURCE)) {
1041                         err = -EACCES;
1042                         goto err_unlock;
1043                 }
1044                 /*
1045                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1046                  * /proc/pid/oom_score_adj instead.
1047                  */
1048                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1049                           current->comm, task_pid_nr(current), task_pid_nr(task),
1050                           task_pid_nr(task));
1051         } else {
1052                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1053                                 !capable(CAP_SYS_RESOURCE)) {
1054                         err = -EACCES;
1055                         goto err_unlock;
1056                 }
1057         }
1058
1059         /*
1060          * Make sure we will check other processes sharing the mm if this is
1061          * not vfrok which wants its own oom_score_adj.
1062          * pin the mm so it doesn't go away and get reused after task_unlock
1063          */
1064         if (!task->vfork_done) {
1065                 struct task_struct *p = find_lock_task_mm(task);
1066
1067                 if (p) {
1068                         if (atomic_read(&p->mm->mm_users) > 1) {
1069                                 mm = p->mm;
1070                                 mmgrab(mm);
1071                         }
1072                         task_unlock(p);
1073                 }
1074         }
1075
1076         task->signal->oom_score_adj = oom_adj;
1077         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1078                 task->signal->oom_score_adj_min = (short)oom_adj;
1079         trace_oom_score_adj_update(task);
1080
1081         if (mm) {
1082                 struct task_struct *p;
1083
1084                 rcu_read_lock();
1085                 for_each_process(p) {
1086                         if (same_thread_group(task, p))
1087                                 continue;
1088
1089                         /* do not touch kernel threads or the global init */
1090                         if (p->flags & PF_KTHREAD || is_global_init(p))
1091                                 continue;
1092
1093                         task_lock(p);
1094                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1095                                 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",
1096                                                 task_pid_nr(p), p->comm,
1097                                                 p->signal->oom_score_adj, oom_adj,
1098                                                 task_pid_nr(task), task->comm);
1099                                 p->signal->oom_score_adj = oom_adj;
1100                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1101                                         p->signal->oom_score_adj_min = (short)oom_adj;
1102                         }
1103                         task_unlock(p);
1104                 }
1105                 rcu_read_unlock();
1106                 mmdrop(mm);
1107         }
1108 err_unlock:
1109         mutex_unlock(&oom_adj_mutex);
1110         put_task_struct(task);
1111         return err;
1112 }
1113
1114 /*
1115  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1116  * kernels.  The effective policy is defined by oom_score_adj, which has a
1117  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1118  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1119  * Processes that become oom disabled via oom_adj will still be oom disabled
1120  * with this implementation.
1121  *
1122  * oom_adj cannot be removed since existing userspace binaries use it.
1123  */
1124 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1125                              size_t count, loff_t *ppos)
1126 {
1127         char buffer[PROC_NUMBUF];
1128         int oom_adj;
1129         int err;
1130
1131         memset(buffer, 0, sizeof(buffer));
1132         if (count > sizeof(buffer) - 1)
1133                 count = sizeof(buffer) - 1;
1134         if (copy_from_user(buffer, buf, count)) {
1135                 err = -EFAULT;
1136                 goto out;
1137         }
1138
1139         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1140         if (err)
1141                 goto out;
1142         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1143              oom_adj != OOM_DISABLE) {
1144                 err = -EINVAL;
1145                 goto out;
1146         }
1147
1148         /*
1149          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1150          * value is always attainable.
1151          */
1152         if (oom_adj == OOM_ADJUST_MAX)
1153                 oom_adj = OOM_SCORE_ADJ_MAX;
1154         else
1155                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1156
1157         err = __set_oom_adj(file, oom_adj, true);
1158 out:
1159         return err < 0 ? err : count;
1160 }
1161
1162 static const struct file_operations proc_oom_adj_operations = {
1163         .read           = oom_adj_read,
1164         .write          = oom_adj_write,
1165         .llseek         = generic_file_llseek,
1166 };
1167
1168 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1169                                         size_t count, loff_t *ppos)
1170 {
1171         struct task_struct *task = get_proc_task(file_inode(file));
1172         char buffer[PROC_NUMBUF];
1173         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1174         size_t len;
1175
1176         if (!task)
1177                 return -ESRCH;
1178         oom_score_adj = task->signal->oom_score_adj;
1179         put_task_struct(task);
1180         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1181         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1182 }
1183
1184 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1185                                         size_t count, loff_t *ppos)
1186 {
1187         char buffer[PROC_NUMBUF];
1188         int oom_score_adj;
1189         int err;
1190
1191         memset(buffer, 0, sizeof(buffer));
1192         if (count > sizeof(buffer) - 1)
1193                 count = sizeof(buffer) - 1;
1194         if (copy_from_user(buffer, buf, count)) {
1195                 err = -EFAULT;
1196                 goto out;
1197         }
1198
1199         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1200         if (err)
1201                 goto out;
1202         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1203                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1204                 err = -EINVAL;
1205                 goto out;
1206         }
1207
1208         err = __set_oom_adj(file, oom_score_adj, false);
1209 out:
1210         return err < 0 ? err : count;
1211 }
1212
1213 static const struct file_operations proc_oom_score_adj_operations = {
1214         .read           = oom_score_adj_read,
1215         .write          = oom_score_adj_write,
1216         .llseek         = default_llseek,
1217 };
1218
1219 #ifdef CONFIG_AUDITSYSCALL
1220 #define TMPBUFLEN 11
1221 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1222                                   size_t count, loff_t *ppos)
1223 {
1224         struct inode * inode = file_inode(file);
1225         struct task_struct *task = get_proc_task(inode);
1226         ssize_t length;
1227         char tmpbuf[TMPBUFLEN];
1228
1229         if (!task)
1230                 return -ESRCH;
1231         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1232                            from_kuid(file->f_cred->user_ns,
1233                                      audit_get_loginuid(task)));
1234         put_task_struct(task);
1235         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1236 }
1237
1238 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1239                                    size_t count, loff_t *ppos)
1240 {
1241         struct inode * inode = file_inode(file);
1242         uid_t loginuid;
1243         kuid_t kloginuid;
1244         int rv;
1245
1246         rcu_read_lock();
1247         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1248                 rcu_read_unlock();
1249                 return -EPERM;
1250         }
1251         rcu_read_unlock();
1252
1253         if (*ppos != 0) {
1254                 /* No partial writes. */
1255                 return -EINVAL;
1256         }
1257
1258         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1259         if (rv < 0)
1260                 return rv;
1261
1262         /* is userspace tring to explicitly UNSET the loginuid? */
1263         if (loginuid == AUDIT_UID_UNSET) {
1264                 kloginuid = INVALID_UID;
1265         } else {
1266                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1267                 if (!uid_valid(kloginuid))
1268                         return -EINVAL;
1269         }
1270
1271         rv = audit_set_loginuid(kloginuid);
1272         if (rv < 0)
1273                 return rv;
1274         return count;
1275 }
1276
1277 static const struct file_operations proc_loginuid_operations = {
1278         .read           = proc_loginuid_read,
1279         .write          = proc_loginuid_write,
1280         .llseek         = generic_file_llseek,
1281 };
1282
1283 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1284                                   size_t count, loff_t *ppos)
1285 {
1286         struct inode * inode = file_inode(file);
1287         struct task_struct *task = get_proc_task(inode);
1288         ssize_t length;
1289         char tmpbuf[TMPBUFLEN];
1290
1291         if (!task)
1292                 return -ESRCH;
1293         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1294                                 audit_get_sessionid(task));
1295         put_task_struct(task);
1296         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1297 }
1298
1299 static const struct file_operations proc_sessionid_operations = {
1300         .read           = proc_sessionid_read,
1301         .llseek         = generic_file_llseek,
1302 };
1303 #endif
1304
1305 #ifdef CONFIG_FAULT_INJECTION
1306 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1307                                       size_t count, loff_t *ppos)
1308 {
1309         struct task_struct *task = get_proc_task(file_inode(file));
1310         char buffer[PROC_NUMBUF];
1311         size_t len;
1312         int make_it_fail;
1313
1314         if (!task)
1315                 return -ESRCH;
1316         make_it_fail = task->make_it_fail;
1317         put_task_struct(task);
1318
1319         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1320
1321         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1322 }
1323
1324 static ssize_t proc_fault_inject_write(struct file * file,
1325                         const char __user * buf, size_t count, loff_t *ppos)
1326 {
1327         struct task_struct *task;
1328         char buffer[PROC_NUMBUF];
1329         int make_it_fail;
1330         int rv;
1331
1332         if (!capable(CAP_SYS_RESOURCE))
1333                 return -EPERM;
1334         memset(buffer, 0, sizeof(buffer));
1335         if (count > sizeof(buffer) - 1)
1336                 count = sizeof(buffer) - 1;
1337         if (copy_from_user(buffer, buf, count))
1338                 return -EFAULT;
1339         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1340         if (rv < 0)
1341                 return rv;
1342         if (make_it_fail < 0 || make_it_fail > 1)
1343                 return -EINVAL;
1344
1345         task = get_proc_task(file_inode(file));
1346         if (!task)
1347                 return -ESRCH;
1348         task->make_it_fail = make_it_fail;
1349         put_task_struct(task);
1350
1351         return count;
1352 }
1353
1354 static const struct file_operations proc_fault_inject_operations = {
1355         .read           = proc_fault_inject_read,
1356         .write          = proc_fault_inject_write,
1357         .llseek         = generic_file_llseek,
1358 };
1359 #endif
1360
1361
1362 #ifdef CONFIG_SCHED_DEBUG
1363 /*
1364  * Print out various scheduling related per-task fields:
1365  */
1366 static int sched_show(struct seq_file *m, void *v)
1367 {
1368         struct inode *inode = m->private;
1369         struct task_struct *p;
1370
1371         p = get_proc_task(inode);
1372         if (!p)
1373                 return -ESRCH;
1374         proc_sched_show_task(p, m);
1375
1376         put_task_struct(p);
1377
1378         return 0;
1379 }
1380
1381 static ssize_t
1382 sched_write(struct file *file, const char __user *buf,
1383             size_t count, loff_t *offset)
1384 {
1385         struct inode *inode = file_inode(file);
1386         struct task_struct *p;
1387
1388         p = get_proc_task(inode);
1389         if (!p)
1390                 return -ESRCH;
1391         proc_sched_set_task(p);
1392
1393         put_task_struct(p);
1394
1395         return count;
1396 }
1397
1398 static int sched_open(struct inode *inode, struct file *filp)
1399 {
1400         return single_open(filp, sched_show, inode);
1401 }
1402
1403 static const struct file_operations proc_pid_sched_operations = {
1404         .open           = sched_open,
1405         .read           = seq_read,
1406         .write          = sched_write,
1407         .llseek         = seq_lseek,
1408         .release        = single_release,
1409 };
1410
1411 #endif
1412
1413 #ifdef CONFIG_SCHED_AUTOGROUP
1414 /*
1415  * Print out autogroup related information:
1416  */
1417 static int sched_autogroup_show(struct seq_file *m, void *v)
1418 {
1419         struct inode *inode = m->private;
1420         struct task_struct *p;
1421
1422         p = get_proc_task(inode);
1423         if (!p)
1424                 return -ESRCH;
1425         proc_sched_autogroup_show_task(p, m);
1426
1427         put_task_struct(p);
1428
1429         return 0;
1430 }
1431
1432 static ssize_t
1433 sched_autogroup_write(struct file *file, const char __user *buf,
1434             size_t count, loff_t *offset)
1435 {
1436         struct inode *inode = file_inode(file);
1437         struct task_struct *p;
1438         char buffer[PROC_NUMBUF];
1439         int nice;
1440         int err;
1441
1442         memset(buffer, 0, sizeof(buffer));
1443         if (count > sizeof(buffer) - 1)
1444                 count = sizeof(buffer) - 1;
1445         if (copy_from_user(buffer, buf, count))
1446                 return -EFAULT;
1447
1448         err = kstrtoint(strstrip(buffer), 0, &nice);
1449         if (err < 0)
1450                 return err;
1451
1452         p = get_proc_task(inode);
1453         if (!p)
1454                 return -ESRCH;
1455
1456         err = proc_sched_autogroup_set_nice(p, nice);
1457         if (err)
1458                 count = err;
1459
1460         put_task_struct(p);
1461
1462         return count;
1463 }
1464
1465 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1466 {
1467         int ret;
1468
1469         ret = single_open(filp, sched_autogroup_show, NULL);
1470         if (!ret) {
1471                 struct seq_file *m = filp->private_data;
1472
1473                 m->private = inode;
1474         }
1475         return ret;
1476 }
1477
1478 static const struct file_operations proc_pid_sched_autogroup_operations = {
1479         .open           = sched_autogroup_open,
1480         .read           = seq_read,
1481         .write          = sched_autogroup_write,
1482         .llseek         = seq_lseek,
1483         .release        = single_release,
1484 };
1485
1486 #endif /* CONFIG_SCHED_AUTOGROUP */
1487
1488 static ssize_t comm_write(struct file *file, const char __user *buf,
1489                                 size_t count, loff_t *offset)
1490 {
1491         struct inode *inode = file_inode(file);
1492         struct task_struct *p;
1493         char buffer[TASK_COMM_LEN];
1494         const size_t maxlen = sizeof(buffer) - 1;
1495
1496         memset(buffer, 0, sizeof(buffer));
1497         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1498                 return -EFAULT;
1499
1500         p = get_proc_task(inode);
1501         if (!p)
1502                 return -ESRCH;
1503
1504         if (same_thread_group(current, p))
1505                 set_task_comm(p, buffer);
1506         else
1507                 count = -EINVAL;
1508
1509         put_task_struct(p);
1510
1511         return count;
1512 }
1513
1514 static int comm_show(struct seq_file *m, void *v)
1515 {
1516         struct inode *inode = m->private;
1517         struct task_struct *p;
1518
1519         p = get_proc_task(inode);
1520         if (!p)
1521                 return -ESRCH;
1522
1523         task_lock(p);
1524         seq_printf(m, "%s\n", p->comm);
1525         task_unlock(p);
1526
1527         put_task_struct(p);
1528
1529         return 0;
1530 }
1531
1532 static int comm_open(struct inode *inode, struct file *filp)
1533 {
1534         return single_open(filp, comm_show, inode);
1535 }
1536
1537 static const struct file_operations proc_pid_set_comm_operations = {
1538         .open           = comm_open,
1539         .read           = seq_read,
1540         .write          = comm_write,
1541         .llseek         = seq_lseek,
1542         .release        = single_release,
1543 };
1544
1545 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1546 {
1547         struct task_struct *task;
1548         struct file *exe_file;
1549
1550         task = get_proc_task(d_inode(dentry));
1551         if (!task)
1552                 return -ENOENT;
1553         exe_file = get_task_exe_file(task);
1554         put_task_struct(task);
1555         if (exe_file) {
1556                 *exe_path = exe_file->f_path;
1557                 path_get(&exe_file->f_path);
1558                 fput(exe_file);
1559                 return 0;
1560         } else
1561                 return -ENOENT;
1562 }
1563
1564 static const char *proc_pid_get_link(struct dentry *dentry,
1565                                      struct inode *inode,
1566                                      struct delayed_call *done)
1567 {
1568         struct path path;
1569         int error = -EACCES;
1570
1571         if (!dentry)
1572                 return ERR_PTR(-ECHILD);
1573
1574         /* Are we allowed to snoop on the tasks file descriptors? */
1575         if (!proc_fd_access_allowed(inode))
1576                 goto out;
1577
1578         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1579         if (error)
1580                 goto out;
1581
1582         nd_jump_link(&path);
1583         return NULL;
1584 out:
1585         return ERR_PTR(error);
1586 }
1587
1588 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1589 {
1590         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1591         char *pathname;
1592         int len;
1593
1594         if (!tmp)
1595                 return -ENOMEM;
1596
1597         pathname = d_path(path, tmp, PAGE_SIZE);
1598         len = PTR_ERR(pathname);
1599         if (IS_ERR(pathname))
1600                 goto out;
1601         len = tmp + PAGE_SIZE - 1 - pathname;
1602
1603         if (len > buflen)
1604                 len = buflen;
1605         if (copy_to_user(buffer, pathname, len))
1606                 len = -EFAULT;
1607  out:
1608         free_page((unsigned long)tmp);
1609         return len;
1610 }
1611
1612 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1613 {
1614         int error = -EACCES;
1615         struct inode *inode = d_inode(dentry);
1616         struct path path;
1617
1618         /* Are we allowed to snoop on the tasks file descriptors? */
1619         if (!proc_fd_access_allowed(inode))
1620                 goto out;
1621
1622         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1623         if (error)
1624                 goto out;
1625
1626         error = do_proc_readlink(&path, buffer, buflen);
1627         path_put(&path);
1628 out:
1629         return error;
1630 }
1631
1632 const struct inode_operations proc_pid_link_inode_operations = {
1633         .readlink       = proc_pid_readlink,
1634         .get_link       = proc_pid_get_link,
1635         .setattr        = proc_setattr,
1636 };
1637
1638
1639 /* building an inode */
1640
1641 void task_dump_owner(struct task_struct *task, mode_t mode,
1642                      kuid_t *ruid, kgid_t *rgid)
1643 {
1644         /* Depending on the state of dumpable compute who should own a
1645          * proc file for a task.
1646          */
1647         const struct cred *cred;
1648         kuid_t uid;
1649         kgid_t gid;
1650
1651         /* Default to the tasks effective ownership */
1652         rcu_read_lock();
1653         cred = __task_cred(task);
1654         uid = cred->euid;
1655         gid = cred->egid;
1656         rcu_read_unlock();
1657
1658         /*
1659          * Before the /proc/pid/status file was created the only way to read
1660          * the effective uid of a /process was to stat /proc/pid.  Reading
1661          * /proc/pid/status is slow enough that procps and other packages
1662          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1663          * made this apply to all per process world readable and executable
1664          * directories.
1665          */
1666         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1667                 struct mm_struct *mm;
1668                 task_lock(task);
1669                 mm = task->mm;
1670                 /* Make non-dumpable tasks owned by some root */
1671                 if (mm) {
1672                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1673                                 struct user_namespace *user_ns = mm->user_ns;
1674
1675                                 uid = make_kuid(user_ns, 0);
1676                                 if (!uid_valid(uid))
1677                                         uid = GLOBAL_ROOT_UID;
1678
1679                                 gid = make_kgid(user_ns, 0);
1680                                 if (!gid_valid(gid))
1681                                         gid = GLOBAL_ROOT_GID;
1682                         }
1683                 } else {
1684                         uid = GLOBAL_ROOT_UID;
1685                         gid = GLOBAL_ROOT_GID;
1686                 }
1687                 task_unlock(task);
1688         }
1689         *ruid = uid;
1690         *rgid = gid;
1691 }
1692
1693 struct inode *proc_pid_make_inode(struct super_block * sb,
1694                                   struct task_struct *task, umode_t mode)
1695 {
1696         struct inode * inode;
1697         struct proc_inode *ei;
1698
1699         /* We need a new inode */
1700
1701         inode = new_inode(sb);
1702         if (!inode)
1703                 goto out;
1704
1705         /* Common stuff */
1706         ei = PROC_I(inode);
1707         inode->i_mode = mode;
1708         inode->i_ino = get_next_ino();
1709         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1710         inode->i_op = &proc_def_inode_operations;
1711
1712         /*
1713          * grab the reference to task.
1714          */
1715         ei->pid = get_task_pid(task, PIDTYPE_PID);
1716         if (!ei->pid)
1717                 goto out_unlock;
1718
1719         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1720         security_task_to_inode(task, inode);
1721
1722 out:
1723         return inode;
1724
1725 out_unlock:
1726         iput(inode);
1727         return NULL;
1728 }
1729
1730 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1731 {
1732         struct inode *inode = d_inode(dentry);
1733         struct task_struct *task;
1734         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1735
1736         generic_fillattr(inode, stat);
1737
1738         rcu_read_lock();
1739         stat->uid = GLOBAL_ROOT_UID;
1740         stat->gid = GLOBAL_ROOT_GID;
1741         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1742         if (task) {
1743                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1744                         rcu_read_unlock();
1745                         /*
1746                          * This doesn't prevent learning whether PID exists,
1747                          * it only makes getattr() consistent with readdir().
1748                          */
1749                         return -ENOENT;
1750                 }
1751                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1752         }
1753         rcu_read_unlock();
1754         return 0;
1755 }
1756
1757 /* dentry stuff */
1758
1759 /*
1760  *      Exceptional case: normally we are not allowed to unhash a busy
1761  * directory. In this case, however, we can do it - no aliasing problems
1762  * due to the way we treat inodes.
1763  *
1764  * Rewrite the inode's ownerships here because the owning task may have
1765  * performed a setuid(), etc.
1766  *
1767  */
1768 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1769 {
1770         struct inode *inode;
1771         struct task_struct *task;
1772
1773         if (flags & LOOKUP_RCU)
1774                 return -ECHILD;
1775
1776         inode = d_inode(dentry);
1777         task = get_proc_task(inode);
1778
1779         if (task) {
1780                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1781
1782                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1783                 security_task_to_inode(task, inode);
1784                 put_task_struct(task);
1785                 return 1;
1786         }
1787         return 0;
1788 }
1789
1790 static inline bool proc_inode_is_dead(struct inode *inode)
1791 {
1792         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1793 }
1794
1795 int pid_delete_dentry(const struct dentry *dentry)
1796 {
1797         /* Is the task we represent dead?
1798          * If so, then don't put the dentry on the lru list,
1799          * kill it immediately.
1800          */
1801         return proc_inode_is_dead(d_inode(dentry));
1802 }
1803
1804 const struct dentry_operations pid_dentry_operations =
1805 {
1806         .d_revalidate   = pid_revalidate,
1807         .d_delete       = pid_delete_dentry,
1808 };
1809
1810 /* Lookups */
1811
1812 /*
1813  * Fill a directory entry.
1814  *
1815  * If possible create the dcache entry and derive our inode number and
1816  * file type from dcache entry.
1817  *
1818  * Since all of the proc inode numbers are dynamically generated, the inode
1819  * numbers do not exist until the inode is cache.  This means creating the
1820  * the dcache entry in readdir is necessary to keep the inode numbers
1821  * reported by readdir in sync with the inode numbers reported
1822  * by stat.
1823  */
1824 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1825         const char *name, int len,
1826         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1827 {
1828         struct dentry *child, *dir = file->f_path.dentry;
1829         struct qstr qname = QSTR_INIT(name, len);
1830         struct inode *inode;
1831         unsigned type;
1832         ino_t ino;
1833
1834         child = d_hash_and_lookup(dir, &qname);
1835         if (!child) {
1836                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1837                 child = d_alloc_parallel(dir, &qname, &wq);
1838                 if (IS_ERR(child))
1839                         goto end_instantiate;
1840                 if (d_in_lookup(child)) {
1841                         int err = instantiate(d_inode(dir), child, task, ptr);
1842                         d_lookup_done(child);
1843                         if (err < 0) {
1844                                 dput(child);
1845                                 goto end_instantiate;
1846                         }
1847                 }
1848         }
1849         inode = d_inode(child);
1850         ino = inode->i_ino;
1851         type = inode->i_mode >> 12;
1852         dput(child);
1853         return dir_emit(ctx, name, len, ino, type);
1854
1855 end_instantiate:
1856         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1857 }
1858
1859 /*
1860  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1861  * which represent vma start and end addresses.
1862  */
1863 static int dname_to_vma_addr(struct dentry *dentry,
1864                              unsigned long *start, unsigned long *end)
1865 {
1866         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1867                 return -EINVAL;
1868
1869         return 0;
1870 }
1871
1872 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1873 {
1874         unsigned long vm_start, vm_end;
1875         bool exact_vma_exists = false;
1876         struct mm_struct *mm = NULL;
1877         struct task_struct *task;
1878         struct inode *inode;
1879         int status = 0;
1880
1881         if (flags & LOOKUP_RCU)
1882                 return -ECHILD;
1883
1884         inode = d_inode(dentry);
1885         task = get_proc_task(inode);
1886         if (!task)
1887                 goto out_notask;
1888
1889         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1890         if (IS_ERR_OR_NULL(mm))
1891                 goto out;
1892
1893         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1894                 down_read(&mm->mmap_sem);
1895                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1896                 up_read(&mm->mmap_sem);
1897         }
1898
1899         mmput(mm);
1900
1901         if (exact_vma_exists) {
1902                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1903
1904                 security_task_to_inode(task, inode);
1905                 status = 1;
1906         }
1907
1908 out:
1909         put_task_struct(task);
1910
1911 out_notask:
1912         return status;
1913 }
1914
1915 static const struct dentry_operations tid_map_files_dentry_operations = {
1916         .d_revalidate   = map_files_d_revalidate,
1917         .d_delete       = pid_delete_dentry,
1918 };
1919
1920 static int map_files_get_link(struct dentry *dentry, struct path *path)
1921 {
1922         unsigned long vm_start, vm_end;
1923         struct vm_area_struct *vma;
1924         struct task_struct *task;
1925         struct mm_struct *mm;
1926         int rc;
1927
1928         rc = -ENOENT;
1929         task = get_proc_task(d_inode(dentry));
1930         if (!task)
1931                 goto out;
1932
1933         mm = get_task_mm(task);
1934         put_task_struct(task);
1935         if (!mm)
1936                 goto out;
1937
1938         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1939         if (rc)
1940                 goto out_mmput;
1941
1942         rc = -ENOENT;
1943         down_read(&mm->mmap_sem);
1944         vma = find_exact_vma(mm, vm_start, vm_end);
1945         if (vma && vma->vm_file) {
1946                 *path = vma->vm_file->f_path;
1947                 path_get(path);
1948                 rc = 0;
1949         }
1950         up_read(&mm->mmap_sem);
1951
1952 out_mmput:
1953         mmput(mm);
1954 out:
1955         return rc;
1956 }
1957
1958 struct map_files_info {
1959         fmode_t         mode;
1960         unsigned int    len;
1961         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
1962 };
1963
1964 /*
1965  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
1966  * symlinks may be used to bypass permissions on ancestor directories in the
1967  * path to the file in question.
1968  */
1969 static const char *
1970 proc_map_files_get_link(struct dentry *dentry,
1971                         struct inode *inode,
1972                         struct delayed_call *done)
1973 {
1974         if (!capable(CAP_SYS_ADMIN))
1975                 return ERR_PTR(-EPERM);
1976
1977         return proc_pid_get_link(dentry, inode, done);
1978 }
1979
1980 /*
1981  * Identical to proc_pid_link_inode_operations except for get_link()
1982  */
1983 static const struct inode_operations proc_map_files_link_inode_operations = {
1984         .readlink       = proc_pid_readlink,
1985         .get_link       = proc_map_files_get_link,
1986         .setattr        = proc_setattr,
1987 };
1988
1989 static int
1990 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
1991                            struct task_struct *task, const void *ptr)
1992 {
1993         fmode_t mode = (fmode_t)(unsigned long)ptr;
1994         struct proc_inode *ei;
1995         struct inode *inode;
1996
1997         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
1998                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
1999                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2000         if (!inode)
2001                 return -ENOENT;
2002
2003         ei = PROC_I(inode);
2004         ei->op.proc_get_link = map_files_get_link;
2005
2006         inode->i_op = &proc_map_files_link_inode_operations;
2007         inode->i_size = 64;
2008
2009         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2010         d_add(dentry, inode);
2011
2012         return 0;
2013 }
2014
2015 static struct dentry *proc_map_files_lookup(struct inode *dir,
2016                 struct dentry *dentry, unsigned int flags)
2017 {
2018         unsigned long vm_start, vm_end;
2019         struct vm_area_struct *vma;
2020         struct task_struct *task;
2021         int result;
2022         struct mm_struct *mm;
2023
2024         result = -ENOENT;
2025         task = get_proc_task(dir);
2026         if (!task)
2027                 goto out;
2028
2029         result = -EACCES;
2030         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2031                 goto out_put_task;
2032
2033         result = -ENOENT;
2034         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2035                 goto out_put_task;
2036
2037         mm = get_task_mm(task);
2038         if (!mm)
2039                 goto out_put_task;
2040
2041         down_read(&mm->mmap_sem);
2042         vma = find_exact_vma(mm, vm_start, vm_end);
2043         if (!vma)
2044                 goto out_no_vma;
2045
2046         if (vma->vm_file)
2047                 result = proc_map_files_instantiate(dir, dentry, task,
2048                                 (void *)(unsigned long)vma->vm_file->f_mode);
2049
2050 out_no_vma:
2051         up_read(&mm->mmap_sem);
2052         mmput(mm);
2053 out_put_task:
2054         put_task_struct(task);
2055 out:
2056         return ERR_PTR(result);
2057 }
2058
2059 static const struct inode_operations proc_map_files_inode_operations = {
2060         .lookup         = proc_map_files_lookup,
2061         .permission     = proc_fd_permission,
2062         .setattr        = proc_setattr,
2063 };
2064
2065 static int
2066 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2067 {
2068         struct vm_area_struct *vma;
2069         struct task_struct *task;
2070         struct mm_struct *mm;
2071         unsigned long nr_files, pos, i;
2072         struct flex_array *fa = NULL;
2073         struct map_files_info info;
2074         struct map_files_info *p;
2075         int ret;
2076
2077         ret = -ENOENT;
2078         task = get_proc_task(file_inode(file));
2079         if (!task)
2080                 goto out;
2081
2082         ret = -EACCES;
2083         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2084                 goto out_put_task;
2085
2086         ret = 0;
2087         if (!dir_emit_dots(file, ctx))
2088                 goto out_put_task;
2089
2090         mm = get_task_mm(task);
2091         if (!mm)
2092                 goto out_put_task;
2093         down_read(&mm->mmap_sem);
2094
2095         nr_files = 0;
2096
2097         /*
2098          * We need two passes here:
2099          *
2100          *  1) Collect vmas of mapped files with mmap_sem taken
2101          *  2) Release mmap_sem and instantiate entries
2102          *
2103          * otherwise we get lockdep complained, since filldir()
2104          * routine might require mmap_sem taken in might_fault().
2105          */
2106
2107         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2108                 if (vma->vm_file && ++pos > ctx->pos)
2109                         nr_files++;
2110         }
2111
2112         if (nr_files) {
2113                 fa = flex_array_alloc(sizeof(info), nr_files,
2114                                         GFP_KERNEL);
2115                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2116                                                 GFP_KERNEL)) {
2117                         ret = -ENOMEM;
2118                         if (fa)
2119                                 flex_array_free(fa);
2120                         up_read(&mm->mmap_sem);
2121                         mmput(mm);
2122                         goto out_put_task;
2123                 }
2124                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2125                                 vma = vma->vm_next) {
2126                         if (!vma->vm_file)
2127                                 continue;
2128                         if (++pos <= ctx->pos)
2129                                 continue;
2130
2131                         info.mode = vma->vm_file->f_mode;
2132                         info.len = snprintf(info.name,
2133                                         sizeof(info.name), "%lx-%lx",
2134                                         vma->vm_start, vma->vm_end);
2135                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2136                                 BUG();
2137                 }
2138         }
2139         up_read(&mm->mmap_sem);
2140
2141         for (i = 0; i < nr_files; i++) {
2142                 p = flex_array_get(fa, i);
2143                 if (!proc_fill_cache(file, ctx,
2144                                       p->name, p->len,
2145                                       proc_map_files_instantiate,
2146                                       task,
2147                                       (void *)(unsigned long)p->mode))
2148                         break;
2149                 ctx->pos++;
2150         }
2151         if (fa)
2152                 flex_array_free(fa);
2153         mmput(mm);
2154
2155 out_put_task:
2156         put_task_struct(task);
2157 out:
2158         return ret;
2159 }
2160
2161 static const struct file_operations proc_map_files_operations = {
2162         .read           = generic_read_dir,
2163         .iterate_shared = proc_map_files_readdir,
2164         .llseek         = generic_file_llseek,
2165 };
2166
2167 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2168 struct timers_private {
2169         struct pid *pid;
2170         struct task_struct *task;
2171         struct sighand_struct *sighand;
2172         struct pid_namespace *ns;
2173         unsigned long flags;
2174 };
2175
2176 static void *timers_start(struct seq_file *m, loff_t *pos)
2177 {
2178         struct timers_private *tp = m->private;
2179
2180         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2181         if (!tp->task)
2182                 return ERR_PTR(-ESRCH);
2183
2184         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2185         if (!tp->sighand)
2186                 return ERR_PTR(-ESRCH);
2187
2188         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2189 }
2190
2191 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2192 {
2193         struct timers_private *tp = m->private;
2194         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2195 }
2196
2197 static void timers_stop(struct seq_file *m, void *v)
2198 {
2199         struct timers_private *tp = m->private;
2200
2201         if (tp->sighand) {
2202                 unlock_task_sighand(tp->task, &tp->flags);
2203                 tp->sighand = NULL;
2204         }
2205
2206         if (tp->task) {
2207                 put_task_struct(tp->task);
2208                 tp->task = NULL;
2209         }
2210 }
2211
2212 static int show_timer(struct seq_file *m, void *v)
2213 {
2214         struct k_itimer *timer;
2215         struct timers_private *tp = m->private;
2216         int notify;
2217         static const char * const nstr[] = {
2218                 [SIGEV_SIGNAL] = "signal",
2219                 [SIGEV_NONE] = "none",
2220                 [SIGEV_THREAD] = "thread",
2221         };
2222
2223         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2224         notify = timer->it_sigev_notify;
2225
2226         seq_printf(m, "ID: %d\n", timer->it_id);
2227         seq_printf(m, "signal: %d/%p\n",
2228                    timer->sigq->info.si_signo,
2229                    timer->sigq->info.si_value.sival_ptr);
2230         seq_printf(m, "notify: %s/%s.%d\n",
2231                    nstr[notify & ~SIGEV_THREAD_ID],
2232                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2233                    pid_nr_ns(timer->it_pid, tp->ns));
2234         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2235
2236         return 0;
2237 }
2238
2239 static const struct seq_operations proc_timers_seq_ops = {
2240         .start  = timers_start,
2241         .next   = timers_next,
2242         .stop   = timers_stop,
2243         .show   = show_timer,
2244 };
2245
2246 static int proc_timers_open(struct inode *inode, struct file *file)
2247 {
2248         struct timers_private *tp;
2249
2250         tp = __seq_open_private(file, &proc_timers_seq_ops,
2251                         sizeof(struct timers_private));
2252         if (!tp)
2253                 return -ENOMEM;
2254
2255         tp->pid = proc_pid(inode);
2256         tp->ns = inode->i_sb->s_fs_info;
2257         return 0;
2258 }
2259
2260 static const struct file_operations proc_timers_operations = {
2261         .open           = proc_timers_open,
2262         .read           = seq_read,
2263         .llseek         = seq_lseek,
2264         .release        = seq_release_private,
2265 };
2266 #endif
2267
2268 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2269                                         size_t count, loff_t *offset)
2270 {
2271         struct inode *inode = file_inode(file);
2272         struct task_struct *p;
2273         u64 slack_ns;
2274         int err;
2275
2276         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2277         if (err < 0)
2278                 return err;
2279
2280         p = get_proc_task(inode);
2281         if (!p)
2282                 return -ESRCH;
2283
2284         if (p != current) {
2285                 if (!capable(CAP_SYS_NICE)) {
2286                         count = -EPERM;
2287                         goto out;
2288                 }
2289
2290                 err = security_task_setscheduler(p);
2291                 if (err) {
2292                         count = err;
2293                         goto out;
2294                 }
2295         }
2296
2297         task_lock(p);
2298         if (slack_ns == 0)
2299                 p->timer_slack_ns = p->default_timer_slack_ns;
2300         else
2301                 p->timer_slack_ns = slack_ns;
2302         task_unlock(p);
2303
2304 out:
2305         put_task_struct(p);
2306
2307         return count;
2308 }
2309
2310 static int timerslack_ns_show(struct seq_file *m, void *v)
2311 {
2312         struct inode *inode = m->private;
2313         struct task_struct *p;
2314         int err = 0;
2315
2316         p = get_proc_task(inode);
2317         if (!p)
2318                 return -ESRCH;
2319
2320         if (p != current) {
2321
2322                 if (!capable(CAP_SYS_NICE)) {
2323                         err = -EPERM;
2324                         goto out;
2325                 }
2326                 err = security_task_getscheduler(p);
2327                 if (err)
2328                         goto out;
2329         }
2330
2331         task_lock(p);
2332         seq_printf(m, "%llu\n", p->timer_slack_ns);
2333         task_unlock(p);
2334
2335 out:
2336         put_task_struct(p);
2337
2338         return err;
2339 }
2340
2341 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2342 {
2343         return single_open(filp, timerslack_ns_show, inode);
2344 }
2345
2346 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2347         .open           = timerslack_ns_open,
2348         .read           = seq_read,
2349         .write          = timerslack_ns_write,
2350         .llseek         = seq_lseek,
2351         .release        = single_release,
2352 };
2353
2354 static int proc_pident_instantiate(struct inode *dir,
2355         struct dentry *dentry, struct task_struct *task, const void *ptr)
2356 {
2357         const struct pid_entry *p = ptr;
2358         struct inode *inode;
2359         struct proc_inode *ei;
2360
2361         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2362         if (!inode)
2363                 goto out;
2364
2365         ei = PROC_I(inode);
2366         if (S_ISDIR(inode->i_mode))
2367                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2368         if (p->iop)
2369                 inode->i_op = p->iop;
2370         if (p->fop)
2371                 inode->i_fop = p->fop;
2372         ei->op = p->op;
2373         d_set_d_op(dentry, &pid_dentry_operations);
2374         d_add(dentry, inode);
2375         /* Close the race of the process dying before we return the dentry */
2376         if (pid_revalidate(dentry, 0))
2377                 return 0;
2378 out:
2379         return -ENOENT;
2380 }
2381
2382 static struct dentry *proc_pident_lookup(struct inode *dir,
2383                                          struct dentry *dentry,
2384                                          const struct pid_entry *ents,
2385                                          unsigned int nents)
2386 {
2387         int error;
2388         struct task_struct *task = get_proc_task(dir);
2389         const struct pid_entry *p, *last;
2390
2391         error = -ENOENT;
2392
2393         if (!task)
2394                 goto out_no_task;
2395
2396         /*
2397          * Yes, it does not scale. And it should not. Don't add
2398          * new entries into /proc/<tgid>/ without very good reasons.
2399          */
2400         last = &ents[nents];
2401         for (p = ents; p < last; p++) {
2402                 if (p->len != dentry->d_name.len)
2403                         continue;
2404                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2405                         break;
2406         }
2407         if (p >= last)
2408                 goto out;
2409
2410         error = proc_pident_instantiate(dir, dentry, task, p);
2411 out:
2412         put_task_struct(task);
2413 out_no_task:
2414         return ERR_PTR(error);
2415 }
2416
2417 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2418                 const struct pid_entry *ents, unsigned int nents)
2419 {
2420         struct task_struct *task = get_proc_task(file_inode(file));
2421         const struct pid_entry *p;
2422
2423         if (!task)
2424                 return -ENOENT;
2425
2426         if (!dir_emit_dots(file, ctx))
2427                 goto out;
2428
2429         if (ctx->pos >= nents + 2)
2430                 goto out;
2431
2432         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2433                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2434                                 proc_pident_instantiate, task, p))
2435                         break;
2436                 ctx->pos++;
2437         }
2438 out:
2439         put_task_struct(task);
2440         return 0;
2441 }
2442
2443 #ifdef CONFIG_SECURITY
2444 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2445                                   size_t count, loff_t *ppos)
2446 {
2447         struct inode * inode = file_inode(file);
2448         char *p = NULL;
2449         ssize_t length;
2450         struct task_struct *task = get_proc_task(inode);
2451
2452         if (!task)
2453                 return -ESRCH;
2454
2455         length = security_getprocattr(task,
2456                                       (char*)file->f_path.dentry->d_name.name,
2457                                       &p);
2458         put_task_struct(task);
2459         if (length > 0)
2460                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2461         kfree(p);
2462         return length;
2463 }
2464
2465 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2466                                    size_t count, loff_t *ppos)
2467 {
2468         struct inode * inode = file_inode(file);
2469         void *page;
2470         ssize_t length;
2471         struct task_struct *task = get_proc_task(inode);
2472
2473         length = -ESRCH;
2474         if (!task)
2475                 goto out_no_task;
2476
2477         /* A task may only write its own attributes. */
2478         length = -EACCES;
2479         if (current != task)
2480                 goto out;
2481
2482         if (count > PAGE_SIZE)
2483                 count = PAGE_SIZE;
2484
2485         /* No partial writes. */
2486         length = -EINVAL;
2487         if (*ppos != 0)
2488                 goto out;
2489
2490         page = memdup_user(buf, count);
2491         if (IS_ERR(page)) {
2492                 length = PTR_ERR(page);
2493                 goto out;
2494         }
2495
2496         /* Guard against adverse ptrace interaction */
2497         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2498         if (length < 0)
2499                 goto out_free;
2500
2501         length = security_setprocattr(file->f_path.dentry->d_name.name,
2502                                       page, count);
2503         mutex_unlock(&current->signal->cred_guard_mutex);
2504 out_free:
2505         kfree(page);
2506 out:
2507         put_task_struct(task);
2508 out_no_task:
2509         return length;
2510 }
2511
2512 static const struct file_operations proc_pid_attr_operations = {
2513         .read           = proc_pid_attr_read,
2514         .write          = proc_pid_attr_write,
2515         .llseek         = generic_file_llseek,
2516 };
2517
2518 static const struct pid_entry attr_dir_stuff[] = {
2519         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2520         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2521         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2522         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2523         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2524         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2525 };
2526
2527 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2528 {
2529         return proc_pident_readdir(file, ctx,
2530                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2531 }
2532
2533 static const struct file_operations proc_attr_dir_operations = {
2534         .read           = generic_read_dir,
2535         .iterate_shared = proc_attr_dir_readdir,
2536         .llseek         = generic_file_llseek,
2537 };
2538
2539 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2540                                 struct dentry *dentry, unsigned int flags)
2541 {
2542         return proc_pident_lookup(dir, dentry,
2543                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2544 }
2545
2546 static const struct inode_operations proc_attr_dir_inode_operations = {
2547         .lookup         = proc_attr_dir_lookup,
2548         .getattr        = pid_getattr,
2549         .setattr        = proc_setattr,
2550 };
2551
2552 #endif
2553
2554 #ifdef CONFIG_ELF_CORE
2555 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2556                                          size_t count, loff_t *ppos)
2557 {
2558         struct task_struct *task = get_proc_task(file_inode(file));
2559         struct mm_struct *mm;
2560         char buffer[PROC_NUMBUF];
2561         size_t len;
2562         int ret;
2563
2564         if (!task)
2565                 return -ESRCH;
2566
2567         ret = 0;
2568         mm = get_task_mm(task);
2569         if (mm) {
2570                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2571                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2572                                 MMF_DUMP_FILTER_SHIFT));
2573                 mmput(mm);
2574                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2575         }
2576
2577         put_task_struct(task);
2578
2579         return ret;
2580 }
2581
2582 static ssize_t proc_coredump_filter_write(struct file *file,
2583                                           const char __user *buf,
2584                                           size_t count,
2585                                           loff_t *ppos)
2586 {
2587         struct task_struct *task;
2588         struct mm_struct *mm;
2589         unsigned int val;
2590         int ret;
2591         int i;
2592         unsigned long mask;
2593
2594         ret = kstrtouint_from_user(buf, count, 0, &val);
2595         if (ret < 0)
2596                 return ret;
2597
2598         ret = -ESRCH;
2599         task = get_proc_task(file_inode(file));
2600         if (!task)
2601                 goto out_no_task;
2602
2603         mm = get_task_mm(task);
2604         if (!mm)
2605                 goto out_no_mm;
2606         ret = 0;
2607
2608         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2609                 if (val & mask)
2610                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2611                 else
2612                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2613         }
2614
2615         mmput(mm);
2616  out_no_mm:
2617         put_task_struct(task);
2618  out_no_task:
2619         if (ret < 0)
2620                 return ret;
2621         return count;
2622 }
2623
2624 static const struct file_operations proc_coredump_filter_operations = {
2625         .read           = proc_coredump_filter_read,
2626         .write          = proc_coredump_filter_write,
2627         .llseek         = generic_file_llseek,
2628 };
2629 #endif
2630
2631 #ifdef CONFIG_TASK_IO_ACCOUNTING
2632 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2633 {
2634         struct task_io_accounting acct = task->ioac;
2635         unsigned long flags;
2636         int result;
2637
2638         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2639         if (result)
2640                 return result;
2641
2642         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2643                 result = -EACCES;
2644                 goto out_unlock;
2645         }
2646
2647         if (whole && lock_task_sighand(task, &flags)) {
2648                 struct task_struct *t = task;
2649
2650                 task_io_accounting_add(&acct, &task->signal->ioac);
2651                 while_each_thread(task, t)
2652                         task_io_accounting_add(&acct, &t->ioac);
2653
2654                 unlock_task_sighand(task, &flags);
2655         }
2656         seq_printf(m,
2657                    "rchar: %llu\n"
2658                    "wchar: %llu\n"
2659                    "syscr: %llu\n"
2660                    "syscw: %llu\n"
2661                    "read_bytes: %llu\n"
2662                    "write_bytes: %llu\n"
2663                    "cancelled_write_bytes: %llu\n",
2664                    (unsigned long long)acct.rchar,
2665                    (unsigned long long)acct.wchar,
2666                    (unsigned long long)acct.syscr,
2667                    (unsigned long long)acct.syscw,
2668                    (unsigned long long)acct.read_bytes,
2669                    (unsigned long long)acct.write_bytes,
2670                    (unsigned long long)acct.cancelled_write_bytes);
2671         result = 0;
2672
2673 out_unlock:
2674         mutex_unlock(&task->signal->cred_guard_mutex);
2675         return result;
2676 }
2677
2678 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2679                                   struct pid *pid, struct task_struct *task)
2680 {
2681         return do_io_accounting(task, m, 0);
2682 }
2683
2684 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2685                                    struct pid *pid, struct task_struct *task)
2686 {
2687         return do_io_accounting(task, m, 1);
2688 }
2689 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2690
2691 #ifdef CONFIG_USER_NS
2692 static int proc_id_map_open(struct inode *inode, struct file *file,
2693         const struct seq_operations *seq_ops)
2694 {
2695         struct user_namespace *ns = NULL;
2696         struct task_struct *task;
2697         struct seq_file *seq;
2698         int ret = -EINVAL;
2699
2700         task = get_proc_task(inode);
2701         if (task) {
2702                 rcu_read_lock();
2703                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2704                 rcu_read_unlock();
2705                 put_task_struct(task);
2706         }
2707         if (!ns)
2708                 goto err;
2709
2710         ret = seq_open(file, seq_ops);
2711         if (ret)
2712                 goto err_put_ns;
2713
2714         seq = file->private_data;
2715         seq->private = ns;
2716
2717         return 0;
2718 err_put_ns:
2719         put_user_ns(ns);
2720 err:
2721         return ret;
2722 }
2723
2724 static int proc_id_map_release(struct inode *inode, struct file *file)
2725 {
2726         struct seq_file *seq = file->private_data;
2727         struct user_namespace *ns = seq->private;
2728         put_user_ns(ns);
2729         return seq_release(inode, file);
2730 }
2731
2732 static int proc_uid_map_open(struct inode *inode, struct file *file)
2733 {
2734         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2735 }
2736
2737 static int proc_gid_map_open(struct inode *inode, struct file *file)
2738 {
2739         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2740 }
2741
2742 static int proc_projid_map_open(struct inode *inode, struct file *file)
2743 {
2744         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2745 }
2746
2747 static const struct file_operations proc_uid_map_operations = {
2748         .open           = proc_uid_map_open,
2749         .write          = proc_uid_map_write,
2750         .read           = seq_read,
2751         .llseek         = seq_lseek,
2752         .release        = proc_id_map_release,
2753 };
2754
2755 static const struct file_operations proc_gid_map_operations = {
2756         .open           = proc_gid_map_open,
2757         .write          = proc_gid_map_write,
2758         .read           = seq_read,
2759         .llseek         = seq_lseek,
2760         .release        = proc_id_map_release,
2761 };
2762
2763 static const struct file_operations proc_projid_map_operations = {
2764         .open           = proc_projid_map_open,
2765         .write          = proc_projid_map_write,
2766         .read           = seq_read,
2767         .llseek         = seq_lseek,
2768         .release        = proc_id_map_release,
2769 };
2770
2771 static int proc_setgroups_open(struct inode *inode, struct file *file)
2772 {
2773         struct user_namespace *ns = NULL;
2774         struct task_struct *task;
2775         int ret;
2776
2777         ret = -ESRCH;
2778         task = get_proc_task(inode);
2779         if (task) {
2780                 rcu_read_lock();
2781                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2782                 rcu_read_unlock();
2783                 put_task_struct(task);
2784         }
2785         if (!ns)
2786                 goto err;
2787
2788         if (file->f_mode & FMODE_WRITE) {
2789                 ret = -EACCES;
2790                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2791                         goto err_put_ns;
2792         }
2793
2794         ret = single_open(file, &proc_setgroups_show, ns);
2795         if (ret)
2796                 goto err_put_ns;
2797
2798         return 0;
2799 err_put_ns:
2800         put_user_ns(ns);
2801 err:
2802         return ret;
2803 }
2804
2805 static int proc_setgroups_release(struct inode *inode, struct file *file)
2806 {
2807         struct seq_file *seq = file->private_data;
2808         struct user_namespace *ns = seq->private;
2809         int ret = single_release(inode, file);
2810         put_user_ns(ns);
2811         return ret;
2812 }
2813
2814 static const struct file_operations proc_setgroups_operations = {
2815         .open           = proc_setgroups_open,
2816         .write          = proc_setgroups_write,
2817         .read           = seq_read,
2818         .llseek         = seq_lseek,
2819         .release        = proc_setgroups_release,
2820 };
2821 #endif /* CONFIG_USER_NS */
2822
2823 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2824                                 struct pid *pid, struct task_struct *task)
2825 {
2826         int err = lock_trace(task);
2827         if (!err) {
2828                 seq_printf(m, "%08x\n", task->personality);
2829                 unlock_trace(task);
2830         }
2831         return err;
2832 }
2833
2834 /*
2835  * Thread groups
2836  */
2837 static const struct file_operations proc_task_operations;
2838 static const struct inode_operations proc_task_inode_operations;
2839
2840 static const struct pid_entry tgid_base_stuff[] = {
2841         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2842         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2843         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2844         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2845         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2846 #ifdef CONFIG_NET
2847         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2848 #endif
2849         REG("environ",    S_IRUSR, proc_environ_operations),
2850         REG("auxv",       S_IRUSR, proc_auxv_operations),
2851         ONE("status",     S_IRUGO, proc_pid_status),
2852         ONE("personality", S_IRUSR, proc_pid_personality),
2853         ONE("limits",     S_IRUGO, proc_pid_limits),
2854 #ifdef CONFIG_SCHED_DEBUG
2855         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2856 #endif
2857 #ifdef CONFIG_SCHED_AUTOGROUP
2858         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2859 #endif
2860         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2861 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2862         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2863 #endif
2864         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2865         ONE("stat",       S_IRUGO, proc_tgid_stat),
2866         ONE("statm",      S_IRUGO, proc_pid_statm),
2867         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2868 #ifdef CONFIG_NUMA
2869         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2870 #endif
2871         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2872         LNK("cwd",        proc_cwd_link),
2873         LNK("root",       proc_root_link),
2874         LNK("exe",        proc_exe_link),
2875         REG("mounts",     S_IRUGO, proc_mounts_operations),
2876         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2877         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2878 #ifdef CONFIG_PROC_PAGE_MONITOR
2879         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2880         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2881         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2882 #endif
2883 #ifdef CONFIG_SECURITY
2884         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2885 #endif
2886 #ifdef CONFIG_KALLSYMS
2887         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2888 #endif
2889 #ifdef CONFIG_STACKTRACE
2890         ONE("stack",      S_IRUSR, proc_pid_stack),
2891 #endif
2892 #ifdef CONFIG_SCHED_INFO
2893         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2894 #endif
2895 #ifdef CONFIG_LATENCYTOP
2896         REG("latency",  S_IRUGO, proc_lstats_operations),
2897 #endif
2898 #ifdef CONFIG_PROC_PID_CPUSET
2899         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2900 #endif
2901 #ifdef CONFIG_CGROUPS
2902         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2903 #endif
2904         ONE("oom_score",  S_IRUGO, proc_oom_score),
2905         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2906         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2907 #ifdef CONFIG_AUDITSYSCALL
2908         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2909         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2910 #endif
2911 #ifdef CONFIG_FAULT_INJECTION
2912         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2913 #endif
2914 #ifdef CONFIG_ELF_CORE
2915         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2916 #endif
2917 #ifdef CONFIG_TASK_IO_ACCOUNTING
2918         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2919 #endif
2920 #ifdef CONFIG_HARDWALL
2921         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2922 #endif
2923 #ifdef CONFIG_USER_NS
2924         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2925         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2926         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2927         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2928 #endif
2929 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2930         REG("timers",     S_IRUGO, proc_timers_operations),
2931 #endif
2932         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2933 };
2934
2935 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2936 {
2937         return proc_pident_readdir(file, ctx,
2938                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2939 }
2940
2941 static const struct file_operations proc_tgid_base_operations = {
2942         .read           = generic_read_dir,
2943         .iterate_shared = proc_tgid_base_readdir,
2944         .llseek         = generic_file_llseek,
2945 };
2946
2947 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
2948 {
2949         return proc_pident_lookup(dir, dentry,
2950                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2951 }
2952
2953 static const struct inode_operations proc_tgid_base_inode_operations = {
2954         .lookup         = proc_tgid_base_lookup,
2955         .getattr        = pid_getattr,
2956         .setattr        = proc_setattr,
2957         .permission     = proc_pid_permission,
2958 };
2959
2960 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2961 {
2962         struct dentry *dentry, *leader, *dir;
2963         char buf[PROC_NUMBUF];
2964         struct qstr name;
2965
2966         name.name = buf;
2967         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2968         /* no ->d_hash() rejects on procfs */
2969         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2970         if (dentry) {
2971                 d_invalidate(dentry);
2972                 dput(dentry);
2973         }
2974
2975         if (pid == tgid)
2976                 return;
2977
2978         name.name = buf;
2979         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2980         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2981         if (!leader)
2982                 goto out;
2983
2984         name.name = "task";
2985         name.len = strlen(name.name);
2986         dir = d_hash_and_lookup(leader, &name);
2987         if (!dir)
2988                 goto out_put_leader;
2989
2990         name.name = buf;
2991         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2992         dentry = d_hash_and_lookup(dir, &name);
2993         if (dentry) {
2994                 d_invalidate(dentry);
2995                 dput(dentry);
2996         }
2997
2998         dput(dir);
2999 out_put_leader:
3000         dput(leader);
3001 out:
3002         return;
3003 }
3004
3005 /**
3006  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3007  * @task: task that should be flushed.
3008  *
3009  * When flushing dentries from proc, one needs to flush them from global
3010  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3011  * in. This call is supposed to do all of this job.
3012  *
3013  * Looks in the dcache for
3014  * /proc/@pid
3015  * /proc/@tgid/task/@pid
3016  * if either directory is present flushes it and all of it'ts children
3017  * from the dcache.
3018  *
3019  * It is safe and reasonable to cache /proc entries for a task until
3020  * that task exits.  After that they just clog up the dcache with
3021  * useless entries, possibly causing useful dcache entries to be
3022  * flushed instead.  This routine is proved to flush those useless
3023  * dcache entries at process exit time.
3024  *
3025  * NOTE: This routine is just an optimization so it does not guarantee
3026  *       that no dcache entries will exist at process exit time it
3027  *       just makes it very unlikely that any will persist.
3028  */
3029
3030 void proc_flush_task(struct task_struct *task)
3031 {
3032         int i;
3033         struct pid *pid, *tgid;
3034         struct upid *upid;
3035
3036         pid = task_pid(task);
3037         tgid = task_tgid(task);
3038
3039         for (i = 0; i <= pid->level; i++) {
3040                 upid = &pid->numbers[i];
3041                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3042                                         tgid->numbers[i].nr);
3043         }
3044 }
3045
3046 static int proc_pid_instantiate(struct inode *dir,
3047                                    struct dentry * dentry,
3048                                    struct task_struct *task, const void *ptr)
3049 {
3050         struct inode *inode;
3051
3052         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3053         if (!inode)
3054                 goto out;
3055
3056         inode->i_op = &proc_tgid_base_inode_operations;
3057         inode->i_fop = &proc_tgid_base_operations;
3058         inode->i_flags|=S_IMMUTABLE;
3059
3060         set_nlink(inode, nlink_tgid);
3061
3062         d_set_d_op(dentry, &pid_dentry_operations);
3063
3064         d_add(dentry, inode);
3065         /* Close the race of the process dying before we return the dentry */
3066         if (pid_revalidate(dentry, 0))
3067                 return 0;
3068 out:
3069         return -ENOENT;
3070 }
3071
3072 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3073 {
3074         int result = -ENOENT;
3075         struct task_struct *task;
3076         unsigned tgid;
3077         struct pid_namespace *ns;
3078
3079         tgid = name_to_int(&dentry->d_name);
3080         if (tgid == ~0U)
3081                 goto out;
3082
3083         ns = dentry->d_sb->s_fs_info;
3084         rcu_read_lock();
3085         task = find_task_by_pid_ns(tgid, ns);
3086         if (task)
3087                 get_task_struct(task);
3088         rcu_read_unlock();
3089         if (!task)
3090                 goto out;
3091
3092         result = proc_pid_instantiate(dir, dentry, task, NULL);
3093         put_task_struct(task);
3094 out:
3095         return ERR_PTR(result);
3096 }
3097
3098 /*
3099  * Find the first task with tgid >= tgid
3100  *
3101  */
3102 struct tgid_iter {
3103         unsigned int tgid;
3104         struct task_struct *task;
3105 };
3106 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3107 {
3108         struct pid *pid;
3109
3110         if (iter.task)
3111                 put_task_struct(iter.task);
3112         rcu_read_lock();
3113 retry:
3114         iter.task = NULL;
3115         pid = find_ge_pid(iter.tgid, ns);
3116         if (pid) {
3117                 iter.tgid = pid_nr_ns(pid, ns);
3118                 iter.task = pid_task(pid, PIDTYPE_PID);
3119                 /* What we to know is if the pid we have find is the
3120                  * pid of a thread_group_leader.  Testing for task
3121                  * being a thread_group_leader is the obvious thing
3122                  * todo but there is a window when it fails, due to
3123                  * the pid transfer logic in de_thread.
3124                  *
3125                  * So we perform the straight forward test of seeing
3126                  * if the pid we have found is the pid of a thread
3127                  * group leader, and don't worry if the task we have
3128                  * found doesn't happen to be a thread group leader.
3129                  * As we don't care in the case of readdir.
3130                  */
3131                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3132                         iter.tgid += 1;
3133                         goto retry;
3134                 }
3135                 get_task_struct(iter.task);
3136         }
3137         rcu_read_unlock();
3138         return iter;
3139 }
3140
3141 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3142
3143 /* for the /proc/ directory itself, after non-process stuff has been done */
3144 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3145 {
3146         struct tgid_iter iter;
3147         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3148         loff_t pos = ctx->pos;
3149
3150         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3151                 return 0;
3152
3153         if (pos == TGID_OFFSET - 2) {
3154                 struct inode *inode = d_inode(ns->proc_self);
3155                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3156                         return 0;
3157                 ctx->pos = pos = pos + 1;
3158         }
3159         if (pos == TGID_OFFSET - 1) {
3160                 struct inode *inode = d_inode(ns->proc_thread_self);
3161                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3162                         return 0;
3163                 ctx->pos = pos = pos + 1;
3164         }
3165         iter.tgid = pos - TGID_OFFSET;
3166         iter.task = NULL;
3167         for (iter = next_tgid(ns, iter);
3168              iter.task;
3169              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3170                 char name[PROC_NUMBUF];
3171                 int len;
3172
3173                 cond_resched();
3174                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3175                         continue;
3176
3177                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3178                 ctx->pos = iter.tgid + TGID_OFFSET;
3179                 if (!proc_fill_cache(file, ctx, name, len,
3180                                      proc_pid_instantiate, iter.task, NULL)) {
3181                         put_task_struct(iter.task);
3182                         return 0;
3183                 }
3184         }
3185         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3186         return 0;
3187 }
3188
3189 /*
3190  * proc_tid_comm_permission is a special permission function exclusively
3191  * used for the node /proc/<pid>/task/<tid>/comm.
3192  * It bypasses generic permission checks in the case where a task of the same
3193  * task group attempts to access the node.
3194  * The rationale behind this is that glibc and bionic access this node for
3195  * cross thread naming (pthread_set/getname_np(!self)). However, if
3196  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3197  * which locks out the cross thread naming implementation.
3198  * This function makes sure that the node is always accessible for members of
3199  * same thread group.
3200  */
3201 static int proc_tid_comm_permission(struct inode *inode, int mask)
3202 {
3203         bool is_same_tgroup;
3204         struct task_struct *task;
3205
3206         task = get_proc_task(inode);
3207         if (!task)
3208                 return -ESRCH;
3209         is_same_tgroup = same_thread_group(current, task);
3210         put_task_struct(task);
3211
3212         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3213                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3214                  * read or written by the members of the corresponding
3215                  * thread group.
3216                  */
3217                 return 0;
3218         }
3219
3220         return generic_permission(inode, mask);
3221 }
3222
3223 static const struct inode_operations proc_tid_comm_inode_operations = {
3224                 .permission = proc_tid_comm_permission,
3225 };
3226
3227 /*
3228  * Tasks
3229  */
3230 static const struct pid_entry tid_base_stuff[] = {
3231         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3232         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3233         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3234 #ifdef CONFIG_NET
3235         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3236 #endif
3237         REG("environ",   S_IRUSR, proc_environ_operations),
3238         REG("auxv",      S_IRUSR, proc_auxv_operations),
3239         ONE("status",    S_IRUGO, proc_pid_status),
3240         ONE("personality", S_IRUSR, proc_pid_personality),
3241         ONE("limits",    S_IRUGO, proc_pid_limits),
3242 #ifdef CONFIG_SCHED_DEBUG
3243         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3244 #endif
3245         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3246                          &proc_tid_comm_inode_operations,
3247                          &proc_pid_set_comm_operations, {}),
3248 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3249         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3250 #endif
3251         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3252         ONE("stat",      S_IRUGO, proc_tid_stat),
3253         ONE("statm",     S_IRUGO, proc_pid_statm),
3254         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3255 #ifdef CONFIG_PROC_CHILDREN
3256         REG("children",  S_IRUGO, proc_tid_children_operations),
3257 #endif
3258 #ifdef CONFIG_NUMA
3259         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3260 #endif
3261         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3262         LNK("cwd",       proc_cwd_link),
3263         LNK("root",      proc_root_link),
3264         LNK("exe",       proc_exe_link),
3265         REG("mounts",    S_IRUGO, proc_mounts_operations),
3266         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3267 #ifdef CONFIG_PROC_PAGE_MONITOR
3268         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3269         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3270         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3271 #endif
3272 #ifdef CONFIG_SECURITY
3273         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3274 #endif
3275 #ifdef CONFIG_KALLSYMS
3276         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3277 #endif
3278 #ifdef CONFIG_STACKTRACE
3279         ONE("stack",      S_IRUSR, proc_pid_stack),
3280 #endif
3281 #ifdef CONFIG_SCHED_INFO
3282         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3283 #endif
3284 #ifdef CONFIG_LATENCYTOP
3285         REG("latency",  S_IRUGO, proc_lstats_operations),
3286 #endif
3287 #ifdef CONFIG_PROC_PID_CPUSET
3288         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3289 #endif
3290 #ifdef CONFIG_CGROUPS
3291         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3292 #endif
3293         ONE("oom_score", S_IRUGO, proc_oom_score),
3294         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3295         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3296 #ifdef CONFIG_AUDITSYSCALL
3297         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3298         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3299 #endif
3300 #ifdef CONFIG_FAULT_INJECTION
3301         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3302 #endif
3303 #ifdef CONFIG_TASK_IO_ACCOUNTING
3304         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3305 #endif
3306 #ifdef CONFIG_HARDWALL
3307         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3308 #endif
3309 #ifdef CONFIG_USER_NS
3310         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3311         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3312         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3313         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3314 #endif
3315 };
3316
3317 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3318 {
3319         return proc_pident_readdir(file, ctx,
3320                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3321 }
3322
3323 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3324 {
3325         return proc_pident_lookup(dir, dentry,
3326                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3327 }
3328
3329 static const struct file_operations proc_tid_base_operations = {
3330         .read           = generic_read_dir,
3331         .iterate_shared = proc_tid_base_readdir,
3332         .llseek         = generic_file_llseek,
3333 };
3334
3335 static const struct inode_operations proc_tid_base_inode_operations = {
3336         .lookup         = proc_tid_base_lookup,
3337         .getattr        = pid_getattr,
3338         .setattr        = proc_setattr,
3339 };
3340
3341 static int proc_task_instantiate(struct inode *dir,
3342         struct dentry *dentry, struct task_struct *task, const void *ptr)
3343 {
3344         struct inode *inode;
3345         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3346
3347         if (!inode)
3348                 goto out;
3349         inode->i_op = &proc_tid_base_inode_operations;
3350         inode->i_fop = &proc_tid_base_operations;
3351         inode->i_flags|=S_IMMUTABLE;
3352
3353         set_nlink(inode, nlink_tid);
3354
3355         d_set_d_op(dentry, &pid_dentry_operations);
3356
3357         d_add(dentry, inode);
3358         /* Close the race of the process dying before we return the dentry */
3359         if (pid_revalidate(dentry, 0))
3360                 return 0;
3361 out:
3362         return -ENOENT;
3363 }
3364
3365 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3366 {
3367         int result = -ENOENT;
3368         struct task_struct *task;
3369         struct task_struct *leader = get_proc_task(dir);
3370         unsigned tid;
3371         struct pid_namespace *ns;
3372
3373         if (!leader)
3374                 goto out_no_task;
3375
3376         tid = name_to_int(&dentry->d_name);
3377         if (tid == ~0U)
3378                 goto out;
3379
3380         ns = dentry->d_sb->s_fs_info;
3381         rcu_read_lock();
3382         task = find_task_by_pid_ns(tid, ns);
3383         if (task)
3384                 get_task_struct(task);
3385         rcu_read_unlock();
3386         if (!task)
3387                 goto out;
3388         if (!same_thread_group(leader, task))
3389                 goto out_drop_task;
3390
3391         result = proc_task_instantiate(dir, dentry, task, NULL);
3392 out_drop_task:
3393         put_task_struct(task);
3394 out:
3395         put_task_struct(leader);
3396 out_no_task:
3397         return ERR_PTR(result);
3398 }
3399
3400 /*
3401  * Find the first tid of a thread group to return to user space.
3402  *
3403  * Usually this is just the thread group leader, but if the users
3404  * buffer was too small or there was a seek into the middle of the
3405  * directory we have more work todo.
3406  *
3407  * In the case of a short read we start with find_task_by_pid.
3408  *
3409  * In the case of a seek we start with the leader and walk nr
3410  * threads past it.
3411  */
3412 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3413                                         struct pid_namespace *ns)
3414 {
3415         struct task_struct *pos, *task;
3416         unsigned long nr = f_pos;
3417
3418         if (nr != f_pos)        /* 32bit overflow? */
3419                 return NULL;
3420
3421         rcu_read_lock();
3422         task = pid_task(pid, PIDTYPE_PID);
3423         if (!task)
3424                 goto fail;
3425
3426         /* Attempt to start with the tid of a thread */
3427         if (tid && nr) {
3428                 pos = find_task_by_pid_ns(tid, ns);
3429                 if (pos && same_thread_group(pos, task))
3430                         goto found;
3431         }
3432
3433         /* If nr exceeds the number of threads there is nothing todo */
3434         if (nr >= get_nr_threads(task))
3435                 goto fail;
3436
3437         /* If we haven't found our starting place yet start
3438          * with the leader and walk nr threads forward.
3439          */
3440         pos = task = task->group_leader;
3441         do {
3442                 if (!nr--)
3443                         goto found;
3444         } while_each_thread(task, pos);
3445 fail:
3446         pos = NULL;
3447         goto out;
3448 found:
3449         get_task_struct(pos);
3450 out:
3451         rcu_read_unlock();
3452         return pos;
3453 }
3454
3455 /*
3456  * Find the next thread in the thread list.
3457  * Return NULL if there is an error or no next thread.
3458  *
3459  * The reference to the input task_struct is released.
3460  */
3461 static struct task_struct *next_tid(struct task_struct *start)
3462 {
3463         struct task_struct *pos = NULL;
3464         rcu_read_lock();
3465         if (pid_alive(start)) {
3466                 pos = next_thread(start);
3467                 if (thread_group_leader(pos))
3468                         pos = NULL;
3469                 else
3470                         get_task_struct(pos);
3471         }
3472         rcu_read_unlock();
3473         put_task_struct(start);
3474         return pos;
3475 }
3476
3477 /* for the /proc/TGID/task/ directories */
3478 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3479 {
3480         struct inode *inode = file_inode(file);
3481         struct task_struct *task;
3482         struct pid_namespace *ns;
3483         int tid;
3484
3485         if (proc_inode_is_dead(inode))
3486                 return -ENOENT;
3487
3488         if (!dir_emit_dots(file, ctx))
3489                 return 0;
3490
3491         /* f_version caches the tgid value that the last readdir call couldn't
3492          * return. lseek aka telldir automagically resets f_version to 0.
3493          */
3494         ns = inode->i_sb->s_fs_info;
3495         tid = (int)file->f_version;
3496         file->f_version = 0;
3497         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3498              task;
3499              task = next_tid(task), ctx->pos++) {
3500                 char name[PROC_NUMBUF];
3501                 int len;
3502                 tid = task_pid_nr_ns(task, ns);
3503                 len = snprintf(name, sizeof(name), "%d", tid);
3504                 if (!proc_fill_cache(file, ctx, name, len,
3505                                 proc_task_instantiate, task, NULL)) {
3506                         /* returning this tgid failed, save it as the first
3507                          * pid for the next readir call */
3508                         file->f_version = (u64)tid;
3509                         put_task_struct(task);
3510                         break;
3511                 }
3512         }
3513
3514         return 0;
3515 }
3516
3517 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3518 {
3519         struct inode *inode = d_inode(dentry);
3520         struct task_struct *p = get_proc_task(inode);
3521         generic_fillattr(inode, stat);
3522
3523         if (p) {
3524                 stat->nlink += get_nr_threads(p);
3525                 put_task_struct(p);
3526         }
3527
3528         return 0;
3529 }
3530
3531 static const struct inode_operations proc_task_inode_operations = {
3532         .lookup         = proc_task_lookup,
3533         .getattr        = proc_task_getattr,
3534         .setattr        = proc_setattr,
3535         .permission     = proc_pid_permission,
3536 };
3537
3538 static const struct file_operations proc_task_operations = {
3539         .read           = generic_read_dir,
3540         .iterate_shared = proc_task_readdir,
3541         .llseek         = generic_file_llseek,
3542 };
3543
3544 void __init set_proc_pid_nlink(void)
3545 {
3546         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3547         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3548 }