fs/proc/base.c

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