fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <linux/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/printk.h>
  77 #include <linux/cgroup.h>
  78 #include <linux/cpuset.h>
  79 #include <linux/audit.h>
  80 #include <linux/poll.h>
  81 #include <linux/nsproxy.h>
  82 #include <linux/oom.h>
  83 #include <linux/elf.h>
  84 #include <linux/pid_namespace.h>
  85 #include <linux/user_namespace.h>
  86 #include <linux/fs_struct.h>
  87 #include <linux/slab.h>
  88 #include <linux/sched/autogroup.h>
  89 #include <linux/sched/mm.h>
  90 #include <linux/sched/coredump.h>
  91 #include <linux/sched/debug.h>
  92 #include <linux/sched/stat.h>
  93 #include <linux/flex_array.h>
  94 #include <linux/posix-timers.h>
  95 #ifdef CONFIG_HARDWALL
  96 #include <asm/hardwall.h>
  97 #endif
  98 #include <trace/events/oom.h>
  99 #include "internal.h"
 100 #include "fd.h"
 101
 102 /* NOTE:
 103  *      Implementing inode permission operations in /proc is almost
 104  *      certainly an error.  Permission checks need to happen during
 105  *      each system call not at open time.  The reason is that most of
 106  *      what we wish to check for permissions in /proc varies at runtime.
 107  *
 108  *      The classic example of a problem is opening file descriptors
 109  *      in /proc for a task before it execs a suid executable.
 110  */
 111
 112 static u8 nlink_tid;
 113 static u8 nlink_tgid;
 114
 115 struct pid_entry {
 116         const char *name;
 117         unsigned int len;
 118         umode_t mode;
 119         const struct inode_operations *iop;
 120         const struct file_operations *fop;
 121         union proc_op op;
 122 };
 123
 124 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 125         .name = (NAME),                                 \
 126         .len  = sizeof(NAME) - 1,                       \
 127         .mode = MODE,                                   \
 128         .iop  = IOP,                                    \
 129         .fop  = FOP,                                    \
 130         .op   = OP,                                     \
 131 }
 132
 133 #define DIR(NAME, MODE, iops, fops)     \
 134         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 135 #define LNK(NAME, get_link)                                     \
 136         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 137                 &proc_pid_link_inode_operations, NULL,          \
 138                 { .proc_get_link = get_link } )
 139 #define REG(NAME, MODE, fops)                           \
 140         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 141 #define ONE(NAME, MODE, show)                           \
 142         NOD(NAME, (S_IFREG|(MODE)),                     \
 143                 NULL, &proc_single_file_operations,     \
 144                 { .proc_show = show } )
 145
 146 /*
 147  * Count the number of hardlinks for the pid_entry table, excluding the .
 148  * and .. links.
 149  */
 150 static unsigned int __init pid_entry_nlink(const struct pid_entry *entries,
 151         unsigned int n)
 152 {
 153         unsigned int i;
 154         unsigned int count;
 155
 156         count = 2;
 157         for (i = 0; i < n; ++i) {
 158                 if (S_ISDIR(entries[i].mode))
 159                         ++count;
 160         }
 161
 162         return count;
 163 }
 164
 165 static int get_task_root(struct task_struct *task, struct path *root)
 166 {
 167         int result = -ENOENT;
 168
 169         task_lock(task);
 170         if (task->fs) {
 171                 get_fs_root(task->fs, root);
 172                 result = 0;
 173         }
 174         task_unlock(task);
 175         return result;
 176 }
 177
 178 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 179 {
 180         struct task_struct *task = get_proc_task(d_inode(dentry));
 181         int result = -ENOENT;
 182
 183         if (task) {
 184                 task_lock(task);
 185                 if (task->fs) {
 186                         get_fs_pwd(task->fs, path);
 187                         result = 0;
 188                 }
 189                 task_unlock(task);
 190                 put_task_struct(task);
 191         }
 192         return result;
 193 }
 194
 195 static int proc_root_link(struct dentry *dentry, struct path *path)
 196 {
 197         struct task_struct *task = get_proc_task(d_inode(dentry));
 198         int result = -ENOENT;
 199
 200         if (task) {
 201                 result = get_task_root(task, path);
 202                 put_task_struct(task);
 203         }
 204         return result;
 205 }
 206
 207 static ssize_t proc_pid_cmdline_read(struct file *file, char __user *buf,
 208                                      size_t _count, loff_t *pos)
 209 {
 210         struct task_struct *tsk;
 211         struct mm_struct *mm;
 212         char *page;
 213         unsigned long count = _count;
 214         unsigned long arg_start, arg_end, env_start, env_end;
 215         unsigned long len1, len2, len;
 216         unsigned long p;
 217         char c;
 218         ssize_t rv;
 219
 220         BUG_ON(*pos < 0);
 221
 222         tsk = get_proc_task(file_inode(file));
 223         if (!tsk)
 224                 return -ESRCH;
 225         mm = get_task_mm(tsk);
 226         put_task_struct(tsk);
 227         if (!mm)
 228                 return 0;
 229         /* Check if process spawned far enough to have cmdline. */
 230         if (!mm->env_end) {
 231                 rv = 0;
 232                 goto out_mmput;
 233         }
 234
 235         page = (char *)__get_free_page(GFP_TEMPORARY);
 236         if (!page) {
 237                 rv = -ENOMEM;
 238                 goto out_mmput;
 239         }
 240
 241         down_read(&mm->mmap_sem);
 242         arg_start = mm->arg_start;
 243         arg_end = mm->arg_end;
 244         env_start = mm->env_start;
 245         env_end = mm->env_end;
 246         up_read(&mm->mmap_sem);
 247
 248         BUG_ON(arg_start > arg_end);
 249         BUG_ON(env_start > env_end);
 250
 251         len1 = arg_end - arg_start;
 252         len2 = env_end - env_start;
 253
 254         /* Empty ARGV. */
 255         if (len1 == 0) {
 256                 rv = 0;
 257                 goto out_free_page;
 258         }
 259         /*
 260          * Inherently racy -- command line shares address space
 261          * with code and data.
 262          */
 263         rv = access_remote_vm(mm, arg_end - 1, &c, 1, 0);
 264         if (rv <= 0)
 265                 goto out_free_page;
 266
 267         rv = 0;
 268
 269         if (c == '\0') {
 270                 /* Command line (set of strings) occupies whole ARGV. */
 271                 if (len1 <= *pos)
 272                         goto out_free_page;
 273
 274                 p = arg_start + *pos;
 275                 len = len1 - *pos;
 276                 while (count > 0 && len > 0) {
 277                         unsigned int _count;
 278                         int nr_read;
 279
 280                         _count = min3(count, len, PAGE_SIZE);
 281                         nr_read = access_remote_vm(mm, p, page, _count, 0);
 282                         if (nr_read < 0)
 283                                 rv = nr_read;
 284                         if (nr_read <= 0)
 285                                 goto out_free_page;
 286
 287                         if (copy_to_user(buf, page, nr_read)) {
 288                                 rv = -EFAULT;
 289                                 goto out_free_page;
 290                         }
 291
 292                         p       += nr_read;
 293                         len     -= nr_read;
 294                         buf     += nr_read;
 295                         count   -= nr_read;
 296                         rv      += nr_read;
 297                 }
 298         } else {
 299                 /*
 300                  * Command line (1 string) occupies ARGV and
 301                  * extends into ENVP.
 302                  */
 303                 struct {
 304                         unsigned long p;
 305                         unsigned long len;
 306                 } cmdline[2] = {
 307                         { .p = arg_start, .len = len1 },
 308                         { .p = env_start, .len = len2 },
 309                 };
 310                 loff_t pos1 = *pos;
 311                 unsigned int i;
 312
 313                 i = 0;
 314                 while (i < 2 && pos1 >= cmdline[i].len) {
 315                         pos1 -= cmdline[i].len;
 316                         i++;
 317                 }
 318                 while (i < 2) {
 319                         p = cmdline[i].p + pos1;
 320                         len = cmdline[i].len - pos1;
 321                         while (count > 0 && len > 0) {
 322                                 unsigned int _count, l;
 323                                 int nr_read;
 324                                 bool final;
 325
 326                                 _count = min3(count, len, PAGE_SIZE);
 327                                 nr_read = access_remote_vm(mm, p, page, _count, 0);
 328                                 if (nr_read < 0)
 329                                         rv = nr_read;
 330                                 if (nr_read <= 0)
 331                                         goto out_free_page;
 332
 333                                 /*
 334                                  * Command line can be shorter than whole ARGV
 335                                  * even if last "marker" byte says it is not.
 336                                  */
 337                                 final = false;
 338                                 l = strnlen(page, nr_read);
 339                                 if (l < nr_read) {
 340                                         nr_read = l;
 341                                         final = true;
 342                                 }
 343
 344                                 if (copy_to_user(buf, page, nr_read)) {
 345                                         rv = -EFAULT;
 346                                         goto out_free_page;
 347                                 }
 348
 349                                 p       += nr_read;
 350                                 len     -= nr_read;
 351                                 buf     += nr_read;
 352                                 count   -= nr_read;
 353                                 rv      += nr_read;
 354
 355                                 if (final)
 356                                         goto out_free_page;
 357                         }
 358
 359                         /* Only first chunk can be read partially. */
 360                         pos1 = 0;
 361                         i++;
 362                 }
 363         }
 364
 365 out_free_page:
 366         free_page((unsigned long)page);
 367 out_mmput:
 368         mmput(mm);
 369         if (rv > 0)
 370                 *pos += rv;
 371         return rv;
 372 }
 373
 374 static const struct file_operations proc_pid_cmdline_ops = {
 375         .read   = proc_pid_cmdline_read,
 376         .llseek = generic_file_llseek,
 377 };
 378
 379 #ifdef CONFIG_KALLSYMS
 380 /*
 381  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 382  * Returns the resolved symbol.  If that fails, simply return the address.
 383  */
 384 static int proc_pid_wchan(struct seq_file *m, struct pid_namespace *ns,
 385                           struct pid *pid, struct task_struct *task)
 386 {
 387         unsigned long wchan;
 388         char symname[KSYM_NAME_LEN];
 389
 390         wchan = get_wchan(task);
 391
 392         if (wchan && ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)
 393                         && !lookup_symbol_name(wchan, symname))
 394                 seq_printf(m, "%s", symname);
 395         else
 396                 seq_putc(m, '0');
 397
 398         return 0;
 399 }
 400 #endif /* CONFIG_KALLSYMS */
 401
 402 static int lock_trace(struct task_struct *task)
 403 {
 404         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 405         if (err)
 406                 return err;
 407         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH_FSCREDS)) {
 408                 mutex_unlock(&task->signal->cred_guard_mutex);
 409                 return -EPERM;
 410         }
 411         return 0;
 412 }
 413
 414 static void unlock_trace(struct task_struct *task)
 415 {
 416         mutex_unlock(&task->signal->cred_guard_mutex);
 417 }
 418
 419 #ifdef CONFIG_STACKTRACE
 420
 421 #define MAX_STACK_TRACE_DEPTH   64
 422
 423 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 424                           struct pid *pid, struct task_struct *task)
 425 {
 426         struct stack_trace trace;
 427         unsigned long *entries;
 428         int err;
 429         int i;
 430
 431         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 432         if (!entries)
 433                 return -ENOMEM;
 434
 435         trace.nr_entries        = 0;
 436         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 437         trace.entries           = entries;
 438         trace.skip              = 0;
 439
 440         err = lock_trace(task);
 441         if (!err) {
 442                 save_stack_trace_tsk(task, &trace);
 443
 444                 for (i = 0; i < trace.nr_entries; i++) {
 445                         seq_printf(m, "[<%pK>] %pB\n",
 446                                    (void *)entries[i], (void *)entries[i]);
 447                 }
 448                 unlock_trace(task);
 449         }
 450         kfree(entries);
 451
 452         return err;
 453 }
 454 #endif
 455
 456 #ifdef CONFIG_SCHED_INFO
 457 /*
 458  * Provides /proc/PID/schedstat
 459  */
 460 static int proc_pid_schedstat(struct seq_file *m, struct pid_namespace *ns,
 461                               struct pid *pid, struct task_struct *task)
 462 {
 463         if (unlikely(!sched_info_on()))
 464                 seq_printf(m, "0 0 0\n");
 465         else
 466                 seq_printf(m, "%llu %llu %lu\n",
 467                    (unsigned long long)task->se.sum_exec_runtime,
 468                    (unsigned long long)task->sched_info.run_delay,
 469                    task->sched_info.pcount);
 470
 471         return 0;
 472 }
 473 #endif
 474
 475 #ifdef CONFIG_LATENCYTOP
 476 static int lstats_show_proc(struct seq_file *m, void *v)
 477 {
 478         int i;
 479         struct inode *inode = m->private;
 480         struct task_struct *task = get_proc_task(inode);
 481
 482         if (!task)
 483                 return -ESRCH;
 484         seq_puts(m, "Latency Top version : v0.1\n");
 485         for (i = 0; i < 32; i++) {
 486                 struct latency_record *lr = &task->latency_record[i];
 487                 if (lr->backtrace[0]) {
 488                         int q;
 489                         seq_printf(m, "%i %li %li",
 490                                    lr->count, lr->time, lr->max);
 491                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 492                                 unsigned long bt = lr->backtrace[q];
 493                                 if (!bt)
 494                                         break;
 495                                 if (bt == ULONG_MAX)
 496                                         break;
 497                                 seq_printf(m, " %ps", (void *)bt);
 498                         }
 499                         seq_putc(m, '\n');
 500                 }
 501
 502         }
 503         put_task_struct(task);
 504         return 0;
 505 }
 506
 507 static int lstats_open(struct inode *inode, struct file *file)
 508 {
 509         return single_open(file, lstats_show_proc, inode);
 510 }
 511
 512 static ssize_t lstats_write(struct file *file, const char __user *buf,
 513                             size_t count, loff_t *offs)
 514 {
 515         struct task_struct *task = get_proc_task(file_inode(file));
 516
 517         if (!task)
 518                 return -ESRCH;
 519         clear_all_latency_tracing(task);
 520         put_task_struct(task);
 521
 522         return count;
 523 }
 524
 525 static const struct file_operations proc_lstats_operations = {
 526         .open           = lstats_open,
 527         .read           = seq_read,
 528         .write          = lstats_write,
 529         .llseek         = seq_lseek,
 530         .release        = single_release,
 531 };
 532
 533 #endif
 534
 535 static int proc_oom_score(struct seq_file *m, struct pid_namespace *ns,
 536                           struct pid *pid, struct task_struct *task)
 537 {
 538         unsigned long totalpages = totalram_pages + total_swap_pages;
 539         unsigned long points = 0;
 540
 541         points = oom_badness(task, NULL, NULL, totalpages) *
 542                                         1000 / totalpages;
 543         seq_printf(m, "%lu\n", points);
 544
 545         return 0;
 546 }
 547
 548 struct limit_names {
 549         const char *name;
 550         const char *unit;
 551 };
 552
 553 static const struct limit_names lnames[RLIM_NLIMITS] = {
 554         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 555         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 556         [RLIMIT_DATA] = {"Max data size", "bytes"},
 557         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 558         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 559         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 560         [RLIMIT_NPROC] = {"Max processes", "processes"},
 561         [RLIMIT_NOFILE] = {"Max open files", "files"},
 562         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 563         [RLIMIT_AS] = {"Max address space", "bytes"},
 564         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 565         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 566         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 567         [RLIMIT_NICE] = {"Max nice priority", NULL},
 568         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 569         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 570 };
 571
 572 /* Display limits for a process */
 573 static int proc_pid_limits(struct seq_file *m, struct pid_namespace *ns,
 574                            struct pid *pid, struct task_struct *task)
 575 {
 576         unsigned int i;
 577         unsigned long flags;
 578
 579         struct rlimit rlim[RLIM_NLIMITS];
 580
 581         if (!lock_task_sighand(task, &flags))
 582                 return 0;
 583         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 584         unlock_task_sighand(task, &flags);
 585
 586         /*
 587          * print the file header
 588          */
 589        seq_printf(m, "%-25s %-20s %-20s %-10s\n",
 590                   "Limit", "Soft Limit", "Hard Limit", "Units");
 591
 592         for (i = 0; i < RLIM_NLIMITS; i++) {
 593                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 594                         seq_printf(m, "%-25s %-20s ",
 595                                    lnames[i].name, "unlimited");
 596                 else
 597                         seq_printf(m, "%-25s %-20lu ",
 598                                    lnames[i].name, rlim[i].rlim_cur);
 599
 600                 if (rlim[i].rlim_max == RLIM_INFINITY)
 601                         seq_printf(m, "%-20s ", "unlimited");
 602                 else
 603                         seq_printf(m, "%-20lu ", rlim[i].rlim_max);
 604
 605                 if (lnames[i].unit)
 606                         seq_printf(m, "%-10s\n", lnames[i].unit);
 607                 else
 608                         seq_putc(m, '\n');
 609         }
 610
 611         return 0;
 612 }
 613
 614 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 615 static int proc_pid_syscall(struct seq_file *m, struct pid_namespace *ns,
 616                             struct pid *pid, struct task_struct *task)
 617 {
 618         long nr;
 619         unsigned long args[6], sp, pc;
 620         int res;
 621
 622         res = lock_trace(task);
 623         if (res)
 624                 return res;
 625
 626         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 627                 seq_puts(m, "running\n");
 628         else if (nr < 0)
 629                 seq_printf(m, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 630         else
 631                 seq_printf(m,
 632                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 633                        nr,
 634                        args[0], args[1], args[2], args[3], args[4], args[5],
 635                        sp, pc);
 636         unlock_trace(task);
 637
 638         return 0;
 639 }
 640 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 641
 642 /************************************************************************/
 643 /*                       Here the fs part begins                        */
 644 /************************************************************************/
 645
 646 /* permission checks */
 647 static int proc_fd_access_allowed(struct inode *inode)
 648 {
 649         struct task_struct *task;
 650         int allowed = 0;
 651         /* Allow access to a task's file descriptors if it is us or we
 652          * may use ptrace attach to the process and find out that
 653          * information.
 654          */
 655         task = get_proc_task(inode);
 656         if (task) {
 657                 allowed = ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 658                 put_task_struct(task);
 659         }
 660         return allowed;
 661 }
 662
 663 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 664 {
 665         int error;
 666         struct inode *inode = d_inode(dentry);
 667
 668         if (attr->ia_valid & ATTR_MODE)
 669                 return -EPERM;
 670
 671         error = setattr_prepare(dentry, attr);
 672         if (error)
 673                 return error;
 674
 675         setattr_copy(inode, attr);
 676         mark_inode_dirty(inode);
 677         return 0;
 678 }
 679
 680 /*
 681  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 682  * or euid/egid (for hide_pid_min=2)?
 683  */
 684 static bool has_pid_permissions(struct pid_namespace *pid,
 685                                  struct task_struct *task,
 686                                  int hide_pid_min)
 687 {
 688         if (pid->hide_pid < hide_pid_min)
 689                 return true;
 690         if (in_group_p(pid->pid_gid))
 691                 return true;
 692         return ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS);
 693 }
 694
 695
 696 static int proc_pid_permission(struct inode *inode, int mask)
 697 {
 698         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 699         struct task_struct *task;
 700         bool has_perms;
 701
 702         task = get_proc_task(inode);
 703         if (!task)
 704                 return -ESRCH;
 705         has_perms = has_pid_permissions(pid, task, HIDEPID_NO_ACCESS);
 706         put_task_struct(task);
 707
 708         if (!has_perms) {
 709                 if (pid->hide_pid == HIDEPID_INVISIBLE) {
 710                         /*
 711                          * Let's make getdents(), stat(), and open()
 712                          * consistent with each other.  If a process
 713                          * may not stat() a file, it shouldn't be seen
 714                          * in procfs at all.
 715                          */
 716                         return -ENOENT;
 717                 }
 718
 719                 return -EPERM;
 720         }
 721         return generic_permission(inode, mask);
 722 }
 723
 724
 725
 726 static const struct inode_operations proc_def_inode_operations = {
 727         .setattr        = proc_setattr,
 728 };
 729
 730 static int proc_single_show(struct seq_file *m, void *v)
 731 {
 732         struct inode *inode = m->private;
 733         struct pid_namespace *ns;
 734         struct pid *pid;
 735         struct task_struct *task;
 736         int ret;
 737
 738         ns = inode->i_sb->s_fs_info;
 739         pid = proc_pid(inode);
 740         task = get_pid_task(pid, PIDTYPE_PID);
 741         if (!task)
 742                 return -ESRCH;
 743
 744         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 745
 746         put_task_struct(task);
 747         return ret;
 748 }
 749
 750 static int proc_single_open(struct inode *inode, struct file *filp)
 751 {
 752         return single_open(filp, proc_single_show, inode);
 753 }
 754
 755 static const struct file_operations proc_single_file_operations = {
 756         .open           = proc_single_open,
 757         .read           = seq_read,
 758         .llseek         = seq_lseek,
 759         .release        = single_release,
 760 };
 761
 762
 763 struct mm_struct *proc_mem_open(struct inode *inode, unsigned int mode)
 764 {
 765         struct task_struct *task = get_proc_task(inode);
 766         struct mm_struct *mm = ERR_PTR(-ESRCH);
 767
 768         if (task) {
 769                 mm = mm_access(task, mode | PTRACE_MODE_FSCREDS);
 770                 put_task_struct(task);
 771
 772                 if (!IS_ERR_OR_NULL(mm)) {
 773                         /* ensure this mm_struct can't be freed */
 774                         mmgrab(mm);
 775                         /* but do not pin its memory */
 776                         mmput(mm);
 777                 }
 778         }
 779
 780         return mm;
 781 }
 782
 783 static int __mem_open(struct inode *inode, struct file *file, unsigned int mode)
 784 {
 785         struct mm_struct *mm = proc_mem_open(inode, mode);
 786
 787         if (IS_ERR(mm))
 788                 return PTR_ERR(mm);
 789
 790         file->private_data = mm;
 791         return 0;
 792 }
 793
 794 static int mem_open(struct inode *inode, struct file *file)
 795 {
 796         int ret = __mem_open(inode, file, PTRACE_MODE_ATTACH);
 797
 798         /* OK to pass negative loff_t, we can catch out-of-range */
 799         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 800
 801         return ret;
 802 }
 803
 804 static ssize_t mem_rw(struct file *file, char __user *buf,
 805                         size_t count, loff_t *ppos, int write)
 806 {
 807         struct mm_struct *mm = file->private_data;
 808         unsigned long addr = *ppos;
 809         ssize_t copied;
 810         char *page;
 811         unsigned int flags;
 812
 813         if (!mm)
 814                 return 0;
 815
 816         page = (char *)__get_free_page(GFP_TEMPORARY);
 817         if (!page)
 818                 return -ENOMEM;
 819
 820         copied = 0;
 821         if (!mmget_not_zero(mm))
 822                 goto free;
 823
 824         flags = FOLL_FORCE | (write ? FOLL_WRITE : 0);
 825
 826         while (count > 0) {
 827                 int this_len = min_t(int, count, PAGE_SIZE);
 828
 829                 if (write && copy_from_user(page, buf, this_len)) {
 830                         copied = -EFAULT;
 831                         break;
 832                 }
 833
 834                 this_len = access_remote_vm(mm, addr, page, this_len, flags);
 835                 if (!this_len) {
 836                         if (!copied)
 837                                 copied = -EIO;
 838                         break;
 839                 }
 840
 841                 if (!write && copy_to_user(buf, page, this_len)) {
 842                         copied = -EFAULT;
 843                         break;
 844                 }
 845
 846                 buf += this_len;
 847                 addr += this_len;
 848                 copied += this_len;
 849                 count -= this_len;
 850         }
 851         *ppos = addr;
 852
 853         mmput(mm);
 854 free:
 855         free_page((unsigned long) page);
 856         return copied;
 857 }
 858
 859 static ssize_t mem_read(struct file *file, char __user *buf,
 860                         size_t count, loff_t *ppos)
 861 {
 862         return mem_rw(file, buf, count, ppos, 0);
 863 }
 864
 865 static ssize_t mem_write(struct file *file, const char __user *buf,
 866                          size_t count, loff_t *ppos)
 867 {
 868         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 869 }
 870
 871 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 872 {
 873         switch (orig) {
 874         case 0:
 875                 file->f_pos = offset;
 876                 break;
 877         case 1:
 878                 file->f_pos += offset;
 879                 break;
 880         default:
 881                 return -EINVAL;
 882         }
 883         force_successful_syscall_return();
 884         return file->f_pos;
 885 }
 886
 887 static int mem_release(struct inode *inode, struct file *file)
 888 {
 889         struct mm_struct *mm = file->private_data;
 890         if (mm)
 891                 mmdrop(mm);
 892         return 0;
 893 }
 894
 895 static const struct file_operations proc_mem_operations = {
 896         .llseek         = mem_lseek,
 897         .read           = mem_read,
 898         .write          = mem_write,
 899         .open           = mem_open,
 900         .release        = mem_release,
 901 };
 902
 903 static int environ_open(struct inode *inode, struct file *file)
 904 {
 905         return __mem_open(inode, file, PTRACE_MODE_READ);
 906 }
 907
 908 static ssize_t environ_read(struct file *file, char __user *buf,
 909                         size_t count, loff_t *ppos)
 910 {
 911         char *page;
 912         unsigned long src = *ppos;
 913         int ret = 0;
 914         struct mm_struct *mm = file->private_data;
 915         unsigned long env_start, env_end;
 916
 917         /* Ensure the process spawned far enough to have an environment. */
 918         if (!mm || !mm->env_end)
 919                 return 0;
 920
 921         page = (char *)__get_free_page(GFP_TEMPORARY);
 922         if (!page)
 923                 return -ENOMEM;
 924
 925         ret = 0;
 926         if (!mmget_not_zero(mm))
 927                 goto free;
 928
 929         down_read(&mm->mmap_sem);
 930         env_start = mm->env_start;
 931         env_end = mm->env_end;
 932         up_read(&mm->mmap_sem);
 933
 934         while (count > 0) {
 935                 size_t this_len, max_len;
 936                 int retval;
 937
 938                 if (src >= (env_end - env_start))
 939                         break;
 940
 941                 this_len = env_end - (env_start + src);
 942
 943                 max_len = min_t(size_t, PAGE_SIZE, count);
 944                 this_len = min(max_len, this_len);
 945
 946                 retval = access_remote_vm(mm, (env_start + src), page, this_len, 0);
 947
 948                 if (retval <= 0) {
 949                         ret = retval;
 950                         break;
 951                 }
 952
 953                 if (copy_to_user(buf, page, retval)) {
 954                         ret = -EFAULT;
 955                         break;
 956                 }
 957
 958                 ret += retval;
 959                 src += retval;
 960                 buf += retval;
 961                 count -= retval;
 962         }
 963         *ppos = src;
 964         mmput(mm);
 965
 966 free:
 967         free_page((unsigned long) page);
 968         return ret;
 969 }
 970
 971 static const struct file_operations proc_environ_operations = {
 972         .open           = environ_open,
 973         .read           = environ_read,
 974         .llseek         = generic_file_llseek,
 975         .release        = mem_release,
 976 };
 977
 978 static int auxv_open(struct inode *inode, struct file *file)
 979 {
 980         return __mem_open(inode, file, PTRACE_MODE_READ_FSCREDS);
 981 }
 982
 983 static ssize_t auxv_read(struct file *file, char __user *buf,
 984                         size_t count, loff_t *ppos)
 985 {
 986         struct mm_struct *mm = file->private_data;
 987         unsigned int nwords = 0;
 988
 989         if (!mm)
 990                 return 0;
 991         do {
 992                 nwords += 2;
 993         } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 994         return simple_read_from_buffer(buf, count, ppos, mm->saved_auxv,
 995                                        nwords * sizeof(mm->saved_auxv[0]));
 996 }
 997
 998 static const struct file_operations proc_auxv_operations = {
 999         .open           = auxv_open,
1000         .read           = auxv_read,
1001         .llseek         = generic_file_llseek,
1002         .release        = mem_release,
1003 };
1004
1005 static ssize_t oom_adj_read(struct file *file, char __user *buf, size_t count,
1006                             loff_t *ppos)
1007 {
1008         struct task_struct *task = get_proc_task(file_inode(file));
1009         char buffer[PROC_NUMBUF];
1010         int oom_adj = OOM_ADJUST_MIN;
1011         size_t len;
1012
1013         if (!task)
1014                 return -ESRCH;
1015         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MAX)
1016                 oom_adj = OOM_ADJUST_MAX;
1017         else
1018                 oom_adj = (task->signal->oom_score_adj * -OOM_DISABLE) /
1019                           OOM_SCORE_ADJ_MAX;
1020         put_task_struct(task);
1021         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_adj);
1022         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1023 }
1024
1025 static int __set_oom_adj(struct file *file, int oom_adj, bool legacy)
1026 {
1027         static DEFINE_MUTEX(oom_adj_mutex);
1028         struct mm_struct *mm = NULL;
1029         struct task_struct *task;
1030         int err = 0;
1031
1032         task = get_proc_task(file_inode(file));
1033         if (!task)
1034                 return -ESRCH;
1035
1036         mutex_lock(&oom_adj_mutex);
1037         if (legacy) {
1038                 if (oom_adj < task->signal->oom_score_adj &&
1039                                 !capable(CAP_SYS_RESOURCE)) {
1040                         err = -EACCES;
1041                         goto err_unlock;
1042                 }
1043                 /*
1044                  * /proc/pid/oom_adj is provided for legacy purposes, ask users to use
1045                  * /proc/pid/oom_score_adj instead.
1046                  */
1047                 pr_warn_once("%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1048                           current->comm, task_pid_nr(current), task_pid_nr(task),
1049                           task_pid_nr(task));
1050         } else {
1051                 if ((short)oom_adj < task->signal->oom_score_adj_min &&
1052                                 !capable(CAP_SYS_RESOURCE)) {
1053                         err = -EACCES;
1054                         goto err_unlock;
1055                 }
1056         }
1057
1058         /*
1059          * Make sure we will check other processes sharing the mm if this is
1060          * not vfrok which wants its own oom_score_adj.
1061          * pin the mm so it doesn't go away and get reused after task_unlock
1062          */
1063         if (!task->vfork_done) {
1064                 struct task_struct *p = find_lock_task_mm(task);
1065
1066                 if (p) {
1067                         if (atomic_read(&p->mm->mm_users) > 1) {
1068                                 mm = p->mm;
1069                                 mmgrab(mm);
1070                         }
1071                         task_unlock(p);
1072                 }
1073         }
1074
1075         task->signal->oom_score_adj = oom_adj;
1076         if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1077                 task->signal->oom_score_adj_min = (short)oom_adj;
1078         trace_oom_score_adj_update(task);
1079
1080         if (mm) {
1081                 struct task_struct *p;
1082
1083                 rcu_read_lock();
1084                 for_each_process(p) {
1085                         if (same_thread_group(task, p))
1086                                 continue;
1087
1088                         /* do not touch kernel threads or the global init */
1089                         if (p->flags & PF_KTHREAD || is_global_init(p))
1090                                 continue;
1091
1092                         task_lock(p);
1093                         if (!p->vfork_done && process_shares_mm(p, mm)) {
1094                                 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",
1095                                                 task_pid_nr(p), p->comm,
1096                                                 p->signal->oom_score_adj, oom_adj,
1097                                                 task_pid_nr(task), task->comm);
1098                                 p->signal->oom_score_adj = oom_adj;
1099                                 if (!legacy && has_capability_noaudit(current, CAP_SYS_RESOURCE))
1100                                         p->signal->oom_score_adj_min = (short)oom_adj;
1101                         }
1102                         task_unlock(p);
1103                 }
1104                 rcu_read_unlock();
1105                 mmdrop(mm);
1106         }
1107 err_unlock:
1108         mutex_unlock(&oom_adj_mutex);
1109         put_task_struct(task);
1110         return err;
1111 }
1112
1113 /*
1114  * /proc/pid/oom_adj exists solely for backwards compatibility with previous
1115  * kernels.  The effective policy is defined by oom_score_adj, which has a
1116  * different scale: oom_adj grew exponentially and oom_score_adj grows linearly.
1117  * Values written to oom_adj are simply mapped linearly to oom_score_adj.
1118  * Processes that become oom disabled via oom_adj will still be oom disabled
1119  * with this implementation.
1120  *
1121  * oom_adj cannot be removed since existing userspace binaries use it.
1122  */
1123 static ssize_t oom_adj_write(struct file *file, const char __user *buf,
1124                              size_t count, loff_t *ppos)
1125 {
1126         char buffer[PROC_NUMBUF];
1127         int oom_adj;
1128         int err;
1129
1130         memset(buffer, 0, sizeof(buffer));
1131         if (count > sizeof(buffer) - 1)
1132                 count = sizeof(buffer) - 1;
1133         if (copy_from_user(buffer, buf, count)) {
1134                 err = -EFAULT;
1135                 goto out;
1136         }
1137
1138         err = kstrtoint(strstrip(buffer), 0, &oom_adj);
1139         if (err)
1140                 goto out;
1141         if ((oom_adj < OOM_ADJUST_MIN || oom_adj > OOM_ADJUST_MAX) &&
1142              oom_adj != OOM_DISABLE) {
1143                 err = -EINVAL;
1144                 goto out;
1145         }
1146
1147         /*
1148          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1149          * value is always attainable.
1150          */
1151         if (oom_adj == OOM_ADJUST_MAX)
1152                 oom_adj = OOM_SCORE_ADJ_MAX;
1153         else
1154                 oom_adj = (oom_adj * OOM_SCORE_ADJ_MAX) / -OOM_DISABLE;
1155
1156         err = __set_oom_adj(file, oom_adj, true);
1157 out:
1158         return err < 0 ? err : count;
1159 }
1160
1161 static const struct file_operations proc_oom_adj_operations = {
1162         .read           = oom_adj_read,
1163         .write          = oom_adj_write,
1164         .llseek         = generic_file_llseek,
1165 };
1166
1167 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1168                                         size_t count, loff_t *ppos)
1169 {
1170         struct task_struct *task = get_proc_task(file_inode(file));
1171         char buffer[PROC_NUMBUF];
1172         short oom_score_adj = OOM_SCORE_ADJ_MIN;
1173         size_t len;
1174
1175         if (!task)
1176                 return -ESRCH;
1177         oom_score_adj = task->signal->oom_score_adj;
1178         put_task_struct(task);
1179         len = snprintf(buffer, sizeof(buffer), "%hd\n", oom_score_adj);
1180         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1181 }
1182
1183 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1184                                         size_t count, loff_t *ppos)
1185 {
1186         char buffer[PROC_NUMBUF];
1187         int oom_score_adj;
1188         int err;
1189
1190         memset(buffer, 0, sizeof(buffer));
1191         if (count > sizeof(buffer) - 1)
1192                 count = sizeof(buffer) - 1;
1193         if (copy_from_user(buffer, buf, count)) {
1194                 err = -EFAULT;
1195                 goto out;
1196         }
1197
1198         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1199         if (err)
1200                 goto out;
1201         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1202                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1203                 err = -EINVAL;
1204                 goto out;
1205         }
1206
1207         err = __set_oom_adj(file, oom_score_adj, false);
1208 out:
1209         return err < 0 ? err : count;
1210 }
1211
1212 static const struct file_operations proc_oom_score_adj_operations = {
1213         .read           = oom_score_adj_read,
1214         .write          = oom_score_adj_write,
1215         .llseek         = default_llseek,
1216 };
1217
1218 #ifdef CONFIG_AUDITSYSCALL
1219 #define TMPBUFLEN 11
1220 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1221                                   size_t count, loff_t *ppos)
1222 {
1223         struct inode * inode = file_inode(file);
1224         struct task_struct *task = get_proc_task(inode);
1225         ssize_t length;
1226         char tmpbuf[TMPBUFLEN];
1227
1228         if (!task)
1229                 return -ESRCH;
1230         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1231                            from_kuid(file->f_cred->user_ns,
1232                                      audit_get_loginuid(task)));
1233         put_task_struct(task);
1234         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1235 }
1236
1237 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1238                                    size_t count, loff_t *ppos)
1239 {
1240         struct inode * inode = file_inode(file);
1241         uid_t loginuid;
1242         kuid_t kloginuid;
1243         int rv;
1244
1245         rcu_read_lock();
1246         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1247                 rcu_read_unlock();
1248                 return -EPERM;
1249         }
1250         rcu_read_unlock();
1251
1252         if (*ppos != 0) {
1253                 /* No partial writes. */
1254                 return -EINVAL;
1255         }
1256
1257         rv = kstrtou32_from_user(buf, count, 10, &loginuid);
1258         if (rv < 0)
1259                 return rv;
1260
1261         /* is userspace tring to explicitly UNSET the loginuid? */
1262         if (loginuid == AUDIT_UID_UNSET) {
1263                 kloginuid = INVALID_UID;
1264         } else {
1265                 kloginuid = make_kuid(file->f_cred->user_ns, loginuid);
1266                 if (!uid_valid(kloginuid))
1267                         return -EINVAL;
1268         }
1269
1270         rv = audit_set_loginuid(kloginuid);
1271         if (rv < 0)
1272                 return rv;
1273         return count;
1274 }
1275
1276 static const struct file_operations proc_loginuid_operations = {
1277         .read           = proc_loginuid_read,
1278         .write          = proc_loginuid_write,
1279         .llseek         = generic_file_llseek,
1280 };
1281
1282 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1283                                   size_t count, loff_t *ppos)
1284 {
1285         struct inode * inode = file_inode(file);
1286         struct task_struct *task = get_proc_task(inode);
1287         ssize_t length;
1288         char tmpbuf[TMPBUFLEN];
1289
1290         if (!task)
1291                 return -ESRCH;
1292         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1293                                 audit_get_sessionid(task));
1294         put_task_struct(task);
1295         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1296 }
1297
1298 static const struct file_operations proc_sessionid_operations = {
1299         .read           = proc_sessionid_read,
1300         .llseek         = generic_file_llseek,
1301 };
1302 #endif
1303
1304 #ifdef CONFIG_FAULT_INJECTION
1305 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1306                                       size_t count, loff_t *ppos)
1307 {
1308         struct task_struct *task = get_proc_task(file_inode(file));
1309         char buffer[PROC_NUMBUF];
1310         size_t len;
1311         int make_it_fail;
1312
1313         if (!task)
1314                 return -ESRCH;
1315         make_it_fail = task->make_it_fail;
1316         put_task_struct(task);
1317
1318         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1319
1320         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1321 }
1322
1323 static ssize_t proc_fault_inject_write(struct file * file,
1324                         const char __user * buf, size_t count, loff_t *ppos)
1325 {
1326         struct task_struct *task;
1327         char buffer[PROC_NUMBUF];
1328         int make_it_fail;
1329         int rv;
1330
1331         if (!capable(CAP_SYS_RESOURCE))
1332                 return -EPERM;
1333         memset(buffer, 0, sizeof(buffer));
1334         if (count > sizeof(buffer) - 1)
1335                 count = sizeof(buffer) - 1;
1336         if (copy_from_user(buffer, buf, count))
1337                 return -EFAULT;
1338         rv = kstrtoint(strstrip(buffer), 0, &make_it_fail);
1339         if (rv < 0)
1340                 return rv;
1341         if (make_it_fail < 0 || make_it_fail > 1)
1342                 return -EINVAL;
1343
1344         task = get_proc_task(file_inode(file));
1345         if (!task)
1346                 return -ESRCH;
1347         task->make_it_fail = make_it_fail;
1348         put_task_struct(task);
1349
1350         return count;
1351 }
1352
1353 static const struct file_operations proc_fault_inject_operations = {
1354         .read           = proc_fault_inject_read,
1355         .write          = proc_fault_inject_write,
1356         .llseek         = generic_file_llseek,
1357 };
1358
1359 static ssize_t proc_fail_nth_write(struct file *file, const char __user *buf,
1360                                    size_t count, loff_t *ppos)
1361 {
1362         struct task_struct *task;
1363         int err;
1364         unsigned int n;
1365
1366         err = kstrtouint_from_user(buf, count, 0, &n);
1367         if (err)
1368                 return err;
1369
1370         task = get_proc_task(file_inode(file));
1371         if (!task)
1372                 return -ESRCH;
1373         WRITE_ONCE(task->fail_nth, n);
1374         put_task_struct(task);
1375
1376         return count;
1377 }
1378
1379 static ssize_t proc_fail_nth_read(struct file *file, char __user *buf,
1380                                   size_t count, loff_t *ppos)
1381 {
1382         struct task_struct *task;
1383         char numbuf[PROC_NUMBUF];
1384         ssize_t len;
1385
1386         task = get_proc_task(file_inode(file));
1387         if (!task)
1388                 return -ESRCH;
1389         len = snprintf(numbuf, sizeof(numbuf), "%u\n",
1390                         READ_ONCE(task->fail_nth));
1391         len = simple_read_from_buffer(buf, count, ppos, numbuf, len);
1392         put_task_struct(task);
1393
1394         return len;
1395 }
1396
1397 static const struct file_operations proc_fail_nth_operations = {
1398         .read           = proc_fail_nth_read,
1399         .write          = proc_fail_nth_write,
1400 };
1401 #endif
1402
1403
1404 #ifdef CONFIG_SCHED_DEBUG
1405 /*
1406  * Print out various scheduling related per-task fields:
1407  */
1408 static int sched_show(struct seq_file *m, void *v)
1409 {
1410         struct inode *inode = m->private;
1411         struct task_struct *p;
1412
1413         p = get_proc_task(inode);
1414         if (!p)
1415                 return -ESRCH;
1416         proc_sched_show_task(p, m);
1417
1418         put_task_struct(p);
1419
1420         return 0;
1421 }
1422
1423 static ssize_t
1424 sched_write(struct file *file, const char __user *buf,
1425             size_t count, loff_t *offset)
1426 {
1427         struct inode *inode = file_inode(file);
1428         struct task_struct *p;
1429
1430         p = get_proc_task(inode);
1431         if (!p)
1432                 return -ESRCH;
1433         proc_sched_set_task(p);
1434
1435         put_task_struct(p);
1436
1437         return count;
1438 }
1439
1440 static int sched_open(struct inode *inode, struct file *filp)
1441 {
1442         return single_open(filp, sched_show, inode);
1443 }
1444
1445 static const struct file_operations proc_pid_sched_operations = {
1446         .open           = sched_open,
1447         .read           = seq_read,
1448         .write          = sched_write,
1449         .llseek         = seq_lseek,
1450         .release        = single_release,
1451 };
1452
1453 #endif
1454
1455 #ifdef CONFIG_SCHED_AUTOGROUP
1456 /*
1457  * Print out autogroup related information:
1458  */
1459 static int sched_autogroup_show(struct seq_file *m, void *v)
1460 {
1461         struct inode *inode = m->private;
1462         struct task_struct *p;
1463
1464         p = get_proc_task(inode);
1465         if (!p)
1466                 return -ESRCH;
1467         proc_sched_autogroup_show_task(p, m);
1468
1469         put_task_struct(p);
1470
1471         return 0;
1472 }
1473
1474 static ssize_t
1475 sched_autogroup_write(struct file *file, const char __user *buf,
1476             size_t count, loff_t *offset)
1477 {
1478         struct inode *inode = file_inode(file);
1479         struct task_struct *p;
1480         char buffer[PROC_NUMBUF];
1481         int nice;
1482         int err;
1483
1484         memset(buffer, 0, sizeof(buffer));
1485         if (count > sizeof(buffer) - 1)
1486                 count = sizeof(buffer) - 1;
1487         if (copy_from_user(buffer, buf, count))
1488                 return -EFAULT;
1489
1490         err = kstrtoint(strstrip(buffer), 0, &nice);
1491         if (err < 0)
1492                 return err;
1493
1494         p = get_proc_task(inode);
1495         if (!p)
1496                 return -ESRCH;
1497
1498         err = proc_sched_autogroup_set_nice(p, nice);
1499         if (err)
1500                 count = err;
1501
1502         put_task_struct(p);
1503
1504         return count;
1505 }
1506
1507 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1508 {
1509         int ret;
1510
1511         ret = single_open(filp, sched_autogroup_show, NULL);
1512         if (!ret) {
1513                 struct seq_file *m = filp->private_data;
1514
1515                 m->private = inode;
1516         }
1517         return ret;
1518 }
1519
1520 static const struct file_operations proc_pid_sched_autogroup_operations = {
1521         .open           = sched_autogroup_open,
1522         .read           = seq_read,
1523         .write          = sched_autogroup_write,
1524         .llseek         = seq_lseek,
1525         .release        = single_release,
1526 };
1527
1528 #endif /* CONFIG_SCHED_AUTOGROUP */
1529
1530 static ssize_t comm_write(struct file *file, const char __user *buf,
1531                                 size_t count, loff_t *offset)
1532 {
1533         struct inode *inode = file_inode(file);
1534         struct task_struct *p;
1535         char buffer[TASK_COMM_LEN];
1536         const size_t maxlen = sizeof(buffer) - 1;
1537
1538         memset(buffer, 0, sizeof(buffer));
1539         if (copy_from_user(buffer, buf, count > maxlen ? maxlen : count))
1540                 return -EFAULT;
1541
1542         p = get_proc_task(inode);
1543         if (!p)
1544                 return -ESRCH;
1545
1546         if (same_thread_group(current, p))
1547                 set_task_comm(p, buffer);
1548         else
1549                 count = -EINVAL;
1550
1551         put_task_struct(p);
1552
1553         return count;
1554 }
1555
1556 static int comm_show(struct seq_file *m, void *v)
1557 {
1558         struct inode *inode = m->private;
1559         struct task_struct *p;
1560
1561         p = get_proc_task(inode);
1562         if (!p)
1563                 return -ESRCH;
1564
1565         task_lock(p);
1566         seq_printf(m, "%s\n", p->comm);
1567         task_unlock(p);
1568
1569         put_task_struct(p);
1570
1571         return 0;
1572 }
1573
1574 static int comm_open(struct inode *inode, struct file *filp)
1575 {
1576         return single_open(filp, comm_show, inode);
1577 }
1578
1579 static const struct file_operations proc_pid_set_comm_operations = {
1580         .open           = comm_open,
1581         .read           = seq_read,
1582         .write          = comm_write,
1583         .llseek         = seq_lseek,
1584         .release        = single_release,
1585 };
1586
1587 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1588 {
1589         struct task_struct *task;
1590         struct file *exe_file;
1591
1592         task = get_proc_task(d_inode(dentry));
1593         if (!task)
1594                 return -ENOENT;
1595         exe_file = get_task_exe_file(task);
1596         put_task_struct(task);
1597         if (exe_file) {
1598                 *exe_path = exe_file->f_path;
1599                 path_get(&exe_file->f_path);
1600                 fput(exe_file);
1601                 return 0;
1602         } else
1603                 return -ENOENT;
1604 }
1605
1606 static const char *proc_pid_get_link(struct dentry *dentry,
1607                                      struct inode *inode,
1608                                      struct delayed_call *done)
1609 {
1610         struct path path;
1611         int error = -EACCES;
1612
1613         if (!dentry)
1614                 return ERR_PTR(-ECHILD);
1615
1616         /* Are we allowed to snoop on the tasks file descriptors? */
1617         if (!proc_fd_access_allowed(inode))
1618                 goto out;
1619
1620         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1621         if (error)
1622                 goto out;
1623
1624         nd_jump_link(&path);
1625         return NULL;
1626 out:
1627         return ERR_PTR(error);
1628 }
1629
1630 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1631 {
1632         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1633         char *pathname;
1634         int len;
1635
1636         if (!tmp)
1637                 return -ENOMEM;
1638
1639         pathname = d_path(path, tmp, PAGE_SIZE);
1640         len = PTR_ERR(pathname);
1641         if (IS_ERR(pathname))
1642                 goto out;
1643         len = tmp + PAGE_SIZE - 1 - pathname;
1644
1645         if (len > buflen)
1646                 len = buflen;
1647         if (copy_to_user(buffer, pathname, len))
1648                 len = -EFAULT;
1649  out:
1650         free_page((unsigned long)tmp);
1651         return len;
1652 }
1653
1654 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1655 {
1656         int error = -EACCES;
1657         struct inode *inode = d_inode(dentry);
1658         struct path path;
1659
1660         /* Are we allowed to snoop on the tasks file descriptors? */
1661         if (!proc_fd_access_allowed(inode))
1662                 goto out;
1663
1664         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1665         if (error)
1666                 goto out;
1667
1668         error = do_proc_readlink(&path, buffer, buflen);
1669         path_put(&path);
1670 out:
1671         return error;
1672 }
1673
1674 const struct inode_operations proc_pid_link_inode_operations = {
1675         .readlink       = proc_pid_readlink,
1676         .get_link       = proc_pid_get_link,
1677         .setattr        = proc_setattr,
1678 };
1679
1680
1681 /* building an inode */
1682
1683 void task_dump_owner(struct task_struct *task, mode_t mode,
1684                      kuid_t *ruid, kgid_t *rgid)
1685 {
1686         /* Depending on the state of dumpable compute who should own a
1687          * proc file for a task.
1688          */
1689         const struct cred *cred;
1690         kuid_t uid;
1691         kgid_t gid;
1692
1693         /* Default to the tasks effective ownership */
1694         rcu_read_lock();
1695         cred = __task_cred(task);
1696         uid = cred->euid;
1697         gid = cred->egid;
1698         rcu_read_unlock();
1699
1700         /*
1701          * Before the /proc/pid/status file was created the only way to read
1702          * the effective uid of a /process was to stat /proc/pid.  Reading
1703          * /proc/pid/status is slow enough that procps and other packages
1704          * kept stating /proc/pid.  To keep the rules in /proc simple I have
1705          * made this apply to all per process world readable and executable
1706          * directories.
1707          */
1708         if (mode != (S_IFDIR|S_IRUGO|S_IXUGO)) {
1709                 struct mm_struct *mm;
1710                 task_lock(task);
1711                 mm = task->mm;
1712                 /* Make non-dumpable tasks owned by some root */
1713                 if (mm) {
1714                         if (get_dumpable(mm) != SUID_DUMP_USER) {
1715                                 struct user_namespace *user_ns = mm->user_ns;
1716
1717                                 uid = make_kuid(user_ns, 0);
1718                                 if (!uid_valid(uid))
1719                                         uid = GLOBAL_ROOT_UID;
1720
1721                                 gid = make_kgid(user_ns, 0);
1722                                 if (!gid_valid(gid))
1723                                         gid = GLOBAL_ROOT_GID;
1724                         }
1725                 } else {
1726                         uid = GLOBAL_ROOT_UID;
1727                         gid = GLOBAL_ROOT_GID;
1728                 }
1729                 task_unlock(task);
1730         }
1731         *ruid = uid;
1732         *rgid = gid;
1733 }
1734
1735 struct inode *proc_pid_make_inode(struct super_block * sb,
1736                                   struct task_struct *task, umode_t mode)
1737 {
1738         struct inode * inode;
1739         struct proc_inode *ei;
1740
1741         /* We need a new inode */
1742
1743         inode = new_inode(sb);
1744         if (!inode)
1745                 goto out;
1746
1747         /* Common stuff */
1748         ei = PROC_I(inode);
1749         inode->i_mode = mode;
1750         inode->i_ino = get_next_ino();
1751         inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode);
1752         inode->i_op = &proc_def_inode_operations;
1753
1754         /*
1755          * grab the reference to task.
1756          */
1757         ei->pid = get_task_pid(task, PIDTYPE_PID);
1758         if (!ei->pid)
1759                 goto out_unlock;
1760
1761         task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1762         security_task_to_inode(task, inode);
1763
1764 out:
1765         return inode;
1766
1767 out_unlock:
1768         iput(inode);
1769         return NULL;
1770 }
1771
1772 int pid_getattr(const struct path *path, struct kstat *stat,
1773                 u32 request_mask, unsigned int query_flags)
1774 {
1775         struct inode *inode = d_inode(path->dentry);
1776         struct task_struct *task;
1777         struct pid_namespace *pid = path->dentry->d_sb->s_fs_info;
1778
1779         generic_fillattr(inode, stat);
1780
1781         rcu_read_lock();
1782         stat->uid = GLOBAL_ROOT_UID;
1783         stat->gid = GLOBAL_ROOT_GID;
1784         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1785         if (task) {
1786                 if (!has_pid_permissions(pid, task, HIDEPID_INVISIBLE)) {
1787                         rcu_read_unlock();
1788                         /*
1789                          * This doesn't prevent learning whether PID exists,
1790                          * it only makes getattr() consistent with readdir().
1791                          */
1792                         return -ENOENT;
1793                 }
1794                 task_dump_owner(task, inode->i_mode, &stat->uid, &stat->gid);
1795         }
1796         rcu_read_unlock();
1797         return 0;
1798 }
1799
1800 /* dentry stuff */
1801
1802 /*
1803  *      Exceptional case: normally we are not allowed to unhash a busy
1804  * directory. In this case, however, we can do it - no aliasing problems
1805  * due to the way we treat inodes.
1806  *
1807  * Rewrite the inode's ownerships here because the owning task may have
1808  * performed a setuid(), etc.
1809  *
1810  */
1811 int pid_revalidate(struct dentry *dentry, unsigned int flags)
1812 {
1813         struct inode *inode;
1814         struct task_struct *task;
1815
1816         if (flags & LOOKUP_RCU)
1817                 return -ECHILD;
1818
1819         inode = d_inode(dentry);
1820         task = get_proc_task(inode);
1821
1822         if (task) {
1823                 task_dump_owner(task, inode->i_mode, &inode->i_uid, &inode->i_gid);
1824
1825                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1826                 security_task_to_inode(task, inode);
1827                 put_task_struct(task);
1828                 return 1;
1829         }
1830         return 0;
1831 }
1832
1833 static inline bool proc_inode_is_dead(struct inode *inode)
1834 {
1835         return !proc_pid(inode)->tasks[PIDTYPE_PID].first;
1836 }
1837
1838 int pid_delete_dentry(const struct dentry *dentry)
1839 {
1840         /* Is the task we represent dead?
1841          * If so, then don't put the dentry on the lru list,
1842          * kill it immediately.
1843          */
1844         return proc_inode_is_dead(d_inode(dentry));
1845 }
1846
1847 const struct dentry_operations pid_dentry_operations =
1848 {
1849         .d_revalidate   = pid_revalidate,
1850         .d_delete       = pid_delete_dentry,
1851 };
1852
1853 /* Lookups */
1854
1855 /*
1856  * Fill a directory entry.
1857  *
1858  * If possible create the dcache entry and derive our inode number and
1859  * file type from dcache entry.
1860  *
1861  * Since all of the proc inode numbers are dynamically generated, the inode
1862  * numbers do not exist until the inode is cache.  This means creating the
1863  * the dcache entry in readdir is necessary to keep the inode numbers
1864  * reported by readdir in sync with the inode numbers reported
1865  * by stat.
1866  */
1867 bool proc_fill_cache(struct file *file, struct dir_context *ctx,
1868         const char *name, int len,
1869         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1870 {
1871         struct dentry *child, *dir = file->f_path.dentry;
1872         struct qstr qname = QSTR_INIT(name, len);
1873         struct inode *inode;
1874         unsigned type;
1875         ino_t ino;
1876
1877         child = d_hash_and_lookup(dir, &qname);
1878         if (!child) {
1879                 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1880                 child = d_alloc_parallel(dir, &qname, &wq);
1881                 if (IS_ERR(child))
1882                         goto end_instantiate;
1883                 if (d_in_lookup(child)) {
1884                         int err = instantiate(d_inode(dir), child, task, ptr);
1885                         d_lookup_done(child);
1886                         if (err < 0) {
1887                                 dput(child);
1888                                 goto end_instantiate;
1889                         }
1890                 }
1891         }
1892         inode = d_inode(child);
1893         ino = inode->i_ino;
1894         type = inode->i_mode >> 12;
1895         dput(child);
1896         return dir_emit(ctx, name, len, ino, type);
1897
1898 end_instantiate:
1899         return dir_emit(ctx, name, len, 1, DT_UNKNOWN);
1900 }
1901
1902 /*
1903  * dname_to_vma_addr - maps a dentry name into two unsigned longs
1904  * which represent vma start and end addresses.
1905  */
1906 static int dname_to_vma_addr(struct dentry *dentry,
1907                              unsigned long *start, unsigned long *end)
1908 {
1909         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
1910                 return -EINVAL;
1911
1912         return 0;
1913 }
1914
1915 static int map_files_d_revalidate(struct dentry *dentry, unsigned int flags)
1916 {
1917         unsigned long vm_start, vm_end;
1918         bool exact_vma_exists = false;
1919         struct mm_struct *mm = NULL;
1920         struct task_struct *task;
1921         struct inode *inode;
1922         int status = 0;
1923
1924         if (flags & LOOKUP_RCU)
1925                 return -ECHILD;
1926
1927         inode = d_inode(dentry);
1928         task = get_proc_task(inode);
1929         if (!task)
1930                 goto out_notask;
1931
1932         mm = mm_access(task, PTRACE_MODE_READ_FSCREDS);
1933         if (IS_ERR_OR_NULL(mm))
1934                 goto out;
1935
1936         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
1937                 down_read(&mm->mmap_sem);
1938                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
1939                 up_read(&mm->mmap_sem);
1940         }
1941
1942         mmput(mm);
1943
1944         if (exact_vma_exists) {
1945                 task_dump_owner(task, 0, &inode->i_uid, &inode->i_gid);
1946
1947                 security_task_to_inode(task, inode);
1948                 status = 1;
1949         }
1950
1951 out:
1952         put_task_struct(task);
1953
1954 out_notask:
1955         return status;
1956 }
1957
1958 static const struct dentry_operations tid_map_files_dentry_operations = {
1959         .d_revalidate   = map_files_d_revalidate,
1960         .d_delete       = pid_delete_dentry,
1961 };
1962
1963 static int map_files_get_link(struct dentry *dentry, struct path *path)
1964 {
1965         unsigned long vm_start, vm_end;
1966         struct vm_area_struct *vma;
1967         struct task_struct *task;
1968         struct mm_struct *mm;
1969         int rc;
1970
1971         rc = -ENOENT;
1972         task = get_proc_task(d_inode(dentry));
1973         if (!task)
1974                 goto out;
1975
1976         mm = get_task_mm(task);
1977         put_task_struct(task);
1978         if (!mm)
1979                 goto out;
1980
1981         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
1982         if (rc)
1983                 goto out_mmput;
1984
1985         rc = -ENOENT;
1986         down_read(&mm->mmap_sem);
1987         vma = find_exact_vma(mm, vm_start, vm_end);
1988         if (vma && vma->vm_file) {
1989                 *path = vma->vm_file->f_path;
1990                 path_get(path);
1991                 rc = 0;
1992         }
1993         up_read(&mm->mmap_sem);
1994
1995 out_mmput:
1996         mmput(mm);
1997 out:
1998         return rc;
1999 }
2000
2001 struct map_files_info {
2002         fmode_t         mode;
2003         unsigned int    len;
2004         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2005 };
2006
2007 /*
2008  * Only allow CAP_SYS_ADMIN to follow the links, due to concerns about how the
2009  * symlinks may be used to bypass permissions on ancestor directories in the
2010  * path to the file in question.
2011  */
2012 static const char *
2013 proc_map_files_get_link(struct dentry *dentry,
2014                         struct inode *inode,
2015                         struct delayed_call *done)
2016 {
2017         if (!capable(CAP_SYS_ADMIN))
2018                 return ERR_PTR(-EPERM);
2019
2020         return proc_pid_get_link(dentry, inode, done);
2021 }
2022
2023 /*
2024  * Identical to proc_pid_link_inode_operations except for get_link()
2025  */
2026 static const struct inode_operations proc_map_files_link_inode_operations = {
2027         .readlink       = proc_pid_readlink,
2028         .get_link       = proc_map_files_get_link,
2029         .setattr        = proc_setattr,
2030 };
2031
2032 static int
2033 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2034                            struct task_struct *task, const void *ptr)
2035 {
2036         fmode_t mode = (fmode_t)(unsigned long)ptr;
2037         struct proc_inode *ei;
2038         struct inode *inode;
2039
2040         inode = proc_pid_make_inode(dir->i_sb, task, S_IFLNK |
2041                                     ((mode & FMODE_READ ) ? S_IRUSR : 0) |
2042                                     ((mode & FMODE_WRITE) ? S_IWUSR : 0));
2043         if (!inode)
2044                 return -ENOENT;
2045
2046         ei = PROC_I(inode);
2047         ei->op.proc_get_link = map_files_get_link;
2048
2049         inode->i_op = &proc_map_files_link_inode_operations;
2050         inode->i_size = 64;
2051
2052         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2053         d_add(dentry, inode);
2054
2055         return 0;
2056 }
2057
2058 static struct dentry *proc_map_files_lookup(struct inode *dir,
2059                 struct dentry *dentry, unsigned int flags)
2060 {
2061         unsigned long vm_start, vm_end;
2062         struct vm_area_struct *vma;
2063         struct task_struct *task;
2064         int result;
2065         struct mm_struct *mm;
2066
2067         result = -ENOENT;
2068         task = get_proc_task(dir);
2069         if (!task)
2070                 goto out;
2071
2072         result = -EACCES;
2073         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2074                 goto out_put_task;
2075
2076         result = -ENOENT;
2077         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2078                 goto out_put_task;
2079
2080         mm = get_task_mm(task);
2081         if (!mm)
2082                 goto out_put_task;
2083
2084         down_read(&mm->mmap_sem);
2085         vma = find_exact_vma(mm, vm_start, vm_end);
2086         if (!vma)
2087                 goto out_no_vma;
2088
2089         if (vma->vm_file)
2090                 result = proc_map_files_instantiate(dir, dentry, task,
2091                                 (void *)(unsigned long)vma->vm_file->f_mode);
2092
2093 out_no_vma:
2094         up_read(&mm->mmap_sem);
2095         mmput(mm);
2096 out_put_task:
2097         put_task_struct(task);
2098 out:
2099         return ERR_PTR(result);
2100 }
2101
2102 static const struct inode_operations proc_map_files_inode_operations = {
2103         .lookup         = proc_map_files_lookup,
2104         .permission     = proc_fd_permission,
2105         .setattr        = proc_setattr,
2106 };
2107
2108 static int
2109 proc_map_files_readdir(struct file *file, struct dir_context *ctx)
2110 {
2111         struct vm_area_struct *vma;
2112         struct task_struct *task;
2113         struct mm_struct *mm;
2114         unsigned long nr_files, pos, i;
2115         struct flex_array *fa = NULL;
2116         struct map_files_info info;
2117         struct map_files_info *p;
2118         int ret;
2119
2120         ret = -ENOENT;
2121         task = get_proc_task(file_inode(file));
2122         if (!task)
2123                 goto out;
2124
2125         ret = -EACCES;
2126         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS))
2127                 goto out_put_task;
2128
2129         ret = 0;
2130         if (!dir_emit_dots(file, ctx))
2131                 goto out_put_task;
2132
2133         mm = get_task_mm(task);
2134         if (!mm)
2135                 goto out_put_task;
2136         down_read(&mm->mmap_sem);
2137
2138         nr_files = 0;
2139
2140         /*
2141          * We need two passes here:
2142          *
2143          *  1) Collect vmas of mapped files with mmap_sem taken
2144          *  2) Release mmap_sem and instantiate entries
2145          *
2146          * otherwise we get lockdep complained, since filldir()
2147          * routine might require mmap_sem taken in might_fault().
2148          */
2149
2150         for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2151                 if (vma->vm_file && ++pos > ctx->pos)
2152                         nr_files++;
2153         }
2154
2155         if (nr_files) {
2156                 fa = flex_array_alloc(sizeof(info), nr_files,
2157                                         GFP_KERNEL);
2158                 if (!fa || flex_array_prealloc(fa, 0, nr_files,
2159                                                 GFP_KERNEL)) {
2160                         ret = -ENOMEM;
2161                         if (fa)
2162                                 flex_array_free(fa);
2163                         up_read(&mm->mmap_sem);
2164                         mmput(mm);
2165                         goto out_put_task;
2166                 }
2167                 for (i = 0, vma = mm->mmap, pos = 2; vma;
2168                                 vma = vma->vm_next) {
2169                         if (!vma->vm_file)
2170                                 continue;
2171                         if (++pos <= ctx->pos)
2172                                 continue;
2173
2174                         info.mode = vma->vm_file->f_mode;
2175                         info.len = snprintf(info.name,
2176                                         sizeof(info.name), "%lx-%lx",
2177                                         vma->vm_start, vma->vm_end);
2178                         if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2179                                 BUG();
2180                 }
2181         }
2182         up_read(&mm->mmap_sem);
2183
2184         for (i = 0; i < nr_files; i++) {
2185                 p = flex_array_get(fa, i);
2186                 if (!proc_fill_cache(file, ctx,
2187                                       p->name, p->len,
2188                                       proc_map_files_instantiate,
2189                                       task,
2190                                       (void *)(unsigned long)p->mode))
2191                         break;
2192                 ctx->pos++;
2193         }
2194         if (fa)
2195                 flex_array_free(fa);
2196         mmput(mm);
2197
2198 out_put_task:
2199         put_task_struct(task);
2200 out:
2201         return ret;
2202 }
2203
2204 static const struct file_operations proc_map_files_operations = {
2205         .read           = generic_read_dir,
2206         .iterate_shared = proc_map_files_readdir,
2207         .llseek         = generic_file_llseek,
2208 };
2209
2210 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2211 struct timers_private {
2212         struct pid *pid;
2213         struct task_struct *task;
2214         struct sighand_struct *sighand;
2215         struct pid_namespace *ns;
2216         unsigned long flags;
2217 };
2218
2219 static void *timers_start(struct seq_file *m, loff_t *pos)
2220 {
2221         struct timers_private *tp = m->private;
2222
2223         tp->task = get_pid_task(tp->pid, PIDTYPE_PID);
2224         if (!tp->task)
2225                 return ERR_PTR(-ESRCH);
2226
2227         tp->sighand = lock_task_sighand(tp->task, &tp->flags);
2228         if (!tp->sighand)
2229                 return ERR_PTR(-ESRCH);
2230
2231         return seq_list_start(&tp->task->signal->posix_timers, *pos);
2232 }
2233
2234 static void *timers_next(struct seq_file *m, void *v, loff_t *pos)
2235 {
2236         struct timers_private *tp = m->private;
2237         return seq_list_next(v, &tp->task->signal->posix_timers, pos);
2238 }
2239
2240 static void timers_stop(struct seq_file *m, void *v)
2241 {
2242         struct timers_private *tp = m->private;
2243
2244         if (tp->sighand) {
2245                 unlock_task_sighand(tp->task, &tp->flags);
2246                 tp->sighand = NULL;
2247         }
2248
2249         if (tp->task) {
2250                 put_task_struct(tp->task);
2251                 tp->task = NULL;
2252         }
2253 }
2254
2255 static int show_timer(struct seq_file *m, void *v)
2256 {
2257         struct k_itimer *timer;
2258         struct timers_private *tp = m->private;
2259         int notify;
2260         static const char * const nstr[] = {
2261                 [SIGEV_SIGNAL] = "signal",
2262                 [SIGEV_NONE] = "none",
2263                 [SIGEV_THREAD] = "thread",
2264         };
2265
2266         timer = list_entry((struct list_head *)v, struct k_itimer, list);
2267         notify = timer->it_sigev_notify;
2268
2269         seq_printf(m, "ID: %d\n", timer->it_id);
2270         seq_printf(m, "signal: %d/%p\n",
2271                    timer->sigq->info.si_signo,
2272                    timer->sigq->info.si_value.sival_ptr);
2273         seq_printf(m, "notify: %s/%s.%d\n",
2274                    nstr[notify & ~SIGEV_THREAD_ID],
2275                    (notify & SIGEV_THREAD_ID) ? "tid" : "pid",
2276                    pid_nr_ns(timer->it_pid, tp->ns));
2277         seq_printf(m, "ClockID: %d\n", timer->it_clock);
2278
2279         return 0;
2280 }
2281
2282 static const struct seq_operations proc_timers_seq_ops = {
2283         .start  = timers_start,
2284         .next   = timers_next,
2285         .stop   = timers_stop,
2286         .show   = show_timer,
2287 };
2288
2289 static int proc_timers_open(struct inode *inode, struct file *file)
2290 {
2291         struct timers_private *tp;
2292
2293         tp = __seq_open_private(file, &proc_timers_seq_ops,
2294                         sizeof(struct timers_private));
2295         if (!tp)
2296                 return -ENOMEM;
2297
2298         tp->pid = proc_pid(inode);
2299         tp->ns = inode->i_sb->s_fs_info;
2300         return 0;
2301 }
2302
2303 static const struct file_operations proc_timers_operations = {
2304         .open           = proc_timers_open,
2305         .read           = seq_read,
2306         .llseek         = seq_lseek,
2307         .release        = seq_release_private,
2308 };
2309 #endif
2310
2311 static ssize_t timerslack_ns_write(struct file *file, const char __user *buf,
2312                                         size_t count, loff_t *offset)
2313 {
2314         struct inode *inode = file_inode(file);
2315         struct task_struct *p;
2316         u64 slack_ns;
2317         int err;
2318
2319         err = kstrtoull_from_user(buf, count, 10, &slack_ns);
2320         if (err < 0)
2321                 return err;
2322
2323         p = get_proc_task(inode);
2324         if (!p)
2325                 return -ESRCH;
2326
2327         if (p != current) {
2328                 if (!capable(CAP_SYS_NICE)) {
2329                         count = -EPERM;
2330                         goto out;
2331                 }
2332
2333                 err = security_task_setscheduler(p);
2334                 if (err) {
2335                         count = err;
2336                         goto out;
2337                 }
2338         }
2339
2340         task_lock(p);
2341         if (slack_ns == 0)
2342                 p->timer_slack_ns = p->default_timer_slack_ns;
2343         else
2344                 p->timer_slack_ns = slack_ns;
2345         task_unlock(p);
2346
2347 out:
2348         put_task_struct(p);
2349
2350         return count;
2351 }
2352
2353 static int timerslack_ns_show(struct seq_file *m, void *v)
2354 {
2355         struct inode *inode = m->private;
2356         struct task_struct *p;
2357         int err = 0;
2358
2359         p = get_proc_task(inode);
2360         if (!p)
2361                 return -ESRCH;
2362
2363         if (p != current) {
2364
2365                 if (!capable(CAP_SYS_NICE)) {
2366                         err = -EPERM;
2367                         goto out;
2368                 }
2369                 err = security_task_getscheduler(p);
2370                 if (err)
2371                         goto out;
2372         }
2373
2374         task_lock(p);
2375         seq_printf(m, "%llu\n", p->timer_slack_ns);
2376         task_unlock(p);
2377
2378 out:
2379         put_task_struct(p);
2380
2381         return err;
2382 }
2383
2384 static int timerslack_ns_open(struct inode *inode, struct file *filp)
2385 {
2386         return single_open(filp, timerslack_ns_show, inode);
2387 }
2388
2389 static const struct file_operations proc_pid_set_timerslack_ns_operations = {
2390         .open           = timerslack_ns_open,
2391         .read           = seq_read,
2392         .write          = timerslack_ns_write,
2393         .llseek         = seq_lseek,
2394         .release        = single_release,
2395 };
2396
2397 static int proc_pident_instantiate(struct inode *dir,
2398         struct dentry *dentry, struct task_struct *task, const void *ptr)
2399 {
2400         const struct pid_entry *p = ptr;
2401         struct inode *inode;
2402         struct proc_inode *ei;
2403
2404         inode = proc_pid_make_inode(dir->i_sb, task, p->mode);
2405         if (!inode)
2406                 goto out;
2407
2408         ei = PROC_I(inode);
2409         if (S_ISDIR(inode->i_mode))
2410                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2411         if (p->iop)
2412                 inode->i_op = p->iop;
2413         if (p->fop)
2414                 inode->i_fop = p->fop;
2415         ei->op = p->op;
2416         d_set_d_op(dentry, &pid_dentry_operations);
2417         d_add(dentry, inode);
2418         /* Close the race of the process dying before we return the dentry */
2419         if (pid_revalidate(dentry, 0))
2420                 return 0;
2421 out:
2422         return -ENOENT;
2423 }
2424
2425 static struct dentry *proc_pident_lookup(struct inode *dir,
2426                                          struct dentry *dentry,
2427                                          const struct pid_entry *ents,
2428                                          unsigned int nents)
2429 {
2430         int error;
2431         struct task_struct *task = get_proc_task(dir);
2432         const struct pid_entry *p, *last;
2433
2434         error = -ENOENT;
2435
2436         if (!task)
2437                 goto out_no_task;
2438
2439         /*
2440          * Yes, it does not scale. And it should not. Don't add
2441          * new entries into /proc/<tgid>/ without very good reasons.
2442          */
2443         last = &ents[nents];
2444         for (p = ents; p < last; p++) {
2445                 if (p->len != dentry->d_name.len)
2446                         continue;
2447                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2448                         break;
2449         }
2450         if (p >= last)
2451                 goto out;
2452
2453         error = proc_pident_instantiate(dir, dentry, task, p);
2454 out:
2455         put_task_struct(task);
2456 out_no_task:
2457         return ERR_PTR(error);
2458 }
2459
2460 static int proc_pident_readdir(struct file *file, struct dir_context *ctx,
2461                 const struct pid_entry *ents, unsigned int nents)
2462 {
2463         struct task_struct *task = get_proc_task(file_inode(file));
2464         const struct pid_entry *p;
2465
2466         if (!task)
2467                 return -ENOENT;
2468
2469         if (!dir_emit_dots(file, ctx))
2470                 goto out;
2471
2472         if (ctx->pos >= nents + 2)
2473                 goto out;
2474
2475         for (p = ents + (ctx->pos - 2); p < ents + nents; p++) {
2476                 if (!proc_fill_cache(file, ctx, p->name, p->len,
2477                                 proc_pident_instantiate, task, p))
2478                         break;
2479                 ctx->pos++;
2480         }
2481 out:
2482         put_task_struct(task);
2483         return 0;
2484 }
2485
2486 #ifdef CONFIG_SECURITY
2487 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2488                                   size_t count, loff_t *ppos)
2489 {
2490         struct inode * inode = file_inode(file);
2491         char *p = NULL;
2492         ssize_t length;
2493         struct task_struct *task = get_proc_task(inode);
2494
2495         if (!task)
2496                 return -ESRCH;
2497
2498         length = security_getprocattr(task,
2499                                       (char*)file->f_path.dentry->d_name.name,
2500                                       &p);
2501         put_task_struct(task);
2502         if (length > 0)
2503                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2504         kfree(p);
2505         return length;
2506 }
2507
2508 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2509                                    size_t count, loff_t *ppos)
2510 {
2511         struct inode * inode = file_inode(file);
2512         void *page;
2513         ssize_t length;
2514         struct task_struct *task = get_proc_task(inode);
2515
2516         length = -ESRCH;
2517         if (!task)
2518                 goto out_no_task;
2519
2520         /* A task may only write its own attributes. */
2521         length = -EACCES;
2522         if (current != task)
2523                 goto out;
2524
2525         if (count > PAGE_SIZE)
2526                 count = PAGE_SIZE;
2527
2528         /* No partial writes. */
2529         length = -EINVAL;
2530         if (*ppos != 0)
2531                 goto out;
2532
2533         page = memdup_user(buf, count);
2534         if (IS_ERR(page)) {
2535                 length = PTR_ERR(page);
2536                 goto out;
2537         }
2538
2539         /* Guard against adverse ptrace interaction */
2540         length = mutex_lock_interruptible(&current->signal->cred_guard_mutex);
2541         if (length < 0)
2542                 goto out_free;
2543
2544         length = security_setprocattr(file->f_path.dentry->d_name.name,
2545                                       page, count);
2546         mutex_unlock(&current->signal->cred_guard_mutex);
2547 out_free:
2548         kfree(page);
2549 out:
2550         put_task_struct(task);
2551 out_no_task:
2552         return length;
2553 }
2554
2555 static const struct file_operations proc_pid_attr_operations = {
2556         .read           = proc_pid_attr_read,
2557         .write          = proc_pid_attr_write,
2558         .llseek         = generic_file_llseek,
2559 };
2560
2561 static const struct pid_entry attr_dir_stuff[] = {
2562         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2563         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2564         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2565         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2566         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2567         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2568 };
2569
2570 static int proc_attr_dir_readdir(struct file *file, struct dir_context *ctx)
2571 {
2572         return proc_pident_readdir(file, ctx,
2573                                    attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2574 }
2575
2576 static const struct file_operations proc_attr_dir_operations = {
2577         .read           = generic_read_dir,
2578         .iterate_shared = proc_attr_dir_readdir,
2579         .llseek         = generic_file_llseek,
2580 };
2581
2582 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2583                                 struct dentry *dentry, unsigned int flags)
2584 {
2585         return proc_pident_lookup(dir, dentry,
2586                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2587 }
2588
2589 static const struct inode_operations proc_attr_dir_inode_operations = {
2590         .lookup         = proc_attr_dir_lookup,
2591         .getattr        = pid_getattr,
2592         .setattr        = proc_setattr,
2593 };
2594
2595 #endif
2596
2597 #ifdef CONFIG_ELF_CORE
2598 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2599                                          size_t count, loff_t *ppos)
2600 {
2601         struct task_struct *task = get_proc_task(file_inode(file));
2602         struct mm_struct *mm;
2603         char buffer[PROC_NUMBUF];
2604         size_t len;
2605         int ret;
2606
2607         if (!task)
2608                 return -ESRCH;
2609
2610         ret = 0;
2611         mm = get_task_mm(task);
2612         if (mm) {
2613                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2614                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2615                                 MMF_DUMP_FILTER_SHIFT));
2616                 mmput(mm);
2617                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2618         }
2619
2620         put_task_struct(task);
2621
2622         return ret;
2623 }
2624
2625 static ssize_t proc_coredump_filter_write(struct file *file,
2626                                           const char __user *buf,
2627                                           size_t count,
2628                                           loff_t *ppos)
2629 {
2630         struct task_struct *task;
2631         struct mm_struct *mm;
2632         unsigned int val;
2633         int ret;
2634         int i;
2635         unsigned long mask;
2636
2637         ret = kstrtouint_from_user(buf, count, 0, &val);
2638         if (ret < 0)
2639                 return ret;
2640
2641         ret = -ESRCH;
2642         task = get_proc_task(file_inode(file));
2643         if (!task)
2644                 goto out_no_task;
2645
2646         mm = get_task_mm(task);
2647         if (!mm)
2648                 goto out_no_mm;
2649         ret = 0;
2650
2651         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2652                 if (val & mask)
2653                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2654                 else
2655                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2656         }
2657
2658         mmput(mm);
2659  out_no_mm:
2660         put_task_struct(task);
2661  out_no_task:
2662         if (ret < 0)
2663                 return ret;
2664         return count;
2665 }
2666
2667 static const struct file_operations proc_coredump_filter_operations = {
2668         .read           = proc_coredump_filter_read,
2669         .write          = proc_coredump_filter_write,
2670         .llseek         = generic_file_llseek,
2671 };
2672 #endif
2673
2674 #ifdef CONFIG_TASK_IO_ACCOUNTING
2675 static int do_io_accounting(struct task_struct *task, struct seq_file *m, int whole)
2676 {
2677         struct task_io_accounting acct = task->ioac;
2678         unsigned long flags;
2679         int result;
2680
2681         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2682         if (result)
2683                 return result;
2684
2685         if (!ptrace_may_access(task, PTRACE_MODE_READ_FSCREDS)) {
2686                 result = -EACCES;
2687                 goto out_unlock;
2688         }
2689
2690         if (whole && lock_task_sighand(task, &flags)) {
2691                 struct task_struct *t = task;
2692
2693                 task_io_accounting_add(&acct, &task->signal->ioac);
2694                 while_each_thread(task, t)
2695                         task_io_accounting_add(&acct, &t->ioac);
2696
2697                 unlock_task_sighand(task, &flags);
2698         }
2699         seq_printf(m,
2700                    "rchar: %llu\n"
2701                    "wchar: %llu\n"
2702                    "syscr: %llu\n"
2703                    "syscw: %llu\n"
2704                    "read_bytes: %llu\n"
2705                    "write_bytes: %llu\n"
2706                    "cancelled_write_bytes: %llu\n",
2707                    (unsigned long long)acct.rchar,
2708                    (unsigned long long)acct.wchar,
2709                    (unsigned long long)acct.syscr,
2710                    (unsigned long long)acct.syscw,
2711                    (unsigned long long)acct.read_bytes,
2712                    (unsigned long long)acct.write_bytes,
2713                    (unsigned long long)acct.cancelled_write_bytes);
2714         result = 0;
2715
2716 out_unlock:
2717         mutex_unlock(&task->signal->cred_guard_mutex);
2718         return result;
2719 }
2720
2721 static int proc_tid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2722                                   struct pid *pid, struct task_struct *task)
2723 {
2724         return do_io_accounting(task, m, 0);
2725 }
2726
2727 static int proc_tgid_io_accounting(struct seq_file *m, struct pid_namespace *ns,
2728                                    struct pid *pid, struct task_struct *task)
2729 {
2730         return do_io_accounting(task, m, 1);
2731 }
2732 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2733
2734 #ifdef CONFIG_USER_NS
2735 static int proc_id_map_open(struct inode *inode, struct file *file,
2736         const struct seq_operations *seq_ops)
2737 {
2738         struct user_namespace *ns = NULL;
2739         struct task_struct *task;
2740         struct seq_file *seq;
2741         int ret = -EINVAL;
2742
2743         task = get_proc_task(inode);
2744         if (task) {
2745                 rcu_read_lock();
2746                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2747                 rcu_read_unlock();
2748                 put_task_struct(task);
2749         }
2750         if (!ns)
2751                 goto err;
2752
2753         ret = seq_open(file, seq_ops);
2754         if (ret)
2755                 goto err_put_ns;
2756
2757         seq = file->private_data;
2758         seq->private = ns;
2759
2760         return 0;
2761 err_put_ns:
2762         put_user_ns(ns);
2763 err:
2764         return ret;
2765 }
2766
2767 static int proc_id_map_release(struct inode *inode, struct file *file)
2768 {
2769         struct seq_file *seq = file->private_data;
2770         struct user_namespace *ns = seq->private;
2771         put_user_ns(ns);
2772         return seq_release(inode, file);
2773 }
2774
2775 static int proc_uid_map_open(struct inode *inode, struct file *file)
2776 {
2777         return proc_id_map_open(inode, file, &proc_uid_seq_operations);
2778 }
2779
2780 static int proc_gid_map_open(struct inode *inode, struct file *file)
2781 {
2782         return proc_id_map_open(inode, file, &proc_gid_seq_operations);
2783 }
2784
2785 static int proc_projid_map_open(struct inode *inode, struct file *file)
2786 {
2787         return proc_id_map_open(inode, file, &proc_projid_seq_operations);
2788 }
2789
2790 static const struct file_operations proc_uid_map_operations = {
2791         .open           = proc_uid_map_open,
2792         .write          = proc_uid_map_write,
2793         .read           = seq_read,
2794         .llseek         = seq_lseek,
2795         .release        = proc_id_map_release,
2796 };
2797
2798 static const struct file_operations proc_gid_map_operations = {
2799         .open           = proc_gid_map_open,
2800         .write          = proc_gid_map_write,
2801         .read           = seq_read,
2802         .llseek         = seq_lseek,
2803         .release        = proc_id_map_release,
2804 };
2805
2806 static const struct file_operations proc_projid_map_operations = {
2807         .open           = proc_projid_map_open,
2808         .write          = proc_projid_map_write,
2809         .read           = seq_read,
2810         .llseek         = seq_lseek,
2811         .release        = proc_id_map_release,
2812 };
2813
2814 static int proc_setgroups_open(struct inode *inode, struct file *file)
2815 {
2816         struct user_namespace *ns = NULL;
2817         struct task_struct *task;
2818         int ret;
2819
2820         ret = -ESRCH;
2821         task = get_proc_task(inode);
2822         if (task) {
2823                 rcu_read_lock();
2824                 ns = get_user_ns(task_cred_xxx(task, user_ns));
2825                 rcu_read_unlock();
2826                 put_task_struct(task);
2827         }
2828         if (!ns)
2829                 goto err;
2830
2831         if (file->f_mode & FMODE_WRITE) {
2832                 ret = -EACCES;
2833                 if (!ns_capable(ns, CAP_SYS_ADMIN))
2834                         goto err_put_ns;
2835         }
2836
2837         ret = single_open(file, &proc_setgroups_show, ns);
2838         if (ret)
2839                 goto err_put_ns;
2840
2841         return 0;
2842 err_put_ns:
2843         put_user_ns(ns);
2844 err:
2845         return ret;
2846 }
2847
2848 static int proc_setgroups_release(struct inode *inode, struct file *file)
2849 {
2850         struct seq_file *seq = file->private_data;
2851         struct user_namespace *ns = seq->private;
2852         int ret = single_release(inode, file);
2853         put_user_ns(ns);
2854         return ret;
2855 }
2856
2857 static const struct file_operations proc_setgroups_operations = {
2858         .open           = proc_setgroups_open,
2859         .write          = proc_setgroups_write,
2860         .read           = seq_read,
2861         .llseek         = seq_lseek,
2862         .release        = proc_setgroups_release,
2863 };
2864 #endif /* CONFIG_USER_NS */
2865
2866 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2867                                 struct pid *pid, struct task_struct *task)
2868 {
2869         int err = lock_trace(task);
2870         if (!err) {
2871                 seq_printf(m, "%08x\n", task->personality);
2872                 unlock_trace(task);
2873         }
2874         return err;
2875 }
2876
2877 #ifdef CONFIG_LIVEPATCH
2878 static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns,
2879                                 struct pid *pid, struct task_struct *task)
2880 {
2881         seq_printf(m, "%d\n", task->patch_state);
2882         return 0;
2883 }
2884 #endif /* CONFIG_LIVEPATCH */
2885
2886 /*
2887  * Thread groups
2888  */
2889 static const struct file_operations proc_task_operations;
2890 static const struct inode_operations proc_task_inode_operations;
2891
2892 static const struct pid_entry tgid_base_stuff[] = {
2893         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2894         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2895         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
2896         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2897         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2898 #ifdef CONFIG_NET
2899         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2900 #endif
2901         REG("environ",    S_IRUSR, proc_environ_operations),
2902         REG("auxv",       S_IRUSR, proc_auxv_operations),
2903         ONE("status",     S_IRUGO, proc_pid_status),
2904         ONE("personality", S_IRUSR, proc_pid_personality),
2905         ONE("limits",     S_IRUGO, proc_pid_limits),
2906 #ifdef CONFIG_SCHED_DEBUG
2907         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2908 #endif
2909 #ifdef CONFIG_SCHED_AUTOGROUP
2910         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2911 #endif
2912         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2913 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2914         ONE("syscall",    S_IRUSR, proc_pid_syscall),
2915 #endif
2916         REG("cmdline",    S_IRUGO, proc_pid_cmdline_ops),
2917         ONE("stat",       S_IRUGO, proc_tgid_stat),
2918         ONE("statm",      S_IRUGO, proc_pid_statm),
2919         REG("maps",       S_IRUGO, proc_pid_maps_operations),
2920 #ifdef CONFIG_NUMA
2921         REG("numa_maps",  S_IRUGO, proc_pid_numa_maps_operations),
2922 #endif
2923         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2924         LNK("cwd",        proc_cwd_link),
2925         LNK("root",       proc_root_link),
2926         LNK("exe",        proc_exe_link),
2927         REG("mounts",     S_IRUGO, proc_mounts_operations),
2928         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2929         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2930 #ifdef CONFIG_PROC_PAGE_MONITOR
2931         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2932         REG("smaps",      S_IRUGO, proc_pid_smaps_operations),
2933         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2934 #endif
2935 #ifdef CONFIG_SECURITY
2936         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2937 #endif
2938 #ifdef CONFIG_KALLSYMS
2939         ONE("wchan",      S_IRUGO, proc_pid_wchan),
2940 #endif
2941 #ifdef CONFIG_STACKTRACE
2942         ONE("stack",      S_IRUSR, proc_pid_stack),
2943 #endif
2944 #ifdef CONFIG_SCHED_INFO
2945         ONE("schedstat",  S_IRUGO, proc_pid_schedstat),
2946 #endif
2947 #ifdef CONFIG_LATENCYTOP
2948         REG("latency",  S_IRUGO, proc_lstats_operations),
2949 #endif
2950 #ifdef CONFIG_PROC_PID_CPUSET
2951         ONE("cpuset",     S_IRUGO, proc_cpuset_show),
2952 #endif
2953 #ifdef CONFIG_CGROUPS
2954         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
2955 #endif
2956         ONE("oom_score",  S_IRUGO, proc_oom_score),
2957         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adj_operations),
2958         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2959 #ifdef CONFIG_AUDITSYSCALL
2960         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2961         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2962 #endif
2963 #ifdef CONFIG_FAULT_INJECTION
2964         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2965         REG("fail-nth", 0644, proc_fail_nth_operations),
2966 #endif
2967 #ifdef CONFIG_ELF_CORE
2968         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2969 #endif
2970 #ifdef CONFIG_TASK_IO_ACCOUNTING
2971         ONE("io",       S_IRUSR, proc_tgid_io_accounting),
2972 #endif
2973 #ifdef CONFIG_HARDWALL
2974         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
2975 #endif
2976 #ifdef CONFIG_USER_NS
2977         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
2978         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
2979         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
2980         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
2981 #endif
2982 #if defined(CONFIG_CHECKPOINT_RESTORE) && defined(CONFIG_POSIX_TIMERS)
2983         REG("timers",     S_IRUGO, proc_timers_operations),
2984 #endif
2985         REG("timerslack_ns", S_IRUGO|S_IWUGO, proc_pid_set_timerslack_ns_operations),
2986 #ifdef CONFIG_LIVEPATCH
2987         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
2988 #endif
2989 };
2990
2991 static int proc_tgid_base_readdir(struct file *file, struct dir_context *ctx)
2992 {
2993         return proc_pident_readdir(file, ctx,
2994                                    tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2995 }
2996
2997 static const struct file_operations proc_tgid_base_operations = {
2998         .read           = generic_read_dir,
2999         .iterate_shared = proc_tgid_base_readdir,
3000         .llseek         = generic_file_llseek,
3001 };
3002
3003 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3004 {
3005         return proc_pident_lookup(dir, dentry,
3006                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3007 }
3008
3009 static const struct inode_operations proc_tgid_base_inode_operations = {
3010         .lookup         = proc_tgid_base_lookup,
3011         .getattr        = pid_getattr,
3012         .setattr        = proc_setattr,
3013         .permission     = proc_pid_permission,
3014 };
3015
3016 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3017 {
3018         struct dentry *dentry, *leader, *dir;
3019         char buf[PROC_NUMBUF];
3020         struct qstr name;
3021
3022         name.name = buf;
3023         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3024         /* no ->d_hash() rejects on procfs */
3025         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3026         if (dentry) {
3027                 d_invalidate(dentry);
3028                 dput(dentry);
3029         }
3030
3031         if (pid == tgid)
3032                 return;
3033
3034         name.name = buf;
3035         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3036         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3037         if (!leader)
3038                 goto out;
3039
3040         name.name = "task";
3041         name.len = strlen(name.name);
3042         dir = d_hash_and_lookup(leader, &name);
3043         if (!dir)
3044                 goto out_put_leader;
3045
3046         name.name = buf;
3047         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3048         dentry = d_hash_and_lookup(dir, &name);
3049         if (dentry) {
3050                 d_invalidate(dentry);
3051                 dput(dentry);
3052         }
3053
3054         dput(dir);
3055 out_put_leader:
3056         dput(leader);
3057 out:
3058         return;
3059 }
3060
3061 /**
3062  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3063  * @task: task that should be flushed.
3064  *
3065  * When flushing dentries from proc, one needs to flush them from global
3066  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3067  * in. This call is supposed to do all of this job.
3068  *
3069  * Looks in the dcache for
3070  * /proc/@pid
3071  * /proc/@tgid/task/@pid
3072  * if either directory is present flushes it and all of it'ts children
3073  * from the dcache.
3074  *
3075  * It is safe and reasonable to cache /proc entries for a task until
3076  * that task exits.  After that they just clog up the dcache with
3077  * useless entries, possibly causing useful dcache entries to be
3078  * flushed instead.  This routine is proved to flush those useless
3079  * dcache entries at process exit time.
3080  *
3081  * NOTE: This routine is just an optimization so it does not guarantee
3082  *       that no dcache entries will exist at process exit time it
3083  *       just makes it very unlikely that any will persist.
3084  */
3085
3086 void proc_flush_task(struct task_struct *task)
3087 {
3088         int i;
3089         struct pid *pid, *tgid;
3090         struct upid *upid;
3091
3092         pid = task_pid(task);
3093         tgid = task_tgid(task);
3094
3095         for (i = 0; i <= pid->level; i++) {
3096                 upid = &pid->numbers[i];
3097                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3098                                         tgid->numbers[i].nr);
3099         }
3100 }
3101
3102 static int proc_pid_instantiate(struct inode *dir,
3103                                    struct dentry * dentry,
3104                                    struct task_struct *task, const void *ptr)
3105 {
3106         struct inode *inode;
3107
3108         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3109         if (!inode)
3110                 goto out;
3111
3112         inode->i_op = &proc_tgid_base_inode_operations;
3113         inode->i_fop = &proc_tgid_base_operations;
3114         inode->i_flags|=S_IMMUTABLE;
3115
3116         set_nlink(inode, nlink_tgid);
3117
3118         d_set_d_op(dentry, &pid_dentry_operations);
3119
3120         d_add(dentry, inode);
3121         /* Close the race of the process dying before we return the dentry */
3122         if (pid_revalidate(dentry, 0))
3123                 return 0;
3124 out:
3125         return -ENOENT;
3126 }
3127
3128 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3129 {
3130         int result = -ENOENT;
3131         struct task_struct *task;
3132         unsigned tgid;
3133         struct pid_namespace *ns;
3134
3135         tgid = name_to_int(&dentry->d_name);
3136         if (tgid == ~0U)
3137                 goto out;
3138
3139         ns = dentry->d_sb->s_fs_info;
3140         rcu_read_lock();
3141         task = find_task_by_pid_ns(tgid, ns);
3142         if (task)
3143                 get_task_struct(task);
3144         rcu_read_unlock();
3145         if (!task)
3146                 goto out;
3147
3148         result = proc_pid_instantiate(dir, dentry, task, NULL);
3149         put_task_struct(task);
3150 out:
3151         return ERR_PTR(result);
3152 }
3153
3154 /*
3155  * Find the first task with tgid >= tgid
3156  *
3157  */
3158 struct tgid_iter {
3159         unsigned int tgid;
3160         struct task_struct *task;
3161 };
3162 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3163 {
3164         struct pid *pid;
3165
3166         if (iter.task)
3167                 put_task_struct(iter.task);
3168         rcu_read_lock();
3169 retry:
3170         iter.task = NULL;
3171         pid = find_ge_pid(iter.tgid, ns);
3172         if (pid) {
3173                 iter.tgid = pid_nr_ns(pid, ns);
3174                 iter.task = pid_task(pid, PIDTYPE_PID);
3175                 /* What we to know is if the pid we have find is the
3176                  * pid of a thread_group_leader.  Testing for task
3177                  * being a thread_group_leader is the obvious thing
3178                  * todo but there is a window when it fails, due to
3179                  * the pid transfer logic in de_thread.
3180                  *
3181                  * So we perform the straight forward test of seeing
3182                  * if the pid we have found is the pid of a thread
3183                  * group leader, and don't worry if the task we have
3184                  * found doesn't happen to be a thread group leader.
3185                  * As we don't care in the case of readdir.
3186                  */
3187                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3188                         iter.tgid += 1;
3189                         goto retry;
3190                 }
3191                 get_task_struct(iter.task);
3192         }
3193         rcu_read_unlock();
3194         return iter;
3195 }
3196
3197 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + 2)
3198
3199 /* for the /proc/ directory itself, after non-process stuff has been done */
3200 int proc_pid_readdir(struct file *file, struct dir_context *ctx)
3201 {
3202         struct tgid_iter iter;
3203         struct pid_namespace *ns = file_inode(file)->i_sb->s_fs_info;
3204         loff_t pos = ctx->pos;
3205
3206         if (pos >= PID_MAX_LIMIT + TGID_OFFSET)
3207                 return 0;
3208
3209         if (pos == TGID_OFFSET - 2) {
3210                 struct inode *inode = d_inode(ns->proc_self);
3211                 if (!dir_emit(ctx, "self", 4, inode->i_ino, DT_LNK))
3212                         return 0;
3213                 ctx->pos = pos = pos + 1;
3214         }
3215         if (pos == TGID_OFFSET - 1) {
3216                 struct inode *inode = d_inode(ns->proc_thread_self);
3217                 if (!dir_emit(ctx, "thread-self", 11, inode->i_ino, DT_LNK))
3218                         return 0;
3219                 ctx->pos = pos = pos + 1;
3220         }
3221         iter.tgid = pos - TGID_OFFSET;
3222         iter.task = NULL;
3223         for (iter = next_tgid(ns, iter);
3224              iter.task;
3225              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3226                 char name[PROC_NUMBUF];
3227                 int len;
3228
3229                 cond_resched();
3230                 if (!has_pid_permissions(ns, iter.task, HIDEPID_INVISIBLE))
3231                         continue;
3232
3233                 len = snprintf(name, sizeof(name), "%d", iter.tgid);
3234                 ctx->pos = iter.tgid + TGID_OFFSET;
3235                 if (!proc_fill_cache(file, ctx, name, len,
3236                                      proc_pid_instantiate, iter.task, NULL)) {
3237                         put_task_struct(iter.task);
3238                         return 0;
3239                 }
3240         }
3241         ctx->pos = PID_MAX_LIMIT + TGID_OFFSET;
3242         return 0;
3243 }
3244
3245 /*
3246  * proc_tid_comm_permission is a special permission function exclusively
3247  * used for the node /proc/<pid>/task/<tid>/comm.
3248  * It bypasses generic permission checks in the case where a task of the same
3249  * task group attempts to access the node.
3250  * The rationale behind this is that glibc and bionic access this node for
3251  * cross thread naming (pthread_set/getname_np(!self)). However, if
3252  * PR_SET_DUMPABLE gets set to 0 this node among others becomes uid=0 gid=0,
3253  * which locks out the cross thread naming implementation.
3254  * This function makes sure that the node is always accessible for members of
3255  * same thread group.
3256  */
3257 static int proc_tid_comm_permission(struct inode *inode, int mask)
3258 {
3259         bool is_same_tgroup;
3260         struct task_struct *task;
3261
3262         task = get_proc_task(inode);
3263         if (!task)
3264                 return -ESRCH;
3265         is_same_tgroup = same_thread_group(current, task);
3266         put_task_struct(task);
3267
3268         if (likely(is_same_tgroup && !(mask & MAY_EXEC))) {
3269                 /* This file (/proc/<pid>/task/<tid>/comm) can always be
3270                  * read or written by the members of the corresponding
3271                  * thread group.
3272                  */
3273                 return 0;
3274         }
3275
3276         return generic_permission(inode, mask);
3277 }
3278
3279 static const struct inode_operations proc_tid_comm_inode_operations = {
3280                 .permission = proc_tid_comm_permission,
3281 };
3282
3283 /*
3284  * Tasks
3285  */
3286 static const struct pid_entry tid_base_stuff[] = {
3287         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3288         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3289         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3290 #ifdef CONFIG_NET
3291         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3292 #endif
3293         REG("environ",   S_IRUSR, proc_environ_operations),
3294         REG("auxv",      S_IRUSR, proc_auxv_operations),
3295         ONE("status",    S_IRUGO, proc_pid_status),
3296         ONE("personality", S_IRUSR, proc_pid_personality),
3297         ONE("limits",    S_IRUGO, proc_pid_limits),
3298 #ifdef CONFIG_SCHED_DEBUG
3299         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3300 #endif
3301         NOD("comm",      S_IFREG|S_IRUGO|S_IWUSR,
3302                          &proc_tid_comm_inode_operations,
3303                          &proc_pid_set_comm_operations, {}),
3304 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3305         ONE("syscall",   S_IRUSR, proc_pid_syscall),
3306 #endif
3307         REG("cmdline",   S_IRUGO, proc_pid_cmdline_ops),
3308         ONE("stat",      S_IRUGO, proc_tid_stat),
3309         ONE("statm",     S_IRUGO, proc_pid_statm),
3310         REG("maps",      S_IRUGO, proc_tid_maps_operations),
3311 #ifdef CONFIG_PROC_CHILDREN
3312         REG("children",  S_IRUGO, proc_tid_children_operations),
3313 #endif
3314 #ifdef CONFIG_NUMA
3315         REG("numa_maps", S_IRUGO, proc_tid_numa_maps_operations),
3316 #endif
3317         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3318         LNK("cwd",       proc_cwd_link),
3319         LNK("root",      proc_root_link),
3320         LNK("exe",       proc_exe_link),
3321         REG("mounts",    S_IRUGO, proc_mounts_operations),
3322         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3323 #ifdef CONFIG_PROC_PAGE_MONITOR
3324         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3325         REG("smaps",     S_IRUGO, proc_tid_smaps_operations),
3326         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
3327 #endif
3328 #ifdef CONFIG_SECURITY
3329         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3330 #endif
3331 #ifdef CONFIG_KALLSYMS
3332         ONE("wchan",     S_IRUGO, proc_pid_wchan),
3333 #endif
3334 #ifdef CONFIG_STACKTRACE
3335         ONE("stack",      S_IRUSR, proc_pid_stack),
3336 #endif
3337 #ifdef CONFIG_SCHED_INFO
3338         ONE("schedstat", S_IRUGO, proc_pid_schedstat),
3339 #endif
3340 #ifdef CONFIG_LATENCYTOP
3341         REG("latency",  S_IRUGO, proc_lstats_operations),
3342 #endif
3343 #ifdef CONFIG_PROC_PID_CPUSET
3344         ONE("cpuset",    S_IRUGO, proc_cpuset_show),
3345 #endif
3346 #ifdef CONFIG_CGROUPS
3347         ONE("cgroup",  S_IRUGO, proc_cgroup_show),
3348 #endif
3349         ONE("oom_score", S_IRUGO, proc_oom_score),
3350         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adj_operations),
3351         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3352 #ifdef CONFIG_AUDITSYSCALL
3353         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3354         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3355 #endif
3356 #ifdef CONFIG_FAULT_INJECTION
3357         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3358         REG("fail-nth", 0644, proc_fail_nth_operations),
3359 #endif
3360 #ifdef CONFIG_TASK_IO_ACCOUNTING
3361         ONE("io",       S_IRUSR, proc_tid_io_accounting),
3362 #endif
3363 #ifdef CONFIG_HARDWALL
3364         ONE("hardwall",   S_IRUGO, proc_pid_hardwall),
3365 #endif
3366 #ifdef CONFIG_USER_NS
3367         REG("uid_map",    S_IRUGO|S_IWUSR, proc_uid_map_operations),
3368         REG("gid_map",    S_IRUGO|S_IWUSR, proc_gid_map_operations),
3369         REG("projid_map", S_IRUGO|S_IWUSR, proc_projid_map_operations),
3370         REG("setgroups",  S_IRUGO|S_IWUSR, proc_setgroups_operations),
3371 #endif
3372 #ifdef CONFIG_LIVEPATCH
3373         ONE("patch_state",  S_IRUSR, proc_pid_patch_state),
3374 #endif
3375 };
3376
3377 static int proc_tid_base_readdir(struct file *file, struct dir_context *ctx)
3378 {
3379         return proc_pident_readdir(file, ctx,
3380                                    tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3381 }
3382
3383 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
3384 {
3385         return proc_pident_lookup(dir, dentry,
3386                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3387 }
3388
3389 static const struct file_operations proc_tid_base_operations = {
3390         .read           = generic_read_dir,
3391         .iterate_shared = proc_tid_base_readdir,
3392         .llseek         = generic_file_llseek,
3393 };
3394
3395 static const struct inode_operations proc_tid_base_inode_operations = {
3396         .lookup         = proc_tid_base_lookup,
3397         .getattr        = pid_getattr,
3398         .setattr        = proc_setattr,
3399 };
3400
3401 static int proc_task_instantiate(struct inode *dir,
3402         struct dentry *dentry, struct task_struct *task, const void *ptr)
3403 {
3404         struct inode *inode;
3405         inode = proc_pid_make_inode(dir->i_sb, task, S_IFDIR | S_IRUGO | S_IXUGO);
3406
3407         if (!inode)
3408                 goto out;
3409         inode->i_op = &proc_tid_base_inode_operations;
3410         inode->i_fop = &proc_tid_base_operations;
3411         inode->i_flags|=S_IMMUTABLE;
3412
3413         set_nlink(inode, nlink_tid);
3414
3415         d_set_d_op(dentry, &pid_dentry_operations);
3416
3417         d_add(dentry, inode);
3418         /* Close the race of the process dying before we return the dentry */
3419         if (pid_revalidate(dentry, 0))
3420                 return 0;
3421 out:
3422         return -ENOENT;
3423 }
3424
3425 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
3426 {
3427         int result = -ENOENT;
3428         struct task_struct *task;
3429         struct task_struct *leader = get_proc_task(dir);
3430         unsigned tid;
3431         struct pid_namespace *ns;
3432
3433         if (!leader)
3434                 goto out_no_task;
3435
3436         tid = name_to_int(&dentry->d_name);
3437         if (tid == ~0U)
3438                 goto out;
3439
3440         ns = dentry->d_sb->s_fs_info;
3441         rcu_read_lock();
3442         task = find_task_by_pid_ns(tid, ns);
3443         if (task)
3444                 get_task_struct(task);
3445         rcu_read_unlock();
3446         if (!task)
3447                 goto out;
3448         if (!same_thread_group(leader, task))
3449                 goto out_drop_task;
3450
3451         result = proc_task_instantiate(dir, dentry, task, NULL);
3452 out_drop_task:
3453         put_task_struct(task);
3454 out:
3455         put_task_struct(leader);
3456 out_no_task:
3457         return ERR_PTR(result);
3458 }
3459
3460 /*
3461  * Find the first tid of a thread group to return to user space.
3462  *
3463  * Usually this is just the thread group leader, but if the users
3464  * buffer was too small or there was a seek into the middle of the
3465  * directory we have more work todo.
3466  *
3467  * In the case of a short read we start with find_task_by_pid.
3468  *
3469  * In the case of a seek we start with the leader and walk nr
3470  * threads past it.
3471  */
3472 static struct task_struct *first_tid(struct pid *pid, int tid, loff_t f_pos,
3473                                         struct pid_namespace *ns)
3474 {
3475         struct task_struct *pos, *task;
3476         unsigned long nr = f_pos;
3477
3478         if (nr != f_pos)        /* 32bit overflow? */
3479                 return NULL;
3480
3481         rcu_read_lock();
3482         task = pid_task(pid, PIDTYPE_PID);
3483         if (!task)
3484                 goto fail;
3485
3486         /* Attempt to start with the tid of a thread */
3487         if (tid && nr) {
3488                 pos = find_task_by_pid_ns(tid, ns);
3489                 if (pos && same_thread_group(pos, task))
3490                         goto found;
3491         }
3492
3493         /* If nr exceeds the number of threads there is nothing todo */
3494         if (nr >= get_nr_threads(task))
3495                 goto fail;
3496
3497         /* If we haven't found our starting place yet start
3498          * with the leader and walk nr threads forward.
3499          */
3500         pos = task = task->group_leader;
3501         do {
3502                 if (!nr--)
3503                         goto found;
3504         } while_each_thread(task, pos);
3505 fail:
3506         pos = NULL;
3507         goto out;
3508 found:
3509         get_task_struct(pos);
3510 out:
3511         rcu_read_unlock();
3512         return pos;
3513 }
3514
3515 /*
3516  * Find the next thread in the thread list.
3517  * Return NULL if there is an error or no next thread.
3518  *
3519  * The reference to the input task_struct is released.
3520  */
3521 static struct task_struct *next_tid(struct task_struct *start)
3522 {
3523         struct task_struct *pos = NULL;
3524         rcu_read_lock();
3525         if (pid_alive(start)) {
3526                 pos = next_thread(start);
3527                 if (thread_group_leader(pos))
3528                         pos = NULL;
3529                 else
3530                         get_task_struct(pos);
3531         }
3532         rcu_read_unlock();
3533         put_task_struct(start);
3534         return pos;
3535 }
3536
3537 /* for the /proc/TGID/task/ directories */
3538 static int proc_task_readdir(struct file *file, struct dir_context *ctx)
3539 {
3540         struct inode *inode = file_inode(file);
3541         struct task_struct *task;
3542         struct pid_namespace *ns;
3543         int tid;
3544
3545         if (proc_inode_is_dead(inode))
3546                 return -ENOENT;
3547
3548         if (!dir_emit_dots(file, ctx))
3549                 return 0;
3550
3551         /* f_version caches the tgid value that the last readdir call couldn't
3552          * return. lseek aka telldir automagically resets f_version to 0.
3553          */
3554         ns = inode->i_sb->s_fs_info;
3555         tid = (int)file->f_version;
3556         file->f_version = 0;
3557         for (task = first_tid(proc_pid(inode), tid, ctx->pos - 2, ns);
3558              task;
3559              task = next_tid(task), ctx->pos++) {
3560                 char name[PROC_NUMBUF];
3561                 int len;
3562                 tid = task_pid_nr_ns(task, ns);
3563                 len = snprintf(name, sizeof(name), "%d", tid);
3564                 if (!proc_fill_cache(file, ctx, name, len,
3565                                 proc_task_instantiate, task, NULL)) {
3566                         /* returning this tgid failed, save it as the first
3567                          * pid for the next readir call */
3568                         file->f_version = (u64)tid;
3569                         put_task_struct(task);
3570                         break;
3571                 }
3572         }
3573
3574         return 0;
3575 }
3576
3577 static int proc_task_getattr(const struct path *path, struct kstat *stat,
3578                              u32 request_mask, unsigned int query_flags)
3579 {
3580         struct inode *inode = d_inode(path->dentry);
3581         struct task_struct *p = get_proc_task(inode);
3582         generic_fillattr(inode, stat);
3583
3584         if (p) {
3585                 stat->nlink += get_nr_threads(p);
3586                 put_task_struct(p);
3587         }
3588
3589         return 0;
3590 }
3591
3592 static const struct inode_operations proc_task_inode_operations = {
3593         .lookup         = proc_task_lookup,
3594         .getattr        = proc_task_getattr,
3595         .setattr        = proc_setattr,
3596         .permission     = proc_pid_permission,
3597 };
3598
3599 static const struct file_operations proc_task_operations = {
3600         .read           = generic_read_dir,
3601         .iterate_shared = proc_task_readdir,
3602         .llseek         = generic_file_llseek,
3603 };
3604
3605 void __init set_proc_pid_nlink(void)
3606 {
3607         nlink_tid = pid_entry_nlink(tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3608         nlink_tgid = pid_entry_nlink(tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3609 }