fs/proc/base.c

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