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/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/fs_struct.h>
  85 #include <linux/slab.h>
  86 #include <linux/flex_array.h>
  87 #ifdef CONFIG_HARDWALL
  88 #include <asm/hardwall.h>
  89 #endif
  90 #include <trace/events/oom.h>
  91 #include "internal.h"
  92
  93 /* NOTE:
  94  *      Implementing inode permission operations in /proc is almost
  95  *      certainly an error.  Permission checks need to happen during
  96  *      each system call not at open time.  The reason is that most of
  97  *      what we wish to check for permissions in /proc varies at runtime.
  98  *
  99  *      The classic example of a problem is opening file descriptors
 100  *      in /proc for a task before it execs a suid executable.
 101  */
 102
 103 struct pid_entry {
 104         char *name;
 105         int len;
 106         umode_t mode;
 107         const struct inode_operations *iop;
 108         const struct file_operations *fop;
 109         union proc_op op;
 110 };
 111
 112 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 113         .name = (NAME),                                 \
 114         .len  = sizeof(NAME) - 1,                       \
 115         .mode = MODE,                                   \
 116         .iop  = IOP,                                    \
 117         .fop  = FOP,                                    \
 118         .op   = OP,                                     \
 119 }
 120
 121 #define DIR(NAME, MODE, iops, fops)     \
 122         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 123 #define LNK(NAME, get_link)                                     \
 124         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 125                 &proc_pid_link_inode_operations, NULL,          \
 126                 { .proc_get_link = get_link } )
 127 #define REG(NAME, MODE, fops)                           \
 128         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 129 #define INF(NAME, MODE, read)                           \
 130         NOD(NAME, (S_IFREG|(MODE)),                     \
 131                 NULL, &proc_info_file_operations,       \
 132                 { .proc_read = read } )
 133 #define ONE(NAME, MODE, show)                           \
 134         NOD(NAME, (S_IFREG|(MODE)),                     \
 135                 NULL, &proc_single_file_operations,     \
 136                 { .proc_show = show } )
 137
 138 static int proc_fd_permission(struct inode *inode, int mask);
 139
 140 /*
 141  * Count the number of hardlinks for the pid_entry table, excluding the .
 142  * and .. links.
 143  */
 144 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 145         unsigned int n)
 146 {
 147         unsigned int i;
 148         unsigned int count;
 149
 150         count = 0;
 151         for (i = 0; i < n; ++i) {
 152                 if (S_ISDIR(entries[i].mode))
 153                         ++count;
 154         }
 155
 156         return count;
 157 }
 158
 159 static int get_task_root(struct task_struct *task, struct path *root)
 160 {
 161         int result = -ENOENT;
 162
 163         task_lock(task);
 164         if (task->fs) {
 165                 get_fs_root(task->fs, root);
 166                 result = 0;
 167         }
 168         task_unlock(task);
 169         return result;
 170 }
 171
 172 static int proc_cwd_link(struct dentry *dentry, struct path *path)
 173 {
 174         struct task_struct *task = get_proc_task(dentry->d_inode);
 175         int result = -ENOENT;
 176
 177         if (task) {
 178                 task_lock(task);
 179                 if (task->fs) {
 180                         get_fs_pwd(task->fs, path);
 181                         result = 0;
 182                 }
 183                 task_unlock(task);
 184                 put_task_struct(task);
 185         }
 186         return result;
 187 }
 188
 189 static int proc_root_link(struct dentry *dentry, struct path *path)
 190 {
 191         struct task_struct *task = get_proc_task(dentry->d_inode);
 192         int result = -ENOENT;
 193
 194         if (task) {
 195                 result = get_task_root(task, path);
 196                 put_task_struct(task);
 197         }
 198         return result;
 199 }
 200
 201 static struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 202 {
 203         struct mm_struct *mm;
 204         int err;
 205
 206         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 207         if (err)
 208                 return ERR_PTR(err);
 209
 210         mm = get_task_mm(task);
 211         if (mm && mm != current->mm &&
 212                         !ptrace_may_access(task, mode)) {
 213                 mmput(mm);
 214                 mm = ERR_PTR(-EACCES);
 215         }
 216         mutex_unlock(&task->signal->cred_guard_mutex);
 217
 218         return mm;
 219 }
 220
 221 struct mm_struct *mm_for_maps(struct task_struct *task)
 222 {
 223         return mm_access(task, PTRACE_MODE_READ);
 224 }
 225
 226 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 227 {
 228         int res = 0;
 229         unsigned int len;
 230         struct mm_struct *mm = get_task_mm(task);
 231         if (!mm)
 232                 goto out;
 233         if (!mm->arg_end)
 234                 goto out_mm;    /* Shh! No looking before we're done */
 235
 236         len = mm->arg_end - mm->arg_start;
 237
 238         if (len > PAGE_SIZE)
 239                 len = PAGE_SIZE;
 240
 241         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 242
 243         // If the nul at the end of args has been overwritten, then
 244         // assume application is using setproctitle(3).
 245         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 246                 len = strnlen(buffer, res);
 247                 if (len < res) {
 248                     res = len;
 249                 } else {
 250                         len = mm->env_end - mm->env_start;
 251                         if (len > PAGE_SIZE - res)
 252                                 len = PAGE_SIZE - res;
 253                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 254                         res = strnlen(buffer, res);
 255                 }
 256         }
 257 out_mm:
 258         mmput(mm);
 259 out:
 260         return res;
 261 }
 262
 263 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 264 {
 265         struct mm_struct *mm = mm_for_maps(task);
 266         int res = PTR_ERR(mm);
 267         if (mm && !IS_ERR(mm)) {
 268                 unsigned int nwords = 0;
 269                 do {
 270                         nwords += 2;
 271                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 272                 res = nwords * sizeof(mm->saved_auxv[0]);
 273                 if (res > PAGE_SIZE)
 274                         res = PAGE_SIZE;
 275                 memcpy(buffer, mm->saved_auxv, res);
 276                 mmput(mm);
 277         }
 278         return res;
 279 }
 280
 281
 282 #ifdef CONFIG_KALLSYMS
 283 /*
 284  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 285  * Returns the resolved symbol.  If that fails, simply return the address.
 286  */
 287 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 288 {
 289         unsigned long wchan;
 290         char symname[KSYM_NAME_LEN];
 291
 292         wchan = get_wchan(task);
 293
 294         if (lookup_symbol_name(wchan, symname) < 0)
 295                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 296                         return 0;
 297                 else
 298                         return sprintf(buffer, "%lu", wchan);
 299         else
 300                 return sprintf(buffer, "%s", symname);
 301 }
 302 #endif /* CONFIG_KALLSYMS */
 303
 304 static int lock_trace(struct task_struct *task)
 305 {
 306         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 307         if (err)
 308                 return err;
 309         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 310                 mutex_unlock(&task->signal->cred_guard_mutex);
 311                 return -EPERM;
 312         }
 313         return 0;
 314 }
 315
 316 static void unlock_trace(struct task_struct *task)
 317 {
 318         mutex_unlock(&task->signal->cred_guard_mutex);
 319 }
 320
 321 #ifdef CONFIG_STACKTRACE
 322
 323 #define MAX_STACK_TRACE_DEPTH   64
 324
 325 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 326                           struct pid *pid, struct task_struct *task)
 327 {
 328         struct stack_trace trace;
 329         unsigned long *entries;
 330         int err;
 331         int i;
 332
 333         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 334         if (!entries)
 335                 return -ENOMEM;
 336
 337         trace.nr_entries        = 0;
 338         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 339         trace.entries           = entries;
 340         trace.skip              = 0;
 341
 342         err = lock_trace(task);
 343         if (!err) {
 344                 save_stack_trace_tsk(task, &trace);
 345
 346                 for (i = 0; i < trace.nr_entries; i++) {
 347                         seq_printf(m, "[<%pK>] %pS\n",
 348                                    (void *)entries[i], (void *)entries[i]);
 349                 }
 350                 unlock_trace(task);
 351         }
 352         kfree(entries);
 353
 354         return err;
 355 }
 356 #endif
 357
 358 #ifdef CONFIG_SCHEDSTATS
 359 /*
 360  * Provides /proc/PID/schedstat
 361  */
 362 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 363 {
 364         return sprintf(buffer, "%llu %llu %lu\n",
 365                         (unsigned long long)task->se.sum_exec_runtime,
 366                         (unsigned long long)task->sched_info.run_delay,
 367                         task->sched_info.pcount);
 368 }
 369 #endif
 370
 371 #ifdef CONFIG_LATENCYTOP
 372 static int lstats_show_proc(struct seq_file *m, void *v)
 373 {
 374         int i;
 375         struct inode *inode = m->private;
 376         struct task_struct *task = get_proc_task(inode);
 377
 378         if (!task)
 379                 return -ESRCH;
 380         seq_puts(m, "Latency Top version : v0.1\n");
 381         for (i = 0; i < 32; i++) {
 382                 struct latency_record *lr = &task->latency_record[i];
 383                 if (lr->backtrace[0]) {
 384                         int q;
 385                         seq_printf(m, "%i %li %li",
 386                                    lr->count, lr->time, lr->max);
 387                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 388                                 unsigned long bt = lr->backtrace[q];
 389                                 if (!bt)
 390                                         break;
 391                                 if (bt == ULONG_MAX)
 392                                         break;
 393                                 seq_printf(m, " %ps", (void *)bt);
 394                         }
 395                         seq_putc(m, '\n');
 396                 }
 397
 398         }
 399         put_task_struct(task);
 400         return 0;
 401 }
 402
 403 static int lstats_open(struct inode *inode, struct file *file)
 404 {
 405         return single_open(file, lstats_show_proc, inode);
 406 }
 407
 408 static ssize_t lstats_write(struct file *file, const char __user *buf,
 409                             size_t count, loff_t *offs)
 410 {
 411         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 412
 413         if (!task)
 414                 return -ESRCH;
 415         clear_all_latency_tracing(task);
 416         put_task_struct(task);
 417
 418         return count;
 419 }
 420
 421 static const struct file_operations proc_lstats_operations = {
 422         .open           = lstats_open,
 423         .read           = seq_read,
 424         .write          = lstats_write,
 425         .llseek         = seq_lseek,
 426         .release        = single_release,
 427 };
 428
 429 #endif
 430
 431 static int proc_oom_score(struct task_struct *task, char *buffer)
 432 {
 433         unsigned long points = 0;
 434
 435         read_lock(&tasklist_lock);
 436         if (pid_alive(task))
 437                 points = oom_badness(task, NULL, NULL,
 438                                         totalram_pages + total_swap_pages);
 439         read_unlock(&tasklist_lock);
 440         return sprintf(buffer, "%lu\n", points);
 441 }
 442
 443 struct limit_names {
 444         char *name;
 445         char *unit;
 446 };
 447
 448 static const struct limit_names lnames[RLIM_NLIMITS] = {
 449         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 450         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 451         [RLIMIT_DATA] = {"Max data size", "bytes"},
 452         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 453         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 454         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 455         [RLIMIT_NPROC] = {"Max processes", "processes"},
 456         [RLIMIT_NOFILE] = {"Max open files", "files"},
 457         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 458         [RLIMIT_AS] = {"Max address space", "bytes"},
 459         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 460         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 461         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 462         [RLIMIT_NICE] = {"Max nice priority", NULL},
 463         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 464         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 465 };
 466
 467 /* Display limits for a process */
 468 static int proc_pid_limits(struct task_struct *task, char *buffer)
 469 {
 470         unsigned int i;
 471         int count = 0;
 472         unsigned long flags;
 473         char *bufptr = buffer;
 474
 475         struct rlimit rlim[RLIM_NLIMITS];
 476
 477         if (!lock_task_sighand(task, &flags))
 478                 return 0;
 479         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 480         unlock_task_sighand(task, &flags);
 481
 482         /*
 483          * print the file header
 484          */
 485         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 486                         "Limit", "Soft Limit", "Hard Limit", "Units");
 487
 488         for (i = 0; i < RLIM_NLIMITS; i++) {
 489                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 490                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 491                                          lnames[i].name, "unlimited");
 492                 else
 493                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 494                                          lnames[i].name, rlim[i].rlim_cur);
 495
 496                 if (rlim[i].rlim_max == RLIM_INFINITY)
 497                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 498                 else
 499                         count += sprintf(&bufptr[count], "%-20lu ",
 500                                          rlim[i].rlim_max);
 501
 502                 if (lnames[i].unit)
 503                         count += sprintf(&bufptr[count], "%-10s\n",
 504                                          lnames[i].unit);
 505                 else
 506                         count += sprintf(&bufptr[count], "\n");
 507         }
 508
 509         return count;
 510 }
 511
 512 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 513 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 514 {
 515         long nr;
 516         unsigned long args[6], sp, pc;
 517         int res = lock_trace(task);
 518         if (res)
 519                 return res;
 520
 521         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 522                 res = sprintf(buffer, "running\n");
 523         else if (nr < 0)
 524                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 525         else
 526                 res = sprintf(buffer,
 527                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 528                        nr,
 529                        args[0], args[1], args[2], args[3], args[4], args[5],
 530                        sp, pc);
 531         unlock_trace(task);
 532         return res;
 533 }
 534 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 535
 536 /************************************************************************/
 537 /*                       Here the fs part begins                        */
 538 /************************************************************************/
 539
 540 /* permission checks */
 541 static int proc_fd_access_allowed(struct inode *inode)
 542 {
 543         struct task_struct *task;
 544         int allowed = 0;
 545         /* Allow access to a task's file descriptors if it is us or we
 546          * may use ptrace attach to the process and find out that
 547          * information.
 548          */
 549         task = get_proc_task(inode);
 550         if (task) {
 551                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 552                 put_task_struct(task);
 553         }
 554         return allowed;
 555 }
 556
 557 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 558 {
 559         int error;
 560         struct inode *inode = dentry->d_inode;
 561
 562         if (attr->ia_valid & ATTR_MODE)
 563                 return -EPERM;
 564
 565         error = inode_change_ok(inode, attr);
 566         if (error)
 567                 return error;
 568
 569         if ((attr->ia_valid & ATTR_SIZE) &&
 570             attr->ia_size != i_size_read(inode)) {
 571                 error = vmtruncate(inode, attr->ia_size);
 572                 if (error)
 573                         return error;
 574         }
 575
 576         setattr_copy(inode, attr);
 577         mark_inode_dirty(inode);
 578         return 0;
 579 }
 580
 581 /*
 582  * May current process learn task's sched/cmdline info (for hide_pid_min=1)
 583  * or euid/egid (for hide_pid_min=2)?
 584  */
 585 static bool has_pid_permissions(struct pid_namespace *pid,
 586                                  struct task_struct *task,
 587                                  int hide_pid_min)
 588 {
 589         if (pid->hide_pid < hide_pid_min)
 590                 return true;
 591         if (in_group_p(pid->pid_gid))
 592                 return true;
 593         return ptrace_may_access(task, PTRACE_MODE_READ);
 594 }
 595
 596
 597 static int proc_pid_permission(struct inode *inode, int mask)
 598 {
 599         struct pid_namespace *pid = inode->i_sb->s_fs_info;
 600         struct task_struct *task;
 601         bool has_perms;
 602
 603         task = get_proc_task(inode);
 604         if (!task)
 605                 return -ESRCH;
 606         has_perms = has_pid_permissions(pid, task, 1);
 607         put_task_struct(task);
 608
 609         if (!has_perms) {
 610                 if (pid->hide_pid == 2) {
 611                         /*
 612                          * Let's make getdents(), stat(), and open()
 613                          * consistent with each other.  If a process
 614                          * may not stat() a file, it shouldn't be seen
 615                          * in procfs at all.
 616                          */
 617                         return -ENOENT;
 618                 }
 619
 620                 return -EPERM;
 621         }
 622         return generic_permission(inode, mask);
 623 }
 624
 625
 626
 627 static const struct inode_operations proc_def_inode_operations = {
 628         .setattr        = proc_setattr,
 629 };
 630
 631 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 632
 633 static ssize_t proc_info_read(struct file * file, char __user * buf,
 634                           size_t count, loff_t *ppos)
 635 {
 636         struct inode * inode = file->f_path.dentry->d_inode;
 637         unsigned long page;
 638         ssize_t length;
 639         struct task_struct *task = get_proc_task(inode);
 640
 641         length = -ESRCH;
 642         if (!task)
 643                 goto out_no_task;
 644
 645         if (count > PROC_BLOCK_SIZE)
 646                 count = PROC_BLOCK_SIZE;
 647
 648         length = -ENOMEM;
 649         if (!(page = __get_free_page(GFP_TEMPORARY)))
 650                 goto out;
 651
 652         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 653
 654         if (length >= 0)
 655                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 656         free_page(page);
 657 out:
 658         put_task_struct(task);
 659 out_no_task:
 660         return length;
 661 }
 662
 663 static const struct file_operations proc_info_file_operations = {
 664         .read           = proc_info_read,
 665         .llseek         = generic_file_llseek,
 666 };
 667
 668 static int proc_single_show(struct seq_file *m, void *v)
 669 {
 670         struct inode *inode = m->private;
 671         struct pid_namespace *ns;
 672         struct pid *pid;
 673         struct task_struct *task;
 674         int ret;
 675
 676         ns = inode->i_sb->s_fs_info;
 677         pid = proc_pid(inode);
 678         task = get_pid_task(pid, PIDTYPE_PID);
 679         if (!task)
 680                 return -ESRCH;
 681
 682         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 683
 684         put_task_struct(task);
 685         return ret;
 686 }
 687
 688 static int proc_single_open(struct inode *inode, struct file *filp)
 689 {
 690         return single_open(filp, proc_single_show, inode);
 691 }
 692
 693 static const struct file_operations proc_single_file_operations = {
 694         .open           = proc_single_open,
 695         .read           = seq_read,
 696         .llseek         = seq_lseek,
 697         .release        = single_release,
 698 };
 699
 700 static int mem_open(struct inode* inode, struct file* file)
 701 {
 702         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 703         struct mm_struct *mm;
 704
 705         if (!task)
 706                 return -ESRCH;
 707
 708         mm = mm_access(task, PTRACE_MODE_ATTACH);
 709         put_task_struct(task);
 710
 711         if (IS_ERR(mm))
 712                 return PTR_ERR(mm);
 713
 714         /* OK to pass negative loff_t, we can catch out-of-range */
 715         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 716         file->private_data = mm;
 717
 718         return 0;
 719 }
 720
 721 static ssize_t mem_read(struct file * file, char __user * buf,
 722                         size_t count, loff_t *ppos)
 723 {
 724         int ret;
 725         char *page;
 726         unsigned long src = *ppos;
 727         struct mm_struct *mm = file->private_data;
 728
 729         if (!mm)
 730                 return 0;
 731
 732         page = (char *)__get_free_page(GFP_TEMPORARY);
 733         if (!page)
 734                 return -ENOMEM;
 735
 736         ret = 0;
 737
 738         while (count > 0) {
 739                 int this_len, retval;
 740
 741                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 742                 retval = access_remote_vm(mm, src, page, this_len, 0);
 743                 if (!retval) {
 744                         if (!ret)
 745                                 ret = -EIO;
 746                         break;
 747                 }
 748
 749                 if (copy_to_user(buf, page, retval)) {
 750                         ret = -EFAULT;
 751                         break;
 752                 }
 753
 754                 ret += retval;
 755                 src += retval;
 756                 buf += retval;
 757                 count -= retval;
 758         }
 759         *ppos = src;
 760
 761         free_page((unsigned long) page);
 762         return ret;
 763 }
 764
 765 static ssize_t mem_write(struct file * file, const char __user *buf,
 766                          size_t count, loff_t *ppos)
 767 {
 768         int copied;
 769         char *page;
 770         unsigned long dst = *ppos;
 771         struct mm_struct *mm = file->private_data;
 772
 773         if (!mm)
 774                 return 0;
 775
 776         page = (char *)__get_free_page(GFP_TEMPORARY);
 777         if (!page)
 778                 return -ENOMEM;
 779
 780         copied = 0;
 781         while (count > 0) {
 782                 int this_len, retval;
 783
 784                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 785                 if (copy_from_user(page, buf, this_len)) {
 786                         copied = -EFAULT;
 787                         break;
 788                 }
 789                 retval = access_remote_vm(mm, dst, page, this_len, 1);
 790                 if (!retval) {
 791                         if (!copied)
 792                                 copied = -EIO;
 793                         break;
 794                 }
 795                 copied += retval;
 796                 buf += retval;
 797                 dst += retval;
 798                 count -= retval;
 799         }
 800         *ppos = dst;
 801
 802         free_page((unsigned long) page);
 803         return copied;
 804 }
 805
 806 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 807 {
 808         switch (orig) {
 809         case 0:
 810                 file->f_pos = offset;
 811                 break;
 812         case 1:
 813                 file->f_pos += offset;
 814                 break;
 815         default:
 816                 return -EINVAL;
 817         }
 818         force_successful_syscall_return();
 819         return file->f_pos;
 820 }
 821
 822 static int mem_release(struct inode *inode, struct file *file)
 823 {
 824         struct mm_struct *mm = file->private_data;
 825
 826         mmput(mm);
 827         return 0;
 828 }
 829
 830 static const struct file_operations proc_mem_operations = {
 831         .llseek         = mem_lseek,
 832         .read           = mem_read,
 833         .write          = mem_write,
 834         .open           = mem_open,
 835         .release        = mem_release,
 836 };
 837
 838 static ssize_t environ_read(struct file *file, char __user *buf,
 839                         size_t count, loff_t *ppos)
 840 {
 841         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 842         char *page;
 843         unsigned long src = *ppos;
 844         int ret = -ESRCH;
 845         struct mm_struct *mm;
 846
 847         if (!task)
 848                 goto out_no_task;
 849
 850         ret = -ENOMEM;
 851         page = (char *)__get_free_page(GFP_TEMPORARY);
 852         if (!page)
 853                 goto out;
 854
 855
 856         mm = mm_for_maps(task);
 857         ret = PTR_ERR(mm);
 858         if (!mm || IS_ERR(mm))
 859                 goto out_free;
 860
 861         ret = 0;
 862         while (count > 0) {
 863                 int this_len, retval, max_len;
 864
 865                 this_len = mm->env_end - (mm->env_start + src);
 866
 867                 if (this_len <= 0)
 868                         break;
 869
 870                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 871                 this_len = (this_len > max_len) ? max_len : this_len;
 872
 873                 retval = access_process_vm(task, (mm->env_start + src),
 874                         page, this_len, 0);
 875
 876                 if (retval <= 0) {
 877                         ret = retval;
 878                         break;
 879                 }
 880
 881                 if (copy_to_user(buf, page, retval)) {
 882                         ret = -EFAULT;
 883                         break;
 884                 }
 885
 886                 ret += retval;
 887                 src += retval;
 888                 buf += retval;
 889                 count -= retval;
 890         }
 891         *ppos = src;
 892
 893         mmput(mm);
 894 out_free:
 895         free_page((unsigned long) page);
 896 out:
 897         put_task_struct(task);
 898 out_no_task:
 899         return ret;
 900 }
 901
 902 static const struct file_operations proc_environ_operations = {
 903         .read           = environ_read,
 904         .llseek         = generic_file_llseek,
 905 };
 906
 907 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 908                                 size_t count, loff_t *ppos)
 909 {
 910         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 911         char buffer[PROC_NUMBUF];
 912         size_t len;
 913         int oom_adjust = OOM_DISABLE;
 914         unsigned long flags;
 915
 916         if (!task)
 917                 return -ESRCH;
 918
 919         if (lock_task_sighand(task, &flags)) {
 920                 oom_adjust = task->signal->oom_adj;
 921                 unlock_task_sighand(task, &flags);
 922         }
 923
 924         put_task_struct(task);
 925
 926         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 927
 928         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 929 }
 930
 931 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 932                                 size_t count, loff_t *ppos)
 933 {
 934         struct task_struct *task;
 935         char buffer[PROC_NUMBUF];
 936         int oom_adjust;
 937         unsigned long flags;
 938         int err;
 939
 940         memset(buffer, 0, sizeof(buffer));
 941         if (count > sizeof(buffer) - 1)
 942                 count = sizeof(buffer) - 1;
 943         if (copy_from_user(buffer, buf, count)) {
 944                 err = -EFAULT;
 945                 goto out;
 946         }
 947
 948         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 949         if (err)
 950                 goto out;
 951         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
 952              oom_adjust != OOM_DISABLE) {
 953                 err = -EINVAL;
 954                 goto out;
 955         }
 956
 957         task = get_proc_task(file->f_path.dentry->d_inode);
 958         if (!task) {
 959                 err = -ESRCH;
 960                 goto out;
 961         }
 962
 963         task_lock(task);
 964         if (!task->mm) {
 965                 err = -EINVAL;
 966                 goto err_task_lock;
 967         }
 968
 969         if (!lock_task_sighand(task, &flags)) {
 970                 err = -ESRCH;
 971                 goto err_task_lock;
 972         }
 973
 974         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
 975                 err = -EACCES;
 976                 goto err_sighand;
 977         }
 978
 979         /*
 980          * Warn that /proc/pid/oom_adj is deprecated, see
 981          * Documentation/feature-removal-schedule.txt.
 982          */
 983         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
 984                   current->comm, task_pid_nr(current), task_pid_nr(task),
 985                   task_pid_nr(task));
 986         task->signal->oom_adj = oom_adjust;
 987         /*
 988          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
 989          * value is always attainable.
 990          */
 991         if (task->signal->oom_adj == OOM_ADJUST_MAX)
 992                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
 993         else
 994                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
 995                                                                 -OOM_DISABLE;
 996         trace_oom_score_adj_update(task);
 997 err_sighand:
 998         unlock_task_sighand(task, &flags);
 999 err_task_lock:
1000         task_unlock(task);
1001         put_task_struct(task);
1002 out:
1003         return err < 0 ? err : count;
1004 }
1005
1006 static const struct file_operations proc_oom_adjust_operations = {
1007         .read           = oom_adjust_read,
1008         .write          = oom_adjust_write,
1009         .llseek         = generic_file_llseek,
1010 };
1011
1012 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1013                                         size_t count, loff_t *ppos)
1014 {
1015         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1016         char buffer[PROC_NUMBUF];
1017         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1018         unsigned long flags;
1019         size_t len;
1020
1021         if (!task)
1022                 return -ESRCH;
1023         if (lock_task_sighand(task, &flags)) {
1024                 oom_score_adj = task->signal->oom_score_adj;
1025                 unlock_task_sighand(task, &flags);
1026         }
1027         put_task_struct(task);
1028         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1029         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1030 }
1031
1032 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1033                                         size_t count, loff_t *ppos)
1034 {
1035         struct task_struct *task;
1036         char buffer[PROC_NUMBUF];
1037         unsigned long flags;
1038         int oom_score_adj;
1039         int err;
1040
1041         memset(buffer, 0, sizeof(buffer));
1042         if (count > sizeof(buffer) - 1)
1043                 count = sizeof(buffer) - 1;
1044         if (copy_from_user(buffer, buf, count)) {
1045                 err = -EFAULT;
1046                 goto out;
1047         }
1048
1049         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1050         if (err)
1051                 goto out;
1052         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1053                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1054                 err = -EINVAL;
1055                 goto out;
1056         }
1057
1058         task = get_proc_task(file->f_path.dentry->d_inode);
1059         if (!task) {
1060                 err = -ESRCH;
1061                 goto out;
1062         }
1063
1064         task_lock(task);
1065         if (!task->mm) {
1066                 err = -EINVAL;
1067                 goto err_task_lock;
1068         }
1069
1070         if (!lock_task_sighand(task, &flags)) {
1071                 err = -ESRCH;
1072                 goto err_task_lock;
1073         }
1074
1075         if (oom_score_adj < task->signal->oom_score_adj_min &&
1076                         !capable(CAP_SYS_RESOURCE)) {
1077                 err = -EACCES;
1078                 goto err_sighand;
1079         }
1080
1081         task->signal->oom_score_adj = oom_score_adj;
1082         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1083                 task->signal->oom_score_adj_min = oom_score_adj;
1084         trace_oom_score_adj_update(task);
1085         /*
1086          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1087          * always attainable.
1088          */
1089         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1090                 task->signal->oom_adj = OOM_DISABLE;
1091         else
1092                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1093                                                         OOM_SCORE_ADJ_MAX;
1094 err_sighand:
1095         unlock_task_sighand(task, &flags);
1096 err_task_lock:
1097         task_unlock(task);
1098         put_task_struct(task);
1099 out:
1100         return err < 0 ? err : count;
1101 }
1102
1103 static const struct file_operations proc_oom_score_adj_operations = {
1104         .read           = oom_score_adj_read,
1105         .write          = oom_score_adj_write,
1106         .llseek         = default_llseek,
1107 };
1108
1109 #ifdef CONFIG_AUDITSYSCALL
1110 #define TMPBUFLEN 21
1111 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1112                                   size_t count, loff_t *ppos)
1113 {
1114         struct inode * inode = file->f_path.dentry->d_inode;
1115         struct task_struct *task = get_proc_task(inode);
1116         ssize_t length;
1117         char tmpbuf[TMPBUFLEN];
1118
1119         if (!task)
1120                 return -ESRCH;
1121         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1122                                 audit_get_loginuid(task));
1123         put_task_struct(task);
1124         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1125 }
1126
1127 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1128                                    size_t count, loff_t *ppos)
1129 {
1130         struct inode * inode = file->f_path.dentry->d_inode;
1131         char *page, *tmp;
1132         ssize_t length;
1133         uid_t loginuid;
1134
1135         if (!capable(CAP_AUDIT_CONTROL))
1136                 return -EPERM;
1137
1138         rcu_read_lock();
1139         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1140                 rcu_read_unlock();
1141                 return -EPERM;
1142         }
1143         rcu_read_unlock();
1144
1145         if (count >= PAGE_SIZE)
1146                 count = PAGE_SIZE - 1;
1147
1148         if (*ppos != 0) {
1149                 /* No partial writes. */
1150                 return -EINVAL;
1151         }
1152         page = (char*)__get_free_page(GFP_TEMPORARY);
1153         if (!page)
1154                 return -ENOMEM;
1155         length = -EFAULT;
1156         if (copy_from_user(page, buf, count))
1157                 goto out_free_page;
1158
1159         page[count] = '\0';
1160         loginuid = simple_strtoul(page, &tmp, 10);
1161         if (tmp == page) {
1162                 length = -EINVAL;
1163                 goto out_free_page;
1164
1165         }
1166         length = audit_set_loginuid(current, loginuid);
1167         if (likely(length == 0))
1168                 length = count;
1169
1170 out_free_page:
1171         free_page((unsigned long) page);
1172         return length;
1173 }
1174
1175 static const struct file_operations proc_loginuid_operations = {
1176         .read           = proc_loginuid_read,
1177         .write          = proc_loginuid_write,
1178         .llseek         = generic_file_llseek,
1179 };
1180
1181 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1182                                   size_t count, loff_t *ppos)
1183 {
1184         struct inode * inode = file->f_path.dentry->d_inode;
1185         struct task_struct *task = get_proc_task(inode);
1186         ssize_t length;
1187         char tmpbuf[TMPBUFLEN];
1188
1189         if (!task)
1190                 return -ESRCH;
1191         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1192                                 audit_get_sessionid(task));
1193         put_task_struct(task);
1194         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1195 }
1196
1197 static const struct file_operations proc_sessionid_operations = {
1198         .read           = proc_sessionid_read,
1199         .llseek         = generic_file_llseek,
1200 };
1201 #endif
1202
1203 #ifdef CONFIG_FAULT_INJECTION
1204 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1205                                       size_t count, loff_t *ppos)
1206 {
1207         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1208         char buffer[PROC_NUMBUF];
1209         size_t len;
1210         int make_it_fail;
1211
1212         if (!task)
1213                 return -ESRCH;
1214         make_it_fail = task->make_it_fail;
1215         put_task_struct(task);
1216
1217         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1218
1219         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1220 }
1221
1222 static ssize_t proc_fault_inject_write(struct file * file,
1223                         const char __user * buf, size_t count, loff_t *ppos)
1224 {
1225         struct task_struct *task;
1226         char buffer[PROC_NUMBUF], *end;
1227         int make_it_fail;
1228
1229         if (!capable(CAP_SYS_RESOURCE))
1230                 return -EPERM;
1231         memset(buffer, 0, sizeof(buffer));
1232         if (count > sizeof(buffer) - 1)
1233                 count = sizeof(buffer) - 1;
1234         if (copy_from_user(buffer, buf, count))
1235                 return -EFAULT;
1236         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1237         if (*end)
1238                 return -EINVAL;
1239         task = get_proc_task(file->f_dentry->d_inode);
1240         if (!task)
1241                 return -ESRCH;
1242         task->make_it_fail = make_it_fail;
1243         put_task_struct(task);
1244
1245         return count;
1246 }
1247
1248 static const struct file_operations proc_fault_inject_operations = {
1249         .read           = proc_fault_inject_read,
1250         .write          = proc_fault_inject_write,
1251         .llseek         = generic_file_llseek,
1252 };
1253 #endif
1254
1255
1256 #ifdef CONFIG_SCHED_DEBUG
1257 /*
1258  * Print out various scheduling related per-task fields:
1259  */
1260 static int sched_show(struct seq_file *m, void *v)
1261 {
1262         struct inode *inode = m->private;
1263         struct task_struct *p;
1264
1265         p = get_proc_task(inode);
1266         if (!p)
1267                 return -ESRCH;
1268         proc_sched_show_task(p, m);
1269
1270         put_task_struct(p);
1271
1272         return 0;
1273 }
1274
1275 static ssize_t
1276 sched_write(struct file *file, const char __user *buf,
1277             size_t count, loff_t *offset)
1278 {
1279         struct inode *inode = file->f_path.dentry->d_inode;
1280         struct task_struct *p;
1281
1282         p = get_proc_task(inode);
1283         if (!p)
1284                 return -ESRCH;
1285         proc_sched_set_task(p);
1286
1287         put_task_struct(p);
1288
1289         return count;
1290 }
1291
1292 static int sched_open(struct inode *inode, struct file *filp)
1293 {
1294         return single_open(filp, sched_show, inode);
1295 }
1296
1297 static const struct file_operations proc_pid_sched_operations = {
1298         .open           = sched_open,
1299         .read           = seq_read,
1300         .write          = sched_write,
1301         .llseek         = seq_lseek,
1302         .release        = single_release,
1303 };
1304
1305 #endif
1306
1307 #ifdef CONFIG_SCHED_AUTOGROUP
1308 /*
1309  * Print out autogroup related information:
1310  */
1311 static int sched_autogroup_show(struct seq_file *m, void *v)
1312 {
1313         struct inode *inode = m->private;
1314         struct task_struct *p;
1315
1316         p = get_proc_task(inode);
1317         if (!p)
1318                 return -ESRCH;
1319         proc_sched_autogroup_show_task(p, m);
1320
1321         put_task_struct(p);
1322
1323         return 0;
1324 }
1325
1326 static ssize_t
1327 sched_autogroup_write(struct file *file, const char __user *buf,
1328             size_t count, loff_t *offset)
1329 {
1330         struct inode *inode = file->f_path.dentry->d_inode;
1331         struct task_struct *p;
1332         char buffer[PROC_NUMBUF];
1333         int nice;
1334         int err;
1335
1336         memset(buffer, 0, sizeof(buffer));
1337         if (count > sizeof(buffer) - 1)
1338                 count = sizeof(buffer) - 1;
1339         if (copy_from_user(buffer, buf, count))
1340                 return -EFAULT;
1341
1342         err = kstrtoint(strstrip(buffer), 0, &nice);
1343         if (err < 0)
1344                 return err;
1345
1346         p = get_proc_task(inode);
1347         if (!p)
1348                 return -ESRCH;
1349
1350         err = nice;
1351         err = proc_sched_autogroup_set_nice(p, &err);
1352         if (err)
1353                 count = err;
1354
1355         put_task_struct(p);
1356
1357         return count;
1358 }
1359
1360 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1361 {
1362         int ret;
1363
1364         ret = single_open(filp, sched_autogroup_show, NULL);
1365         if (!ret) {
1366                 struct seq_file *m = filp->private_data;
1367
1368                 m->private = inode;
1369         }
1370         return ret;
1371 }
1372
1373 static const struct file_operations proc_pid_sched_autogroup_operations = {
1374         .open           = sched_autogroup_open,
1375         .read           = seq_read,
1376         .write          = sched_autogroup_write,
1377         .llseek         = seq_lseek,
1378         .release        = single_release,
1379 };
1380
1381 #endif /* CONFIG_SCHED_AUTOGROUP */
1382
1383 static ssize_t comm_write(struct file *file, const char __user *buf,
1384                                 size_t count, loff_t *offset)
1385 {
1386         struct inode *inode = file->f_path.dentry->d_inode;
1387         struct task_struct *p;
1388         char buffer[TASK_COMM_LEN];
1389
1390         memset(buffer, 0, sizeof(buffer));
1391         if (count > sizeof(buffer) - 1)
1392                 count = sizeof(buffer) - 1;
1393         if (copy_from_user(buffer, buf, count))
1394                 return -EFAULT;
1395
1396         p = get_proc_task(inode);
1397         if (!p)
1398                 return -ESRCH;
1399
1400         if (same_thread_group(current, p))
1401                 set_task_comm(p, buffer);
1402         else
1403                 count = -EINVAL;
1404
1405         put_task_struct(p);
1406
1407         return count;
1408 }
1409
1410 static int comm_show(struct seq_file *m, void *v)
1411 {
1412         struct inode *inode = m->private;
1413         struct task_struct *p;
1414
1415         p = get_proc_task(inode);
1416         if (!p)
1417                 return -ESRCH;
1418
1419         task_lock(p);
1420         seq_printf(m, "%s\n", p->comm);
1421         task_unlock(p);
1422
1423         put_task_struct(p);
1424
1425         return 0;
1426 }
1427
1428 static int comm_open(struct inode *inode, struct file *filp)
1429 {
1430         return single_open(filp, comm_show, inode);
1431 }
1432
1433 static const struct file_operations proc_pid_set_comm_operations = {
1434         .open           = comm_open,
1435         .read           = seq_read,
1436         .write          = comm_write,
1437         .llseek         = seq_lseek,
1438         .release        = single_release,
1439 };
1440
1441 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1442 {
1443         struct task_struct *task;
1444         struct mm_struct *mm;
1445         struct file *exe_file;
1446
1447         task = get_proc_task(dentry->d_inode);
1448         if (!task)
1449                 return -ENOENT;
1450         mm = get_task_mm(task);
1451         put_task_struct(task);
1452         if (!mm)
1453                 return -ENOENT;
1454         exe_file = get_mm_exe_file(mm);
1455         mmput(mm);
1456         if (exe_file) {
1457                 *exe_path = exe_file->f_path;
1458                 path_get(&exe_file->f_path);
1459                 fput(exe_file);
1460                 return 0;
1461         } else
1462                 return -ENOENT;
1463 }
1464
1465 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1466 {
1467         struct inode *inode = dentry->d_inode;
1468         int error = -EACCES;
1469
1470         /* We don't need a base pointer in the /proc filesystem */
1471         path_put(&nd->path);
1472
1473         /* Are we allowed to snoop on the tasks file descriptors? */
1474         if (!proc_fd_access_allowed(inode))
1475                 goto out;
1476
1477         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1478 out:
1479         return ERR_PTR(error);
1480 }
1481
1482 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1483 {
1484         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1485         char *pathname;
1486         int len;
1487
1488         if (!tmp)
1489                 return -ENOMEM;
1490
1491         pathname = d_path(path, tmp, PAGE_SIZE);
1492         len = PTR_ERR(pathname);
1493         if (IS_ERR(pathname))
1494                 goto out;
1495         len = tmp + PAGE_SIZE - 1 - pathname;
1496
1497         if (len > buflen)
1498                 len = buflen;
1499         if (copy_to_user(buffer, pathname, len))
1500                 len = -EFAULT;
1501  out:
1502         free_page((unsigned long)tmp);
1503         return len;
1504 }
1505
1506 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1507 {
1508         int error = -EACCES;
1509         struct inode *inode = dentry->d_inode;
1510         struct path path;
1511
1512         /* Are we allowed to snoop on the tasks file descriptors? */
1513         if (!proc_fd_access_allowed(inode))
1514                 goto out;
1515
1516         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1517         if (error)
1518                 goto out;
1519
1520         error = do_proc_readlink(&path, buffer, buflen);
1521         path_put(&path);
1522 out:
1523         return error;
1524 }
1525
1526 static const struct inode_operations proc_pid_link_inode_operations = {
1527         .readlink       = proc_pid_readlink,
1528         .follow_link    = proc_pid_follow_link,
1529         .setattr        = proc_setattr,
1530 };
1531
1532
1533 /* building an inode */
1534
1535 static int task_dumpable(struct task_struct *task)
1536 {
1537         int dumpable = 0;
1538         struct mm_struct *mm;
1539
1540         task_lock(task);
1541         mm = task->mm;
1542         if (mm)
1543                 dumpable = get_dumpable(mm);
1544         task_unlock(task);
1545         if(dumpable == 1)
1546                 return 1;
1547         return 0;
1548 }
1549
1550 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1551 {
1552         struct inode * inode;
1553         struct proc_inode *ei;
1554         const struct cred *cred;
1555
1556         /* We need a new inode */
1557
1558         inode = new_inode(sb);
1559         if (!inode)
1560                 goto out;
1561
1562         /* Common stuff */
1563         ei = PROC_I(inode);
1564         inode->i_ino = get_next_ino();
1565         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1566         inode->i_op = &proc_def_inode_operations;
1567
1568         /*
1569          * grab the reference to task.
1570          */
1571         ei->pid = get_task_pid(task, PIDTYPE_PID);
1572         if (!ei->pid)
1573                 goto out_unlock;
1574
1575         if (task_dumpable(task)) {
1576                 rcu_read_lock();
1577                 cred = __task_cred(task);
1578                 inode->i_uid = cred->euid;
1579                 inode->i_gid = cred->egid;
1580                 rcu_read_unlock();
1581         }
1582         security_task_to_inode(task, inode);
1583
1584 out:
1585         return inode;
1586
1587 out_unlock:
1588         iput(inode);
1589         return NULL;
1590 }
1591
1592 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1593 {
1594         struct inode *inode = dentry->d_inode;
1595         struct task_struct *task;
1596         const struct cred *cred;
1597         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1598
1599         generic_fillattr(inode, stat);
1600
1601         rcu_read_lock();
1602         stat->uid = 0;
1603         stat->gid = 0;
1604         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1605         if (task) {
1606                 if (!has_pid_permissions(pid, task, 2)) {
1607                         rcu_read_unlock();
1608                         /*
1609                          * This doesn't prevent learning whether PID exists,
1610                          * it only makes getattr() consistent with readdir().
1611                          */
1612                         return -ENOENT;
1613                 }
1614                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1615                     task_dumpable(task)) {
1616                         cred = __task_cred(task);
1617                         stat->uid = cred->euid;
1618                         stat->gid = cred->egid;
1619                 }
1620         }
1621         rcu_read_unlock();
1622         return 0;
1623 }
1624
1625 /* dentry stuff */
1626
1627 /*
1628  *      Exceptional case: normally we are not allowed to unhash a busy
1629  * directory. In this case, however, we can do it - no aliasing problems
1630  * due to the way we treat inodes.
1631  *
1632  * Rewrite the inode's ownerships here because the owning task may have
1633  * performed a setuid(), etc.
1634  *
1635  * Before the /proc/pid/status file was created the only way to read
1636  * the effective uid of a /process was to stat /proc/pid.  Reading
1637  * /proc/pid/status is slow enough that procps and other packages
1638  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1639  * made this apply to all per process world readable and executable
1640  * directories.
1641  */
1642 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1643 {
1644         struct inode *inode;
1645         struct task_struct *task;
1646         const struct cred *cred;
1647
1648         if (nd && nd->flags & LOOKUP_RCU)
1649                 return -ECHILD;
1650
1651         inode = dentry->d_inode;
1652         task = get_proc_task(inode);
1653
1654         if (task) {
1655                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1656                     task_dumpable(task)) {
1657                         rcu_read_lock();
1658                         cred = __task_cred(task);
1659                         inode->i_uid = cred->euid;
1660                         inode->i_gid = cred->egid;
1661                         rcu_read_unlock();
1662                 } else {
1663                         inode->i_uid = 0;
1664                         inode->i_gid = 0;
1665                 }
1666                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1667                 security_task_to_inode(task, inode);
1668                 put_task_struct(task);
1669                 return 1;
1670         }
1671         d_drop(dentry);
1672         return 0;
1673 }
1674
1675 static int pid_delete_dentry(const struct dentry * dentry)
1676 {
1677         /* Is the task we represent dead?
1678          * If so, then don't put the dentry on the lru list,
1679          * kill it immediately.
1680          */
1681         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1682 }
1683
1684 const struct dentry_operations pid_dentry_operations =
1685 {
1686         .d_revalidate   = pid_revalidate,
1687         .d_delete       = pid_delete_dentry,
1688 };
1689
1690 /* Lookups */
1691
1692 /*
1693  * Fill a directory entry.
1694  *
1695  * If possible create the dcache entry and derive our inode number and
1696  * file type from dcache entry.
1697  *
1698  * Since all of the proc inode numbers are dynamically generated, the inode
1699  * numbers do not exist until the inode is cache.  This means creating the
1700  * the dcache entry in readdir is necessary to keep the inode numbers
1701  * reported by readdir in sync with the inode numbers reported
1702  * by stat.
1703  */
1704 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1705         const char *name, int len,
1706         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1707 {
1708         struct dentry *child, *dir = filp->f_path.dentry;
1709         struct inode *inode;
1710         struct qstr qname;
1711         ino_t ino = 0;
1712         unsigned type = DT_UNKNOWN;
1713
1714         qname.name = name;
1715         qname.len  = len;
1716         qname.hash = full_name_hash(name, len);
1717
1718         child = d_lookup(dir, &qname);
1719         if (!child) {
1720                 struct dentry *new;
1721                 new = d_alloc(dir, &qname);
1722                 if (new) {
1723                         child = instantiate(dir->d_inode, new, task, ptr);
1724                         if (child)
1725                                 dput(new);
1726                         else
1727                                 child = new;
1728                 }
1729         }
1730         if (!child || IS_ERR(child) || !child->d_inode)
1731                 goto end_instantiate;
1732         inode = child->d_inode;
1733         if (inode) {
1734                 ino = inode->i_ino;
1735                 type = inode->i_mode >> 12;
1736         }
1737         dput(child);
1738 end_instantiate:
1739         if (!ino)
1740                 ino = find_inode_number(dir, &qname);
1741         if (!ino)
1742                 ino = 1;
1743         return filldir(dirent, name, len, filp->f_pos, ino, type);
1744 }
1745
1746 static unsigned name_to_int(struct dentry *dentry)
1747 {
1748         const char *name = dentry->d_name.name;
1749         int len = dentry->d_name.len;
1750         unsigned n = 0;
1751
1752         if (len > 1 && *name == '0')
1753                 goto out;
1754         while (len-- > 0) {
1755                 unsigned c = *name++ - '0';
1756                 if (c > 9)
1757                         goto out;
1758                 if (n >= (~0U-9)/10)
1759                         goto out;
1760                 n *= 10;
1761                 n += c;
1762         }
1763         return n;
1764 out:
1765         return ~0U;
1766 }
1767
1768 #define PROC_FDINFO_MAX 64
1769
1770 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1771 {
1772         struct task_struct *task = get_proc_task(inode);
1773         struct files_struct *files = NULL;
1774         struct file *file;
1775         int fd = proc_fd(inode);
1776
1777         if (task) {
1778                 files = get_files_struct(task);
1779                 put_task_struct(task);
1780         }
1781         if (files) {
1782                 /*
1783                  * We are not taking a ref to the file structure, so we must
1784                  * hold ->file_lock.
1785                  */
1786                 spin_lock(&files->file_lock);
1787                 file = fcheck_files(files, fd);
1788                 if (file) {
1789                         unsigned int f_flags;
1790                         struct fdtable *fdt;
1791
1792                         fdt = files_fdtable(files);
1793                         f_flags = file->f_flags & ~O_CLOEXEC;
1794                         if (FD_ISSET(fd, fdt->close_on_exec))
1795                                 f_flags |= O_CLOEXEC;
1796
1797                         if (path) {
1798                                 *path = file->f_path;
1799                                 path_get(&file->f_path);
1800                         }
1801                         if (info)
1802                                 snprintf(info, PROC_FDINFO_MAX,
1803                                          "pos:\t%lli\n"
1804                                          "flags:\t0%o\n",
1805                                          (long long) file->f_pos,
1806                                          f_flags);
1807                         spin_unlock(&files->file_lock);
1808                         put_files_struct(files);
1809                         return 0;
1810                 }
1811                 spin_unlock(&files->file_lock);
1812                 put_files_struct(files);
1813         }
1814         return -ENOENT;
1815 }
1816
1817 static int proc_fd_link(struct dentry *dentry, struct path *path)
1818 {
1819         return proc_fd_info(dentry->d_inode, path, NULL);
1820 }
1821
1822 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1823 {
1824         struct inode *inode;
1825         struct task_struct *task;
1826         int fd;
1827         struct files_struct *files;
1828         const struct cred *cred;
1829
1830         if (nd && nd->flags & LOOKUP_RCU)
1831                 return -ECHILD;
1832
1833         inode = dentry->d_inode;
1834         task = get_proc_task(inode);
1835         fd = proc_fd(inode);
1836
1837         if (task) {
1838                 files = get_files_struct(task);
1839                 if (files) {
1840                         rcu_read_lock();
1841                         if (fcheck_files(files, fd)) {
1842                                 rcu_read_unlock();
1843                                 put_files_struct(files);
1844                                 if (task_dumpable(task)) {
1845                                         rcu_read_lock();
1846                                         cred = __task_cred(task);
1847                                         inode->i_uid = cred->euid;
1848                                         inode->i_gid = cred->egid;
1849                                         rcu_read_unlock();
1850                                 } else {
1851                                         inode->i_uid = 0;
1852                                         inode->i_gid = 0;
1853                                 }
1854                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1855                                 security_task_to_inode(task, inode);
1856                                 put_task_struct(task);
1857                                 return 1;
1858                         }
1859                         rcu_read_unlock();
1860                         put_files_struct(files);
1861                 }
1862                 put_task_struct(task);
1863         }
1864         d_drop(dentry);
1865         return 0;
1866 }
1867
1868 static const struct dentry_operations tid_fd_dentry_operations =
1869 {
1870         .d_revalidate   = tid_fd_revalidate,
1871         .d_delete       = pid_delete_dentry,
1872 };
1873
1874 static struct dentry *proc_fd_instantiate(struct inode *dir,
1875         struct dentry *dentry, struct task_struct *task, const void *ptr)
1876 {
1877         unsigned fd = *(const unsigned *)ptr;
1878         struct file *file;
1879         struct files_struct *files;
1880         struct inode *inode;
1881         struct proc_inode *ei;
1882         struct dentry *error = ERR_PTR(-ENOENT);
1883
1884         inode = proc_pid_make_inode(dir->i_sb, task);
1885         if (!inode)
1886                 goto out;
1887         ei = PROC_I(inode);
1888         ei->fd = fd;
1889         files = get_files_struct(task);
1890         if (!files)
1891                 goto out_iput;
1892         inode->i_mode = S_IFLNK;
1893
1894         /*
1895          * We are not taking a ref to the file structure, so we must
1896          * hold ->file_lock.
1897          */
1898         spin_lock(&files->file_lock);
1899         file = fcheck_files(files, fd);
1900         if (!file)
1901                 goto out_unlock;
1902         if (file->f_mode & FMODE_READ)
1903                 inode->i_mode |= S_IRUSR | S_IXUSR;
1904         if (file->f_mode & FMODE_WRITE)
1905                 inode->i_mode |= S_IWUSR | S_IXUSR;
1906         spin_unlock(&files->file_lock);
1907         put_files_struct(files);
1908
1909         inode->i_op = &proc_pid_link_inode_operations;
1910         inode->i_size = 64;
1911         ei->op.proc_get_link = proc_fd_link;
1912         d_set_d_op(dentry, &tid_fd_dentry_operations);
1913         d_add(dentry, inode);
1914         /* Close the race of the process dying before we return the dentry */
1915         if (tid_fd_revalidate(dentry, NULL))
1916                 error = NULL;
1917
1918  out:
1919         return error;
1920 out_unlock:
1921         spin_unlock(&files->file_lock);
1922         put_files_struct(files);
1923 out_iput:
1924         iput(inode);
1925         goto out;
1926 }
1927
1928 static struct dentry *proc_lookupfd_common(struct inode *dir,
1929                                            struct dentry *dentry,
1930                                            instantiate_t instantiate)
1931 {
1932         struct task_struct *task = get_proc_task(dir);
1933         unsigned fd = name_to_int(dentry);
1934         struct dentry *result = ERR_PTR(-ENOENT);
1935
1936         if (!task)
1937                 goto out_no_task;
1938         if (fd == ~0U)
1939                 goto out;
1940
1941         result = instantiate(dir, dentry, task, &fd);
1942 out:
1943         put_task_struct(task);
1944 out_no_task:
1945         return result;
1946 }
1947
1948 static int proc_readfd_common(struct file * filp, void * dirent,
1949                               filldir_t filldir, instantiate_t instantiate)
1950 {
1951         struct dentry *dentry = filp->f_path.dentry;
1952         struct inode *inode = dentry->d_inode;
1953         struct task_struct *p = get_proc_task(inode);
1954         unsigned int fd, ino;
1955         int retval;
1956         struct files_struct * files;
1957
1958         retval = -ENOENT;
1959         if (!p)
1960                 goto out_no_task;
1961         retval = 0;
1962
1963         fd = filp->f_pos;
1964         switch (fd) {
1965                 case 0:
1966                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1967                                 goto out;
1968                         filp->f_pos++;
1969                 case 1:
1970                         ino = parent_ino(dentry);
1971                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1972                                 goto out;
1973                         filp->f_pos++;
1974                 default:
1975                         files = get_files_struct(p);
1976                         if (!files)
1977                                 goto out;
1978                         rcu_read_lock();
1979                         for (fd = filp->f_pos-2;
1980                              fd < files_fdtable(files)->max_fds;
1981                              fd++, filp->f_pos++) {
1982                                 char name[PROC_NUMBUF];
1983                                 int len;
1984
1985                                 if (!fcheck_files(files, fd))
1986                                         continue;
1987                                 rcu_read_unlock();
1988
1989                                 len = snprintf(name, sizeof(name), "%d", fd);
1990                                 if (proc_fill_cache(filp, dirent, filldir,
1991                                                     name, len, instantiate,
1992                                                     p, &fd) < 0) {
1993                                         rcu_read_lock();
1994                                         break;
1995                                 }
1996                                 rcu_read_lock();
1997                         }
1998                         rcu_read_unlock();
1999                         put_files_struct(files);
2000         }
2001 out:
2002         put_task_struct(p);
2003 out_no_task:
2004         return retval;
2005 }
2006
2007 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2008                                     struct nameidata *nd)
2009 {
2010         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2011 }
2012
2013 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2014 {
2015         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2016 }
2017
2018 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2019                                       size_t len, loff_t *ppos)
2020 {
2021         char tmp[PROC_FDINFO_MAX];
2022         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2023         if (!err)
2024                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2025         return err;
2026 }
2027
2028 static const struct file_operations proc_fdinfo_file_operations = {
2029         .open           = nonseekable_open,
2030         .read           = proc_fdinfo_read,
2031         .llseek         = no_llseek,
2032 };
2033
2034 static const struct file_operations proc_fd_operations = {
2035         .read           = generic_read_dir,
2036         .readdir        = proc_readfd,
2037         .llseek         = default_llseek,
2038 };
2039
2040 #ifdef CONFIG_CHECKPOINT_RESTORE
2041
2042 /*
2043  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2044  * which represent vma start and end addresses.
2045  */
2046 static int dname_to_vma_addr(struct dentry *dentry,
2047                              unsigned long *start, unsigned long *end)
2048 {
2049         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2050                 return -EINVAL;
2051
2052         return 0;
2053 }
2054
2055 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2056 {
2057         unsigned long vm_start, vm_end;
2058         bool exact_vma_exists = false;
2059         struct mm_struct *mm = NULL;
2060         struct task_struct *task;
2061         const struct cred *cred;
2062         struct inode *inode;
2063         int status = 0;
2064
2065         if (nd && nd->flags & LOOKUP_RCU)
2066                 return -ECHILD;
2067
2068         if (!capable(CAP_SYS_ADMIN)) {
2069                 status = -EACCES;
2070                 goto out_notask;
2071         }
2072
2073         inode = dentry->d_inode;
2074         task = get_proc_task(inode);
2075         if (!task)
2076                 goto out_notask;
2077
2078         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2079                 goto out;
2080
2081         mm = get_task_mm(task);
2082         if (!mm)
2083                 goto out;
2084
2085         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2086                 down_read(&mm->mmap_sem);
2087                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2088                 up_read(&mm->mmap_sem);
2089         }
2090
2091         mmput(mm);
2092
2093         if (exact_vma_exists) {
2094                 if (task_dumpable(task)) {
2095                         rcu_read_lock();
2096                         cred = __task_cred(task);
2097                         inode->i_uid = cred->euid;
2098                         inode->i_gid = cred->egid;
2099                         rcu_read_unlock();
2100                 } else {
2101                         inode->i_uid = 0;
2102                         inode->i_gid = 0;
2103                 }
2104                 security_task_to_inode(task, inode);
2105                 status = 1;
2106         }
2107
2108 out:
2109         put_task_struct(task);
2110
2111 out_notask:
2112         if (status <= 0)
2113                 d_drop(dentry);
2114
2115         return status;
2116 }
2117
2118 static const struct dentry_operations tid_map_files_dentry_operations = {
2119         .d_revalidate   = map_files_d_revalidate,
2120         .d_delete       = pid_delete_dentry,
2121 };
2122
2123 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2124 {
2125         unsigned long vm_start, vm_end;
2126         struct vm_area_struct *vma;
2127         struct task_struct *task;
2128         struct mm_struct *mm;
2129         int rc;
2130
2131         rc = -ENOENT;
2132         task = get_proc_task(dentry->d_inode);
2133         if (!task)
2134                 goto out;
2135
2136         mm = get_task_mm(task);
2137         put_task_struct(task);
2138         if (!mm)
2139                 goto out;
2140
2141         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2142         if (rc)
2143                 goto out_mmput;
2144
2145         down_read(&mm->mmap_sem);
2146         vma = find_exact_vma(mm, vm_start, vm_end);
2147         if (vma && vma->vm_file) {
2148                 *path = vma->vm_file->f_path;
2149                 path_get(path);
2150                 rc = 0;
2151         }
2152         up_read(&mm->mmap_sem);
2153
2154 out_mmput:
2155         mmput(mm);
2156 out:
2157         return rc;
2158 }
2159
2160 struct map_files_info {
2161         struct file     *file;
2162         unsigned long   len;
2163         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2164 };
2165
2166 static struct dentry *
2167 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2168                            struct task_struct *task, const void *ptr)
2169 {
2170         const struct file *file = ptr;
2171         struct proc_inode *ei;
2172         struct inode *inode;
2173
2174         if (!file)
2175                 return ERR_PTR(-ENOENT);
2176
2177         inode = proc_pid_make_inode(dir->i_sb, task);
2178         if (!inode)
2179                 return ERR_PTR(-ENOENT);
2180
2181         ei = PROC_I(inode);
2182         ei->op.proc_get_link = proc_map_files_get_link;
2183
2184         inode->i_op = &proc_pid_link_inode_operations;
2185         inode->i_size = 64;
2186         inode->i_mode = S_IFLNK;
2187
2188         if (file->f_mode & FMODE_READ)
2189                 inode->i_mode |= S_IRUSR;
2190         if (file->f_mode & FMODE_WRITE)
2191                 inode->i_mode |= S_IWUSR;
2192
2193         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2194         d_add(dentry, inode);
2195
2196         return NULL;
2197 }
2198
2199 static struct dentry *proc_map_files_lookup(struct inode *dir,
2200                 struct dentry *dentry, struct nameidata *nd)
2201 {
2202         unsigned long vm_start, vm_end;
2203         struct vm_area_struct *vma;
2204         struct task_struct *task;
2205         struct dentry *result;
2206         struct mm_struct *mm;
2207
2208         result = ERR_PTR(-EACCES);
2209         if (!capable(CAP_SYS_ADMIN))
2210                 goto out;
2211
2212         result = ERR_PTR(-ENOENT);
2213         task = get_proc_task(dir);
2214         if (!task)
2215                 goto out;
2216
2217         result = ERR_PTR(-EACCES);
2218         if (lock_trace(task))
2219                 goto out_put_task;
2220
2221         result = ERR_PTR(-ENOENT);
2222         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2223                 goto out_unlock;
2224
2225         mm = get_task_mm(task);
2226         if (!mm)
2227                 goto out_unlock;
2228
2229         down_read(&mm->mmap_sem);
2230         vma = find_exact_vma(mm, vm_start, vm_end);
2231         if (!vma)
2232                 goto out_no_vma;
2233
2234         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2235
2236 out_no_vma:
2237         up_read(&mm->mmap_sem);
2238         mmput(mm);
2239 out_unlock:
2240         unlock_trace(task);
2241 out_put_task:
2242         put_task_struct(task);
2243 out:
2244         return result;
2245 }
2246
2247 static const struct inode_operations proc_map_files_inode_operations = {
2248         .lookup         = proc_map_files_lookup,
2249         .permission     = proc_fd_permission,
2250         .setattr        = proc_setattr,
2251 };
2252
2253 static int
2254 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2255 {
2256         struct dentry *dentry = filp->f_path.dentry;
2257         struct inode *inode = dentry->d_inode;
2258         struct vm_area_struct *vma;
2259         struct task_struct *task;
2260         struct mm_struct *mm;
2261         ino_t ino;
2262         int ret;
2263
2264         ret = -EACCES;
2265         if (!capable(CAP_SYS_ADMIN))
2266                 goto out;
2267
2268         ret = -ENOENT;
2269         task = get_proc_task(inode);
2270         if (!task)
2271                 goto out;
2272
2273         ret = -EACCES;
2274         if (lock_trace(task))
2275                 goto out_put_task;
2276
2277         ret = 0;
2278         switch (filp->f_pos) {
2279         case 0:
2280                 ino = inode->i_ino;
2281                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2282                         goto out_unlock;
2283                 filp->f_pos++;
2284         case 1:
2285                 ino = parent_ino(dentry);
2286                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2287                         goto out_unlock;
2288                 filp->f_pos++;
2289         default:
2290         {
2291                 unsigned long nr_files, pos, i;
2292                 struct flex_array *fa = NULL;
2293                 struct map_files_info info;
2294                 struct map_files_info *p;
2295
2296                 mm = get_task_mm(task);
2297                 if (!mm)
2298                         goto out_unlock;
2299                 down_read(&mm->mmap_sem);
2300
2301                 nr_files = 0;
2302
2303                 /*
2304                  * We need two passes here:
2305                  *
2306                  *  1) Collect vmas of mapped files with mmap_sem taken
2307                  *  2) Release mmap_sem and instantiate entries
2308                  *
2309                  * otherwise we get lockdep complained, since filldir()
2310                  * routine might require mmap_sem taken in might_fault().
2311                  */
2312
2313                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2314                         if (vma->vm_file && ++pos > filp->f_pos)
2315                                 nr_files++;
2316                 }
2317
2318                 if (nr_files) {
2319                         fa = flex_array_alloc(sizeof(info), nr_files,
2320                                                 GFP_KERNEL);
2321                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2322                                                         GFP_KERNEL)) {
2323                                 ret = -ENOMEM;
2324                                 if (fa)
2325                                         flex_array_free(fa);
2326                                 up_read(&mm->mmap_sem);
2327                                 mmput(mm);
2328                                 goto out_unlock;
2329                         }
2330                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2331                                         vma = vma->vm_next) {
2332                                 if (!vma->vm_file)
2333                                         continue;
2334                                 if (++pos <= filp->f_pos)
2335                                         continue;
2336
2337                                 get_file(vma->vm_file);
2338                                 info.file = vma->vm_file;
2339                                 info.len = snprintf(info.name,
2340                                                 sizeof(info.name), "%lx-%lx",
2341                                                 vma->vm_start, vma->vm_end);
2342                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2343                                         BUG();
2344                         }
2345                 }
2346                 up_read(&mm->mmap_sem);
2347
2348                 for (i = 0; i < nr_files; i++) {
2349                         p = flex_array_get(fa, i);
2350                         ret = proc_fill_cache(filp, dirent, filldir,
2351                                               p->name, p->len,
2352                                               proc_map_files_instantiate,
2353                                               task, p->file);
2354                         if (ret)
2355                                 break;
2356                         filp->f_pos++;
2357                         fput(p->file);
2358                 }
2359                 for (; i < nr_files; i++) {
2360                         /*
2361                          * In case of error don't forget
2362                          * to put rest of file refs.
2363                          */
2364                         p = flex_array_get(fa, i);
2365                         fput(p->file);
2366                 }
2367                 if (fa)
2368                         flex_array_free(fa);
2369                 mmput(mm);
2370         }
2371         }
2372
2373 out_unlock:
2374         unlock_trace(task);
2375 out_put_task:
2376         put_task_struct(task);
2377 out:
2378         return ret;
2379 }
2380
2381 static const struct file_operations proc_map_files_operations = {
2382         .read           = generic_read_dir,
2383         .readdir        = proc_map_files_readdir,
2384         .llseek         = default_llseek,
2385 };
2386
2387 #endif /* CONFIG_CHECKPOINT_RESTORE */
2388
2389 /*
2390  * /proc/pid/fd needs a special permission handler so that a process can still
2391  * access /proc/self/fd after it has executed a setuid().
2392  */
2393 static int proc_fd_permission(struct inode *inode, int mask)
2394 {
2395         int rv = generic_permission(inode, mask);
2396         if (rv == 0)
2397                 return 0;
2398         if (task_pid(current) == proc_pid(inode))
2399                 rv = 0;
2400         return rv;
2401 }
2402
2403 /*
2404  * proc directories can do almost nothing..
2405  */
2406 static const struct inode_operations proc_fd_inode_operations = {
2407         .lookup         = proc_lookupfd,
2408         .permission     = proc_fd_permission,
2409         .setattr        = proc_setattr,
2410 };
2411
2412 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2413         struct dentry *dentry, struct task_struct *task, const void *ptr)
2414 {
2415         unsigned fd = *(unsigned *)ptr;
2416         struct inode *inode;
2417         struct proc_inode *ei;
2418         struct dentry *error = ERR_PTR(-ENOENT);
2419
2420         inode = proc_pid_make_inode(dir->i_sb, task);
2421         if (!inode)
2422                 goto out;
2423         ei = PROC_I(inode);
2424         ei->fd = fd;
2425         inode->i_mode = S_IFREG | S_IRUSR;
2426         inode->i_fop = &proc_fdinfo_file_operations;
2427         d_set_d_op(dentry, &tid_fd_dentry_operations);
2428         d_add(dentry, inode);
2429         /* Close the race of the process dying before we return the dentry */
2430         if (tid_fd_revalidate(dentry, NULL))
2431                 error = NULL;
2432
2433  out:
2434         return error;
2435 }
2436
2437 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2438                                         struct dentry *dentry,
2439                                         struct nameidata *nd)
2440 {
2441         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2442 }
2443
2444 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2445 {
2446         return proc_readfd_common(filp, dirent, filldir,
2447                                   proc_fdinfo_instantiate);
2448 }
2449
2450 static const struct file_operations proc_fdinfo_operations = {
2451         .read           = generic_read_dir,
2452         .readdir        = proc_readfdinfo,
2453         .llseek         = default_llseek,
2454 };
2455
2456 /*
2457  * proc directories can do almost nothing..
2458  */
2459 static const struct inode_operations proc_fdinfo_inode_operations = {
2460         .lookup         = proc_lookupfdinfo,
2461         .setattr        = proc_setattr,
2462 };
2463
2464
2465 static struct dentry *proc_pident_instantiate(struct inode *dir,
2466         struct dentry *dentry, struct task_struct *task, const void *ptr)
2467 {
2468         const struct pid_entry *p = ptr;
2469         struct inode *inode;
2470         struct proc_inode *ei;
2471         struct dentry *error = ERR_PTR(-ENOENT);
2472
2473         inode = proc_pid_make_inode(dir->i_sb, task);
2474         if (!inode)
2475                 goto out;
2476
2477         ei = PROC_I(inode);
2478         inode->i_mode = p->mode;
2479         if (S_ISDIR(inode->i_mode))
2480                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2481         if (p->iop)
2482                 inode->i_op = p->iop;
2483         if (p->fop)
2484                 inode->i_fop = p->fop;
2485         ei->op = p->op;
2486         d_set_d_op(dentry, &pid_dentry_operations);
2487         d_add(dentry, inode);
2488         /* Close the race of the process dying before we return the dentry */
2489         if (pid_revalidate(dentry, NULL))
2490                 error = NULL;
2491 out:
2492         return error;
2493 }
2494
2495 static struct dentry *proc_pident_lookup(struct inode *dir,
2496                                          struct dentry *dentry,
2497                                          const struct pid_entry *ents,
2498                                          unsigned int nents)
2499 {
2500         struct dentry *error;
2501         struct task_struct *task = get_proc_task(dir);
2502         const struct pid_entry *p, *last;
2503
2504         error = ERR_PTR(-ENOENT);
2505
2506         if (!task)
2507                 goto out_no_task;
2508
2509         /*
2510          * Yes, it does not scale. And it should not. Don't add
2511          * new entries into /proc/<tgid>/ without very good reasons.
2512          */
2513         last = &ents[nents - 1];
2514         for (p = ents; p <= last; p++) {
2515                 if (p->len != dentry->d_name.len)
2516                         continue;
2517                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2518                         break;
2519         }
2520         if (p > last)
2521                 goto out;
2522
2523         error = proc_pident_instantiate(dir, dentry, task, p);
2524 out:
2525         put_task_struct(task);
2526 out_no_task:
2527         return error;
2528 }
2529
2530 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2531         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2532 {
2533         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2534                                 proc_pident_instantiate, task, p);
2535 }
2536
2537 static int proc_pident_readdir(struct file *filp,
2538                 void *dirent, filldir_t filldir,
2539                 const struct pid_entry *ents, unsigned int nents)
2540 {
2541         int i;
2542         struct dentry *dentry = filp->f_path.dentry;
2543         struct inode *inode = dentry->d_inode;
2544         struct task_struct *task = get_proc_task(inode);
2545         const struct pid_entry *p, *last;
2546         ino_t ino;
2547         int ret;
2548
2549         ret = -ENOENT;
2550         if (!task)
2551                 goto out_no_task;
2552
2553         ret = 0;
2554         i = filp->f_pos;
2555         switch (i) {
2556         case 0:
2557                 ino = inode->i_ino;
2558                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2559                         goto out;
2560                 i++;
2561                 filp->f_pos++;
2562                 /* fall through */
2563         case 1:
2564                 ino = parent_ino(dentry);
2565                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2566                         goto out;
2567                 i++;
2568                 filp->f_pos++;
2569                 /* fall through */
2570         default:
2571                 i -= 2;
2572                 if (i >= nents) {
2573                         ret = 1;
2574                         goto out;
2575                 }
2576                 p = ents + i;
2577                 last = &ents[nents - 1];
2578                 while (p <= last) {
2579                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2580                                 goto out;
2581                         filp->f_pos++;
2582                         p++;
2583                 }
2584         }
2585
2586         ret = 1;
2587 out:
2588         put_task_struct(task);
2589 out_no_task:
2590         return ret;
2591 }
2592
2593 #ifdef CONFIG_SECURITY
2594 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2595                                   size_t count, loff_t *ppos)
2596 {
2597         struct inode * inode = file->f_path.dentry->d_inode;
2598         char *p = NULL;
2599         ssize_t length;
2600         struct task_struct *task = get_proc_task(inode);
2601
2602         if (!task)
2603                 return -ESRCH;
2604
2605         length = security_getprocattr(task,
2606                                       (char*)file->f_path.dentry->d_name.name,
2607                                       &p);
2608         put_task_struct(task);
2609         if (length > 0)
2610                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2611         kfree(p);
2612         return length;
2613 }
2614
2615 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2616                                    size_t count, loff_t *ppos)
2617 {
2618         struct inode * inode = file->f_path.dentry->d_inode;
2619         char *page;
2620         ssize_t length;
2621         struct task_struct *task = get_proc_task(inode);
2622
2623         length = -ESRCH;
2624         if (!task)
2625                 goto out_no_task;
2626         if (count > PAGE_SIZE)
2627                 count = PAGE_SIZE;
2628
2629         /* No partial writes. */
2630         length = -EINVAL;
2631         if (*ppos != 0)
2632                 goto out;
2633
2634         length = -ENOMEM;
2635         page = (char*)__get_free_page(GFP_TEMPORARY);
2636         if (!page)
2637                 goto out;
2638
2639         length = -EFAULT;
2640         if (copy_from_user(page, buf, count))
2641                 goto out_free;
2642
2643         /* Guard against adverse ptrace interaction */
2644         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2645         if (length < 0)
2646                 goto out_free;
2647
2648         length = security_setprocattr(task,
2649                                       (char*)file->f_path.dentry->d_name.name,
2650                                       (void*)page, count);
2651         mutex_unlock(&task->signal->cred_guard_mutex);
2652 out_free:
2653         free_page((unsigned long) page);
2654 out:
2655         put_task_struct(task);
2656 out_no_task:
2657         return length;
2658 }
2659
2660 static const struct file_operations proc_pid_attr_operations = {
2661         .read           = proc_pid_attr_read,
2662         .write          = proc_pid_attr_write,
2663         .llseek         = generic_file_llseek,
2664 };
2665
2666 static const struct pid_entry attr_dir_stuff[] = {
2667         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2668         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2669         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2670         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2671         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2672         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2673 };
2674
2675 static int proc_attr_dir_readdir(struct file * filp,
2676                              void * dirent, filldir_t filldir)
2677 {
2678         return proc_pident_readdir(filp,dirent,filldir,
2679                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2680 }
2681
2682 static const struct file_operations proc_attr_dir_operations = {
2683         .read           = generic_read_dir,
2684         .readdir        = proc_attr_dir_readdir,
2685         .llseek         = default_llseek,
2686 };
2687
2688 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2689                                 struct dentry *dentry, struct nameidata *nd)
2690 {
2691         return proc_pident_lookup(dir, dentry,
2692                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2693 }
2694
2695 static const struct inode_operations proc_attr_dir_inode_operations = {
2696         .lookup         = proc_attr_dir_lookup,
2697         .getattr        = pid_getattr,
2698         .setattr        = proc_setattr,
2699 };
2700
2701 #endif
2702
2703 #ifdef CONFIG_ELF_CORE
2704 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2705                                          size_t count, loff_t *ppos)
2706 {
2707         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2708         struct mm_struct *mm;
2709         char buffer[PROC_NUMBUF];
2710         size_t len;
2711         int ret;
2712
2713         if (!task)
2714                 return -ESRCH;
2715
2716         ret = 0;
2717         mm = get_task_mm(task);
2718         if (mm) {
2719                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2720                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2721                                 MMF_DUMP_FILTER_SHIFT));
2722                 mmput(mm);
2723                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2724         }
2725
2726         put_task_struct(task);
2727
2728         return ret;
2729 }
2730
2731 static ssize_t proc_coredump_filter_write(struct file *file,
2732                                           const char __user *buf,
2733                                           size_t count,
2734                                           loff_t *ppos)
2735 {
2736         struct task_struct *task;
2737         struct mm_struct *mm;
2738         char buffer[PROC_NUMBUF], *end;
2739         unsigned int val;
2740         int ret;
2741         int i;
2742         unsigned long mask;
2743
2744         ret = -EFAULT;
2745         memset(buffer, 0, sizeof(buffer));
2746         if (count > sizeof(buffer) - 1)
2747                 count = sizeof(buffer) - 1;
2748         if (copy_from_user(buffer, buf, count))
2749                 goto out_no_task;
2750
2751         ret = -EINVAL;
2752         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2753         if (*end == '\n')
2754                 end++;
2755         if (end - buffer == 0)
2756                 goto out_no_task;
2757
2758         ret = -ESRCH;
2759         task = get_proc_task(file->f_dentry->d_inode);
2760         if (!task)
2761                 goto out_no_task;
2762
2763         ret = end - buffer;
2764         mm = get_task_mm(task);
2765         if (!mm)
2766                 goto out_no_mm;
2767
2768         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2769                 if (val & mask)
2770                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2771                 else
2772                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2773         }
2774
2775         mmput(mm);
2776  out_no_mm:
2777         put_task_struct(task);
2778  out_no_task:
2779         return ret;
2780 }
2781
2782 static const struct file_operations proc_coredump_filter_operations = {
2783         .read           = proc_coredump_filter_read,
2784         .write          = proc_coredump_filter_write,
2785         .llseek         = generic_file_llseek,
2786 };
2787 #endif
2788
2789 /*
2790  * /proc/self:
2791  */
2792 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2793                               int buflen)
2794 {
2795         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2796         pid_t tgid = task_tgid_nr_ns(current, ns);
2797         char tmp[PROC_NUMBUF];
2798         if (!tgid)
2799                 return -ENOENT;
2800         sprintf(tmp, "%d", tgid);
2801         return vfs_readlink(dentry,buffer,buflen,tmp);
2802 }
2803
2804 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2805 {
2806         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2807         pid_t tgid = task_tgid_nr_ns(current, ns);
2808         char *name = ERR_PTR(-ENOENT);
2809         if (tgid) {
2810                 name = __getname();
2811                 if (!name)
2812                         name = ERR_PTR(-ENOMEM);
2813                 else
2814                         sprintf(name, "%d", tgid);
2815         }
2816         nd_set_link(nd, name);
2817         return NULL;
2818 }
2819
2820 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2821                                 void *cookie)
2822 {
2823         char *s = nd_get_link(nd);
2824         if (!IS_ERR(s))
2825                 __putname(s);
2826 }
2827
2828 static const struct inode_operations proc_self_inode_operations = {
2829         .readlink       = proc_self_readlink,
2830         .follow_link    = proc_self_follow_link,
2831         .put_link       = proc_self_put_link,
2832 };
2833
2834 /*
2835  * proc base
2836  *
2837  * These are the directory entries in the root directory of /proc
2838  * that properly belong to the /proc filesystem, as they describe
2839  * describe something that is process related.
2840  */
2841 static const struct pid_entry proc_base_stuff[] = {
2842         NOD("self", S_IFLNK|S_IRWXUGO,
2843                 &proc_self_inode_operations, NULL, {}),
2844 };
2845
2846 static struct dentry *proc_base_instantiate(struct inode *dir,
2847         struct dentry *dentry, struct task_struct *task, const void *ptr)
2848 {
2849         const struct pid_entry *p = ptr;
2850         struct inode *inode;
2851         struct proc_inode *ei;
2852         struct dentry *error;
2853
2854         /* Allocate the inode */
2855         error = ERR_PTR(-ENOMEM);
2856         inode = new_inode(dir->i_sb);
2857         if (!inode)
2858                 goto out;
2859
2860         /* Initialize the inode */
2861         ei = PROC_I(inode);
2862         inode->i_ino = get_next_ino();
2863         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2864
2865         /*
2866          * grab the reference to the task.
2867          */
2868         ei->pid = get_task_pid(task, PIDTYPE_PID);
2869         if (!ei->pid)
2870                 goto out_iput;
2871
2872         inode->i_mode = p->mode;
2873         if (S_ISDIR(inode->i_mode))
2874                 set_nlink(inode, 2);
2875         if (S_ISLNK(inode->i_mode))
2876                 inode->i_size = 64;
2877         if (p->iop)
2878                 inode->i_op = p->iop;
2879         if (p->fop)
2880                 inode->i_fop = p->fop;
2881         ei->op = p->op;
2882         d_add(dentry, inode);
2883         error = NULL;
2884 out:
2885         return error;
2886 out_iput:
2887         iput(inode);
2888         goto out;
2889 }
2890
2891 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2892 {
2893         struct dentry *error;
2894         struct task_struct *task = get_proc_task(dir);
2895         const struct pid_entry *p, *last;
2896
2897         error = ERR_PTR(-ENOENT);
2898
2899         if (!task)
2900                 goto out_no_task;
2901
2902         /* Lookup the directory entry */
2903         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2904         for (p = proc_base_stuff; p <= last; p++) {
2905                 if (p->len != dentry->d_name.len)
2906                         continue;
2907                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2908                         break;
2909         }
2910         if (p > last)
2911                 goto out;
2912
2913         error = proc_base_instantiate(dir, dentry, task, p);
2914
2915 out:
2916         put_task_struct(task);
2917 out_no_task:
2918         return error;
2919 }
2920
2921 static int proc_base_fill_cache(struct file *filp, void *dirent,
2922         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2923 {
2924         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2925                                 proc_base_instantiate, task, p);
2926 }
2927
2928 #ifdef CONFIG_TASK_IO_ACCOUNTING
2929 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2930 {
2931         struct task_io_accounting acct = task->ioac;
2932         unsigned long flags;
2933         int result;
2934
2935         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2936         if (result)
2937                 return result;
2938
2939         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2940                 result = -EACCES;
2941                 goto out_unlock;
2942         }
2943
2944         if (whole && lock_task_sighand(task, &flags)) {
2945                 struct task_struct *t = task;
2946
2947                 task_io_accounting_add(&acct, &task->signal->ioac);
2948                 while_each_thread(task, t)
2949                         task_io_accounting_add(&acct, &t->ioac);
2950
2951                 unlock_task_sighand(task, &flags);
2952         }
2953         result = sprintf(buffer,
2954                         "rchar: %llu\n"
2955                         "wchar: %llu\n"
2956                         "syscr: %llu\n"
2957                         "syscw: %llu\n"
2958                         "read_bytes: %llu\n"
2959                         "write_bytes: %llu\n"
2960                         "cancelled_write_bytes: %llu\n",
2961                         (unsigned long long)acct.rchar,
2962                         (unsigned long long)acct.wchar,
2963                         (unsigned long long)acct.syscr,
2964                         (unsigned long long)acct.syscw,
2965                         (unsigned long long)acct.read_bytes,
2966                         (unsigned long long)acct.write_bytes,
2967                         (unsigned long long)acct.cancelled_write_bytes);
2968 out_unlock:
2969         mutex_unlock(&task->signal->cred_guard_mutex);
2970         return result;
2971 }
2972
2973 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2974 {
2975         return do_io_accounting(task, buffer, 0);
2976 }
2977
2978 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2979 {
2980         return do_io_accounting(task, buffer, 1);
2981 }
2982 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2983
2984 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2985                                 struct pid *pid, struct task_struct *task)
2986 {
2987         int err = lock_trace(task);
2988         if (!err) {
2989                 seq_printf(m, "%08x\n", task->personality);
2990                 unlock_trace(task);
2991         }
2992         return err;
2993 }
2994
2995 /*
2996  * Thread groups
2997  */
2998 static const struct file_operations proc_task_operations;
2999 static const struct inode_operations proc_task_inode_operations;
3000
3001 static const struct pid_entry tgid_base_stuff[] = {
3002         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3003         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3004 #ifdef CONFIG_CHECKPOINT_RESTORE
3005         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3006 #endif
3007         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3008         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3009 #ifdef CONFIG_NET
3010         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3011 #endif
3012         REG("environ",    S_IRUSR, proc_environ_operations),
3013         INF("auxv",       S_IRUSR, proc_pid_auxv),
3014         ONE("status",     S_IRUGO, proc_pid_status),
3015         ONE("personality", S_IRUGO, proc_pid_personality),
3016         INF("limits",     S_IRUGO, proc_pid_limits),
3017 #ifdef CONFIG_SCHED_DEBUG
3018         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3019 #endif
3020 #ifdef CONFIG_SCHED_AUTOGROUP
3021         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3022 #endif
3023         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3024 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3025         INF("syscall",    S_IRUGO, proc_pid_syscall),
3026 #endif
3027         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3028         ONE("stat",       S_IRUGO, proc_tgid_stat),
3029         ONE("statm",      S_IRUGO, proc_pid_statm),
3030         REG("maps",       S_IRUGO, proc_maps_operations),
3031 #ifdef CONFIG_NUMA
3032         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
3033 #endif
3034         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3035         LNK("cwd",        proc_cwd_link),
3036         LNK("root",       proc_root_link),
3037         LNK("exe",        proc_exe_link),
3038         REG("mounts",     S_IRUGO, proc_mounts_operations),
3039         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3040         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3041 #ifdef CONFIG_PROC_PAGE_MONITOR
3042         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3043         REG("smaps",      S_IRUGO, proc_smaps_operations),
3044         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3045 #endif
3046 #ifdef CONFIG_SECURITY
3047         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3048 #endif
3049 #ifdef CONFIG_KALLSYMS
3050         INF("wchan",      S_IRUGO, proc_pid_wchan),
3051 #endif
3052 #ifdef CONFIG_STACKTRACE
3053         ONE("stack",      S_IRUGO, proc_pid_stack),
3054 #endif
3055 #ifdef CONFIG_SCHEDSTATS
3056         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3057 #endif
3058 #ifdef CONFIG_LATENCYTOP
3059         REG("latency",  S_IRUGO, proc_lstats_operations),
3060 #endif
3061 #ifdef CONFIG_PROC_PID_CPUSET
3062         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3063 #endif
3064 #ifdef CONFIG_CGROUPS
3065         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3066 #endif
3067         INF("oom_score",  S_IRUGO, proc_oom_score),
3068         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3069         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3070 #ifdef CONFIG_AUDITSYSCALL
3071         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3072         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3073 #endif
3074 #ifdef CONFIG_FAULT_INJECTION
3075         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3076 #endif
3077 #ifdef CONFIG_ELF_CORE
3078         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3079 #endif
3080 #ifdef CONFIG_TASK_IO_ACCOUNTING
3081         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3082 #endif
3083 #ifdef CONFIG_HARDWALL
3084         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3085 #endif
3086 };
3087
3088 static int proc_tgid_base_readdir(struct file * filp,
3089                              void * dirent, filldir_t filldir)
3090 {
3091         return proc_pident_readdir(filp,dirent,filldir,
3092                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3093 }
3094
3095 static const struct file_operations proc_tgid_base_operations = {
3096         .read           = generic_read_dir,
3097         .readdir        = proc_tgid_base_readdir,
3098         .llseek         = default_llseek,
3099 };
3100
3101 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3102         return proc_pident_lookup(dir, dentry,
3103                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3104 }
3105
3106 static const struct inode_operations proc_tgid_base_inode_operations = {
3107         .lookup         = proc_tgid_base_lookup,
3108         .getattr        = pid_getattr,
3109         .setattr        = proc_setattr,
3110         .permission     = proc_pid_permission,
3111 };
3112
3113 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3114 {
3115         struct dentry *dentry, *leader, *dir;
3116         char buf[PROC_NUMBUF];
3117         struct qstr name;
3118
3119         name.name = buf;
3120         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3121         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3122         if (dentry) {
3123                 shrink_dcache_parent(dentry);
3124                 d_drop(dentry);
3125                 dput(dentry);
3126         }
3127
3128         name.name = buf;
3129         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3130         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3131         if (!leader)
3132                 goto out;
3133
3134         name.name = "task";
3135         name.len = strlen(name.name);
3136         dir = d_hash_and_lookup(leader, &name);
3137         if (!dir)
3138                 goto out_put_leader;
3139
3140         name.name = buf;
3141         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3142         dentry = d_hash_and_lookup(dir, &name);
3143         if (dentry) {
3144                 shrink_dcache_parent(dentry);
3145                 d_drop(dentry);
3146                 dput(dentry);
3147         }
3148
3149         dput(dir);
3150 out_put_leader:
3151         dput(leader);
3152 out:
3153         return;
3154 }
3155
3156 /**
3157  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3158  * @task: task that should be flushed.
3159  *
3160  * When flushing dentries from proc, one needs to flush them from global
3161  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3162  * in. This call is supposed to do all of this job.
3163  *
3164  * Looks in the dcache for
3165  * /proc/@pid
3166  * /proc/@tgid/task/@pid
3167  * if either directory is present flushes it and all of it'ts children
3168  * from the dcache.
3169  *
3170  * It is safe and reasonable to cache /proc entries for a task until
3171  * that task exits.  After that they just clog up the dcache with
3172  * useless entries, possibly causing useful dcache entries to be
3173  * flushed instead.  This routine is proved to flush those useless
3174  * dcache entries at process exit time.
3175  *
3176  * NOTE: This routine is just an optimization so it does not guarantee
3177  *       that no dcache entries will exist at process exit time it
3178  *       just makes it very unlikely that any will persist.
3179  */
3180
3181 void proc_flush_task(struct task_struct *task)
3182 {
3183         int i;
3184         struct pid *pid, *tgid;
3185         struct upid *upid;
3186
3187         pid = task_pid(task);
3188         tgid = task_tgid(task);
3189
3190         for (i = 0; i <= pid->level; i++) {
3191                 upid = &pid->numbers[i];
3192                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3193                                         tgid->numbers[i].nr);
3194         }
3195
3196         upid = &pid->numbers[pid->level];
3197         if (upid->nr == 1)
3198                 pid_ns_release_proc(upid->ns);
3199 }
3200
3201 static struct dentry *proc_pid_instantiate(struct inode *dir,
3202                                            struct dentry * dentry,
3203                                            struct task_struct *task, const void *ptr)
3204 {
3205         struct dentry *error = ERR_PTR(-ENOENT);
3206         struct inode *inode;
3207
3208         inode = proc_pid_make_inode(dir->i_sb, task);
3209         if (!inode)
3210                 goto out;
3211
3212         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3213         inode->i_op = &proc_tgid_base_inode_operations;
3214         inode->i_fop = &proc_tgid_base_operations;
3215         inode->i_flags|=S_IMMUTABLE;
3216
3217         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3218                                                   ARRAY_SIZE(tgid_base_stuff)));
3219
3220         d_set_d_op(dentry, &pid_dentry_operations);
3221
3222         d_add(dentry, inode);
3223         /* Close the race of the process dying before we return the dentry */
3224         if (pid_revalidate(dentry, NULL))
3225                 error = NULL;
3226 out:
3227         return error;
3228 }
3229
3230 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3231 {
3232         struct dentry *result;
3233         struct task_struct *task;
3234         unsigned tgid;
3235         struct pid_namespace *ns;
3236
3237         result = proc_base_lookup(dir, dentry);
3238         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3239                 goto out;
3240
3241         tgid = name_to_int(dentry);
3242         if (tgid == ~0U)
3243                 goto out;
3244
3245         ns = dentry->d_sb->s_fs_info;
3246         rcu_read_lock();
3247         task = find_task_by_pid_ns(tgid, ns);
3248         if (task)
3249                 get_task_struct(task);
3250         rcu_read_unlock();
3251         if (!task)
3252                 goto out;
3253
3254         result = proc_pid_instantiate(dir, dentry, task, NULL);
3255         put_task_struct(task);
3256 out:
3257         return result;
3258 }
3259
3260 /*
3261  * Find the first task with tgid >= tgid
3262  *
3263  */
3264 struct tgid_iter {
3265         unsigned int tgid;
3266         struct task_struct *task;
3267 };
3268 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3269 {
3270         struct pid *pid;
3271
3272         if (iter.task)
3273                 put_task_struct(iter.task);
3274         rcu_read_lock();
3275 retry:
3276         iter.task = NULL;
3277         pid = find_ge_pid(iter.tgid, ns);
3278         if (pid) {
3279                 iter.tgid = pid_nr_ns(pid, ns);
3280                 iter.task = pid_task(pid, PIDTYPE_PID);
3281                 /* What we to know is if the pid we have find is the
3282                  * pid of a thread_group_leader.  Testing for task
3283                  * being a thread_group_leader is the obvious thing
3284                  * todo but there is a window when it fails, due to
3285                  * the pid transfer logic in de_thread.
3286                  *
3287                  * So we perform the straight forward test of seeing
3288                  * if the pid we have found is the pid of a thread
3289                  * group leader, and don't worry if the task we have
3290                  * found doesn't happen to be a thread group leader.
3291                  * As we don't care in the case of readdir.
3292                  */
3293                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3294                         iter.tgid += 1;
3295                         goto retry;
3296                 }
3297                 get_task_struct(iter.task);
3298         }
3299         rcu_read_unlock();
3300         return iter;
3301 }
3302
3303 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3304
3305 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3306         struct tgid_iter iter)
3307 {
3308         char name[PROC_NUMBUF];
3309         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3310         return proc_fill_cache(filp, dirent, filldir, name, len,
3311                                 proc_pid_instantiate, iter.task, NULL);
3312 }
3313
3314 static int fake_filldir(void *buf, const char *name, int namelen,
3315                         loff_t offset, u64 ino, unsigned d_type)
3316 {
3317         return 0;
3318 }
3319
3320 /* for the /proc/ directory itself, after non-process stuff has been done */
3321 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3322 {
3323         unsigned int nr;
3324         struct task_struct *reaper;
3325         struct tgid_iter iter;
3326         struct pid_namespace *ns;
3327         filldir_t __filldir;
3328
3329         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3330                 goto out_no_task;
3331         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3332
3333         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3334         if (!reaper)
3335                 goto out_no_task;
3336
3337         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3338                 const struct pid_entry *p = &proc_base_stuff[nr];
3339                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3340                         goto out;
3341         }
3342
3343         ns = filp->f_dentry->d_sb->s_fs_info;
3344         iter.task = NULL;
3345         iter.tgid = filp->f_pos - TGID_OFFSET;
3346         for (iter = next_tgid(ns, iter);
3347              iter.task;
3348              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3349                 if (has_pid_permissions(ns, iter.task, 2))
3350                         __filldir = filldir;
3351                 else
3352                         __filldir = fake_filldir;
3353
3354                 filp->f_pos = iter.tgid + TGID_OFFSET;
3355                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3356                         put_task_struct(iter.task);
3357                         goto out;
3358                 }
3359         }
3360         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3361 out:
3362         put_task_struct(reaper);
3363 out_no_task:
3364         return 0;
3365 }
3366
3367 /*
3368  * Tasks
3369  */
3370 static const struct pid_entry tid_base_stuff[] = {
3371         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3372         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3373         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3374         REG("environ",   S_IRUSR, proc_environ_operations),
3375         INF("auxv",      S_IRUSR, proc_pid_auxv),
3376         ONE("status",    S_IRUGO, proc_pid_status),
3377         ONE("personality", S_IRUGO, proc_pid_personality),
3378         INF("limits",    S_IRUGO, proc_pid_limits),
3379 #ifdef CONFIG_SCHED_DEBUG
3380         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3381 #endif
3382         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3383 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3384         INF("syscall",   S_IRUGO, proc_pid_syscall),
3385 #endif
3386         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3387         ONE("stat",      S_IRUGO, proc_tid_stat),
3388         ONE("statm",     S_IRUGO, proc_pid_statm),
3389         REG("maps",      S_IRUGO, proc_maps_operations),
3390 #ifdef CONFIG_NUMA
3391         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3392 #endif
3393         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3394         LNK("cwd",       proc_cwd_link),
3395         LNK("root",      proc_root_link),
3396         LNK("exe",       proc_exe_link),
3397         REG("mounts",    S_IRUGO, proc_mounts_operations),
3398         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3399 #ifdef CONFIG_PROC_PAGE_MONITOR
3400         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3401         REG("smaps",     S_IRUGO, proc_smaps_operations),
3402         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3403 #endif
3404 #ifdef CONFIG_SECURITY
3405         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3406 #endif
3407 #ifdef CONFIG_KALLSYMS
3408         INF("wchan",     S_IRUGO, proc_pid_wchan),
3409 #endif
3410 #ifdef CONFIG_STACKTRACE
3411         ONE("stack",      S_IRUGO, proc_pid_stack),
3412 #endif
3413 #ifdef CONFIG_SCHEDSTATS
3414         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3415 #endif
3416 #ifdef CONFIG_LATENCYTOP
3417         REG("latency",  S_IRUGO, proc_lstats_operations),
3418 #endif
3419 #ifdef CONFIG_PROC_PID_CPUSET
3420         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3421 #endif
3422 #ifdef CONFIG_CGROUPS
3423         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3424 #endif
3425         INF("oom_score", S_IRUGO, proc_oom_score),
3426         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3427         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3428 #ifdef CONFIG_AUDITSYSCALL
3429         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3430         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3431 #endif
3432 #ifdef CONFIG_FAULT_INJECTION
3433         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3434 #endif
3435 #ifdef CONFIG_TASK_IO_ACCOUNTING
3436         INF("io",       S_IRUSR, proc_tid_io_accounting),
3437 #endif
3438 #ifdef CONFIG_HARDWALL
3439         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3440 #endif
3441 };
3442
3443 static int proc_tid_base_readdir(struct file * filp,
3444                              void * dirent, filldir_t filldir)
3445 {
3446         return proc_pident_readdir(filp,dirent,filldir,
3447                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3448 }
3449
3450 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3451         return proc_pident_lookup(dir, dentry,
3452                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3453 }
3454
3455 static const struct file_operations proc_tid_base_operations = {
3456         .read           = generic_read_dir,
3457         .readdir        = proc_tid_base_readdir,
3458         .llseek         = default_llseek,
3459 };
3460
3461 static const struct inode_operations proc_tid_base_inode_operations = {
3462         .lookup         = proc_tid_base_lookup,
3463         .getattr        = pid_getattr,
3464         .setattr        = proc_setattr,
3465 };
3466
3467 static struct dentry *proc_task_instantiate(struct inode *dir,
3468         struct dentry *dentry, struct task_struct *task, const void *ptr)
3469 {
3470         struct dentry *error = ERR_PTR(-ENOENT);
3471         struct inode *inode;
3472         inode = proc_pid_make_inode(dir->i_sb, task);
3473
3474         if (!inode)
3475                 goto out;
3476         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3477         inode->i_op = &proc_tid_base_inode_operations;
3478         inode->i_fop = &proc_tid_base_operations;
3479         inode->i_flags|=S_IMMUTABLE;
3480
3481         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3482                                                   ARRAY_SIZE(tid_base_stuff)));
3483
3484         d_set_d_op(dentry, &pid_dentry_operations);
3485
3486         d_add(dentry, inode);
3487         /* Close the race of the process dying before we return the dentry */
3488         if (pid_revalidate(dentry, NULL))
3489                 error = NULL;
3490 out:
3491         return error;
3492 }
3493
3494 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3495 {
3496         struct dentry *result = ERR_PTR(-ENOENT);
3497         struct task_struct *task;
3498         struct task_struct *leader = get_proc_task(dir);
3499         unsigned tid;
3500         struct pid_namespace *ns;
3501
3502         if (!leader)
3503                 goto out_no_task;
3504
3505         tid = name_to_int(dentry);
3506         if (tid == ~0U)
3507                 goto out;
3508
3509         ns = dentry->d_sb->s_fs_info;
3510         rcu_read_lock();
3511         task = find_task_by_pid_ns(tid, ns);
3512         if (task)
3513                 get_task_struct(task);
3514         rcu_read_unlock();
3515         if (!task)
3516                 goto out;
3517         if (!same_thread_group(leader, task))
3518                 goto out_drop_task;
3519
3520         result = proc_task_instantiate(dir, dentry, task, NULL);
3521 out_drop_task:
3522         put_task_struct(task);
3523 out:
3524         put_task_struct(leader);
3525 out_no_task:
3526         return result;
3527 }
3528
3529 /*
3530  * Find the first tid of a thread group to return to user space.
3531  *
3532  * Usually this is just the thread group leader, but if the users
3533  * buffer was too small or there was a seek into the middle of the
3534  * directory we have more work todo.
3535  *
3536  * In the case of a short read we start with find_task_by_pid.
3537  *
3538  * In the case of a seek we start with the leader and walk nr
3539  * threads past it.
3540  */
3541 static struct task_struct *first_tid(struct task_struct *leader,
3542                 int tid, int nr, struct pid_namespace *ns)
3543 {
3544         struct task_struct *pos;
3545
3546         rcu_read_lock();
3547         /* Attempt to start with the pid of a thread */
3548         if (tid && (nr > 0)) {
3549                 pos = find_task_by_pid_ns(tid, ns);
3550                 if (pos && (pos->group_leader == leader))
3551                         goto found;
3552         }
3553
3554         /* If nr exceeds the number of threads there is nothing todo */
3555         pos = NULL;
3556         if (nr && nr >= get_nr_threads(leader))
3557                 goto out;
3558
3559         /* If we haven't found our starting place yet start
3560          * with the leader and walk nr threads forward.
3561          */
3562         for (pos = leader; nr > 0; --nr) {
3563                 pos = next_thread(pos);
3564                 if (pos == leader) {
3565                         pos = NULL;
3566                         goto out;
3567                 }
3568         }
3569 found:
3570         get_task_struct(pos);
3571 out:
3572         rcu_read_unlock();
3573         return pos;
3574 }
3575
3576 /*
3577  * Find the next thread in the thread list.
3578  * Return NULL if there is an error or no next thread.
3579  *
3580  * The reference to the input task_struct is released.
3581  */
3582 static struct task_struct *next_tid(struct task_struct *start)
3583 {
3584         struct task_struct *pos = NULL;
3585         rcu_read_lock();
3586         if (pid_alive(start)) {
3587                 pos = next_thread(start);
3588                 if (thread_group_leader(pos))
3589                         pos = NULL;
3590                 else
3591                         get_task_struct(pos);
3592         }
3593         rcu_read_unlock();
3594         put_task_struct(start);
3595         return pos;
3596 }
3597
3598 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3599         struct task_struct *task, int tid)
3600 {
3601         char name[PROC_NUMBUF];
3602         int len = snprintf(name, sizeof(name), "%d", tid);
3603         return proc_fill_cache(filp, dirent, filldir, name, len,
3604                                 proc_task_instantiate, task, NULL);
3605 }
3606
3607 /* for the /proc/TGID/task/ directories */
3608 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3609 {
3610         struct dentry *dentry = filp->f_path.dentry;
3611         struct inode *inode = dentry->d_inode;
3612         struct task_struct *leader = NULL;
3613         struct task_struct *task;
3614         int retval = -ENOENT;
3615         ino_t ino;
3616         int tid;
3617         struct pid_namespace *ns;
3618
3619         task = get_proc_task(inode);
3620         if (!task)
3621                 goto out_no_task;
3622         rcu_read_lock();
3623         if (pid_alive(task)) {
3624                 leader = task->group_leader;
3625                 get_task_struct(leader);
3626         }
3627         rcu_read_unlock();
3628         put_task_struct(task);
3629         if (!leader)
3630                 goto out_no_task;
3631         retval = 0;
3632
3633         switch ((unsigned long)filp->f_pos) {
3634         case 0:
3635                 ino = inode->i_ino;
3636                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3637                         goto out;
3638                 filp->f_pos++;
3639                 /* fall through */
3640         case 1:
3641                 ino = parent_ino(dentry);
3642                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3643                         goto out;
3644                 filp->f_pos++;
3645                 /* fall through */
3646         }
3647
3648         /* f_version caches the tgid value that the last readdir call couldn't
3649          * return. lseek aka telldir automagically resets f_version to 0.
3650          */
3651         ns = filp->f_dentry->d_sb->s_fs_info;
3652         tid = (int)filp->f_version;
3653         filp->f_version = 0;
3654         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3655              task;
3656              task = next_tid(task), filp->f_pos++) {
3657                 tid = task_pid_nr_ns(task, ns);
3658                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3659                         /* returning this tgid failed, save it as the first
3660                          * pid for the next readir call */
3661                         filp->f_version = (u64)tid;
3662                         put_task_struct(task);
3663                         break;
3664                 }
3665         }
3666 out:
3667         put_task_struct(leader);
3668 out_no_task:
3669         return retval;
3670 }
3671
3672 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3673 {
3674         struct inode *inode = dentry->d_inode;
3675         struct task_struct *p = get_proc_task(inode);
3676         generic_fillattr(inode, stat);
3677
3678         if (p) {
3679                 stat->nlink += get_nr_threads(p);
3680                 put_task_struct(p);
3681         }
3682
3683         return 0;
3684 }
3685
3686 static const struct inode_operations proc_task_inode_operations = {
3687         .lookup         = proc_task_lookup,
3688         .getattr        = proc_task_getattr,
3689         .setattr        = proc_setattr,
3690         .permission     = proc_pid_permission,
3691 };
3692
3693 static const struct file_operations proc_task_operations = {
3694         .read           = generic_read_dir,
3695         .readdir        = proc_task_readdir,
3696         .llseek         = default_llseek,
3697 };