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         rcu_read_lock();
1136         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1137                 rcu_read_unlock();
1138                 return -EPERM;
1139         }
1140         rcu_read_unlock();
1141
1142         if (count >= PAGE_SIZE)
1143                 count = PAGE_SIZE - 1;
1144
1145         if (*ppos != 0) {
1146                 /* No partial writes. */
1147                 return -EINVAL;
1148         }
1149         page = (char*)__get_free_page(GFP_TEMPORARY);
1150         if (!page)
1151                 return -ENOMEM;
1152         length = -EFAULT;
1153         if (copy_from_user(page, buf, count))
1154                 goto out_free_page;
1155
1156         page[count] = '\0';
1157         loginuid = simple_strtoul(page, &tmp, 10);
1158         if (tmp == page) {
1159                 length = -EINVAL;
1160                 goto out_free_page;
1161
1162         }
1163         length = audit_set_loginuid(loginuid);
1164         if (likely(length == 0))
1165                 length = count;
1166
1167 out_free_page:
1168         free_page((unsigned long) page);
1169         return length;
1170 }
1171
1172 static const struct file_operations proc_loginuid_operations = {
1173         .read           = proc_loginuid_read,
1174         .write          = proc_loginuid_write,
1175         .llseek         = generic_file_llseek,
1176 };
1177
1178 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1179                                   size_t count, loff_t *ppos)
1180 {
1181         struct inode * inode = file->f_path.dentry->d_inode;
1182         struct task_struct *task = get_proc_task(inode);
1183         ssize_t length;
1184         char tmpbuf[TMPBUFLEN];
1185
1186         if (!task)
1187                 return -ESRCH;
1188         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1189                                 audit_get_sessionid(task));
1190         put_task_struct(task);
1191         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1192 }
1193
1194 static const struct file_operations proc_sessionid_operations = {
1195         .read           = proc_sessionid_read,
1196         .llseek         = generic_file_llseek,
1197 };
1198 #endif
1199
1200 #ifdef CONFIG_FAULT_INJECTION
1201 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1202                                       size_t count, loff_t *ppos)
1203 {
1204         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1205         char buffer[PROC_NUMBUF];
1206         size_t len;
1207         int make_it_fail;
1208
1209         if (!task)
1210                 return -ESRCH;
1211         make_it_fail = task->make_it_fail;
1212         put_task_struct(task);
1213
1214         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1215
1216         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1217 }
1218
1219 static ssize_t proc_fault_inject_write(struct file * file,
1220                         const char __user * buf, size_t count, loff_t *ppos)
1221 {
1222         struct task_struct *task;
1223         char buffer[PROC_NUMBUF], *end;
1224         int make_it_fail;
1225
1226         if (!capable(CAP_SYS_RESOURCE))
1227                 return -EPERM;
1228         memset(buffer, 0, sizeof(buffer));
1229         if (count > sizeof(buffer) - 1)
1230                 count = sizeof(buffer) - 1;
1231         if (copy_from_user(buffer, buf, count))
1232                 return -EFAULT;
1233         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1234         if (*end)
1235                 return -EINVAL;
1236         task = get_proc_task(file->f_dentry->d_inode);
1237         if (!task)
1238                 return -ESRCH;
1239         task->make_it_fail = make_it_fail;
1240         put_task_struct(task);
1241
1242         return count;
1243 }
1244
1245 static const struct file_operations proc_fault_inject_operations = {
1246         .read           = proc_fault_inject_read,
1247         .write          = proc_fault_inject_write,
1248         .llseek         = generic_file_llseek,
1249 };
1250 #endif
1251
1252
1253 #ifdef CONFIG_SCHED_DEBUG
1254 /*
1255  * Print out various scheduling related per-task fields:
1256  */
1257 static int sched_show(struct seq_file *m, void *v)
1258 {
1259         struct inode *inode = m->private;
1260         struct task_struct *p;
1261
1262         p = get_proc_task(inode);
1263         if (!p)
1264                 return -ESRCH;
1265         proc_sched_show_task(p, m);
1266
1267         put_task_struct(p);
1268
1269         return 0;
1270 }
1271
1272 static ssize_t
1273 sched_write(struct file *file, const char __user *buf,
1274             size_t count, loff_t *offset)
1275 {
1276         struct inode *inode = file->f_path.dentry->d_inode;
1277         struct task_struct *p;
1278
1279         p = get_proc_task(inode);
1280         if (!p)
1281                 return -ESRCH;
1282         proc_sched_set_task(p);
1283
1284         put_task_struct(p);
1285
1286         return count;
1287 }
1288
1289 static int sched_open(struct inode *inode, struct file *filp)
1290 {
1291         return single_open(filp, sched_show, inode);
1292 }
1293
1294 static const struct file_operations proc_pid_sched_operations = {
1295         .open           = sched_open,
1296         .read           = seq_read,
1297         .write          = sched_write,
1298         .llseek         = seq_lseek,
1299         .release        = single_release,
1300 };
1301
1302 #endif
1303
1304 #ifdef CONFIG_SCHED_AUTOGROUP
1305 /*
1306  * Print out autogroup related information:
1307  */
1308 static int sched_autogroup_show(struct seq_file *m, void *v)
1309 {
1310         struct inode *inode = m->private;
1311         struct task_struct *p;
1312
1313         p = get_proc_task(inode);
1314         if (!p)
1315                 return -ESRCH;
1316         proc_sched_autogroup_show_task(p, m);
1317
1318         put_task_struct(p);
1319
1320         return 0;
1321 }
1322
1323 static ssize_t
1324 sched_autogroup_write(struct file *file, const char __user *buf,
1325             size_t count, loff_t *offset)
1326 {
1327         struct inode *inode = file->f_path.dentry->d_inode;
1328         struct task_struct *p;
1329         char buffer[PROC_NUMBUF];
1330         int nice;
1331         int err;
1332
1333         memset(buffer, 0, sizeof(buffer));
1334         if (count > sizeof(buffer) - 1)
1335                 count = sizeof(buffer) - 1;
1336         if (copy_from_user(buffer, buf, count))
1337                 return -EFAULT;
1338
1339         err = kstrtoint(strstrip(buffer), 0, &nice);
1340         if (err < 0)
1341                 return err;
1342
1343         p = get_proc_task(inode);
1344         if (!p)
1345                 return -ESRCH;
1346
1347         err = nice;
1348         err = proc_sched_autogroup_set_nice(p, &err);
1349         if (err)
1350                 count = err;
1351
1352         put_task_struct(p);
1353
1354         return count;
1355 }
1356
1357 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1358 {
1359         int ret;
1360
1361         ret = single_open(filp, sched_autogroup_show, NULL);
1362         if (!ret) {
1363                 struct seq_file *m = filp->private_data;
1364
1365                 m->private = inode;
1366         }
1367         return ret;
1368 }
1369
1370 static const struct file_operations proc_pid_sched_autogroup_operations = {
1371         .open           = sched_autogroup_open,
1372         .read           = seq_read,
1373         .write          = sched_autogroup_write,
1374         .llseek         = seq_lseek,
1375         .release        = single_release,
1376 };
1377
1378 #endif /* CONFIG_SCHED_AUTOGROUP */
1379
1380 static ssize_t comm_write(struct file *file, const char __user *buf,
1381                                 size_t count, loff_t *offset)
1382 {
1383         struct inode *inode = file->f_path.dentry->d_inode;
1384         struct task_struct *p;
1385         char buffer[TASK_COMM_LEN];
1386
1387         memset(buffer, 0, sizeof(buffer));
1388         if (count > sizeof(buffer) - 1)
1389                 count = sizeof(buffer) - 1;
1390         if (copy_from_user(buffer, buf, count))
1391                 return -EFAULT;
1392
1393         p = get_proc_task(inode);
1394         if (!p)
1395                 return -ESRCH;
1396
1397         if (same_thread_group(current, p))
1398                 set_task_comm(p, buffer);
1399         else
1400                 count = -EINVAL;
1401
1402         put_task_struct(p);
1403
1404         return count;
1405 }
1406
1407 static int comm_show(struct seq_file *m, void *v)
1408 {
1409         struct inode *inode = m->private;
1410         struct task_struct *p;
1411
1412         p = get_proc_task(inode);
1413         if (!p)
1414                 return -ESRCH;
1415
1416         task_lock(p);
1417         seq_printf(m, "%s\n", p->comm);
1418         task_unlock(p);
1419
1420         put_task_struct(p);
1421
1422         return 0;
1423 }
1424
1425 static int comm_open(struct inode *inode, struct file *filp)
1426 {
1427         return single_open(filp, comm_show, inode);
1428 }
1429
1430 static const struct file_operations proc_pid_set_comm_operations = {
1431         .open           = comm_open,
1432         .read           = seq_read,
1433         .write          = comm_write,
1434         .llseek         = seq_lseek,
1435         .release        = single_release,
1436 };
1437
1438 static int proc_exe_link(struct dentry *dentry, struct path *exe_path)
1439 {
1440         struct task_struct *task;
1441         struct mm_struct *mm;
1442         struct file *exe_file;
1443
1444         task = get_proc_task(dentry->d_inode);
1445         if (!task)
1446                 return -ENOENT;
1447         mm = get_task_mm(task);
1448         put_task_struct(task);
1449         if (!mm)
1450                 return -ENOENT;
1451         exe_file = get_mm_exe_file(mm);
1452         mmput(mm);
1453         if (exe_file) {
1454                 *exe_path = exe_file->f_path;
1455                 path_get(&exe_file->f_path);
1456                 fput(exe_file);
1457                 return 0;
1458         } else
1459                 return -ENOENT;
1460 }
1461
1462 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1463 {
1464         struct inode *inode = dentry->d_inode;
1465         int error = -EACCES;
1466
1467         /* We don't need a base pointer in the /proc filesystem */
1468         path_put(&nd->path);
1469
1470         /* Are we allowed to snoop on the tasks file descriptors? */
1471         if (!proc_fd_access_allowed(inode))
1472                 goto out;
1473
1474         error = PROC_I(inode)->op.proc_get_link(dentry, &nd->path);
1475 out:
1476         return ERR_PTR(error);
1477 }
1478
1479 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1480 {
1481         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1482         char *pathname;
1483         int len;
1484
1485         if (!tmp)
1486                 return -ENOMEM;
1487
1488         pathname = d_path(path, tmp, PAGE_SIZE);
1489         len = PTR_ERR(pathname);
1490         if (IS_ERR(pathname))
1491                 goto out;
1492         len = tmp + PAGE_SIZE - 1 - pathname;
1493
1494         if (len > buflen)
1495                 len = buflen;
1496         if (copy_to_user(buffer, pathname, len))
1497                 len = -EFAULT;
1498  out:
1499         free_page((unsigned long)tmp);
1500         return len;
1501 }
1502
1503 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1504 {
1505         int error = -EACCES;
1506         struct inode *inode = dentry->d_inode;
1507         struct path path;
1508
1509         /* Are we allowed to snoop on the tasks file descriptors? */
1510         if (!proc_fd_access_allowed(inode))
1511                 goto out;
1512
1513         error = PROC_I(inode)->op.proc_get_link(dentry, &path);
1514         if (error)
1515                 goto out;
1516
1517         error = do_proc_readlink(&path, buffer, buflen);
1518         path_put(&path);
1519 out:
1520         return error;
1521 }
1522
1523 static const struct inode_operations proc_pid_link_inode_operations = {
1524         .readlink       = proc_pid_readlink,
1525         .follow_link    = proc_pid_follow_link,
1526         .setattr        = proc_setattr,
1527 };
1528
1529
1530 /* building an inode */
1531
1532 static int task_dumpable(struct task_struct *task)
1533 {
1534         int dumpable = 0;
1535         struct mm_struct *mm;
1536
1537         task_lock(task);
1538         mm = task->mm;
1539         if (mm)
1540                 dumpable = get_dumpable(mm);
1541         task_unlock(task);
1542         if(dumpable == 1)
1543                 return 1;
1544         return 0;
1545 }
1546
1547 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1548 {
1549         struct inode * inode;
1550         struct proc_inode *ei;
1551         const struct cred *cred;
1552
1553         /* We need a new inode */
1554
1555         inode = new_inode(sb);
1556         if (!inode)
1557                 goto out;
1558
1559         /* Common stuff */
1560         ei = PROC_I(inode);
1561         inode->i_ino = get_next_ino();
1562         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1563         inode->i_op = &proc_def_inode_operations;
1564
1565         /*
1566          * grab the reference to task.
1567          */
1568         ei->pid = get_task_pid(task, PIDTYPE_PID);
1569         if (!ei->pid)
1570                 goto out_unlock;
1571
1572         if (task_dumpable(task)) {
1573                 rcu_read_lock();
1574                 cred = __task_cred(task);
1575                 inode->i_uid = cred->euid;
1576                 inode->i_gid = cred->egid;
1577                 rcu_read_unlock();
1578         }
1579         security_task_to_inode(task, inode);
1580
1581 out:
1582         return inode;
1583
1584 out_unlock:
1585         iput(inode);
1586         return NULL;
1587 }
1588
1589 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1590 {
1591         struct inode *inode = dentry->d_inode;
1592         struct task_struct *task;
1593         const struct cred *cred;
1594         struct pid_namespace *pid = dentry->d_sb->s_fs_info;
1595
1596         generic_fillattr(inode, stat);
1597
1598         rcu_read_lock();
1599         stat->uid = 0;
1600         stat->gid = 0;
1601         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1602         if (task) {
1603                 if (!has_pid_permissions(pid, task, 2)) {
1604                         rcu_read_unlock();
1605                         /*
1606                          * This doesn't prevent learning whether PID exists,
1607                          * it only makes getattr() consistent with readdir().
1608                          */
1609                         return -ENOENT;
1610                 }
1611                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1612                     task_dumpable(task)) {
1613                         cred = __task_cred(task);
1614                         stat->uid = cred->euid;
1615                         stat->gid = cred->egid;
1616                 }
1617         }
1618         rcu_read_unlock();
1619         return 0;
1620 }
1621
1622 /* dentry stuff */
1623
1624 /*
1625  *      Exceptional case: normally we are not allowed to unhash a busy
1626  * directory. In this case, however, we can do it - no aliasing problems
1627  * due to the way we treat inodes.
1628  *
1629  * Rewrite the inode's ownerships here because the owning task may have
1630  * performed a setuid(), etc.
1631  *
1632  * Before the /proc/pid/status file was created the only way to read
1633  * the effective uid of a /process was to stat /proc/pid.  Reading
1634  * /proc/pid/status is slow enough that procps and other packages
1635  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1636  * made this apply to all per process world readable and executable
1637  * directories.
1638  */
1639 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1640 {
1641         struct inode *inode;
1642         struct task_struct *task;
1643         const struct cred *cred;
1644
1645         if (nd && nd->flags & LOOKUP_RCU)
1646                 return -ECHILD;
1647
1648         inode = dentry->d_inode;
1649         task = get_proc_task(inode);
1650
1651         if (task) {
1652                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1653                     task_dumpable(task)) {
1654                         rcu_read_lock();
1655                         cred = __task_cred(task);
1656                         inode->i_uid = cred->euid;
1657                         inode->i_gid = cred->egid;
1658                         rcu_read_unlock();
1659                 } else {
1660                         inode->i_uid = 0;
1661                         inode->i_gid = 0;
1662                 }
1663                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1664                 security_task_to_inode(task, inode);
1665                 put_task_struct(task);
1666                 return 1;
1667         }
1668         d_drop(dentry);
1669         return 0;
1670 }
1671
1672 static int pid_delete_dentry(const struct dentry * dentry)
1673 {
1674         /* Is the task we represent dead?
1675          * If so, then don't put the dentry on the lru list,
1676          * kill it immediately.
1677          */
1678         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1679 }
1680
1681 const struct dentry_operations pid_dentry_operations =
1682 {
1683         .d_revalidate   = pid_revalidate,
1684         .d_delete       = pid_delete_dentry,
1685 };
1686
1687 /* Lookups */
1688
1689 /*
1690  * Fill a directory entry.
1691  *
1692  * If possible create the dcache entry and derive our inode number and
1693  * file type from dcache entry.
1694  *
1695  * Since all of the proc inode numbers are dynamically generated, the inode
1696  * numbers do not exist until the inode is cache.  This means creating the
1697  * the dcache entry in readdir is necessary to keep the inode numbers
1698  * reported by readdir in sync with the inode numbers reported
1699  * by stat.
1700  */
1701 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1702         const char *name, int len,
1703         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1704 {
1705         struct dentry *child, *dir = filp->f_path.dentry;
1706         struct inode *inode;
1707         struct qstr qname;
1708         ino_t ino = 0;
1709         unsigned type = DT_UNKNOWN;
1710
1711         qname.name = name;
1712         qname.len  = len;
1713         qname.hash = full_name_hash(name, len);
1714
1715         child = d_lookup(dir, &qname);
1716         if (!child) {
1717                 struct dentry *new;
1718                 new = d_alloc(dir, &qname);
1719                 if (new) {
1720                         child = instantiate(dir->d_inode, new, task, ptr);
1721                         if (child)
1722                                 dput(new);
1723                         else
1724                                 child = new;
1725                 }
1726         }
1727         if (!child || IS_ERR(child) || !child->d_inode)
1728                 goto end_instantiate;
1729         inode = child->d_inode;
1730         if (inode) {
1731                 ino = inode->i_ino;
1732                 type = inode->i_mode >> 12;
1733         }
1734         dput(child);
1735 end_instantiate:
1736         if (!ino)
1737                 ino = find_inode_number(dir, &qname);
1738         if (!ino)
1739                 ino = 1;
1740         return filldir(dirent, name, len, filp->f_pos, ino, type);
1741 }
1742
1743 static unsigned name_to_int(struct dentry *dentry)
1744 {
1745         const char *name = dentry->d_name.name;
1746         int len = dentry->d_name.len;
1747         unsigned n = 0;
1748
1749         if (len > 1 && *name == '0')
1750                 goto out;
1751         while (len-- > 0) {
1752                 unsigned c = *name++ - '0';
1753                 if (c > 9)
1754                         goto out;
1755                 if (n >= (~0U-9)/10)
1756                         goto out;
1757                 n *= 10;
1758                 n += c;
1759         }
1760         return n;
1761 out:
1762         return ~0U;
1763 }
1764
1765 #define PROC_FDINFO_MAX 64
1766
1767 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1768 {
1769         struct task_struct *task = get_proc_task(inode);
1770         struct files_struct *files = NULL;
1771         struct file *file;
1772         int fd = proc_fd(inode);
1773
1774         if (task) {
1775                 files = get_files_struct(task);
1776                 put_task_struct(task);
1777         }
1778         if (files) {
1779                 /*
1780                  * We are not taking a ref to the file structure, so we must
1781                  * hold ->file_lock.
1782                  */
1783                 spin_lock(&files->file_lock);
1784                 file = fcheck_files(files, fd);
1785                 if (file) {
1786                         unsigned int f_flags;
1787                         struct fdtable *fdt;
1788
1789                         fdt = files_fdtable(files);
1790                         f_flags = file->f_flags & ~O_CLOEXEC;
1791                         if (FD_ISSET(fd, fdt->close_on_exec))
1792                                 f_flags |= O_CLOEXEC;
1793
1794                         if (path) {
1795                                 *path = file->f_path;
1796                                 path_get(&file->f_path);
1797                         }
1798                         if (info)
1799                                 snprintf(info, PROC_FDINFO_MAX,
1800                                          "pos:\t%lli\n"
1801                                          "flags:\t0%o\n",
1802                                          (long long) file->f_pos,
1803                                          f_flags);
1804                         spin_unlock(&files->file_lock);
1805                         put_files_struct(files);
1806                         return 0;
1807                 }
1808                 spin_unlock(&files->file_lock);
1809                 put_files_struct(files);
1810         }
1811         return -ENOENT;
1812 }
1813
1814 static int proc_fd_link(struct dentry *dentry, struct path *path)
1815 {
1816         return proc_fd_info(dentry->d_inode, path, NULL);
1817 }
1818
1819 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1820 {
1821         struct inode *inode;
1822         struct task_struct *task;
1823         int fd;
1824         struct files_struct *files;
1825         const struct cred *cred;
1826
1827         if (nd && nd->flags & LOOKUP_RCU)
1828                 return -ECHILD;
1829
1830         inode = dentry->d_inode;
1831         task = get_proc_task(inode);
1832         fd = proc_fd(inode);
1833
1834         if (task) {
1835                 files = get_files_struct(task);
1836                 if (files) {
1837                         rcu_read_lock();
1838                         if (fcheck_files(files, fd)) {
1839                                 rcu_read_unlock();
1840                                 put_files_struct(files);
1841                                 if (task_dumpable(task)) {
1842                                         rcu_read_lock();
1843                                         cred = __task_cred(task);
1844                                         inode->i_uid = cred->euid;
1845                                         inode->i_gid = cred->egid;
1846                                         rcu_read_unlock();
1847                                 } else {
1848                                         inode->i_uid = 0;
1849                                         inode->i_gid = 0;
1850                                 }
1851                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1852                                 security_task_to_inode(task, inode);
1853                                 put_task_struct(task);
1854                                 return 1;
1855                         }
1856                         rcu_read_unlock();
1857                         put_files_struct(files);
1858                 }
1859                 put_task_struct(task);
1860         }
1861         d_drop(dentry);
1862         return 0;
1863 }
1864
1865 static const struct dentry_operations tid_fd_dentry_operations =
1866 {
1867         .d_revalidate   = tid_fd_revalidate,
1868         .d_delete       = pid_delete_dentry,
1869 };
1870
1871 static struct dentry *proc_fd_instantiate(struct inode *dir,
1872         struct dentry *dentry, struct task_struct *task, const void *ptr)
1873 {
1874         unsigned fd = *(const unsigned *)ptr;
1875         struct file *file;
1876         struct files_struct *files;
1877         struct inode *inode;
1878         struct proc_inode *ei;
1879         struct dentry *error = ERR_PTR(-ENOENT);
1880
1881         inode = proc_pid_make_inode(dir->i_sb, task);
1882         if (!inode)
1883                 goto out;
1884         ei = PROC_I(inode);
1885         ei->fd = fd;
1886         files = get_files_struct(task);
1887         if (!files)
1888                 goto out_iput;
1889         inode->i_mode = S_IFLNK;
1890
1891         /*
1892          * We are not taking a ref to the file structure, so we must
1893          * hold ->file_lock.
1894          */
1895         spin_lock(&files->file_lock);
1896         file = fcheck_files(files, fd);
1897         if (!file)
1898                 goto out_unlock;
1899         if (file->f_mode & FMODE_READ)
1900                 inode->i_mode |= S_IRUSR | S_IXUSR;
1901         if (file->f_mode & FMODE_WRITE)
1902                 inode->i_mode |= S_IWUSR | S_IXUSR;
1903         spin_unlock(&files->file_lock);
1904         put_files_struct(files);
1905
1906         inode->i_op = &proc_pid_link_inode_operations;
1907         inode->i_size = 64;
1908         ei->op.proc_get_link = proc_fd_link;
1909         d_set_d_op(dentry, &tid_fd_dentry_operations);
1910         d_add(dentry, inode);
1911         /* Close the race of the process dying before we return the dentry */
1912         if (tid_fd_revalidate(dentry, NULL))
1913                 error = NULL;
1914
1915  out:
1916         return error;
1917 out_unlock:
1918         spin_unlock(&files->file_lock);
1919         put_files_struct(files);
1920 out_iput:
1921         iput(inode);
1922         goto out;
1923 }
1924
1925 static struct dentry *proc_lookupfd_common(struct inode *dir,
1926                                            struct dentry *dentry,
1927                                            instantiate_t instantiate)
1928 {
1929         struct task_struct *task = get_proc_task(dir);
1930         unsigned fd = name_to_int(dentry);
1931         struct dentry *result = ERR_PTR(-ENOENT);
1932
1933         if (!task)
1934                 goto out_no_task;
1935         if (fd == ~0U)
1936                 goto out;
1937
1938         result = instantiate(dir, dentry, task, &fd);
1939 out:
1940         put_task_struct(task);
1941 out_no_task:
1942         return result;
1943 }
1944
1945 static int proc_readfd_common(struct file * filp, void * dirent,
1946                               filldir_t filldir, instantiate_t instantiate)
1947 {
1948         struct dentry *dentry = filp->f_path.dentry;
1949         struct inode *inode = dentry->d_inode;
1950         struct task_struct *p = get_proc_task(inode);
1951         unsigned int fd, ino;
1952         int retval;
1953         struct files_struct * files;
1954
1955         retval = -ENOENT;
1956         if (!p)
1957                 goto out_no_task;
1958         retval = 0;
1959
1960         fd = filp->f_pos;
1961         switch (fd) {
1962                 case 0:
1963                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1964                                 goto out;
1965                         filp->f_pos++;
1966                 case 1:
1967                         ino = parent_ino(dentry);
1968                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1969                                 goto out;
1970                         filp->f_pos++;
1971                 default:
1972                         files = get_files_struct(p);
1973                         if (!files)
1974                                 goto out;
1975                         rcu_read_lock();
1976                         for (fd = filp->f_pos-2;
1977                              fd < files_fdtable(files)->max_fds;
1978                              fd++, filp->f_pos++) {
1979                                 char name[PROC_NUMBUF];
1980                                 int len;
1981
1982                                 if (!fcheck_files(files, fd))
1983                                         continue;
1984                                 rcu_read_unlock();
1985
1986                                 len = snprintf(name, sizeof(name), "%d", fd);
1987                                 if (proc_fill_cache(filp, dirent, filldir,
1988                                                     name, len, instantiate,
1989                                                     p, &fd) < 0) {
1990                                         rcu_read_lock();
1991                                         break;
1992                                 }
1993                                 rcu_read_lock();
1994                         }
1995                         rcu_read_unlock();
1996                         put_files_struct(files);
1997         }
1998 out:
1999         put_task_struct(p);
2000 out_no_task:
2001         return retval;
2002 }
2003
2004 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2005                                     struct nameidata *nd)
2006 {
2007         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2008 }
2009
2010 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2011 {
2012         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2013 }
2014
2015 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2016                                       size_t len, loff_t *ppos)
2017 {
2018         char tmp[PROC_FDINFO_MAX];
2019         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2020         if (!err)
2021                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2022         return err;
2023 }
2024
2025 static const struct file_operations proc_fdinfo_file_operations = {
2026         .open           = nonseekable_open,
2027         .read           = proc_fdinfo_read,
2028         .llseek         = no_llseek,
2029 };
2030
2031 static const struct file_operations proc_fd_operations = {
2032         .read           = generic_read_dir,
2033         .readdir        = proc_readfd,
2034         .llseek         = default_llseek,
2035 };
2036
2037 #ifdef CONFIG_CHECKPOINT_RESTORE
2038
2039 /*
2040  * dname_to_vma_addr - maps a dentry name into two unsigned longs
2041  * which represent vma start and end addresses.
2042  */
2043 static int dname_to_vma_addr(struct dentry *dentry,
2044                              unsigned long *start, unsigned long *end)
2045 {
2046         if (sscanf(dentry->d_name.name, "%lx-%lx", start, end) != 2)
2047                 return -EINVAL;
2048
2049         return 0;
2050 }
2051
2052 static int map_files_d_revalidate(struct dentry *dentry, struct nameidata *nd)
2053 {
2054         unsigned long vm_start, vm_end;
2055         bool exact_vma_exists = false;
2056         struct mm_struct *mm = NULL;
2057         struct task_struct *task;
2058         const struct cred *cred;
2059         struct inode *inode;
2060         int status = 0;
2061
2062         if (nd && nd->flags & LOOKUP_RCU)
2063                 return -ECHILD;
2064
2065         if (!capable(CAP_SYS_ADMIN)) {
2066                 status = -EACCES;
2067                 goto out_notask;
2068         }
2069
2070         inode = dentry->d_inode;
2071         task = get_proc_task(inode);
2072         if (!task)
2073                 goto out_notask;
2074
2075         if (!ptrace_may_access(task, PTRACE_MODE_READ))
2076                 goto out;
2077
2078         mm = get_task_mm(task);
2079         if (!mm)
2080                 goto out;
2081
2082         if (!dname_to_vma_addr(dentry, &vm_start, &vm_end)) {
2083                 down_read(&mm->mmap_sem);
2084                 exact_vma_exists = !!find_exact_vma(mm, vm_start, vm_end);
2085                 up_read(&mm->mmap_sem);
2086         }
2087
2088         mmput(mm);
2089
2090         if (exact_vma_exists) {
2091                 if (task_dumpable(task)) {
2092                         rcu_read_lock();
2093                         cred = __task_cred(task);
2094                         inode->i_uid = cred->euid;
2095                         inode->i_gid = cred->egid;
2096                         rcu_read_unlock();
2097                 } else {
2098                         inode->i_uid = 0;
2099                         inode->i_gid = 0;
2100                 }
2101                 security_task_to_inode(task, inode);
2102                 status = 1;
2103         }
2104
2105 out:
2106         put_task_struct(task);
2107
2108 out_notask:
2109         if (status <= 0)
2110                 d_drop(dentry);
2111
2112         return status;
2113 }
2114
2115 static const struct dentry_operations tid_map_files_dentry_operations = {
2116         .d_revalidate   = map_files_d_revalidate,
2117         .d_delete       = pid_delete_dentry,
2118 };
2119
2120 static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
2121 {
2122         unsigned long vm_start, vm_end;
2123         struct vm_area_struct *vma;
2124         struct task_struct *task;
2125         struct mm_struct *mm;
2126         int rc;
2127
2128         rc = -ENOENT;
2129         task = get_proc_task(dentry->d_inode);
2130         if (!task)
2131                 goto out;
2132
2133         mm = get_task_mm(task);
2134         put_task_struct(task);
2135         if (!mm)
2136                 goto out;
2137
2138         rc = dname_to_vma_addr(dentry, &vm_start, &vm_end);
2139         if (rc)
2140                 goto out_mmput;
2141
2142         down_read(&mm->mmap_sem);
2143         vma = find_exact_vma(mm, vm_start, vm_end);
2144         if (vma && vma->vm_file) {
2145                 *path = vma->vm_file->f_path;
2146                 path_get(path);
2147                 rc = 0;
2148         }
2149         up_read(&mm->mmap_sem);
2150
2151 out_mmput:
2152         mmput(mm);
2153 out:
2154         return rc;
2155 }
2156
2157 struct map_files_info {
2158         struct file     *file;
2159         unsigned long   len;
2160         unsigned char   name[4*sizeof(long)+2]; /* max: %lx-%lx\0 */
2161 };
2162
2163 static struct dentry *
2164 proc_map_files_instantiate(struct inode *dir, struct dentry *dentry,
2165                            struct task_struct *task, const void *ptr)
2166 {
2167         const struct file *file = ptr;
2168         struct proc_inode *ei;
2169         struct inode *inode;
2170
2171         if (!file)
2172                 return ERR_PTR(-ENOENT);
2173
2174         inode = proc_pid_make_inode(dir->i_sb, task);
2175         if (!inode)
2176                 return ERR_PTR(-ENOENT);
2177
2178         ei = PROC_I(inode);
2179         ei->op.proc_get_link = proc_map_files_get_link;
2180
2181         inode->i_op = &proc_pid_link_inode_operations;
2182         inode->i_size = 64;
2183         inode->i_mode = S_IFLNK;
2184
2185         if (file->f_mode & FMODE_READ)
2186                 inode->i_mode |= S_IRUSR;
2187         if (file->f_mode & FMODE_WRITE)
2188                 inode->i_mode |= S_IWUSR;
2189
2190         d_set_d_op(dentry, &tid_map_files_dentry_operations);
2191         d_add(dentry, inode);
2192
2193         return NULL;
2194 }
2195
2196 static struct dentry *proc_map_files_lookup(struct inode *dir,
2197                 struct dentry *dentry, struct nameidata *nd)
2198 {
2199         unsigned long vm_start, vm_end;
2200         struct vm_area_struct *vma;
2201         struct task_struct *task;
2202         struct dentry *result;
2203         struct mm_struct *mm;
2204
2205         result = ERR_PTR(-EACCES);
2206         if (!capable(CAP_SYS_ADMIN))
2207                 goto out;
2208
2209         result = ERR_PTR(-ENOENT);
2210         task = get_proc_task(dir);
2211         if (!task)
2212                 goto out;
2213
2214         result = ERR_PTR(-EACCES);
2215         if (lock_trace(task))
2216                 goto out_put_task;
2217
2218         result = ERR_PTR(-ENOENT);
2219         if (dname_to_vma_addr(dentry, &vm_start, &vm_end))
2220                 goto out_unlock;
2221
2222         mm = get_task_mm(task);
2223         if (!mm)
2224                 goto out_unlock;
2225
2226         down_read(&mm->mmap_sem);
2227         vma = find_exact_vma(mm, vm_start, vm_end);
2228         if (!vma)
2229                 goto out_no_vma;
2230
2231         result = proc_map_files_instantiate(dir, dentry, task, vma->vm_file);
2232
2233 out_no_vma:
2234         up_read(&mm->mmap_sem);
2235         mmput(mm);
2236 out_unlock:
2237         unlock_trace(task);
2238 out_put_task:
2239         put_task_struct(task);
2240 out:
2241         return result;
2242 }
2243
2244 static const struct inode_operations proc_map_files_inode_operations = {
2245         .lookup         = proc_map_files_lookup,
2246         .permission     = proc_fd_permission,
2247         .setattr        = proc_setattr,
2248 };
2249
2250 static int
2251 proc_map_files_readdir(struct file *filp, void *dirent, filldir_t filldir)
2252 {
2253         struct dentry *dentry = filp->f_path.dentry;
2254         struct inode *inode = dentry->d_inode;
2255         struct vm_area_struct *vma;
2256         struct task_struct *task;
2257         struct mm_struct *mm;
2258         ino_t ino;
2259         int ret;
2260
2261         ret = -EACCES;
2262         if (!capable(CAP_SYS_ADMIN))
2263                 goto out;
2264
2265         ret = -ENOENT;
2266         task = get_proc_task(inode);
2267         if (!task)
2268                 goto out;
2269
2270         ret = -EACCES;
2271         if (lock_trace(task))
2272                 goto out_put_task;
2273
2274         ret = 0;
2275         switch (filp->f_pos) {
2276         case 0:
2277                 ino = inode->i_ino;
2278                 if (filldir(dirent, ".", 1, 0, ino, DT_DIR) < 0)
2279                         goto out_unlock;
2280                 filp->f_pos++;
2281         case 1:
2282                 ino = parent_ino(dentry);
2283                 if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2284                         goto out_unlock;
2285                 filp->f_pos++;
2286         default:
2287         {
2288                 unsigned long nr_files, pos, i;
2289                 struct flex_array *fa = NULL;
2290                 struct map_files_info info;
2291                 struct map_files_info *p;
2292
2293                 mm = get_task_mm(task);
2294                 if (!mm)
2295                         goto out_unlock;
2296                 down_read(&mm->mmap_sem);
2297
2298                 nr_files = 0;
2299
2300                 /*
2301                  * We need two passes here:
2302                  *
2303                  *  1) Collect vmas of mapped files with mmap_sem taken
2304                  *  2) Release mmap_sem and instantiate entries
2305                  *
2306                  * otherwise we get lockdep complained, since filldir()
2307                  * routine might require mmap_sem taken in might_fault().
2308                  */
2309
2310                 for (vma = mm->mmap, pos = 2; vma; vma = vma->vm_next) {
2311                         if (vma->vm_file && ++pos > filp->f_pos)
2312                                 nr_files++;
2313                 }
2314
2315                 if (nr_files) {
2316                         fa = flex_array_alloc(sizeof(info), nr_files,
2317                                                 GFP_KERNEL);
2318                         if (!fa || flex_array_prealloc(fa, 0, nr_files,
2319                                                         GFP_KERNEL)) {
2320                                 ret = -ENOMEM;
2321                                 if (fa)
2322                                         flex_array_free(fa);
2323                                 up_read(&mm->mmap_sem);
2324                                 mmput(mm);
2325                                 goto out_unlock;
2326                         }
2327                         for (i = 0, vma = mm->mmap, pos = 2; vma;
2328                                         vma = vma->vm_next) {
2329                                 if (!vma->vm_file)
2330                                         continue;
2331                                 if (++pos <= filp->f_pos)
2332                                         continue;
2333
2334                                 get_file(vma->vm_file);
2335                                 info.file = vma->vm_file;
2336                                 info.len = snprintf(info.name,
2337                                                 sizeof(info.name), "%lx-%lx",
2338                                                 vma->vm_start, vma->vm_end);
2339                                 if (flex_array_put(fa, i++, &info, GFP_KERNEL))
2340                                         BUG();
2341                         }
2342                 }
2343                 up_read(&mm->mmap_sem);
2344
2345                 for (i = 0; i < nr_files; i++) {
2346                         p = flex_array_get(fa, i);
2347                         ret = proc_fill_cache(filp, dirent, filldir,
2348                                               p->name, p->len,
2349                                               proc_map_files_instantiate,
2350                                               task, p->file);
2351                         if (ret)
2352                                 break;
2353                         filp->f_pos++;
2354                         fput(p->file);
2355                 }
2356                 for (; i < nr_files; i++) {
2357                         /*
2358                          * In case of error don't forget
2359                          * to put rest of file refs.
2360                          */
2361                         p = flex_array_get(fa, i);
2362                         fput(p->file);
2363                 }
2364                 if (fa)
2365                         flex_array_free(fa);
2366                 mmput(mm);
2367         }
2368         }
2369
2370 out_unlock:
2371         unlock_trace(task);
2372 out_put_task:
2373         put_task_struct(task);
2374 out:
2375         return ret;
2376 }
2377
2378 static const struct file_operations proc_map_files_operations = {
2379         .read           = generic_read_dir,
2380         .readdir        = proc_map_files_readdir,
2381         .llseek         = default_llseek,
2382 };
2383
2384 #endif /* CONFIG_CHECKPOINT_RESTORE */
2385
2386 /*
2387  * /proc/pid/fd needs a special permission handler so that a process can still
2388  * access /proc/self/fd after it has executed a setuid().
2389  */
2390 static int proc_fd_permission(struct inode *inode, int mask)
2391 {
2392         int rv = generic_permission(inode, mask);
2393         if (rv == 0)
2394                 return 0;
2395         if (task_pid(current) == proc_pid(inode))
2396                 rv = 0;
2397         return rv;
2398 }
2399
2400 /*
2401  * proc directories can do almost nothing..
2402  */
2403 static const struct inode_operations proc_fd_inode_operations = {
2404         .lookup         = proc_lookupfd,
2405         .permission     = proc_fd_permission,
2406         .setattr        = proc_setattr,
2407 };
2408
2409 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2410         struct dentry *dentry, struct task_struct *task, const void *ptr)
2411 {
2412         unsigned fd = *(unsigned *)ptr;
2413         struct inode *inode;
2414         struct proc_inode *ei;
2415         struct dentry *error = ERR_PTR(-ENOENT);
2416
2417         inode = proc_pid_make_inode(dir->i_sb, task);
2418         if (!inode)
2419                 goto out;
2420         ei = PROC_I(inode);
2421         ei->fd = fd;
2422         inode->i_mode = S_IFREG | S_IRUSR;
2423         inode->i_fop = &proc_fdinfo_file_operations;
2424         d_set_d_op(dentry, &tid_fd_dentry_operations);
2425         d_add(dentry, inode);
2426         /* Close the race of the process dying before we return the dentry */
2427         if (tid_fd_revalidate(dentry, NULL))
2428                 error = NULL;
2429
2430  out:
2431         return error;
2432 }
2433
2434 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2435                                         struct dentry *dentry,
2436                                         struct nameidata *nd)
2437 {
2438         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2439 }
2440
2441 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2442 {
2443         return proc_readfd_common(filp, dirent, filldir,
2444                                   proc_fdinfo_instantiate);
2445 }
2446
2447 static const struct file_operations proc_fdinfo_operations = {
2448         .read           = generic_read_dir,
2449         .readdir        = proc_readfdinfo,
2450         .llseek         = default_llseek,
2451 };
2452
2453 /*
2454  * proc directories can do almost nothing..
2455  */
2456 static const struct inode_operations proc_fdinfo_inode_operations = {
2457         .lookup         = proc_lookupfdinfo,
2458         .setattr        = proc_setattr,
2459 };
2460
2461
2462 static struct dentry *proc_pident_instantiate(struct inode *dir,
2463         struct dentry *dentry, struct task_struct *task, const void *ptr)
2464 {
2465         const struct pid_entry *p = ptr;
2466         struct inode *inode;
2467         struct proc_inode *ei;
2468         struct dentry *error = ERR_PTR(-ENOENT);
2469
2470         inode = proc_pid_make_inode(dir->i_sb, task);
2471         if (!inode)
2472                 goto out;
2473
2474         ei = PROC_I(inode);
2475         inode->i_mode = p->mode;
2476         if (S_ISDIR(inode->i_mode))
2477                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2478         if (p->iop)
2479                 inode->i_op = p->iop;
2480         if (p->fop)
2481                 inode->i_fop = p->fop;
2482         ei->op = p->op;
2483         d_set_d_op(dentry, &pid_dentry_operations);
2484         d_add(dentry, inode);
2485         /* Close the race of the process dying before we return the dentry */
2486         if (pid_revalidate(dentry, NULL))
2487                 error = NULL;
2488 out:
2489         return error;
2490 }
2491
2492 static struct dentry *proc_pident_lookup(struct inode *dir,
2493                                          struct dentry *dentry,
2494                                          const struct pid_entry *ents,
2495                                          unsigned int nents)
2496 {
2497         struct dentry *error;
2498         struct task_struct *task = get_proc_task(dir);
2499         const struct pid_entry *p, *last;
2500
2501         error = ERR_PTR(-ENOENT);
2502
2503         if (!task)
2504                 goto out_no_task;
2505
2506         /*
2507          * Yes, it does not scale. And it should not. Don't add
2508          * new entries into /proc/<tgid>/ without very good reasons.
2509          */
2510         last = &ents[nents - 1];
2511         for (p = ents; p <= last; p++) {
2512                 if (p->len != dentry->d_name.len)
2513                         continue;
2514                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2515                         break;
2516         }
2517         if (p > last)
2518                 goto out;
2519
2520         error = proc_pident_instantiate(dir, dentry, task, p);
2521 out:
2522         put_task_struct(task);
2523 out_no_task:
2524         return error;
2525 }
2526
2527 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2528         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2529 {
2530         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2531                                 proc_pident_instantiate, task, p);
2532 }
2533
2534 static int proc_pident_readdir(struct file *filp,
2535                 void *dirent, filldir_t filldir,
2536                 const struct pid_entry *ents, unsigned int nents)
2537 {
2538         int i;
2539         struct dentry *dentry = filp->f_path.dentry;
2540         struct inode *inode = dentry->d_inode;
2541         struct task_struct *task = get_proc_task(inode);
2542         const struct pid_entry *p, *last;
2543         ino_t ino;
2544         int ret;
2545
2546         ret = -ENOENT;
2547         if (!task)
2548                 goto out_no_task;
2549
2550         ret = 0;
2551         i = filp->f_pos;
2552         switch (i) {
2553         case 0:
2554                 ino = inode->i_ino;
2555                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2556                         goto out;
2557                 i++;
2558                 filp->f_pos++;
2559                 /* fall through */
2560         case 1:
2561                 ino = parent_ino(dentry);
2562                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2563                         goto out;
2564                 i++;
2565                 filp->f_pos++;
2566                 /* fall through */
2567         default:
2568                 i -= 2;
2569                 if (i >= nents) {
2570                         ret = 1;
2571                         goto out;
2572                 }
2573                 p = ents + i;
2574                 last = &ents[nents - 1];
2575                 while (p <= last) {
2576                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2577                                 goto out;
2578                         filp->f_pos++;
2579                         p++;
2580                 }
2581         }
2582
2583         ret = 1;
2584 out:
2585         put_task_struct(task);
2586 out_no_task:
2587         return ret;
2588 }
2589
2590 #ifdef CONFIG_SECURITY
2591 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2592                                   size_t count, loff_t *ppos)
2593 {
2594         struct inode * inode = file->f_path.dentry->d_inode;
2595         char *p = NULL;
2596         ssize_t length;
2597         struct task_struct *task = get_proc_task(inode);
2598
2599         if (!task)
2600                 return -ESRCH;
2601
2602         length = security_getprocattr(task,
2603                                       (char*)file->f_path.dentry->d_name.name,
2604                                       &p);
2605         put_task_struct(task);
2606         if (length > 0)
2607                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2608         kfree(p);
2609         return length;
2610 }
2611
2612 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2613                                    size_t count, loff_t *ppos)
2614 {
2615         struct inode * inode = file->f_path.dentry->d_inode;
2616         char *page;
2617         ssize_t length;
2618         struct task_struct *task = get_proc_task(inode);
2619
2620         length = -ESRCH;
2621         if (!task)
2622                 goto out_no_task;
2623         if (count > PAGE_SIZE)
2624                 count = PAGE_SIZE;
2625
2626         /* No partial writes. */
2627         length = -EINVAL;
2628         if (*ppos != 0)
2629                 goto out;
2630
2631         length = -ENOMEM;
2632         page = (char*)__get_free_page(GFP_TEMPORARY);
2633         if (!page)
2634                 goto out;
2635
2636         length = -EFAULT;
2637         if (copy_from_user(page, buf, count))
2638                 goto out_free;
2639
2640         /* Guard against adverse ptrace interaction */
2641         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2642         if (length < 0)
2643                 goto out_free;
2644
2645         length = security_setprocattr(task,
2646                                       (char*)file->f_path.dentry->d_name.name,
2647                                       (void*)page, count);
2648         mutex_unlock(&task->signal->cred_guard_mutex);
2649 out_free:
2650         free_page((unsigned long) page);
2651 out:
2652         put_task_struct(task);
2653 out_no_task:
2654         return length;
2655 }
2656
2657 static const struct file_operations proc_pid_attr_operations = {
2658         .read           = proc_pid_attr_read,
2659         .write          = proc_pid_attr_write,
2660         .llseek         = generic_file_llseek,
2661 };
2662
2663 static const struct pid_entry attr_dir_stuff[] = {
2664         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2665         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2666         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2667         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2668         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2669         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2670 };
2671
2672 static int proc_attr_dir_readdir(struct file * filp,
2673                              void * dirent, filldir_t filldir)
2674 {
2675         return proc_pident_readdir(filp,dirent,filldir,
2676                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2677 }
2678
2679 static const struct file_operations proc_attr_dir_operations = {
2680         .read           = generic_read_dir,
2681         .readdir        = proc_attr_dir_readdir,
2682         .llseek         = default_llseek,
2683 };
2684
2685 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2686                                 struct dentry *dentry, struct nameidata *nd)
2687 {
2688         return proc_pident_lookup(dir, dentry,
2689                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2690 }
2691
2692 static const struct inode_operations proc_attr_dir_inode_operations = {
2693         .lookup         = proc_attr_dir_lookup,
2694         .getattr        = pid_getattr,
2695         .setattr        = proc_setattr,
2696 };
2697
2698 #endif
2699
2700 #ifdef CONFIG_ELF_CORE
2701 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2702                                          size_t count, loff_t *ppos)
2703 {
2704         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2705         struct mm_struct *mm;
2706         char buffer[PROC_NUMBUF];
2707         size_t len;
2708         int ret;
2709
2710         if (!task)
2711                 return -ESRCH;
2712
2713         ret = 0;
2714         mm = get_task_mm(task);
2715         if (mm) {
2716                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2717                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2718                                 MMF_DUMP_FILTER_SHIFT));
2719                 mmput(mm);
2720                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2721         }
2722
2723         put_task_struct(task);
2724
2725         return ret;
2726 }
2727
2728 static ssize_t proc_coredump_filter_write(struct file *file,
2729                                           const char __user *buf,
2730                                           size_t count,
2731                                           loff_t *ppos)
2732 {
2733         struct task_struct *task;
2734         struct mm_struct *mm;
2735         char buffer[PROC_NUMBUF], *end;
2736         unsigned int val;
2737         int ret;
2738         int i;
2739         unsigned long mask;
2740
2741         ret = -EFAULT;
2742         memset(buffer, 0, sizeof(buffer));
2743         if (count > sizeof(buffer) - 1)
2744                 count = sizeof(buffer) - 1;
2745         if (copy_from_user(buffer, buf, count))
2746                 goto out_no_task;
2747
2748         ret = -EINVAL;
2749         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2750         if (*end == '\n')
2751                 end++;
2752         if (end - buffer == 0)
2753                 goto out_no_task;
2754
2755         ret = -ESRCH;
2756         task = get_proc_task(file->f_dentry->d_inode);
2757         if (!task)
2758                 goto out_no_task;
2759
2760         ret = end - buffer;
2761         mm = get_task_mm(task);
2762         if (!mm)
2763                 goto out_no_mm;
2764
2765         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2766                 if (val & mask)
2767                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2768                 else
2769                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2770         }
2771
2772         mmput(mm);
2773  out_no_mm:
2774         put_task_struct(task);
2775  out_no_task:
2776         return ret;
2777 }
2778
2779 static const struct file_operations proc_coredump_filter_operations = {
2780         .read           = proc_coredump_filter_read,
2781         .write          = proc_coredump_filter_write,
2782         .llseek         = generic_file_llseek,
2783 };
2784 #endif
2785
2786 /*
2787  * /proc/self:
2788  */
2789 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2790                               int buflen)
2791 {
2792         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2793         pid_t tgid = task_tgid_nr_ns(current, ns);
2794         char tmp[PROC_NUMBUF];
2795         if (!tgid)
2796                 return -ENOENT;
2797         sprintf(tmp, "%d", tgid);
2798         return vfs_readlink(dentry,buffer,buflen,tmp);
2799 }
2800
2801 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2802 {
2803         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2804         pid_t tgid = task_tgid_nr_ns(current, ns);
2805         char *name = ERR_PTR(-ENOENT);
2806         if (tgid) {
2807                 name = __getname();
2808                 if (!name)
2809                         name = ERR_PTR(-ENOMEM);
2810                 else
2811                         sprintf(name, "%d", tgid);
2812         }
2813         nd_set_link(nd, name);
2814         return NULL;
2815 }
2816
2817 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2818                                 void *cookie)
2819 {
2820         char *s = nd_get_link(nd);
2821         if (!IS_ERR(s))
2822                 __putname(s);
2823 }
2824
2825 static const struct inode_operations proc_self_inode_operations = {
2826         .readlink       = proc_self_readlink,
2827         .follow_link    = proc_self_follow_link,
2828         .put_link       = proc_self_put_link,
2829 };
2830
2831 /*
2832  * proc base
2833  *
2834  * These are the directory entries in the root directory of /proc
2835  * that properly belong to the /proc filesystem, as they describe
2836  * describe something that is process related.
2837  */
2838 static const struct pid_entry proc_base_stuff[] = {
2839         NOD("self", S_IFLNK|S_IRWXUGO,
2840                 &proc_self_inode_operations, NULL, {}),
2841 };
2842
2843 static struct dentry *proc_base_instantiate(struct inode *dir,
2844         struct dentry *dentry, struct task_struct *task, const void *ptr)
2845 {
2846         const struct pid_entry *p = ptr;
2847         struct inode *inode;
2848         struct proc_inode *ei;
2849         struct dentry *error;
2850
2851         /* Allocate the inode */
2852         error = ERR_PTR(-ENOMEM);
2853         inode = new_inode(dir->i_sb);
2854         if (!inode)
2855                 goto out;
2856
2857         /* Initialize the inode */
2858         ei = PROC_I(inode);
2859         inode->i_ino = get_next_ino();
2860         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2861
2862         /*
2863          * grab the reference to the task.
2864          */
2865         ei->pid = get_task_pid(task, PIDTYPE_PID);
2866         if (!ei->pid)
2867                 goto out_iput;
2868
2869         inode->i_mode = p->mode;
2870         if (S_ISDIR(inode->i_mode))
2871                 set_nlink(inode, 2);
2872         if (S_ISLNK(inode->i_mode))
2873                 inode->i_size = 64;
2874         if (p->iop)
2875                 inode->i_op = p->iop;
2876         if (p->fop)
2877                 inode->i_fop = p->fop;
2878         ei->op = p->op;
2879         d_add(dentry, inode);
2880         error = NULL;
2881 out:
2882         return error;
2883 out_iput:
2884         iput(inode);
2885         goto out;
2886 }
2887
2888 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2889 {
2890         struct dentry *error;
2891         struct task_struct *task = get_proc_task(dir);
2892         const struct pid_entry *p, *last;
2893
2894         error = ERR_PTR(-ENOENT);
2895
2896         if (!task)
2897                 goto out_no_task;
2898
2899         /* Lookup the directory entry */
2900         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2901         for (p = proc_base_stuff; p <= last; p++) {
2902                 if (p->len != dentry->d_name.len)
2903                         continue;
2904                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2905                         break;
2906         }
2907         if (p > last)
2908                 goto out;
2909
2910         error = proc_base_instantiate(dir, dentry, task, p);
2911
2912 out:
2913         put_task_struct(task);
2914 out_no_task:
2915         return error;
2916 }
2917
2918 static int proc_base_fill_cache(struct file *filp, void *dirent,
2919         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2920 {
2921         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2922                                 proc_base_instantiate, task, p);
2923 }
2924
2925 #ifdef CONFIG_TASK_IO_ACCOUNTING
2926 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2927 {
2928         struct task_io_accounting acct = task->ioac;
2929         unsigned long flags;
2930         int result;
2931
2932         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2933         if (result)
2934                 return result;
2935
2936         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2937                 result = -EACCES;
2938                 goto out_unlock;
2939         }
2940
2941         if (whole && lock_task_sighand(task, &flags)) {
2942                 struct task_struct *t = task;
2943
2944                 task_io_accounting_add(&acct, &task->signal->ioac);
2945                 while_each_thread(task, t)
2946                         task_io_accounting_add(&acct, &t->ioac);
2947
2948                 unlock_task_sighand(task, &flags);
2949         }
2950         result = sprintf(buffer,
2951                         "rchar: %llu\n"
2952                         "wchar: %llu\n"
2953                         "syscr: %llu\n"
2954                         "syscw: %llu\n"
2955                         "read_bytes: %llu\n"
2956                         "write_bytes: %llu\n"
2957                         "cancelled_write_bytes: %llu\n",
2958                         (unsigned long long)acct.rchar,
2959                         (unsigned long long)acct.wchar,
2960                         (unsigned long long)acct.syscr,
2961                         (unsigned long long)acct.syscw,
2962                         (unsigned long long)acct.read_bytes,
2963                         (unsigned long long)acct.write_bytes,
2964                         (unsigned long long)acct.cancelled_write_bytes);
2965 out_unlock:
2966         mutex_unlock(&task->signal->cred_guard_mutex);
2967         return result;
2968 }
2969
2970 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2971 {
2972         return do_io_accounting(task, buffer, 0);
2973 }
2974
2975 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2976 {
2977         return do_io_accounting(task, buffer, 1);
2978 }
2979 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2980
2981 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2982                                 struct pid *pid, struct task_struct *task)
2983 {
2984         int err = lock_trace(task);
2985         if (!err) {
2986                 seq_printf(m, "%08x\n", task->personality);
2987                 unlock_trace(task);
2988         }
2989         return err;
2990 }
2991
2992 /*
2993  * Thread groups
2994  */
2995 static const struct file_operations proc_task_operations;
2996 static const struct inode_operations proc_task_inode_operations;
2997
2998 static const struct pid_entry tgid_base_stuff[] = {
2999         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
3000         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3001 #ifdef CONFIG_CHECKPOINT_RESTORE
3002         DIR("map_files",  S_IRUSR|S_IXUSR, proc_map_files_inode_operations, proc_map_files_operations),
3003 #endif
3004         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3005         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3006 #ifdef CONFIG_NET
3007         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
3008 #endif
3009         REG("environ",    S_IRUSR, proc_environ_operations),
3010         INF("auxv",       S_IRUSR, proc_pid_auxv),
3011         ONE("status",     S_IRUGO, proc_pid_status),
3012         ONE("personality", S_IRUGO, proc_pid_personality),
3013         INF("limits",     S_IRUGO, proc_pid_limits),
3014 #ifdef CONFIG_SCHED_DEBUG
3015         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3016 #endif
3017 #ifdef CONFIG_SCHED_AUTOGROUP
3018         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
3019 #endif
3020         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3021 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3022         INF("syscall",    S_IRUGO, proc_pid_syscall),
3023 #endif
3024         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
3025         ONE("stat",       S_IRUGO, proc_tgid_stat),
3026         ONE("statm",      S_IRUGO, proc_pid_statm),
3027         REG("maps",       S_IRUGO, proc_maps_operations),
3028 #ifdef CONFIG_NUMA
3029         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
3030 #endif
3031         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
3032         LNK("cwd",        proc_cwd_link),
3033         LNK("root",       proc_root_link),
3034         LNK("exe",        proc_exe_link),
3035         REG("mounts",     S_IRUGO, proc_mounts_operations),
3036         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3037         REG("mountstats", S_IRUSR, proc_mountstats_operations),
3038 #ifdef CONFIG_PROC_PAGE_MONITOR
3039         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3040         REG("smaps",      S_IRUGO, proc_smaps_operations),
3041         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3042 #endif
3043 #ifdef CONFIG_SECURITY
3044         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3045 #endif
3046 #ifdef CONFIG_KALLSYMS
3047         INF("wchan",      S_IRUGO, proc_pid_wchan),
3048 #endif
3049 #ifdef CONFIG_STACKTRACE
3050         ONE("stack",      S_IRUGO, proc_pid_stack),
3051 #endif
3052 #ifdef CONFIG_SCHEDSTATS
3053         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
3054 #endif
3055 #ifdef CONFIG_LATENCYTOP
3056         REG("latency",  S_IRUGO, proc_lstats_operations),
3057 #endif
3058 #ifdef CONFIG_PROC_PID_CPUSET
3059         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
3060 #endif
3061 #ifdef CONFIG_CGROUPS
3062         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3063 #endif
3064         INF("oom_score",  S_IRUGO, proc_oom_score),
3065         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3066         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3067 #ifdef CONFIG_AUDITSYSCALL
3068         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
3069         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3070 #endif
3071 #ifdef CONFIG_FAULT_INJECTION
3072         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3073 #endif
3074 #ifdef CONFIG_ELF_CORE
3075         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
3076 #endif
3077 #ifdef CONFIG_TASK_IO_ACCOUNTING
3078         INF("io",       S_IRUSR, proc_tgid_io_accounting),
3079 #endif
3080 #ifdef CONFIG_HARDWALL
3081         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3082 #endif
3083 };
3084
3085 static int proc_tgid_base_readdir(struct file * filp,
3086                              void * dirent, filldir_t filldir)
3087 {
3088         return proc_pident_readdir(filp,dirent,filldir,
3089                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
3090 }
3091
3092 static const struct file_operations proc_tgid_base_operations = {
3093         .read           = generic_read_dir,
3094         .readdir        = proc_tgid_base_readdir,
3095         .llseek         = default_llseek,
3096 };
3097
3098 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3099         return proc_pident_lookup(dir, dentry,
3100                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
3101 }
3102
3103 static const struct inode_operations proc_tgid_base_inode_operations = {
3104         .lookup         = proc_tgid_base_lookup,
3105         .getattr        = pid_getattr,
3106         .setattr        = proc_setattr,
3107         .permission     = proc_pid_permission,
3108 };
3109
3110 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
3111 {
3112         struct dentry *dentry, *leader, *dir;
3113         char buf[PROC_NUMBUF];
3114         struct qstr name;
3115
3116         name.name = buf;
3117         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3118         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
3119         if (dentry) {
3120                 shrink_dcache_parent(dentry);
3121                 d_drop(dentry);
3122                 dput(dentry);
3123         }
3124
3125         name.name = buf;
3126         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
3127         leader = d_hash_and_lookup(mnt->mnt_root, &name);
3128         if (!leader)
3129                 goto out;
3130
3131         name.name = "task";
3132         name.len = strlen(name.name);
3133         dir = d_hash_and_lookup(leader, &name);
3134         if (!dir)
3135                 goto out_put_leader;
3136
3137         name.name = buf;
3138         name.len = snprintf(buf, sizeof(buf), "%d", pid);
3139         dentry = d_hash_and_lookup(dir, &name);
3140         if (dentry) {
3141                 shrink_dcache_parent(dentry);
3142                 d_drop(dentry);
3143                 dput(dentry);
3144         }
3145
3146         dput(dir);
3147 out_put_leader:
3148         dput(leader);
3149 out:
3150         return;
3151 }
3152
3153 /**
3154  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
3155  * @task: task that should be flushed.
3156  *
3157  * When flushing dentries from proc, one needs to flush them from global
3158  * proc (proc_mnt) and from all the namespaces' procs this task was seen
3159  * in. This call is supposed to do all of this job.
3160  *
3161  * Looks in the dcache for
3162  * /proc/@pid
3163  * /proc/@tgid/task/@pid
3164  * if either directory is present flushes it and all of it'ts children
3165  * from the dcache.
3166  *
3167  * It is safe and reasonable to cache /proc entries for a task until
3168  * that task exits.  After that they just clog up the dcache with
3169  * useless entries, possibly causing useful dcache entries to be
3170  * flushed instead.  This routine is proved to flush those useless
3171  * dcache entries at process exit time.
3172  *
3173  * NOTE: This routine is just an optimization so it does not guarantee
3174  *       that no dcache entries will exist at process exit time it
3175  *       just makes it very unlikely that any will persist.
3176  */
3177
3178 void proc_flush_task(struct task_struct *task)
3179 {
3180         int i;
3181         struct pid *pid, *tgid;
3182         struct upid *upid;
3183
3184         pid = task_pid(task);
3185         tgid = task_tgid(task);
3186
3187         for (i = 0; i <= pid->level; i++) {
3188                 upid = &pid->numbers[i];
3189                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
3190                                         tgid->numbers[i].nr);
3191         }
3192
3193         upid = &pid->numbers[pid->level];
3194         if (upid->nr == 1)
3195                 pid_ns_release_proc(upid->ns);
3196 }
3197
3198 static struct dentry *proc_pid_instantiate(struct inode *dir,
3199                                            struct dentry * dentry,
3200                                            struct task_struct *task, const void *ptr)
3201 {
3202         struct dentry *error = ERR_PTR(-ENOENT);
3203         struct inode *inode;
3204
3205         inode = proc_pid_make_inode(dir->i_sb, task);
3206         if (!inode)
3207                 goto out;
3208
3209         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3210         inode->i_op = &proc_tgid_base_inode_operations;
3211         inode->i_fop = &proc_tgid_base_operations;
3212         inode->i_flags|=S_IMMUTABLE;
3213
3214         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
3215                                                   ARRAY_SIZE(tgid_base_stuff)));
3216
3217         d_set_d_op(dentry, &pid_dentry_operations);
3218
3219         d_add(dentry, inode);
3220         /* Close the race of the process dying before we return the dentry */
3221         if (pid_revalidate(dentry, NULL))
3222                 error = NULL;
3223 out:
3224         return error;
3225 }
3226
3227 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3228 {
3229         struct dentry *result;
3230         struct task_struct *task;
3231         unsigned tgid;
3232         struct pid_namespace *ns;
3233
3234         result = proc_base_lookup(dir, dentry);
3235         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3236                 goto out;
3237
3238         tgid = name_to_int(dentry);
3239         if (tgid == ~0U)
3240                 goto out;
3241
3242         ns = dentry->d_sb->s_fs_info;
3243         rcu_read_lock();
3244         task = find_task_by_pid_ns(tgid, ns);
3245         if (task)
3246                 get_task_struct(task);
3247         rcu_read_unlock();
3248         if (!task)
3249                 goto out;
3250
3251         result = proc_pid_instantiate(dir, dentry, task, NULL);
3252         put_task_struct(task);
3253 out:
3254         return result;
3255 }
3256
3257 /*
3258  * Find the first task with tgid >= tgid
3259  *
3260  */
3261 struct tgid_iter {
3262         unsigned int tgid;
3263         struct task_struct *task;
3264 };
3265 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3266 {
3267         struct pid *pid;
3268
3269         if (iter.task)
3270                 put_task_struct(iter.task);
3271         rcu_read_lock();
3272 retry:
3273         iter.task = NULL;
3274         pid = find_ge_pid(iter.tgid, ns);
3275         if (pid) {
3276                 iter.tgid = pid_nr_ns(pid, ns);
3277                 iter.task = pid_task(pid, PIDTYPE_PID);
3278                 /* What we to know is if the pid we have find is the
3279                  * pid of a thread_group_leader.  Testing for task
3280                  * being a thread_group_leader is the obvious thing
3281                  * todo but there is a window when it fails, due to
3282                  * the pid transfer logic in de_thread.
3283                  *
3284                  * So we perform the straight forward test of seeing
3285                  * if the pid we have found is the pid of a thread
3286                  * group leader, and don't worry if the task we have
3287                  * found doesn't happen to be a thread group leader.
3288                  * As we don't care in the case of readdir.
3289                  */
3290                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3291                         iter.tgid += 1;
3292                         goto retry;
3293                 }
3294                 get_task_struct(iter.task);
3295         }
3296         rcu_read_unlock();
3297         return iter;
3298 }
3299
3300 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3301
3302 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3303         struct tgid_iter iter)
3304 {
3305         char name[PROC_NUMBUF];
3306         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3307         return proc_fill_cache(filp, dirent, filldir, name, len,
3308                                 proc_pid_instantiate, iter.task, NULL);
3309 }
3310
3311 static int fake_filldir(void *buf, const char *name, int namelen,
3312                         loff_t offset, u64 ino, unsigned d_type)
3313 {
3314         return 0;
3315 }
3316
3317 /* for the /proc/ directory itself, after non-process stuff has been done */
3318 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3319 {
3320         unsigned int nr;
3321         struct task_struct *reaper;
3322         struct tgid_iter iter;
3323         struct pid_namespace *ns;
3324         filldir_t __filldir;
3325
3326         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3327                 goto out_no_task;
3328         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3329
3330         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3331         if (!reaper)
3332                 goto out_no_task;
3333
3334         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3335                 const struct pid_entry *p = &proc_base_stuff[nr];
3336                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3337                         goto out;
3338         }
3339
3340         ns = filp->f_dentry->d_sb->s_fs_info;
3341         iter.task = NULL;
3342         iter.tgid = filp->f_pos - TGID_OFFSET;
3343         for (iter = next_tgid(ns, iter);
3344              iter.task;
3345              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3346                 if (has_pid_permissions(ns, iter.task, 2))
3347                         __filldir = filldir;
3348                 else
3349                         __filldir = fake_filldir;
3350
3351                 filp->f_pos = iter.tgid + TGID_OFFSET;
3352                 if (proc_pid_fill_cache(filp, dirent, __filldir, iter) < 0) {
3353                         put_task_struct(iter.task);
3354                         goto out;
3355                 }
3356         }
3357         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3358 out:
3359         put_task_struct(reaper);
3360 out_no_task:
3361         return 0;
3362 }
3363
3364 /*
3365  * Tasks
3366  */
3367 static const struct pid_entry tid_base_stuff[] = {
3368         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3369         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3370         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3371         REG("environ",   S_IRUSR, proc_environ_operations),
3372         INF("auxv",      S_IRUSR, proc_pid_auxv),
3373         ONE("status",    S_IRUGO, proc_pid_status),
3374         ONE("personality", S_IRUGO, proc_pid_personality),
3375         INF("limits",    S_IRUGO, proc_pid_limits),
3376 #ifdef CONFIG_SCHED_DEBUG
3377         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3378 #endif
3379         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3380 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3381         INF("syscall",   S_IRUGO, proc_pid_syscall),
3382 #endif
3383         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3384         ONE("stat",      S_IRUGO, proc_tid_stat),
3385         ONE("statm",     S_IRUGO, proc_pid_statm),
3386         REG("maps",      S_IRUGO, proc_maps_operations),
3387 #ifdef CONFIG_NUMA
3388         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3389 #endif
3390         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3391         LNK("cwd",       proc_cwd_link),
3392         LNK("root",      proc_root_link),
3393         LNK("exe",       proc_exe_link),
3394         REG("mounts",    S_IRUGO, proc_mounts_operations),
3395         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3396 #ifdef CONFIG_PROC_PAGE_MONITOR
3397         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3398         REG("smaps",     S_IRUGO, proc_smaps_operations),
3399         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3400 #endif
3401 #ifdef CONFIG_SECURITY
3402         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3403 #endif
3404 #ifdef CONFIG_KALLSYMS
3405         INF("wchan",     S_IRUGO, proc_pid_wchan),
3406 #endif
3407 #ifdef CONFIG_STACKTRACE
3408         ONE("stack",      S_IRUGO, proc_pid_stack),
3409 #endif
3410 #ifdef CONFIG_SCHEDSTATS
3411         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3412 #endif
3413 #ifdef CONFIG_LATENCYTOP
3414         REG("latency",  S_IRUGO, proc_lstats_operations),
3415 #endif
3416 #ifdef CONFIG_PROC_PID_CPUSET
3417         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3418 #endif
3419 #ifdef CONFIG_CGROUPS
3420         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3421 #endif
3422         INF("oom_score", S_IRUGO, proc_oom_score),
3423         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3424         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3425 #ifdef CONFIG_AUDITSYSCALL
3426         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3427         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3428 #endif
3429 #ifdef CONFIG_FAULT_INJECTION
3430         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3431 #endif
3432 #ifdef CONFIG_TASK_IO_ACCOUNTING
3433         INF("io",       S_IRUSR, proc_tid_io_accounting),
3434 #endif
3435 #ifdef CONFIG_HARDWALL
3436         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3437 #endif
3438 };
3439
3440 static int proc_tid_base_readdir(struct file * filp,
3441                              void * dirent, filldir_t filldir)
3442 {
3443         return proc_pident_readdir(filp,dirent,filldir,
3444                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3445 }
3446
3447 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3448         return proc_pident_lookup(dir, dentry,
3449                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3450 }
3451
3452 static const struct file_operations proc_tid_base_operations = {
3453         .read           = generic_read_dir,
3454         .readdir        = proc_tid_base_readdir,
3455         .llseek         = default_llseek,
3456 };
3457
3458 static const struct inode_operations proc_tid_base_inode_operations = {
3459         .lookup         = proc_tid_base_lookup,
3460         .getattr        = pid_getattr,
3461         .setattr        = proc_setattr,
3462 };
3463
3464 static struct dentry *proc_task_instantiate(struct inode *dir,
3465         struct dentry *dentry, struct task_struct *task, const void *ptr)
3466 {
3467         struct dentry *error = ERR_PTR(-ENOENT);
3468         struct inode *inode;
3469         inode = proc_pid_make_inode(dir->i_sb, task);
3470
3471         if (!inode)
3472                 goto out;
3473         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3474         inode->i_op = &proc_tid_base_inode_operations;
3475         inode->i_fop = &proc_tid_base_operations;
3476         inode->i_flags|=S_IMMUTABLE;
3477
3478         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3479                                                   ARRAY_SIZE(tid_base_stuff)));
3480
3481         d_set_d_op(dentry, &pid_dentry_operations);
3482
3483         d_add(dentry, inode);
3484         /* Close the race of the process dying before we return the dentry */
3485         if (pid_revalidate(dentry, NULL))
3486                 error = NULL;
3487 out:
3488         return error;
3489 }
3490
3491 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3492 {
3493         struct dentry *result = ERR_PTR(-ENOENT);
3494         struct task_struct *task;
3495         struct task_struct *leader = get_proc_task(dir);
3496         unsigned tid;
3497         struct pid_namespace *ns;
3498
3499         if (!leader)
3500                 goto out_no_task;
3501
3502         tid = name_to_int(dentry);
3503         if (tid == ~0U)
3504                 goto out;
3505
3506         ns = dentry->d_sb->s_fs_info;
3507         rcu_read_lock();
3508         task = find_task_by_pid_ns(tid, ns);
3509         if (task)
3510                 get_task_struct(task);
3511         rcu_read_unlock();
3512         if (!task)
3513                 goto out;
3514         if (!same_thread_group(leader, task))
3515                 goto out_drop_task;
3516
3517         result = proc_task_instantiate(dir, dentry, task, NULL);
3518 out_drop_task:
3519         put_task_struct(task);
3520 out:
3521         put_task_struct(leader);
3522 out_no_task:
3523         return result;
3524 }
3525
3526 /*
3527  * Find the first tid of a thread group to return to user space.
3528  *
3529  * Usually this is just the thread group leader, but if the users
3530  * buffer was too small or there was a seek into the middle of the
3531  * directory we have more work todo.
3532  *
3533  * In the case of a short read we start with find_task_by_pid.
3534  *
3535  * In the case of a seek we start with the leader and walk nr
3536  * threads past it.
3537  */
3538 static struct task_struct *first_tid(struct task_struct *leader,
3539                 int tid, int nr, struct pid_namespace *ns)
3540 {
3541         struct task_struct *pos;
3542
3543         rcu_read_lock();
3544         /* Attempt to start with the pid of a thread */
3545         if (tid && (nr > 0)) {
3546                 pos = find_task_by_pid_ns(tid, ns);
3547                 if (pos && (pos->group_leader == leader))
3548                         goto found;
3549         }
3550
3551         /* If nr exceeds the number of threads there is nothing todo */
3552         pos = NULL;
3553         if (nr && nr >= get_nr_threads(leader))
3554                 goto out;
3555
3556         /* If we haven't found our starting place yet start
3557          * with the leader and walk nr threads forward.
3558          */
3559         for (pos = leader; nr > 0; --nr) {
3560                 pos = next_thread(pos);
3561                 if (pos == leader) {
3562                         pos = NULL;
3563                         goto out;
3564                 }
3565         }
3566 found:
3567         get_task_struct(pos);
3568 out:
3569         rcu_read_unlock();
3570         return pos;
3571 }
3572
3573 /*
3574  * Find the next thread in the thread list.
3575  * Return NULL if there is an error or no next thread.
3576  *
3577  * The reference to the input task_struct is released.
3578  */
3579 static struct task_struct *next_tid(struct task_struct *start)
3580 {
3581         struct task_struct *pos = NULL;
3582         rcu_read_lock();
3583         if (pid_alive(start)) {
3584                 pos = next_thread(start);
3585                 if (thread_group_leader(pos))
3586                         pos = NULL;
3587                 else
3588                         get_task_struct(pos);
3589         }
3590         rcu_read_unlock();
3591         put_task_struct(start);
3592         return pos;
3593 }
3594
3595 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3596         struct task_struct *task, int tid)
3597 {
3598         char name[PROC_NUMBUF];
3599         int len = snprintf(name, sizeof(name), "%d", tid);
3600         return proc_fill_cache(filp, dirent, filldir, name, len,
3601                                 proc_task_instantiate, task, NULL);
3602 }
3603
3604 /* for the /proc/TGID/task/ directories */
3605 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3606 {
3607         struct dentry *dentry = filp->f_path.dentry;
3608         struct inode *inode = dentry->d_inode;
3609         struct task_struct *leader = NULL;
3610         struct task_struct *task;
3611         int retval = -ENOENT;
3612         ino_t ino;
3613         int tid;
3614         struct pid_namespace *ns;
3615
3616         task = get_proc_task(inode);
3617         if (!task)
3618                 goto out_no_task;
3619         rcu_read_lock();
3620         if (pid_alive(task)) {
3621                 leader = task->group_leader;
3622                 get_task_struct(leader);
3623         }
3624         rcu_read_unlock();
3625         put_task_struct(task);
3626         if (!leader)
3627                 goto out_no_task;
3628         retval = 0;
3629
3630         switch ((unsigned long)filp->f_pos) {
3631         case 0:
3632                 ino = inode->i_ino;
3633                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3634                         goto out;
3635                 filp->f_pos++;
3636                 /* fall through */
3637         case 1:
3638                 ino = parent_ino(dentry);
3639                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3640                         goto out;
3641                 filp->f_pos++;
3642                 /* fall through */
3643         }
3644
3645         /* f_version caches the tgid value that the last readdir call couldn't
3646          * return. lseek aka telldir automagically resets f_version to 0.
3647          */
3648         ns = filp->f_dentry->d_sb->s_fs_info;
3649         tid = (int)filp->f_version;
3650         filp->f_version = 0;
3651         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3652              task;
3653              task = next_tid(task), filp->f_pos++) {
3654                 tid = task_pid_nr_ns(task, ns);
3655                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3656                         /* returning this tgid failed, save it as the first
3657                          * pid for the next readir call */
3658                         filp->f_version = (u64)tid;
3659                         put_task_struct(task);
3660                         break;
3661                 }
3662         }
3663 out:
3664         put_task_struct(leader);
3665 out_no_task:
3666         return retval;
3667 }
3668
3669 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3670 {
3671         struct inode *inode = dentry->d_inode;
3672         struct task_struct *p = get_proc_task(inode);
3673         generic_fillattr(inode, stat);
3674
3675         if (p) {
3676                 stat->nlink += get_nr_threads(p);
3677                 put_task_struct(p);
3678         }
3679
3680         return 0;
3681 }
3682
3683 static const struct inode_operations proc_task_inode_operations = {
3684         .lookup         = proc_task_lookup,
3685         .getattr        = proc_task_getattr,
3686         .setattr        = proc_setattr,
3687         .permission     = proc_pid_permission,
3688 };
3689
3690 static const struct file_operations proc_task_operations = {
3691         .read           = generic_read_dir,
3692         .readdir        = proc_task_readdir,
3693         .llseek         = default_llseek,
3694 };