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