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