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