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