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                         unsigned int f_flags;
1924                         struct fdtable *fdt;
1925
1926                         fdt = files_fdtable(files);
1927                         f_flags = file->f_flags & ~O_CLOEXEC;
1928                         if (FD_ISSET(fd, fdt->close_on_exec))
1929                                 f_flags |= O_CLOEXEC;
1930
1931                         if (path) {
1932                                 *path = file->f_path;
1933                                 path_get(&file->f_path);
1934                         }
1935                         if (info)
1936                                 snprintf(info, PROC_FDINFO_MAX,
1937                                          "pos:\t%lli\n"
1938                                          "flags:\t0%o\n",
1939                                          (long long) file->f_pos,
1940                                          f_flags);
1941                         spin_unlock(&files->file_lock);
1942                         put_files_struct(files);
1943                         return 0;
1944                 }
1945                 spin_unlock(&files->file_lock);
1946                 put_files_struct(files);
1947         }
1948         return -ENOENT;
1949 }
1950
1951 static int proc_fd_link(struct inode *inode, struct path *path)
1952 {
1953         return proc_fd_info(inode, path, NULL);
1954 }
1955
1956 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1957 {
1958         struct inode *inode;
1959         struct task_struct *task;
1960         int fd;
1961         struct files_struct *files;
1962         const struct cred *cred;
1963
1964         if (nd && nd->flags & LOOKUP_RCU)
1965                 return -ECHILD;
1966
1967         inode = dentry->d_inode;
1968         task = get_proc_task(inode);
1969         fd = proc_fd(inode);
1970
1971         if (task) {
1972                 files = get_files_struct(task);
1973                 if (files) {
1974                         rcu_read_lock();
1975                         if (fcheck_files(files, fd)) {
1976                                 rcu_read_unlock();
1977                                 put_files_struct(files);
1978                                 if (task_dumpable(task)) {
1979                                         rcu_read_lock();
1980                                         cred = __task_cred(task);
1981                                         inode->i_uid = cred->euid;
1982                                         inode->i_gid = cred->egid;
1983                                         rcu_read_unlock();
1984                                 } else {
1985                                         inode->i_uid = 0;
1986                                         inode->i_gid = 0;
1987                                 }
1988                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1989                                 security_task_to_inode(task, inode);
1990                                 put_task_struct(task);
1991                                 return 1;
1992                         }
1993                         rcu_read_unlock();
1994                         put_files_struct(files);
1995                 }
1996                 put_task_struct(task);
1997         }
1998         d_drop(dentry);
1999         return 0;
2000 }
2001
2002 static const struct dentry_operations tid_fd_dentry_operations =
2003 {
2004         .d_revalidate   = tid_fd_revalidate,
2005         .d_delete       = pid_delete_dentry,
2006 };
2007
2008 static struct dentry *proc_fd_instantiate(struct inode *dir,
2009         struct dentry *dentry, struct task_struct *task, const void *ptr)
2010 {
2011         unsigned fd = *(const unsigned *)ptr;
2012         struct file *file;
2013         struct files_struct *files;
2014         struct inode *inode;
2015         struct proc_inode *ei;
2016         struct dentry *error = ERR_PTR(-ENOENT);
2017
2018         inode = proc_pid_make_inode(dir->i_sb, task);
2019         if (!inode)
2020                 goto out;
2021         ei = PROC_I(inode);
2022         ei->fd = fd;
2023         files = get_files_struct(task);
2024         if (!files)
2025                 goto out_iput;
2026         inode->i_mode = S_IFLNK;
2027
2028         /*
2029          * We are not taking a ref to the file structure, so we must
2030          * hold ->file_lock.
2031          */
2032         spin_lock(&files->file_lock);
2033         file = fcheck_files(files, fd);
2034         if (!file)
2035                 goto out_unlock;
2036         if (file->f_mode & FMODE_READ)
2037                 inode->i_mode |= S_IRUSR | S_IXUSR;
2038         if (file->f_mode & FMODE_WRITE)
2039                 inode->i_mode |= S_IWUSR | S_IXUSR;
2040         spin_unlock(&files->file_lock);
2041         put_files_struct(files);
2042
2043         inode->i_op = &proc_pid_link_inode_operations;
2044         inode->i_size = 64;
2045         ei->op.proc_get_link = proc_fd_link;
2046         d_set_d_op(dentry, &tid_fd_dentry_operations);
2047         d_add(dentry, inode);
2048         /* Close the race of the process dying before we return the dentry */
2049         if (tid_fd_revalidate(dentry, NULL))
2050                 error = NULL;
2051
2052  out:
2053         return error;
2054 out_unlock:
2055         spin_unlock(&files->file_lock);
2056         put_files_struct(files);
2057 out_iput:
2058         iput(inode);
2059         goto out;
2060 }
2061
2062 static struct dentry *proc_lookupfd_common(struct inode *dir,
2063                                            struct dentry *dentry,
2064                                            instantiate_t instantiate)
2065 {
2066         struct task_struct *task = get_proc_task(dir);
2067         unsigned fd = name_to_int(dentry);
2068         struct dentry *result = ERR_PTR(-ENOENT);
2069
2070         if (!task)
2071                 goto out_no_task;
2072         if (fd == ~0U)
2073                 goto out;
2074
2075         result = instantiate(dir, dentry, task, &fd);
2076 out:
2077         put_task_struct(task);
2078 out_no_task:
2079         return result;
2080 }
2081
2082 static int proc_readfd_common(struct file * filp, void * dirent,
2083                               filldir_t filldir, instantiate_t instantiate)
2084 {
2085         struct dentry *dentry = filp->f_path.dentry;
2086         struct inode *inode = dentry->d_inode;
2087         struct task_struct *p = get_proc_task(inode);
2088         unsigned int fd, ino;
2089         int retval;
2090         struct files_struct * files;
2091
2092         retval = -ENOENT;
2093         if (!p)
2094                 goto out_no_task;
2095         retval = 0;
2096
2097         fd = filp->f_pos;
2098         switch (fd) {
2099                 case 0:
2100                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2101                                 goto out;
2102                         filp->f_pos++;
2103                 case 1:
2104                         ino = parent_ino(dentry);
2105                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2106                                 goto out;
2107                         filp->f_pos++;
2108                 default:
2109                         files = get_files_struct(p);
2110                         if (!files)
2111                                 goto out;
2112                         rcu_read_lock();
2113                         for (fd = filp->f_pos-2;
2114                              fd < files_fdtable(files)->max_fds;
2115                              fd++, filp->f_pos++) {
2116                                 char name[PROC_NUMBUF];
2117                                 int len;
2118
2119                                 if (!fcheck_files(files, fd))
2120                                         continue;
2121                                 rcu_read_unlock();
2122
2123                                 len = snprintf(name, sizeof(name), "%d", fd);
2124                                 if (proc_fill_cache(filp, dirent, filldir,
2125                                                     name, len, instantiate,
2126                                                     p, &fd) < 0) {
2127                                         rcu_read_lock();
2128                                         break;
2129                                 }
2130                                 rcu_read_lock();
2131                         }
2132                         rcu_read_unlock();
2133                         put_files_struct(files);
2134         }
2135 out:
2136         put_task_struct(p);
2137 out_no_task:
2138         return retval;
2139 }
2140
2141 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2142                                     struct nameidata *nd)
2143 {
2144         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2145 }
2146
2147 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2148 {
2149         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2150 }
2151
2152 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2153                                       size_t len, loff_t *ppos)
2154 {
2155         char tmp[PROC_FDINFO_MAX];
2156         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2157         if (!err)
2158                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2159         return err;
2160 }
2161
2162 static const struct file_operations proc_fdinfo_file_operations = {
2163         .open           = nonseekable_open,
2164         .read           = proc_fdinfo_read,
2165         .llseek         = no_llseek,
2166 };
2167
2168 static const struct file_operations proc_fd_operations = {
2169         .read           = generic_read_dir,
2170         .readdir        = proc_readfd,
2171         .llseek         = default_llseek,
2172 };
2173
2174 /*
2175  * /proc/pid/fd needs a special permission handler so that a process can still
2176  * access /proc/self/fd after it has executed a setuid().
2177  */
2178 static int proc_fd_permission(struct inode *inode, int mask, unsigned int flags)
2179 {
2180         int rv = generic_permission(inode, mask, flags, NULL);
2181         if (rv == 0)
2182                 return 0;
2183         if (task_pid(current) == proc_pid(inode))
2184                 rv = 0;
2185         return rv;
2186 }
2187
2188 /*
2189  * proc directories can do almost nothing..
2190  */
2191 static const struct inode_operations proc_fd_inode_operations = {
2192         .lookup         = proc_lookupfd,
2193         .permission     = proc_fd_permission,
2194         .setattr        = proc_setattr,
2195 };
2196
2197 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2198         struct dentry *dentry, struct task_struct *task, const void *ptr)
2199 {
2200         unsigned fd = *(unsigned *)ptr;
2201         struct inode *inode;
2202         struct proc_inode *ei;
2203         struct dentry *error = ERR_PTR(-ENOENT);
2204
2205         inode = proc_pid_make_inode(dir->i_sb, task);
2206         if (!inode)
2207                 goto out;
2208         ei = PROC_I(inode);
2209         ei->fd = fd;
2210         inode->i_mode = S_IFREG | S_IRUSR;
2211         inode->i_fop = &proc_fdinfo_file_operations;
2212         d_set_d_op(dentry, &tid_fd_dentry_operations);
2213         d_add(dentry, inode);
2214         /* Close the race of the process dying before we return the dentry */
2215         if (tid_fd_revalidate(dentry, NULL))
2216                 error = NULL;
2217
2218  out:
2219         return error;
2220 }
2221
2222 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2223                                         struct dentry *dentry,
2224                                         struct nameidata *nd)
2225 {
2226         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2227 }
2228
2229 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2230 {
2231         return proc_readfd_common(filp, dirent, filldir,
2232                                   proc_fdinfo_instantiate);
2233 }
2234
2235 static const struct file_operations proc_fdinfo_operations = {
2236         .read           = generic_read_dir,
2237         .readdir        = proc_readfdinfo,
2238         .llseek         = default_llseek,
2239 };
2240
2241 /*
2242  * proc directories can do almost nothing..
2243  */
2244 static const struct inode_operations proc_fdinfo_inode_operations = {
2245         .lookup         = proc_lookupfdinfo,
2246         .setattr        = proc_setattr,
2247 };
2248
2249
2250 static struct dentry *proc_pident_instantiate(struct inode *dir,
2251         struct dentry *dentry, struct task_struct *task, const void *ptr)
2252 {
2253         const struct pid_entry *p = ptr;
2254         struct inode *inode;
2255         struct proc_inode *ei;
2256         struct dentry *error = ERR_PTR(-ENOENT);
2257
2258         inode = proc_pid_make_inode(dir->i_sb, task);
2259         if (!inode)
2260                 goto out;
2261
2262         ei = PROC_I(inode);
2263         inode->i_mode = p->mode;
2264         if (S_ISDIR(inode->i_mode))
2265                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
2266         if (p->iop)
2267                 inode->i_op = p->iop;
2268         if (p->fop)
2269                 inode->i_fop = p->fop;
2270         ei->op = p->op;
2271         d_set_d_op(dentry, &pid_dentry_operations);
2272         d_add(dentry, inode);
2273         /* Close the race of the process dying before we return the dentry */
2274         if (pid_revalidate(dentry, NULL))
2275                 error = NULL;
2276 out:
2277         return error;
2278 }
2279
2280 static struct dentry *proc_pident_lookup(struct inode *dir,
2281                                          struct dentry *dentry,
2282                                          const struct pid_entry *ents,
2283                                          unsigned int nents)
2284 {
2285         struct dentry *error;
2286         struct task_struct *task = get_proc_task(dir);
2287         const struct pid_entry *p, *last;
2288
2289         error = ERR_PTR(-ENOENT);
2290
2291         if (!task)
2292                 goto out_no_task;
2293
2294         /*
2295          * Yes, it does not scale. And it should not. Don't add
2296          * new entries into /proc/<tgid>/ without very good reasons.
2297          */
2298         last = &ents[nents - 1];
2299         for (p = ents; p <= last; p++) {
2300                 if (p->len != dentry->d_name.len)
2301                         continue;
2302                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2303                         break;
2304         }
2305         if (p > last)
2306                 goto out;
2307
2308         error = proc_pident_instantiate(dir, dentry, task, p);
2309 out:
2310         put_task_struct(task);
2311 out_no_task:
2312         return error;
2313 }
2314
2315 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2316         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2317 {
2318         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2319                                 proc_pident_instantiate, task, p);
2320 }
2321
2322 static int proc_pident_readdir(struct file *filp,
2323                 void *dirent, filldir_t filldir,
2324                 const struct pid_entry *ents, unsigned int nents)
2325 {
2326         int i;
2327         struct dentry *dentry = filp->f_path.dentry;
2328         struct inode *inode = dentry->d_inode;
2329         struct task_struct *task = get_proc_task(inode);
2330         const struct pid_entry *p, *last;
2331         ino_t ino;
2332         int ret;
2333
2334         ret = -ENOENT;
2335         if (!task)
2336                 goto out_no_task;
2337
2338         ret = 0;
2339         i = filp->f_pos;
2340         switch (i) {
2341         case 0:
2342                 ino = inode->i_ino;
2343                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2344                         goto out;
2345                 i++;
2346                 filp->f_pos++;
2347                 /* fall through */
2348         case 1:
2349                 ino = parent_ino(dentry);
2350                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2351                         goto out;
2352                 i++;
2353                 filp->f_pos++;
2354                 /* fall through */
2355         default:
2356                 i -= 2;
2357                 if (i >= nents) {
2358                         ret = 1;
2359                         goto out;
2360                 }
2361                 p = ents + i;
2362                 last = &ents[nents - 1];
2363                 while (p <= last) {
2364                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2365                                 goto out;
2366                         filp->f_pos++;
2367                         p++;
2368                 }
2369         }
2370
2371         ret = 1;
2372 out:
2373         put_task_struct(task);
2374 out_no_task:
2375         return ret;
2376 }
2377
2378 #ifdef CONFIG_SECURITY
2379 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2380                                   size_t count, loff_t *ppos)
2381 {
2382         struct inode * inode = file->f_path.dentry->d_inode;
2383         char *p = NULL;
2384         ssize_t length;
2385         struct task_struct *task = get_proc_task(inode);
2386
2387         if (!task)
2388                 return -ESRCH;
2389
2390         length = security_getprocattr(task,
2391                                       (char*)file->f_path.dentry->d_name.name,
2392                                       &p);
2393         put_task_struct(task);
2394         if (length > 0)
2395                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2396         kfree(p);
2397         return length;
2398 }
2399
2400 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2401                                    size_t count, loff_t *ppos)
2402 {
2403         struct inode * inode = file->f_path.dentry->d_inode;
2404         char *page;
2405         ssize_t length;
2406         struct task_struct *task = get_proc_task(inode);
2407
2408         length = -ESRCH;
2409         if (!task)
2410                 goto out_no_task;
2411         if (count > PAGE_SIZE)
2412                 count = PAGE_SIZE;
2413
2414         /* No partial writes. */
2415         length = -EINVAL;
2416         if (*ppos != 0)
2417                 goto out;
2418
2419         length = -ENOMEM;
2420         page = (char*)__get_free_page(GFP_TEMPORARY);
2421         if (!page)
2422                 goto out;
2423
2424         length = -EFAULT;
2425         if (copy_from_user(page, buf, count))
2426                 goto out_free;
2427
2428         /* Guard against adverse ptrace interaction */
2429         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2430         if (length < 0)
2431                 goto out_free;
2432
2433         length = security_setprocattr(task,
2434                                       (char*)file->f_path.dentry->d_name.name,
2435                                       (void*)page, count);
2436         mutex_unlock(&task->signal->cred_guard_mutex);
2437 out_free:
2438         free_page((unsigned long) page);
2439 out:
2440         put_task_struct(task);
2441 out_no_task:
2442         return length;
2443 }
2444
2445 static const struct file_operations proc_pid_attr_operations = {
2446         .read           = proc_pid_attr_read,
2447         .write          = proc_pid_attr_write,
2448         .llseek         = generic_file_llseek,
2449 };
2450
2451 static const struct pid_entry attr_dir_stuff[] = {
2452         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2453         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2454         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2455         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2456         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2457         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2458 };
2459
2460 static int proc_attr_dir_readdir(struct file * filp,
2461                              void * dirent, filldir_t filldir)
2462 {
2463         return proc_pident_readdir(filp,dirent,filldir,
2464                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2465 }
2466
2467 static const struct file_operations proc_attr_dir_operations = {
2468         .read           = generic_read_dir,
2469         .readdir        = proc_attr_dir_readdir,
2470         .llseek         = default_llseek,
2471 };
2472
2473 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2474                                 struct dentry *dentry, struct nameidata *nd)
2475 {
2476         return proc_pident_lookup(dir, dentry,
2477                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2478 }
2479
2480 static const struct inode_operations proc_attr_dir_inode_operations = {
2481         .lookup         = proc_attr_dir_lookup,
2482         .getattr        = pid_getattr,
2483         .setattr        = proc_setattr,
2484 };
2485
2486 #endif
2487
2488 #ifdef CONFIG_ELF_CORE
2489 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2490                                          size_t count, loff_t *ppos)
2491 {
2492         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2493         struct mm_struct *mm;
2494         char buffer[PROC_NUMBUF];
2495         size_t len;
2496         int ret;
2497
2498         if (!task)
2499                 return -ESRCH;
2500
2501         ret = 0;
2502         mm = get_task_mm(task);
2503         if (mm) {
2504                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2505                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2506                                 MMF_DUMP_FILTER_SHIFT));
2507                 mmput(mm);
2508                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2509         }
2510
2511         put_task_struct(task);
2512
2513         return ret;
2514 }
2515
2516 static ssize_t proc_coredump_filter_write(struct file *file,
2517                                           const char __user *buf,
2518                                           size_t count,
2519                                           loff_t *ppos)
2520 {
2521         struct task_struct *task;
2522         struct mm_struct *mm;
2523         char buffer[PROC_NUMBUF], *end;
2524         unsigned int val;
2525         int ret;
2526         int i;
2527         unsigned long mask;
2528
2529         ret = -EFAULT;
2530         memset(buffer, 0, sizeof(buffer));
2531         if (count > sizeof(buffer) - 1)
2532                 count = sizeof(buffer) - 1;
2533         if (copy_from_user(buffer, buf, count))
2534                 goto out_no_task;
2535
2536         ret = -EINVAL;
2537         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2538         if (*end == '\n')
2539                 end++;
2540         if (end - buffer == 0)
2541                 goto out_no_task;
2542
2543         ret = -ESRCH;
2544         task = get_proc_task(file->f_dentry->d_inode);
2545         if (!task)
2546                 goto out_no_task;
2547
2548         ret = end - buffer;
2549         mm = get_task_mm(task);
2550         if (!mm)
2551                 goto out_no_mm;
2552
2553         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2554                 if (val & mask)
2555                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2556                 else
2557                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2558         }
2559
2560         mmput(mm);
2561  out_no_mm:
2562         put_task_struct(task);
2563  out_no_task:
2564         return ret;
2565 }
2566
2567 static const struct file_operations proc_coredump_filter_operations = {
2568         .read           = proc_coredump_filter_read,
2569         .write          = proc_coredump_filter_write,
2570         .llseek         = generic_file_llseek,
2571 };
2572 #endif
2573
2574 /*
2575  * /proc/self:
2576  */
2577 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2578                               int buflen)
2579 {
2580         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2581         pid_t tgid = task_tgid_nr_ns(current, ns);
2582         char tmp[PROC_NUMBUF];
2583         if (!tgid)
2584                 return -ENOENT;
2585         sprintf(tmp, "%d", tgid);
2586         return vfs_readlink(dentry,buffer,buflen,tmp);
2587 }
2588
2589 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2590 {
2591         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2592         pid_t tgid = task_tgid_nr_ns(current, ns);
2593         char *name = ERR_PTR(-ENOENT);
2594         if (tgid) {
2595                 name = __getname();
2596                 if (!name)
2597                         name = ERR_PTR(-ENOMEM);
2598                 else
2599                         sprintf(name, "%d", tgid);
2600         }
2601         nd_set_link(nd, name);
2602         return NULL;
2603 }
2604
2605 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2606                                 void *cookie)
2607 {
2608         char *s = nd_get_link(nd);
2609         if (!IS_ERR(s))
2610                 __putname(s);
2611 }
2612
2613 static const struct inode_operations proc_self_inode_operations = {
2614         .readlink       = proc_self_readlink,
2615         .follow_link    = proc_self_follow_link,
2616         .put_link       = proc_self_put_link,
2617 };
2618
2619 /*
2620  * proc base
2621  *
2622  * These are the directory entries in the root directory of /proc
2623  * that properly belong to the /proc filesystem, as they describe
2624  * describe something that is process related.
2625  */
2626 static const struct pid_entry proc_base_stuff[] = {
2627         NOD("self", S_IFLNK|S_IRWXUGO,
2628                 &proc_self_inode_operations, NULL, {}),
2629 };
2630
2631 static struct dentry *proc_base_instantiate(struct inode *dir,
2632         struct dentry *dentry, struct task_struct *task, const void *ptr)
2633 {
2634         const struct pid_entry *p = ptr;
2635         struct inode *inode;
2636         struct proc_inode *ei;
2637         struct dentry *error;
2638
2639         /* Allocate the inode */
2640         error = ERR_PTR(-ENOMEM);
2641         inode = new_inode(dir->i_sb);
2642         if (!inode)
2643                 goto out;
2644
2645         /* Initialize the inode */
2646         ei = PROC_I(inode);
2647         inode->i_ino = get_next_ino();
2648         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2649
2650         /*
2651          * grab the reference to the task.
2652          */
2653         ei->pid = get_task_pid(task, PIDTYPE_PID);
2654         if (!ei->pid)
2655                 goto out_iput;
2656
2657         inode->i_mode = p->mode;
2658         if (S_ISDIR(inode->i_mode))
2659                 inode->i_nlink = 2;
2660         if (S_ISLNK(inode->i_mode))
2661                 inode->i_size = 64;
2662         if (p->iop)
2663                 inode->i_op = p->iop;
2664         if (p->fop)
2665                 inode->i_fop = p->fop;
2666         ei->op = p->op;
2667         d_add(dentry, inode);
2668         error = NULL;
2669 out:
2670         return error;
2671 out_iput:
2672         iput(inode);
2673         goto out;
2674 }
2675
2676 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2677 {
2678         struct dentry *error;
2679         struct task_struct *task = get_proc_task(dir);
2680         const struct pid_entry *p, *last;
2681
2682         error = ERR_PTR(-ENOENT);
2683
2684         if (!task)
2685                 goto out_no_task;
2686
2687         /* Lookup the directory entry */
2688         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2689         for (p = proc_base_stuff; p <= last; p++) {
2690                 if (p->len != dentry->d_name.len)
2691                         continue;
2692                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2693                         break;
2694         }
2695         if (p > last)
2696                 goto out;
2697
2698         error = proc_base_instantiate(dir, dentry, task, p);
2699
2700 out:
2701         put_task_struct(task);
2702 out_no_task:
2703         return error;
2704 }
2705
2706 static int proc_base_fill_cache(struct file *filp, void *dirent,
2707         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2708 {
2709         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2710                                 proc_base_instantiate, task, p);
2711 }
2712
2713 #ifdef CONFIG_TASK_IO_ACCOUNTING
2714 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2715 {
2716         struct task_io_accounting acct = task->ioac;
2717         unsigned long flags;
2718         int result;
2719
2720         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2721         if (result)
2722                 return result;
2723
2724         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2725                 result = -EACCES;
2726                 goto out_unlock;
2727         }
2728
2729         if (whole && lock_task_sighand(task, &flags)) {
2730                 struct task_struct *t = task;
2731
2732                 task_io_accounting_add(&acct, &task->signal->ioac);
2733                 while_each_thread(task, t)
2734                         task_io_accounting_add(&acct, &t->ioac);
2735
2736                 unlock_task_sighand(task, &flags);
2737         }
2738         result = sprintf(buffer,
2739                         "rchar: %llu\n"
2740                         "wchar: %llu\n"
2741                         "syscr: %llu\n"
2742                         "syscw: %llu\n"
2743                         "read_bytes: %llu\n"
2744                         "write_bytes: %llu\n"
2745                         "cancelled_write_bytes: %llu\n",
2746                         (unsigned long long)acct.rchar,
2747                         (unsigned long long)acct.wchar,
2748                         (unsigned long long)acct.syscr,
2749                         (unsigned long long)acct.syscw,
2750                         (unsigned long long)acct.read_bytes,
2751                         (unsigned long long)acct.write_bytes,
2752                         (unsigned long long)acct.cancelled_write_bytes);
2753 out_unlock:
2754         mutex_unlock(&task->signal->cred_guard_mutex);
2755         return result;
2756 }
2757
2758 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2759 {
2760         return do_io_accounting(task, buffer, 0);
2761 }
2762
2763 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2764 {
2765         return do_io_accounting(task, buffer, 1);
2766 }
2767 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2768
2769 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2770                                 struct pid *pid, struct task_struct *task)
2771 {
2772         int err = lock_trace(task);
2773         if (!err) {
2774                 seq_printf(m, "%08x\n", task->personality);
2775                 unlock_trace(task);
2776         }
2777         return err;
2778 }
2779
2780 /*
2781  * Thread groups
2782  */
2783 static const struct file_operations proc_task_operations;
2784 static const struct inode_operations proc_task_inode_operations;
2785
2786 static const struct pid_entry tgid_base_stuff[] = {
2787         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2788         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2789         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2790         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2791 #ifdef CONFIG_NET
2792         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2793 #endif
2794         REG("environ",    S_IRUSR, proc_environ_operations),
2795         INF("auxv",       S_IRUSR, proc_pid_auxv),
2796         ONE("status",     S_IRUGO, proc_pid_status),
2797         ONE("personality", S_IRUGO, proc_pid_personality),
2798         INF("limits",     S_IRUGO, proc_pid_limits),
2799 #ifdef CONFIG_SCHED_DEBUG
2800         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2801 #endif
2802 #ifdef CONFIG_SCHED_AUTOGROUP
2803         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2804 #endif
2805         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2806 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2807         INF("syscall",    S_IRUGO, proc_pid_syscall),
2808 #endif
2809         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2810         ONE("stat",       S_IRUGO, proc_tgid_stat),
2811         ONE("statm",      S_IRUGO, proc_pid_statm),
2812         REG("maps",       S_IRUGO, proc_maps_operations),
2813 #ifdef CONFIG_NUMA
2814         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2815 #endif
2816         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2817         LNK("cwd",        proc_cwd_link),
2818         LNK("root",       proc_root_link),
2819         LNK("exe",        proc_exe_link),
2820         REG("mounts",     S_IRUGO, proc_mounts_operations),
2821         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2822         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2823 #ifdef CONFIG_PROC_PAGE_MONITOR
2824         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2825         REG("smaps",      S_IRUGO, proc_smaps_operations),
2826         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2827 #endif
2828 #ifdef CONFIG_SECURITY
2829         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2830 #endif
2831 #ifdef CONFIG_KALLSYMS
2832         INF("wchan",      S_IRUGO, proc_pid_wchan),
2833 #endif
2834 #ifdef CONFIG_STACKTRACE
2835         ONE("stack",      S_IRUGO, proc_pid_stack),
2836 #endif
2837 #ifdef CONFIG_SCHEDSTATS
2838         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2839 #endif
2840 #ifdef CONFIG_LATENCYTOP
2841         REG("latency",  S_IRUGO, proc_lstats_operations),
2842 #endif
2843 #ifdef CONFIG_PROC_PID_CPUSET
2844         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2845 #endif
2846 #ifdef CONFIG_CGROUPS
2847         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2848 #endif
2849         INF("oom_score",  S_IRUGO, proc_oom_score),
2850         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2851         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2852 #ifdef CONFIG_AUDITSYSCALL
2853         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2854         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2855 #endif
2856 #ifdef CONFIG_FAULT_INJECTION
2857         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2858 #endif
2859 #ifdef CONFIG_ELF_CORE
2860         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2861 #endif
2862 #ifdef CONFIG_TASK_IO_ACCOUNTING
2863         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2864 #endif
2865 #ifdef CONFIG_HARDWALL
2866         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2867 #endif
2868 };
2869
2870 static int proc_tgid_base_readdir(struct file * filp,
2871                              void * dirent, filldir_t filldir)
2872 {
2873         return proc_pident_readdir(filp,dirent,filldir,
2874                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2875 }
2876
2877 static const struct file_operations proc_tgid_base_operations = {
2878         .read           = generic_read_dir,
2879         .readdir        = proc_tgid_base_readdir,
2880         .llseek         = default_llseek,
2881 };
2882
2883 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2884         return proc_pident_lookup(dir, dentry,
2885                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2886 }
2887
2888 static const struct inode_operations proc_tgid_base_inode_operations = {
2889         .lookup         = proc_tgid_base_lookup,
2890         .getattr        = pid_getattr,
2891         .setattr        = proc_setattr,
2892 };
2893
2894 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2895 {
2896         struct dentry *dentry, *leader, *dir;
2897         char buf[PROC_NUMBUF];
2898         struct qstr name;
2899
2900         name.name = buf;
2901         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2902         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2903         if (dentry) {
2904                 shrink_dcache_parent(dentry);
2905                 d_drop(dentry);
2906                 dput(dentry);
2907         }
2908
2909         name.name = buf;
2910         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2911         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2912         if (!leader)
2913                 goto out;
2914
2915         name.name = "task";
2916         name.len = strlen(name.name);
2917         dir = d_hash_and_lookup(leader, &name);
2918         if (!dir)
2919                 goto out_put_leader;
2920
2921         name.name = buf;
2922         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2923         dentry = d_hash_and_lookup(dir, &name);
2924         if (dentry) {
2925                 shrink_dcache_parent(dentry);
2926                 d_drop(dentry);
2927                 dput(dentry);
2928         }
2929
2930         dput(dir);
2931 out_put_leader:
2932         dput(leader);
2933 out:
2934         return;
2935 }
2936
2937 /**
2938  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2939  * @task: task that should be flushed.
2940  *
2941  * When flushing dentries from proc, one needs to flush them from global
2942  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2943  * in. This call is supposed to do all of this job.
2944  *
2945  * Looks in the dcache for
2946  * /proc/@pid
2947  * /proc/@tgid/task/@pid
2948  * if either directory is present flushes it and all of it'ts children
2949  * from the dcache.
2950  *
2951  * It is safe and reasonable to cache /proc entries for a task until
2952  * that task exits.  After that they just clog up the dcache with
2953  * useless entries, possibly causing useful dcache entries to be
2954  * flushed instead.  This routine is proved to flush those useless
2955  * dcache entries at process exit time.
2956  *
2957  * NOTE: This routine is just an optimization so it does not guarantee
2958  *       that no dcache entries will exist at process exit time it
2959  *       just makes it very unlikely that any will persist.
2960  */
2961
2962 void proc_flush_task(struct task_struct *task)
2963 {
2964         int i;
2965         struct pid *pid, *tgid;
2966         struct upid *upid;
2967
2968         pid = task_pid(task);
2969         tgid = task_tgid(task);
2970
2971         for (i = 0; i <= pid->level; i++) {
2972                 upid = &pid->numbers[i];
2973                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2974                                         tgid->numbers[i].nr);
2975         }
2976
2977         upid = &pid->numbers[pid->level];
2978         if (upid->nr == 1)
2979                 pid_ns_release_proc(upid->ns);
2980 }
2981
2982 static struct dentry *proc_pid_instantiate(struct inode *dir,
2983                                            struct dentry * dentry,
2984                                            struct task_struct *task, const void *ptr)
2985 {
2986         struct dentry *error = ERR_PTR(-ENOENT);
2987         struct inode *inode;
2988
2989         inode = proc_pid_make_inode(dir->i_sb, task);
2990         if (!inode)
2991                 goto out;
2992
2993         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2994         inode->i_op = &proc_tgid_base_inode_operations;
2995         inode->i_fop = &proc_tgid_base_operations;
2996         inode->i_flags|=S_IMMUTABLE;
2997
2998         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2999                 ARRAY_SIZE(tgid_base_stuff));
3000
3001         d_set_d_op(dentry, &pid_dentry_operations);
3002
3003         d_add(dentry, inode);
3004         /* Close the race of the process dying before we return the dentry */
3005         if (pid_revalidate(dentry, NULL))
3006                 error = NULL;
3007 out:
3008         return error;
3009 }
3010
3011 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3012 {
3013         struct dentry *result;
3014         struct task_struct *task;
3015         unsigned tgid;
3016         struct pid_namespace *ns;
3017
3018         result = proc_base_lookup(dir, dentry);
3019         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
3020                 goto out;
3021
3022         tgid = name_to_int(dentry);
3023         if (tgid == ~0U)
3024                 goto out;
3025
3026         ns = dentry->d_sb->s_fs_info;
3027         rcu_read_lock();
3028         task = find_task_by_pid_ns(tgid, ns);
3029         if (task)
3030                 get_task_struct(task);
3031         rcu_read_unlock();
3032         if (!task)
3033                 goto out;
3034
3035         result = proc_pid_instantiate(dir, dentry, task, NULL);
3036         put_task_struct(task);
3037 out:
3038         return result;
3039 }
3040
3041 /*
3042  * Find the first task with tgid >= tgid
3043  *
3044  */
3045 struct tgid_iter {
3046         unsigned int tgid;
3047         struct task_struct *task;
3048 };
3049 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
3050 {
3051         struct pid *pid;
3052
3053         if (iter.task)
3054                 put_task_struct(iter.task);
3055         rcu_read_lock();
3056 retry:
3057         iter.task = NULL;
3058         pid = find_ge_pid(iter.tgid, ns);
3059         if (pid) {
3060                 iter.tgid = pid_nr_ns(pid, ns);
3061                 iter.task = pid_task(pid, PIDTYPE_PID);
3062                 /* What we to know is if the pid we have find is the
3063                  * pid of a thread_group_leader.  Testing for task
3064                  * being a thread_group_leader is the obvious thing
3065                  * todo but there is a window when it fails, due to
3066                  * the pid transfer logic in de_thread.
3067                  *
3068                  * So we perform the straight forward test of seeing
3069                  * if the pid we have found is the pid of a thread
3070                  * group leader, and don't worry if the task we have
3071                  * found doesn't happen to be a thread group leader.
3072                  * As we don't care in the case of readdir.
3073                  */
3074                 if (!iter.task || !has_group_leader_pid(iter.task)) {
3075                         iter.tgid += 1;
3076                         goto retry;
3077                 }
3078                 get_task_struct(iter.task);
3079         }
3080         rcu_read_unlock();
3081         return iter;
3082 }
3083
3084 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
3085
3086 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3087         struct tgid_iter iter)
3088 {
3089         char name[PROC_NUMBUF];
3090         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
3091         return proc_fill_cache(filp, dirent, filldir, name, len,
3092                                 proc_pid_instantiate, iter.task, NULL);
3093 }
3094
3095 /* for the /proc/ directory itself, after non-process stuff has been done */
3096 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3097 {
3098         unsigned int nr;
3099         struct task_struct *reaper;
3100         struct tgid_iter iter;
3101         struct pid_namespace *ns;
3102
3103         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3104                 goto out_no_task;
3105         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3106
3107         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3108         if (!reaper)
3109                 goto out_no_task;
3110
3111         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3112                 const struct pid_entry *p = &proc_base_stuff[nr];
3113                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3114                         goto out;
3115         }
3116
3117         ns = filp->f_dentry->d_sb->s_fs_info;
3118         iter.task = NULL;
3119         iter.tgid = filp->f_pos - TGID_OFFSET;
3120         for (iter = next_tgid(ns, iter);
3121              iter.task;
3122              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3123                 filp->f_pos = iter.tgid + TGID_OFFSET;
3124                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3125                         put_task_struct(iter.task);
3126                         goto out;
3127                 }
3128         }
3129         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3130 out:
3131         put_task_struct(reaper);
3132 out_no_task:
3133         return 0;
3134 }
3135
3136 /*
3137  * Tasks
3138  */
3139 static const struct pid_entry tid_base_stuff[] = {
3140         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3141         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3142         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3143         REG("environ",   S_IRUSR, proc_environ_operations),
3144         INF("auxv",      S_IRUSR, proc_pid_auxv),
3145         ONE("status",    S_IRUGO, proc_pid_status),
3146         ONE("personality", S_IRUGO, proc_pid_personality),
3147         INF("limits",    S_IRUGO, proc_pid_limits),
3148 #ifdef CONFIG_SCHED_DEBUG
3149         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3150 #endif
3151         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3152 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3153         INF("syscall",   S_IRUGO, proc_pid_syscall),
3154 #endif
3155         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3156         ONE("stat",      S_IRUGO, proc_tid_stat),
3157         ONE("statm",     S_IRUGO, proc_pid_statm),
3158         REG("maps",      S_IRUGO, proc_maps_operations),
3159 #ifdef CONFIG_NUMA
3160         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3161 #endif
3162         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3163         LNK("cwd",       proc_cwd_link),
3164         LNK("root",      proc_root_link),
3165         LNK("exe",       proc_exe_link),
3166         REG("mounts",    S_IRUGO, proc_mounts_operations),
3167         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3168 #ifdef CONFIG_PROC_PAGE_MONITOR
3169         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3170         REG("smaps",     S_IRUGO, proc_smaps_operations),
3171         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3172 #endif
3173 #ifdef CONFIG_SECURITY
3174         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3175 #endif
3176 #ifdef CONFIG_KALLSYMS
3177         INF("wchan",     S_IRUGO, proc_pid_wchan),
3178 #endif
3179 #ifdef CONFIG_STACKTRACE
3180         ONE("stack",      S_IRUGO, proc_pid_stack),
3181 #endif
3182 #ifdef CONFIG_SCHEDSTATS
3183         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3184 #endif
3185 #ifdef CONFIG_LATENCYTOP
3186         REG("latency",  S_IRUGO, proc_lstats_operations),
3187 #endif
3188 #ifdef CONFIG_PROC_PID_CPUSET
3189         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3190 #endif
3191 #ifdef CONFIG_CGROUPS
3192         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3193 #endif
3194         INF("oom_score", S_IRUGO, proc_oom_score),
3195         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3196         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3197 #ifdef CONFIG_AUDITSYSCALL
3198         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3199         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3200 #endif
3201 #ifdef CONFIG_FAULT_INJECTION
3202         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3203 #endif
3204 #ifdef CONFIG_TASK_IO_ACCOUNTING
3205         INF("io",       S_IRUSR, proc_tid_io_accounting),
3206 #endif
3207 #ifdef CONFIG_HARDWALL
3208         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3209 #endif
3210 };
3211
3212 static int proc_tid_base_readdir(struct file * filp,
3213                              void * dirent, filldir_t filldir)
3214 {
3215         return proc_pident_readdir(filp,dirent,filldir,
3216                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3217 }
3218
3219 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3220         return proc_pident_lookup(dir, dentry,
3221                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3222 }
3223
3224 static const struct file_operations proc_tid_base_operations = {
3225         .read           = generic_read_dir,
3226         .readdir        = proc_tid_base_readdir,
3227         .llseek         = default_llseek,
3228 };
3229
3230 static const struct inode_operations proc_tid_base_inode_operations = {
3231         .lookup         = proc_tid_base_lookup,
3232         .getattr        = pid_getattr,
3233         .setattr        = proc_setattr,
3234 };
3235
3236 static struct dentry *proc_task_instantiate(struct inode *dir,
3237         struct dentry *dentry, struct task_struct *task, const void *ptr)
3238 {
3239         struct dentry *error = ERR_PTR(-ENOENT);
3240         struct inode *inode;
3241         inode = proc_pid_make_inode(dir->i_sb, task);
3242
3243         if (!inode)
3244                 goto out;
3245         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3246         inode->i_op = &proc_tid_base_inode_operations;
3247         inode->i_fop = &proc_tid_base_operations;
3248         inode->i_flags|=S_IMMUTABLE;
3249
3250         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
3251                 ARRAY_SIZE(tid_base_stuff));
3252
3253         d_set_d_op(dentry, &pid_dentry_operations);
3254
3255         d_add(dentry, inode);
3256         /* Close the race of the process dying before we return the dentry */
3257         if (pid_revalidate(dentry, NULL))
3258                 error = NULL;
3259 out:
3260         return error;
3261 }
3262
3263 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3264 {
3265         struct dentry *result = ERR_PTR(-ENOENT);
3266         struct task_struct *task;
3267         struct task_struct *leader = get_proc_task(dir);
3268         unsigned tid;
3269         struct pid_namespace *ns;
3270
3271         if (!leader)
3272                 goto out_no_task;
3273
3274         tid = name_to_int(dentry);
3275         if (tid == ~0U)
3276                 goto out;
3277
3278         ns = dentry->d_sb->s_fs_info;
3279         rcu_read_lock();
3280         task = find_task_by_pid_ns(tid, ns);
3281         if (task)
3282                 get_task_struct(task);
3283         rcu_read_unlock();
3284         if (!task)
3285                 goto out;
3286         if (!same_thread_group(leader, task))
3287                 goto out_drop_task;
3288
3289         result = proc_task_instantiate(dir, dentry, task, NULL);
3290 out_drop_task:
3291         put_task_struct(task);
3292 out:
3293         put_task_struct(leader);
3294 out_no_task:
3295         return result;
3296 }
3297
3298 /*
3299  * Find the first tid of a thread group to return to user space.
3300  *
3301  * Usually this is just the thread group leader, but if the users
3302  * buffer was too small or there was a seek into the middle of the
3303  * directory we have more work todo.
3304  *
3305  * In the case of a short read we start with find_task_by_pid.
3306  *
3307  * In the case of a seek we start with the leader and walk nr
3308  * threads past it.
3309  */
3310 static struct task_struct *first_tid(struct task_struct *leader,
3311                 int tid, int nr, struct pid_namespace *ns)
3312 {
3313         struct task_struct *pos;
3314
3315         rcu_read_lock();
3316         /* Attempt to start with the pid of a thread */
3317         if (tid && (nr > 0)) {
3318                 pos = find_task_by_pid_ns(tid, ns);
3319                 if (pos && (pos->group_leader == leader))
3320                         goto found;
3321         }
3322
3323         /* If nr exceeds the number of threads there is nothing todo */
3324         pos = NULL;
3325         if (nr && nr >= get_nr_threads(leader))
3326                 goto out;
3327
3328         /* If we haven't found our starting place yet start
3329          * with the leader and walk nr threads forward.
3330          */
3331         for (pos = leader; nr > 0; --nr) {
3332                 pos = next_thread(pos);
3333                 if (pos == leader) {
3334                         pos = NULL;
3335                         goto out;
3336                 }
3337         }
3338 found:
3339         get_task_struct(pos);
3340 out:
3341         rcu_read_unlock();
3342         return pos;
3343 }
3344
3345 /*
3346  * Find the next thread in the thread list.
3347  * Return NULL if there is an error or no next thread.
3348  *
3349  * The reference to the input task_struct is released.
3350  */
3351 static struct task_struct *next_tid(struct task_struct *start)
3352 {
3353         struct task_struct *pos = NULL;
3354         rcu_read_lock();
3355         if (pid_alive(start)) {
3356                 pos = next_thread(start);
3357                 if (thread_group_leader(pos))
3358                         pos = NULL;
3359                 else
3360                         get_task_struct(pos);
3361         }
3362         rcu_read_unlock();
3363         put_task_struct(start);
3364         return pos;
3365 }
3366
3367 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3368         struct task_struct *task, int tid)
3369 {
3370         char name[PROC_NUMBUF];
3371         int len = snprintf(name, sizeof(name), "%d", tid);
3372         return proc_fill_cache(filp, dirent, filldir, name, len,
3373                                 proc_task_instantiate, task, NULL);
3374 }
3375
3376 /* for the /proc/TGID/task/ directories */
3377 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3378 {
3379         struct dentry *dentry = filp->f_path.dentry;
3380         struct inode *inode = dentry->d_inode;
3381         struct task_struct *leader = NULL;
3382         struct task_struct *task;
3383         int retval = -ENOENT;
3384         ino_t ino;
3385         int tid;
3386         struct pid_namespace *ns;
3387
3388         task = get_proc_task(inode);
3389         if (!task)
3390                 goto out_no_task;
3391         rcu_read_lock();
3392         if (pid_alive(task)) {
3393                 leader = task->group_leader;
3394                 get_task_struct(leader);
3395         }
3396         rcu_read_unlock();
3397         put_task_struct(task);
3398         if (!leader)
3399                 goto out_no_task;
3400         retval = 0;
3401
3402         switch ((unsigned long)filp->f_pos) {
3403         case 0:
3404                 ino = inode->i_ino;
3405                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3406                         goto out;
3407                 filp->f_pos++;
3408                 /* fall through */
3409         case 1:
3410                 ino = parent_ino(dentry);
3411                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3412                         goto out;
3413                 filp->f_pos++;
3414                 /* fall through */
3415         }
3416
3417         /* f_version caches the tgid value that the last readdir call couldn't
3418          * return. lseek aka telldir automagically resets f_version to 0.
3419          */
3420         ns = filp->f_dentry->d_sb->s_fs_info;
3421         tid = (int)filp->f_version;
3422         filp->f_version = 0;
3423         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3424              task;
3425              task = next_tid(task), filp->f_pos++) {
3426                 tid = task_pid_nr_ns(task, ns);
3427                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3428                         /* returning this tgid failed, save it as the first
3429                          * pid for the next readir call */
3430                         filp->f_version = (u64)tid;
3431                         put_task_struct(task);
3432                         break;
3433                 }
3434         }
3435 out:
3436         put_task_struct(leader);
3437 out_no_task:
3438         return retval;
3439 }
3440
3441 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3442 {
3443         struct inode *inode = dentry->d_inode;
3444         struct task_struct *p = get_proc_task(inode);
3445         generic_fillattr(inode, stat);
3446
3447         if (p) {
3448                 stat->nlink += get_nr_threads(p);
3449                 put_task_struct(p);
3450         }
3451
3452         return 0;
3453 }
3454
3455 static const struct inode_operations proc_task_inode_operations = {
3456         .lookup         = proc_task_lookup,
3457         .getattr        = proc_task_getattr,
3458         .setattr        = proc_setattr,
3459 };
3460
3461 static const struct file_operations proc_task_operations = {
3462         .read           = generic_read_dir,
3463         .readdir        = proc_task_readdir,
3464         .llseek         = default_llseek,
3465 };