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/rcupdate.h>
  67 #include <linux/kallsyms.h>
  68 #include <linux/stacktrace.h>
  69 #include <linux/resource.h>
  70 #include <linux/module.h>
  71 #include <linux/mount.h>
  72 #include <linux/security.h>
  73 #include <linux/ptrace.h>
  74 #include <linux/tracehook.h>
  75 #include <linux/cgroup.h>
  76 #include <linux/cpuset.h>
  77 #include <linux/audit.h>
  78 #include <linux/poll.h>
  79 #include <linux/nsproxy.h>
  80 #include <linux/oom.h>
  81 #include <linux/elf.h>
  82 #include <linux/pid_namespace.h>
  83 #include <linux/fs_struct.h>
  84 #include "internal.h"
  85
  86 /* NOTE:
  87  *      Implementing inode permission operations in /proc is almost
  88  *      certainly an error.  Permission checks need to happen during
  89  *      each system call not at open time.  The reason is that most of
  90  *      what we wish to check for permissions in /proc varies at runtime.
  91  *
  92  *      The classic example of a problem is opening file descriptors
  93  *      in /proc for a task before it execs a suid executable.
  94  */
  95
  96 struct pid_entry {
  97         char *name;
  98         int len;
  99         mode_t mode;
 100         const struct inode_operations *iop;
 101         const struct file_operations *fop;
 102         union proc_op op;
 103 };
 104
 105 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 106         .name = (NAME),                                 \
 107         .len  = sizeof(NAME) - 1,                       \
 108         .mode = MODE,                                   \
 109         .iop  = IOP,                                    \
 110         .fop  = FOP,                                    \
 111         .op   = OP,                                     \
 112 }
 113
 114 #define DIR(NAME, MODE, iops, fops)     \
 115         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 116 #define LNK(NAME, get_link)                                     \
 117         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 118                 &proc_pid_link_inode_operations, NULL,          \
 119                 { .proc_get_link = get_link } )
 120 #define REG(NAME, MODE, fops)                           \
 121         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 122 #define INF(NAME, MODE, read)                           \
 123         NOD(NAME, (S_IFREG|(MODE)),                     \
 124                 NULL, &proc_info_file_operations,       \
 125                 { .proc_read = read } )
 126 #define ONE(NAME, MODE, show)                           \
 127         NOD(NAME, (S_IFREG|(MODE)),                     \
 128                 NULL, &proc_single_file_operations,     \
 129                 { .proc_show = show } )
 130
 131 /*
 132  * Count the number of hardlinks for the pid_entry table, excluding the .
 133  * and .. links.
 134  */
 135 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 136         unsigned int n)
 137 {
 138         unsigned int i;
 139         unsigned int count;
 140
 141         count = 0;
 142         for (i = 0; i < n; ++i) {
 143                 if (S_ISDIR(entries[i].mode))
 144                         ++count;
 145         }
 146
 147         return count;
 148 }
 149
 150 static int get_fs_path(struct task_struct *task, struct path *path, bool root)
 151 {
 152         struct fs_struct *fs;
 153         int result = -ENOENT;
 154
 155         task_lock(task);
 156         fs = task->fs;
 157         if (fs) {
 158                 read_lock(&fs->lock);
 159                 *path = root ? fs->root : fs->pwd;
 160                 path_get(path);
 161                 read_unlock(&fs->lock);
 162                 result = 0;
 163         }
 164         task_unlock(task);
 165         return result;
 166 }
 167
 168 static int get_nr_threads(struct task_struct *tsk)
 169 {
 170         unsigned long flags;
 171         int count = 0;
 172
 173         if (lock_task_sighand(tsk, &flags)) {
 174                 count = atomic_read(&tsk->signal->count);
 175                 unlock_task_sighand(tsk, &flags);
 176         }
 177         return count;
 178 }
 179
 180 static int proc_cwd_link(struct inode *inode, struct path *path)
 181 {
 182         struct task_struct *task = get_proc_task(inode);
 183         int result = -ENOENT;
 184
 185         if (task) {
 186                 result = get_fs_path(task, path, 0);
 187                 put_task_struct(task);
 188         }
 189         return result;
 190 }
 191
 192 static int proc_root_link(struct inode *inode, struct path *path)
 193 {
 194         struct task_struct *task = get_proc_task(inode);
 195         int result = -ENOENT;
 196
 197         if (task) {
 198                 result = get_fs_path(task, path, 1);
 199                 put_task_struct(task);
 200         }
 201         return result;
 202 }
 203
 204 /*
 205  * Return zero if current may access user memory in @task, -error if not.
 206  */
 207 static int check_mem_permission(struct task_struct *task)
 208 {
 209         /*
 210          * A task can always look at itself, in case it chooses
 211          * to use system calls instead of load instructions.
 212          */
 213         if (task == current)
 214                 return 0;
 215
 216         /*
 217          * If current is actively ptrace'ing, and would also be
 218          * permitted to freshly attach with ptrace now, permit it.
 219          */
 220         if (task_is_stopped_or_traced(task)) {
 221                 int match;
 222                 rcu_read_lock();
 223                 match = (tracehook_tracer_task(task) == current);
 224                 rcu_read_unlock();
 225                 if (match && ptrace_may_access(task, PTRACE_MODE_ATTACH))
 226                         return 0;
 227         }
 228
 229         /*
 230          * Noone else is allowed.
 231          */
 232         return -EPERM;
 233 }
 234
 235 struct mm_struct *mm_for_maps(struct task_struct *task)
 236 {
 237         struct mm_struct *mm;
 238
 239         if (mutex_lock_killable(&task->cred_guard_mutex))
 240                 return NULL;
 241
 242         mm = get_task_mm(task);
 243         if (mm && mm != current->mm &&
 244                         !ptrace_may_access(task, PTRACE_MODE_READ)) {
 245                 mmput(mm);
 246                 mm = NULL;
 247         }
 248         mutex_unlock(&task->cred_guard_mutex);
 249
 250         return mm;
 251 }
 252
 253 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 254 {
 255         int res = 0;
 256         unsigned int len;
 257         struct mm_struct *mm = get_task_mm(task);
 258         if (!mm)
 259                 goto out;
 260         if (!mm->arg_end)
 261                 goto out_mm;    /* Shh! No looking before we're done */
 262
 263         len = mm->arg_end - mm->arg_start;
 264
 265         if (len > PAGE_SIZE)
 266                 len = PAGE_SIZE;
 267
 268         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 269
 270         // If the nul at the end of args has been overwritten, then
 271         // assume application is using setproctitle(3).
 272         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 273                 len = strnlen(buffer, res);
 274                 if (len < res) {
 275                     res = len;
 276                 } else {
 277                         len = mm->env_end - mm->env_start;
 278                         if (len > PAGE_SIZE - res)
 279                                 len = PAGE_SIZE - res;
 280                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 281                         res = strnlen(buffer, res);
 282                 }
 283         }
 284 out_mm:
 285         mmput(mm);
 286 out:
 287         return res;
 288 }
 289
 290 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 291 {
 292         int res = 0;
 293         struct mm_struct *mm = get_task_mm(task);
 294         if (mm) {
 295                 unsigned int nwords = 0;
 296                 do {
 297                         nwords += 2;
 298                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 299                 res = nwords * sizeof(mm->saved_auxv[0]);
 300                 if (res > PAGE_SIZE)
 301                         res = PAGE_SIZE;
 302                 memcpy(buffer, mm->saved_auxv, res);
 303                 mmput(mm);
 304         }
 305         return res;
 306 }
 307
 308
 309 #ifdef CONFIG_KALLSYMS
 310 /*
 311  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 312  * Returns the resolved symbol.  If that fails, simply return the address.
 313  */
 314 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 315 {
 316         unsigned long wchan;
 317         char symname[KSYM_NAME_LEN];
 318
 319         wchan = get_wchan(task);
 320
 321         if (lookup_symbol_name(wchan, symname) < 0)
 322                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 323                         return 0;
 324                 else
 325                         return sprintf(buffer, "%lu", wchan);
 326         else
 327                 return sprintf(buffer, "%s", symname);
 328 }
 329 #endif /* CONFIG_KALLSYMS */
 330
 331 #ifdef CONFIG_STACKTRACE
 332
 333 #define MAX_STACK_TRACE_DEPTH   64
 334
 335 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 336                           struct pid *pid, struct task_struct *task)
 337 {
 338         struct stack_trace trace;
 339         unsigned long *entries;
 340         int i;
 341
 342         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 343         if (!entries)
 344                 return -ENOMEM;
 345
 346         trace.nr_entries        = 0;
 347         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 348         trace.entries           = entries;
 349         trace.skip              = 0;
 350         save_stack_trace_tsk(task, &trace);
 351
 352         for (i = 0; i < trace.nr_entries; i++) {
 353                 seq_printf(m, "[<%p>] %pS\n",
 354                            (void *)entries[i], (void *)entries[i]);
 355         }
 356         kfree(entries);
 357
 358         return 0;
 359 }
 360 #endif
 361
 362 #ifdef CONFIG_SCHEDSTATS
 363 /*
 364  * Provides /proc/PID/schedstat
 365  */
 366 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 367 {
 368         return sprintf(buffer, "%llu %llu %lu\n",
 369                         (unsigned long long)task->se.sum_exec_runtime,
 370                         (unsigned long long)task->sched_info.run_delay,
 371                         task->sched_info.pcount);
 372 }
 373 #endif
 374
 375 #ifdef CONFIG_LATENCYTOP
 376 static int lstats_show_proc(struct seq_file *m, void *v)
 377 {
 378         int i;
 379         struct inode *inode = m->private;
 380         struct task_struct *task = get_proc_task(inode);
 381
 382         if (!task)
 383                 return -ESRCH;
 384         seq_puts(m, "Latency Top version : v0.1\n");
 385         for (i = 0; i < 32; i++) {
 386                 if (task->latency_record[i].backtrace[0]) {
 387                         int q;
 388                         seq_printf(m, "%i %li %li ",
 389                                 task->latency_record[i].count,
 390                                 task->latency_record[i].time,
 391                                 task->latency_record[i].max);
 392                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 393                                 char sym[KSYM_SYMBOL_LEN];
 394                                 char *c;
 395                                 if (!task->latency_record[i].backtrace[q])
 396                                         break;
 397                                 if (task->latency_record[i].backtrace[q] == ULONG_MAX)
 398                                         break;
 399                                 sprint_symbol(sym, task->latency_record[i].backtrace[q]);
 400                                 c = strchr(sym, '+');
 401                                 if (c)
 402                                         *c = 0;
 403                                 seq_printf(m, "%s ", sym);
 404                         }
 405                         seq_printf(m, "\n");
 406                 }
 407
 408         }
 409         put_task_struct(task);
 410         return 0;
 411 }
 412
 413 static int lstats_open(struct inode *inode, struct file *file)
 414 {
 415         return single_open(file, lstats_show_proc, inode);
 416 }
 417
 418 static ssize_t lstats_write(struct file *file, const char __user *buf,
 419                             size_t count, loff_t *offs)
 420 {
 421         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 422
 423         if (!task)
 424                 return -ESRCH;
 425         clear_all_latency_tracing(task);
 426         put_task_struct(task);
 427
 428         return count;
 429 }
 430
 431 static const struct file_operations proc_lstats_operations = {
 432         .open           = lstats_open,
 433         .read           = seq_read,
 434         .write          = lstats_write,
 435         .llseek         = seq_lseek,
 436         .release        = single_release,
 437 };
 438
 439 #endif
 440
 441 /* The badness from the OOM killer */
 442 unsigned long badness(struct task_struct *p, unsigned long uptime);
 443 static int proc_oom_score(struct task_struct *task, char *buffer)
 444 {
 445         unsigned long points;
 446         struct timespec uptime;
 447
 448         do_posix_clock_monotonic_gettime(&uptime);
 449         read_lock(&tasklist_lock);
 450         points = badness(task, uptime.tv_sec);
 451         read_unlock(&tasklist_lock);
 452         return sprintf(buffer, "%lu\n", points);
 453 }
 454
 455 struct limit_names {
 456         char *name;
 457         char *unit;
 458 };
 459
 460 static const struct limit_names lnames[RLIM_NLIMITS] = {
 461         [RLIMIT_CPU] = {"Max cpu time", "ms"},
 462         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 463         [RLIMIT_DATA] = {"Max data size", "bytes"},
 464         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 465         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 466         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 467         [RLIMIT_NPROC] = {"Max processes", "processes"},
 468         [RLIMIT_NOFILE] = {"Max open files", "files"},
 469         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 470         [RLIMIT_AS] = {"Max address space", "bytes"},
 471         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 472         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 473         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 474         [RLIMIT_NICE] = {"Max nice priority", NULL},
 475         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 476         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 477 };
 478
 479 /* Display limits for a process */
 480 static int proc_pid_limits(struct task_struct *task, char *buffer)
 481 {
 482         unsigned int i;
 483         int count = 0;
 484         unsigned long flags;
 485         char *bufptr = buffer;
 486
 487         struct rlimit rlim[RLIM_NLIMITS];
 488
 489         if (!lock_task_sighand(task, &flags))
 490                 return 0;
 491         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 492         unlock_task_sighand(task, &flags);
 493
 494         /*
 495          * print the file header
 496          */
 497         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 498                         "Limit", "Soft Limit", "Hard Limit", "Units");
 499
 500         for (i = 0; i < RLIM_NLIMITS; i++) {
 501                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 502                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 503                                          lnames[i].name, "unlimited");
 504                 else
 505                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 506                                          lnames[i].name, rlim[i].rlim_cur);
 507
 508                 if (rlim[i].rlim_max == RLIM_INFINITY)
 509                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 510                 else
 511                         count += sprintf(&bufptr[count], "%-20lu ",
 512                                          rlim[i].rlim_max);
 513
 514                 if (lnames[i].unit)
 515                         count += sprintf(&bufptr[count], "%-10s\n",
 516                                          lnames[i].unit);
 517                 else
 518                         count += sprintf(&bufptr[count], "\n");
 519         }
 520
 521         return count;
 522 }
 523
 524 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 525 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 526 {
 527         long nr;
 528         unsigned long args[6], sp, pc;
 529
 530         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 531                 return sprintf(buffer, "running\n");
 532
 533         if (nr < 0)
 534                 return sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 535
 536         return sprintf(buffer,
 537                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 538                        nr,
 539                        args[0], args[1], args[2], args[3], args[4], args[5],
 540                        sp, pc);
 541 }
 542 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 543
 544 /************************************************************************/
 545 /*                       Here the fs part begins                        */
 546 /************************************************************************/
 547
 548 /* permission checks */
 549 static int proc_fd_access_allowed(struct inode *inode)
 550 {
 551         struct task_struct *task;
 552         int allowed = 0;
 553         /* Allow access to a task's file descriptors if it is us or we
 554          * may use ptrace attach to the process and find out that
 555          * information.
 556          */
 557         task = get_proc_task(inode);
 558         if (task) {
 559                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 560                 put_task_struct(task);
 561         }
 562         return allowed;
 563 }
 564
 565 static int proc_setattr(struct dentry *dentry, struct iattr *attr)
 566 {
 567         int error;
 568         struct inode *inode = dentry->d_inode;
 569
 570         if (attr->ia_valid & ATTR_MODE)
 571                 return -EPERM;
 572
 573         error = inode_change_ok(inode, attr);
 574         if (!error)
 575                 error = inode_setattr(inode, attr);
 576         return error;
 577 }
 578
 579 static const struct inode_operations proc_def_inode_operations = {
 580         .setattr        = proc_setattr,
 581 };
 582
 583 static int mounts_open_common(struct inode *inode, struct file *file,
 584                               const struct seq_operations *op)
 585 {
 586         struct task_struct *task = get_proc_task(inode);
 587         struct nsproxy *nsp;
 588         struct mnt_namespace *ns = NULL;
 589         struct path root;
 590         struct proc_mounts *p;
 591         int ret = -EINVAL;
 592
 593         if (task) {
 594                 rcu_read_lock();
 595                 nsp = task_nsproxy(task);
 596                 if (nsp) {
 597                         ns = nsp->mnt_ns;
 598                         if (ns)
 599                                 get_mnt_ns(ns);
 600                 }
 601                 rcu_read_unlock();
 602                 if (ns && get_fs_path(task, &root, 1) == 0)
 603                         ret = 0;
 604                 put_task_struct(task);
 605         }
 606
 607         if (!ns)
 608                 goto err;
 609         if (ret)
 610                 goto err_put_ns;
 611
 612         ret = -ENOMEM;
 613         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 614         if (!p)
 615                 goto err_put_path;
 616
 617         file->private_data = &p->m;
 618         ret = seq_open(file, op);
 619         if (ret)
 620                 goto err_free;
 621
 622         p->m.private = p;
 623         p->ns = ns;
 624         p->root = root;
 625         p->event = ns->event;
 626
 627         return 0;
 628
 629  err_free:
 630         kfree(p);
 631  err_put_path:
 632         path_put(&root);
 633  err_put_ns:
 634         put_mnt_ns(ns);
 635  err:
 636         return ret;
 637 }
 638
 639 static int mounts_release(struct inode *inode, struct file *file)
 640 {
 641         struct proc_mounts *p = file->private_data;
 642         path_put(&p->root);
 643         put_mnt_ns(p->ns);
 644         return seq_release(inode, file);
 645 }
 646
 647 static unsigned mounts_poll(struct file *file, poll_table *wait)
 648 {
 649         struct proc_mounts *p = file->private_data;
 650         struct mnt_namespace *ns = p->ns;
 651         unsigned res = POLLIN | POLLRDNORM;
 652
 653         poll_wait(file, &ns->poll, wait);
 654
 655         spin_lock(&vfsmount_lock);
 656         if (p->event != ns->event) {
 657                 p->event = ns->event;
 658                 res |= POLLERR | POLLPRI;
 659         }
 660         spin_unlock(&vfsmount_lock);
 661
 662         return res;
 663 }
 664
 665 static int mounts_open(struct inode *inode, struct file *file)
 666 {
 667         return mounts_open_common(inode, file, &mounts_op);
 668 }
 669
 670 static const struct file_operations proc_mounts_operations = {
 671         .open           = mounts_open,
 672         .read           = seq_read,
 673         .llseek         = seq_lseek,
 674         .release        = mounts_release,
 675         .poll           = mounts_poll,
 676 };
 677
 678 static int mountinfo_open(struct inode *inode, struct file *file)
 679 {
 680         return mounts_open_common(inode, file, &mountinfo_op);
 681 }
 682
 683 static const struct file_operations proc_mountinfo_operations = {
 684         .open           = mountinfo_open,
 685         .read           = seq_read,
 686         .llseek         = seq_lseek,
 687         .release        = mounts_release,
 688         .poll           = mounts_poll,
 689 };
 690
 691 static int mountstats_open(struct inode *inode, struct file *file)
 692 {
 693         return mounts_open_common(inode, file, &mountstats_op);
 694 }
 695
 696 static const struct file_operations proc_mountstats_operations = {
 697         .open           = mountstats_open,
 698         .read           = seq_read,
 699         .llseek         = seq_lseek,
 700         .release        = mounts_release,
 701 };
 702
 703 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 704
 705 static ssize_t proc_info_read(struct file * file, char __user * buf,
 706                           size_t count, loff_t *ppos)
 707 {
 708         struct inode * inode = file->f_path.dentry->d_inode;
 709         unsigned long page;
 710         ssize_t length;
 711         struct task_struct *task = get_proc_task(inode);
 712
 713         length = -ESRCH;
 714         if (!task)
 715                 goto out_no_task;
 716
 717         if (count > PROC_BLOCK_SIZE)
 718                 count = PROC_BLOCK_SIZE;
 719
 720         length = -ENOMEM;
 721         if (!(page = __get_free_page(GFP_TEMPORARY)))
 722                 goto out;
 723
 724         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 725
 726         if (length >= 0)
 727                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 728         free_page(page);
 729 out:
 730         put_task_struct(task);
 731 out_no_task:
 732         return length;
 733 }
 734
 735 static const struct file_operations proc_info_file_operations = {
 736         .read           = proc_info_read,
 737 };
 738
 739 static int proc_single_show(struct seq_file *m, void *v)
 740 {
 741         struct inode *inode = m->private;
 742         struct pid_namespace *ns;
 743         struct pid *pid;
 744         struct task_struct *task;
 745         int ret;
 746
 747         ns = inode->i_sb->s_fs_info;
 748         pid = proc_pid(inode);
 749         task = get_pid_task(pid, PIDTYPE_PID);
 750         if (!task)
 751                 return -ESRCH;
 752
 753         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 754
 755         put_task_struct(task);
 756         return ret;
 757 }
 758
 759 static int proc_single_open(struct inode *inode, struct file *filp)
 760 {
 761         int ret;
 762         ret = single_open(filp, proc_single_show, NULL);
 763         if (!ret) {
 764                 struct seq_file *m = filp->private_data;
 765
 766                 m->private = inode;
 767         }
 768         return ret;
 769 }
 770
 771 static const struct file_operations proc_single_file_operations = {
 772         .open           = proc_single_open,
 773         .read           = seq_read,
 774         .llseek         = seq_lseek,
 775         .release        = single_release,
 776 };
 777
 778 static int mem_open(struct inode* inode, struct file* file)
 779 {
 780         file->private_data = (void*)((long)current->self_exec_id);
 781         return 0;
 782 }
 783
 784 static ssize_t mem_read(struct file * file, char __user * buf,
 785                         size_t count, loff_t *ppos)
 786 {
 787         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 788         char *page;
 789         unsigned long src = *ppos;
 790         int ret = -ESRCH;
 791         struct mm_struct *mm;
 792
 793         if (!task)
 794                 goto out_no_task;
 795
 796         if (check_mem_permission(task))
 797                 goto out;
 798
 799         ret = -ENOMEM;
 800         page = (char *)__get_free_page(GFP_TEMPORARY);
 801         if (!page)
 802                 goto out;
 803
 804         ret = 0;
 805
 806         mm = get_task_mm(task);
 807         if (!mm)
 808                 goto out_free;
 809
 810         ret = -EIO;
 811
 812         if (file->private_data != (void*)((long)current->self_exec_id))
 813                 goto out_put;
 814
 815         ret = 0;
 816
 817         while (count > 0) {
 818                 int this_len, retval;
 819
 820                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 821                 retval = access_process_vm(task, src, page, this_len, 0);
 822                 if (!retval || check_mem_permission(task)) {
 823                         if (!ret)
 824                                 ret = -EIO;
 825                         break;
 826                 }
 827
 828                 if (copy_to_user(buf, page, retval)) {
 829                         ret = -EFAULT;
 830                         break;
 831                 }
 832
 833                 ret += retval;
 834                 src += retval;
 835                 buf += retval;
 836                 count -= retval;
 837         }
 838         *ppos = src;
 839
 840 out_put:
 841         mmput(mm);
 842 out_free:
 843         free_page((unsigned long) page);
 844 out:
 845         put_task_struct(task);
 846 out_no_task:
 847         return ret;
 848 }
 849
 850 #define mem_write NULL
 851
 852 #ifndef mem_write
 853 /* This is a security hazard */
 854 static ssize_t mem_write(struct file * file, const char __user *buf,
 855                          size_t count, loff_t *ppos)
 856 {
 857         int copied;
 858         char *page;
 859         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 860         unsigned long dst = *ppos;
 861
 862         copied = -ESRCH;
 863         if (!task)
 864                 goto out_no_task;
 865
 866         if (check_mem_permission(task))
 867                 goto out;
 868
 869         copied = -ENOMEM;
 870         page = (char *)__get_free_page(GFP_TEMPORARY);
 871         if (!page)
 872                 goto out;
 873
 874         copied = 0;
 875         while (count > 0) {
 876                 int this_len, retval;
 877
 878                 this_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 879                 if (copy_from_user(page, buf, this_len)) {
 880                         copied = -EFAULT;
 881                         break;
 882                 }
 883                 retval = access_process_vm(task, dst, page, this_len, 1);
 884                 if (!retval) {
 885                         if (!copied)
 886                                 copied = -EIO;
 887                         break;
 888                 }
 889                 copied += retval;
 890                 buf += retval;
 891                 dst += retval;
 892                 count -= retval;
 893         }
 894         *ppos = dst;
 895         free_page((unsigned long) page);
 896 out:
 897         put_task_struct(task);
 898 out_no_task:
 899         return copied;
 900 }
 901 #endif
 902
 903 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 904 {
 905         switch (orig) {
 906         case 0:
 907                 file->f_pos = offset;
 908                 break;
 909         case 1:
 910                 file->f_pos += offset;
 911                 break;
 912         default:
 913                 return -EINVAL;
 914         }
 915         force_successful_syscall_return();
 916         return file->f_pos;
 917 }
 918
 919 static const struct file_operations proc_mem_operations = {
 920         .llseek         = mem_lseek,
 921         .read           = mem_read,
 922         .write          = mem_write,
 923         .open           = mem_open,
 924 };
 925
 926 static ssize_t environ_read(struct file *file, char __user *buf,
 927                         size_t count, loff_t *ppos)
 928 {
 929         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 930         char *page;
 931         unsigned long src = *ppos;
 932         int ret = -ESRCH;
 933         struct mm_struct *mm;
 934
 935         if (!task)
 936                 goto out_no_task;
 937
 938         if (!ptrace_may_access(task, PTRACE_MODE_READ))
 939                 goto out;
 940
 941         ret = -ENOMEM;
 942         page = (char *)__get_free_page(GFP_TEMPORARY);
 943         if (!page)
 944                 goto out;
 945
 946         ret = 0;
 947
 948         mm = get_task_mm(task);
 949         if (!mm)
 950                 goto out_free;
 951
 952         while (count > 0) {
 953                 int this_len, retval, max_len;
 954
 955                 this_len = mm->env_end - (mm->env_start + src);
 956
 957                 if (this_len <= 0)
 958                         break;
 959
 960                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 961                 this_len = (this_len > max_len) ? max_len : this_len;
 962
 963                 retval = access_process_vm(task, (mm->env_start + src),
 964                         page, this_len, 0);
 965
 966                 if (retval <= 0) {
 967                         ret = retval;
 968                         break;
 969                 }
 970
 971                 if (copy_to_user(buf, page, retval)) {
 972                         ret = -EFAULT;
 973                         break;
 974                 }
 975
 976                 ret += retval;
 977                 src += retval;
 978                 buf += retval;
 979                 count -= retval;
 980         }
 981         *ppos = src;
 982
 983         mmput(mm);
 984 out_free:
 985         free_page((unsigned long) page);
 986 out:
 987         put_task_struct(task);
 988 out_no_task:
 989         return ret;
 990 }
 991
 992 static const struct file_operations proc_environ_operations = {
 993         .read           = environ_read,
 994 };
 995
 996 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 997                                 size_t count, loff_t *ppos)
 998 {
 999         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1000         char buffer[PROC_NUMBUF];
1001         size_t len;
1002         int oom_adjust;
1003
1004         if (!task)
1005                 return -ESRCH;
1006         task_lock(task);
1007         if (task->mm)
1008                 oom_adjust = task->mm->oom_adj;
1009         else
1010                 oom_adjust = OOM_DISABLE;
1011         task_unlock(task);
1012         put_task_struct(task);
1013
1014         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
1015
1016         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1017 }
1018
1019 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
1020                                 size_t count, loff_t *ppos)
1021 {
1022         struct task_struct *task;
1023         char buffer[PROC_NUMBUF], *end;
1024         int oom_adjust;
1025
1026         memset(buffer, 0, sizeof(buffer));
1027         if (count > sizeof(buffer) - 1)
1028                 count = sizeof(buffer) - 1;
1029         if (copy_from_user(buffer, buf, count))
1030                 return -EFAULT;
1031         oom_adjust = simple_strtol(buffer, &end, 0);
1032         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1033              oom_adjust != OOM_DISABLE)
1034                 return -EINVAL;
1035         if (*end == '\n')
1036                 end++;
1037         task = get_proc_task(file->f_path.dentry->d_inode);
1038         if (!task)
1039                 return -ESRCH;
1040         task_lock(task);
1041         if (!task->mm) {
1042                 task_unlock(task);
1043                 put_task_struct(task);
1044                 return -EINVAL;
1045         }
1046         if (oom_adjust < task->mm->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1047                 task_unlock(task);
1048                 put_task_struct(task);
1049                 return -EACCES;
1050         }
1051         task->mm->oom_adj = oom_adjust;
1052         task_unlock(task);
1053         put_task_struct(task);
1054         if (end - buffer == 0)
1055                 return -EIO;
1056         return end - buffer;
1057 }
1058
1059 static const struct file_operations proc_oom_adjust_operations = {
1060         .read           = oom_adjust_read,
1061         .write          = oom_adjust_write,
1062 };
1063
1064 #ifdef CONFIG_AUDITSYSCALL
1065 #define TMPBUFLEN 21
1066 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1067                                   size_t count, loff_t *ppos)
1068 {
1069         struct inode * inode = file->f_path.dentry->d_inode;
1070         struct task_struct *task = get_proc_task(inode);
1071         ssize_t length;
1072         char tmpbuf[TMPBUFLEN];
1073
1074         if (!task)
1075                 return -ESRCH;
1076         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1077                                 audit_get_loginuid(task));
1078         put_task_struct(task);
1079         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1080 }
1081
1082 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1083                                    size_t count, loff_t *ppos)
1084 {
1085         struct inode * inode = file->f_path.dentry->d_inode;
1086         char *page, *tmp;
1087         ssize_t length;
1088         uid_t loginuid;
1089
1090         if (!capable(CAP_AUDIT_CONTROL))
1091                 return -EPERM;
1092
1093         if (current != pid_task(proc_pid(inode), PIDTYPE_PID))
1094                 return -EPERM;
1095
1096         if (count >= PAGE_SIZE)
1097                 count = PAGE_SIZE - 1;
1098
1099         if (*ppos != 0) {
1100                 /* No partial writes. */
1101                 return -EINVAL;
1102         }
1103         page = (char*)__get_free_page(GFP_TEMPORARY);
1104         if (!page)
1105                 return -ENOMEM;
1106         length = -EFAULT;
1107         if (copy_from_user(page, buf, count))
1108                 goto out_free_page;
1109
1110         page[count] = '\0';
1111         loginuid = simple_strtoul(page, &tmp, 10);
1112         if (tmp == page) {
1113                 length = -EINVAL;
1114                 goto out_free_page;
1115
1116         }
1117         length = audit_set_loginuid(current, loginuid);
1118         if (likely(length == 0))
1119                 length = count;
1120
1121 out_free_page:
1122         free_page((unsigned long) page);
1123         return length;
1124 }
1125
1126 static const struct file_operations proc_loginuid_operations = {
1127         .read           = proc_loginuid_read,
1128         .write          = proc_loginuid_write,
1129 };
1130
1131 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1132                                   size_t count, loff_t *ppos)
1133 {
1134         struct inode * inode = file->f_path.dentry->d_inode;
1135         struct task_struct *task = get_proc_task(inode);
1136         ssize_t length;
1137         char tmpbuf[TMPBUFLEN];
1138
1139         if (!task)
1140                 return -ESRCH;
1141         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1142                                 audit_get_sessionid(task));
1143         put_task_struct(task);
1144         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1145 }
1146
1147 static const struct file_operations proc_sessionid_operations = {
1148         .read           = proc_sessionid_read,
1149 };
1150 #endif
1151
1152 #ifdef CONFIG_FAULT_INJECTION
1153 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1154                                       size_t count, loff_t *ppos)
1155 {
1156         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1157         char buffer[PROC_NUMBUF];
1158         size_t len;
1159         int make_it_fail;
1160
1161         if (!task)
1162                 return -ESRCH;
1163         make_it_fail = task->make_it_fail;
1164         put_task_struct(task);
1165
1166         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1167
1168         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1169 }
1170
1171 static ssize_t proc_fault_inject_write(struct file * file,
1172                         const char __user * buf, size_t count, loff_t *ppos)
1173 {
1174         struct task_struct *task;
1175         char buffer[PROC_NUMBUF], *end;
1176         int make_it_fail;
1177
1178         if (!capable(CAP_SYS_RESOURCE))
1179                 return -EPERM;
1180         memset(buffer, 0, sizeof(buffer));
1181         if (count > sizeof(buffer) - 1)
1182                 count = sizeof(buffer) - 1;
1183         if (copy_from_user(buffer, buf, count))
1184                 return -EFAULT;
1185         make_it_fail = simple_strtol(buffer, &end, 0);
1186         if (*end == '\n')
1187                 end++;
1188         task = get_proc_task(file->f_dentry->d_inode);
1189         if (!task)
1190                 return -ESRCH;
1191         task->make_it_fail = make_it_fail;
1192         put_task_struct(task);
1193         if (end - buffer == 0)
1194                 return -EIO;
1195         return end - buffer;
1196 }
1197
1198 static const struct file_operations proc_fault_inject_operations = {
1199         .read           = proc_fault_inject_read,
1200         .write          = proc_fault_inject_write,
1201 };
1202 #endif
1203
1204
1205 #ifdef CONFIG_SCHED_DEBUG
1206 /*
1207  * Print out various scheduling related per-task fields:
1208  */
1209 static int sched_show(struct seq_file *m, void *v)
1210 {
1211         struct inode *inode = m->private;
1212         struct task_struct *p;
1213
1214         p = get_proc_task(inode);
1215         if (!p)
1216                 return -ESRCH;
1217         proc_sched_show_task(p, m);
1218
1219         put_task_struct(p);
1220
1221         return 0;
1222 }
1223
1224 static ssize_t
1225 sched_write(struct file *file, const char __user *buf,
1226             size_t count, loff_t *offset)
1227 {
1228         struct inode *inode = file->f_path.dentry->d_inode;
1229         struct task_struct *p;
1230
1231         p = get_proc_task(inode);
1232         if (!p)
1233                 return -ESRCH;
1234         proc_sched_set_task(p);
1235
1236         put_task_struct(p);
1237
1238         return count;
1239 }
1240
1241 static int sched_open(struct inode *inode, struct file *filp)
1242 {
1243         int ret;
1244
1245         ret = single_open(filp, sched_show, NULL);
1246         if (!ret) {
1247                 struct seq_file *m = filp->private_data;
1248
1249                 m->private = inode;
1250         }
1251         return ret;
1252 }
1253
1254 static const struct file_operations proc_pid_sched_operations = {
1255         .open           = sched_open,
1256         .read           = seq_read,
1257         .write          = sched_write,
1258         .llseek         = seq_lseek,
1259         .release        = single_release,
1260 };
1261
1262 #endif
1263
1264 /*
1265  * We added or removed a vma mapping the executable. The vmas are only mapped
1266  * during exec and are not mapped with the mmap system call.
1267  * Callers must hold down_write() on the mm's mmap_sem for these
1268  */
1269 void added_exe_file_vma(struct mm_struct *mm)
1270 {
1271         mm->num_exe_file_vmas++;
1272 }
1273
1274 void removed_exe_file_vma(struct mm_struct *mm)
1275 {
1276         mm->num_exe_file_vmas--;
1277         if ((mm->num_exe_file_vmas == 0) && mm->exe_file){
1278                 fput(mm->exe_file);
1279                 mm->exe_file = NULL;
1280         }
1281
1282 }
1283
1284 void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
1285 {
1286         if (new_exe_file)
1287                 get_file(new_exe_file);
1288         if (mm->exe_file)
1289                 fput(mm->exe_file);
1290         mm->exe_file = new_exe_file;
1291         mm->num_exe_file_vmas = 0;
1292 }
1293
1294 struct file *get_mm_exe_file(struct mm_struct *mm)
1295 {
1296         struct file *exe_file;
1297
1298         /* We need mmap_sem to protect against races with removal of
1299          * VM_EXECUTABLE vmas */
1300         down_read(&mm->mmap_sem);
1301         exe_file = mm->exe_file;
1302         if (exe_file)
1303                 get_file(exe_file);
1304         up_read(&mm->mmap_sem);
1305         return exe_file;
1306 }
1307
1308 void dup_mm_exe_file(struct mm_struct *oldmm, struct mm_struct *newmm)
1309 {
1310         /* It's safe to write the exe_file pointer without exe_file_lock because
1311          * this is called during fork when the task is not yet in /proc */
1312         newmm->exe_file = get_mm_exe_file(oldmm);
1313 }
1314
1315 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1316 {
1317         struct task_struct *task;
1318         struct mm_struct *mm;
1319         struct file *exe_file;
1320
1321         task = get_proc_task(inode);
1322         if (!task)
1323                 return -ENOENT;
1324         mm = get_task_mm(task);
1325         put_task_struct(task);
1326         if (!mm)
1327                 return -ENOENT;
1328         exe_file = get_mm_exe_file(mm);
1329         mmput(mm);
1330         if (exe_file) {
1331                 *exe_path = exe_file->f_path;
1332                 path_get(&exe_file->f_path);
1333                 fput(exe_file);
1334                 return 0;
1335         } else
1336                 return -ENOENT;
1337 }
1338
1339 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1340 {
1341         struct inode *inode = dentry->d_inode;
1342         int error = -EACCES;
1343
1344         /* We don't need a base pointer in the /proc filesystem */
1345         path_put(&nd->path);
1346
1347         /* Are we allowed to snoop on the tasks file descriptors? */
1348         if (!proc_fd_access_allowed(inode))
1349                 goto out;
1350
1351         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1352         nd->last_type = LAST_BIND;
1353 out:
1354         return ERR_PTR(error);
1355 }
1356
1357 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1358 {
1359         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1360         char *pathname;
1361         int len;
1362
1363         if (!tmp)
1364                 return -ENOMEM;
1365
1366         pathname = d_path(path, tmp, PAGE_SIZE);
1367         len = PTR_ERR(pathname);
1368         if (IS_ERR(pathname))
1369                 goto out;
1370         len = tmp + PAGE_SIZE - 1 - pathname;
1371
1372         if (len > buflen)
1373                 len = buflen;
1374         if (copy_to_user(buffer, pathname, len))
1375                 len = -EFAULT;
1376  out:
1377         free_page((unsigned long)tmp);
1378         return len;
1379 }
1380
1381 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1382 {
1383         int error = -EACCES;
1384         struct inode *inode = dentry->d_inode;
1385         struct path path;
1386
1387         /* Are we allowed to snoop on the tasks file descriptors? */
1388         if (!proc_fd_access_allowed(inode))
1389                 goto out;
1390
1391         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1392         if (error)
1393                 goto out;
1394
1395         error = do_proc_readlink(&path, buffer, buflen);
1396         path_put(&path);
1397 out:
1398         return error;
1399 }
1400
1401 static const struct inode_operations proc_pid_link_inode_operations = {
1402         .readlink       = proc_pid_readlink,
1403         .follow_link    = proc_pid_follow_link,
1404         .setattr        = proc_setattr,
1405 };
1406
1407
1408 /* building an inode */
1409
1410 static int task_dumpable(struct task_struct *task)
1411 {
1412         int dumpable = 0;
1413         struct mm_struct *mm;
1414
1415         task_lock(task);
1416         mm = task->mm;
1417         if (mm)
1418                 dumpable = get_dumpable(mm);
1419         task_unlock(task);
1420         if(dumpable == 1)
1421                 return 1;
1422         return 0;
1423 }
1424
1425
1426 static struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1427 {
1428         struct inode * inode;
1429         struct proc_inode *ei;
1430         const struct cred *cred;
1431
1432         /* We need a new inode */
1433
1434         inode = new_inode(sb);
1435         if (!inode)
1436                 goto out;
1437
1438         /* Common stuff */
1439         ei = PROC_I(inode);
1440         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1441         inode->i_op = &proc_def_inode_operations;
1442
1443         /*
1444          * grab the reference to task.
1445          */
1446         ei->pid = get_task_pid(task, PIDTYPE_PID);
1447         if (!ei->pid)
1448                 goto out_unlock;
1449
1450         if (task_dumpable(task)) {
1451                 rcu_read_lock();
1452                 cred = __task_cred(task);
1453                 inode->i_uid = cred->euid;
1454                 inode->i_gid = cred->egid;
1455                 rcu_read_unlock();
1456         }
1457         security_task_to_inode(task, inode);
1458
1459 out:
1460         return inode;
1461
1462 out_unlock:
1463         iput(inode);
1464         return NULL;
1465 }
1466
1467 static int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1468 {
1469         struct inode *inode = dentry->d_inode;
1470         struct task_struct *task;
1471         const struct cred *cred;
1472
1473         generic_fillattr(inode, stat);
1474
1475         rcu_read_lock();
1476         stat->uid = 0;
1477         stat->gid = 0;
1478         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1479         if (task) {
1480                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1481                     task_dumpable(task)) {
1482                         cred = __task_cred(task);
1483                         stat->uid = cred->euid;
1484                         stat->gid = cred->egid;
1485                 }
1486         }
1487         rcu_read_unlock();
1488         return 0;
1489 }
1490
1491 /* dentry stuff */
1492
1493 /*
1494  *      Exceptional case: normally we are not allowed to unhash a busy
1495  * directory. In this case, however, we can do it - no aliasing problems
1496  * due to the way we treat inodes.
1497  *
1498  * Rewrite the inode's ownerships here because the owning task may have
1499  * performed a setuid(), etc.
1500  *
1501  * Before the /proc/pid/status file was created the only way to read
1502  * the effective uid of a /process was to stat /proc/pid.  Reading
1503  * /proc/pid/status is slow enough that procps and other packages
1504  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1505  * made this apply to all per process world readable and executable
1506  * directories.
1507  */
1508 static int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1509 {
1510         struct inode *inode = dentry->d_inode;
1511         struct task_struct *task = get_proc_task(inode);
1512         const struct cred *cred;
1513
1514         if (task) {
1515                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1516                     task_dumpable(task)) {
1517                         rcu_read_lock();
1518                         cred = __task_cred(task);
1519                         inode->i_uid = cred->euid;
1520                         inode->i_gid = cred->egid;
1521                         rcu_read_unlock();
1522                 } else {
1523                         inode->i_uid = 0;
1524                         inode->i_gid = 0;
1525                 }
1526                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1527                 security_task_to_inode(task, inode);
1528                 put_task_struct(task);
1529                 return 1;
1530         }
1531         d_drop(dentry);
1532         return 0;
1533 }
1534
1535 static int pid_delete_dentry(struct dentry * dentry)
1536 {
1537         /* Is the task we represent dead?
1538          * If so, then don't put the dentry on the lru list,
1539          * kill it immediately.
1540          */
1541         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1542 }
1543
1544 static const struct dentry_operations pid_dentry_operations =
1545 {
1546         .d_revalidate   = pid_revalidate,
1547         .d_delete       = pid_delete_dentry,
1548 };
1549
1550 /* Lookups */
1551
1552 typedef struct dentry *instantiate_t(struct inode *, struct dentry *,
1553                                 struct task_struct *, const void *);
1554
1555 /*
1556  * Fill a directory entry.
1557  *
1558  * If possible create the dcache entry and derive our inode number and
1559  * file type from dcache entry.
1560  *
1561  * Since all of the proc inode numbers are dynamically generated, the inode
1562  * numbers do not exist until the inode is cache.  This means creating the
1563  * the dcache entry in readdir is necessary to keep the inode numbers
1564  * reported by readdir in sync with the inode numbers reported
1565  * by stat.
1566  */
1567 static int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1568         char *name, int len,
1569         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1570 {
1571         struct dentry *child, *dir = filp->f_path.dentry;
1572         struct inode *inode;
1573         struct qstr qname;
1574         ino_t ino = 0;
1575         unsigned type = DT_UNKNOWN;
1576
1577         qname.name = name;
1578         qname.len  = len;
1579         qname.hash = full_name_hash(name, len);
1580
1581         child = d_lookup(dir, &qname);
1582         if (!child) {
1583                 struct dentry *new;
1584                 new = d_alloc(dir, &qname);
1585                 if (new) {
1586                         child = instantiate(dir->d_inode, new, task, ptr);
1587                         if (child)
1588                                 dput(new);
1589                         else
1590                                 child = new;
1591                 }
1592         }
1593         if (!child || IS_ERR(child) || !child->d_inode)
1594                 goto end_instantiate;
1595         inode = child->d_inode;
1596         if (inode) {
1597                 ino = inode->i_ino;
1598                 type = inode->i_mode >> 12;
1599         }
1600         dput(child);
1601 end_instantiate:
1602         if (!ino)
1603                 ino = find_inode_number(dir, &qname);
1604         if (!ino)
1605                 ino = 1;
1606         return filldir(dirent, name, len, filp->f_pos, ino, type);
1607 }
1608
1609 static unsigned name_to_int(struct dentry *dentry)
1610 {
1611         const char *name = dentry->d_name.name;
1612         int len = dentry->d_name.len;
1613         unsigned n = 0;
1614
1615         if (len > 1 && *name == '0')
1616                 goto out;
1617         while (len-- > 0) {
1618                 unsigned c = *name++ - '0';
1619                 if (c > 9)
1620                         goto out;
1621                 if (n >= (~0U-9)/10)
1622                         goto out;
1623                 n *= 10;
1624                 n += c;
1625         }
1626         return n;
1627 out:
1628         return ~0U;
1629 }
1630
1631 #define PROC_FDINFO_MAX 64
1632
1633 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1634 {
1635         struct task_struct *task = get_proc_task(inode);
1636         struct files_struct *files = NULL;
1637         struct file *file;
1638         int fd = proc_fd(inode);
1639
1640         if (task) {
1641                 files = get_files_struct(task);
1642                 put_task_struct(task);
1643         }
1644         if (files) {
1645                 /*
1646                  * We are not taking a ref to the file structure, so we must
1647                  * hold ->file_lock.
1648                  */
1649                 spin_lock(&files->file_lock);
1650                 file = fcheck_files(files, fd);
1651                 if (file) {
1652                         if (path) {
1653                                 *path = file->f_path;
1654                                 path_get(&file->f_path);
1655                         }
1656                         if (info)
1657                                 snprintf(info, PROC_FDINFO_MAX,
1658                                          "pos:\t%lli\n"
1659                                          "flags:\t0%o\n",
1660                                          (long long) file->f_pos,
1661                                          file->f_flags);
1662                         spin_unlock(&files->file_lock);
1663                         put_files_struct(files);
1664                         return 0;
1665                 }
1666                 spin_unlock(&files->file_lock);
1667                 put_files_struct(files);
1668         }
1669         return -ENOENT;
1670 }
1671
1672 static int proc_fd_link(struct inode *inode, struct path *path)
1673 {
1674         return proc_fd_info(inode, path, NULL);
1675 }
1676
1677 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1678 {
1679         struct inode *inode = dentry->d_inode;
1680         struct task_struct *task = get_proc_task(inode);
1681         int fd = proc_fd(inode);
1682         struct files_struct *files;
1683         const struct cred *cred;
1684
1685         if (task) {
1686                 files = get_files_struct(task);
1687                 if (files) {
1688                         rcu_read_lock();
1689                         if (fcheck_files(files, fd)) {
1690                                 rcu_read_unlock();
1691                                 put_files_struct(files);
1692                                 if (task_dumpable(task)) {
1693                                         rcu_read_lock();
1694                                         cred = __task_cred(task);
1695                                         inode->i_uid = cred->euid;
1696                                         inode->i_gid = cred->egid;
1697                                         rcu_read_unlock();
1698                                 } else {
1699                                         inode->i_uid = 0;
1700                                         inode->i_gid = 0;
1701                                 }
1702                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1703                                 security_task_to_inode(task, inode);
1704                                 put_task_struct(task);
1705                                 return 1;
1706                         }
1707                         rcu_read_unlock();
1708                         put_files_struct(files);
1709                 }
1710                 put_task_struct(task);
1711         }
1712         d_drop(dentry);
1713         return 0;
1714 }
1715
1716 static const struct dentry_operations tid_fd_dentry_operations =
1717 {
1718         .d_revalidate   = tid_fd_revalidate,
1719         .d_delete       = pid_delete_dentry,
1720 };
1721
1722 static struct dentry *proc_fd_instantiate(struct inode *dir,
1723         struct dentry *dentry, struct task_struct *task, const void *ptr)
1724 {
1725         unsigned fd = *(const unsigned *)ptr;
1726         struct file *file;
1727         struct files_struct *files;
1728         struct inode *inode;
1729         struct proc_inode *ei;
1730         struct dentry *error = ERR_PTR(-ENOENT);
1731
1732         inode = proc_pid_make_inode(dir->i_sb, task);
1733         if (!inode)
1734                 goto out;
1735         ei = PROC_I(inode);
1736         ei->fd = fd;
1737         files = get_files_struct(task);
1738         if (!files)
1739                 goto out_iput;
1740         inode->i_mode = S_IFLNK;
1741
1742         /*
1743          * We are not taking a ref to the file structure, so we must
1744          * hold ->file_lock.
1745          */
1746         spin_lock(&files->file_lock);
1747         file = fcheck_files(files, fd);
1748         if (!file)
1749                 goto out_unlock;
1750         if (file->f_mode & FMODE_READ)
1751                 inode->i_mode |= S_IRUSR | S_IXUSR;
1752         if (file->f_mode & FMODE_WRITE)
1753                 inode->i_mode |= S_IWUSR | S_IXUSR;
1754         spin_unlock(&files->file_lock);
1755         put_files_struct(files);
1756
1757         inode->i_op = &proc_pid_link_inode_operations;
1758         inode->i_size = 64;
1759         ei->op.proc_get_link = proc_fd_link;
1760         dentry->d_op = &tid_fd_dentry_operations;
1761         d_add(dentry, inode);
1762         /* Close the race of the process dying before we return the dentry */
1763         if (tid_fd_revalidate(dentry, NULL))
1764                 error = NULL;
1765
1766  out:
1767         return error;
1768 out_unlock:
1769         spin_unlock(&files->file_lock);
1770         put_files_struct(files);
1771 out_iput:
1772         iput(inode);
1773         goto out;
1774 }
1775
1776 static struct dentry *proc_lookupfd_common(struct inode *dir,
1777                                            struct dentry *dentry,
1778                                            instantiate_t instantiate)
1779 {
1780         struct task_struct *task = get_proc_task(dir);
1781         unsigned fd = name_to_int(dentry);
1782         struct dentry *result = ERR_PTR(-ENOENT);
1783
1784         if (!task)
1785                 goto out_no_task;
1786         if (fd == ~0U)
1787                 goto out;
1788
1789         result = instantiate(dir, dentry, task, &fd);
1790 out:
1791         put_task_struct(task);
1792 out_no_task:
1793         return result;
1794 }
1795
1796 static int proc_readfd_common(struct file * filp, void * dirent,
1797                               filldir_t filldir, instantiate_t instantiate)
1798 {
1799         struct dentry *dentry = filp->f_path.dentry;
1800         struct inode *inode = dentry->d_inode;
1801         struct task_struct *p = get_proc_task(inode);
1802         unsigned int fd, ino;
1803         int retval;
1804         struct files_struct * files;
1805
1806         retval = -ENOENT;
1807         if (!p)
1808                 goto out_no_task;
1809         retval = 0;
1810
1811         fd = filp->f_pos;
1812         switch (fd) {
1813                 case 0:
1814                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
1815                                 goto out;
1816                         filp->f_pos++;
1817                 case 1:
1818                         ino = parent_ino(dentry);
1819                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
1820                                 goto out;
1821                         filp->f_pos++;
1822                 default:
1823                         files = get_files_struct(p);
1824                         if (!files)
1825                                 goto out;
1826                         rcu_read_lock();
1827                         for (fd = filp->f_pos-2;
1828                              fd < files_fdtable(files)->max_fds;
1829                              fd++, filp->f_pos++) {
1830                                 char name[PROC_NUMBUF];
1831                                 int len;
1832
1833                                 if (!fcheck_files(files, fd))
1834                                         continue;
1835                                 rcu_read_unlock();
1836
1837                                 len = snprintf(name, sizeof(name), "%d", fd);
1838                                 if (proc_fill_cache(filp, dirent, filldir,
1839                                                     name, len, instantiate,
1840                                                     p, &fd) < 0) {
1841                                         rcu_read_lock();
1842                                         break;
1843                                 }
1844                                 rcu_read_lock();
1845                         }
1846                         rcu_read_unlock();
1847                         put_files_struct(files);
1848         }
1849 out:
1850         put_task_struct(p);
1851 out_no_task:
1852         return retval;
1853 }
1854
1855 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
1856                                     struct nameidata *nd)
1857 {
1858         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
1859 }
1860
1861 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
1862 {
1863         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
1864 }
1865
1866 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
1867                                       size_t len, loff_t *ppos)
1868 {
1869         char tmp[PROC_FDINFO_MAX];
1870         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
1871         if (!err)
1872                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
1873         return err;
1874 }
1875
1876 static const struct file_operations proc_fdinfo_file_operations = {
1877         .open           = nonseekable_open,
1878         .read           = proc_fdinfo_read,
1879 };
1880
1881 static const struct file_operations proc_fd_operations = {
1882         .read           = generic_read_dir,
1883         .readdir        = proc_readfd,
1884 };
1885
1886 /*
1887  * /proc/pid/fd needs a special permission handler so that a process can still
1888  * access /proc/self/fd after it has executed a setuid().
1889  */
1890 static int proc_fd_permission(struct inode *inode, int mask)
1891 {
1892         int rv;
1893
1894         rv = generic_permission(inode, mask, NULL);
1895         if (rv == 0)
1896                 return 0;
1897         if (task_pid(current) == proc_pid(inode))
1898                 rv = 0;
1899         return rv;
1900 }
1901
1902 /*
1903  * proc directories can do almost nothing..
1904  */
1905 static const struct inode_operations proc_fd_inode_operations = {
1906         .lookup         = proc_lookupfd,
1907         .permission     = proc_fd_permission,
1908         .setattr        = proc_setattr,
1909 };
1910
1911 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
1912         struct dentry *dentry, struct task_struct *task, const void *ptr)
1913 {
1914         unsigned fd = *(unsigned *)ptr;
1915         struct inode *inode;
1916         struct proc_inode *ei;
1917         struct dentry *error = ERR_PTR(-ENOENT);
1918
1919         inode = proc_pid_make_inode(dir->i_sb, task);
1920         if (!inode)
1921                 goto out;
1922         ei = PROC_I(inode);
1923         ei->fd = fd;
1924         inode->i_mode = S_IFREG | S_IRUSR;
1925         inode->i_fop = &proc_fdinfo_file_operations;
1926         dentry->d_op = &tid_fd_dentry_operations;
1927         d_add(dentry, inode);
1928         /* Close the race of the process dying before we return the dentry */
1929         if (tid_fd_revalidate(dentry, NULL))
1930                 error = NULL;
1931
1932  out:
1933         return error;
1934 }
1935
1936 static struct dentry *proc_lookupfdinfo(struct inode *dir,
1937                                         struct dentry *dentry,
1938                                         struct nameidata *nd)
1939 {
1940         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
1941 }
1942
1943 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
1944 {
1945         return proc_readfd_common(filp, dirent, filldir,
1946                                   proc_fdinfo_instantiate);
1947 }
1948
1949 static const struct file_operations proc_fdinfo_operations = {
1950         .read           = generic_read_dir,
1951         .readdir        = proc_readfdinfo,
1952 };
1953
1954 /*
1955  * proc directories can do almost nothing..
1956  */
1957 static const struct inode_operations proc_fdinfo_inode_operations = {
1958         .lookup         = proc_lookupfdinfo,
1959         .setattr        = proc_setattr,
1960 };
1961
1962
1963 static struct dentry *proc_pident_instantiate(struct inode *dir,
1964         struct dentry *dentry, struct task_struct *task, const void *ptr)
1965 {
1966         const struct pid_entry *p = ptr;
1967         struct inode *inode;
1968         struct proc_inode *ei;
1969         struct dentry *error = ERR_PTR(-ENOENT);
1970
1971         inode = proc_pid_make_inode(dir->i_sb, task);
1972         if (!inode)
1973                 goto out;
1974
1975         ei = PROC_I(inode);
1976         inode->i_mode = p->mode;
1977         if (S_ISDIR(inode->i_mode))
1978                 inode->i_nlink = 2;     /* Use getattr to fix if necessary */
1979         if (p->iop)
1980                 inode->i_op = p->iop;
1981         if (p->fop)
1982                 inode->i_fop = p->fop;
1983         ei->op = p->op;
1984         dentry->d_op = &pid_dentry_operations;
1985         d_add(dentry, inode);
1986         /* Close the race of the process dying before we return the dentry */
1987         if (pid_revalidate(dentry, NULL))
1988                 error = NULL;
1989 out:
1990         return error;
1991 }
1992
1993 static struct dentry *proc_pident_lookup(struct inode *dir,
1994                                          struct dentry *dentry,
1995                                          const struct pid_entry *ents,
1996                                          unsigned int nents)
1997 {
1998         struct dentry *error;
1999         struct task_struct *task = get_proc_task(dir);
2000         const struct pid_entry *p, *last;
2001
2002         error = ERR_PTR(-ENOENT);
2003
2004         if (!task)
2005                 goto out_no_task;
2006
2007         /*
2008          * Yes, it does not scale. And it should not. Don't add
2009          * new entries into /proc/<tgid>/ without very good reasons.
2010          */
2011         last = &ents[nents - 1];
2012         for (p = ents; p <= last; p++) {
2013                 if (p->len != dentry->d_name.len)
2014                         continue;
2015                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2016                         break;
2017         }
2018         if (p > last)
2019                 goto out;
2020
2021         error = proc_pident_instantiate(dir, dentry, task, p);
2022 out:
2023         put_task_struct(task);
2024 out_no_task:
2025         return error;
2026 }
2027
2028 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2029         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2030 {
2031         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2032                                 proc_pident_instantiate, task, p);
2033 }
2034
2035 static int proc_pident_readdir(struct file *filp,
2036                 void *dirent, filldir_t filldir,
2037                 const struct pid_entry *ents, unsigned int nents)
2038 {
2039         int i;
2040         struct dentry *dentry = filp->f_path.dentry;
2041         struct inode *inode = dentry->d_inode;
2042         struct task_struct *task = get_proc_task(inode);
2043         const struct pid_entry *p, *last;
2044         ino_t ino;
2045         int ret;
2046
2047         ret = -ENOENT;
2048         if (!task)
2049                 goto out_no_task;
2050
2051         ret = 0;
2052         i = filp->f_pos;
2053         switch (i) {
2054         case 0:
2055                 ino = inode->i_ino;
2056                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2057                         goto out;
2058                 i++;
2059                 filp->f_pos++;
2060                 /* fall through */
2061         case 1:
2062                 ino = parent_ino(dentry);
2063                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2064                         goto out;
2065                 i++;
2066                 filp->f_pos++;
2067                 /* fall through */
2068         default:
2069                 i -= 2;
2070                 if (i >= nents) {
2071                         ret = 1;
2072                         goto out;
2073                 }
2074                 p = ents + i;
2075                 last = &ents[nents - 1];
2076                 while (p <= last) {
2077                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2078                                 goto out;
2079                         filp->f_pos++;
2080                         p++;
2081                 }
2082         }
2083
2084         ret = 1;
2085 out:
2086         put_task_struct(task);
2087 out_no_task:
2088         return ret;
2089 }
2090
2091 #ifdef CONFIG_SECURITY
2092 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2093                                   size_t count, loff_t *ppos)
2094 {
2095         struct inode * inode = file->f_path.dentry->d_inode;
2096         char *p = NULL;
2097         ssize_t length;
2098         struct task_struct *task = get_proc_task(inode);
2099
2100         if (!task)
2101                 return -ESRCH;
2102
2103         length = security_getprocattr(task,
2104                                       (char*)file->f_path.dentry->d_name.name,
2105                                       &p);
2106         put_task_struct(task);
2107         if (length > 0)
2108                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2109         kfree(p);
2110         return length;
2111 }
2112
2113 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2114                                    size_t count, loff_t *ppos)
2115 {
2116         struct inode * inode = file->f_path.dentry->d_inode;
2117         char *page;
2118         ssize_t length;
2119         struct task_struct *task = get_proc_task(inode);
2120
2121         length = -ESRCH;
2122         if (!task)
2123                 goto out_no_task;
2124         if (count > PAGE_SIZE)
2125                 count = PAGE_SIZE;
2126
2127         /* No partial writes. */
2128         length = -EINVAL;
2129         if (*ppos != 0)
2130                 goto out;
2131
2132         length = -ENOMEM;
2133         page = (char*)__get_free_page(GFP_TEMPORARY);
2134         if (!page)
2135                 goto out;
2136
2137         length = -EFAULT;
2138         if (copy_from_user(page, buf, count))
2139                 goto out_free;
2140
2141         /* Guard against adverse ptrace interaction */
2142         length = mutex_lock_interruptible(&task->cred_guard_mutex);
2143         if (length < 0)
2144                 goto out_free;
2145
2146         length = security_setprocattr(task,
2147                                       (char*)file->f_path.dentry->d_name.name,
2148                                       (void*)page, count);
2149         mutex_unlock(&task->cred_guard_mutex);
2150 out_free:
2151         free_page((unsigned long) page);
2152 out:
2153         put_task_struct(task);
2154 out_no_task:
2155         return length;
2156 }
2157
2158 static const struct file_operations proc_pid_attr_operations = {
2159         .read           = proc_pid_attr_read,
2160         .write          = proc_pid_attr_write,
2161 };
2162
2163 static const struct pid_entry attr_dir_stuff[] = {
2164         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2165         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2166         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2167         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2168         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2169         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2170 };
2171
2172 static int proc_attr_dir_readdir(struct file * filp,
2173                              void * dirent, filldir_t filldir)
2174 {
2175         return proc_pident_readdir(filp,dirent,filldir,
2176                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2177 }
2178
2179 static const struct file_operations proc_attr_dir_operations = {
2180         .read           = generic_read_dir,
2181         .readdir        = proc_attr_dir_readdir,
2182 };
2183
2184 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2185                                 struct dentry *dentry, struct nameidata *nd)
2186 {
2187         return proc_pident_lookup(dir, dentry,
2188                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2189 }
2190
2191 static const struct inode_operations proc_attr_dir_inode_operations = {
2192         .lookup         = proc_attr_dir_lookup,
2193         .getattr        = pid_getattr,
2194         .setattr        = proc_setattr,
2195 };
2196
2197 #endif
2198
2199 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2200 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2201                                          size_t count, loff_t *ppos)
2202 {
2203         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2204         struct mm_struct *mm;
2205         char buffer[PROC_NUMBUF];
2206         size_t len;
2207         int ret;
2208
2209         if (!task)
2210                 return -ESRCH;
2211
2212         ret = 0;
2213         mm = get_task_mm(task);
2214         if (mm) {
2215                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2216                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2217                                 MMF_DUMP_FILTER_SHIFT));
2218                 mmput(mm);
2219                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2220         }
2221
2222         put_task_struct(task);
2223
2224         return ret;
2225 }
2226
2227 static ssize_t proc_coredump_filter_write(struct file *file,
2228                                           const char __user *buf,
2229                                           size_t count,
2230                                           loff_t *ppos)
2231 {
2232         struct task_struct *task;
2233         struct mm_struct *mm;
2234         char buffer[PROC_NUMBUF], *end;
2235         unsigned int val;
2236         int ret;
2237         int i;
2238         unsigned long mask;
2239
2240         ret = -EFAULT;
2241         memset(buffer, 0, sizeof(buffer));
2242         if (count > sizeof(buffer) - 1)
2243                 count = sizeof(buffer) - 1;
2244         if (copy_from_user(buffer, buf, count))
2245                 goto out_no_task;
2246
2247         ret = -EINVAL;
2248         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2249         if (*end == '\n')
2250                 end++;
2251         if (end - buffer == 0)
2252                 goto out_no_task;
2253
2254         ret = -ESRCH;
2255         task = get_proc_task(file->f_dentry->d_inode);
2256         if (!task)
2257                 goto out_no_task;
2258
2259         ret = end - buffer;
2260         mm = get_task_mm(task);
2261         if (!mm)
2262                 goto out_no_mm;
2263
2264         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2265                 if (val & mask)
2266                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2267                 else
2268                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2269         }
2270
2271         mmput(mm);
2272  out_no_mm:
2273         put_task_struct(task);
2274  out_no_task:
2275         return ret;
2276 }
2277
2278 static const struct file_operations proc_coredump_filter_operations = {
2279         .read           = proc_coredump_filter_read,
2280         .write          = proc_coredump_filter_write,
2281 };
2282 #endif
2283
2284 /*
2285  * /proc/self:
2286  */
2287 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2288                               int buflen)
2289 {
2290         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2291         pid_t tgid = task_tgid_nr_ns(current, ns);
2292         char tmp[PROC_NUMBUF];
2293         if (!tgid)
2294                 return -ENOENT;
2295         sprintf(tmp, "%d", tgid);
2296         return vfs_readlink(dentry,buffer,buflen,tmp);
2297 }
2298
2299 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2300 {
2301         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2302         pid_t tgid = task_tgid_nr_ns(current, ns);
2303         char tmp[PROC_NUMBUF];
2304         if (!tgid)
2305                 return ERR_PTR(-ENOENT);
2306         sprintf(tmp, "%d", task_tgid_nr_ns(current, ns));
2307         return ERR_PTR(vfs_follow_link(nd,tmp));
2308 }
2309
2310 static const struct inode_operations proc_self_inode_operations = {
2311         .readlink       = proc_self_readlink,
2312         .follow_link    = proc_self_follow_link,
2313 };
2314
2315 /*
2316  * proc base
2317  *
2318  * These are the directory entries in the root directory of /proc
2319  * that properly belong to the /proc filesystem, as they describe
2320  * describe something that is process related.
2321  */
2322 static const struct pid_entry proc_base_stuff[] = {
2323         NOD("self", S_IFLNK|S_IRWXUGO,
2324                 &proc_self_inode_operations, NULL, {}),
2325 };
2326
2327 /*
2328  *      Exceptional case: normally we are not allowed to unhash a busy
2329  * directory. In this case, however, we can do it - no aliasing problems
2330  * due to the way we treat inodes.
2331  */
2332 static int proc_base_revalidate(struct dentry *dentry, struct nameidata *nd)
2333 {
2334         struct inode *inode = dentry->d_inode;
2335         struct task_struct *task = get_proc_task(inode);
2336         if (task) {
2337                 put_task_struct(task);
2338                 return 1;
2339         }
2340         d_drop(dentry);
2341         return 0;
2342 }
2343
2344 static const struct dentry_operations proc_base_dentry_operations =
2345 {
2346         .d_revalidate   = proc_base_revalidate,
2347         .d_delete       = pid_delete_dentry,
2348 };
2349
2350 static struct dentry *proc_base_instantiate(struct inode *dir,
2351         struct dentry *dentry, struct task_struct *task, const void *ptr)
2352 {
2353         const struct pid_entry *p = ptr;
2354         struct inode *inode;
2355         struct proc_inode *ei;
2356         struct dentry *error = ERR_PTR(-EINVAL);
2357
2358         /* Allocate the inode */
2359         error = ERR_PTR(-ENOMEM);
2360         inode = new_inode(dir->i_sb);
2361         if (!inode)
2362                 goto out;
2363
2364         /* Initialize the inode */
2365         ei = PROC_I(inode);
2366         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2367
2368         /*
2369          * grab the reference to the task.
2370          */
2371         ei->pid = get_task_pid(task, PIDTYPE_PID);
2372         if (!ei->pid)
2373                 goto out_iput;
2374
2375         inode->i_mode = p->mode;
2376         if (S_ISDIR(inode->i_mode))
2377                 inode->i_nlink = 2;
2378         if (S_ISLNK(inode->i_mode))
2379                 inode->i_size = 64;
2380         if (p->iop)
2381                 inode->i_op = p->iop;
2382         if (p->fop)
2383                 inode->i_fop = p->fop;
2384         ei->op = p->op;
2385         dentry->d_op = &proc_base_dentry_operations;
2386         d_add(dentry, inode);
2387         error = NULL;
2388 out:
2389         return error;
2390 out_iput:
2391         iput(inode);
2392         goto out;
2393 }
2394
2395 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2396 {
2397         struct dentry *error;
2398         struct task_struct *task = get_proc_task(dir);
2399         const struct pid_entry *p, *last;
2400
2401         error = ERR_PTR(-ENOENT);
2402
2403         if (!task)
2404                 goto out_no_task;
2405
2406         /* Lookup the directory entry */
2407         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2408         for (p = proc_base_stuff; p <= last; p++) {
2409                 if (p->len != dentry->d_name.len)
2410                         continue;
2411                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2412                         break;
2413         }
2414         if (p > last)
2415                 goto out;
2416
2417         error = proc_base_instantiate(dir, dentry, task, p);
2418
2419 out:
2420         put_task_struct(task);
2421 out_no_task:
2422         return error;
2423 }
2424
2425 static int proc_base_fill_cache(struct file *filp, void *dirent,
2426         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2427 {
2428         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2429                                 proc_base_instantiate, task, p);
2430 }
2431
2432 #ifdef CONFIG_TASK_IO_ACCOUNTING
2433 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2434 {
2435         struct task_io_accounting acct = task->ioac;
2436         unsigned long flags;
2437
2438         if (whole && lock_task_sighand(task, &flags)) {
2439                 struct task_struct *t = task;
2440
2441                 task_io_accounting_add(&acct, &task->signal->ioac);
2442                 while_each_thread(task, t)
2443                         task_io_accounting_add(&acct, &t->ioac);
2444
2445                 unlock_task_sighand(task, &flags);
2446         }
2447         return sprintf(buffer,
2448                         "rchar: %llu\n"
2449                         "wchar: %llu\n"
2450                         "syscr: %llu\n"
2451                         "syscw: %llu\n"
2452                         "read_bytes: %llu\n"
2453                         "write_bytes: %llu\n"
2454                         "cancelled_write_bytes: %llu\n",
2455                         (unsigned long long)acct.rchar,
2456                         (unsigned long long)acct.wchar,
2457                         (unsigned long long)acct.syscr,
2458                         (unsigned long long)acct.syscw,
2459                         (unsigned long long)acct.read_bytes,
2460                         (unsigned long long)acct.write_bytes,
2461                         (unsigned long long)acct.cancelled_write_bytes);
2462 }
2463
2464 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2465 {
2466         return do_io_accounting(task, buffer, 0);
2467 }
2468
2469 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2470 {
2471         return do_io_accounting(task, buffer, 1);
2472 }
2473 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2474
2475 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2476                                 struct pid *pid, struct task_struct *task)
2477 {
2478         seq_printf(m, "%08x\n", task->personality);
2479         return 0;
2480 }
2481
2482 /*
2483  * Thread groups
2484  */
2485 static const struct file_operations proc_task_operations;
2486 static const struct inode_operations proc_task_inode_operations;
2487
2488 static const struct pid_entry tgid_base_stuff[] = {
2489         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2490         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2491         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2492 #ifdef CONFIG_NET
2493         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2494 #endif
2495         REG("environ",    S_IRUSR, proc_environ_operations),
2496         INF("auxv",       S_IRUSR, proc_pid_auxv),
2497         ONE("status",     S_IRUGO, proc_pid_status),
2498         ONE("personality", S_IRUSR, proc_pid_personality),
2499         INF("limits",     S_IRUSR, proc_pid_limits),
2500 #ifdef CONFIG_SCHED_DEBUG
2501         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2502 #endif
2503 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2504         INF("syscall",    S_IRUSR, proc_pid_syscall),
2505 #endif
2506         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2507         ONE("stat",       S_IRUGO, proc_tgid_stat),
2508         ONE("statm",      S_IRUGO, proc_pid_statm),
2509         REG("maps",       S_IRUGO, proc_maps_operations),
2510 #ifdef CONFIG_NUMA
2511         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2512 #endif
2513         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2514         LNK("cwd",        proc_cwd_link),
2515         LNK("root",       proc_root_link),
2516         LNK("exe",        proc_exe_link),
2517         REG("mounts",     S_IRUGO, proc_mounts_operations),
2518         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2519         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2520 #ifdef CONFIG_PROC_PAGE_MONITOR
2521         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2522         REG("smaps",      S_IRUGO, proc_smaps_operations),
2523         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2524 #endif
2525 #ifdef CONFIG_SECURITY
2526         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2527 #endif
2528 #ifdef CONFIG_KALLSYMS
2529         INF("wchan",      S_IRUGO, proc_pid_wchan),
2530 #endif
2531 #ifdef CONFIG_STACKTRACE
2532         ONE("stack",      S_IRUSR, proc_pid_stack),
2533 #endif
2534 #ifdef CONFIG_SCHEDSTATS
2535         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2536 #endif
2537 #ifdef CONFIG_LATENCYTOP
2538         REG("latency",  S_IRUGO, proc_lstats_operations),
2539 #endif
2540 #ifdef CONFIG_PROC_PID_CPUSET
2541         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2542 #endif
2543 #ifdef CONFIG_CGROUPS
2544         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2545 #endif
2546         INF("oom_score",  S_IRUGO, proc_oom_score),
2547         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2548 #ifdef CONFIG_AUDITSYSCALL
2549         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2550         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2551 #endif
2552 #ifdef CONFIG_FAULT_INJECTION
2553         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2554 #endif
2555 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
2556         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2557 #endif
2558 #ifdef CONFIG_TASK_IO_ACCOUNTING
2559         INF("io",       S_IRUGO, proc_tgid_io_accounting),
2560 #endif
2561 };
2562
2563 static int proc_tgid_base_readdir(struct file * filp,
2564                              void * dirent, filldir_t filldir)
2565 {
2566         return proc_pident_readdir(filp,dirent,filldir,
2567                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2568 }
2569
2570 static const struct file_operations proc_tgid_base_operations = {
2571         .read           = generic_read_dir,
2572         .readdir        = proc_tgid_base_readdir,
2573 };
2574
2575 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2576         return proc_pident_lookup(dir, dentry,
2577                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2578 }
2579
2580 static const struct inode_operations proc_tgid_base_inode_operations = {
2581         .lookup         = proc_tgid_base_lookup,
2582         .getattr        = pid_getattr,
2583         .setattr        = proc_setattr,
2584 };
2585
2586 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2587 {
2588         struct dentry *dentry, *leader, *dir;
2589         char buf[PROC_NUMBUF];
2590         struct qstr name;
2591
2592         name.name = buf;
2593         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2594         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2595         if (dentry) {
2596                 if (!(current->flags & PF_EXITING))
2597                         shrink_dcache_parent(dentry);
2598                 d_drop(dentry);
2599                 dput(dentry);
2600         }
2601
2602         if (tgid == 0)
2603                 goto out;
2604
2605         name.name = buf;
2606         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2607         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2608         if (!leader)
2609                 goto out;
2610
2611         name.name = "task";
2612         name.len = strlen(name.name);
2613         dir = d_hash_and_lookup(leader, &name);
2614         if (!dir)
2615                 goto out_put_leader;
2616
2617         name.name = buf;
2618         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2619         dentry = d_hash_and_lookup(dir, &name);
2620         if (dentry) {
2621                 shrink_dcache_parent(dentry);
2622                 d_drop(dentry);
2623                 dput(dentry);
2624         }
2625
2626         dput(dir);
2627 out_put_leader:
2628         dput(leader);
2629 out:
2630         return;
2631 }
2632
2633 /**
2634  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2635  * @task: task that should be flushed.
2636  *
2637  * When flushing dentries from proc, one needs to flush them from global
2638  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2639  * in. This call is supposed to do all of this job.
2640  *
2641  * Looks in the dcache for
2642  * /proc/@pid
2643  * /proc/@tgid/task/@pid
2644  * if either directory is present flushes it and all of it'ts children
2645  * from the dcache.
2646  *
2647  * It is safe and reasonable to cache /proc entries for a task until
2648  * that task exits.  After that they just clog up the dcache with
2649  * useless entries, possibly causing useful dcache entries to be
2650  * flushed instead.  This routine is proved to flush those useless
2651  * dcache entries at process exit time.
2652  *
2653  * NOTE: This routine is just an optimization so it does not guarantee
2654  *       that no dcache entries will exist at process exit time it
2655  *       just makes it very unlikely that any will persist.
2656  */
2657
2658 void proc_flush_task(struct task_struct *task)
2659 {
2660         int i;
2661         struct pid *pid, *tgid = NULL;
2662         struct upid *upid;
2663
2664         pid = task_pid(task);
2665         if (thread_group_leader(task))
2666                 tgid = task_tgid(task);
2667
2668         for (i = 0; i <= pid->level; i++) {
2669                 upid = &pid->numbers[i];
2670                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2671                         tgid ? tgid->numbers[i].nr : 0);
2672         }
2673
2674         upid = &pid->numbers[pid->level];
2675         if (upid->nr == 1)
2676                 pid_ns_release_proc(upid->ns);
2677 }
2678
2679 static struct dentry *proc_pid_instantiate(struct inode *dir,
2680                                            struct dentry * dentry,
2681                                            struct task_struct *task, const void *ptr)
2682 {
2683         struct dentry *error = ERR_PTR(-ENOENT);
2684         struct inode *inode;
2685
2686         inode = proc_pid_make_inode(dir->i_sb, task);
2687         if (!inode)
2688                 goto out;
2689
2690         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2691         inode->i_op = &proc_tgid_base_inode_operations;
2692         inode->i_fop = &proc_tgid_base_operations;
2693         inode->i_flags|=S_IMMUTABLE;
2694
2695         inode->i_nlink = 2 + pid_entry_count_dirs(tgid_base_stuff,
2696                 ARRAY_SIZE(tgid_base_stuff));
2697
2698         dentry->d_op = &pid_dentry_operations;
2699
2700         d_add(dentry, inode);
2701         /* Close the race of the process dying before we return the dentry */
2702         if (pid_revalidate(dentry, NULL))
2703                 error = NULL;
2704 out:
2705         return error;
2706 }
2707
2708 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2709 {
2710         struct dentry *result = ERR_PTR(-ENOENT);
2711         struct task_struct *task;
2712         unsigned tgid;
2713         struct pid_namespace *ns;
2714
2715         result = proc_base_lookup(dir, dentry);
2716         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2717                 goto out;
2718
2719         tgid = name_to_int(dentry);
2720         if (tgid == ~0U)
2721                 goto out;
2722
2723         ns = dentry->d_sb->s_fs_info;
2724         rcu_read_lock();
2725         task = find_task_by_pid_ns(tgid, ns);
2726         if (task)
2727                 get_task_struct(task);
2728         rcu_read_unlock();
2729         if (!task)
2730                 goto out;
2731
2732         result = proc_pid_instantiate(dir, dentry, task, NULL);
2733         put_task_struct(task);
2734 out:
2735         return result;
2736 }
2737
2738 /*
2739  * Find the first task with tgid >= tgid
2740  *
2741  */
2742 struct tgid_iter {
2743         unsigned int tgid;
2744         struct task_struct *task;
2745 };
2746 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2747 {
2748         struct pid *pid;
2749
2750         if (iter.task)
2751                 put_task_struct(iter.task);
2752         rcu_read_lock();
2753 retry:
2754         iter.task = NULL;
2755         pid = find_ge_pid(iter.tgid, ns);
2756         if (pid) {
2757                 iter.tgid = pid_nr_ns(pid, ns);
2758                 iter.task = pid_task(pid, PIDTYPE_PID);
2759                 /* What we to know is if the pid we have find is the
2760                  * pid of a thread_group_leader.  Testing for task
2761                  * being a thread_group_leader is the obvious thing
2762                  * todo but there is a window when it fails, due to
2763                  * the pid transfer logic in de_thread.
2764                  *
2765                  * So we perform the straight forward test of seeing
2766                  * if the pid we have found is the pid of a thread
2767                  * group leader, and don't worry if the task we have
2768                  * found doesn't happen to be a thread group leader.
2769                  * As we don't care in the case of readdir.
2770                  */
2771                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2772                         iter.tgid += 1;
2773                         goto retry;
2774                 }
2775                 get_task_struct(iter.task);
2776         }
2777         rcu_read_unlock();
2778         return iter;
2779 }
2780
2781 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2782
2783 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2784         struct tgid_iter iter)
2785 {
2786         char name[PROC_NUMBUF];
2787         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2788         return proc_fill_cache(filp, dirent, filldir, name, len,
2789                                 proc_pid_instantiate, iter.task, NULL);
2790 }
2791
2792 /* for the /proc/ directory itself, after non-process stuff has been done */
2793 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
2794 {
2795         unsigned int nr = filp->f_pos - FIRST_PROCESS_ENTRY;
2796         struct task_struct *reaper = get_proc_task(filp->f_path.dentry->d_inode);
2797         struct tgid_iter iter;
2798         struct pid_namespace *ns;
2799
2800         if (!reaper)
2801                 goto out_no_task;
2802
2803         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
2804                 const struct pid_entry *p = &proc_base_stuff[nr];
2805                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
2806                         goto out;
2807         }
2808
2809         ns = filp->f_dentry->d_sb->s_fs_info;
2810         iter.task = NULL;
2811         iter.tgid = filp->f_pos - TGID_OFFSET;
2812         for (iter = next_tgid(ns, iter);
2813              iter.task;
2814              iter.tgid += 1, iter = next_tgid(ns, iter)) {
2815                 filp->f_pos = iter.tgid + TGID_OFFSET;
2816                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
2817                         put_task_struct(iter.task);
2818                         goto out;
2819                 }
2820         }
2821         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
2822 out:
2823         put_task_struct(reaper);
2824 out_no_task:
2825         return 0;
2826 }
2827
2828 /*
2829  * Tasks
2830  */
2831 static const struct pid_entry tid_base_stuff[] = {
2832         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2833         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fd_operations),
2834         REG("environ",   S_IRUSR, proc_environ_operations),
2835         INF("auxv",      S_IRUSR, proc_pid_auxv),
2836         ONE("status",    S_IRUGO, proc_pid_status),
2837         ONE("personality", S_IRUSR, proc_pid_personality),
2838         INF("limits",    S_IRUSR, proc_pid_limits),
2839 #ifdef CONFIG_SCHED_DEBUG
2840         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2841 #endif
2842 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2843         INF("syscall",   S_IRUSR, proc_pid_syscall),
2844 #endif
2845         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
2846         ONE("stat",      S_IRUGO, proc_tid_stat),
2847         ONE("statm",     S_IRUGO, proc_pid_statm),
2848         REG("maps",      S_IRUGO, proc_maps_operations),
2849 #ifdef CONFIG_NUMA
2850         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
2851 #endif
2852         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
2853         LNK("cwd",       proc_cwd_link),
2854         LNK("root",      proc_root_link),
2855         LNK("exe",       proc_exe_link),
2856         REG("mounts",    S_IRUGO, proc_mounts_operations),
2857         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2858 #ifdef CONFIG_PROC_PAGE_MONITOR
2859         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2860         REG("smaps",     S_IRUGO, proc_smaps_operations),
2861         REG("pagemap",    S_IRUSR, proc_pagemap_operations),
2862 #endif
2863 #ifdef CONFIG_SECURITY
2864         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2865 #endif
2866 #ifdef CONFIG_KALLSYMS
2867         INF("wchan",     S_IRUGO, proc_pid_wchan),
2868 #endif
2869 #ifdef CONFIG_STACKTRACE
2870         ONE("stack",      S_IRUSR, proc_pid_stack),
2871 #endif
2872 #ifdef CONFIG_SCHEDSTATS
2873         INF("schedstat", S_IRUGO, proc_pid_schedstat),
2874 #endif
2875 #ifdef CONFIG_LATENCYTOP
2876         REG("latency",  S_IRUGO, proc_lstats_operations),
2877 #endif
2878 #ifdef CONFIG_PROC_PID_CPUSET
2879         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
2880 #endif
2881 #ifdef CONFIG_CGROUPS
2882         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2883 #endif
2884         INF("oom_score", S_IRUGO, proc_oom_score),
2885         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2886 #ifdef CONFIG_AUDITSYSCALL
2887         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
2888         REG("sessionid",  S_IRUSR, proc_sessionid_operations),
2889 #endif
2890 #ifdef CONFIG_FAULT_INJECTION
2891         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2892 #endif
2893 #ifdef CONFIG_TASK_IO_ACCOUNTING
2894         INF("io",       S_IRUGO, proc_tid_io_accounting),
2895 #endif
2896 };
2897
2898 static int proc_tid_base_readdir(struct file * filp,
2899                              void * dirent, filldir_t filldir)
2900 {
2901         return proc_pident_readdir(filp,dirent,filldir,
2902                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
2903 }
2904
2905 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2906         return proc_pident_lookup(dir, dentry,
2907                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
2908 }
2909
2910 static const struct file_operations proc_tid_base_operations = {
2911         .read           = generic_read_dir,
2912         .readdir        = proc_tid_base_readdir,
2913 };
2914
2915 static const struct inode_operations proc_tid_base_inode_operations = {
2916         .lookup         = proc_tid_base_lookup,
2917         .getattr        = pid_getattr,
2918         .setattr        = proc_setattr,
2919 };
2920
2921 static struct dentry *proc_task_instantiate(struct inode *dir,
2922         struct dentry *dentry, struct task_struct *task, const void *ptr)
2923 {
2924         struct dentry *error = ERR_PTR(-ENOENT);
2925         struct inode *inode;
2926         inode = proc_pid_make_inode(dir->i_sb, task);
2927
2928         if (!inode)
2929                 goto out;
2930         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2931         inode->i_op = &proc_tid_base_inode_operations;
2932         inode->i_fop = &proc_tid_base_operations;
2933         inode->i_flags|=S_IMMUTABLE;
2934
2935         inode->i_nlink = 2 + pid_entry_count_dirs(tid_base_stuff,
2936                 ARRAY_SIZE(tid_base_stuff));
2937
2938         dentry->d_op = &pid_dentry_operations;
2939
2940         d_add(dentry, inode);
2941         /* Close the race of the process dying before we return the dentry */
2942         if (pid_revalidate(dentry, NULL))
2943                 error = NULL;
2944 out:
2945         return error;
2946 }
2947
2948 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2949 {
2950         struct dentry *result = ERR_PTR(-ENOENT);
2951         struct task_struct *task;
2952         struct task_struct *leader = get_proc_task(dir);
2953         unsigned tid;
2954         struct pid_namespace *ns;
2955
2956         if (!leader)
2957                 goto out_no_task;
2958
2959         tid = name_to_int(dentry);
2960         if (tid == ~0U)
2961                 goto out;
2962
2963         ns = dentry->d_sb->s_fs_info;
2964         rcu_read_lock();
2965         task = find_task_by_pid_ns(tid, ns);
2966         if (task)
2967                 get_task_struct(task);
2968         rcu_read_unlock();
2969         if (!task)
2970                 goto out;
2971         if (!same_thread_group(leader, task))
2972                 goto out_drop_task;
2973
2974         result = proc_task_instantiate(dir, dentry, task, NULL);
2975 out_drop_task:
2976         put_task_struct(task);
2977 out:
2978         put_task_struct(leader);
2979 out_no_task:
2980         return result;
2981 }
2982
2983 /*
2984  * Find the first tid of a thread group to return to user space.
2985  *
2986  * Usually this is just the thread group leader, but if the users
2987  * buffer was too small or there was a seek into the middle of the
2988  * directory we have more work todo.
2989  *
2990  * In the case of a short read we start with find_task_by_pid.
2991  *
2992  * In the case of a seek we start with the leader and walk nr
2993  * threads past it.
2994  */
2995 static struct task_struct *first_tid(struct task_struct *leader,
2996                 int tid, int nr, struct pid_namespace *ns)
2997 {
2998         struct task_struct *pos;
2999
3000         rcu_read_lock();
3001         /* Attempt to start with the pid of a thread */
3002         if (tid && (nr > 0)) {
3003                 pos = find_task_by_pid_ns(tid, ns);
3004                 if (pos && (pos->group_leader == leader))
3005                         goto found;
3006         }
3007
3008         /* If nr exceeds the number of threads there is nothing todo */
3009         pos = NULL;
3010         if (nr && nr >= get_nr_threads(leader))
3011                 goto out;
3012
3013         /* If we haven't found our starting place yet start
3014          * with the leader and walk nr threads forward.
3015          */
3016         for (pos = leader; nr > 0; --nr) {
3017                 pos = next_thread(pos);
3018                 if (pos == leader) {
3019                         pos = NULL;
3020                         goto out;
3021                 }
3022         }
3023 found:
3024         get_task_struct(pos);
3025 out:
3026         rcu_read_unlock();
3027         return pos;
3028 }
3029
3030 /*
3031  * Find the next thread in the thread list.
3032  * Return NULL if there is an error or no next thread.
3033  *
3034  * The reference to the input task_struct is released.
3035  */
3036 static struct task_struct *next_tid(struct task_struct *start)
3037 {
3038         struct task_struct *pos = NULL;
3039         rcu_read_lock();
3040         if (pid_alive(start)) {
3041                 pos = next_thread(start);
3042                 if (thread_group_leader(pos))
3043                         pos = NULL;
3044                 else
3045                         get_task_struct(pos);
3046         }
3047         rcu_read_unlock();
3048         put_task_struct(start);
3049         return pos;
3050 }
3051
3052 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3053         struct task_struct *task, int tid)
3054 {
3055         char name[PROC_NUMBUF];
3056         int len = snprintf(name, sizeof(name), "%d", tid);
3057         return proc_fill_cache(filp, dirent, filldir, name, len,
3058                                 proc_task_instantiate, task, NULL);
3059 }
3060
3061 /* for the /proc/TGID/task/ directories */
3062 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3063 {
3064         struct dentry *dentry = filp->f_path.dentry;
3065         struct inode *inode = dentry->d_inode;
3066         struct task_struct *leader = NULL;
3067         struct task_struct *task;
3068         int retval = -ENOENT;
3069         ino_t ino;
3070         int tid;
3071         struct pid_namespace *ns;
3072
3073         task = get_proc_task(inode);
3074         if (!task)
3075                 goto out_no_task;
3076         rcu_read_lock();
3077         if (pid_alive(task)) {
3078                 leader = task->group_leader;
3079                 get_task_struct(leader);
3080         }
3081         rcu_read_unlock();
3082         put_task_struct(task);
3083         if (!leader)
3084                 goto out_no_task;
3085         retval = 0;
3086
3087         switch ((unsigned long)filp->f_pos) {
3088         case 0:
3089                 ino = inode->i_ino;
3090                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3091                         goto out;
3092                 filp->f_pos++;
3093                 /* fall through */
3094         case 1:
3095                 ino = parent_ino(dentry);
3096                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3097                         goto out;
3098                 filp->f_pos++;
3099                 /* fall through */
3100         }
3101
3102         /* f_version caches the tgid value that the last readdir call couldn't
3103          * return. lseek aka telldir automagically resets f_version to 0.
3104          */
3105         ns = filp->f_dentry->d_sb->s_fs_info;
3106         tid = (int)filp->f_version;
3107         filp->f_version = 0;
3108         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3109              task;
3110              task = next_tid(task), filp->f_pos++) {
3111                 tid = task_pid_nr_ns(task, ns);
3112                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3113                         /* returning this tgid failed, save it as the first
3114                          * pid for the next readir call */
3115                         filp->f_version = (u64)tid;
3116                         put_task_struct(task);
3117                         break;
3118                 }
3119         }
3120 out:
3121         put_task_struct(leader);
3122 out_no_task:
3123         return retval;
3124 }
3125
3126 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3127 {
3128         struct inode *inode = dentry->d_inode;
3129         struct task_struct *p = get_proc_task(inode);
3130         generic_fillattr(inode, stat);
3131
3132         if (p) {
3133                 stat->nlink += get_nr_threads(p);
3134                 put_task_struct(p);
3135         }
3136
3137         return 0;
3138 }
3139
3140 static const struct inode_operations proc_task_inode_operations = {
3141         .lookup         = proc_task_lookup,
3142         .getattr        = proc_task_getattr,
3143         .setattr        = proc_setattr,
3144 };
3145
3146 static const struct file_operations proc_task_operations = {
3147         .read           = generic_read_dir,
3148         .readdir        = proc_task_readdir,
3149 };