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