fs/proc/base.c

   1 /*
   2  *  linux/fs/proc/base.c
   3  *
   4  *  Copyright (C) 1991, 1992 Linus Torvalds
   5  *
   6  *  proc base directory handling functions
   7  *
   8  *  1999, Al Viro. Rewritten. Now it covers the whole per-process part.
   9  *  Instead of using magical inumbers to determine the kind of object
  10  *  we allocate and fill in-core inodes upon lookup. They don't even
  11  *  go into icache. We cache the reference to task_struct upon lookup too.
  12  *  Eventually it should become a filesystem in its own. We don't use the
  13  *  rest of procfs anymore.
  14  *
  15  *
  16  *  Changelog:
  17  *  17-Jan-2005
  18  *  Allan Bezerra
  19  *  Bruna Moreira <bruna.moreira@indt.org.br>
  20  *  Edjard Mota <edjard.mota@indt.org.br>
  21  *  Ilias Biris <ilias.biris@indt.org.br>
  22  *  Mauricio Lin <mauricio.lin@indt.org.br>
  23  *
  24  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  25  *
  26  *  A new process specific entry (smaps) included in /proc. It shows the
  27  *  size of rss for each memory area. The maps entry lacks information
  28  *  about physical memory size (rss) for each mapped file, i.e.,
  29  *  rss information for executables and library files.
  30  *  This additional information is useful for any tools that need to know
  31  *  about physical memory consumption for a process specific library.
  32  *
  33  *  Changelog:
  34  *  21-Feb-2005
  35  *  Embedded Linux Lab - 10LE Instituto Nokia de Tecnologia - INdT
  36  *  Pud inclusion in the page table walking.
  37  *
  38  *  ChangeLog:
  39  *  10-Mar-2005
  40  *  10LE Instituto Nokia de Tecnologia - INdT:
  41  *  A better way to walks through the page table as suggested by Hugh Dickins.
  42  *
  43  *  Simo Piiroinen <simo.piiroinen@nokia.com>:
  44  *  Smaps information related to shared, private, clean and dirty pages.
  45  *
  46  *  Paul Mundt <paul.mundt@nokia.com>:
  47  *  Overall revision about smaps.
  48  */
  49
  50 #include <asm/uaccess.h>
  51
  52 #include <linux/errno.h>
  53 #include <linux/time.h>
  54 #include <linux/proc_fs.h>
  55 #include <linux/stat.h>
  56 #include <linux/task_io_accounting_ops.h>
  57 #include <linux/init.h>
  58 #include <linux/capability.h>
  59 #include <linux/file.h>
  60 #include <linux/fdtable.h>
  61 #include <linux/string.h>
  62 #include <linux/seq_file.h>
  63 #include <linux/namei.h>
  64 #include <linux/mnt_namespace.h>
  65 #include <linux/mm.h>
  66 #include <linux/swap.h>
  67 #include <linux/rcupdate.h>
  68 #include <linux/kallsyms.h>
  69 #include <linux/stacktrace.h>
  70 #include <linux/resource.h>
  71 #include <linux/module.h>
  72 #include <linux/mount.h>
  73 #include <linux/security.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/tracehook.h>
  76 #include <linux/cgroup.h>
  77 #include <linux/cpuset.h>
  78 #include <linux/audit.h>
  79 #include <linux/poll.h>
  80 #include <linux/nsproxy.h>
  81 #include <linux/oom.h>
  82 #include <linux/elf.h>
  83 #include <linux/pid_namespace.h>
  84 #include <linux/fs_struct.h>
  85 #include <linux/slab.h>
  86 #ifdef CONFIG_HARDWALL
  87 #include <asm/hardwall.h>
  88 #endif
  89 #include "internal.h"
  90
  91 /* NOTE:
  92  *      Implementing inode permission operations in /proc is almost
  93  *      certainly an error.  Permission checks need to happen during
  94  *      each system call not at open time.  The reason is that most of
  95  *      what we wish to check for permissions in /proc varies at runtime.
  96  *
  97  *      The classic example of a problem is opening file descriptors
  98  *      in /proc for a task before it execs a suid executable.
  99  */
 100
 101 struct pid_entry {
 102         char *name;
 103         int len;
 104         mode_t mode;
 105         const struct inode_operations *iop;
 106         const struct file_operations *fop;
 107         union proc_op op;
 108 };
 109
 110 #define NOD(NAME, MODE, IOP, FOP, OP) {                 \
 111         .name = (NAME),                                 \
 112         .len  = sizeof(NAME) - 1,                       \
 113         .mode = MODE,                                   \
 114         .iop  = IOP,                                    \
 115         .fop  = FOP,                                    \
 116         .op   = OP,                                     \
 117 }
 118
 119 #define DIR(NAME, MODE, iops, fops)     \
 120         NOD(NAME, (S_IFDIR|(MODE)), &iops, &fops, {} )
 121 #define LNK(NAME, get_link)                                     \
 122         NOD(NAME, (S_IFLNK|S_IRWXUGO),                          \
 123                 &proc_pid_link_inode_operations, NULL,          \
 124                 { .proc_get_link = get_link } )
 125 #define REG(NAME, MODE, fops)                           \
 126         NOD(NAME, (S_IFREG|(MODE)), NULL, &fops, {})
 127 #define INF(NAME, MODE, read)                           \
 128         NOD(NAME, (S_IFREG|(MODE)),                     \
 129                 NULL, &proc_info_file_operations,       \
 130                 { .proc_read = read } )
 131 #define ONE(NAME, MODE, show)                           \
 132         NOD(NAME, (S_IFREG|(MODE)),                     \
 133                 NULL, &proc_single_file_operations,     \
 134                 { .proc_show = show } )
 135
 136 /*
 137  * Count the number of hardlinks for the pid_entry table, excluding the .
 138  * and .. links.
 139  */
 140 static unsigned int pid_entry_count_dirs(const struct pid_entry *entries,
 141         unsigned int n)
 142 {
 143         unsigned int i;
 144         unsigned int count;
 145
 146         count = 0;
 147         for (i = 0; i < n; ++i) {
 148                 if (S_ISDIR(entries[i].mode))
 149                         ++count;
 150         }
 151
 152         return count;
 153 }
 154
 155 static int get_task_root(struct task_struct *task, struct path *root)
 156 {
 157         int result = -ENOENT;
 158
 159         task_lock(task);
 160         if (task->fs) {
 161                 get_fs_root(task->fs, root);
 162                 result = 0;
 163         }
 164         task_unlock(task);
 165         return result;
 166 }
 167
 168 static int proc_cwd_link(struct inode *inode, struct path *path)
 169 {
 170         struct task_struct *task = get_proc_task(inode);
 171         int result = -ENOENT;
 172
 173         if (task) {
 174                 task_lock(task);
 175                 if (task->fs) {
 176                         get_fs_pwd(task->fs, path);
 177                         result = 0;
 178                 }
 179                 task_unlock(task);
 180                 put_task_struct(task);
 181         }
 182         return result;
 183 }
 184
 185 static int proc_root_link(struct inode *inode, struct path *path)
 186 {
 187         struct task_struct *task = get_proc_task(inode);
 188         int result = -ENOENT;
 189
 190         if (task) {
 191                 result = get_task_root(task, path);
 192                 put_task_struct(task);
 193         }
 194         return result;
 195 }
 196
 197 static struct mm_struct *mm_access(struct task_struct *task, unsigned int mode)
 198 {
 199         struct mm_struct *mm;
 200         int err;
 201
 202         err =  mutex_lock_killable(&task->signal->cred_guard_mutex);
 203         if (err)
 204                 return ERR_PTR(err);
 205
 206         mm = get_task_mm(task);
 207         if (mm && mm != current->mm &&
 208                         !ptrace_may_access(task, mode)) {
 209                 mmput(mm);
 210                 mm = ERR_PTR(-EACCES);
 211         }
 212         mutex_unlock(&task->signal->cred_guard_mutex);
 213
 214         return mm;
 215 }
 216
 217 struct mm_struct *mm_for_maps(struct task_struct *task)
 218 {
 219         return mm_access(task, PTRACE_MODE_READ);
 220 }
 221
 222 static int proc_pid_cmdline(struct task_struct *task, char * buffer)
 223 {
 224         int res = 0;
 225         unsigned int len;
 226         struct mm_struct *mm = get_task_mm(task);
 227         if (!mm)
 228                 goto out;
 229         if (!mm->arg_end)
 230                 goto out_mm;    /* Shh! No looking before we're done */
 231
 232         len = mm->arg_end - mm->arg_start;
 233
 234         if (len > PAGE_SIZE)
 235                 len = PAGE_SIZE;
 236
 237         res = access_process_vm(task, mm->arg_start, buffer, len, 0);
 238
 239         // If the nul at the end of args has been overwritten, then
 240         // assume application is using setproctitle(3).
 241         if (res > 0 && buffer[res-1] != '\0' && len < PAGE_SIZE) {
 242                 len = strnlen(buffer, res);
 243                 if (len < res) {
 244                     res = len;
 245                 } else {
 246                         len = mm->env_end - mm->env_start;
 247                         if (len > PAGE_SIZE - res)
 248                                 len = PAGE_SIZE - res;
 249                         res += access_process_vm(task, mm->env_start, buffer+res, len, 0);
 250                         res = strnlen(buffer, res);
 251                 }
 252         }
 253 out_mm:
 254         mmput(mm);
 255 out:
 256         return res;
 257 }
 258
 259 static int proc_pid_auxv(struct task_struct *task, char *buffer)
 260 {
 261         struct mm_struct *mm = mm_for_maps(task);
 262         int res = PTR_ERR(mm);
 263         if (mm && !IS_ERR(mm)) {
 264                 unsigned int nwords = 0;
 265                 do {
 266                         nwords += 2;
 267                 } while (mm->saved_auxv[nwords - 2] != 0); /* AT_NULL */
 268                 res = nwords * sizeof(mm->saved_auxv[0]);
 269                 if (res > PAGE_SIZE)
 270                         res = PAGE_SIZE;
 271                 memcpy(buffer, mm->saved_auxv, res);
 272                 mmput(mm);
 273         }
 274         return res;
 275 }
 276
 277
 278 #ifdef CONFIG_KALLSYMS
 279 /*
 280  * Provides a wchan file via kallsyms in a proper one-value-per-file format.
 281  * Returns the resolved symbol.  If that fails, simply return the address.
 282  */
 283 static int proc_pid_wchan(struct task_struct *task, char *buffer)
 284 {
 285         unsigned long wchan;
 286         char symname[KSYM_NAME_LEN];
 287
 288         wchan = get_wchan(task);
 289
 290         if (lookup_symbol_name(wchan, symname) < 0)
 291                 if (!ptrace_may_access(task, PTRACE_MODE_READ))
 292                         return 0;
 293                 else
 294                         return sprintf(buffer, "%lu", wchan);
 295         else
 296                 return sprintf(buffer, "%s", symname);
 297 }
 298 #endif /* CONFIG_KALLSYMS */
 299
 300 static int lock_trace(struct task_struct *task)
 301 {
 302         int err = mutex_lock_killable(&task->signal->cred_guard_mutex);
 303         if (err)
 304                 return err;
 305         if (!ptrace_may_access(task, PTRACE_MODE_ATTACH)) {
 306                 mutex_unlock(&task->signal->cred_guard_mutex);
 307                 return -EPERM;
 308         }
 309         return 0;
 310 }
 311
 312 static void unlock_trace(struct task_struct *task)
 313 {
 314         mutex_unlock(&task->signal->cred_guard_mutex);
 315 }
 316
 317 #ifdef CONFIG_STACKTRACE
 318
 319 #define MAX_STACK_TRACE_DEPTH   64
 320
 321 static int proc_pid_stack(struct seq_file *m, struct pid_namespace *ns,
 322                           struct pid *pid, struct task_struct *task)
 323 {
 324         struct stack_trace trace;
 325         unsigned long *entries;
 326         int err;
 327         int i;
 328
 329         entries = kmalloc(MAX_STACK_TRACE_DEPTH * sizeof(*entries), GFP_KERNEL);
 330         if (!entries)
 331                 return -ENOMEM;
 332
 333         trace.nr_entries        = 0;
 334         trace.max_entries       = MAX_STACK_TRACE_DEPTH;
 335         trace.entries           = entries;
 336         trace.skip              = 0;
 337
 338         err = lock_trace(task);
 339         if (!err) {
 340                 save_stack_trace_tsk(task, &trace);
 341
 342                 for (i = 0; i < trace.nr_entries; i++) {
 343                         seq_printf(m, "[<%pK>] %pS\n",
 344                                    (void *)entries[i], (void *)entries[i]);
 345                 }
 346                 unlock_trace(task);
 347         }
 348         kfree(entries);
 349
 350         return err;
 351 }
 352 #endif
 353
 354 #ifdef CONFIG_SCHEDSTATS
 355 /*
 356  * Provides /proc/PID/schedstat
 357  */
 358 static int proc_pid_schedstat(struct task_struct *task, char *buffer)
 359 {
 360         return sprintf(buffer, "%llu %llu %lu\n",
 361                         (unsigned long long)task->se.sum_exec_runtime,
 362                         (unsigned long long)task->sched_info.run_delay,
 363                         task->sched_info.pcount);
 364 }
 365 #endif
 366
 367 #ifdef CONFIG_LATENCYTOP
 368 static int lstats_show_proc(struct seq_file *m, void *v)
 369 {
 370         int i;
 371         struct inode *inode = m->private;
 372         struct task_struct *task = get_proc_task(inode);
 373
 374         if (!task)
 375                 return -ESRCH;
 376         seq_puts(m, "Latency Top version : v0.1\n");
 377         for (i = 0; i < 32; i++) {
 378                 struct latency_record *lr = &task->latency_record[i];
 379                 if (lr->backtrace[0]) {
 380                         int q;
 381                         seq_printf(m, "%i %li %li",
 382                                    lr->count, lr->time, lr->max);
 383                         for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
 384                                 unsigned long bt = lr->backtrace[q];
 385                                 if (!bt)
 386                                         break;
 387                                 if (bt == ULONG_MAX)
 388                                         break;
 389                                 seq_printf(m, " %ps", (void *)bt);
 390                         }
 391                         seq_putc(m, '\n');
 392                 }
 393
 394         }
 395         put_task_struct(task);
 396         return 0;
 397 }
 398
 399 static int lstats_open(struct inode *inode, struct file *file)
 400 {
 401         return single_open(file, lstats_show_proc, inode);
 402 }
 403
 404 static ssize_t lstats_write(struct file *file, const char __user *buf,
 405                             size_t count, loff_t *offs)
 406 {
 407         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 408
 409         if (!task)
 410                 return -ESRCH;
 411         clear_all_latency_tracing(task);
 412         put_task_struct(task);
 413
 414         return count;
 415 }
 416
 417 static const struct file_operations proc_lstats_operations = {
 418         .open           = lstats_open,
 419         .read           = seq_read,
 420         .write          = lstats_write,
 421         .llseek         = seq_lseek,
 422         .release        = single_release,
 423 };
 424
 425 #endif
 426
 427 static int proc_oom_score(struct task_struct *task, char *buffer)
 428 {
 429         unsigned long points = 0;
 430
 431         read_lock(&tasklist_lock);
 432         if (pid_alive(task))
 433                 points = oom_badness(task, NULL, NULL,
 434                                         totalram_pages + total_swap_pages);
 435         read_unlock(&tasklist_lock);
 436         return sprintf(buffer, "%lu\n", points);
 437 }
 438
 439 struct limit_names {
 440         char *name;
 441         char *unit;
 442 };
 443
 444 static const struct limit_names lnames[RLIM_NLIMITS] = {
 445         [RLIMIT_CPU] = {"Max cpu time", "seconds"},
 446         [RLIMIT_FSIZE] = {"Max file size", "bytes"},
 447         [RLIMIT_DATA] = {"Max data size", "bytes"},
 448         [RLIMIT_STACK] = {"Max stack size", "bytes"},
 449         [RLIMIT_CORE] = {"Max core file size", "bytes"},
 450         [RLIMIT_RSS] = {"Max resident set", "bytes"},
 451         [RLIMIT_NPROC] = {"Max processes", "processes"},
 452         [RLIMIT_NOFILE] = {"Max open files", "files"},
 453         [RLIMIT_MEMLOCK] = {"Max locked memory", "bytes"},
 454         [RLIMIT_AS] = {"Max address space", "bytes"},
 455         [RLIMIT_LOCKS] = {"Max file locks", "locks"},
 456         [RLIMIT_SIGPENDING] = {"Max pending signals", "signals"},
 457         [RLIMIT_MSGQUEUE] = {"Max msgqueue size", "bytes"},
 458         [RLIMIT_NICE] = {"Max nice priority", NULL},
 459         [RLIMIT_RTPRIO] = {"Max realtime priority", NULL},
 460         [RLIMIT_RTTIME] = {"Max realtime timeout", "us"},
 461 };
 462
 463 /* Display limits for a process */
 464 static int proc_pid_limits(struct task_struct *task, char *buffer)
 465 {
 466         unsigned int i;
 467         int count = 0;
 468         unsigned long flags;
 469         char *bufptr = buffer;
 470
 471         struct rlimit rlim[RLIM_NLIMITS];
 472
 473         if (!lock_task_sighand(task, &flags))
 474                 return 0;
 475         memcpy(rlim, task->signal->rlim, sizeof(struct rlimit) * RLIM_NLIMITS);
 476         unlock_task_sighand(task, &flags);
 477
 478         /*
 479          * print the file header
 480          */
 481         count += sprintf(&bufptr[count], "%-25s %-20s %-20s %-10s\n",
 482                         "Limit", "Soft Limit", "Hard Limit", "Units");
 483
 484         for (i = 0; i < RLIM_NLIMITS; i++) {
 485                 if (rlim[i].rlim_cur == RLIM_INFINITY)
 486                         count += sprintf(&bufptr[count], "%-25s %-20s ",
 487                                          lnames[i].name, "unlimited");
 488                 else
 489                         count += sprintf(&bufptr[count], "%-25s %-20lu ",
 490                                          lnames[i].name, rlim[i].rlim_cur);
 491
 492                 if (rlim[i].rlim_max == RLIM_INFINITY)
 493                         count += sprintf(&bufptr[count], "%-20s ", "unlimited");
 494                 else
 495                         count += sprintf(&bufptr[count], "%-20lu ",
 496                                          rlim[i].rlim_max);
 497
 498                 if (lnames[i].unit)
 499                         count += sprintf(&bufptr[count], "%-10s\n",
 500                                          lnames[i].unit);
 501                 else
 502                         count += sprintf(&bufptr[count], "\n");
 503         }
 504
 505         return count;
 506 }
 507
 508 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 509 static int proc_pid_syscall(struct task_struct *task, char *buffer)
 510 {
 511         long nr;
 512         unsigned long args[6], sp, pc;
 513         int res = lock_trace(task);
 514         if (res)
 515                 return res;
 516
 517         if (task_current_syscall(task, &nr, args, 6, &sp, &pc))
 518                 res = sprintf(buffer, "running\n");
 519         else if (nr < 0)
 520                 res = sprintf(buffer, "%ld 0x%lx 0x%lx\n", nr, sp, pc);
 521         else
 522                 res = sprintf(buffer,
 523                        "%ld 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx 0x%lx\n",
 524                        nr,
 525                        args[0], args[1], args[2], args[3], args[4], args[5],
 526                        sp, pc);
 527         unlock_trace(task);
 528         return res;
 529 }
 530 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
 531
 532 /************************************************************************/
 533 /*                       Here the fs part begins                        */
 534 /************************************************************************/
 535
 536 /* permission checks */
 537 static int proc_fd_access_allowed(struct inode *inode)
 538 {
 539         struct task_struct *task;
 540         int allowed = 0;
 541         /* Allow access to a task's file descriptors if it is us or we
 542          * may use ptrace attach to the process and find out that
 543          * information.
 544          */
 545         task = get_proc_task(inode);
 546         if (task) {
 547                 allowed = ptrace_may_access(task, PTRACE_MODE_READ);
 548                 put_task_struct(task);
 549         }
 550         return allowed;
 551 }
 552
 553 int proc_setattr(struct dentry *dentry, struct iattr *attr)
 554 {
 555         int error;
 556         struct inode *inode = dentry->d_inode;
 557
 558         if (attr->ia_valid & ATTR_MODE)
 559                 return -EPERM;
 560
 561         error = inode_change_ok(inode, attr);
 562         if (error)
 563                 return error;
 564
 565         if ((attr->ia_valid & ATTR_SIZE) &&
 566             attr->ia_size != i_size_read(inode)) {
 567                 error = vmtruncate(inode, attr->ia_size);
 568                 if (error)
 569                         return error;
 570         }
 571
 572         setattr_copy(inode, attr);
 573         mark_inode_dirty(inode);
 574         return 0;
 575 }
 576
 577 static const struct inode_operations proc_def_inode_operations = {
 578         .setattr        = proc_setattr,
 579 };
 580
 581 static int mounts_open_common(struct inode *inode, struct file *file,
 582                               const struct seq_operations *op)
 583 {
 584         struct task_struct *task = get_proc_task(inode);
 585         struct nsproxy *nsp;
 586         struct mnt_namespace *ns = NULL;
 587         struct path root;
 588         struct proc_mounts *p;
 589         int ret = -EINVAL;
 590
 591         if (task) {
 592                 rcu_read_lock();
 593                 nsp = task_nsproxy(task);
 594                 if (nsp) {
 595                         ns = nsp->mnt_ns;
 596                         if (ns)
 597                                 get_mnt_ns(ns);
 598                 }
 599                 rcu_read_unlock();
 600                 if (ns && get_task_root(task, &root) == 0)
 601                         ret = 0;
 602                 put_task_struct(task);
 603         }
 604
 605         if (!ns)
 606                 goto err;
 607         if (ret)
 608                 goto err_put_ns;
 609
 610         ret = -ENOMEM;
 611         p = kmalloc(sizeof(struct proc_mounts), GFP_KERNEL);
 612         if (!p)
 613                 goto err_put_path;
 614
 615         file->private_data = &p->m;
 616         ret = seq_open(file, op);
 617         if (ret)
 618                 goto err_free;
 619
 620         p->m.private = p;
 621         p->ns = ns;
 622         p->root = root;
 623         p->m.poll_event = ns->event;
 624
 625         return 0;
 626
 627  err_free:
 628         kfree(p);
 629  err_put_path:
 630         path_put(&root);
 631  err_put_ns:
 632         put_mnt_ns(ns);
 633  err:
 634         return ret;
 635 }
 636
 637 static int mounts_release(struct inode *inode, struct file *file)
 638 {
 639         struct proc_mounts *p = file->private_data;
 640         path_put(&p->root);
 641         put_mnt_ns(p->ns);
 642         return seq_release(inode, file);
 643 }
 644
 645 static unsigned mounts_poll(struct file *file, poll_table *wait)
 646 {
 647         struct proc_mounts *p = file->private_data;
 648         unsigned res = POLLIN | POLLRDNORM;
 649
 650         poll_wait(file, &p->ns->poll, wait);
 651         if (mnt_had_events(p))
 652                 res |= POLLERR | POLLPRI;
 653
 654         return res;
 655 }
 656
 657 static int mounts_open(struct inode *inode, struct file *file)
 658 {
 659         return mounts_open_common(inode, file, &mounts_op);
 660 }
 661
 662 static const struct file_operations proc_mounts_operations = {
 663         .open           = mounts_open,
 664         .read           = seq_read,
 665         .llseek         = seq_lseek,
 666         .release        = mounts_release,
 667         .poll           = mounts_poll,
 668 };
 669
 670 static int mountinfo_open(struct inode *inode, struct file *file)
 671 {
 672         return mounts_open_common(inode, file, &mountinfo_op);
 673 }
 674
 675 static const struct file_operations proc_mountinfo_operations = {
 676         .open           = mountinfo_open,
 677         .read           = seq_read,
 678         .llseek         = seq_lseek,
 679         .release        = mounts_release,
 680         .poll           = mounts_poll,
 681 };
 682
 683 static int mountstats_open(struct inode *inode, struct file *file)
 684 {
 685         return mounts_open_common(inode, file, &mountstats_op);
 686 }
 687
 688 static const struct file_operations proc_mountstats_operations = {
 689         .open           = mountstats_open,
 690         .read           = seq_read,
 691         .llseek         = seq_lseek,
 692         .release        = mounts_release,
 693 };
 694
 695 #define PROC_BLOCK_SIZE (3*1024)                /* 4K page size but our output routines use some slack for overruns */
 696
 697 static ssize_t proc_info_read(struct file * file, char __user * buf,
 698                           size_t count, loff_t *ppos)
 699 {
 700         struct inode * inode = file->f_path.dentry->d_inode;
 701         unsigned long page;
 702         ssize_t length;
 703         struct task_struct *task = get_proc_task(inode);
 704
 705         length = -ESRCH;
 706         if (!task)
 707                 goto out_no_task;
 708
 709         if (count > PROC_BLOCK_SIZE)
 710                 count = PROC_BLOCK_SIZE;
 711
 712         length = -ENOMEM;
 713         if (!(page = __get_free_page(GFP_TEMPORARY)))
 714                 goto out;
 715
 716         length = PROC_I(inode)->op.proc_read(task, (char*)page);
 717
 718         if (length >= 0)
 719                 length = simple_read_from_buffer(buf, count, ppos, (char *)page, length);
 720         free_page(page);
 721 out:
 722         put_task_struct(task);
 723 out_no_task:
 724         return length;
 725 }
 726
 727 static const struct file_operations proc_info_file_operations = {
 728         .read           = proc_info_read,
 729         .llseek         = generic_file_llseek,
 730 };
 731
 732 static int proc_single_show(struct seq_file *m, void *v)
 733 {
 734         struct inode *inode = m->private;
 735         struct pid_namespace *ns;
 736         struct pid *pid;
 737         struct task_struct *task;
 738         int ret;
 739
 740         ns = inode->i_sb->s_fs_info;
 741         pid = proc_pid(inode);
 742         task = get_pid_task(pid, PIDTYPE_PID);
 743         if (!task)
 744                 return -ESRCH;
 745
 746         ret = PROC_I(inode)->op.proc_show(m, ns, pid, task);
 747
 748         put_task_struct(task);
 749         return ret;
 750 }
 751
 752 static int proc_single_open(struct inode *inode, struct file *filp)
 753 {
 754         return single_open(filp, proc_single_show, inode);
 755 }
 756
 757 static const struct file_operations proc_single_file_operations = {
 758         .open           = proc_single_open,
 759         .read           = seq_read,
 760         .llseek         = seq_lseek,
 761         .release        = single_release,
 762 };
 763
 764 static int mem_open(struct inode* inode, struct file* file)
 765 {
 766         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 767         struct mm_struct *mm;
 768
 769         if (!task)
 770                 return -ESRCH;
 771
 772         mm = mm_access(task, PTRACE_MODE_ATTACH);
 773         put_task_struct(task);
 774
 775         if (IS_ERR(mm))
 776                 return PTR_ERR(mm);
 777
 778         if (mm) {
 779                 /* ensure this mm_struct can't be freed */
 780                 atomic_inc(&mm->mm_count);
 781                 /* but do not pin its memory */
 782                 mmput(mm);
 783         }
 784
 785         /* OK to pass negative loff_t, we can catch out-of-range */
 786         file->f_mode |= FMODE_UNSIGNED_OFFSET;
 787         file->private_data = mm;
 788
 789         return 0;
 790 }
 791
 792 static ssize_t mem_rw(struct file *file, char __user *buf,
 793                         size_t count, loff_t *ppos, int write)
 794 {
 795         struct mm_struct *mm = file->private_data;
 796         unsigned long addr = *ppos;
 797         ssize_t copied;
 798         char *page;
 799
 800         if (!mm)
 801                 return 0;
 802
 803         page = (char *)__get_free_page(GFP_TEMPORARY);
 804         if (!page)
 805                 return -ENOMEM;
 806
 807         copied = 0;
 808         if (!atomic_inc_not_zero(&mm->mm_users))
 809                 goto free;
 810
 811         while (count > 0) {
 812                 int this_len = min_t(int, count, PAGE_SIZE);
 813
 814                 if (write && copy_from_user(page, buf, this_len)) {
 815                         copied = -EFAULT;
 816                         break;
 817                 }
 818
 819                 this_len = access_remote_vm(mm, addr, page, this_len, write);
 820                 if (!this_len) {
 821                         if (!copied)
 822                                 copied = -EIO;
 823                         break;
 824                 }
 825
 826                 if (!write && copy_to_user(buf, page, this_len)) {
 827                         copied = -EFAULT;
 828                         break;
 829                 }
 830
 831                 buf += this_len;
 832                 addr += this_len;
 833                 copied += this_len;
 834                 count -= this_len;
 835         }
 836         *ppos = addr;
 837
 838         mmput(mm);
 839 free:
 840         free_page((unsigned long) page);
 841         return copied;
 842 }
 843
 844 static ssize_t mem_read(struct file *file, char __user *buf,
 845                         size_t count, loff_t *ppos)
 846 {
 847         return mem_rw(file, buf, count, ppos, 0);
 848 }
 849
 850 static ssize_t mem_write(struct file *file, const char __user *buf,
 851                          size_t count, loff_t *ppos)
 852 {
 853         return mem_rw(file, (char __user*)buf, count, ppos, 1);
 854 }
 855
 856 loff_t mem_lseek(struct file *file, loff_t offset, int orig)
 857 {
 858         switch (orig) {
 859         case 0:
 860                 file->f_pos = offset;
 861                 break;
 862         case 1:
 863                 file->f_pos += offset;
 864                 break;
 865         default:
 866                 return -EINVAL;
 867         }
 868         force_successful_syscall_return();
 869         return file->f_pos;
 870 }
 871
 872 static int mem_release(struct inode *inode, struct file *file)
 873 {
 874         struct mm_struct *mm = file->private_data;
 875         if (mm)
 876                 mmdrop(mm);
 877         return 0;
 878 }
 879
 880 static const struct file_operations proc_mem_operations = {
 881         .llseek         = mem_lseek,
 882         .read           = mem_read,
 883         .write          = mem_write,
 884         .open           = mem_open,
 885         .release        = mem_release,
 886 };
 887
 888 static ssize_t environ_read(struct file *file, char __user *buf,
 889                         size_t count, loff_t *ppos)
 890 {
 891         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
 892         char *page;
 893         unsigned long src = *ppos;
 894         int ret = -ESRCH;
 895         struct mm_struct *mm;
 896
 897         if (!task)
 898                 goto out_no_task;
 899
 900         ret = -ENOMEM;
 901         page = (char *)__get_free_page(GFP_TEMPORARY);
 902         if (!page)
 903                 goto out;
 904
 905
 906         mm = mm_for_maps(task);
 907         ret = PTR_ERR(mm);
 908         if (!mm || IS_ERR(mm))
 909                 goto out_free;
 910
 911         ret = 0;
 912         while (count > 0) {
 913                 int this_len, retval, max_len;
 914
 915                 this_len = mm->env_end - (mm->env_start + src);
 916
 917                 if (this_len <= 0)
 918                         break;
 919
 920                 max_len = (count > PAGE_SIZE) ? PAGE_SIZE : count;
 921                 this_len = (this_len > max_len) ? max_len : this_len;
 922
 923                 retval = access_process_vm(task, (mm->env_start + src),
 924                         page, this_len, 0);
 925
 926                 if (retval <= 0) {
 927                         ret = retval;
 928                         break;
 929                 }
 930
 931                 if (copy_to_user(buf, page, retval)) {
 932                         ret = -EFAULT;
 933                         break;
 934                 }
 935
 936                 ret += retval;
 937                 src += retval;
 938                 buf += retval;
 939                 count -= retval;
 940         }
 941         *ppos = src;
 942
 943         mmput(mm);
 944 out_free:
 945         free_page((unsigned long) page);
 946 out:
 947         put_task_struct(task);
 948 out_no_task:
 949         return ret;
 950 }
 951
 952 static const struct file_operations proc_environ_operations = {
 953         .read           = environ_read,
 954         .llseek         = generic_file_llseek,
 955 };
 956
 957 static ssize_t oom_adjust_read(struct file *file, char __user *buf,
 958                                 size_t count, loff_t *ppos)
 959 {
 960         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
 961         char buffer[PROC_NUMBUF];
 962         size_t len;
 963         int oom_adjust = OOM_DISABLE;
 964         unsigned long flags;
 965
 966         if (!task)
 967                 return -ESRCH;
 968
 969         if (lock_task_sighand(task, &flags)) {
 970                 oom_adjust = task->signal->oom_adj;
 971                 unlock_task_sighand(task, &flags);
 972         }
 973
 974         put_task_struct(task);
 975
 976         len = snprintf(buffer, sizeof(buffer), "%i\n", oom_adjust);
 977
 978         return simple_read_from_buffer(buf, count, ppos, buffer, len);
 979 }
 980
 981 static ssize_t oom_adjust_write(struct file *file, const char __user *buf,
 982                                 size_t count, loff_t *ppos)
 983 {
 984         struct task_struct *task;
 985         char buffer[PROC_NUMBUF];
 986         int oom_adjust;
 987         unsigned long flags;
 988         int err;
 989
 990         memset(buffer, 0, sizeof(buffer));
 991         if (count > sizeof(buffer) - 1)
 992                 count = sizeof(buffer) - 1;
 993         if (copy_from_user(buffer, buf, count)) {
 994                 err = -EFAULT;
 995                 goto out;
 996         }
 997
 998         err = kstrtoint(strstrip(buffer), 0, &oom_adjust);
 999         if (err)
1000                 goto out;
1001         if ((oom_adjust < OOM_ADJUST_MIN || oom_adjust > OOM_ADJUST_MAX) &&
1002              oom_adjust != OOM_DISABLE) {
1003                 err = -EINVAL;
1004                 goto out;
1005         }
1006
1007         task = get_proc_task(file->f_path.dentry->d_inode);
1008         if (!task) {
1009                 err = -ESRCH;
1010                 goto out;
1011         }
1012
1013         task_lock(task);
1014         if (!task->mm) {
1015                 err = -EINVAL;
1016                 goto err_task_lock;
1017         }
1018
1019         if (!lock_task_sighand(task, &flags)) {
1020                 err = -ESRCH;
1021                 goto err_task_lock;
1022         }
1023
1024         if (oom_adjust < task->signal->oom_adj && !capable(CAP_SYS_RESOURCE)) {
1025                 err = -EACCES;
1026                 goto err_sighand;
1027         }
1028
1029         /*
1030          * Warn that /proc/pid/oom_adj is deprecated, see
1031          * Documentation/feature-removal-schedule.txt.
1032          */
1033         printk_once(KERN_WARNING "%s (%d): /proc/%d/oom_adj is deprecated, please use /proc/%d/oom_score_adj instead.\n",
1034                   current->comm, task_pid_nr(current), task_pid_nr(task),
1035                   task_pid_nr(task));
1036         task->signal->oom_adj = oom_adjust;
1037         /*
1038          * Scale /proc/pid/oom_score_adj appropriately ensuring that a maximum
1039          * value is always attainable.
1040          */
1041         if (task->signal->oom_adj == OOM_ADJUST_MAX)
1042                 task->signal->oom_score_adj = OOM_SCORE_ADJ_MAX;
1043         else
1044                 task->signal->oom_score_adj = (oom_adjust * OOM_SCORE_ADJ_MAX) /
1045                                                                 -OOM_DISABLE;
1046 err_sighand:
1047         unlock_task_sighand(task, &flags);
1048 err_task_lock:
1049         task_unlock(task);
1050         put_task_struct(task);
1051 out:
1052         return err < 0 ? err : count;
1053 }
1054
1055 static const struct file_operations proc_oom_adjust_operations = {
1056         .read           = oom_adjust_read,
1057         .write          = oom_adjust_write,
1058         .llseek         = generic_file_llseek,
1059 };
1060
1061 static ssize_t oom_score_adj_read(struct file *file, char __user *buf,
1062                                         size_t count, loff_t *ppos)
1063 {
1064         struct task_struct *task = get_proc_task(file->f_path.dentry->d_inode);
1065         char buffer[PROC_NUMBUF];
1066         int oom_score_adj = OOM_SCORE_ADJ_MIN;
1067         unsigned long flags;
1068         size_t len;
1069
1070         if (!task)
1071                 return -ESRCH;
1072         if (lock_task_sighand(task, &flags)) {
1073                 oom_score_adj = task->signal->oom_score_adj;
1074                 unlock_task_sighand(task, &flags);
1075         }
1076         put_task_struct(task);
1077         len = snprintf(buffer, sizeof(buffer), "%d\n", oom_score_adj);
1078         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1079 }
1080
1081 static ssize_t oom_score_adj_write(struct file *file, const char __user *buf,
1082                                         size_t count, loff_t *ppos)
1083 {
1084         struct task_struct *task;
1085         char buffer[PROC_NUMBUF];
1086         unsigned long flags;
1087         int oom_score_adj;
1088         int err;
1089
1090         memset(buffer, 0, sizeof(buffer));
1091         if (count > sizeof(buffer) - 1)
1092                 count = sizeof(buffer) - 1;
1093         if (copy_from_user(buffer, buf, count)) {
1094                 err = -EFAULT;
1095                 goto out;
1096         }
1097
1098         err = kstrtoint(strstrip(buffer), 0, &oom_score_adj);
1099         if (err)
1100                 goto out;
1101         if (oom_score_adj < OOM_SCORE_ADJ_MIN ||
1102                         oom_score_adj > OOM_SCORE_ADJ_MAX) {
1103                 err = -EINVAL;
1104                 goto out;
1105         }
1106
1107         task = get_proc_task(file->f_path.dentry->d_inode);
1108         if (!task) {
1109                 err = -ESRCH;
1110                 goto out;
1111         }
1112
1113         task_lock(task);
1114         if (!task->mm) {
1115                 err = -EINVAL;
1116                 goto err_task_lock;
1117         }
1118
1119         if (!lock_task_sighand(task, &flags)) {
1120                 err = -ESRCH;
1121                 goto err_task_lock;
1122         }
1123
1124         if (oom_score_adj < task->signal->oom_score_adj_min &&
1125                         !capable(CAP_SYS_RESOURCE)) {
1126                 err = -EACCES;
1127                 goto err_sighand;
1128         }
1129
1130         task->signal->oom_score_adj = oom_score_adj;
1131         if (has_capability_noaudit(current, CAP_SYS_RESOURCE))
1132                 task->signal->oom_score_adj_min = oom_score_adj;
1133         /*
1134          * Scale /proc/pid/oom_adj appropriately ensuring that OOM_DISABLE is
1135          * always attainable.
1136          */
1137         if (task->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
1138                 task->signal->oom_adj = OOM_DISABLE;
1139         else
1140                 task->signal->oom_adj = (oom_score_adj * OOM_ADJUST_MAX) /
1141                                                         OOM_SCORE_ADJ_MAX;
1142 err_sighand:
1143         unlock_task_sighand(task, &flags);
1144 err_task_lock:
1145         task_unlock(task);
1146         put_task_struct(task);
1147 out:
1148         return err < 0 ? err : count;
1149 }
1150
1151 static const struct file_operations proc_oom_score_adj_operations = {
1152         .read           = oom_score_adj_read,
1153         .write          = oom_score_adj_write,
1154         .llseek         = default_llseek,
1155 };
1156
1157 #ifdef CONFIG_AUDITSYSCALL
1158 #define TMPBUFLEN 21
1159 static ssize_t proc_loginuid_read(struct file * file, char __user * buf,
1160                                   size_t count, loff_t *ppos)
1161 {
1162         struct inode * inode = file->f_path.dentry->d_inode;
1163         struct task_struct *task = get_proc_task(inode);
1164         ssize_t length;
1165         char tmpbuf[TMPBUFLEN];
1166
1167         if (!task)
1168                 return -ESRCH;
1169         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1170                                 audit_get_loginuid(task));
1171         put_task_struct(task);
1172         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1173 }
1174
1175 static ssize_t proc_loginuid_write(struct file * file, const char __user * buf,
1176                                    size_t count, loff_t *ppos)
1177 {
1178         struct inode * inode = file->f_path.dentry->d_inode;
1179         char *page, *tmp;
1180         ssize_t length;
1181         uid_t loginuid;
1182
1183         if (!capable(CAP_AUDIT_CONTROL))
1184                 return -EPERM;
1185
1186         rcu_read_lock();
1187         if (current != pid_task(proc_pid(inode), PIDTYPE_PID)) {
1188                 rcu_read_unlock();
1189                 return -EPERM;
1190         }
1191         rcu_read_unlock();
1192
1193         if (count >= PAGE_SIZE)
1194                 count = PAGE_SIZE - 1;
1195
1196         if (*ppos != 0) {
1197                 /* No partial writes. */
1198                 return -EINVAL;
1199         }
1200         page = (char*)__get_free_page(GFP_TEMPORARY);
1201         if (!page)
1202                 return -ENOMEM;
1203         length = -EFAULT;
1204         if (copy_from_user(page, buf, count))
1205                 goto out_free_page;
1206
1207         page[count] = '\0';
1208         loginuid = simple_strtoul(page, &tmp, 10);
1209         if (tmp == page) {
1210                 length = -EINVAL;
1211                 goto out_free_page;
1212
1213         }
1214         length = audit_set_loginuid(current, loginuid);
1215         if (likely(length == 0))
1216                 length = count;
1217
1218 out_free_page:
1219         free_page((unsigned long) page);
1220         return length;
1221 }
1222
1223 static const struct file_operations proc_loginuid_operations = {
1224         .read           = proc_loginuid_read,
1225         .write          = proc_loginuid_write,
1226         .llseek         = generic_file_llseek,
1227 };
1228
1229 static ssize_t proc_sessionid_read(struct file * file, char __user * buf,
1230                                   size_t count, loff_t *ppos)
1231 {
1232         struct inode * inode = file->f_path.dentry->d_inode;
1233         struct task_struct *task = get_proc_task(inode);
1234         ssize_t length;
1235         char tmpbuf[TMPBUFLEN];
1236
1237         if (!task)
1238                 return -ESRCH;
1239         length = scnprintf(tmpbuf, TMPBUFLEN, "%u",
1240                                 audit_get_sessionid(task));
1241         put_task_struct(task);
1242         return simple_read_from_buffer(buf, count, ppos, tmpbuf, length);
1243 }
1244
1245 static const struct file_operations proc_sessionid_operations = {
1246         .read           = proc_sessionid_read,
1247         .llseek         = generic_file_llseek,
1248 };
1249 #endif
1250
1251 #ifdef CONFIG_FAULT_INJECTION
1252 static ssize_t proc_fault_inject_read(struct file * file, char __user * buf,
1253                                       size_t count, loff_t *ppos)
1254 {
1255         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
1256         char buffer[PROC_NUMBUF];
1257         size_t len;
1258         int make_it_fail;
1259
1260         if (!task)
1261                 return -ESRCH;
1262         make_it_fail = task->make_it_fail;
1263         put_task_struct(task);
1264
1265         len = snprintf(buffer, sizeof(buffer), "%i\n", make_it_fail);
1266
1267         return simple_read_from_buffer(buf, count, ppos, buffer, len);
1268 }
1269
1270 static ssize_t proc_fault_inject_write(struct file * file,
1271                         const char __user * buf, size_t count, loff_t *ppos)
1272 {
1273         struct task_struct *task;
1274         char buffer[PROC_NUMBUF], *end;
1275         int make_it_fail;
1276
1277         if (!capable(CAP_SYS_RESOURCE))
1278                 return -EPERM;
1279         memset(buffer, 0, sizeof(buffer));
1280         if (count > sizeof(buffer) - 1)
1281                 count = sizeof(buffer) - 1;
1282         if (copy_from_user(buffer, buf, count))
1283                 return -EFAULT;
1284         make_it_fail = simple_strtol(strstrip(buffer), &end, 0);
1285         if (*end)
1286                 return -EINVAL;
1287         task = get_proc_task(file->f_dentry->d_inode);
1288         if (!task)
1289                 return -ESRCH;
1290         task->make_it_fail = make_it_fail;
1291         put_task_struct(task);
1292
1293         return count;
1294 }
1295
1296 static const struct file_operations proc_fault_inject_operations = {
1297         .read           = proc_fault_inject_read,
1298         .write          = proc_fault_inject_write,
1299         .llseek         = generic_file_llseek,
1300 };
1301 #endif
1302
1303
1304 #ifdef CONFIG_SCHED_DEBUG
1305 /*
1306  * Print out various scheduling related per-task fields:
1307  */
1308 static int sched_show(struct seq_file *m, void *v)
1309 {
1310         struct inode *inode = m->private;
1311         struct task_struct *p;
1312
1313         p = get_proc_task(inode);
1314         if (!p)
1315                 return -ESRCH;
1316         proc_sched_show_task(p, m);
1317
1318         put_task_struct(p);
1319
1320         return 0;
1321 }
1322
1323 static ssize_t
1324 sched_write(struct file *file, const char __user *buf,
1325             size_t count, loff_t *offset)
1326 {
1327         struct inode *inode = file->f_path.dentry->d_inode;
1328         struct task_struct *p;
1329
1330         p = get_proc_task(inode);
1331         if (!p)
1332                 return -ESRCH;
1333         proc_sched_set_task(p);
1334
1335         put_task_struct(p);
1336
1337         return count;
1338 }
1339
1340 static int sched_open(struct inode *inode, struct file *filp)
1341 {
1342         return single_open(filp, sched_show, inode);
1343 }
1344
1345 static const struct file_operations proc_pid_sched_operations = {
1346         .open           = sched_open,
1347         .read           = seq_read,
1348         .write          = sched_write,
1349         .llseek         = seq_lseek,
1350         .release        = single_release,
1351 };
1352
1353 #endif
1354
1355 #ifdef CONFIG_SCHED_AUTOGROUP
1356 /*
1357  * Print out autogroup related information:
1358  */
1359 static int sched_autogroup_show(struct seq_file *m, void *v)
1360 {
1361         struct inode *inode = m->private;
1362         struct task_struct *p;
1363
1364         p = get_proc_task(inode);
1365         if (!p)
1366                 return -ESRCH;
1367         proc_sched_autogroup_show_task(p, m);
1368
1369         put_task_struct(p);
1370
1371         return 0;
1372 }
1373
1374 static ssize_t
1375 sched_autogroup_write(struct file *file, const char __user *buf,
1376             size_t count, loff_t *offset)
1377 {
1378         struct inode *inode = file->f_path.dentry->d_inode;
1379         struct task_struct *p;
1380         char buffer[PROC_NUMBUF];
1381         int nice;
1382         int err;
1383
1384         memset(buffer, 0, sizeof(buffer));
1385         if (count > sizeof(buffer) - 1)
1386                 count = sizeof(buffer) - 1;
1387         if (copy_from_user(buffer, buf, count))
1388                 return -EFAULT;
1389
1390         err = kstrtoint(strstrip(buffer), 0, &nice);
1391         if (err < 0)
1392                 return err;
1393
1394         p = get_proc_task(inode);
1395         if (!p)
1396                 return -ESRCH;
1397
1398         err = nice;
1399         err = proc_sched_autogroup_set_nice(p, &err);
1400         if (err)
1401                 count = err;
1402
1403         put_task_struct(p);
1404
1405         return count;
1406 }
1407
1408 static int sched_autogroup_open(struct inode *inode, struct file *filp)
1409 {
1410         int ret;
1411
1412         ret = single_open(filp, sched_autogroup_show, NULL);
1413         if (!ret) {
1414                 struct seq_file *m = filp->private_data;
1415
1416                 m->private = inode;
1417         }
1418         return ret;
1419 }
1420
1421 static const struct file_operations proc_pid_sched_autogroup_operations = {
1422         .open           = sched_autogroup_open,
1423         .read           = seq_read,
1424         .write          = sched_autogroup_write,
1425         .llseek         = seq_lseek,
1426         .release        = single_release,
1427 };
1428
1429 #endif /* CONFIG_SCHED_AUTOGROUP */
1430
1431 static ssize_t comm_write(struct file *file, const char __user *buf,
1432                                 size_t count, loff_t *offset)
1433 {
1434         struct inode *inode = file->f_path.dentry->d_inode;
1435         struct task_struct *p;
1436         char buffer[TASK_COMM_LEN];
1437
1438         memset(buffer, 0, sizeof(buffer));
1439         if (count > sizeof(buffer) - 1)
1440                 count = sizeof(buffer) - 1;
1441         if (copy_from_user(buffer, buf, count))
1442                 return -EFAULT;
1443
1444         p = get_proc_task(inode);
1445         if (!p)
1446                 return -ESRCH;
1447
1448         if (same_thread_group(current, p))
1449                 set_task_comm(p, buffer);
1450         else
1451                 count = -EINVAL;
1452
1453         put_task_struct(p);
1454
1455         return count;
1456 }
1457
1458 static int comm_show(struct seq_file *m, void *v)
1459 {
1460         struct inode *inode = m->private;
1461         struct task_struct *p;
1462
1463         p = get_proc_task(inode);
1464         if (!p)
1465                 return -ESRCH;
1466
1467         task_lock(p);
1468         seq_printf(m, "%s\n", p->comm);
1469         task_unlock(p);
1470
1471         put_task_struct(p);
1472
1473         return 0;
1474 }
1475
1476 static int comm_open(struct inode *inode, struct file *filp)
1477 {
1478         return single_open(filp, comm_show, inode);
1479 }
1480
1481 static const struct file_operations proc_pid_set_comm_operations = {
1482         .open           = comm_open,
1483         .read           = seq_read,
1484         .write          = comm_write,
1485         .llseek         = seq_lseek,
1486         .release        = single_release,
1487 };
1488
1489 static int proc_exe_link(struct inode *inode, struct path *exe_path)
1490 {
1491         struct task_struct *task;
1492         struct mm_struct *mm;
1493         struct file *exe_file;
1494
1495         task = get_proc_task(inode);
1496         if (!task)
1497                 return -ENOENT;
1498         mm = get_task_mm(task);
1499         put_task_struct(task);
1500         if (!mm)
1501                 return -ENOENT;
1502         exe_file = get_mm_exe_file(mm);
1503         mmput(mm);
1504         if (exe_file) {
1505                 *exe_path = exe_file->f_path;
1506                 path_get(&exe_file->f_path);
1507                 fput(exe_file);
1508                 return 0;
1509         } else
1510                 return -ENOENT;
1511 }
1512
1513 static void *proc_pid_follow_link(struct dentry *dentry, struct nameidata *nd)
1514 {
1515         struct inode *inode = dentry->d_inode;
1516         int error = -EACCES;
1517
1518         /* We don't need a base pointer in the /proc filesystem */
1519         path_put(&nd->path);
1520
1521         /* Are we allowed to snoop on the tasks file descriptors? */
1522         if (!proc_fd_access_allowed(inode))
1523                 goto out;
1524
1525         error = PROC_I(inode)->op.proc_get_link(inode, &nd->path);
1526 out:
1527         return ERR_PTR(error);
1528 }
1529
1530 static int do_proc_readlink(struct path *path, char __user *buffer, int buflen)
1531 {
1532         char *tmp = (char*)__get_free_page(GFP_TEMPORARY);
1533         char *pathname;
1534         int len;
1535
1536         if (!tmp)
1537                 return -ENOMEM;
1538
1539         pathname = d_path(path, tmp, PAGE_SIZE);
1540         len = PTR_ERR(pathname);
1541         if (IS_ERR(pathname))
1542                 goto out;
1543         len = tmp + PAGE_SIZE - 1 - pathname;
1544
1545         if (len > buflen)
1546                 len = buflen;
1547         if (copy_to_user(buffer, pathname, len))
1548                 len = -EFAULT;
1549  out:
1550         free_page((unsigned long)tmp);
1551         return len;
1552 }
1553
1554 static int proc_pid_readlink(struct dentry * dentry, char __user * buffer, int buflen)
1555 {
1556         int error = -EACCES;
1557         struct inode *inode = dentry->d_inode;
1558         struct path path;
1559
1560         /* Are we allowed to snoop on the tasks file descriptors? */
1561         if (!proc_fd_access_allowed(inode))
1562                 goto out;
1563
1564         error = PROC_I(inode)->op.proc_get_link(inode, &path);
1565         if (error)
1566                 goto out;
1567
1568         error = do_proc_readlink(&path, buffer, buflen);
1569         path_put(&path);
1570 out:
1571         return error;
1572 }
1573
1574 static const struct inode_operations proc_pid_link_inode_operations = {
1575         .readlink       = proc_pid_readlink,
1576         .follow_link    = proc_pid_follow_link,
1577         .setattr        = proc_setattr,
1578 };
1579
1580
1581 /* building an inode */
1582
1583 static int task_dumpable(struct task_struct *task)
1584 {
1585         int dumpable = 0;
1586         struct mm_struct *mm;
1587
1588         task_lock(task);
1589         mm = task->mm;
1590         if (mm)
1591                 dumpable = get_dumpable(mm);
1592         task_unlock(task);
1593         if(dumpable == 1)
1594                 return 1;
1595         return 0;
1596 }
1597
1598 struct inode *proc_pid_make_inode(struct super_block * sb, struct task_struct *task)
1599 {
1600         struct inode * inode;
1601         struct proc_inode *ei;
1602         const struct cred *cred;
1603
1604         /* We need a new inode */
1605
1606         inode = new_inode(sb);
1607         if (!inode)
1608                 goto out;
1609
1610         /* Common stuff */
1611         ei = PROC_I(inode);
1612         inode->i_ino = get_next_ino();
1613         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1614         inode->i_op = &proc_def_inode_operations;
1615
1616         /*
1617          * grab the reference to task.
1618          */
1619         ei->pid = get_task_pid(task, PIDTYPE_PID);
1620         if (!ei->pid)
1621                 goto out_unlock;
1622
1623         if (task_dumpable(task)) {
1624                 rcu_read_lock();
1625                 cred = __task_cred(task);
1626                 inode->i_uid = cred->euid;
1627                 inode->i_gid = cred->egid;
1628                 rcu_read_unlock();
1629         }
1630         security_task_to_inode(task, inode);
1631
1632 out:
1633         return inode;
1634
1635 out_unlock:
1636         iput(inode);
1637         return NULL;
1638 }
1639
1640 int pid_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
1641 {
1642         struct inode *inode = dentry->d_inode;
1643         struct task_struct *task;
1644         const struct cred *cred;
1645
1646         generic_fillattr(inode, stat);
1647
1648         rcu_read_lock();
1649         stat->uid = 0;
1650         stat->gid = 0;
1651         task = pid_task(proc_pid(inode), PIDTYPE_PID);
1652         if (task) {
1653                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1654                     task_dumpable(task)) {
1655                         cred = __task_cred(task);
1656                         stat->uid = cred->euid;
1657                         stat->gid = cred->egid;
1658                 }
1659         }
1660         rcu_read_unlock();
1661         return 0;
1662 }
1663
1664 /* dentry stuff */
1665
1666 /*
1667  *      Exceptional case: normally we are not allowed to unhash a busy
1668  * directory. In this case, however, we can do it - no aliasing problems
1669  * due to the way we treat inodes.
1670  *
1671  * Rewrite the inode's ownerships here because the owning task may have
1672  * performed a setuid(), etc.
1673  *
1674  * Before the /proc/pid/status file was created the only way to read
1675  * the effective uid of a /process was to stat /proc/pid.  Reading
1676  * /proc/pid/status is slow enough that procps and other packages
1677  * kept stating /proc/pid.  To keep the rules in /proc simple I have
1678  * made this apply to all per process world readable and executable
1679  * directories.
1680  */
1681 int pid_revalidate(struct dentry *dentry, struct nameidata *nd)
1682 {
1683         struct inode *inode;
1684         struct task_struct *task;
1685         const struct cred *cred;
1686
1687         if (nd && nd->flags & LOOKUP_RCU)
1688                 return -ECHILD;
1689
1690         inode = dentry->d_inode;
1691         task = get_proc_task(inode);
1692
1693         if (task) {
1694                 if ((inode->i_mode == (S_IFDIR|S_IRUGO|S_IXUGO)) ||
1695                     task_dumpable(task)) {
1696                         rcu_read_lock();
1697                         cred = __task_cred(task);
1698                         inode->i_uid = cred->euid;
1699                         inode->i_gid = cred->egid;
1700                         rcu_read_unlock();
1701                 } else {
1702                         inode->i_uid = 0;
1703                         inode->i_gid = 0;
1704                 }
1705                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1706                 security_task_to_inode(task, inode);
1707                 put_task_struct(task);
1708                 return 1;
1709         }
1710         d_drop(dentry);
1711         return 0;
1712 }
1713
1714 static int pid_delete_dentry(const struct dentry * dentry)
1715 {
1716         /* Is the task we represent dead?
1717          * If so, then don't put the dentry on the lru list,
1718          * kill it immediately.
1719          */
1720         return !proc_pid(dentry->d_inode)->tasks[PIDTYPE_PID].first;
1721 }
1722
1723 const struct dentry_operations pid_dentry_operations =
1724 {
1725         .d_revalidate   = pid_revalidate,
1726         .d_delete       = pid_delete_dentry,
1727 };
1728
1729 /* Lookups */
1730
1731 /*
1732  * Fill a directory entry.
1733  *
1734  * If possible create the dcache entry and derive our inode number and
1735  * file type from dcache entry.
1736  *
1737  * Since all of the proc inode numbers are dynamically generated, the inode
1738  * numbers do not exist until the inode is cache.  This means creating the
1739  * the dcache entry in readdir is necessary to keep the inode numbers
1740  * reported by readdir in sync with the inode numbers reported
1741  * by stat.
1742  */
1743 int proc_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
1744         const char *name, int len,
1745         instantiate_t instantiate, struct task_struct *task, const void *ptr)
1746 {
1747         struct dentry *child, *dir = filp->f_path.dentry;
1748         struct inode *inode;
1749         struct qstr qname;
1750         ino_t ino = 0;
1751         unsigned type = DT_UNKNOWN;
1752
1753         qname.name = name;
1754         qname.len  = len;
1755         qname.hash = full_name_hash(name, len);
1756
1757         child = d_lookup(dir, &qname);
1758         if (!child) {
1759                 struct dentry *new;
1760                 new = d_alloc(dir, &qname);
1761                 if (new) {
1762                         child = instantiate(dir->d_inode, new, task, ptr);
1763                         if (child)
1764                                 dput(new);
1765                         else
1766                                 child = new;
1767                 }
1768         }
1769         if (!child || IS_ERR(child) || !child->d_inode)
1770                 goto end_instantiate;
1771         inode = child->d_inode;
1772         if (inode) {
1773                 ino = inode->i_ino;
1774                 type = inode->i_mode >> 12;
1775         }
1776         dput(child);
1777 end_instantiate:
1778         if (!ino)
1779                 ino = find_inode_number(dir, &qname);
1780         if (!ino)
1781                 ino = 1;
1782         return filldir(dirent, name, len, filp->f_pos, ino, type);
1783 }
1784
1785 static unsigned name_to_int(struct dentry *dentry)
1786 {
1787         const char *name = dentry->d_name.name;
1788         int len = dentry->d_name.len;
1789         unsigned n = 0;
1790
1791         if (len > 1 && *name == '0')
1792                 goto out;
1793         while (len-- > 0) {
1794                 unsigned c = *name++ - '0';
1795                 if (c > 9)
1796                         goto out;
1797                 if (n >= (~0U-9)/10)
1798                         goto out;
1799                 n *= 10;
1800                 n += c;
1801         }
1802         return n;
1803 out:
1804         return ~0U;
1805 }
1806
1807 #define PROC_FDINFO_MAX 64
1808
1809 static int proc_fd_info(struct inode *inode, struct path *path, char *info)
1810 {
1811         struct task_struct *task = get_proc_task(inode);
1812         struct files_struct *files = NULL;
1813         struct file *file;
1814         int fd = proc_fd(inode);
1815
1816         if (task) {
1817                 files = get_files_struct(task);
1818                 put_task_struct(task);
1819         }
1820         if (files) {
1821                 /*
1822                  * We are not taking a ref to the file structure, so we must
1823                  * hold ->file_lock.
1824                  */
1825                 spin_lock(&files->file_lock);
1826                 file = fcheck_files(files, fd);
1827                 if (file) {
1828                         unsigned int f_flags;
1829                         struct fdtable *fdt;
1830
1831                         fdt = files_fdtable(files);
1832                         f_flags = file->f_flags & ~O_CLOEXEC;
1833                         if (FD_ISSET(fd, fdt->close_on_exec))
1834                                 f_flags |= O_CLOEXEC;
1835
1836                         if (path) {
1837                                 *path = file->f_path;
1838                                 path_get(&file->f_path);
1839                         }
1840                         if (info)
1841                                 snprintf(info, PROC_FDINFO_MAX,
1842                                          "pos:\t%lli\n"
1843                                          "flags:\t0%o\n",
1844                                          (long long) file->f_pos,
1845                                          f_flags);
1846                         spin_unlock(&files->file_lock);
1847                         put_files_struct(files);
1848                         return 0;
1849                 }
1850                 spin_unlock(&files->file_lock);
1851                 put_files_struct(files);
1852         }
1853         return -ENOENT;
1854 }
1855
1856 static int proc_fd_link(struct inode *inode, struct path *path)
1857 {
1858         return proc_fd_info(inode, path, NULL);
1859 }
1860
1861 static int tid_fd_revalidate(struct dentry *dentry, struct nameidata *nd)
1862 {
1863         struct inode *inode;
1864         struct task_struct *task;
1865         int fd;
1866         struct files_struct *files;
1867         const struct cred *cred;
1868
1869         if (nd && nd->flags & LOOKUP_RCU)
1870                 return -ECHILD;
1871
1872         inode = dentry->d_inode;
1873         task = get_proc_task(inode);
1874         fd = proc_fd(inode);
1875
1876         if (task) {
1877                 files = get_files_struct(task);
1878                 if (files) {
1879                         rcu_read_lock();
1880                         if (fcheck_files(files, fd)) {
1881                                 rcu_read_unlock();
1882                                 put_files_struct(files);
1883                                 if (task_dumpable(task)) {
1884                                         rcu_read_lock();
1885                                         cred = __task_cred(task);
1886                                         inode->i_uid = cred->euid;
1887                                         inode->i_gid = cred->egid;
1888                                         rcu_read_unlock();
1889                                 } else {
1890                                         inode->i_uid = 0;
1891                                         inode->i_gid = 0;
1892                                 }
1893                                 inode->i_mode &= ~(S_ISUID | S_ISGID);
1894                                 security_task_to_inode(task, inode);
1895                                 put_task_struct(task);
1896                                 return 1;
1897                         }
1898                         rcu_read_unlock();
1899                         put_files_struct(files);
1900                 }
1901                 put_task_struct(task);
1902         }
1903         d_drop(dentry);
1904         return 0;
1905 }
1906
1907 static const struct dentry_operations tid_fd_dentry_operations =
1908 {
1909         .d_revalidate   = tid_fd_revalidate,
1910         .d_delete       = pid_delete_dentry,
1911 };
1912
1913 static struct dentry *proc_fd_instantiate(struct inode *dir,
1914         struct dentry *dentry, struct task_struct *task, const void *ptr)
1915 {
1916         unsigned fd = *(const unsigned *)ptr;
1917         struct file *file;
1918         struct files_struct *files;
1919         struct inode *inode;
1920         struct proc_inode *ei;
1921         struct dentry *error = ERR_PTR(-ENOENT);
1922
1923         inode = proc_pid_make_inode(dir->i_sb, task);
1924         if (!inode)
1925                 goto out;
1926         ei = PROC_I(inode);
1927         ei->fd = fd;
1928         files = get_files_struct(task);
1929         if (!files)
1930                 goto out_iput;
1931         inode->i_mode = S_IFLNK;
1932
1933         /*
1934          * We are not taking a ref to the file structure, so we must
1935          * hold ->file_lock.
1936          */
1937         spin_lock(&files->file_lock);
1938         file = fcheck_files(files, fd);
1939         if (!file)
1940                 goto out_unlock;
1941         if (file->f_mode & FMODE_READ)
1942                 inode->i_mode |= S_IRUSR | S_IXUSR;
1943         if (file->f_mode & FMODE_WRITE)
1944                 inode->i_mode |= S_IWUSR | S_IXUSR;
1945         spin_unlock(&files->file_lock);
1946         put_files_struct(files);
1947
1948         inode->i_op = &proc_pid_link_inode_operations;
1949         inode->i_size = 64;
1950         ei->op.proc_get_link = proc_fd_link;
1951         d_set_d_op(dentry, &tid_fd_dentry_operations);
1952         d_add(dentry, inode);
1953         /* Close the race of the process dying before we return the dentry */
1954         if (tid_fd_revalidate(dentry, NULL))
1955                 error = NULL;
1956
1957  out:
1958         return error;
1959 out_unlock:
1960         spin_unlock(&files->file_lock);
1961         put_files_struct(files);
1962 out_iput:
1963         iput(inode);
1964         goto out;
1965 }
1966
1967 static struct dentry *proc_lookupfd_common(struct inode *dir,
1968                                            struct dentry *dentry,
1969                                            instantiate_t instantiate)
1970 {
1971         struct task_struct *task = get_proc_task(dir);
1972         unsigned fd = name_to_int(dentry);
1973         struct dentry *result = ERR_PTR(-ENOENT);
1974
1975         if (!task)
1976                 goto out_no_task;
1977         if (fd == ~0U)
1978                 goto out;
1979
1980         result = instantiate(dir, dentry, task, &fd);
1981 out:
1982         put_task_struct(task);
1983 out_no_task:
1984         return result;
1985 }
1986
1987 static int proc_readfd_common(struct file * filp, void * dirent,
1988                               filldir_t filldir, instantiate_t instantiate)
1989 {
1990         struct dentry *dentry = filp->f_path.dentry;
1991         struct inode *inode = dentry->d_inode;
1992         struct task_struct *p = get_proc_task(inode);
1993         unsigned int fd, ino;
1994         int retval;
1995         struct files_struct * files;
1996
1997         retval = -ENOENT;
1998         if (!p)
1999                 goto out_no_task;
2000         retval = 0;
2001
2002         fd = filp->f_pos;
2003         switch (fd) {
2004                 case 0:
2005                         if (filldir(dirent, ".", 1, 0, inode->i_ino, DT_DIR) < 0)
2006                                 goto out;
2007                         filp->f_pos++;
2008                 case 1:
2009                         ino = parent_ino(dentry);
2010                         if (filldir(dirent, "..", 2, 1, ino, DT_DIR) < 0)
2011                                 goto out;
2012                         filp->f_pos++;
2013                 default:
2014                         files = get_files_struct(p);
2015                         if (!files)
2016                                 goto out;
2017                         rcu_read_lock();
2018                         for (fd = filp->f_pos-2;
2019                              fd < files_fdtable(files)->max_fds;
2020                              fd++, filp->f_pos++) {
2021                                 char name[PROC_NUMBUF];
2022                                 int len;
2023
2024                                 if (!fcheck_files(files, fd))
2025                                         continue;
2026                                 rcu_read_unlock();
2027
2028                                 len = snprintf(name, sizeof(name), "%d", fd);
2029                                 if (proc_fill_cache(filp, dirent, filldir,
2030                                                     name, len, instantiate,
2031                                                     p, &fd) < 0) {
2032                                         rcu_read_lock();
2033                                         break;
2034                                 }
2035                                 rcu_read_lock();
2036                         }
2037                         rcu_read_unlock();
2038                         put_files_struct(files);
2039         }
2040 out:
2041         put_task_struct(p);
2042 out_no_task:
2043         return retval;
2044 }
2045
2046 static struct dentry *proc_lookupfd(struct inode *dir, struct dentry *dentry,
2047                                     struct nameidata *nd)
2048 {
2049         return proc_lookupfd_common(dir, dentry, proc_fd_instantiate);
2050 }
2051
2052 static int proc_readfd(struct file *filp, void *dirent, filldir_t filldir)
2053 {
2054         return proc_readfd_common(filp, dirent, filldir, proc_fd_instantiate);
2055 }
2056
2057 static ssize_t proc_fdinfo_read(struct file *file, char __user *buf,
2058                                       size_t len, loff_t *ppos)
2059 {
2060         char tmp[PROC_FDINFO_MAX];
2061         int err = proc_fd_info(file->f_path.dentry->d_inode, NULL, tmp);
2062         if (!err)
2063                 err = simple_read_from_buffer(buf, len, ppos, tmp, strlen(tmp));
2064         return err;
2065 }
2066
2067 static const struct file_operations proc_fdinfo_file_operations = {
2068         .open           = nonseekable_open,
2069         .read           = proc_fdinfo_read,
2070         .llseek         = no_llseek,
2071 };
2072
2073 static const struct file_operations proc_fd_operations = {
2074         .read           = generic_read_dir,
2075         .readdir        = proc_readfd,
2076         .llseek         = default_llseek,
2077 };
2078
2079 /*
2080  * /proc/pid/fd needs a special permission handler so that a process can still
2081  * access /proc/self/fd after it has executed a setuid().
2082  */
2083 static int proc_fd_permission(struct inode *inode, int mask)
2084 {
2085         int rv = generic_permission(inode, mask);
2086         if (rv == 0)
2087                 return 0;
2088         if (task_pid(current) == proc_pid(inode))
2089                 rv = 0;
2090         return rv;
2091 }
2092
2093 /*
2094  * proc directories can do almost nothing..
2095  */
2096 static const struct inode_operations proc_fd_inode_operations = {
2097         .lookup         = proc_lookupfd,
2098         .permission     = proc_fd_permission,
2099         .setattr        = proc_setattr,
2100 };
2101
2102 static struct dentry *proc_fdinfo_instantiate(struct inode *dir,
2103         struct dentry *dentry, struct task_struct *task, const void *ptr)
2104 {
2105         unsigned fd = *(unsigned *)ptr;
2106         struct inode *inode;
2107         struct proc_inode *ei;
2108         struct dentry *error = ERR_PTR(-ENOENT);
2109
2110         inode = proc_pid_make_inode(dir->i_sb, task);
2111         if (!inode)
2112                 goto out;
2113         ei = PROC_I(inode);
2114         ei->fd = fd;
2115         inode->i_mode = S_IFREG | S_IRUSR;
2116         inode->i_fop = &proc_fdinfo_file_operations;
2117         d_set_d_op(dentry, &tid_fd_dentry_operations);
2118         d_add(dentry, inode);
2119         /* Close the race of the process dying before we return the dentry */
2120         if (tid_fd_revalidate(dentry, NULL))
2121                 error = NULL;
2122
2123  out:
2124         return error;
2125 }
2126
2127 static struct dentry *proc_lookupfdinfo(struct inode *dir,
2128                                         struct dentry *dentry,
2129                                         struct nameidata *nd)
2130 {
2131         return proc_lookupfd_common(dir, dentry, proc_fdinfo_instantiate);
2132 }
2133
2134 static int proc_readfdinfo(struct file *filp, void *dirent, filldir_t filldir)
2135 {
2136         return proc_readfd_common(filp, dirent, filldir,
2137                                   proc_fdinfo_instantiate);
2138 }
2139
2140 static const struct file_operations proc_fdinfo_operations = {
2141         .read           = generic_read_dir,
2142         .readdir        = proc_readfdinfo,
2143         .llseek         = default_llseek,
2144 };
2145
2146 /*
2147  * proc directories can do almost nothing..
2148  */
2149 static const struct inode_operations proc_fdinfo_inode_operations = {
2150         .lookup         = proc_lookupfdinfo,
2151         .setattr        = proc_setattr,
2152 };
2153
2154
2155 static struct dentry *proc_pident_instantiate(struct inode *dir,
2156         struct dentry *dentry, struct task_struct *task, const void *ptr)
2157 {
2158         const struct pid_entry *p = ptr;
2159         struct inode *inode;
2160         struct proc_inode *ei;
2161         struct dentry *error = ERR_PTR(-ENOENT);
2162
2163         inode = proc_pid_make_inode(dir->i_sb, task);
2164         if (!inode)
2165                 goto out;
2166
2167         ei = PROC_I(inode);
2168         inode->i_mode = p->mode;
2169         if (S_ISDIR(inode->i_mode))
2170                 set_nlink(inode, 2);    /* Use getattr to fix if necessary */
2171         if (p->iop)
2172                 inode->i_op = p->iop;
2173         if (p->fop)
2174                 inode->i_fop = p->fop;
2175         ei->op = p->op;
2176         d_set_d_op(dentry, &pid_dentry_operations);
2177         d_add(dentry, inode);
2178         /* Close the race of the process dying before we return the dentry */
2179         if (pid_revalidate(dentry, NULL))
2180                 error = NULL;
2181 out:
2182         return error;
2183 }
2184
2185 static struct dentry *proc_pident_lookup(struct inode *dir,
2186                                          struct dentry *dentry,
2187                                          const struct pid_entry *ents,
2188                                          unsigned int nents)
2189 {
2190         struct dentry *error;
2191         struct task_struct *task = get_proc_task(dir);
2192         const struct pid_entry *p, *last;
2193
2194         error = ERR_PTR(-ENOENT);
2195
2196         if (!task)
2197                 goto out_no_task;
2198
2199         /*
2200          * Yes, it does not scale. And it should not. Don't add
2201          * new entries into /proc/<tgid>/ without very good reasons.
2202          */
2203         last = &ents[nents - 1];
2204         for (p = ents; p <= last; p++) {
2205                 if (p->len != dentry->d_name.len)
2206                         continue;
2207                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2208                         break;
2209         }
2210         if (p > last)
2211                 goto out;
2212
2213         error = proc_pident_instantiate(dir, dentry, task, p);
2214 out:
2215         put_task_struct(task);
2216 out_no_task:
2217         return error;
2218 }
2219
2220 static int proc_pident_fill_cache(struct file *filp, void *dirent,
2221         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2222 {
2223         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2224                                 proc_pident_instantiate, task, p);
2225 }
2226
2227 static int proc_pident_readdir(struct file *filp,
2228                 void *dirent, filldir_t filldir,
2229                 const struct pid_entry *ents, unsigned int nents)
2230 {
2231         int i;
2232         struct dentry *dentry = filp->f_path.dentry;
2233         struct inode *inode = dentry->d_inode;
2234         struct task_struct *task = get_proc_task(inode);
2235         const struct pid_entry *p, *last;
2236         ino_t ino;
2237         int ret;
2238
2239         ret = -ENOENT;
2240         if (!task)
2241                 goto out_no_task;
2242
2243         ret = 0;
2244         i = filp->f_pos;
2245         switch (i) {
2246         case 0:
2247                 ino = inode->i_ino;
2248                 if (filldir(dirent, ".", 1, i, ino, DT_DIR) < 0)
2249                         goto out;
2250                 i++;
2251                 filp->f_pos++;
2252                 /* fall through */
2253         case 1:
2254                 ino = parent_ino(dentry);
2255                 if (filldir(dirent, "..", 2, i, ino, DT_DIR) < 0)
2256                         goto out;
2257                 i++;
2258                 filp->f_pos++;
2259                 /* fall through */
2260         default:
2261                 i -= 2;
2262                 if (i >= nents) {
2263                         ret = 1;
2264                         goto out;
2265                 }
2266                 p = ents + i;
2267                 last = &ents[nents - 1];
2268                 while (p <= last) {
2269                         if (proc_pident_fill_cache(filp, dirent, filldir, task, p) < 0)
2270                                 goto out;
2271                         filp->f_pos++;
2272                         p++;
2273                 }
2274         }
2275
2276         ret = 1;
2277 out:
2278         put_task_struct(task);
2279 out_no_task:
2280         return ret;
2281 }
2282
2283 #ifdef CONFIG_SECURITY
2284 static ssize_t proc_pid_attr_read(struct file * file, char __user * buf,
2285                                   size_t count, loff_t *ppos)
2286 {
2287         struct inode * inode = file->f_path.dentry->d_inode;
2288         char *p = NULL;
2289         ssize_t length;
2290         struct task_struct *task = get_proc_task(inode);
2291
2292         if (!task)
2293                 return -ESRCH;
2294
2295         length = security_getprocattr(task,
2296                                       (char*)file->f_path.dentry->d_name.name,
2297                                       &p);
2298         put_task_struct(task);
2299         if (length > 0)
2300                 length = simple_read_from_buffer(buf, count, ppos, p, length);
2301         kfree(p);
2302         return length;
2303 }
2304
2305 static ssize_t proc_pid_attr_write(struct file * file, const char __user * buf,
2306                                    size_t count, loff_t *ppos)
2307 {
2308         struct inode * inode = file->f_path.dentry->d_inode;
2309         char *page;
2310         ssize_t length;
2311         struct task_struct *task = get_proc_task(inode);
2312
2313         length = -ESRCH;
2314         if (!task)
2315                 goto out_no_task;
2316         if (count > PAGE_SIZE)
2317                 count = PAGE_SIZE;
2318
2319         /* No partial writes. */
2320         length = -EINVAL;
2321         if (*ppos != 0)
2322                 goto out;
2323
2324         length = -ENOMEM;
2325         page = (char*)__get_free_page(GFP_TEMPORARY);
2326         if (!page)
2327                 goto out;
2328
2329         length = -EFAULT;
2330         if (copy_from_user(page, buf, count))
2331                 goto out_free;
2332
2333         /* Guard against adverse ptrace interaction */
2334         length = mutex_lock_interruptible(&task->signal->cred_guard_mutex);
2335         if (length < 0)
2336                 goto out_free;
2337
2338         length = security_setprocattr(task,
2339                                       (char*)file->f_path.dentry->d_name.name,
2340                                       (void*)page, count);
2341         mutex_unlock(&task->signal->cred_guard_mutex);
2342 out_free:
2343         free_page((unsigned long) page);
2344 out:
2345         put_task_struct(task);
2346 out_no_task:
2347         return length;
2348 }
2349
2350 static const struct file_operations proc_pid_attr_operations = {
2351         .read           = proc_pid_attr_read,
2352         .write          = proc_pid_attr_write,
2353         .llseek         = generic_file_llseek,
2354 };
2355
2356 static const struct pid_entry attr_dir_stuff[] = {
2357         REG("current",    S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2358         REG("prev",       S_IRUGO,         proc_pid_attr_operations),
2359         REG("exec",       S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2360         REG("fscreate",   S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2361         REG("keycreate",  S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2362         REG("sockcreate", S_IRUGO|S_IWUGO, proc_pid_attr_operations),
2363 };
2364
2365 static int proc_attr_dir_readdir(struct file * filp,
2366                              void * dirent, filldir_t filldir)
2367 {
2368         return proc_pident_readdir(filp,dirent,filldir,
2369                                    attr_dir_stuff,ARRAY_SIZE(attr_dir_stuff));
2370 }
2371
2372 static const struct file_operations proc_attr_dir_operations = {
2373         .read           = generic_read_dir,
2374         .readdir        = proc_attr_dir_readdir,
2375         .llseek         = default_llseek,
2376 };
2377
2378 static struct dentry *proc_attr_dir_lookup(struct inode *dir,
2379                                 struct dentry *dentry, struct nameidata *nd)
2380 {
2381         return proc_pident_lookup(dir, dentry,
2382                                   attr_dir_stuff, ARRAY_SIZE(attr_dir_stuff));
2383 }
2384
2385 static const struct inode_operations proc_attr_dir_inode_operations = {
2386         .lookup         = proc_attr_dir_lookup,
2387         .getattr        = pid_getattr,
2388         .setattr        = proc_setattr,
2389 };
2390
2391 #endif
2392
2393 #ifdef CONFIG_ELF_CORE
2394 static ssize_t proc_coredump_filter_read(struct file *file, char __user *buf,
2395                                          size_t count, loff_t *ppos)
2396 {
2397         struct task_struct *task = get_proc_task(file->f_dentry->d_inode);
2398         struct mm_struct *mm;
2399         char buffer[PROC_NUMBUF];
2400         size_t len;
2401         int ret;
2402
2403         if (!task)
2404                 return -ESRCH;
2405
2406         ret = 0;
2407         mm = get_task_mm(task);
2408         if (mm) {
2409                 len = snprintf(buffer, sizeof(buffer), "%08lx\n",
2410                                ((mm->flags & MMF_DUMP_FILTER_MASK) >>
2411                                 MMF_DUMP_FILTER_SHIFT));
2412                 mmput(mm);
2413                 ret = simple_read_from_buffer(buf, count, ppos, buffer, len);
2414         }
2415
2416         put_task_struct(task);
2417
2418         return ret;
2419 }
2420
2421 static ssize_t proc_coredump_filter_write(struct file *file,
2422                                           const char __user *buf,
2423                                           size_t count,
2424                                           loff_t *ppos)
2425 {
2426         struct task_struct *task;
2427         struct mm_struct *mm;
2428         char buffer[PROC_NUMBUF], *end;
2429         unsigned int val;
2430         int ret;
2431         int i;
2432         unsigned long mask;
2433
2434         ret = -EFAULT;
2435         memset(buffer, 0, sizeof(buffer));
2436         if (count > sizeof(buffer) - 1)
2437                 count = sizeof(buffer) - 1;
2438         if (copy_from_user(buffer, buf, count))
2439                 goto out_no_task;
2440
2441         ret = -EINVAL;
2442         val = (unsigned int)simple_strtoul(buffer, &end, 0);
2443         if (*end == '\n')
2444                 end++;
2445         if (end - buffer == 0)
2446                 goto out_no_task;
2447
2448         ret = -ESRCH;
2449         task = get_proc_task(file->f_dentry->d_inode);
2450         if (!task)
2451                 goto out_no_task;
2452
2453         ret = end - buffer;
2454         mm = get_task_mm(task);
2455         if (!mm)
2456                 goto out_no_mm;
2457
2458         for (i = 0, mask = 1; i < MMF_DUMP_FILTER_BITS; i++, mask <<= 1) {
2459                 if (val & mask)
2460                         set_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2461                 else
2462                         clear_bit(i + MMF_DUMP_FILTER_SHIFT, &mm->flags);
2463         }
2464
2465         mmput(mm);
2466  out_no_mm:
2467         put_task_struct(task);
2468  out_no_task:
2469         return ret;
2470 }
2471
2472 static const struct file_operations proc_coredump_filter_operations = {
2473         .read           = proc_coredump_filter_read,
2474         .write          = proc_coredump_filter_write,
2475         .llseek         = generic_file_llseek,
2476 };
2477 #endif
2478
2479 /*
2480  * /proc/self:
2481  */
2482 static int proc_self_readlink(struct dentry *dentry, char __user *buffer,
2483                               int buflen)
2484 {
2485         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2486         pid_t tgid = task_tgid_nr_ns(current, ns);
2487         char tmp[PROC_NUMBUF];
2488         if (!tgid)
2489                 return -ENOENT;
2490         sprintf(tmp, "%d", tgid);
2491         return vfs_readlink(dentry,buffer,buflen,tmp);
2492 }
2493
2494 static void *proc_self_follow_link(struct dentry *dentry, struct nameidata *nd)
2495 {
2496         struct pid_namespace *ns = dentry->d_sb->s_fs_info;
2497         pid_t tgid = task_tgid_nr_ns(current, ns);
2498         char *name = ERR_PTR(-ENOENT);
2499         if (tgid) {
2500                 name = __getname();
2501                 if (!name)
2502                         name = ERR_PTR(-ENOMEM);
2503                 else
2504                         sprintf(name, "%d", tgid);
2505         }
2506         nd_set_link(nd, name);
2507         return NULL;
2508 }
2509
2510 static void proc_self_put_link(struct dentry *dentry, struct nameidata *nd,
2511                                 void *cookie)
2512 {
2513         char *s = nd_get_link(nd);
2514         if (!IS_ERR(s))
2515                 __putname(s);
2516 }
2517
2518 static const struct inode_operations proc_self_inode_operations = {
2519         .readlink       = proc_self_readlink,
2520         .follow_link    = proc_self_follow_link,
2521         .put_link       = proc_self_put_link,
2522 };
2523
2524 /*
2525  * proc base
2526  *
2527  * These are the directory entries in the root directory of /proc
2528  * that properly belong to the /proc filesystem, as they describe
2529  * describe something that is process related.
2530  */
2531 static const struct pid_entry proc_base_stuff[] = {
2532         NOD("self", S_IFLNK|S_IRWXUGO,
2533                 &proc_self_inode_operations, NULL, {}),
2534 };
2535
2536 static struct dentry *proc_base_instantiate(struct inode *dir,
2537         struct dentry *dentry, struct task_struct *task, const void *ptr)
2538 {
2539         const struct pid_entry *p = ptr;
2540         struct inode *inode;
2541         struct proc_inode *ei;
2542         struct dentry *error;
2543
2544         /* Allocate the inode */
2545         error = ERR_PTR(-ENOMEM);
2546         inode = new_inode(dir->i_sb);
2547         if (!inode)
2548                 goto out;
2549
2550         /* Initialize the inode */
2551         ei = PROC_I(inode);
2552         inode->i_ino = get_next_ino();
2553         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2554
2555         /*
2556          * grab the reference to the task.
2557          */
2558         ei->pid = get_task_pid(task, PIDTYPE_PID);
2559         if (!ei->pid)
2560                 goto out_iput;
2561
2562         inode->i_mode = p->mode;
2563         if (S_ISDIR(inode->i_mode))
2564                 set_nlink(inode, 2);
2565         if (S_ISLNK(inode->i_mode))
2566                 inode->i_size = 64;
2567         if (p->iop)
2568                 inode->i_op = p->iop;
2569         if (p->fop)
2570                 inode->i_fop = p->fop;
2571         ei->op = p->op;
2572         d_add(dentry, inode);
2573         error = NULL;
2574 out:
2575         return error;
2576 out_iput:
2577         iput(inode);
2578         goto out;
2579 }
2580
2581 static struct dentry *proc_base_lookup(struct inode *dir, struct dentry *dentry)
2582 {
2583         struct dentry *error;
2584         struct task_struct *task = get_proc_task(dir);
2585         const struct pid_entry *p, *last;
2586
2587         error = ERR_PTR(-ENOENT);
2588
2589         if (!task)
2590                 goto out_no_task;
2591
2592         /* Lookup the directory entry */
2593         last = &proc_base_stuff[ARRAY_SIZE(proc_base_stuff) - 1];
2594         for (p = proc_base_stuff; p <= last; p++) {
2595                 if (p->len != dentry->d_name.len)
2596                         continue;
2597                 if (!memcmp(dentry->d_name.name, p->name, p->len))
2598                         break;
2599         }
2600         if (p > last)
2601                 goto out;
2602
2603         error = proc_base_instantiate(dir, dentry, task, p);
2604
2605 out:
2606         put_task_struct(task);
2607 out_no_task:
2608         return error;
2609 }
2610
2611 static int proc_base_fill_cache(struct file *filp, void *dirent,
2612         filldir_t filldir, struct task_struct *task, const struct pid_entry *p)
2613 {
2614         return proc_fill_cache(filp, dirent, filldir, p->name, p->len,
2615                                 proc_base_instantiate, task, p);
2616 }
2617
2618 #ifdef CONFIG_TASK_IO_ACCOUNTING
2619 static int do_io_accounting(struct task_struct *task, char *buffer, int whole)
2620 {
2621         struct task_io_accounting acct = task->ioac;
2622         unsigned long flags;
2623         int result;
2624
2625         result = mutex_lock_killable(&task->signal->cred_guard_mutex);
2626         if (result)
2627                 return result;
2628
2629         if (!ptrace_may_access(task, PTRACE_MODE_READ)) {
2630                 result = -EACCES;
2631                 goto out_unlock;
2632         }
2633
2634         if (whole && lock_task_sighand(task, &flags)) {
2635                 struct task_struct *t = task;
2636
2637                 task_io_accounting_add(&acct, &task->signal->ioac);
2638                 while_each_thread(task, t)
2639                         task_io_accounting_add(&acct, &t->ioac);
2640
2641                 unlock_task_sighand(task, &flags);
2642         }
2643         result = sprintf(buffer,
2644                         "rchar: %llu\n"
2645                         "wchar: %llu\n"
2646                         "syscr: %llu\n"
2647                         "syscw: %llu\n"
2648                         "read_bytes: %llu\n"
2649                         "write_bytes: %llu\n"
2650                         "cancelled_write_bytes: %llu\n",
2651                         (unsigned long long)acct.rchar,
2652                         (unsigned long long)acct.wchar,
2653                         (unsigned long long)acct.syscr,
2654                         (unsigned long long)acct.syscw,
2655                         (unsigned long long)acct.read_bytes,
2656                         (unsigned long long)acct.write_bytes,
2657                         (unsigned long long)acct.cancelled_write_bytes);
2658 out_unlock:
2659         mutex_unlock(&task->signal->cred_guard_mutex);
2660         return result;
2661 }
2662
2663 static int proc_tid_io_accounting(struct task_struct *task, char *buffer)
2664 {
2665         return do_io_accounting(task, buffer, 0);
2666 }
2667
2668 static int proc_tgid_io_accounting(struct task_struct *task, char *buffer)
2669 {
2670         return do_io_accounting(task, buffer, 1);
2671 }
2672 #endif /* CONFIG_TASK_IO_ACCOUNTING */
2673
2674 static int proc_pid_personality(struct seq_file *m, struct pid_namespace *ns,
2675                                 struct pid *pid, struct task_struct *task)
2676 {
2677         int err = lock_trace(task);
2678         if (!err) {
2679                 seq_printf(m, "%08x\n", task->personality);
2680                 unlock_trace(task);
2681         }
2682         return err;
2683 }
2684
2685 /*
2686  * Thread groups
2687  */
2688 static const struct file_operations proc_task_operations;
2689 static const struct inode_operations proc_task_inode_operations;
2690
2691 static const struct pid_entry tgid_base_stuff[] = {
2692         DIR("task",       S_IRUGO|S_IXUGO, proc_task_inode_operations, proc_task_operations),
2693         DIR("fd",         S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
2694         DIR("fdinfo",     S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
2695         DIR("ns",         S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
2696 #ifdef CONFIG_NET
2697         DIR("net",        S_IRUGO|S_IXUGO, proc_net_inode_operations, proc_net_operations),
2698 #endif
2699         REG("environ",    S_IRUSR, proc_environ_operations),
2700         INF("auxv",       S_IRUSR, proc_pid_auxv),
2701         ONE("status",     S_IRUGO, proc_pid_status),
2702         ONE("personality", S_IRUGO, proc_pid_personality),
2703         INF("limits",     S_IRUGO, proc_pid_limits),
2704 #ifdef CONFIG_SCHED_DEBUG
2705         REG("sched",      S_IRUGO|S_IWUSR, proc_pid_sched_operations),
2706 #endif
2707 #ifdef CONFIG_SCHED_AUTOGROUP
2708         REG("autogroup",  S_IRUGO|S_IWUSR, proc_pid_sched_autogroup_operations),
2709 #endif
2710         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
2711 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
2712         INF("syscall",    S_IRUGO, proc_pid_syscall),
2713 #endif
2714         INF("cmdline",    S_IRUGO, proc_pid_cmdline),
2715         ONE("stat",       S_IRUGO, proc_tgid_stat),
2716         ONE("statm",      S_IRUGO, proc_pid_statm),
2717         REG("maps",       S_IRUGO, proc_maps_operations),
2718 #ifdef CONFIG_NUMA
2719         REG("numa_maps",  S_IRUGO, proc_numa_maps_operations),
2720 #endif
2721         REG("mem",        S_IRUSR|S_IWUSR, proc_mem_operations),
2722         LNK("cwd",        proc_cwd_link),
2723         LNK("root",       proc_root_link),
2724         LNK("exe",        proc_exe_link),
2725         REG("mounts",     S_IRUGO, proc_mounts_operations),
2726         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
2727         REG("mountstats", S_IRUSR, proc_mountstats_operations),
2728 #ifdef CONFIG_PROC_PAGE_MONITOR
2729         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
2730         REG("smaps",      S_IRUGO, proc_smaps_operations),
2731         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
2732 #endif
2733 #ifdef CONFIG_SECURITY
2734         DIR("attr",       S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
2735 #endif
2736 #ifdef CONFIG_KALLSYMS
2737         INF("wchan",      S_IRUGO, proc_pid_wchan),
2738 #endif
2739 #ifdef CONFIG_STACKTRACE
2740         ONE("stack",      S_IRUGO, proc_pid_stack),
2741 #endif
2742 #ifdef CONFIG_SCHEDSTATS
2743         INF("schedstat",  S_IRUGO, proc_pid_schedstat),
2744 #endif
2745 #ifdef CONFIG_LATENCYTOP
2746         REG("latency",  S_IRUGO, proc_lstats_operations),
2747 #endif
2748 #ifdef CONFIG_PROC_PID_CPUSET
2749         REG("cpuset",     S_IRUGO, proc_cpuset_operations),
2750 #endif
2751 #ifdef CONFIG_CGROUPS
2752         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
2753 #endif
2754         INF("oom_score",  S_IRUGO, proc_oom_score),
2755         REG("oom_adj",    S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
2756         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
2757 #ifdef CONFIG_AUDITSYSCALL
2758         REG("loginuid",   S_IWUSR|S_IRUGO, proc_loginuid_operations),
2759         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
2760 #endif
2761 #ifdef CONFIG_FAULT_INJECTION
2762         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
2763 #endif
2764 #ifdef CONFIG_ELF_CORE
2765         REG("coredump_filter", S_IRUGO|S_IWUSR, proc_coredump_filter_operations),
2766 #endif
2767 #ifdef CONFIG_TASK_IO_ACCOUNTING
2768         INF("io",       S_IRUSR, proc_tgid_io_accounting),
2769 #endif
2770 #ifdef CONFIG_HARDWALL
2771         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
2772 #endif
2773 };
2774
2775 static int proc_tgid_base_readdir(struct file * filp,
2776                              void * dirent, filldir_t filldir)
2777 {
2778         return proc_pident_readdir(filp,dirent,filldir,
2779                                    tgid_base_stuff,ARRAY_SIZE(tgid_base_stuff));
2780 }
2781
2782 static const struct file_operations proc_tgid_base_operations = {
2783         .read           = generic_read_dir,
2784         .readdir        = proc_tgid_base_readdir,
2785         .llseek         = default_llseek,
2786 };
2787
2788 static struct dentry *proc_tgid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
2789         return proc_pident_lookup(dir, dentry,
2790                                   tgid_base_stuff, ARRAY_SIZE(tgid_base_stuff));
2791 }
2792
2793 static const struct inode_operations proc_tgid_base_inode_operations = {
2794         .lookup         = proc_tgid_base_lookup,
2795         .getattr        = pid_getattr,
2796         .setattr        = proc_setattr,
2797 };
2798
2799 static void proc_flush_task_mnt(struct vfsmount *mnt, pid_t pid, pid_t tgid)
2800 {
2801         struct dentry *dentry, *leader, *dir;
2802         char buf[PROC_NUMBUF];
2803         struct qstr name;
2804
2805         name.name = buf;
2806         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2807         dentry = d_hash_and_lookup(mnt->mnt_root, &name);
2808         if (dentry) {
2809                 shrink_dcache_parent(dentry);
2810                 d_drop(dentry);
2811                 dput(dentry);
2812         }
2813
2814         name.name = buf;
2815         name.len = snprintf(buf, sizeof(buf), "%d", tgid);
2816         leader = d_hash_and_lookup(mnt->mnt_root, &name);
2817         if (!leader)
2818                 goto out;
2819
2820         name.name = "task";
2821         name.len = strlen(name.name);
2822         dir = d_hash_and_lookup(leader, &name);
2823         if (!dir)
2824                 goto out_put_leader;
2825
2826         name.name = buf;
2827         name.len = snprintf(buf, sizeof(buf), "%d", pid);
2828         dentry = d_hash_and_lookup(dir, &name);
2829         if (dentry) {
2830                 shrink_dcache_parent(dentry);
2831                 d_drop(dentry);
2832                 dput(dentry);
2833         }
2834
2835         dput(dir);
2836 out_put_leader:
2837         dput(leader);
2838 out:
2839         return;
2840 }
2841
2842 /**
2843  * proc_flush_task -  Remove dcache entries for @task from the /proc dcache.
2844  * @task: task that should be flushed.
2845  *
2846  * When flushing dentries from proc, one needs to flush them from global
2847  * proc (proc_mnt) and from all the namespaces' procs this task was seen
2848  * in. This call is supposed to do all of this job.
2849  *
2850  * Looks in the dcache for
2851  * /proc/@pid
2852  * /proc/@tgid/task/@pid
2853  * if either directory is present flushes it and all of it'ts children
2854  * from the dcache.
2855  *
2856  * It is safe and reasonable to cache /proc entries for a task until
2857  * that task exits.  After that they just clog up the dcache with
2858  * useless entries, possibly causing useful dcache entries to be
2859  * flushed instead.  This routine is proved to flush those useless
2860  * dcache entries at process exit time.
2861  *
2862  * NOTE: This routine is just an optimization so it does not guarantee
2863  *       that no dcache entries will exist at process exit time it
2864  *       just makes it very unlikely that any will persist.
2865  */
2866
2867 void proc_flush_task(struct task_struct *task)
2868 {
2869         int i;
2870         struct pid *pid, *tgid;
2871         struct upid *upid;
2872
2873         pid = task_pid(task);
2874         tgid = task_tgid(task);
2875
2876         for (i = 0; i <= pid->level; i++) {
2877                 upid = &pid->numbers[i];
2878                 proc_flush_task_mnt(upid->ns->proc_mnt, upid->nr,
2879                                         tgid->numbers[i].nr);
2880         }
2881
2882         upid = &pid->numbers[pid->level];
2883         if (upid->nr == 1)
2884                 pid_ns_release_proc(upid->ns);
2885 }
2886
2887 static struct dentry *proc_pid_instantiate(struct inode *dir,
2888                                            struct dentry * dentry,
2889                                            struct task_struct *task, const void *ptr)
2890 {
2891         struct dentry *error = ERR_PTR(-ENOENT);
2892         struct inode *inode;
2893
2894         inode = proc_pid_make_inode(dir->i_sb, task);
2895         if (!inode)
2896                 goto out;
2897
2898         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
2899         inode->i_op = &proc_tgid_base_inode_operations;
2900         inode->i_fop = &proc_tgid_base_operations;
2901         inode->i_flags|=S_IMMUTABLE;
2902
2903         set_nlink(inode, 2 + pid_entry_count_dirs(tgid_base_stuff,
2904                                                   ARRAY_SIZE(tgid_base_stuff)));
2905
2906         d_set_d_op(dentry, &pid_dentry_operations);
2907
2908         d_add(dentry, inode);
2909         /* Close the race of the process dying before we return the dentry */
2910         if (pid_revalidate(dentry, NULL))
2911                 error = NULL;
2912 out:
2913         return error;
2914 }
2915
2916 struct dentry *proc_pid_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
2917 {
2918         struct dentry *result;
2919         struct task_struct *task;
2920         unsigned tgid;
2921         struct pid_namespace *ns;
2922
2923         result = proc_base_lookup(dir, dentry);
2924         if (!IS_ERR(result) || PTR_ERR(result) != -ENOENT)
2925                 goto out;
2926
2927         tgid = name_to_int(dentry);
2928         if (tgid == ~0U)
2929                 goto out;
2930
2931         ns = dentry->d_sb->s_fs_info;
2932         rcu_read_lock();
2933         task = find_task_by_pid_ns(tgid, ns);
2934         if (task)
2935                 get_task_struct(task);
2936         rcu_read_unlock();
2937         if (!task)
2938                 goto out;
2939
2940         result = proc_pid_instantiate(dir, dentry, task, NULL);
2941         put_task_struct(task);
2942 out:
2943         return result;
2944 }
2945
2946 /*
2947  * Find the first task with tgid >= tgid
2948  *
2949  */
2950 struct tgid_iter {
2951         unsigned int tgid;
2952         struct task_struct *task;
2953 };
2954 static struct tgid_iter next_tgid(struct pid_namespace *ns, struct tgid_iter iter)
2955 {
2956         struct pid *pid;
2957
2958         if (iter.task)
2959                 put_task_struct(iter.task);
2960         rcu_read_lock();
2961 retry:
2962         iter.task = NULL;
2963         pid = find_ge_pid(iter.tgid, ns);
2964         if (pid) {
2965                 iter.tgid = pid_nr_ns(pid, ns);
2966                 iter.task = pid_task(pid, PIDTYPE_PID);
2967                 /* What we to know is if the pid we have find is the
2968                  * pid of a thread_group_leader.  Testing for task
2969                  * being a thread_group_leader is the obvious thing
2970                  * todo but there is a window when it fails, due to
2971                  * the pid transfer logic in de_thread.
2972                  *
2973                  * So we perform the straight forward test of seeing
2974                  * if the pid we have found is the pid of a thread
2975                  * group leader, and don't worry if the task we have
2976                  * found doesn't happen to be a thread group leader.
2977                  * As we don't care in the case of readdir.
2978                  */
2979                 if (!iter.task || !has_group_leader_pid(iter.task)) {
2980                         iter.tgid += 1;
2981                         goto retry;
2982                 }
2983                 get_task_struct(iter.task);
2984         }
2985         rcu_read_unlock();
2986         return iter;
2987 }
2988
2989 #define TGID_OFFSET (FIRST_PROCESS_ENTRY + ARRAY_SIZE(proc_base_stuff))
2990
2991 static int proc_pid_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
2992         struct tgid_iter iter)
2993 {
2994         char name[PROC_NUMBUF];
2995         int len = snprintf(name, sizeof(name), "%d", iter.tgid);
2996         return proc_fill_cache(filp, dirent, filldir, name, len,
2997                                 proc_pid_instantiate, iter.task, NULL);
2998 }
2999
3000 /* for the /proc/ directory itself, after non-process stuff has been done */
3001 int proc_pid_readdir(struct file * filp, void * dirent, filldir_t filldir)
3002 {
3003         unsigned int nr;
3004         struct task_struct *reaper;
3005         struct tgid_iter iter;
3006         struct pid_namespace *ns;
3007
3008         if (filp->f_pos >= PID_MAX_LIMIT + TGID_OFFSET)
3009                 goto out_no_task;
3010         nr = filp->f_pos - FIRST_PROCESS_ENTRY;
3011
3012         reaper = get_proc_task(filp->f_path.dentry->d_inode);
3013         if (!reaper)
3014                 goto out_no_task;
3015
3016         for (; nr < ARRAY_SIZE(proc_base_stuff); filp->f_pos++, nr++) {
3017                 const struct pid_entry *p = &proc_base_stuff[nr];
3018                 if (proc_base_fill_cache(filp, dirent, filldir, reaper, p) < 0)
3019                         goto out;
3020         }
3021
3022         ns = filp->f_dentry->d_sb->s_fs_info;
3023         iter.task = NULL;
3024         iter.tgid = filp->f_pos - TGID_OFFSET;
3025         for (iter = next_tgid(ns, iter);
3026              iter.task;
3027              iter.tgid += 1, iter = next_tgid(ns, iter)) {
3028                 filp->f_pos = iter.tgid + TGID_OFFSET;
3029                 if (proc_pid_fill_cache(filp, dirent, filldir, iter) < 0) {
3030                         put_task_struct(iter.task);
3031                         goto out;
3032                 }
3033         }
3034         filp->f_pos = PID_MAX_LIMIT + TGID_OFFSET;
3035 out:
3036         put_task_struct(reaper);
3037 out_no_task:
3038         return 0;
3039 }
3040
3041 /*
3042  * Tasks
3043  */
3044 static const struct pid_entry tid_base_stuff[] = {
3045         DIR("fd",        S_IRUSR|S_IXUSR, proc_fd_inode_operations, proc_fd_operations),
3046         DIR("fdinfo",    S_IRUSR|S_IXUSR, proc_fdinfo_inode_operations, proc_fdinfo_operations),
3047         DIR("ns",        S_IRUSR|S_IXUGO, proc_ns_dir_inode_operations, proc_ns_dir_operations),
3048         REG("environ",   S_IRUSR, proc_environ_operations),
3049         INF("auxv",      S_IRUSR, proc_pid_auxv),
3050         ONE("status",    S_IRUGO, proc_pid_status),
3051         ONE("personality", S_IRUGO, proc_pid_personality),
3052         INF("limits",    S_IRUGO, proc_pid_limits),
3053 #ifdef CONFIG_SCHED_DEBUG
3054         REG("sched",     S_IRUGO|S_IWUSR, proc_pid_sched_operations),
3055 #endif
3056         REG("comm",      S_IRUGO|S_IWUSR, proc_pid_set_comm_operations),
3057 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
3058         INF("syscall",   S_IRUGO, proc_pid_syscall),
3059 #endif
3060         INF("cmdline",   S_IRUGO, proc_pid_cmdline),
3061         ONE("stat",      S_IRUGO, proc_tid_stat),
3062         ONE("statm",     S_IRUGO, proc_pid_statm),
3063         REG("maps",      S_IRUGO, proc_maps_operations),
3064 #ifdef CONFIG_NUMA
3065         REG("numa_maps", S_IRUGO, proc_numa_maps_operations),
3066 #endif
3067         REG("mem",       S_IRUSR|S_IWUSR, proc_mem_operations),
3068         LNK("cwd",       proc_cwd_link),
3069         LNK("root",      proc_root_link),
3070         LNK("exe",       proc_exe_link),
3071         REG("mounts",    S_IRUGO, proc_mounts_operations),
3072         REG("mountinfo",  S_IRUGO, proc_mountinfo_operations),
3073 #ifdef CONFIG_PROC_PAGE_MONITOR
3074         REG("clear_refs", S_IWUSR, proc_clear_refs_operations),
3075         REG("smaps",     S_IRUGO, proc_smaps_operations),
3076         REG("pagemap",    S_IRUGO, proc_pagemap_operations),
3077 #endif
3078 #ifdef CONFIG_SECURITY
3079         DIR("attr",      S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations),
3080 #endif
3081 #ifdef CONFIG_KALLSYMS
3082         INF("wchan",     S_IRUGO, proc_pid_wchan),
3083 #endif
3084 #ifdef CONFIG_STACKTRACE
3085         ONE("stack",      S_IRUGO, proc_pid_stack),
3086 #endif
3087 #ifdef CONFIG_SCHEDSTATS
3088         INF("schedstat", S_IRUGO, proc_pid_schedstat),
3089 #endif
3090 #ifdef CONFIG_LATENCYTOP
3091         REG("latency",  S_IRUGO, proc_lstats_operations),
3092 #endif
3093 #ifdef CONFIG_PROC_PID_CPUSET
3094         REG("cpuset",    S_IRUGO, proc_cpuset_operations),
3095 #endif
3096 #ifdef CONFIG_CGROUPS
3097         REG("cgroup",  S_IRUGO, proc_cgroup_operations),
3098 #endif
3099         INF("oom_score", S_IRUGO, proc_oom_score),
3100         REG("oom_adj",   S_IRUGO|S_IWUSR, proc_oom_adjust_operations),
3101         REG("oom_score_adj", S_IRUGO|S_IWUSR, proc_oom_score_adj_operations),
3102 #ifdef CONFIG_AUDITSYSCALL
3103         REG("loginuid",  S_IWUSR|S_IRUGO, proc_loginuid_operations),
3104         REG("sessionid",  S_IRUGO, proc_sessionid_operations),
3105 #endif
3106 #ifdef CONFIG_FAULT_INJECTION
3107         REG("make-it-fail", S_IRUGO|S_IWUSR, proc_fault_inject_operations),
3108 #endif
3109 #ifdef CONFIG_TASK_IO_ACCOUNTING
3110         INF("io",       S_IRUSR, proc_tid_io_accounting),
3111 #endif
3112 #ifdef CONFIG_HARDWALL
3113         INF("hardwall",   S_IRUGO, proc_pid_hardwall),
3114 #endif
3115 };
3116
3117 static int proc_tid_base_readdir(struct file * filp,
3118                              void * dirent, filldir_t filldir)
3119 {
3120         return proc_pident_readdir(filp,dirent,filldir,
3121                                    tid_base_stuff,ARRAY_SIZE(tid_base_stuff));
3122 }
3123
3124 static struct dentry *proc_tid_base_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd){
3125         return proc_pident_lookup(dir, dentry,
3126                                   tid_base_stuff, ARRAY_SIZE(tid_base_stuff));
3127 }
3128
3129 static const struct file_operations proc_tid_base_operations = {
3130         .read           = generic_read_dir,
3131         .readdir        = proc_tid_base_readdir,
3132         .llseek         = default_llseek,
3133 };
3134
3135 static const struct inode_operations proc_tid_base_inode_operations = {
3136         .lookup         = proc_tid_base_lookup,
3137         .getattr        = pid_getattr,
3138         .setattr        = proc_setattr,
3139 };
3140
3141 static struct dentry *proc_task_instantiate(struct inode *dir,
3142         struct dentry *dentry, struct task_struct *task, const void *ptr)
3143 {
3144         struct dentry *error = ERR_PTR(-ENOENT);
3145         struct inode *inode;
3146         inode = proc_pid_make_inode(dir->i_sb, task);
3147
3148         if (!inode)
3149                 goto out;
3150         inode->i_mode = S_IFDIR|S_IRUGO|S_IXUGO;
3151         inode->i_op = &proc_tid_base_inode_operations;
3152         inode->i_fop = &proc_tid_base_operations;
3153         inode->i_flags|=S_IMMUTABLE;
3154
3155         set_nlink(inode, 2 + pid_entry_count_dirs(tid_base_stuff,
3156                                                   ARRAY_SIZE(tid_base_stuff)));
3157
3158         d_set_d_op(dentry, &pid_dentry_operations);
3159
3160         d_add(dentry, inode);
3161         /* Close the race of the process dying before we return the dentry */
3162         if (pid_revalidate(dentry, NULL))
3163                 error = NULL;
3164 out:
3165         return error;
3166 }
3167
3168 static struct dentry *proc_task_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
3169 {
3170         struct dentry *result = ERR_PTR(-ENOENT);
3171         struct task_struct *task;
3172         struct task_struct *leader = get_proc_task(dir);
3173         unsigned tid;
3174         struct pid_namespace *ns;
3175
3176         if (!leader)
3177                 goto out_no_task;
3178
3179         tid = name_to_int(dentry);
3180         if (tid == ~0U)
3181                 goto out;
3182
3183         ns = dentry->d_sb->s_fs_info;
3184         rcu_read_lock();
3185         task = find_task_by_pid_ns(tid, ns);
3186         if (task)
3187                 get_task_struct(task);
3188         rcu_read_unlock();
3189         if (!task)
3190                 goto out;
3191         if (!same_thread_group(leader, task))
3192                 goto out_drop_task;
3193
3194         result = proc_task_instantiate(dir, dentry, task, NULL);
3195 out_drop_task:
3196         put_task_struct(task);
3197 out:
3198         put_task_struct(leader);
3199 out_no_task:
3200         return result;
3201 }
3202
3203 /*
3204  * Find the first tid of a thread group to return to user space.
3205  *
3206  * Usually this is just the thread group leader, but if the users
3207  * buffer was too small or there was a seek into the middle of the
3208  * directory we have more work todo.
3209  *
3210  * In the case of a short read we start with find_task_by_pid.
3211  *
3212  * In the case of a seek we start with the leader and walk nr
3213  * threads past it.
3214  */
3215 static struct task_struct *first_tid(struct task_struct *leader,
3216                 int tid, int nr, struct pid_namespace *ns)
3217 {
3218         struct task_struct *pos;
3219
3220         rcu_read_lock();
3221         /* Attempt to start with the pid of a thread */
3222         if (tid && (nr > 0)) {
3223                 pos = find_task_by_pid_ns(tid, ns);
3224                 if (pos && (pos->group_leader == leader))
3225                         goto found;
3226         }
3227
3228         /* If nr exceeds the number of threads there is nothing todo */
3229         pos = NULL;
3230         if (nr && nr >= get_nr_threads(leader))
3231                 goto out;
3232
3233         /* If we haven't found our starting place yet start
3234          * with the leader and walk nr threads forward.
3235          */
3236         for (pos = leader; nr > 0; --nr) {
3237                 pos = next_thread(pos);
3238                 if (pos == leader) {
3239                         pos = NULL;
3240                         goto out;
3241                 }
3242         }
3243 found:
3244         get_task_struct(pos);
3245 out:
3246         rcu_read_unlock();
3247         return pos;
3248 }
3249
3250 /*
3251  * Find the next thread in the thread list.
3252  * Return NULL if there is an error or no next thread.
3253  *
3254  * The reference to the input task_struct is released.
3255  */
3256 static struct task_struct *next_tid(struct task_struct *start)
3257 {
3258         struct task_struct *pos = NULL;
3259         rcu_read_lock();
3260         if (pid_alive(start)) {
3261                 pos = next_thread(start);
3262                 if (thread_group_leader(pos))
3263                         pos = NULL;
3264                 else
3265                         get_task_struct(pos);
3266         }
3267         rcu_read_unlock();
3268         put_task_struct(start);
3269         return pos;
3270 }
3271
3272 static int proc_task_fill_cache(struct file *filp, void *dirent, filldir_t filldir,
3273         struct task_struct *task, int tid)
3274 {
3275         char name[PROC_NUMBUF];
3276         int len = snprintf(name, sizeof(name), "%d", tid);
3277         return proc_fill_cache(filp, dirent, filldir, name, len,
3278                                 proc_task_instantiate, task, NULL);
3279 }
3280
3281 /* for the /proc/TGID/task/ directories */
3282 static int proc_task_readdir(struct file * filp, void * dirent, filldir_t filldir)
3283 {
3284         struct dentry *dentry = filp->f_path.dentry;
3285         struct inode *inode = dentry->d_inode;
3286         struct task_struct *leader = NULL;
3287         struct task_struct *task;
3288         int retval = -ENOENT;
3289         ino_t ino;
3290         int tid;
3291         struct pid_namespace *ns;
3292
3293         task = get_proc_task(inode);
3294         if (!task)
3295                 goto out_no_task;
3296         rcu_read_lock();
3297         if (pid_alive(task)) {
3298                 leader = task->group_leader;
3299                 get_task_struct(leader);
3300         }
3301         rcu_read_unlock();
3302         put_task_struct(task);
3303         if (!leader)
3304                 goto out_no_task;
3305         retval = 0;
3306
3307         switch ((unsigned long)filp->f_pos) {
3308         case 0:
3309                 ino = inode->i_ino;
3310                 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) < 0)
3311                         goto out;
3312                 filp->f_pos++;
3313                 /* fall through */
3314         case 1:
3315                 ino = parent_ino(dentry);
3316                 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) < 0)
3317                         goto out;
3318                 filp->f_pos++;
3319                 /* fall through */
3320         }
3321
3322         /* f_version caches the tgid value that the last readdir call couldn't
3323          * return. lseek aka telldir automagically resets f_version to 0.
3324          */
3325         ns = filp->f_dentry->d_sb->s_fs_info;
3326         tid = (int)filp->f_version;
3327         filp->f_version = 0;
3328         for (task = first_tid(leader, tid, filp->f_pos - 2, ns);
3329              task;
3330              task = next_tid(task), filp->f_pos++) {
3331                 tid = task_pid_nr_ns(task, ns);
3332                 if (proc_task_fill_cache(filp, dirent, filldir, task, tid) < 0) {
3333                         /* returning this tgid failed, save it as the first
3334                          * pid for the next readir call */
3335                         filp->f_version = (u64)tid;
3336                         put_task_struct(task);
3337                         break;
3338                 }
3339         }
3340 out:
3341         put_task_struct(leader);
3342 out_no_task:
3343         return retval;
3344 }
3345
3346 static int proc_task_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
3347 {
3348         struct inode *inode = dentry->d_inode;
3349         struct task_struct *p = get_proc_task(inode);
3350         generic_fillattr(inode, stat);
3351
3352         if (p) {
3353                 stat->nlink += get_nr_threads(p);
3354                 put_task_struct(p);
3355         }
3356
3357         return 0;
3358 }
3359
3360 static const struct inode_operations proc_task_inode_operations = {
3361         .lookup         = proc_task_lookup,
3362         .getattr        = proc_task_getattr,
3363         .setattr        = proc_setattr,
3364 };
3365
3366 static const struct file_operations proc_task_operations = {
3367         .read           = generic_read_dir,
3368         .readdir        = proc_task_readdir,
3369         .llseek         = default_llseek,
3370 };